2015-04-28 15:27:52 +02:00
|
|
|
/*
|
|
|
|
* Block layer I/O functions
|
|
|
|
*
|
|
|
|
* Copyright (c) 2003 Fabrice Bellard
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
|
|
* in the Software without restriction, including without limitation the rights
|
|
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
|
|
* furnished to do so, subject to the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice shall be included in
|
|
|
|
* all copies or substantial portions of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
|
|
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
|
|
* THE SOFTWARE.
|
|
|
|
*/
|
|
|
|
|
2016-01-18 19:01:42 +01:00
|
|
|
#include "qemu/osdep.h"
|
2015-04-28 15:27:52 +02:00
|
|
|
#include "trace.h"
|
2015-10-19 17:53:21 +02:00
|
|
|
#include "sysemu/block-backend.h"
|
2018-02-16 17:50:12 +01:00
|
|
|
#include "block/aio-wait.h"
|
2015-04-28 15:27:52 +02:00
|
|
|
#include "block/blockjob.h"
|
2017-05-08 16:13:03 +02:00
|
|
|
#include "block/blockjob_int.h"
|
2015-04-28 15:27:52 +02:00
|
|
|
#include "block/block_int.h"
|
2016-03-20 18:16:19 +01:00
|
|
|
#include "qemu/cutils.h"
|
include/qemu/osdep.h: Don't include qapi/error.h
Commit 57cb38b included qapi/error.h into qemu/osdep.h to get the
Error typedef. Since then, we've moved to include qemu/osdep.h
everywhere. Its file comment explains: "To avoid getting into
possible circular include dependencies, this file should not include
any other QEMU headers, with the exceptions of config-host.h,
compiler.h, os-posix.h and os-win32.h, all of which are doing a
similar job to this file and are under similar constraints."
qapi/error.h doesn't do a similar job, and it doesn't adhere to
similar constraints: it includes qapi-types.h. That's in excess of
100KiB of crap most .c files don't actually need.
Add the typedef to qemu/typedefs.h, and include that instead of
qapi/error.h. Include qapi/error.h in .c files that need it and don't
get it now. Include qapi-types.h in qom/object.h for uint16List.
Update scripts/clean-includes accordingly. Update it further to match
reality: replace config.h by config-target.h, add sysemu/os-posix.h,
sysemu/os-win32.h. Update the list of includes in the qemu/osdep.h
comment quoted above similarly.
This reduces the number of objects depending on qapi/error.h from "all
of them" to less than a third. Unfortunately, the number depending on
qapi-types.h shrinks only a little. More work is needed for that one.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
[Fix compilation without the spice devel packages. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2016-03-14 09:01:28 +01:00
|
|
|
#include "qapi/error.h"
|
2015-03-17 18:29:20 +01:00
|
|
|
#include "qemu/error-report.h"
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
|
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
/* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
|
|
|
|
#define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
|
|
|
|
|
2016-06-01 23:10:03 +02:00
|
|
|
static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
|
2017-06-09 12:18:08 +02:00
|
|
|
int64_t offset, int bytes, BdrvRequestFlags flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-12-07 13:03:13 +01:00
|
|
|
void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-11-29 11:25:10 +01:00
|
|
|
BdrvChild *c, *next;
|
2016-03-21 12:56:44 +01:00
|
|
|
|
2017-11-29 11:25:10 +01:00
|
|
|
QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
|
2017-12-07 13:03:13 +01:00
|
|
|
if (c == ignore) {
|
|
|
|
continue;
|
|
|
|
}
|
2016-03-22 12:05:35 +01:00
|
|
|
if (c->role->drained_begin) {
|
|
|
|
c->role->drained_begin(c);
|
|
|
|
}
|
2016-04-07 18:33:33 +02:00
|
|
|
}
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-12-07 13:03:13 +01:00
|
|
|
void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore)
|
2016-04-07 18:33:33 +02:00
|
|
|
{
|
2017-11-29 11:25:10 +01:00
|
|
|
BdrvChild *c, *next;
|
2016-03-21 12:56:44 +01:00
|
|
|
|
2017-11-29 11:25:10 +01:00
|
|
|
QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
|
2017-12-07 13:03:13 +01:00
|
|
|
if (c == ignore) {
|
|
|
|
continue;
|
|
|
|
}
|
2016-03-22 12:05:35 +01:00
|
|
|
if (c->role->drained_end) {
|
|
|
|
c->role->drained_end(c);
|
|
|
|
}
|
2016-03-21 12:56:44 +01:00
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2018-03-22 14:11:20 +01:00
|
|
|
static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore)
|
|
|
|
{
|
|
|
|
BdrvChild *c, *next;
|
|
|
|
bool busy = false;
|
|
|
|
|
|
|
|
QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
|
|
|
|
if (c == ignore) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (c->role->drained_poll) {
|
|
|
|
busy |= c->role->drained_poll(c);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return busy;
|
|
|
|
}
|
|
|
|
|
2016-06-24 00:37:23 +02:00
|
|
|
static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src)
|
|
|
|
{
|
|
|
|
dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer);
|
|
|
|
dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer);
|
|
|
|
dst->opt_mem_alignment = MAX(dst->opt_mem_alignment,
|
|
|
|
src->opt_mem_alignment);
|
|
|
|
dst->min_mem_alignment = MAX(dst->min_mem_alignment,
|
|
|
|
src->min_mem_alignment);
|
|
|
|
dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov);
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
|
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
Error *local_err = NULL;
|
|
|
|
|
|
|
|
memset(&bs->bl, 0, sizeof(bs->bl));
|
|
|
|
|
|
|
|
if (!drv) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2016-06-24 00:37:18 +02:00
|
|
|
/* Default alignment based on whether driver has byte interface */
|
block: Support byte-based aio callbacks
We are gradually moving away from sector-based interfaces, towards
byte-based. Add new sector-based aio callbacks for read and write,
to match the fact that bdrv_aio_pdiscard is already byte-based.
Ideally, drivers should be converted to use coroutine callbacks
rather than aio; but that is not quite as trivial (and if we were
to do that conversion, the null-aio driver would disappear), so for
the short term, converting the signature but keeping things with
aio is easier. However, we CAN declare that a driver that uses
the byte-based aio interfaces now defaults to byte-based
operations, and must explicitly provide a refresh_limits override
to stick with larger alignments (making the alignment issues more
obvious directly in the drivers touched in the next few patches).
Once all drivers are converted, the sector-based aio callbacks will
be removed; in the meantime, a FIXME comment is added due to a
slight inefficiency that will be touched up as part of that later
cleanup.
Simplify some instances of 'bs->drv' into 'drv' while touching this,
since the local variable already exists to reduce typing.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2018-04-24 21:25:01 +02:00
|
|
|
bs->bl.request_alignment = (drv->bdrv_co_preadv ||
|
|
|
|
drv->bdrv_aio_preadv) ? 1 : 512;
|
2016-06-24 00:37:18 +02:00
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/* Take some limits from the children as a default */
|
|
|
|
if (bs->file) {
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_refresh_limits(bs->file->bs, &local_err);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (local_err) {
|
|
|
|
error_propagate(errp, local_err);
|
|
|
|
return;
|
|
|
|
}
|
2016-06-24 00:37:23 +02:00
|
|
|
bdrv_merge_limits(&bs->bl, &bs->file->bs->bl);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
2015-05-12 16:30:55 +02:00
|
|
|
bs->bl.min_mem_alignment = 512;
|
block: align bounce buffers to page
The following sequence
int fd = open(argv[1], O_RDWR | O_CREAT | O_DIRECT, 0644);
for (i = 0; i < 100000; i++)
write(fd, buf, 4096);
performs 5% better if buf is aligned to 4096 bytes.
The difference is quite reliable.
On the other hand we do not want at the moment to enforce bounce
buffering if guest request is aligned to 512 bytes.
The patch changes default bounce buffer optimal alignment to
MAX(page size, 4k). 4k is chosen as maximal known sector size on real
HDD.
The justification of the performance improve is quite interesting.
From the kernel point of view each request to the disk was split
by two. This could be seen by blktrace like this:
9,0 11 1 0.000000000 11151 Q WS 312737792 + 1023 [qemu-img]
9,0 11 2 0.000007938 11151 Q WS 312738815 + 8 [qemu-img]
9,0 11 3 0.000030735 11151 Q WS 312738823 + 1016 [qemu-img]
9,0 11 4 0.000032482 11151 Q WS 312739839 + 8 [qemu-img]
9,0 11 5 0.000041379 11151 Q WS 312739847 + 1016 [qemu-img]
9,0 11 6 0.000042818 11151 Q WS 312740863 + 8 [qemu-img]
9,0 11 7 0.000051236 11151 Q WS 312740871 + 1017 [qemu-img]
9,0 5 1 0.169071519 11151 Q WS 312741888 + 1023 [qemu-img]
After the patch the pattern becomes normal:
9,0 6 1 0.000000000 12422 Q WS 314834944 + 1024 [qemu-img]
9,0 6 2 0.000038527 12422 Q WS 314835968 + 1024 [qemu-img]
9,0 6 3 0.000072849 12422 Q WS 314836992 + 1024 [qemu-img]
9,0 6 4 0.000106276 12422 Q WS 314838016 + 1024 [qemu-img]
and the amount of requests sent to disk (could be calculated counting
number of lines in the output of blktrace) is reduced about 2 times.
Both qemu-img and qemu-io are affected while qemu-kvm is not. The guest
does his job well and real requests comes properly aligned (to page).
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431441056-26198-3-git-send-email-den@openvz.org
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2015-05-12 16:30:56 +02:00
|
|
|
bs->bl.opt_mem_alignment = getpagesize();
|
2015-07-09 11:56:44 +02:00
|
|
|
|
|
|
|
/* Safe default since most protocols use readv()/writev()/etc */
|
|
|
|
bs->bl.max_iov = IOV_MAX;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2015-06-17 14:55:21 +02:00
|
|
|
if (bs->backing) {
|
|
|
|
bdrv_refresh_limits(bs->backing->bs, &local_err);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (local_err) {
|
|
|
|
error_propagate(errp, local_err);
|
|
|
|
return;
|
|
|
|
}
|
2016-06-24 00:37:23 +02:00
|
|
|
bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Then let the driver override it */
|
|
|
|
if (drv->bdrv_refresh_limits) {
|
|
|
|
drv->bdrv_refresh_limits(bs, errp);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* The copy-on-read flag is actually a reference count so multiple users may
|
|
|
|
* use the feature without worrying about clobbering its previous state.
|
|
|
|
* Copy-on-read stays enabled until all users have called to disable it.
|
|
|
|
*/
|
|
|
|
void bdrv_enable_copy_on_read(BlockDriverState *bs)
|
|
|
|
{
|
2017-06-05 14:38:50 +02:00
|
|
|
atomic_inc(&bs->copy_on_read);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_disable_copy_on_read(BlockDriverState *bs)
|
|
|
|
{
|
2017-06-05 14:38:50 +02:00
|
|
|
int old = atomic_fetch_dec(&bs->copy_on_read);
|
|
|
|
assert(old >= 1);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-06-29 15:27:45 +02:00
|
|
|
typedef struct {
|
|
|
|
Coroutine *co;
|
|
|
|
BlockDriverState *bs;
|
|
|
|
bool done;
|
2017-09-23 13:14:09 +02:00
|
|
|
bool begin;
|
2017-12-06 17:05:44 +01:00
|
|
|
bool recursive;
|
2018-03-23 12:40:41 +01:00
|
|
|
bool poll;
|
2017-12-07 13:03:13 +01:00
|
|
|
BdrvChild *parent;
|
2017-06-29 15:27:45 +02:00
|
|
|
} BdrvCoDrainData;
|
|
|
|
|
|
|
|
static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
|
|
|
|
{
|
|
|
|
BdrvCoDrainData *data = opaque;
|
|
|
|
BlockDriverState *bs = data->bs;
|
|
|
|
|
2017-09-23 13:14:09 +02:00
|
|
|
if (data->begin) {
|
2017-09-23 13:14:10 +02:00
|
|
|
bs->drv->bdrv_co_drain_begin(bs);
|
2017-09-23 13:14:09 +02:00
|
|
|
} else {
|
|
|
|
bs->drv->bdrv_co_drain_end(bs);
|
|
|
|
}
|
2017-06-29 15:27:45 +02:00
|
|
|
|
|
|
|
/* Set data->done before reading bs->wakeup. */
|
|
|
|
atomic_mb_set(&data->done, true);
|
|
|
|
bdrv_wakeup(bs);
|
|
|
|
}
|
|
|
|
|
2017-12-05 12:52:09 +01:00
|
|
|
/* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
|
2017-12-14 10:40:20 +01:00
|
|
|
static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
|
2017-06-29 15:27:45 +02:00
|
|
|
{
|
2017-09-23 13:14:09 +02:00
|
|
|
BdrvCoDrainData data = { .bs = bs, .done = false, .begin = begin};
|
2017-06-29 15:27:45 +02:00
|
|
|
|
2017-09-23 13:14:10 +02:00
|
|
|
if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) ||
|
2017-09-23 13:14:09 +02:00
|
|
|
(!begin && !bs->drv->bdrv_co_drain_end)) {
|
2017-06-29 15:27:45 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
data.co = qemu_coroutine_create(bdrv_drain_invoke_entry, &data);
|
|
|
|
bdrv_coroutine_enter(bs, data.co);
|
|
|
|
BDRV_POLL_WHILE(bs, !data.done);
|
|
|
|
}
|
|
|
|
|
2018-03-22 10:57:14 +01:00
|
|
|
/* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
|
2018-03-23 12:40:41 +01:00
|
|
|
bool bdrv_drain_poll(BlockDriverState *bs, bool recursive,
|
|
|
|
BdrvChild *ignore_parent)
|
2018-03-22 14:11:20 +01:00
|
|
|
{
|
2018-03-23 12:40:41 +01:00
|
|
|
BdrvChild *child, *next;
|
|
|
|
|
2018-03-22 14:11:20 +01:00
|
|
|
if (bdrv_parent_drained_poll(bs, ignore_parent)) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2018-03-23 12:40:41 +01:00
|
|
|
if (atomic_read(&bs->in_flight)) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (recursive) {
|
|
|
|
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
|
|
|
|
if (bdrv_drain_poll(child->bs, recursive, child)) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
2018-03-22 14:11:20 +01:00
|
|
|
}
|
|
|
|
|
2018-03-23 12:40:41 +01:00
|
|
|
static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive,
|
2018-03-22 14:11:20 +01:00
|
|
|
BdrvChild *ignore_parent)
|
2018-03-22 10:57:14 +01:00
|
|
|
{
|
|
|
|
/* Execute pending BHs first and check everything else only after the BHs
|
|
|
|
* have executed. */
|
|
|
|
while (aio_poll(bs->aio_context, false));
|
2018-03-22 14:11:20 +01:00
|
|
|
|
2018-03-23 12:40:41 +01:00
|
|
|
return bdrv_drain_poll(bs, recursive, ignore_parent);
|
2018-03-22 10:57:14 +01:00
|
|
|
}
|
|
|
|
|
2017-12-06 17:05:44 +01:00
|
|
|
static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
|
2018-03-23 12:40:41 +01:00
|
|
|
BdrvChild *parent, bool poll);
|
2017-12-06 17:05:44 +01:00
|
|
|
static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
|
|
|
|
BdrvChild *parent);
|
2017-12-07 13:03:13 +01:00
|
|
|
|
2016-04-05 13:20:52 +02:00
|
|
|
static void bdrv_co_drain_bh_cb(void *opaque)
|
|
|
|
{
|
|
|
|
BdrvCoDrainData *data = opaque;
|
|
|
|
Coroutine *co = data->co;
|
2016-10-27 12:48:52 +02:00
|
|
|
BlockDriverState *bs = data->bs;
|
2016-04-05 13:20:52 +02:00
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2017-09-23 13:14:09 +02:00
|
|
|
if (data->begin) {
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(bs, data->recursive, data->parent, data->poll);
|
2017-09-23 13:14:09 +02:00
|
|
|
} else {
|
2017-12-06 17:05:44 +01:00
|
|
|
bdrv_do_drained_end(bs, data->recursive, data->parent);
|
2017-09-23 13:14:09 +02:00
|
|
|
}
|
|
|
|
|
2016-04-05 13:20:52 +02:00
|
|
|
data->done = true;
|
2017-02-13 14:52:31 +01:00
|
|
|
aio_co_wake(co);
|
2016-04-05 13:20:52 +02:00
|
|
|
}
|
|
|
|
|
2017-09-23 13:14:09 +02:00
|
|
|
static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
|
2017-12-06 17:05:44 +01:00
|
|
|
bool begin, bool recursive,
|
2018-03-23 12:40:41 +01:00
|
|
|
BdrvChild *parent, bool poll)
|
2016-04-05 13:20:52 +02:00
|
|
|
{
|
|
|
|
BdrvCoDrainData data;
|
|
|
|
|
|
|
|
/* Calling bdrv_drain() from a BH ensures the current coroutine yields and
|
2018-03-22 16:28:33 +01:00
|
|
|
* other coroutines run if they were queued by aio_co_enter(). */
|
2016-04-05 13:20:52 +02:00
|
|
|
|
|
|
|
assert(qemu_in_coroutine());
|
|
|
|
data = (BdrvCoDrainData) {
|
|
|
|
.co = qemu_coroutine_self(),
|
|
|
|
.bs = bs,
|
|
|
|
.done = false,
|
2017-09-23 13:14:09 +02:00
|
|
|
.begin = begin,
|
2017-12-06 17:05:44 +01:00
|
|
|
.recursive = recursive,
|
2017-12-07 13:03:13 +01:00
|
|
|
.parent = parent,
|
2018-03-23 12:40:41 +01:00
|
|
|
.poll = poll,
|
2016-04-05 13:20:52 +02:00
|
|
|
};
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
2016-10-03 18:14:16 +02:00
|
|
|
aio_bh_schedule_oneshot(bdrv_get_aio_context(bs),
|
|
|
|
bdrv_co_drain_bh_cb, &data);
|
2016-04-05 13:20:52 +02:00
|
|
|
|
|
|
|
qemu_coroutine_yield();
|
|
|
|
/* If we are resumed from some other event (such as an aio completion or a
|
|
|
|
* timer callback), it is a bug in the caller that should be fixed. */
|
|
|
|
assert(data.done);
|
|
|
|
}
|
|
|
|
|
2017-12-18 16:05:48 +01:00
|
|
|
void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
|
2018-03-23 12:40:41 +01:00
|
|
|
BdrvChild *parent, bool poll)
|
2016-05-23 16:08:55 +02:00
|
|
|
{
|
2017-12-06 17:05:44 +01:00
|
|
|
BdrvChild *child, *next;
|
|
|
|
|
2016-10-27 12:48:53 +02:00
|
|
|
if (qemu_in_coroutine()) {
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_co_yield_to_drain(bs, true, recursive, parent, poll);
|
2016-10-27 12:48:53 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2017-12-06 11:00:59 +01:00
|
|
|
/* Stop things in parent-to-child order */
|
2017-06-05 14:38:51 +02:00
|
|
|
if (atomic_fetch_inc(&bs->quiesce_counter) == 0) {
|
2016-05-23 16:08:55 +02:00
|
|
|
aio_disable_external(bdrv_get_aio_context(bs));
|
|
|
|
}
|
|
|
|
|
2017-12-07 13:03:13 +01:00
|
|
|
bdrv_parent_drained_begin(bs, parent);
|
2017-12-14 10:40:20 +01:00
|
|
|
bdrv_drain_invoke(bs, true);
|
2018-03-22 14:35:58 +01:00
|
|
|
|
2017-12-06 17:05:44 +01:00
|
|
|
if (recursive) {
|
2017-12-18 16:05:48 +01:00
|
|
|
bs->recursive_quiesce_counter++;
|
2017-12-06 17:05:44 +01:00
|
|
|
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(child->bs, true, child, false);
|
2017-12-06 17:05:44 +01:00
|
|
|
}
|
|
|
|
}
|
2018-03-23 12:40:41 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Wait for drained requests to finish.
|
|
|
|
*
|
|
|
|
* Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
|
|
|
|
* call is needed so things in this AioContext can make progress even
|
|
|
|
* though we don't return to the main AioContext loop - this automatically
|
|
|
|
* includes other nodes in the same AioContext and therefore all child
|
|
|
|
* nodes.
|
|
|
|
*/
|
|
|
|
if (poll) {
|
|
|
|
BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent));
|
|
|
|
}
|
2016-05-23 16:08:55 +02:00
|
|
|
}
|
|
|
|
|
2017-12-07 13:03:13 +01:00
|
|
|
void bdrv_drained_begin(BlockDriverState *bs)
|
|
|
|
{
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(bs, false, NULL, true);
|
2017-12-06 17:05:44 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_subtree_drained_begin(BlockDriverState *bs)
|
|
|
|
{
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(bs, true, NULL, true);
|
2017-12-07 13:03:13 +01:00
|
|
|
}
|
|
|
|
|
2017-12-18 16:05:48 +01:00
|
|
|
void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
|
|
|
|
BdrvChild *parent)
|
2016-05-23 16:08:55 +02:00
|
|
|
{
|
2017-12-06 17:05:44 +01:00
|
|
|
BdrvChild *child, *next;
|
2017-12-13 18:14:18 +01:00
|
|
|
int old_quiesce_counter;
|
|
|
|
|
2017-09-23 13:14:09 +02:00
|
|
|
if (qemu_in_coroutine()) {
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_co_yield_to_drain(bs, false, recursive, parent, false);
|
2017-09-23 13:14:09 +02:00
|
|
|
return;
|
|
|
|
}
|
2016-05-23 16:08:55 +02:00
|
|
|
assert(bs->quiesce_counter > 0);
|
2017-12-13 18:14:18 +01:00
|
|
|
old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter);
|
2016-05-23 16:08:55 +02:00
|
|
|
|
2017-12-06 11:00:59 +01:00
|
|
|
/* Re-enable things in child-to-parent order */
|
2017-12-14 10:40:20 +01:00
|
|
|
bdrv_drain_invoke(bs, false);
|
2017-12-07 13:03:13 +01:00
|
|
|
bdrv_parent_drained_end(bs, parent);
|
2017-12-13 18:14:18 +01:00
|
|
|
if (old_quiesce_counter == 1) {
|
|
|
|
aio_enable_external(bdrv_get_aio_context(bs));
|
|
|
|
}
|
2017-12-06 17:05:44 +01:00
|
|
|
|
|
|
|
if (recursive) {
|
2017-12-18 16:05:48 +01:00
|
|
|
bs->recursive_quiesce_counter--;
|
2017-12-06 17:05:44 +01:00
|
|
|
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
|
|
|
|
bdrv_do_drained_end(child->bs, true, child);
|
|
|
|
}
|
|
|
|
}
|
2016-05-23 16:08:55 +02:00
|
|
|
}
|
|
|
|
|
2017-12-07 13:03:13 +01:00
|
|
|
void bdrv_drained_end(BlockDriverState *bs)
|
|
|
|
{
|
2017-12-06 17:05:44 +01:00
|
|
|
bdrv_do_drained_end(bs, false, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_subtree_drained_end(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
bdrv_do_drained_end(bs, true, NULL);
|
2017-12-07 13:03:13 +01:00
|
|
|
}
|
|
|
|
|
2017-12-18 16:05:48 +01:00
|
|
|
void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < new_parent->recursive_quiesce_counter; i++) {
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(child->bs, true, child, true);
|
2017-12-18 16:05:48 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < old_parent->recursive_quiesce_counter; i++) {
|
|
|
|
bdrv_do_drained_end(child->bs, true, child);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/*
|
2015-11-09 11:16:53 +01:00
|
|
|
* Wait for pending requests to complete on a single BlockDriverState subtree,
|
|
|
|
* and suspend block driver's internal I/O until next request arrives.
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
|
|
|
* Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
|
|
|
|
* AioContext.
|
|
|
|
*/
|
2016-04-07 18:33:32 +02:00
|
|
|
void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2016-05-23 16:08:55 +02:00
|
|
|
assert(qemu_in_coroutine());
|
|
|
|
bdrv_drained_begin(bs);
|
|
|
|
bdrv_drained_end(bs);
|
2016-04-07 18:33:32 +02:00
|
|
|
}
|
2015-06-10 13:38:17 +02:00
|
|
|
|
2016-04-07 18:33:32 +02:00
|
|
|
void bdrv_drain(BlockDriverState *bs)
|
|
|
|
{
|
2016-05-23 16:08:55 +02:00
|
|
|
bdrv_drained_begin(bs);
|
|
|
|
bdrv_drained_end(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-12-14 11:25:16 +01:00
|
|
|
static void bdrv_drain_assert_idle(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
BdrvChild *child, *next;
|
|
|
|
|
|
|
|
assert(atomic_read(&bs->in_flight) == 0);
|
|
|
|
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
|
|
|
|
bdrv_drain_assert_idle(child->bs);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/*
|
|
|
|
* Wait for pending requests to complete across all BlockDriverStates
|
|
|
|
*
|
|
|
|
* This function does not flush data to disk, use bdrv_flush_all() for that
|
|
|
|
* after calling this function.
|
2016-10-28 09:08:02 +02:00
|
|
|
*
|
|
|
|
* This pauses all block jobs and disables external clients. It must
|
|
|
|
* be paired with bdrv_drain_all_end().
|
|
|
|
*
|
|
|
|
* NOTE: no new block jobs or BlockDriverStates can be created between
|
|
|
|
* the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2016-10-28 09:08:02 +02:00
|
|
|
void bdrv_drain_all_begin(void)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2016-03-22 18:58:50 +01:00
|
|
|
BlockDriverState *bs;
|
2016-05-20 18:49:07 +02:00
|
|
|
BdrvNextIterator it;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-12-15 09:33:21 +01:00
|
|
|
/* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread
|
|
|
|
* or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on
|
|
|
|
* nodes in several different AioContexts, so make sure we're in the main
|
|
|
|
* context. */
|
|
|
|
assert(qemu_get_current_aio_context() == qemu_get_aio_context());
|
|
|
|
|
2016-05-20 18:49:07 +02:00
|
|
|
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
|
2015-04-28 15:27:52 +02:00
|
|
|
AioContext *aio_context = bdrv_get_aio_context(bs);
|
|
|
|
|
|
|
|
aio_context_acquire(aio_context);
|
2018-03-23 12:40:41 +01:00
|
|
|
bdrv_do_drained_begin(bs, true, NULL, true);
|
2015-04-28 15:27:52 +02:00
|
|
|
aio_context_release(aio_context);
|
|
|
|
}
|
|
|
|
|
2017-12-14 11:25:16 +01:00
|
|
|
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
|
|
|
|
bdrv_drain_assert_idle(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2016-10-28 09:08:02 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_drain_all_end(void)
|
|
|
|
{
|
|
|
|
BlockDriverState *bs;
|
|
|
|
BdrvNextIterator it;
|
|
|
|
|
2016-05-20 18:49:07 +02:00
|
|
|
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
|
2015-04-28 15:27:52 +02:00
|
|
|
AioContext *aio_context = bdrv_get_aio_context(bs);
|
|
|
|
|
|
|
|
aio_context_acquire(aio_context);
|
2017-12-14 10:27:23 +01:00
|
|
|
bdrv_do_drained_end(bs, true, NULL);
|
2015-04-28 15:27:52 +02:00
|
|
|
aio_context_release(aio_context);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-10-28 09:08:02 +02:00
|
|
|
void bdrv_drain_all(void)
|
|
|
|
{
|
|
|
|
bdrv_drain_all_begin();
|
|
|
|
bdrv_drain_all_end();
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/**
|
|
|
|
* Remove an active request from the tracked requests list
|
|
|
|
*
|
|
|
|
* This function should be called when a tracked request is completing.
|
|
|
|
*/
|
|
|
|
static void tracked_request_end(BdrvTrackedRequest *req)
|
|
|
|
{
|
|
|
|
if (req->serialising) {
|
2017-06-05 14:38:53 +02:00
|
|
|
atomic_dec(&req->bs->serialising_in_flight);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_lock(&req->bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
QLIST_REMOVE(req, list);
|
|
|
|
qemu_co_queue_restart_all(&req->wait_queue);
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_unlock(&req->bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Add an active request to the tracked requests list
|
|
|
|
*/
|
|
|
|
static void tracked_request_begin(BdrvTrackedRequest *req,
|
|
|
|
BlockDriverState *bs,
|
|
|
|
int64_t offset,
|
2015-11-09 11:16:46 +01:00
|
|
|
unsigned int bytes,
|
|
|
|
enum BdrvTrackedRequestType type)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
*req = (BdrvTrackedRequest){
|
|
|
|
.bs = bs,
|
|
|
|
.offset = offset,
|
|
|
|
.bytes = bytes,
|
2015-11-09 11:16:46 +01:00
|
|
|
.type = type,
|
2015-04-28 15:27:52 +02:00
|
|
|
.co = qemu_coroutine_self(),
|
|
|
|
.serialising = false,
|
|
|
|
.overlap_offset = offset,
|
|
|
|
.overlap_bytes = bytes,
|
|
|
|
};
|
|
|
|
|
|
|
|
qemu_co_queue_init(&req->wait_queue);
|
|
|
|
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_lock(&bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_unlock(&bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
|
|
|
|
{
|
|
|
|
int64_t overlap_offset = req->offset & ~(align - 1);
|
|
|
|
unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
|
|
|
|
- overlap_offset;
|
|
|
|
|
|
|
|
if (!req->serialising) {
|
2017-06-05 14:38:53 +02:00
|
|
|
atomic_inc(&req->bs->serialising_in_flight);
|
2015-04-28 15:27:52 +02:00
|
|
|
req->serialising = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
|
|
|
|
req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
|
|
|
|
}
|
|
|
|
|
2016-06-02 11:41:52 +02:00
|
|
|
/**
|
|
|
|
* Round a region to cluster boundaries
|
|
|
|
*/
|
|
|
|
void bdrv_round_to_clusters(BlockDriverState *bs,
|
2017-10-12 05:46:59 +02:00
|
|
|
int64_t offset, int64_t bytes,
|
2016-06-02 11:41:52 +02:00
|
|
|
int64_t *cluster_offset,
|
2017-10-12 05:46:59 +02:00
|
|
|
int64_t *cluster_bytes)
|
2016-06-02 11:41:52 +02:00
|
|
|
{
|
|
|
|
BlockDriverInfo bdi;
|
|
|
|
|
|
|
|
if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
|
|
|
|
*cluster_offset = offset;
|
|
|
|
*cluster_bytes = bytes;
|
|
|
|
} else {
|
|
|
|
int64_t c = bdi.cluster_size;
|
|
|
|
*cluster_offset = QEMU_ALIGN_DOWN(offset, c);
|
|
|
|
*cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
static int bdrv_get_cluster_size(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
BlockDriverInfo bdi;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = bdrv_get_info(bs, &bdi);
|
|
|
|
if (ret < 0 || bdi.cluster_size == 0) {
|
2016-06-24 00:37:24 +02:00
|
|
|
return bs->bl.request_alignment;
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
|
|
|
return bdi.cluster_size;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool tracked_request_overlaps(BdrvTrackedRequest *req,
|
|
|
|
int64_t offset, unsigned int bytes)
|
|
|
|
{
|
|
|
|
/* aaaa bbbb */
|
|
|
|
if (offset >= req->overlap_offset + req->overlap_bytes) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
/* bbbb aaaa */
|
|
|
|
if (req->overlap_offset >= offset + bytes) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
void bdrv_inc_in_flight(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
atomic_inc(&bs->in_flight);
|
|
|
|
}
|
|
|
|
|
2016-10-27 12:49:05 +02:00
|
|
|
void bdrv_wakeup(BlockDriverState *bs)
|
|
|
|
{
|
2018-02-16 17:50:12 +01:00
|
|
|
aio_wait_kick(bdrv_get_aio_wait(bs));
|
2016-10-27 12:49:05 +02:00
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
void bdrv_dec_in_flight(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
atomic_dec(&bs->in_flight);
|
2016-10-27 12:49:05 +02:00
|
|
|
bdrv_wakeup(bs);
|
2016-10-27 12:48:52 +02:00
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
|
|
|
|
{
|
|
|
|
BlockDriverState *bs = self->bs;
|
|
|
|
BdrvTrackedRequest *req;
|
|
|
|
bool retry;
|
|
|
|
bool waited = false;
|
|
|
|
|
2017-06-05 14:38:53 +02:00
|
|
|
if (!atomic_read(&bs->serialising_in_flight)) {
|
2015-04-28 15:27:52 +02:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
do {
|
|
|
|
retry = false;
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_lock(&bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
QLIST_FOREACH(req, &bs->tracked_requests, list) {
|
|
|
|
if (req == self || (!req->serialising && !self->serialising)) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (tracked_request_overlaps(req, self->overlap_offset,
|
|
|
|
self->overlap_bytes))
|
|
|
|
{
|
|
|
|
/* Hitting this means there was a reentrant request, for
|
|
|
|
* example, a block driver issuing nested requests. This must
|
|
|
|
* never happen since it means deadlock.
|
|
|
|
*/
|
|
|
|
assert(qemu_coroutine_self() != req->co);
|
|
|
|
|
|
|
|
/* If the request is already (indirectly) waiting for us, or
|
|
|
|
* will wait for us as soon as it wakes up, then just go on
|
|
|
|
* (instead of producing a deadlock in the former case). */
|
|
|
|
if (!req->waiting_for) {
|
|
|
|
self->waiting_for = req;
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
self->waiting_for = NULL;
|
|
|
|
retry = true;
|
|
|
|
waited = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_unlock(&bs->reqs_lock);
|
2015-04-28 15:27:52 +02:00
|
|
|
} while (retry);
|
|
|
|
|
|
|
|
return waited;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
|
|
|
|
size_t size)
|
|
|
|
{
|
|
|
|
if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
|
|
|
|
return -EIO;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!bdrv_is_inserted(bs)) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (offset < 0) {
|
|
|
|
return -EIO;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
typedef struct RwCo {
|
2016-05-30 15:00:40 +02:00
|
|
|
BdrvChild *child;
|
2015-04-28 15:27:52 +02:00
|
|
|
int64_t offset;
|
|
|
|
QEMUIOVector *qiov;
|
|
|
|
bool is_write;
|
|
|
|
int ret;
|
|
|
|
BdrvRequestFlags flags;
|
|
|
|
} RwCo;
|
|
|
|
|
|
|
|
static void coroutine_fn bdrv_rw_co_entry(void *opaque)
|
|
|
|
{
|
|
|
|
RwCo *rwco = opaque;
|
|
|
|
|
|
|
|
if (!rwco->is_write) {
|
2016-06-20 21:31:46 +02:00
|
|
|
rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset,
|
2016-04-25 14:57:23 +02:00
|
|
|
rwco->qiov->size, rwco->qiov,
|
|
|
|
rwco->flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
2016-06-20 21:31:46 +02:00
|
|
|
rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset,
|
2016-04-25 14:57:23 +02:00
|
|
|
rwco->qiov->size, rwco->qiov,
|
|
|
|
rwco->flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Process a vectored synchronous request using coroutines
|
|
|
|
*/
|
2016-05-30 15:00:40 +02:00
|
|
|
static int bdrv_prwv_co(BdrvChild *child, int64_t offset,
|
2015-04-28 15:27:52 +02:00
|
|
|
QEMUIOVector *qiov, bool is_write,
|
|
|
|
BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
Coroutine *co;
|
|
|
|
RwCo rwco = {
|
2016-05-30 15:00:40 +02:00
|
|
|
.child = child,
|
2015-04-28 15:27:52 +02:00
|
|
|
.offset = offset,
|
|
|
|
.qiov = qiov,
|
|
|
|
.is_write = is_write,
|
|
|
|
.ret = NOT_DONE,
|
|
|
|
.flags = flags,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (qemu_in_coroutine()) {
|
|
|
|
/* Fast-path if already in coroutine context */
|
|
|
|
bdrv_rw_co_entry(&rwco);
|
|
|
|
} else {
|
coroutine: move entry argument to qemu_coroutine_create
In practice the entry argument is always known at creation time, and
it is confusing that sometimes qemu_coroutine_enter is used with a
non-NULL argument to re-enter a coroutine (this happens in
block/sheepdog.c and tests/test-coroutine.c). So pass the opaque value
at creation time, for consistency with e.g. aio_bh_new.
Mostly done with the following semantic patch:
@ entry1 @
expression entry, arg, co;
@@
- co = qemu_coroutine_create(entry);
+ co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry2 @
expression entry, arg;
identifier co;
@@
- Coroutine *co = qemu_coroutine_create(entry);
+ Coroutine *co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry3 @
expression entry, arg;
@@
- qemu_coroutine_enter(qemu_coroutine_create(entry), arg);
+ qemu_coroutine_enter(qemu_coroutine_create(entry, arg));
@ reentry @
expression co;
@@
- qemu_coroutine_enter(co, NULL);
+ qemu_coroutine_enter(co);
except for the aforementioned few places where the semantic patch
stumbled (as expected) and for test_co_queue, which would otherwise
produce an uninitialized variable warning.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-07-04 19:10:01 +02:00
|
|
|
co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco);
|
block: Use bdrv_coroutine_enter to start I/O coroutines
BDRV_POLL_WHILE waits for the started I/O by releasing bs's ctx then polling
the main context, which relies on the yielded coroutine continuing on bs->ctx
before notifying qemu_aio_context with bdrv_wakeup().
Thus, using qemu_coroutine_enter to start I/O is wrong because if the coroutine
is entered from main loop, co->ctx will be qemu_aio_context, as a result of the
"release, poll, acquire" loop of BDRV_POLL_WHILE, race conditions happen when
both main thread and the iothread access the same BDS:
main loop iothread
-----------------------------------------------------------------------
blockdev_snapshot
aio_context_acquire(bs->ctx)
virtio_scsi_data_plane_handle_cmd
bdrv_drained_begin(bs->ctx)
bdrv_flush(bs)
bdrv_co_flush(bs) aio_context_acquire(bs->ctx).enter
...
qemu_coroutine_yield(co)
BDRV_POLL_WHILE()
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx).return
...
aio_co_wake(co)
aio_poll(qemu_aio_context) ...
co_schedule_bh_cb() ...
qemu_coroutine_enter(co) ...
/* (A) bdrv_co_flush(bs) /* (B) I/O on bs */
continues... */
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx)
Note that in above case, bdrv_drained_begin() doesn't do the "release,
poll, acquire" in BDRV_POLL_WHILE, because bs->in_flight == 0.
Fix this by using bdrv_coroutine_enter and enter coroutine in the right
context.
iotests 109 output is updated because the coroutine reenter flow during
mirror job complete is different (now through co_queue_wakeup, instead
of the unconditional qemu_coroutine_switch before), making the end job
len different.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
2017-04-10 14:20:17 +02:00
|
|
|
bdrv_coroutine_enter(child->bs, co);
|
2016-10-27 12:48:55 +02:00
|
|
|
BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
return rwco.ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Process a synchronous request using coroutines
|
|
|
|
*/
|
2016-05-30 15:00:40 +02:00
|
|
|
static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf,
|
2015-04-28 15:27:52 +02:00
|
|
|
int nb_sectors, bool is_write, BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = (void *)buf,
|
|
|
|
.iov_len = nb_sectors * BDRV_SECTOR_SIZE,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-05-30 15:00:40 +02:00
|
|
|
return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS,
|
2015-04-28 15:27:52 +02:00
|
|
|
&qiov, is_write, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* return < 0 if error. See bdrv_write() for the return codes */
|
2016-05-30 16:48:35 +02:00
|
|
|
int bdrv_read(BdrvChild *child, int64_t sector_num,
|
2015-04-28 15:27:52 +02:00
|
|
|
uint8_t *buf, int nb_sectors)
|
|
|
|
{
|
2016-05-30 15:00:40 +02:00
|
|
|
return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Return < 0 if error. Important errors are:
|
|
|
|
-EIO generic I/O error (may happen for all errors)
|
|
|
|
-ENOMEDIUM No media inserted.
|
|
|
|
-EINVAL Invalid sector number or nb_sectors
|
|
|
|
-EACCES Trying to write a read-only device
|
|
|
|
*/
|
2016-05-31 14:42:08 +02:00
|
|
|
int bdrv_write(BdrvChild *child, int64_t sector_num,
|
2015-04-28 15:27:52 +02:00
|
|
|
const uint8_t *buf, int nb_sectors)
|
|
|
|
{
|
2016-05-30 15:00:40 +02:00
|
|
|
return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-06-16 15:13:15 +02:00
|
|
|
int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
|
2017-06-09 12:18:08 +02:00
|
|
|
int bytes, BdrvRequestFlags flags)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2016-06-01 23:10:04 +02:00
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = NULL,
|
2017-06-09 12:18:08 +02:00
|
|
|
.iov_len = bytes,
|
2016-06-01 23:10:04 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-05-30 15:00:40 +02:00
|
|
|
return bdrv_prwv_co(child, offset, &qiov, true,
|
2016-06-01 23:10:04 +02:00
|
|
|
BDRV_REQ_ZERO_WRITE | flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2016-06-01 23:10:04 +02:00
|
|
|
* Completely zero out a block device with the help of bdrv_pwrite_zeroes.
|
2015-04-28 15:27:52 +02:00
|
|
|
* The operation is sped up by checking the block status and only writing
|
|
|
|
* zeroes to the device if they currently do not return zeroes. Optional
|
2016-06-01 23:10:04 +02:00
|
|
|
* flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
|
block: Honor BDRV_REQ_FUA during write_zeroes
The block layer has a couple of cases where it can lose
Force Unit Access semantics when writing a large block of
zeroes, such that the request returns before the zeroes
have been guaranteed to land on underlying media.
SCSI does not support FUA during WRITESAME(10/16); FUA is only
supported if it falls back to WRITE(10/16). But where the
underlying device is new enough to not need a fallback, it
means that any upper layer request with FUA semantics was
silently ignoring BDRV_REQ_FUA.
Conversely, NBD has situations where it can support FUA but not
ZERO_WRITE; when that happens, the generic block layer fallback
to bdrv_driver_pwritev() (or the older bdrv_co_writev() in qemu
2.6) was losing the FUA flag.
The problem of losing flags unrelated to ZERO_WRITE has been
latent in bdrv_co_do_write_zeroes() since commit aa7bfbff, but
back then, it did not matter because there was no FUA flag. It
became observable when commit 93f5e6d8 paved the way for flags
that can impact correctness, when we should have been using
bdrv_co_writev_flags() with modified flags. Compare to commit
9eeb6dd, which got flag manipulation right in
bdrv_co_do_zero_pwritev().
Symptoms: I tested with qemu-io with default writethrough cache
(which is supposed to use FUA semantics on every write), and
targetted an NBD client connected to a server that intentionally
did not advertise NBD_FLAG_SEND_FUA. When doing 'write 0 512',
the NBD client sent two operations (NBD_CMD_WRITE then
NBD_CMD_FLUSH) to get the fallback FUA semantics; but when doing
'write -z 0 512', the NBD client sent only NBD_CMD_WRITE.
The fix is do to a cleanup bdrv_co_flush() at the end of the
operation if any step in the middle relied on a BDS that does
not natively support FUA for that step (note that we don't
need to flush after every operation, if the operation is broken
into chunks based on bounce-buffer sizing). Each BDS gains a
new flag .supported_zero_flags, which parallels the use of
.supported_write_flags but only when accessing a zero write
operation (the flags MUST be different, because of SCSI having
different semantics based on WRITE vs. WRITESAME; and also
because BDRV_REQ_MAY_UNMAP only makes sense on zero writes).
Also fix some documentation to describe -ENOTSUP semantics,
particularly since iscsi depends on those semantics.
Down the road, we may want to add a driver where its
.bdrv_co_pwritev() honors all three of BDRV_REQ_FUA,
BDRV_REQ_ZERO_WRITE, and BDRV_REQ_MAY_UNMAP, and advertise
this via bs->supported_write_flags for blocks opened by that
driver; such a driver should NOT supply .bdrv_co_write_zeroes
nor .supported_zero_flags. But none of the drivers touched
in this patch want to do that (the act of writing zeroes is
different enough from normal writes to deserve a second
callback).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-05-04 00:39:07 +02:00
|
|
|
* BDRV_REQ_FUA).
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
|
|
|
* Returns < 0 on error, 0 on success. For error codes see bdrv_write().
|
|
|
|
*/
|
2016-06-16 15:13:15 +02:00
|
|
|
int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
int ret;
|
|
|
|
int64_t target_size, bytes, offset = 0;
|
2016-06-16 15:13:15 +02:00
|
|
|
BlockDriverState *bs = child->bs;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-10-12 05:47:01 +02:00
|
|
|
target_size = bdrv_getlength(bs);
|
|
|
|
if (target_size < 0) {
|
|
|
|
return target_size;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
for (;;) {
|
2017-10-12 05:47:01 +02:00
|
|
|
bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES);
|
|
|
|
if (bytes <= 0) {
|
2015-04-28 15:27:52 +02:00
|
|
|
return 0;
|
|
|
|
}
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
2017-10-12 05:47:01 +02:00
|
|
|
error_report("error getting block status at offset %" PRId64 ": %s",
|
|
|
|
offset, strerror(-ret));
|
2015-04-28 15:27:52 +02:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
if (ret & BDRV_BLOCK_ZERO) {
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
offset += bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
continue;
|
|
|
|
}
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
ret = bdrv_pwrite_zeroes(child, offset, bytes, flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
2017-10-12 05:47:01 +02:00
|
|
|
error_report("error writing zeroes at offset %" PRId64 ": %s",
|
|
|
|
offset, strerror(-ret));
|
2015-04-28 15:27:52 +02:00
|
|
|
return ret;
|
|
|
|
}
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
offset += bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-06-20 18:24:02 +02:00
|
|
|
int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
|
2016-06-09 16:36:00 +02:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2016-05-30 15:00:40 +02:00
|
|
|
ret = bdrv_prwv_co(child, offset, qiov, false, 0);
|
2016-06-09 16:36:00 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
return qiov->size;
|
|
|
|
}
|
|
|
|
|
2016-06-20 18:24:02 +02:00
|
|
|
int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = (void *)buf,
|
|
|
|
.iov_len = bytes,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (bytes < 0) {
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-06-20 18:24:02 +02:00
|
|
|
return bdrv_preadv(child, offset, &qiov);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-06-20 20:09:15 +02:00
|
|
|
int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2016-05-30 15:00:40 +02:00
|
|
|
ret = bdrv_prwv_co(child, offset, qiov, true, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
return qiov->size;
|
|
|
|
}
|
|
|
|
|
2016-06-20 20:09:15 +02:00
|
|
|
int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = (void *) buf,
|
|
|
|
.iov_len = bytes,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (bytes < 0) {
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-06-20 20:09:15 +02:00
|
|
|
return bdrv_pwritev(child, offset, &qiov);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Writes to the file and ensures that no writes are reordered across this
|
|
|
|
* request (acts as a barrier)
|
|
|
|
*
|
|
|
|
* Returns 0 on success, -errno in error cases.
|
|
|
|
*/
|
2016-06-20 20:09:15 +02:00
|
|
|
int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
|
|
|
|
const void *buf, int count)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2016-06-20 20:09:15 +02:00
|
|
|
ret = bdrv_pwrite(child, offset, buf, count);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-20 20:09:15 +02:00
|
|
|
ret = bdrv_flush(child->bs);
|
2016-03-04 14:16:51 +01:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-04-25 14:13:12 +02:00
|
|
|
typedef struct CoroutineIOCompletion {
|
|
|
|
Coroutine *coroutine;
|
|
|
|
int ret;
|
|
|
|
} CoroutineIOCompletion;
|
|
|
|
|
|
|
|
static void bdrv_co_io_em_complete(void *opaque, int ret)
|
|
|
|
{
|
|
|
|
CoroutineIOCompletion *co = opaque;
|
|
|
|
|
|
|
|
co->ret = ret;
|
2017-02-13 14:52:32 +01:00
|
|
|
aio_co_wake(co->coroutine);
|
2016-04-25 14:13:12 +02:00
|
|
|
}
|
|
|
|
|
2016-04-25 11:46:41 +02:00
|
|
|
static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
|
|
|
|
uint64_t offset, uint64_t bytes,
|
|
|
|
QEMUIOVector *qiov, int flags)
|
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
2016-04-25 11:25:18 +02:00
|
|
|
int64_t sector_num;
|
|
|
|
unsigned int nb_sectors;
|
|
|
|
|
2016-06-13 20:56:35 +02:00
|
|
|
assert(!(flags & ~BDRV_REQ_MASK));
|
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2016-04-25 11:25:18 +02:00
|
|
|
if (drv->bdrv_co_preadv) {
|
|
|
|
return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
|
|
|
|
}
|
|
|
|
|
2018-04-24 21:25:06 +02:00
|
|
|
if (drv->bdrv_aio_preadv) {
|
2016-04-25 14:13:12 +02:00
|
|
|
BlockAIOCB *acb;
|
|
|
|
CoroutineIOCompletion co = {
|
|
|
|
.coroutine = qemu_coroutine_self(),
|
|
|
|
};
|
|
|
|
|
2018-04-24 21:25:06 +02:00
|
|
|
acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags,
|
|
|
|
bdrv_co_io_em_complete, &co);
|
2016-04-25 14:13:12 +02:00
|
|
|
if (acb == NULL) {
|
|
|
|
return -EIO;
|
|
|
|
} else {
|
|
|
|
qemu_coroutine_yield();
|
|
|
|
return co.ret;
|
|
|
|
}
|
|
|
|
}
|
2018-04-24 21:25:06 +02:00
|
|
|
|
|
|
|
sector_num = offset >> BDRV_SECTOR_BITS;
|
|
|
|
nb_sectors = bytes >> BDRV_SECTOR_BITS;
|
|
|
|
|
|
|
|
assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
|
|
|
|
assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
|
|
|
|
assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
|
|
|
|
assert(drv->bdrv_co_readv);
|
|
|
|
|
|
|
|
return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
|
2016-04-25 11:46:41 +02:00
|
|
|
}
|
|
|
|
|
2016-04-25 12:13:39 +02:00
|
|
|
static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
|
|
|
|
uint64_t offset, uint64_t bytes,
|
|
|
|
QEMUIOVector *qiov, int flags)
|
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
2016-04-25 11:25:18 +02:00
|
|
|
int64_t sector_num;
|
|
|
|
unsigned int nb_sectors;
|
2016-04-25 12:13:39 +02:00
|
|
|
int ret;
|
|
|
|
|
2016-06-13 20:56:35 +02:00
|
|
|
assert(!(flags & ~BDRV_REQ_MASK));
|
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2016-04-25 11:25:18 +02:00
|
|
|
if (drv->bdrv_co_pwritev) {
|
2016-06-07 15:51:28 +02:00
|
|
|
ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
|
|
|
|
flags & bs->supported_write_flags);
|
|
|
|
flags &= ~bs->supported_write_flags;
|
2016-04-25 11:25:18 +02:00
|
|
|
goto emulate_flags;
|
|
|
|
}
|
|
|
|
|
2018-04-24 21:25:06 +02:00
|
|
|
if (drv->bdrv_aio_pwritev) {
|
2016-04-25 14:13:12 +02:00
|
|
|
BlockAIOCB *acb;
|
|
|
|
CoroutineIOCompletion co = {
|
|
|
|
.coroutine = qemu_coroutine_self(),
|
|
|
|
};
|
|
|
|
|
2018-04-24 21:25:06 +02:00
|
|
|
acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov,
|
|
|
|
flags & bs->supported_write_flags,
|
|
|
|
bdrv_co_io_em_complete, &co);
|
|
|
|
flags &= ~bs->supported_write_flags;
|
2016-04-25 14:13:12 +02:00
|
|
|
if (acb == NULL) {
|
2016-04-25 11:25:18 +02:00
|
|
|
ret = -EIO;
|
2016-04-25 14:13:12 +02:00
|
|
|
} else {
|
|
|
|
qemu_coroutine_yield();
|
2016-04-25 11:25:18 +02:00
|
|
|
ret = co.ret;
|
2016-04-25 14:13:12 +02:00
|
|
|
}
|
2018-04-24 21:25:06 +02:00
|
|
|
goto emulate_flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
sector_num = offset >> BDRV_SECTOR_BITS;
|
|
|
|
nb_sectors = bytes >> BDRV_SECTOR_BITS;
|
|
|
|
|
|
|
|
assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
|
|
|
|
assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
|
|
|
|
assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
|
|
|
|
|
2018-04-25 00:01:57 +02:00
|
|
|
assert(drv->bdrv_co_writev);
|
|
|
|
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov,
|
|
|
|
flags & bs->supported_write_flags);
|
|
|
|
flags &= ~bs->supported_write_flags;
|
2016-04-25 12:13:39 +02:00
|
|
|
|
2016-04-25 11:25:18 +02:00
|
|
|
emulate_flags:
|
2016-05-04 00:39:06 +02:00
|
|
|
if (ret == 0 && (flags & BDRV_REQ_FUA)) {
|
2016-04-25 12:13:39 +02:00
|
|
|
ret = bdrv_co_flush(bs);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-07-22 10:17:42 +02:00
|
|
|
static int coroutine_fn
|
|
|
|
bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
|
|
|
|
uint64_t bytes, QEMUIOVector *qiov)
|
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2016-07-22 10:17:42 +02:00
|
|
|
if (!drv->bdrv_co_pwritev_compressed) {
|
|
|
|
return -ENOTSUP;
|
|
|
|
}
|
|
|
|
|
|
|
|
return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov);
|
|
|
|
}
|
|
|
|
|
2017-02-09 15:58:43 +01:00
|
|
|
static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child,
|
2016-06-02 11:41:52 +02:00
|
|
|
int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-02-09 15:58:43 +01:00
|
|
|
BlockDriverState *bs = child->bs;
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/* Perform I/O through a temporary buffer so that users who scribble over
|
|
|
|
* their read buffer while the operation is in progress do not end up
|
|
|
|
* modifying the image file. This is critical for zero-copy guest I/O
|
|
|
|
* where anything might happen inside guest memory.
|
|
|
|
*/
|
|
|
|
void *bounce_buffer;
|
|
|
|
|
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
struct iovec iov;
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
QEMUIOVector local_qiov;
|
2016-06-02 11:41:52 +02:00
|
|
|
int64_t cluster_offset;
|
2017-10-12 05:46:59 +02:00
|
|
|
int64_t cluster_bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
size_t skip_bytes;
|
|
|
|
int ret;
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
|
|
|
|
BDRV_REQUEST_MAX_BYTES);
|
|
|
|
unsigned int progress = 0;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2017-04-07 12:29:05 +02:00
|
|
|
/* FIXME We cannot require callers to have write permissions when all they
|
|
|
|
* are doing is a read request. If we did things right, write permissions
|
|
|
|
* would be obtained anyway, but internally by the copy-on-read code. As
|
2017-09-28 21:13:25 +02:00
|
|
|
* long as it is implemented here rather than in a separate filter driver,
|
2017-04-07 12:29:05 +02:00
|
|
|
* the copy-on-read code doesn't have its own BdrvChild, however, for which
|
|
|
|
* it could request permissions. Therefore we have to bypass the permission
|
|
|
|
* system for the moment. */
|
|
|
|
// assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
|
2017-02-09 16:49:53 +01:00
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/* Cover entire cluster so no additional backing file I/O is required when
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
* allocating cluster in the image file. Note that this value may exceed
|
|
|
|
* BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
|
|
|
|
* is one reason we loop rather than doing it all at once.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2016-06-02 11:41:52 +02:00
|
|
|
bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
skip_bytes = offset - cluster_offset;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-06-02 11:41:52 +02:00
|
|
|
trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
|
|
|
|
cluster_offset, cluster_bytes);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
bounce_buffer = qemu_try_blockalign(bs,
|
|
|
|
MIN(MIN(max_transfer, cluster_bytes),
|
|
|
|
MAX_BOUNCE_BUFFER));
|
2015-04-28 15:27:52 +02:00
|
|
|
if (bounce_buffer == NULL) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
while (cluster_bytes) {
|
|
|
|
int64_t pnum;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
ret = bdrv_is_allocated(bs, cluster_offset,
|
|
|
|
MIN(cluster_bytes, max_transfer), &pnum);
|
|
|
|
if (ret < 0) {
|
|
|
|
/* Safe to treat errors in querying allocation as if
|
|
|
|
* unallocated; we'll probably fail again soon on the
|
|
|
|
* read, but at least that will set a decent errno.
|
|
|
|
*/
|
|
|
|
pnum = MIN(cluster_bytes, max_transfer);
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
assert(skip_bytes < pnum);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
if (ret <= 0) {
|
|
|
|
/* Must copy-on-read; use the bounce buffer */
|
|
|
|
iov.iov_base = bounce_buffer;
|
|
|
|
iov.iov_len = pnum = MIN(pnum, MAX_BOUNCE_BUFFER);
|
|
|
|
qemu_iovec_init_external(&local_qiov, &iov, 1);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
ret = bdrv_driver_preadv(bs, cluster_offset, pnum,
|
|
|
|
&local_qiov, 0);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
bdrv_debug_event(bs, BLKDBG_COR_WRITE);
|
|
|
|
if (drv->bdrv_co_pwrite_zeroes &&
|
|
|
|
buffer_is_zero(bounce_buffer, pnum)) {
|
|
|
|
/* FIXME: Should we (perhaps conditionally) be setting
|
|
|
|
* BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
|
|
|
|
* that still correctly reads as zero? */
|
2018-04-21 15:29:24 +02:00
|
|
|
ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum,
|
|
|
|
BDRV_REQ_WRITE_UNCHANGED);
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
} else {
|
|
|
|
/* This does not change the data on the disk, it is not
|
|
|
|
* necessary to flush even in cache=writethrough mode.
|
|
|
|
*/
|
|
|
|
ret = bdrv_driver_pwritev(bs, cluster_offset, pnum,
|
2018-04-21 15:29:24 +02:00
|
|
|
&local_qiov,
|
|
|
|
BDRV_REQ_WRITE_UNCHANGED);
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (ret < 0) {
|
|
|
|
/* It might be okay to ignore write errors for guest
|
|
|
|
* requests. If this is a deliberate copy-on-read
|
|
|
|
* then we don't want to ignore the error. Simply
|
|
|
|
* report it in all cases.
|
|
|
|
*/
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
qemu_iovec_from_buf(qiov, progress, bounce_buffer + skip_bytes,
|
|
|
|
pnum - skip_bytes);
|
|
|
|
} else {
|
|
|
|
/* Read directly into the destination */
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov);
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, progress, pnum - skip_bytes);
|
|
|
|
ret = bdrv_driver_preadv(bs, offset + progress, local_qiov.size,
|
|
|
|
&local_qiov, 0);
|
|
|
|
qemu_iovec_destroy(&local_qiov);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
cluster_offset += pnum;
|
|
|
|
cluster_bytes -= pnum;
|
|
|
|
progress += pnum - skip_bytes;
|
|
|
|
skip_bytes = 0;
|
|
|
|
}
|
|
|
|
ret = 0;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
err:
|
|
|
|
qemu_vfree(bounce_buffer);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Forwards an already correctly aligned request to the BlockDriver. This
|
2016-07-15 20:31:59 +02:00
|
|
|
* handles copy on read, zeroing after EOF, and fragmentation of large
|
|
|
|
* reads; any other features must be implemented by the caller.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2017-02-09 15:58:43 +01:00
|
|
|
static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
|
|
|
|
int64_t align, QEMUIOVector *qiov, int flags)
|
|
|
|
{
|
2017-02-09 15:58:43 +01:00
|
|
|
BlockDriverState *bs = child->bs;
|
2016-06-01 17:13:47 +02:00
|
|
|
int64_t total_bytes, max_bytes;
|
2016-07-15 20:31:59 +02:00
|
|
|
int ret = 0;
|
|
|
|
uint64_t bytes_remaining = bytes;
|
|
|
|
int max_transfer;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-06-03 16:17:28 +02:00
|
|
|
assert(is_power_of_2(align));
|
|
|
|
assert((offset & (align - 1)) == 0);
|
|
|
|
assert((bytes & (align - 1)) == 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
assert(!qiov || bytes == qiov->size);
|
2016-03-21 15:11:42 +01:00
|
|
|
assert((bs->open_flags & BDRV_O_NO_IO) == 0);
|
2016-07-15 20:31:59 +02:00
|
|
|
max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
|
|
|
|
align);
|
2016-06-24 00:37:06 +02:00
|
|
|
|
|
|
|
/* TODO: We would need a per-BDS .supported_read_flags and
|
|
|
|
* potential fallback support, if we ever implement any read flags
|
|
|
|
* to pass through to drivers. For now, there aren't any
|
|
|
|
* passthrough flags. */
|
|
|
|
assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ)));
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
/* Handle Copy on Read and associated serialisation */
|
|
|
|
if (flags & BDRV_REQ_COPY_ON_READ) {
|
|
|
|
/* If we touch the same cluster it counts as an overlap. This
|
|
|
|
* guarantees that allocating writes will be serialized and not race
|
|
|
|
* with each other for the same cluster. For example, in copy-on-read
|
|
|
|
* it ensures that the CoR read and write operations are atomic and
|
|
|
|
* guest writes cannot interleave between them. */
|
|
|
|
mark_request_serialising(req, bdrv_get_cluster_size(bs));
|
|
|
|
}
|
|
|
|
|
2015-12-01 10:36:28 +01:00
|
|
|
if (!(flags & BDRV_REQ_NO_SERIALISING)) {
|
|
|
|
wait_serialising_requests(req);
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (flags & BDRV_REQ_COPY_ON_READ) {
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
int64_t pnum;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-10-12 05:47:18 +02:00
|
|
|
ret = bdrv_is_allocated(bs, offset, bytes, &pnum);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2017-10-12 05:47:18 +02:00
|
|
|
if (!ret || pnum != bytes) {
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_co_do_copy_on_readv(child, offset, bytes, qiov);
|
2015-04-28 15:27:52 +02:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-07-15 20:31:59 +02:00
|
|
|
/* Forward the request to the BlockDriver, possibly fragmenting it */
|
2016-06-01 17:13:47 +02:00
|
|
|
total_bytes = bdrv_getlength(bs);
|
|
|
|
if (total_bytes < 0) {
|
|
|
|
ret = total_bytes;
|
|
|
|
goto out;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-06-01 17:13:47 +02:00
|
|
|
max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
|
2016-07-15 20:31:59 +02:00
|
|
|
if (bytes <= max_bytes && bytes <= max_transfer) {
|
2016-06-01 17:13:47 +02:00
|
|
|
ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
|
2016-07-15 20:31:59 +02:00
|
|
|
goto out;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-07-15 20:31:59 +02:00
|
|
|
while (bytes_remaining) {
|
|
|
|
int num;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-07-15 20:31:59 +02:00
|
|
|
if (max_bytes) {
|
|
|
|
QEMUIOVector local_qiov;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-07-15 20:31:59 +02:00
|
|
|
num = MIN(bytes_remaining, MIN(max_bytes, max_transfer));
|
|
|
|
assert(num);
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov);
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-07-15 20:31:59 +02:00
|
|
|
ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining,
|
|
|
|
num, &local_qiov, 0);
|
|
|
|
max_bytes -= num;
|
|
|
|
qemu_iovec_destroy(&local_qiov);
|
|
|
|
} else {
|
|
|
|
num = bytes_remaining;
|
|
|
|
ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0,
|
|
|
|
bytes_remaining);
|
|
|
|
}
|
|
|
|
if (ret < 0) {
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
bytes_remaining -= num;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
2016-07-15 20:31:59 +02:00
|
|
|
return ret < 0 ? ret : 0;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Handle a read request in coroutine context
|
|
|
|
*/
|
2016-06-20 21:31:46 +02:00
|
|
|
int coroutine_fn bdrv_co_preadv(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
|
|
|
|
BdrvRequestFlags flags)
|
|
|
|
{
|
2016-06-20 21:31:46 +02:00
|
|
|
BlockDriverState *bs = child->bs;
|
2015-04-28 15:27:52 +02:00
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
BdrvTrackedRequest req;
|
|
|
|
|
2016-06-24 00:37:24 +02:00
|
|
|
uint64_t align = bs->bl.request_alignment;
|
2015-04-28 15:27:52 +02:00
|
|
|
uint8_t *head_buf = NULL;
|
|
|
|
uint8_t *tail_buf = NULL;
|
|
|
|
QEMUIOVector local_qiov;
|
|
|
|
bool use_local_qiov = false;
|
|
|
|
int ret;
|
|
|
|
|
2017-08-04 12:50:36 +02:00
|
|
|
trace_bdrv_co_preadv(child->bs, offset, bytes, flags);
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = bdrv_check_byte_request(bs, offset, bytes);
|
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
|
|
|
|
2015-09-08 05:28:32 +02:00
|
|
|
/* Don't do copy-on-read if we read data before write operation */
|
2017-06-05 14:38:50 +02:00
|
|
|
if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) {
|
2015-04-28 15:27:52 +02:00
|
|
|
flags |= BDRV_REQ_COPY_ON_READ;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Align read if necessary by padding qiov */
|
|
|
|
if (offset & (align - 1)) {
|
|
|
|
head_buf = qemu_blockalign(bs, align);
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov + 2);
|
|
|
|
qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
|
|
|
|
use_local_qiov = true;
|
|
|
|
|
|
|
|
bytes += offset & (align - 1);
|
|
|
|
offset = offset & ~(align - 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((offset + bytes) & (align - 1)) {
|
|
|
|
if (!use_local_qiov) {
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov + 1);
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
|
|
|
|
use_local_qiov = true;
|
|
|
|
}
|
|
|
|
tail_buf = qemu_blockalign(bs, align);
|
|
|
|
qemu_iovec_add(&local_qiov, tail_buf,
|
|
|
|
align - ((offset + bytes) & (align - 1)));
|
|
|
|
|
|
|
|
bytes = ROUND_UP(bytes, align);
|
|
|
|
}
|
|
|
|
|
2015-11-09 11:16:46 +01:00
|
|
|
tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_preadv(child, &req, offset, bytes, align,
|
2015-04-28 15:27:52 +02:00
|
|
|
use_local_qiov ? &local_qiov : qiov,
|
|
|
|
flags);
|
|
|
|
tracked_request_end(&req);
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (use_local_qiov) {
|
|
|
|
qemu_iovec_destroy(&local_qiov);
|
|
|
|
qemu_vfree(head_buf);
|
|
|
|
qemu_vfree(tail_buf);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-05-30 13:52:34 +02:00
|
|
|
static int coroutine_fn bdrv_co_do_readv(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
|
|
|
BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2016-06-20 21:31:46 +02:00
|
|
|
return bdrv_co_preadv(child, sector_num << BDRV_SECTOR_BITS,
|
2016-04-25 14:57:23 +02:00
|
|
|
nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-05-24 17:21:22 +02:00
|
|
|
int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num,
|
|
|
|
int nb_sectors, QEMUIOVector *qiov)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2016-05-30 13:52:34 +02:00
|
|
|
return bdrv_co_do_readv(child, sector_num, nb_sectors, qiov, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-06-01 23:10:03 +02:00
|
|
|
static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
|
2017-06-09 12:18:08 +02:00
|
|
|
int64_t offset, int bytes, BdrvRequestFlags flags)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {0};
|
|
|
|
int ret = 0;
|
block: Honor BDRV_REQ_FUA during write_zeroes
The block layer has a couple of cases where it can lose
Force Unit Access semantics when writing a large block of
zeroes, such that the request returns before the zeroes
have been guaranteed to land on underlying media.
SCSI does not support FUA during WRITESAME(10/16); FUA is only
supported if it falls back to WRITE(10/16). But where the
underlying device is new enough to not need a fallback, it
means that any upper layer request with FUA semantics was
silently ignoring BDRV_REQ_FUA.
Conversely, NBD has situations where it can support FUA but not
ZERO_WRITE; when that happens, the generic block layer fallback
to bdrv_driver_pwritev() (or the older bdrv_co_writev() in qemu
2.6) was losing the FUA flag.
The problem of losing flags unrelated to ZERO_WRITE has been
latent in bdrv_co_do_write_zeroes() since commit aa7bfbff, but
back then, it did not matter because there was no FUA flag. It
became observable when commit 93f5e6d8 paved the way for flags
that can impact correctness, when we should have been using
bdrv_co_writev_flags() with modified flags. Compare to commit
9eeb6dd, which got flag manipulation right in
bdrv_co_do_zero_pwritev().
Symptoms: I tested with qemu-io with default writethrough cache
(which is supposed to use FUA semantics on every write), and
targetted an NBD client connected to a server that intentionally
did not advertise NBD_FLAG_SEND_FUA. When doing 'write 0 512',
the NBD client sent two operations (NBD_CMD_WRITE then
NBD_CMD_FLUSH) to get the fallback FUA semantics; but when doing
'write -z 0 512', the NBD client sent only NBD_CMD_WRITE.
The fix is do to a cleanup bdrv_co_flush() at the end of the
operation if any step in the middle relied on a BDS that does
not natively support FUA for that step (note that we don't
need to flush after every operation, if the operation is broken
into chunks based on bounce-buffer sizing). Each BDS gains a
new flag .supported_zero_flags, which parallels the use of
.supported_write_flags but only when accessing a zero write
operation (the flags MUST be different, because of SCSI having
different semantics based on WRITE vs. WRITESAME; and also
because BDRV_REQ_MAY_UNMAP only makes sense on zero writes).
Also fix some documentation to describe -ENOTSUP semantics,
particularly since iscsi depends on those semantics.
Down the road, we may want to add a driver where its
.bdrv_co_pwritev() honors all three of BDRV_REQ_FUA,
BDRV_REQ_ZERO_WRITE, and BDRV_REQ_MAY_UNMAP, and advertise
this via bs->supported_write_flags for blocks opened by that
driver; such a driver should NOT supply .bdrv_co_write_zeroes
nor .supported_zero_flags. But none of the drivers touched
in this patch want to do that (the act of writing zeroes is
different enough from normal writes to deserve a second
callback).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-05-04 00:39:07 +02:00
|
|
|
bool need_flush = false;
|
2016-05-26 05:48:45 +02:00
|
|
|
int head = 0;
|
|
|
|
int tail = 0;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-06-01 23:10:02 +02:00
|
|
|
int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
|
2016-06-24 00:37:24 +02:00
|
|
|
int alignment = MAX(bs->bl.pwrite_zeroes_alignment,
|
|
|
|
bs->bl.request_alignment);
|
block: Perform copy-on-read in loop
Improve our braindead copy-on-read implementation. Pre-patch,
we have multiple issues:
- we create a bounce buffer and perform a write for the entire
request, even if the active image already has 99% of the
clusters occupied, and really only needs to copy-on-read the
remaining 1% of the clusters
- our bounce buffer was as large as the read request, and can
needlessly exhaust our memory by using double the memory of
the request size (the original request plus our bounce buffer),
rather than a capped maximum overhead beyond the original
- if a driver has a max_transfer limit, we are bypassing the
normal code in bdrv_aligned_preadv() that fragments to that
limit, and instead attempt to read the entire buffer from the
driver in one go, which some drivers may assert on
- a client can request a large request of nearly 2G such that
rounding the request out to cluster boundaries results in a
byte count larger than 2G. While this cannot exceed 32 bits,
it DOES have some follow-on problems:
-- the call to bdrv_driver_pread() can assert for exceeding
BDRV_REQUEST_MAX_BYTES, if the driver is old and lacks
.bdrv_co_preadv
-- if the buffer is all zeroes, the subsequent call to
bdrv_co_do_pwrite_zeroes is a no-op due to a negative size,
which means we did not actually copy on read
Fix all of these issues by breaking up the action into a loop,
where each iteration is capped to sane limits. Also, querying
the allocation status allows us to optimize: when data is
already present in the active layer, we don't need to bounce.
Note that the code has a telling comment that copy-on-read
should probably be a filter driver rather than a bolt-on hack
in io.c; but that remains a task for another day.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-05 21:02:47 +02:00
|
|
|
int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER);
|
2016-06-01 23:10:03 +02:00
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2016-07-21 21:34:48 +02:00
|
|
|
assert(alignment % bs->bl.request_alignment == 0);
|
|
|
|
head = offset % alignment;
|
2017-06-09 12:18:08 +02:00
|
|
|
tail = (offset + bytes) % alignment;
|
2016-07-21 21:34:48 +02:00
|
|
|
max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment);
|
|
|
|
assert(max_write_zeroes >= bs->bl.request_alignment);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
while (bytes > 0 && !ret) {
|
|
|
|
int num = bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
/* Align request. Block drivers can expect the "bulk" of the request
|
2016-05-26 05:48:45 +02:00
|
|
|
* to be aligned, and that unaligned requests do not cross cluster
|
|
|
|
* boundaries.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2016-05-26 05:48:45 +02:00
|
|
|
if (head) {
|
2016-11-17 21:13:56 +01:00
|
|
|
/* Make a small request up to the first aligned sector. For
|
|
|
|
* convenience, limit this request to max_transfer even if
|
|
|
|
* we don't need to fall back to writes. */
|
2017-06-09 12:18:08 +02:00
|
|
|
num = MIN(MIN(bytes, max_transfer), alignment - head);
|
2016-11-17 21:13:56 +01:00
|
|
|
head = (head + num) % alignment;
|
|
|
|
assert(num < max_write_zeroes);
|
2016-06-01 23:10:03 +02:00
|
|
|
} else if (tail && num > alignment) {
|
2016-05-26 05:48:45 +02:00
|
|
|
/* Shorten the request to the last aligned sector. */
|
|
|
|
num -= tail;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* limit request size */
|
|
|
|
if (num > max_write_zeroes) {
|
|
|
|
num = max_write_zeroes;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = -ENOTSUP;
|
|
|
|
/* First try the efficient write zeroes operation */
|
2016-06-01 23:10:03 +02:00
|
|
|
if (drv->bdrv_co_pwrite_zeroes) {
|
|
|
|
ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
|
|
|
|
flags & bs->supported_zero_flags);
|
|
|
|
if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
|
|
|
|
!(bs->supported_zero_flags & BDRV_REQ_FUA)) {
|
|
|
|
need_flush = true;
|
|
|
|
}
|
block: Honor BDRV_REQ_FUA during write_zeroes
The block layer has a couple of cases where it can lose
Force Unit Access semantics when writing a large block of
zeroes, such that the request returns before the zeroes
have been guaranteed to land on underlying media.
SCSI does not support FUA during WRITESAME(10/16); FUA is only
supported if it falls back to WRITE(10/16). But where the
underlying device is new enough to not need a fallback, it
means that any upper layer request with FUA semantics was
silently ignoring BDRV_REQ_FUA.
Conversely, NBD has situations where it can support FUA but not
ZERO_WRITE; when that happens, the generic block layer fallback
to bdrv_driver_pwritev() (or the older bdrv_co_writev() in qemu
2.6) was losing the FUA flag.
The problem of losing flags unrelated to ZERO_WRITE has been
latent in bdrv_co_do_write_zeroes() since commit aa7bfbff, but
back then, it did not matter because there was no FUA flag. It
became observable when commit 93f5e6d8 paved the way for flags
that can impact correctness, when we should have been using
bdrv_co_writev_flags() with modified flags. Compare to commit
9eeb6dd, which got flag manipulation right in
bdrv_co_do_zero_pwritev().
Symptoms: I tested with qemu-io with default writethrough cache
(which is supposed to use FUA semantics on every write), and
targetted an NBD client connected to a server that intentionally
did not advertise NBD_FLAG_SEND_FUA. When doing 'write 0 512',
the NBD client sent two operations (NBD_CMD_WRITE then
NBD_CMD_FLUSH) to get the fallback FUA semantics; but when doing
'write -z 0 512', the NBD client sent only NBD_CMD_WRITE.
The fix is do to a cleanup bdrv_co_flush() at the end of the
operation if any step in the middle relied on a BDS that does
not natively support FUA for that step (note that we don't
need to flush after every operation, if the operation is broken
into chunks based on bounce-buffer sizing). Each BDS gains a
new flag .supported_zero_flags, which parallels the use of
.supported_write_flags but only when accessing a zero write
operation (the flags MUST be different, because of SCSI having
different semantics based on WRITE vs. WRITESAME; and also
because BDRV_REQ_MAY_UNMAP only makes sense on zero writes).
Also fix some documentation to describe -ENOTSUP semantics,
particularly since iscsi depends on those semantics.
Down the road, we may want to add a driver where its
.bdrv_co_pwritev() honors all three of BDRV_REQ_FUA,
BDRV_REQ_ZERO_WRITE, and BDRV_REQ_MAY_UNMAP, and advertise
this via bs->supported_write_flags for blocks opened by that
driver; such a driver should NOT supply .bdrv_co_write_zeroes
nor .supported_zero_flags. But none of the drivers touched
in this patch want to do that (the act of writing zeroes is
different enough from normal writes to deserve a second
callback).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-05-04 00:39:07 +02:00
|
|
|
} else {
|
|
|
|
assert(!bs->supported_zero_flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (ret == -ENOTSUP) {
|
|
|
|
/* Fall back to bounce buffer if write zeroes is unsupported */
|
block: Honor BDRV_REQ_FUA during write_zeroes
The block layer has a couple of cases where it can lose
Force Unit Access semantics when writing a large block of
zeroes, such that the request returns before the zeroes
have been guaranteed to land on underlying media.
SCSI does not support FUA during WRITESAME(10/16); FUA is only
supported if it falls back to WRITE(10/16). But where the
underlying device is new enough to not need a fallback, it
means that any upper layer request with FUA semantics was
silently ignoring BDRV_REQ_FUA.
Conversely, NBD has situations where it can support FUA but not
ZERO_WRITE; when that happens, the generic block layer fallback
to bdrv_driver_pwritev() (or the older bdrv_co_writev() in qemu
2.6) was losing the FUA flag.
The problem of losing flags unrelated to ZERO_WRITE has been
latent in bdrv_co_do_write_zeroes() since commit aa7bfbff, but
back then, it did not matter because there was no FUA flag. It
became observable when commit 93f5e6d8 paved the way for flags
that can impact correctness, when we should have been using
bdrv_co_writev_flags() with modified flags. Compare to commit
9eeb6dd, which got flag manipulation right in
bdrv_co_do_zero_pwritev().
Symptoms: I tested with qemu-io with default writethrough cache
(which is supposed to use FUA semantics on every write), and
targetted an NBD client connected to a server that intentionally
did not advertise NBD_FLAG_SEND_FUA. When doing 'write 0 512',
the NBD client sent two operations (NBD_CMD_WRITE then
NBD_CMD_FLUSH) to get the fallback FUA semantics; but when doing
'write -z 0 512', the NBD client sent only NBD_CMD_WRITE.
The fix is do to a cleanup bdrv_co_flush() at the end of the
operation if any step in the middle relied on a BDS that does
not natively support FUA for that step (note that we don't
need to flush after every operation, if the operation is broken
into chunks based on bounce-buffer sizing). Each BDS gains a
new flag .supported_zero_flags, which parallels the use of
.supported_write_flags but only when accessing a zero write
operation (the flags MUST be different, because of SCSI having
different semantics based on WRITE vs. WRITESAME; and also
because BDRV_REQ_MAY_UNMAP only makes sense on zero writes).
Also fix some documentation to describe -ENOTSUP semantics,
particularly since iscsi depends on those semantics.
Down the road, we may want to add a driver where its
.bdrv_co_pwritev() honors all three of BDRV_REQ_FUA,
BDRV_REQ_ZERO_WRITE, and BDRV_REQ_MAY_UNMAP, and advertise
this via bs->supported_write_flags for blocks opened by that
driver; such a driver should NOT supply .bdrv_co_write_zeroes
nor .supported_zero_flags. But none of the drivers touched
in this patch want to do that (the act of writing zeroes is
different enough from normal writes to deserve a second
callback).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-05-04 00:39:07 +02:00
|
|
|
BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
|
|
|
|
|
|
|
|
if ((flags & BDRV_REQ_FUA) &&
|
|
|
|
!(bs->supported_write_flags & BDRV_REQ_FUA)) {
|
|
|
|
/* No need for bdrv_driver_pwrite() to do a fallback
|
|
|
|
* flush on each chunk; use just one at the end */
|
|
|
|
write_flags &= ~BDRV_REQ_FUA;
|
|
|
|
need_flush = true;
|
|
|
|
}
|
2016-06-24 00:37:19 +02:00
|
|
|
num = MIN(num, max_transfer);
|
2016-06-01 23:10:03 +02:00
|
|
|
iov.iov_len = num;
|
2015-04-28 15:27:52 +02:00
|
|
|
if (iov.iov_base == NULL) {
|
2016-06-01 23:10:03 +02:00
|
|
|
iov.iov_base = qemu_try_blockalign(bs, num);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (iov.iov_base == NULL) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto fail;
|
|
|
|
}
|
2016-06-01 23:10:03 +02:00
|
|
|
memset(iov.iov_base, 0, num);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
|
|
|
|
2016-06-01 23:10:03 +02:00
|
|
|
ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
/* Keep bounce buffer around if it is big enough for all
|
|
|
|
* all future requests.
|
|
|
|
*/
|
2016-06-24 00:37:19 +02:00
|
|
|
if (num < max_transfer) {
|
2015-04-28 15:27:52 +02:00
|
|
|
qemu_vfree(iov.iov_base);
|
|
|
|
iov.iov_base = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-06-01 23:10:03 +02:00
|
|
|
offset += num;
|
2017-06-09 12:18:08 +02:00
|
|
|
bytes -= num;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
fail:
|
block: Honor BDRV_REQ_FUA during write_zeroes
The block layer has a couple of cases where it can lose
Force Unit Access semantics when writing a large block of
zeroes, such that the request returns before the zeroes
have been guaranteed to land on underlying media.
SCSI does not support FUA during WRITESAME(10/16); FUA is only
supported if it falls back to WRITE(10/16). But where the
underlying device is new enough to not need a fallback, it
means that any upper layer request with FUA semantics was
silently ignoring BDRV_REQ_FUA.
Conversely, NBD has situations where it can support FUA but not
ZERO_WRITE; when that happens, the generic block layer fallback
to bdrv_driver_pwritev() (or the older bdrv_co_writev() in qemu
2.6) was losing the FUA flag.
The problem of losing flags unrelated to ZERO_WRITE has been
latent in bdrv_co_do_write_zeroes() since commit aa7bfbff, but
back then, it did not matter because there was no FUA flag. It
became observable when commit 93f5e6d8 paved the way for flags
that can impact correctness, when we should have been using
bdrv_co_writev_flags() with modified flags. Compare to commit
9eeb6dd, which got flag manipulation right in
bdrv_co_do_zero_pwritev().
Symptoms: I tested with qemu-io with default writethrough cache
(which is supposed to use FUA semantics on every write), and
targetted an NBD client connected to a server that intentionally
did not advertise NBD_FLAG_SEND_FUA. When doing 'write 0 512',
the NBD client sent two operations (NBD_CMD_WRITE then
NBD_CMD_FLUSH) to get the fallback FUA semantics; but when doing
'write -z 0 512', the NBD client sent only NBD_CMD_WRITE.
The fix is do to a cleanup bdrv_co_flush() at the end of the
operation if any step in the middle relied on a BDS that does
not natively support FUA for that step (note that we don't
need to flush after every operation, if the operation is broken
into chunks based on bounce-buffer sizing). Each BDS gains a
new flag .supported_zero_flags, which parallels the use of
.supported_write_flags but only when accessing a zero write
operation (the flags MUST be different, because of SCSI having
different semantics based on WRITE vs. WRITESAME; and also
because BDRV_REQ_MAY_UNMAP only makes sense on zero writes).
Also fix some documentation to describe -ENOTSUP semantics,
particularly since iscsi depends on those semantics.
Down the road, we may want to add a driver where its
.bdrv_co_pwritev() honors all three of BDRV_REQ_FUA,
BDRV_REQ_ZERO_WRITE, and BDRV_REQ_MAY_UNMAP, and advertise
this via bs->supported_write_flags for blocks opened by that
driver; such a driver should NOT supply .bdrv_co_write_zeroes
nor .supported_zero_flags. But none of the drivers touched
in this patch want to do that (the act of writing zeroes is
different enough from normal writes to deserve a second
callback).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-05-04 00:39:07 +02:00
|
|
|
if (ret == 0 && need_flush) {
|
|
|
|
ret = bdrv_co_flush(bs);
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
qemu_vfree(iov.iov_base);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
* Forwards an already correctly aligned write request to the BlockDriver,
|
|
|
|
* after possibly fragmenting it.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2017-02-09 15:58:43 +01:00
|
|
|
static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
|
2016-06-24 00:37:05 +02:00
|
|
|
int64_t align, QEMUIOVector *qiov, int flags)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-02-09 15:58:43 +01:00
|
|
|
BlockDriverState *bs = child->bs;
|
2015-04-28 15:27:52 +02:00
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
bool waited;
|
|
|
|
int ret;
|
|
|
|
|
2016-06-03 18:42:51 +02:00
|
|
|
int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
uint64_t bytes_remaining = bytes;
|
|
|
|
int max_transfer;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2017-06-28 14:05:10 +02:00
|
|
|
if (bdrv_has_readonly_bitmaps(bs)) {
|
|
|
|
return -EPERM;
|
|
|
|
}
|
|
|
|
|
2016-06-24 00:37:05 +02:00
|
|
|
assert(is_power_of_2(align));
|
|
|
|
assert((offset & (align - 1)) == 0);
|
|
|
|
assert((bytes & (align - 1)) == 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
assert(!qiov || bytes == qiov->size);
|
2016-03-21 15:11:42 +01:00
|
|
|
assert((bs->open_flags & BDRV_O_NO_IO) == 0);
|
2016-06-13 20:56:35 +02:00
|
|
|
assert(!(flags & ~BDRV_REQ_MASK));
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
|
|
|
|
align);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
waited = wait_serialising_requests(req);
|
|
|
|
assert(!waited || !req->serialising);
|
|
|
|
assert(req->overlap_offset <= offset);
|
|
|
|
assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
|
2018-04-21 15:29:23 +02:00
|
|
|
if (flags & BDRV_REQ_WRITE_UNCHANGED) {
|
|
|
|
assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
|
|
|
|
} else {
|
|
|
|
assert(child->perm & BLK_PERM_WRITE);
|
|
|
|
}
|
2017-04-11 17:52:26 +02:00
|
|
|
assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
|
|
|
|
|
|
|
|
if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
|
2016-06-01 23:10:13 +02:00
|
|
|
!(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
|
2015-04-28 15:27:52 +02:00
|
|
|
qemu_iovec_is_zero(qiov)) {
|
|
|
|
flags |= BDRV_REQ_ZERO_WRITE;
|
|
|
|
if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
|
|
|
|
flags |= BDRV_REQ_MAY_UNMAP;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ret < 0) {
|
|
|
|
/* Do nothing, write notifier decided to fail this request */
|
|
|
|
} else if (flags & BDRV_REQ_ZERO_WRITE) {
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
|
2016-06-03 18:42:51 +02:00
|
|
|
ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
|
2016-07-22 10:17:49 +02:00
|
|
|
} else if (flags & BDRV_REQ_WRITE_COMPRESSED) {
|
|
|
|
ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov);
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
} else if (bytes <= max_transfer) {
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV);
|
2016-04-25 12:13:39 +02:00
|
|
|
ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
} else {
|
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV);
|
|
|
|
while (bytes_remaining) {
|
|
|
|
int num = MIN(bytes_remaining, max_transfer);
|
|
|
|
QEMUIOVector local_qiov;
|
|
|
|
int local_flags = flags;
|
|
|
|
|
|
|
|
assert(num);
|
|
|
|
if (num < bytes_remaining && (flags & BDRV_REQ_FUA) &&
|
|
|
|
!(bs->supported_write_flags & BDRV_REQ_FUA)) {
|
|
|
|
/* If FUA is going to be emulated by flush, we only
|
|
|
|
* need to flush on the last iteration */
|
|
|
|
local_flags &= ~BDRV_REQ_FUA;
|
|
|
|
}
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov);
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
|
|
|
|
|
|
|
|
ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining,
|
|
|
|
num, &local_qiov, local_flags);
|
|
|
|
qemu_iovec_destroy(&local_qiov);
|
|
|
|
if (ret < 0) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
bytes_remaining -= num;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-06-05 14:39:01 +02:00
|
|
|
atomic_inc(&bs->write_gen);
|
2017-09-25 16:55:25 +02:00
|
|
|
bdrv_set_dirty(bs, offset, bytes);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-06-05 14:39:00 +02:00
|
|
|
stat64_max(&bs->wr_highest_offset, offset + bytes);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (ret >= 0) {
|
2016-06-03 18:42:51 +02:00
|
|
|
bs->total_sectors = MAX(bs->total_sectors, end_sector);
|
block: Fragment writes to max transfer length
Drivers should be able to rely on the block layer honoring the
max transfer length, rather than needing to return -EINVAL
(iscsi) or manually fragment things (nbd). We already fragment
write zeroes at the block layer; this patch adds the fragmentation
for normal writes, after requests have been aligned (fragmenting
before alignment would lead to multiple unaligned requests, rather
than just the head and tail).
When fragmenting a large request where FUA was requested, but
where we know that FUA is implemented by flushing all requests
rather than the given request, then we can still get by with
only one flush. Note, however, that we need a followup patch
to the raw format driver to avoid a regression in the number of
flushes actually issued.
The return value was previously nebulous on success (sometimes
zero, sometimes the length written); since we never have a short
write, and since fragmenting may store yet another positive
value in 'ret', change the function to always return 0 on success,
matching what we do in bdrv_aligned_preadv().
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 1468607524-19021-4-git-send-email-eblake@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-07-15 20:32:01 +02:00
|
|
|
ret = 0;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-02-09 15:58:43 +01:00
|
|
|
static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child,
|
2015-05-13 15:12:00 +02:00
|
|
|
int64_t offset,
|
|
|
|
unsigned int bytes,
|
|
|
|
BdrvRequestFlags flags,
|
|
|
|
BdrvTrackedRequest *req)
|
|
|
|
{
|
2017-02-09 15:58:43 +01:00
|
|
|
BlockDriverState *bs = child->bs;
|
2015-05-13 15:12:00 +02:00
|
|
|
uint8_t *buf = NULL;
|
|
|
|
QEMUIOVector local_qiov;
|
|
|
|
struct iovec iov;
|
2016-06-24 00:37:24 +02:00
|
|
|
uint64_t align = bs->bl.request_alignment;
|
2015-05-13 15:12:00 +02:00
|
|
|
unsigned int head_padding_bytes, tail_padding_bytes;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
head_padding_bytes = offset & (align - 1);
|
2017-04-26 10:40:46 +02:00
|
|
|
tail_padding_bytes = (align - (offset + bytes)) & (align - 1);
|
2015-05-13 15:12:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
assert(flags & BDRV_REQ_ZERO_WRITE);
|
|
|
|
if (head_padding_bytes || tail_padding_bytes) {
|
|
|
|
buf = qemu_blockalign(bs, align);
|
|
|
|
iov = (struct iovec) {
|
|
|
|
.iov_base = buf,
|
|
|
|
.iov_len = align,
|
|
|
|
};
|
|
|
|
qemu_iovec_init_external(&local_qiov, &iov, 1);
|
|
|
|
}
|
|
|
|
if (head_padding_bytes) {
|
|
|
|
uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
|
|
|
|
|
|
|
|
/* RMW the unaligned part before head. */
|
|
|
|
mark_request_serialising(req, align);
|
|
|
|
wait_serialising_requests(req);
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_preadv(child, req, offset & ~(align - 1), align,
|
2015-05-13 15:12:00 +02:00
|
|
|
align, &local_qiov, 0);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
|
2015-05-13 15:12:00 +02:00
|
|
|
|
|
|
|
memset(buf + head_padding_bytes, 0, zero_bytes);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_pwritev(child, req, offset & ~(align - 1), align,
|
2016-06-24 00:37:05 +02:00
|
|
|
align, &local_qiov,
|
2015-05-13 15:12:00 +02:00
|
|
|
flags & ~BDRV_REQ_ZERO_WRITE);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
offset += zero_bytes;
|
|
|
|
bytes -= zero_bytes;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(!bytes || (offset & (align - 1)) == 0);
|
|
|
|
if (bytes >= align) {
|
|
|
|
/* Write the aligned part in the middle. */
|
|
|
|
uint64_t aligned_bytes = bytes & ~(align - 1);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align,
|
2015-05-13 15:12:00 +02:00
|
|
|
NULL, flags);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
bytes -= aligned_bytes;
|
|
|
|
offset += aligned_bytes;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(!bytes || (offset & (align - 1)) == 0);
|
|
|
|
if (bytes) {
|
|
|
|
assert(align == tail_padding_bytes + bytes);
|
|
|
|
/* RMW the unaligned part after tail. */
|
|
|
|
mark_request_serialising(req, align);
|
|
|
|
wait_serialising_requests(req);
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_preadv(child, req, offset, align,
|
2015-05-13 15:12:00 +02:00
|
|
|
align, &local_qiov, 0);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
|
2015-05-13 15:12:00 +02:00
|
|
|
|
|
|
|
memset(buf, 0, bytes);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_pwritev(child, req, offset, align, align,
|
2015-05-13 15:12:00 +02:00
|
|
|
&local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
|
|
|
|
}
|
|
|
|
fail:
|
|
|
|
qemu_vfree(buf);
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/*
|
|
|
|
* Handle a write request in coroutine context
|
|
|
|
*/
|
2016-06-20 21:31:46 +02:00
|
|
|
int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
|
|
|
|
BdrvRequestFlags flags)
|
|
|
|
{
|
2016-06-20 21:31:46 +02:00
|
|
|
BlockDriverState *bs = child->bs;
|
2015-04-28 15:27:52 +02:00
|
|
|
BdrvTrackedRequest req;
|
2016-06-24 00:37:24 +02:00
|
|
|
uint64_t align = bs->bl.request_alignment;
|
2015-04-28 15:27:52 +02:00
|
|
|
uint8_t *head_buf = NULL;
|
|
|
|
uint8_t *tail_buf = NULL;
|
|
|
|
QEMUIOVector local_qiov;
|
|
|
|
bool use_local_qiov = false;
|
|
|
|
int ret;
|
|
|
|
|
2017-08-04 12:50:36 +02:00
|
|
|
trace_bdrv_co_pwritev(child->bs, offset, bytes, flags);
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
if (!bs->drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
if (bs->read_only) {
|
2015-05-07 17:45:48 +02:00
|
|
|
return -EPERM;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2016-01-13 15:56:06 +01:00
|
|
|
assert(!(bs->open_flags & BDRV_O_INACTIVE));
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
ret = bdrv_check_byte_request(bs, offset, bytes);
|
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
/*
|
|
|
|
* Align write if necessary by performing a read-modify-write cycle.
|
|
|
|
* Pad qiov with the read parts and be sure to have a tracked request not
|
|
|
|
* only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
|
|
|
|
*/
|
2015-11-09 11:16:46 +01:00
|
|
|
tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2018-02-14 17:09:20 +01:00
|
|
|
if (flags & BDRV_REQ_ZERO_WRITE) {
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req);
|
2015-05-13 15:12:00 +02:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
if (offset & (align - 1)) {
|
|
|
|
QEMUIOVector head_qiov;
|
|
|
|
struct iovec head_iov;
|
|
|
|
|
|
|
|
mark_request_serialising(&req, align);
|
|
|
|
wait_serialising_requests(&req);
|
|
|
|
|
|
|
|
head_buf = qemu_blockalign(bs, align);
|
|
|
|
head_iov = (struct iovec) {
|
|
|
|
.iov_base = head_buf,
|
|
|
|
.iov_len = align,
|
|
|
|
};
|
|
|
|
qemu_iovec_init_external(&head_qiov, &head_iov, 1);
|
|
|
|
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_preadv(child, &req, offset & ~(align - 1), align,
|
2015-04-28 15:27:52 +02:00
|
|
|
align, &head_qiov, 0);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov + 2);
|
|
|
|
qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
|
|
|
|
use_local_qiov = true;
|
|
|
|
|
|
|
|
bytes += offset & (align - 1);
|
|
|
|
offset = offset & ~(align - 1);
|
2016-05-30 13:31:13 +02:00
|
|
|
|
|
|
|
/* We have read the tail already if the request is smaller
|
|
|
|
* than one aligned block.
|
|
|
|
*/
|
|
|
|
if (bytes < align) {
|
|
|
|
qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
|
|
|
|
bytes = align;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if ((offset + bytes) & (align - 1)) {
|
|
|
|
QEMUIOVector tail_qiov;
|
|
|
|
struct iovec tail_iov;
|
|
|
|
size_t tail_bytes;
|
|
|
|
bool waited;
|
|
|
|
|
|
|
|
mark_request_serialising(&req, align);
|
|
|
|
waited = wait_serialising_requests(&req);
|
|
|
|
assert(!waited || !use_local_qiov);
|
|
|
|
|
|
|
|
tail_buf = qemu_blockalign(bs, align);
|
|
|
|
tail_iov = (struct iovec) {
|
|
|
|
.iov_base = tail_buf,
|
|
|
|
.iov_len = align,
|
|
|
|
};
|
|
|
|
qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
|
|
|
|
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_preadv(child, &req, (offset + bytes) & ~(align - 1),
|
|
|
|
align, align, &tail_qiov, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
goto fail;
|
|
|
|
}
|
2015-06-16 14:19:22 +02:00
|
|
|
bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (!use_local_qiov) {
|
|
|
|
qemu_iovec_init(&local_qiov, qiov->niov + 1);
|
|
|
|
qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
|
|
|
|
use_local_qiov = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
tail_bytes = (offset + bytes) & (align - 1);
|
|
|
|
qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
|
|
|
|
|
|
|
|
bytes = ROUND_UP(bytes, align);
|
|
|
|
}
|
|
|
|
|
2017-02-09 15:58:43 +01:00
|
|
|
ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align,
|
2016-07-22 10:17:49 +02:00
|
|
|
use_local_qiov ? &local_qiov : qiov,
|
|
|
|
flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
fail:
|
|
|
|
|
|
|
|
if (use_local_qiov) {
|
|
|
|
qemu_iovec_destroy(&local_qiov);
|
|
|
|
}
|
|
|
|
qemu_vfree(head_buf);
|
|
|
|
qemu_vfree(tail_buf);
|
2015-05-13 15:12:00 +02:00
|
|
|
out:
|
|
|
|
tracked_request_end(&req);
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-05-30 13:52:34 +02:00
|
|
|
static int coroutine_fn bdrv_co_do_writev(BdrvChild *child,
|
2015-04-28 15:27:52 +02:00
|
|
|
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
|
|
|
|
BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2016-06-20 21:31:46 +02:00
|
|
|
return bdrv_co_pwritev(child, sector_num << BDRV_SECTOR_BITS,
|
2016-04-25 14:57:23 +02:00
|
|
|
nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-05-24 17:21:22 +02:00
|
|
|
int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num,
|
2015-04-28 15:27:52 +02:00
|
|
|
int nb_sectors, QEMUIOVector *qiov)
|
|
|
|
{
|
2016-05-30 13:52:34 +02:00
|
|
|
return bdrv_co_do_writev(child, sector_num, nb_sectors, qiov, 0);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-06-20 21:31:46 +02:00
|
|
|
int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
|
2017-06-09 12:18:08 +02:00
|
|
|
int bytes, BdrvRequestFlags flags)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-06-09 12:18:08 +02:00
|
|
|
trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-06-20 21:31:46 +02:00
|
|
|
if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
|
2015-04-28 15:27:52 +02:00
|
|
|
flags &= ~BDRV_REQ_MAY_UNMAP;
|
|
|
|
}
|
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
return bdrv_co_pwritev(child, offset, bytes, NULL,
|
2016-06-01 23:10:04 +02:00
|
|
|
BDRV_REQ_ZERO_WRITE | flags);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-09-23 03:45:50 +02:00
|
|
|
/*
|
|
|
|
* Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
|
|
|
|
*/
|
|
|
|
int bdrv_flush_all(void)
|
|
|
|
{
|
|
|
|
BdrvNextIterator it;
|
|
|
|
BlockDriverState *bs = NULL;
|
|
|
|
int result = 0;
|
|
|
|
|
|
|
|
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
|
|
|
|
AioContext *aio_context = bdrv_get_aio_context(bs);
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
aio_context_acquire(aio_context);
|
|
|
|
ret = bdrv_flush(bs);
|
|
|
|
if (ret < 0 && !result) {
|
|
|
|
result = ret;
|
|
|
|
}
|
|
|
|
aio_context_release(aio_context);
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2017-10-12 05:47:05 +02:00
|
|
|
typedef struct BdrvCoBlockStatusData {
|
2015-04-28 15:27:52 +02:00
|
|
|
BlockDriverState *bs;
|
|
|
|
BlockDriverState *base;
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
bool want_zero;
|
2017-10-12 05:47:05 +02:00
|
|
|
int64_t offset;
|
|
|
|
int64_t bytes;
|
|
|
|
int64_t *pnum;
|
|
|
|
int64_t *map;
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
BlockDriverState **file;
|
2017-10-12 05:47:05 +02:00
|
|
|
int ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
bool done;
|
2017-10-12 05:47:05 +02:00
|
|
|
} BdrvCoBlockStatusData;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2018-02-13 21:26:43 +01:00
|
|
|
int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs,
|
|
|
|
bool want_zero,
|
|
|
|
int64_t offset,
|
|
|
|
int64_t bytes,
|
|
|
|
int64_t *pnum,
|
|
|
|
int64_t *map,
|
|
|
|
BlockDriverState **file)
|
2017-07-13 17:30:28 +02:00
|
|
|
{
|
|
|
|
assert(bs->file && bs->file->bs);
|
2018-02-13 21:26:43 +01:00
|
|
|
*pnum = bytes;
|
|
|
|
*map = offset;
|
2017-07-13 17:30:28 +02:00
|
|
|
*file = bs->file->bs;
|
2018-02-13 21:26:43 +01:00
|
|
|
return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
|
2017-07-13 17:30:28 +02:00
|
|
|
}
|
|
|
|
|
2018-02-13 21:26:43 +01:00
|
|
|
int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs,
|
|
|
|
bool want_zero,
|
|
|
|
int64_t offset,
|
|
|
|
int64_t bytes,
|
|
|
|
int64_t *pnum,
|
|
|
|
int64_t *map,
|
|
|
|
BlockDriverState **file)
|
2017-07-13 17:30:28 +02:00
|
|
|
{
|
|
|
|
assert(bs->backing && bs->backing->bs);
|
2018-02-13 21:26:43 +01:00
|
|
|
*pnum = bytes;
|
|
|
|
*map = offset;
|
2017-07-13 17:30:28 +02:00
|
|
|
*file = bs->backing->bs;
|
2018-02-13 21:26:43 +01:00
|
|
|
return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
|
2017-07-13 17:30:28 +02:00
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/*
|
|
|
|
* Returns the allocation status of the specified sectors.
|
|
|
|
* Drivers not implementing the functionality are assumed to not support
|
|
|
|
* backing files, hence all their sectors are reported as allocated.
|
|
|
|
*
|
block: Add .bdrv_co_block_status() callback
We are gradually moving away from sector-based interfaces, towards
byte-based. Now that the block layer exposes byte-based allocation,
it's time to tackle the drivers. Add a new callback that operates
on as small as byte boundaries. Subsequent patches will then update
individual drivers, then finally remove .bdrv_co_get_block_status().
The new code also passes through the 'want_zero' hint, which will
allow subsequent patches to further optimize callers that only care
about how much of the image is allocated (want_zero is false),
rather than full details about runs of zeroes and which offsets the
allocation actually maps to (want_zero is true). As part of this
effort, fix another part of the documentation: the claim in commit
4c41cb4 that BDRV_BLOCK_ALLOCATED is short for 'DATA || ZERO' is a
lie at the block layer (see commit e88ae2264), even though it is
how the bit is computed from the driver layer. After all, there
are intentionally cases where we return ZERO but not ALLOCATED at
the block layer, when we know that a read sees zero because the
backing file is too short. Note that the driver interface is thus
slightly different than the public interface with regards to which
bits will be set, and what guarantees are provided on input.
We also add an assertion that any driver using the new callback will
make progress (the only time pnum will be 0 is if the block layer
already handled an out-of-bounds request, or if there is an error);
the old driver interface did not provide this guarantee, which
could lead to some inf-loops in drastic corner-case failures.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2018-02-13 21:26:41 +01:00
|
|
|
* If 'want_zero' is true, the caller is querying for mapping
|
|
|
|
* purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
|
|
|
|
* _ZERO where possible; otherwise, the result favors larger 'pnum',
|
|
|
|
* with a focus on accurate BDRV_BLOCK_ALLOCATED.
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
*
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
* If 'offset' is beyond the end of the disk image the return value is
|
2017-05-05 04:14:59 +02:00
|
|
|
* BDRV_BLOCK_EOF and 'pnum' is set to 0.
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
* 'bytes' is the max value 'pnum' should be set to. If bytes goes
|
2017-05-05 04:14:59 +02:00
|
|
|
* beyond the end of the disk image it will be clamped; if 'pnum' is set to
|
|
|
|
* the end of the image, then the returned value will include BDRV_BLOCK_EOF.
|
2016-01-26 04:58:48 +01:00
|
|
|
*
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
* 'pnum' is set to the number of bytes (including and immediately
|
|
|
|
* following the specified offset) that are easily known to be in the
|
|
|
|
* same allocated/unallocated state. Note that a second call starting
|
|
|
|
* at the original offset plus returned pnum may have the same status.
|
|
|
|
* The returned value is non-zero on success except at end-of-file.
|
|
|
|
*
|
|
|
|
* Returns negative errno on failure. Otherwise, if the
|
|
|
|
* BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
|
|
|
|
* set to the host mapping and BDS corresponding to the guest offset.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs,
|
|
|
|
bool want_zero,
|
|
|
|
int64_t offset, int64_t bytes,
|
|
|
|
int64_t *pnum, int64_t *map,
|
|
|
|
BlockDriverState **file)
|
|
|
|
{
|
|
|
|
int64_t total_size;
|
|
|
|
int64_t n; /* bytes */
|
2017-10-12 05:47:17 +02:00
|
|
|
int ret;
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
int64_t local_map = 0;
|
2017-10-12 05:46:57 +02:00
|
|
|
BlockDriverState *local_file = NULL;
|
2017-10-12 05:47:17 +02:00
|
|
|
int64_t aligned_offset, aligned_bytes;
|
|
|
|
uint32_t align;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-10-12 05:46:57 +02:00
|
|
|
assert(pnum);
|
|
|
|
*pnum = 0;
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
total_size = bdrv_getlength(bs);
|
|
|
|
if (total_size < 0) {
|
|
|
|
ret = total_size;
|
2017-10-12 05:46:57 +02:00
|
|
|
goto early_out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (offset >= total_size) {
|
2017-10-12 05:46:57 +02:00
|
|
|
ret = BDRV_BLOCK_EOF;
|
|
|
|
goto early_out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (!bytes) {
|
2017-10-12 05:46:57 +02:00
|
|
|
ret = 0;
|
|
|
|
goto early_out;
|
2017-10-05 21:02:44 +02:00
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
n = total_size - offset;
|
|
|
|
if (n < bytes) {
|
|
|
|
bytes = n;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
/* Must be non-NULL or bdrv_getlength() would have failed */
|
|
|
|
assert(bs->drv);
|
2018-02-13 21:27:01 +01:00
|
|
|
if (!bs->drv->bdrv_co_block_status) {
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
*pnum = bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (offset + bytes == total_size) {
|
2017-05-05 04:14:59 +02:00
|
|
|
ret |= BDRV_BLOCK_EOF;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
if (bs->drv->protocol_name) {
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
ret |= BDRV_BLOCK_OFFSET_VALID;
|
|
|
|
local_map = offset;
|
2017-10-12 05:46:57 +02:00
|
|
|
local_file = bs;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2017-10-12 05:46:57 +02:00
|
|
|
goto early_out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
2017-10-12 05:47:17 +02:00
|
|
|
|
|
|
|
/* Round out to request_alignment boundaries */
|
block: Add .bdrv_co_block_status() callback
We are gradually moving away from sector-based interfaces, towards
byte-based. Now that the block layer exposes byte-based allocation,
it's time to tackle the drivers. Add a new callback that operates
on as small as byte boundaries. Subsequent patches will then update
individual drivers, then finally remove .bdrv_co_get_block_status().
The new code also passes through the 'want_zero' hint, which will
allow subsequent patches to further optimize callers that only care
about how much of the image is allocated (want_zero is false),
rather than full details about runs of zeroes and which offsets the
allocation actually maps to (want_zero is true). As part of this
effort, fix another part of the documentation: the claim in commit
4c41cb4 that BDRV_BLOCK_ALLOCATED is short for 'DATA || ZERO' is a
lie at the block layer (see commit e88ae2264), even though it is
how the bit is computed from the driver layer. After all, there
are intentionally cases where we return ZERO but not ALLOCATED at
the block layer, when we know that a read sees zero because the
backing file is too short. Note that the driver interface is thus
slightly different than the public interface with regards to which
bits will be set, and what guarantees are provided on input.
We also add an assertion that any driver using the new callback will
make progress (the only time pnum will be 0 is if the block layer
already handled an out-of-bounds request, or if there is an error);
the old driver interface did not provide this guarantee, which
could lead to some inf-loops in drastic corner-case failures.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2018-02-13 21:26:41 +01:00
|
|
|
align = bs->bl.request_alignment;
|
2017-10-12 05:47:17 +02:00
|
|
|
aligned_offset = QEMU_ALIGN_DOWN(offset, align);
|
|
|
|
aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset;
|
|
|
|
|
2018-02-13 21:27:01 +01:00
|
|
|
ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset,
|
|
|
|
aligned_bytes, pnum, &local_map,
|
|
|
|
&local_file);
|
|
|
|
if (ret < 0) {
|
|
|
|
*pnum = 0;
|
|
|
|
goto out;
|
2017-10-12 05:47:17 +02:00
|
|
|
}
|
|
|
|
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
/*
|
2018-02-13 21:27:01 +01:00
|
|
|
* The driver's result must be a non-zero multiple of request_alignment.
|
2017-10-12 05:47:17 +02:00
|
|
|
* Clamp pnum and adjust map to original request.
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
*/
|
2018-02-13 21:27:01 +01:00
|
|
|
assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) &&
|
|
|
|
align > offset - aligned_offset);
|
2017-10-12 05:47:17 +02:00
|
|
|
*pnum -= offset - aligned_offset;
|
|
|
|
if (*pnum > bytes) {
|
|
|
|
*pnum = bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (ret & BDRV_BLOCK_OFFSET_VALID) {
|
2017-10-12 05:47:17 +02:00
|
|
|
local_map += offset - aligned_offset;
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (ret & BDRV_BLOCK_RAW) {
|
2017-10-12 05:46:57 +02:00
|
|
|
assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file);
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
ret = bdrv_co_block_status(local_file, want_zero, local_map,
|
|
|
|
*pnum, pnum, &local_map, &local_file);
|
2016-10-27 12:48:52 +02:00
|
|
|
goto out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
|
|
|
|
ret |= BDRV_BLOCK_ALLOCATED;
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
} else if (want_zero) {
|
2015-04-28 15:27:52 +02:00
|
|
|
if (bdrv_unallocated_blocks_are_zero(bs)) {
|
|
|
|
ret |= BDRV_BLOCK_ZERO;
|
2015-06-17 14:55:21 +02:00
|
|
|
} else if (bs->backing) {
|
|
|
|
BlockDriverState *bs2 = bs->backing->bs;
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
int64_t size2 = bdrv_getlength(bs2);
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (size2 >= 0 && offset >= size2) {
|
2015-04-28 15:27:52 +02:00
|
|
|
ret |= BDRV_BLOCK_ZERO;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
if (want_zero && local_file && local_file != bs &&
|
2015-04-28 15:27:52 +02:00
|
|
|
(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
|
|
|
|
(ret & BDRV_BLOCK_OFFSET_VALID)) {
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
int64_t file_pnum;
|
|
|
|
int ret2;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
ret2 = bdrv_co_block_status(local_file, want_zero, local_map,
|
|
|
|
*pnum, &file_pnum, NULL, NULL);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret2 >= 0) {
|
|
|
|
/* Ignore errors. This is just providing extra information, it
|
|
|
|
* is useful but not necessary.
|
|
|
|
*/
|
block: Exploit BDRV_BLOCK_EOF for larger zero blocks
When we have a BDS with unallocated clusters, but asking the status
of its underlying bs->file or backing layer encounters an end-of-file
condition, we know that the rest of the unallocated area will read as
zeroes. However, pre-patch, this required two separate calls to
bdrv_get_block_status(), as the first call stops at the point where
the underlying file ends. Thanks to BDRV_BLOCK_EOF, we can now widen
the results of the primary status if the secondary status already
includes BDRV_BLOCK_ZERO.
In turn, this fixes a TODO mentioned in iotest 154, where we can now
see that all sectors in a partial cluster at the end of a file read
as zero when coupling the shorter backing file's status along with our
knowledge that the remaining sectors came from an unallocated cluster.
Also, note that the loop in bdrv_co_get_block_status_above() had an
inefficent exit: in cases where the active layer sets BDRV_BLOCK_ZERO
but does NOT set BDRV_BLOCK_ALLOCATED (namely, where we know we read
zeroes merely because our unallocated clusters lie beyond the backing
file's shorter length), we still ended up probing the backing layer
even though we already had a good answer.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170505021500.19315-3-eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
2017-05-05 04:15:00 +02:00
|
|
|
if (ret2 & BDRV_BLOCK_EOF &&
|
|
|
|
(!file_pnum || ret2 & BDRV_BLOCK_ZERO)) {
|
|
|
|
/*
|
|
|
|
* It is valid for the format block driver to read
|
|
|
|
* beyond the end of the underlying file's current
|
|
|
|
* size; such areas read as zero.
|
|
|
|
*/
|
2015-04-28 15:27:52 +02:00
|
|
|
ret |= BDRV_BLOCK_ZERO;
|
|
|
|
} else {
|
|
|
|
/* Limit request to the range reported by the protocol driver */
|
|
|
|
*pnum = file_pnum;
|
|
|
|
ret |= (ret2 & BDRV_BLOCK_ZERO);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
out:
|
|
|
|
bdrv_dec_in_flight(bs);
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (ret >= 0 && offset + *pnum == total_size) {
|
2017-05-05 04:14:59 +02:00
|
|
|
ret |= BDRV_BLOCK_EOF;
|
|
|
|
}
|
2017-10-12 05:46:57 +02:00
|
|
|
early_out:
|
|
|
|
if (file) {
|
|
|
|
*file = local_file;
|
|
|
|
}
|
block: Switch bdrv_co_get_block_status() to byte-based
We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based. Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses. For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:04 +02:00
|
|
|
if (map) {
|
|
|
|
*map = local_map;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-10-12 05:47:07 +02:00
|
|
|
static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs,
|
|
|
|
BlockDriverState *base,
|
|
|
|
bool want_zero,
|
|
|
|
int64_t offset,
|
|
|
|
int64_t bytes,
|
|
|
|
int64_t *pnum,
|
|
|
|
int64_t *map,
|
|
|
|
BlockDriverState **file)
|
2015-06-08 07:56:07 +02:00
|
|
|
{
|
|
|
|
BlockDriverState *p;
|
2017-10-12 05:47:07 +02:00
|
|
|
int ret = 0;
|
block: Exploit BDRV_BLOCK_EOF for larger zero blocks
When we have a BDS with unallocated clusters, but asking the status
of its underlying bs->file or backing layer encounters an end-of-file
condition, we know that the rest of the unallocated area will read as
zeroes. However, pre-patch, this required two separate calls to
bdrv_get_block_status(), as the first call stops at the point where
the underlying file ends. Thanks to BDRV_BLOCK_EOF, we can now widen
the results of the primary status if the secondary status already
includes BDRV_BLOCK_ZERO.
In turn, this fixes a TODO mentioned in iotest 154, where we can now
see that all sectors in a partial cluster at the end of a file read
as zero when coupling the shorter backing file's status along with our
knowledge that the remaining sectors came from an unallocated cluster.
Also, note that the loop in bdrv_co_get_block_status_above() had an
inefficent exit: in cases where the active layer sets BDRV_BLOCK_ZERO
but does NOT set BDRV_BLOCK_ALLOCATED (namely, where we know we read
zeroes merely because our unallocated clusters lie beyond the backing
file's shorter length), we still ended up probing the backing layer
even though we already had a good answer.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170505021500.19315-3-eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
2017-05-05 04:15:00 +02:00
|
|
|
bool first = true;
|
2015-06-08 07:56:07 +02:00
|
|
|
|
|
|
|
assert(bs != base);
|
2015-06-17 14:55:21 +02:00
|
|
|
for (p = bs; p != base; p = backing_bs(p)) {
|
2017-10-12 05:47:07 +02:00
|
|
|
ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map,
|
|
|
|
file);
|
block: Exploit BDRV_BLOCK_EOF for larger zero blocks
When we have a BDS with unallocated clusters, but asking the status
of its underlying bs->file or backing layer encounters an end-of-file
condition, we know that the rest of the unallocated area will read as
zeroes. However, pre-patch, this required two separate calls to
bdrv_get_block_status(), as the first call stops at the point where
the underlying file ends. Thanks to BDRV_BLOCK_EOF, we can now widen
the results of the primary status if the secondary status already
includes BDRV_BLOCK_ZERO.
In turn, this fixes a TODO mentioned in iotest 154, where we can now
see that all sectors in a partial cluster at the end of a file read
as zero when coupling the shorter backing file's status along with our
knowledge that the remaining sectors came from an unallocated cluster.
Also, note that the loop in bdrv_co_get_block_status_above() had an
inefficent exit: in cases where the active layer sets BDRV_BLOCK_ZERO
but does NOT set BDRV_BLOCK_ALLOCATED (namely, where we know we read
zeroes merely because our unallocated clusters lie beyond the backing
file's shorter length), we still ended up probing the backing layer
even though we already had a good answer.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170505021500.19315-3-eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
2017-05-05 04:15:00 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) {
|
|
|
|
/*
|
|
|
|
* Reading beyond the end of the file continues to read
|
|
|
|
* zeroes, but we can only widen the result to the
|
|
|
|
* unallocated length we learned from an earlier
|
|
|
|
* iteration.
|
|
|
|
*/
|
2017-10-12 05:47:07 +02:00
|
|
|
*pnum = bytes;
|
block: Exploit BDRV_BLOCK_EOF for larger zero blocks
When we have a BDS with unallocated clusters, but asking the status
of its underlying bs->file or backing layer encounters an end-of-file
condition, we know that the rest of the unallocated area will read as
zeroes. However, pre-patch, this required two separate calls to
bdrv_get_block_status(), as the first call stops at the point where
the underlying file ends. Thanks to BDRV_BLOCK_EOF, we can now widen
the results of the primary status if the secondary status already
includes BDRV_BLOCK_ZERO.
In turn, this fixes a TODO mentioned in iotest 154, where we can now
see that all sectors in a partial cluster at the end of a file read
as zero when coupling the shorter backing file's status along with our
knowledge that the remaining sectors came from an unallocated cluster.
Also, note that the loop in bdrv_co_get_block_status_above() had an
inefficent exit: in cases where the active layer sets BDRV_BLOCK_ZERO
but does NOT set BDRV_BLOCK_ALLOCATED (namely, where we know we read
zeroes merely because our unallocated clusters lie beyond the backing
file's shorter length), we still ended up probing the backing layer
even though we already had a good answer.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170505021500.19315-3-eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
2017-05-05 04:15:00 +02:00
|
|
|
}
|
|
|
|
if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) {
|
2015-06-08 07:56:07 +02:00
|
|
|
break;
|
|
|
|
}
|
2017-10-12 05:47:07 +02:00
|
|
|
/* [offset, pnum] unallocated on this layer, which could be only
|
|
|
|
* the first part of [offset, bytes]. */
|
|
|
|
bytes = MIN(bytes, *pnum);
|
block: Exploit BDRV_BLOCK_EOF for larger zero blocks
When we have a BDS with unallocated clusters, but asking the status
of its underlying bs->file or backing layer encounters an end-of-file
condition, we know that the rest of the unallocated area will read as
zeroes. However, pre-patch, this required two separate calls to
bdrv_get_block_status(), as the first call stops at the point where
the underlying file ends. Thanks to BDRV_BLOCK_EOF, we can now widen
the results of the primary status if the secondary status already
includes BDRV_BLOCK_ZERO.
In turn, this fixes a TODO mentioned in iotest 154, where we can now
see that all sectors in a partial cluster at the end of a file read
as zero when coupling the shorter backing file's status along with our
knowledge that the remaining sectors came from an unallocated cluster.
Also, note that the loop in bdrv_co_get_block_status_above() had an
inefficent exit: in cases where the active layer sets BDRV_BLOCK_ZERO
but does NOT set BDRV_BLOCK_ALLOCATED (namely, where we know we read
zeroes merely because our unallocated clusters lie beyond the backing
file's shorter length), we still ended up probing the backing layer
even though we already had a good answer.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170505021500.19315-3-eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
2017-05-05 04:15:00 +02:00
|
|
|
first = false;
|
2015-06-08 07:56:07 +02:00
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
block: Convert bdrv_get_block_status_above() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status_above()
to bdrv_block_status_above() ensures that the compiler enforces that
all callers are updated. Likewise, since it a byte interface allows
an offset mapping that might not be sector aligned, split the mapping
out of the return value and into a pass-by-reference parameter. For
now, the io.c layer still assert()s that all uses are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status in the drivers.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), plus
updates for the new split return interface. But some code,
particularly bdrv_block_status(), gets a lot simpler because it no
longer has to mess with sectors. Likewise, mirror code no longer
computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop
an assertion about alignment because the loop no longer depends on
alignment (never mind that we don't really have a driver that
reports sub-sector alignments, so it's not really possible to test
the effect of sub-sector mirroring). Fix a neighboring assertion to
use is_power_of_2 while there.
For ease of review, bdrv_get_block_status() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:08 +02:00
|
|
|
/* Coroutine wrapper for bdrv_block_status_above() */
|
2017-10-12 05:47:07 +02:00
|
|
|
static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-10-12 05:47:05 +02:00
|
|
|
BdrvCoBlockStatusData *data = opaque;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-10-12 05:47:07 +02:00
|
|
|
data->ret = bdrv_co_block_status_above(data->bs, data->base,
|
|
|
|
data->want_zero,
|
|
|
|
data->offset, data->bytes,
|
|
|
|
data->pnum, data->map, data->file);
|
2015-04-28 15:27:52 +02:00
|
|
|
data->done = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2017-10-12 05:47:07 +02:00
|
|
|
* Synchronous wrapper around bdrv_co_block_status_above().
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
2017-10-12 05:47:07 +02:00
|
|
|
* See bdrv_co_block_status_above() for details.
|
2015-04-28 15:27:52 +02:00
|
|
|
*/
|
2017-10-12 05:47:06 +02:00
|
|
|
static int bdrv_common_block_status_above(BlockDriverState *bs,
|
|
|
|
BlockDriverState *base,
|
|
|
|
bool want_zero, int64_t offset,
|
|
|
|
int64_t bytes, int64_t *pnum,
|
|
|
|
int64_t *map,
|
|
|
|
BlockDriverState **file)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
Coroutine *co;
|
2017-10-12 05:47:05 +02:00
|
|
|
BdrvCoBlockStatusData data = {
|
2015-04-28 15:27:52 +02:00
|
|
|
.bs = bs,
|
2015-06-08 07:56:07 +02:00
|
|
|
.base = base,
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
.want_zero = want_zero,
|
2017-10-12 05:47:06 +02:00
|
|
|
.offset = offset,
|
|
|
|
.bytes = bytes,
|
|
|
|
.pnum = pnum,
|
|
|
|
.map = map,
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
.file = file,
|
2015-04-28 15:27:52 +02:00
|
|
|
.done = false,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (qemu_in_coroutine()) {
|
|
|
|
/* Fast-path if already in coroutine context */
|
2017-10-12 05:47:07 +02:00
|
|
|
bdrv_block_status_above_co_entry(&data);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
2017-10-12 05:47:07 +02:00
|
|
|
co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data);
|
block: Use bdrv_coroutine_enter to start I/O coroutines
BDRV_POLL_WHILE waits for the started I/O by releasing bs's ctx then polling
the main context, which relies on the yielded coroutine continuing on bs->ctx
before notifying qemu_aio_context with bdrv_wakeup().
Thus, using qemu_coroutine_enter to start I/O is wrong because if the coroutine
is entered from main loop, co->ctx will be qemu_aio_context, as a result of the
"release, poll, acquire" loop of BDRV_POLL_WHILE, race conditions happen when
both main thread and the iothread access the same BDS:
main loop iothread
-----------------------------------------------------------------------
blockdev_snapshot
aio_context_acquire(bs->ctx)
virtio_scsi_data_plane_handle_cmd
bdrv_drained_begin(bs->ctx)
bdrv_flush(bs)
bdrv_co_flush(bs) aio_context_acquire(bs->ctx).enter
...
qemu_coroutine_yield(co)
BDRV_POLL_WHILE()
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx).return
...
aio_co_wake(co)
aio_poll(qemu_aio_context) ...
co_schedule_bh_cb() ...
qemu_coroutine_enter(co) ...
/* (A) bdrv_co_flush(bs) /* (B) I/O on bs */
continues... */
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx)
Note that in above case, bdrv_drained_begin() doesn't do the "release,
poll, acquire" in BDRV_POLL_WHILE, because bs->in_flight == 0.
Fix this by using bdrv_coroutine_enter and enter coroutine in the right
context.
iotests 109 output is updated because the coroutine reenter flow during
mirror job complete is different (now through co_queue_wakeup, instead
of the unconditional qemu_coroutine_switch before), making the end job
len different.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
2017-04-10 14:20:17 +02:00
|
|
|
bdrv_coroutine_enter(bs, co);
|
2016-10-27 12:48:55 +02:00
|
|
|
BDRV_POLL_WHILE(bs, !data.done);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2017-10-12 05:47:06 +02:00
|
|
|
return data.ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
block: Convert bdrv_get_block_status_above() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status_above()
to bdrv_block_status_above() ensures that the compiler enforces that
all callers are updated. Likewise, since it a byte interface allows
an offset mapping that might not be sector aligned, split the mapping
out of the return value and into a pass-by-reference parameter. For
now, the io.c layer still assert()s that all uses are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status in the drivers.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), plus
updates for the new split return interface. But some code,
particularly bdrv_block_status(), gets a lot simpler because it no
longer has to mess with sectors. Likewise, mirror code no longer
computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop
an assertion about alignment because the loop no longer depends on
alignment (never mind that we don't really have a driver that
reports sub-sector alignments, so it's not really possible to test
the effect of sub-sector mirroring). Fix a neighboring assertion to
use is_power_of_2 while there.
For ease of review, bdrv_get_block_status() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:08 +02:00
|
|
|
int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base,
|
|
|
|
int64_t offset, int64_t bytes, int64_t *pnum,
|
|
|
|
int64_t *map, BlockDriverState **file)
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
{
|
block: Convert bdrv_get_block_status_above() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status_above()
to bdrv_block_status_above() ensures that the compiler enforces that
all callers are updated. Likewise, since it a byte interface allows
an offset mapping that might not be sector aligned, split the mapping
out of the return value and into a pass-by-reference parameter. For
now, the io.c layer still assert()s that all uses are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status in the drivers.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), plus
updates for the new split return interface. But some code,
particularly bdrv_block_status(), gets a lot simpler because it no
longer has to mess with sectors. Likewise, mirror code no longer
computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop
an assertion about alignment because the loop no longer depends on
alignment (never mind that we don't really have a driver that
reports sub-sector alignments, so it's not really possible to test
the effect of sub-sector mirroring). Fix a neighboring assertion to
use is_power_of_2 while there.
For ease of review, bdrv_get_block_status() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:08 +02:00
|
|
|
return bdrv_common_block_status_above(bs, base, true, offset, bytes,
|
|
|
|
pnum, map, file);
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
}
|
|
|
|
|
block: Convert bdrv_get_block_status() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated. For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value. We'll have some glue code that
munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention. But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:03 +02:00
|
|
|
int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
|
|
|
|
int64_t *pnum, int64_t *map, BlockDriverState **file)
|
2015-06-08 07:56:07 +02:00
|
|
|
{
|
block: Convert bdrv_get_block_status_above() to bytes
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status_above()
to bdrv_block_status_above() ensures that the compiler enforces that
all callers are updated. Likewise, since it a byte interface allows
an offset mapping that might not be sector aligned, split the mapping
out of the return value and into a pass-by-reference parameter. For
now, the io.c layer still assert()s that all uses are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status in the drivers.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), plus
updates for the new split return interface. But some code,
particularly bdrv_block_status(), gets a lot simpler because it no
longer has to mess with sectors. Likewise, mirror code no longer
computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop
an assertion about alignment because the loop no longer depends on
alignment (never mind that we don't really have a driver that
reports sub-sector alignments, so it's not really possible to test
the effect of sub-sector mirroring). Fix a neighboring assertion to
use is_power_of_2 while there.
For ease of review, bdrv_get_block_status() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:47:08 +02:00
|
|
|
return bdrv_block_status_above(bs, backing_bs(bs),
|
|
|
|
offset, bytes, pnum, map, file);
|
2015-06-08 07:56:07 +02:00
|
|
|
}
|
|
|
|
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset,
|
|
|
|
int64_t bytes, int64_t *pnum)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2017-10-12 05:47:06 +02:00
|
|
|
int ret;
|
|
|
|
int64_t dummy;
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
|
2017-10-12 05:47:06 +02:00
|
|
|
ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset,
|
|
|
|
bytes, pnum ? pnum : &dummy, NULL,
|
block: Add flag to avoid wasted work in bdrv_is_allocated()
Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping. In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero. Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.
This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 05:46:58 +02:00
|
|
|
NULL);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
return !!(ret & BDRV_BLOCK_ALLOCATED);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
|
|
|
|
*
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
* Return true if (a prefix of) the given range is allocated in any image
|
|
|
|
* between BASE and TOP (inclusive). BASE can be NULL to check if the given
|
|
|
|
* offset is allocated in any image of the chain. Return false otherwise,
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
* or negative errno on failure.
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
* 'pnum' is set to the number of bytes (including and immediately
|
|
|
|
* following the specified offset) that are known to be in the same
|
|
|
|
* allocated/unallocated state. Note that a subsequent call starting
|
|
|
|
* at 'offset + *pnum' may return the same allocation status (in other
|
|
|
|
* words, the result is not necessarily the maximum possible range);
|
|
|
|
* but 'pnum' will only be 0 when end of file is reached.
|
2015-04-28 15:27:52 +02:00
|
|
|
*
|
|
|
|
*/
|
|
|
|
int bdrv_is_allocated_above(BlockDriverState *top,
|
|
|
|
BlockDriverState *base,
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
int64_t offset, int64_t bytes, int64_t *pnum)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
BlockDriverState *intermediate;
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
int ret;
|
|
|
|
int64_t n = bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
intermediate = top;
|
|
|
|
while (intermediate && intermediate != base) {
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
int64_t pnum_inter;
|
2017-07-07 14:44:58 +02:00
|
|
|
int64_t size_inter;
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
block: Make bdrv_is_allocated() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned
on input and that *pnum is sector-aligned on return to the caller,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, this code adds usages like
DIV_ROUND_UP(,BDRV_SECTOR_SIZE) to callers that still want aligned
values, where the call might reasonbly give non-aligned results
in the future; on the other hand, no rounding is needed for callers
that should just continue to work with byte alignment.
For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_is_allocated(). But
some code, particularly bdrv_commit(), gets a lot simpler because it
no longer has to mess with sectors; also, it is now possible to pass
NULL if the caller does not care how much of the image is allocated
beyond the initial offset. Leave comments where we can further
simplify once a later patch eliminates the need for sector-aligned
requests through bdrv_is_allocated().
For ease of review, bdrv_is_allocated_above() will be tackled
separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:57 +02:00
|
|
|
}
|
|
|
|
if (ret) {
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
*pnum = pnum_inter;
|
2015-04-28 15:27:52 +02:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
size_inter = bdrv_getlength(intermediate);
|
2017-07-07 14:44:58 +02:00
|
|
|
if (size_inter < 0) {
|
|
|
|
return size_inter;
|
|
|
|
}
|
block: Make bdrv_is_allocated_above() byte-based
We are gradually moving away from sector-based interfaces, towards
byte-based. In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.
Changing the signature of the function to use int64_t *pnum ensures
that the compiler enforces that all callers are updated. For now,
the io.c layer still assert()s that all callers are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status. Therefore, for the most part this patch is just the
addition of scaling at the callers followed by inverse scaling at
bdrv_is_allocated(). But some code, particularly stream_run(),
gets a lot simpler because it no longer has to mess with sectors.
Leave comments where we can further simplify by switching to
byte-based iterations, once later patches eliminate the need for
sector-aligned operations.
For ease of review, bdrv_is_allocated() was tackled separately.
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-07-07 14:44:59 +02:00
|
|
|
if (n > pnum_inter &&
|
|
|
|
(intermediate == top || offset + pnum_inter < size_inter)) {
|
|
|
|
n = pnum_inter;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2015-06-17 14:55:21 +02:00
|
|
|
intermediate = backing_bs(intermediate);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
*pnum = n;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-06-09 16:24:44 +02:00
|
|
|
typedef struct BdrvVmstateCo {
|
|
|
|
BlockDriverState *bs;
|
|
|
|
QEMUIOVector *qiov;
|
|
|
|
int64_t pos;
|
|
|
|
bool is_read;
|
|
|
|
int ret;
|
|
|
|
} BdrvVmstateCo;
|
|
|
|
|
|
|
|
static int coroutine_fn
|
|
|
|
bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
|
|
|
|
bool is_read)
|
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
2017-05-22 15:57:01 +02:00
|
|
|
int ret = -ENOTSUP;
|
|
|
|
|
|
|
|
bdrv_inc_in_flight(bs);
|
2016-06-09 16:24:44 +02:00
|
|
|
|
|
|
|
if (!drv) {
|
2017-05-22 15:57:01 +02:00
|
|
|
ret = -ENOMEDIUM;
|
2016-06-09 16:24:44 +02:00
|
|
|
} else if (drv->bdrv_load_vmstate) {
|
2017-05-22 15:57:01 +02:00
|
|
|
if (is_read) {
|
|
|
|
ret = drv->bdrv_load_vmstate(bs, qiov, pos);
|
|
|
|
} else {
|
|
|
|
ret = drv->bdrv_save_vmstate(bs, qiov, pos);
|
|
|
|
}
|
2016-06-09 16:24:44 +02:00
|
|
|
} else if (bs->file) {
|
2017-05-22 15:57:01 +02:00
|
|
|
ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read);
|
2016-06-09 16:24:44 +02:00
|
|
|
}
|
|
|
|
|
2017-05-22 15:57:01 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
|
|
|
return ret;
|
2016-06-09 16:24:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
|
|
|
|
{
|
|
|
|
BdrvVmstateCo *co = opaque;
|
|
|
|
co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int
|
|
|
|
bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
|
|
|
|
bool is_read)
|
|
|
|
{
|
|
|
|
if (qemu_in_coroutine()) {
|
|
|
|
return bdrv_co_rw_vmstate(bs, qiov, pos, is_read);
|
|
|
|
} else {
|
|
|
|
BdrvVmstateCo data = {
|
|
|
|
.bs = bs,
|
|
|
|
.qiov = qiov,
|
|
|
|
.pos = pos,
|
|
|
|
.is_read = is_read,
|
|
|
|
.ret = -EINPROGRESS,
|
|
|
|
};
|
coroutine: move entry argument to qemu_coroutine_create
In practice the entry argument is always known at creation time, and
it is confusing that sometimes qemu_coroutine_enter is used with a
non-NULL argument to re-enter a coroutine (this happens in
block/sheepdog.c and tests/test-coroutine.c). So pass the opaque value
at creation time, for consistency with e.g. aio_bh_new.
Mostly done with the following semantic patch:
@ entry1 @
expression entry, arg, co;
@@
- co = qemu_coroutine_create(entry);
+ co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry2 @
expression entry, arg;
identifier co;
@@
- Coroutine *co = qemu_coroutine_create(entry);
+ Coroutine *co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry3 @
expression entry, arg;
@@
- qemu_coroutine_enter(qemu_coroutine_create(entry), arg);
+ qemu_coroutine_enter(qemu_coroutine_create(entry, arg));
@ reentry @
expression co;
@@
- qemu_coroutine_enter(co, NULL);
+ qemu_coroutine_enter(co);
except for the aforementioned few places where the semantic patch
stumbled (as expected) and for test_co_queue, which would otherwise
produce an uninitialized variable warning.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-07-04 19:10:01 +02:00
|
|
|
Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data);
|
2016-06-09 16:24:44 +02:00
|
|
|
|
block: Use bdrv_coroutine_enter to start I/O coroutines
BDRV_POLL_WHILE waits for the started I/O by releasing bs's ctx then polling
the main context, which relies on the yielded coroutine continuing on bs->ctx
before notifying qemu_aio_context with bdrv_wakeup().
Thus, using qemu_coroutine_enter to start I/O is wrong because if the coroutine
is entered from main loop, co->ctx will be qemu_aio_context, as a result of the
"release, poll, acquire" loop of BDRV_POLL_WHILE, race conditions happen when
both main thread and the iothread access the same BDS:
main loop iothread
-----------------------------------------------------------------------
blockdev_snapshot
aio_context_acquire(bs->ctx)
virtio_scsi_data_plane_handle_cmd
bdrv_drained_begin(bs->ctx)
bdrv_flush(bs)
bdrv_co_flush(bs) aio_context_acquire(bs->ctx).enter
...
qemu_coroutine_yield(co)
BDRV_POLL_WHILE()
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx).return
...
aio_co_wake(co)
aio_poll(qemu_aio_context) ...
co_schedule_bh_cb() ...
qemu_coroutine_enter(co) ...
/* (A) bdrv_co_flush(bs) /* (B) I/O on bs */
continues... */
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx)
Note that in above case, bdrv_drained_begin() doesn't do the "release,
poll, acquire" in BDRV_POLL_WHILE, because bs->in_flight == 0.
Fix this by using bdrv_coroutine_enter and enter coroutine in the right
context.
iotests 109 output is updated because the coroutine reenter flow during
mirror job complete is different (now through co_queue_wakeup, instead
of the unconditional qemu_coroutine_switch before), making the end job
len different.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
2017-04-10 14:20:17 +02:00
|
|
|
bdrv_coroutine_enter(bs, co);
|
2017-05-22 15:57:02 +02:00
|
|
|
BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS);
|
2016-06-09 16:24:44 +02:00
|
|
|
return data.ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
|
|
|
|
int64_t pos, int size)
|
|
|
|
{
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = (void *) buf,
|
|
|
|
.iov_len = size,
|
|
|
|
};
|
2016-06-10 17:57:26 +02:00
|
|
|
int ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-06-10 17:57:26 +02:00
|
|
|
|
|
|
|
ret = bdrv_writev_vmstate(bs, &qiov, pos);
|
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
return size;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
|
|
|
|
{
|
2016-06-09 16:24:44 +02:00
|
|
|
return bdrv_rw_vmstate(bs, qiov, pos, false);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
|
|
|
|
int64_t pos, int size)
|
2016-06-09 16:50:16 +02:00
|
|
|
{
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
struct iovec iov = {
|
|
|
|
.iov_base = buf,
|
|
|
|
.iov_len = size,
|
|
|
|
};
|
2016-06-10 17:57:26 +02:00
|
|
|
int ret;
|
2016-06-09 16:50:16 +02:00
|
|
|
|
|
|
|
qemu_iovec_init_external(&qiov, &iov, 1);
|
2016-06-10 17:57:26 +02:00
|
|
|
ret = bdrv_readv_vmstate(bs, &qiov, pos);
|
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
return size;
|
2016-06-09 16:50:16 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2016-06-09 16:24:44 +02:00
|
|
|
return bdrv_rw_vmstate(bs, qiov, pos, true);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/**************************************************************/
|
|
|
|
/* async I/Os */
|
|
|
|
|
|
|
|
void bdrv_aio_cancel(BlockAIOCB *acb)
|
|
|
|
{
|
|
|
|
qemu_aio_ref(acb);
|
|
|
|
bdrv_aio_cancel_async(acb);
|
|
|
|
while (acb->refcnt > 1) {
|
|
|
|
if (acb->aiocb_info->get_aio_context) {
|
|
|
|
aio_poll(acb->aiocb_info->get_aio_context(acb), true);
|
|
|
|
} else if (acb->bs) {
|
2017-02-13 14:52:29 +01:00
|
|
|
/* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
|
|
|
|
* assert that we're not using an I/O thread. Thread-safe
|
|
|
|
* code should use bdrv_aio_cancel_async exclusively.
|
|
|
|
*/
|
|
|
|
assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context());
|
2015-04-28 15:27:52 +02:00
|
|
|
aio_poll(bdrv_get_aio_context(acb->bs), true);
|
|
|
|
} else {
|
|
|
|
abort();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
qemu_aio_unref(acb);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Async version of aio cancel. The caller is not blocked if the acb implements
|
|
|
|
* cancel_async, otherwise we do nothing and let the request normally complete.
|
|
|
|
* In either case the completion callback must be called. */
|
|
|
|
void bdrv_aio_cancel_async(BlockAIOCB *acb)
|
|
|
|
{
|
|
|
|
if (acb->aiocb_info->cancel_async) {
|
|
|
|
acb->aiocb_info->cancel_async(acb);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**************************************************************/
|
|
|
|
/* Coroutine block device emulation */
|
|
|
|
|
2016-05-30 15:00:40 +02:00
|
|
|
typedef struct FlushCo {
|
|
|
|
BlockDriverState *bs;
|
|
|
|
int ret;
|
|
|
|
} FlushCo;
|
|
|
|
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
static void coroutine_fn bdrv_flush_co_entry(void *opaque)
|
|
|
|
{
|
2016-05-30 15:00:40 +02:00
|
|
|
FlushCo *rwco = opaque;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
rwco->ret = bdrv_co_flush(rwco->bs);
|
|
|
|
}
|
|
|
|
|
|
|
|
int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
|
|
|
|
{
|
2017-04-10 15:00:50 +02:00
|
|
|
int current_gen;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
bdrv_inc_in_flight(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-04-26 04:03:49 +02:00
|
|
|
if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
|
2015-06-23 12:44:57 +02:00
|
|
|
bdrv_is_sg(bs)) {
|
2017-04-10 15:00:50 +02:00
|
|
|
goto early_exit;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_lock(&bs->reqs_lock);
|
2017-06-05 14:39:01 +02:00
|
|
|
current_gen = atomic_read(&bs->write_gen);
|
2016-07-18 21:39:52 +02:00
|
|
|
|
|
|
|
/* Wait until any previous flushes are completed */
|
2016-10-27 12:48:52 +02:00
|
|
|
while (bs->active_flush_req) {
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock);
|
2016-07-18 21:39:52 +02:00
|
|
|
}
|
|
|
|
|
2017-06-05 14:39:02 +02:00
|
|
|
/* Flushes reach this point in nondecreasing current_gen order. */
|
2016-10-27 12:48:52 +02:00
|
|
|
bs->active_flush_req = true;
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_unlock(&bs->reqs_lock);
|
2016-07-18 21:39:52 +02:00
|
|
|
|
2016-03-14 08:44:53 +01:00
|
|
|
/* Write back all layers by calling one driver function */
|
|
|
|
if (bs->drv->bdrv_co_flush) {
|
|
|
|
ret = bs->drv->bdrv_co_flush(bs);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
/* Write back cached data to the OS even with cache=unsafe */
|
|
|
|
BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
|
|
|
|
if (bs->drv->bdrv_co_flush_to_os) {
|
|
|
|
ret = bs->drv->bdrv_co_flush_to_os(bs);
|
|
|
|
if (ret < 0) {
|
2015-11-09 11:16:47 +01:00
|
|
|
goto out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* But don't actually force it to the disk with cache=unsafe */
|
|
|
|
if (bs->open_flags & BDRV_O_NO_FLUSH) {
|
|
|
|
goto flush_parent;
|
|
|
|
}
|
|
|
|
|
2016-07-18 21:39:52 +02:00
|
|
|
/* Check if we really need to flush anything */
|
|
|
|
if (bs->flushed_gen == current_gen) {
|
|
|
|
goto flush_parent;
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!bs->drv) {
|
|
|
|
/* bs->drv->bdrv_co_flush() might have ejected the BDS
|
|
|
|
* (even in case of apparent success) */
|
|
|
|
ret = -ENOMEDIUM;
|
|
|
|
goto out;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
if (bs->drv->bdrv_co_flush_to_disk) {
|
|
|
|
ret = bs->drv->bdrv_co_flush_to_disk(bs);
|
|
|
|
} else if (bs->drv->bdrv_aio_flush) {
|
|
|
|
BlockAIOCB *acb;
|
|
|
|
CoroutineIOCompletion co = {
|
|
|
|
.coroutine = qemu_coroutine_self(),
|
|
|
|
};
|
|
|
|
|
|
|
|
acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
|
|
|
|
if (acb == NULL) {
|
|
|
|
ret = -EIO;
|
|
|
|
} else {
|
|
|
|
qemu_coroutine_yield();
|
|
|
|
ret = co.ret;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Some block drivers always operate in either writethrough or unsafe
|
|
|
|
* mode and don't support bdrv_flush therefore. Usually qemu doesn't
|
|
|
|
* know how the server works (because the behaviour is hardcoded or
|
|
|
|
* depends on server-side configuration), so we can't ensure that
|
|
|
|
* everything is safe on disk. Returning an error doesn't work because
|
|
|
|
* that would break guests even if the server operates in writethrough
|
|
|
|
* mode.
|
|
|
|
*
|
|
|
|
* Let's hope the user knows what he's doing.
|
|
|
|
*/
|
|
|
|
ret = 0;
|
|
|
|
}
|
2016-07-18 21:39:52 +02:00
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
2015-11-09 11:16:47 +01:00
|
|
|
goto out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
|
|
|
|
* in the case of cache=unsafe, so there are no useless flushes.
|
|
|
|
*/
|
|
|
|
flush_parent:
|
2015-11-09 11:16:47 +01:00
|
|
|
ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
|
|
|
|
out:
|
2016-07-18 21:39:52 +02:00
|
|
|
/* Notify any pending flushes that we have completed */
|
2016-11-05 00:03:15 +01:00
|
|
|
if (ret == 0) {
|
|
|
|
bs->flushed_gen = current_gen;
|
|
|
|
}
|
2017-06-05 14:39:02 +02:00
|
|
|
|
|
|
|
qemu_co_mutex_lock(&bs->reqs_lock);
|
2016-10-27 12:48:52 +02:00
|
|
|
bs->active_flush_req = false;
|
2016-08-17 20:06:54 +02:00
|
|
|
/* Return value is ignored - it's ok if wait queue is empty */
|
|
|
|
qemu_co_queue_next(&bs->flush_queue);
|
2017-06-05 14:39:02 +02:00
|
|
|
qemu_co_mutex_unlock(&bs->reqs_lock);
|
2016-07-18 21:39:52 +02:00
|
|
|
|
2017-04-10 15:00:50 +02:00
|
|
|
early_exit:
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2015-11-09 11:16:47 +01:00
|
|
|
return ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int bdrv_flush(BlockDriverState *bs)
|
|
|
|
{
|
|
|
|
Coroutine *co;
|
2016-05-30 15:00:40 +02:00
|
|
|
FlushCo flush_co = {
|
2015-04-28 15:27:52 +02:00
|
|
|
.bs = bs,
|
|
|
|
.ret = NOT_DONE,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (qemu_in_coroutine()) {
|
|
|
|
/* Fast-path if already in coroutine context */
|
2016-05-30 15:00:40 +02:00
|
|
|
bdrv_flush_co_entry(&flush_co);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
coroutine: move entry argument to qemu_coroutine_create
In practice the entry argument is always known at creation time, and
it is confusing that sometimes qemu_coroutine_enter is used with a
non-NULL argument to re-enter a coroutine (this happens in
block/sheepdog.c and tests/test-coroutine.c). So pass the opaque value
at creation time, for consistency with e.g. aio_bh_new.
Mostly done with the following semantic patch:
@ entry1 @
expression entry, arg, co;
@@
- co = qemu_coroutine_create(entry);
+ co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry2 @
expression entry, arg;
identifier co;
@@
- Coroutine *co = qemu_coroutine_create(entry);
+ Coroutine *co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry3 @
expression entry, arg;
@@
- qemu_coroutine_enter(qemu_coroutine_create(entry), arg);
+ qemu_coroutine_enter(qemu_coroutine_create(entry, arg));
@ reentry @
expression co;
@@
- qemu_coroutine_enter(co, NULL);
+ qemu_coroutine_enter(co);
except for the aforementioned few places where the semantic patch
stumbled (as expected) and for test_co_queue, which would otherwise
produce an uninitialized variable warning.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-07-04 19:10:01 +02:00
|
|
|
co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co);
|
block: Use bdrv_coroutine_enter to start I/O coroutines
BDRV_POLL_WHILE waits for the started I/O by releasing bs's ctx then polling
the main context, which relies on the yielded coroutine continuing on bs->ctx
before notifying qemu_aio_context with bdrv_wakeup().
Thus, using qemu_coroutine_enter to start I/O is wrong because if the coroutine
is entered from main loop, co->ctx will be qemu_aio_context, as a result of the
"release, poll, acquire" loop of BDRV_POLL_WHILE, race conditions happen when
both main thread and the iothread access the same BDS:
main loop iothread
-----------------------------------------------------------------------
blockdev_snapshot
aio_context_acquire(bs->ctx)
virtio_scsi_data_plane_handle_cmd
bdrv_drained_begin(bs->ctx)
bdrv_flush(bs)
bdrv_co_flush(bs) aio_context_acquire(bs->ctx).enter
...
qemu_coroutine_yield(co)
BDRV_POLL_WHILE()
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx).return
...
aio_co_wake(co)
aio_poll(qemu_aio_context) ...
co_schedule_bh_cb() ...
qemu_coroutine_enter(co) ...
/* (A) bdrv_co_flush(bs) /* (B) I/O on bs */
continues... */
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx)
Note that in above case, bdrv_drained_begin() doesn't do the "release,
poll, acquire" in BDRV_POLL_WHILE, because bs->in_flight == 0.
Fix this by using bdrv_coroutine_enter and enter coroutine in the right
context.
iotests 109 output is updated because the coroutine reenter flow during
mirror job complete is different (now through co_queue_wakeup, instead
of the unconditional qemu_coroutine_switch before), making the end job
len different.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
2017-04-10 14:20:17 +02:00
|
|
|
bdrv_coroutine_enter(bs, co);
|
2016-10-27 12:48:55 +02:00
|
|
|
BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-05-30 15:00:40 +02:00
|
|
|
return flush_co.ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
typedef struct DiscardCo {
|
|
|
|
BlockDriverState *bs;
|
2016-07-16 01:22:51 +02:00
|
|
|
int64_t offset;
|
2017-06-09 12:18:08 +02:00
|
|
|
int bytes;
|
2015-04-28 15:27:52 +02:00
|
|
|
int ret;
|
|
|
|
} DiscardCo;
|
2016-07-16 01:22:51 +02:00
|
|
|
static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
DiscardCo *rwco = opaque;
|
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
rwco->ret = bdrv_co_pdiscard(rwco->bs, rwco->offset, rwco->bytes);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-07-16 01:22:50 +02:00
|
|
|
int coroutine_fn bdrv_co_pdiscard(BlockDriverState *bs, int64_t offset,
|
2017-06-09 12:18:08 +02:00
|
|
|
int bytes)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
2015-11-09 11:16:48 +01:00
|
|
|
BdrvTrackedRequest req;
|
2016-07-16 01:22:50 +02:00
|
|
|
int max_pdiscard, ret;
|
block: Pass unaligned discard requests to drivers
Discard is advisory, so rounding the requests to alignment
boundaries is never semantically wrong from the data that
the guest sees. But at least the Dell Equallogic iSCSI SANs
has an interesting property that its advertised discard
alignment is 15M, yet documents that discarding a sequence
of 1M slices will eventually result in the 15M page being
marked as discarded, and it is possible to observe which
pages have been discarded.
Between commits 9f1963b and b8d0a980, we converted the block
layer to a byte-based interface that ultimately ignores any
unaligned head or tail based on the driver's advertised
discard granularity, which means that qemu 2.7 refuses to
pass any discard request smaller than 15M down to the Dell
Equallogic hardware. This is a slight regression in behavior
compared to earlier qemu, where a guest executing discards
in power-of-2 chunks used to be able to get every page
discarded, but is now left with various pages still allocated
because the guest requests did not align with the hardware's
15M pages.
Since the SCSI specification says nothing about a minimum
discard granularity, and only documents the preferred
alignment, it is best if the block layer gives the driver
every bit of information about discard requests, rather than
rounding it to alignment boundaries early.
Rework the block layer discard algorithm to mirror the write
zero algorithm: always peel off any unaligned head or tail
and manage that in isolation, then do the bulk of the request
on an aligned boundary. The fallback when the driver returns
-ENOTSUP for an unaligned request is to silently ignore that
portion of the discard request; but for devices that can pass
the partial request all the way down to hardware, this can
result in the hardware coalescing requests and discarding
aligned pages after all.
Reported by: Peter Lieven <pl@kamp.de>
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-11-17 21:13:58 +01:00
|
|
|
int head, tail, align;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
if (!bs->drv) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
|
2017-06-28 14:05:10 +02:00
|
|
|
if (bdrv_has_readonly_bitmaps(bs)) {
|
|
|
|
return -EPERM;
|
|
|
|
}
|
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
ret = bdrv_check_byte_request(bs, offset, bytes);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (ret < 0) {
|
|
|
|
return ret;
|
|
|
|
} else if (bs->read_only) {
|
2015-05-07 17:45:48 +02:00
|
|
|
return -EPERM;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2016-01-13 15:56:06 +01:00
|
|
|
assert(!(bs->open_flags & BDRV_O_INACTIVE));
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
/* Do nothing if disabled. */
|
|
|
|
if (!(bs->open_flags & BDRV_O_UNMAP)) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-07-16 01:23:06 +02:00
|
|
|
if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) {
|
2015-04-28 15:27:52 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
block: Pass unaligned discard requests to drivers
Discard is advisory, so rounding the requests to alignment
boundaries is never semantically wrong from the data that
the guest sees. But at least the Dell Equallogic iSCSI SANs
has an interesting property that its advertised discard
alignment is 15M, yet documents that discarding a sequence
of 1M slices will eventually result in the 15M page being
marked as discarded, and it is possible to observe which
pages have been discarded.
Between commits 9f1963b and b8d0a980, we converted the block
layer to a byte-based interface that ultimately ignores any
unaligned head or tail based on the driver's advertised
discard granularity, which means that qemu 2.7 refuses to
pass any discard request smaller than 15M down to the Dell
Equallogic hardware. This is a slight regression in behavior
compared to earlier qemu, where a guest executing discards
in power-of-2 chunks used to be able to get every page
discarded, but is now left with various pages still allocated
because the guest requests did not align with the hardware's
15M pages.
Since the SCSI specification says nothing about a minimum
discard granularity, and only documents the preferred
alignment, it is best if the block layer gives the driver
every bit of information about discard requests, rather than
rounding it to alignment boundaries early.
Rework the block layer discard algorithm to mirror the write
zero algorithm: always peel off any unaligned head or tail
and manage that in isolation, then do the bulk of the request
on an aligned boundary. The fallback when the driver returns
-ENOTSUP for an unaligned request is to silently ignore that
portion of the discard request; but for devices that can pass
the partial request all the way down to hardware, this can
result in the hardware coalescing requests and discarding
aligned pages after all.
Reported by: Peter Lieven <pl@kamp.de>
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-11-17 21:13:58 +01:00
|
|
|
/* Discard is advisory, but some devices track and coalesce
|
|
|
|
* unaligned requests, so we must pass everything down rather than
|
|
|
|
* round here. Still, most devices will just silently ignore
|
|
|
|
* unaligned requests (by returning -ENOTSUP), so we must fragment
|
|
|
|
* the request accordingly. */
|
2016-07-16 01:23:06 +02:00
|
|
|
align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment);
|
2016-07-21 21:34:48 +02:00
|
|
|
assert(align % bs->bl.request_alignment == 0);
|
|
|
|
head = offset % align;
|
2017-06-09 12:18:08 +02:00
|
|
|
tail = (offset + bytes) % align;
|
2016-07-16 01:22:50 +02:00
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
2017-06-09 12:18:08 +02:00
|
|
|
tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD);
|
2015-06-08 07:56:10 +02:00
|
|
|
|
2016-06-16 18:09:41 +02:00
|
|
|
ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
|
|
|
|
if (ret < 0) {
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-07-16 01:22:50 +02:00
|
|
|
max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX),
|
|
|
|
align);
|
block: Pass unaligned discard requests to drivers
Discard is advisory, so rounding the requests to alignment
boundaries is never semantically wrong from the data that
the guest sees. But at least the Dell Equallogic iSCSI SANs
has an interesting property that its advertised discard
alignment is 15M, yet documents that discarding a sequence
of 1M slices will eventually result in the 15M page being
marked as discarded, and it is possible to observe which
pages have been discarded.
Between commits 9f1963b and b8d0a980, we converted the block
layer to a byte-based interface that ultimately ignores any
unaligned head or tail based on the driver's advertised
discard granularity, which means that qemu 2.7 refuses to
pass any discard request smaller than 15M down to the Dell
Equallogic hardware. This is a slight regression in behavior
compared to earlier qemu, where a guest executing discards
in power-of-2 chunks used to be able to get every page
discarded, but is now left with various pages still allocated
because the guest requests did not align with the hardware's
15M pages.
Since the SCSI specification says nothing about a minimum
discard granularity, and only documents the preferred
alignment, it is best if the block layer gives the driver
every bit of information about discard requests, rather than
rounding it to alignment boundaries early.
Rework the block layer discard algorithm to mirror the write
zero algorithm: always peel off any unaligned head or tail
and manage that in isolation, then do the bulk of the request
on an aligned boundary. The fallback when the driver returns
-ENOTSUP for an unaligned request is to silently ignore that
portion of the discard request; but for devices that can pass
the partial request all the way down to hardware, this can
result in the hardware coalescing requests and discarding
aligned pages after all.
Reported by: Peter Lieven <pl@kamp.de>
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-11-17 21:13:58 +01:00
|
|
|
assert(max_pdiscard >= bs->bl.request_alignment);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
while (bytes > 0) {
|
|
|
|
int num = bytes;
|
block: Pass unaligned discard requests to drivers
Discard is advisory, so rounding the requests to alignment
boundaries is never semantically wrong from the data that
the guest sees. But at least the Dell Equallogic iSCSI SANs
has an interesting property that its advertised discard
alignment is 15M, yet documents that discarding a sequence
of 1M slices will eventually result in the 15M page being
marked as discarded, and it is possible to observe which
pages have been discarded.
Between commits 9f1963b and b8d0a980, we converted the block
layer to a byte-based interface that ultimately ignores any
unaligned head or tail based on the driver's advertised
discard granularity, which means that qemu 2.7 refuses to
pass any discard request smaller than 15M down to the Dell
Equallogic hardware. This is a slight regression in behavior
compared to earlier qemu, where a guest executing discards
in power-of-2 chunks used to be able to get every page
discarded, but is now left with various pages still allocated
because the guest requests did not align with the hardware's
15M pages.
Since the SCSI specification says nothing about a minimum
discard granularity, and only documents the preferred
alignment, it is best if the block layer gives the driver
every bit of information about discard requests, rather than
rounding it to alignment boundaries early.
Rework the block layer discard algorithm to mirror the write
zero algorithm: always peel off any unaligned head or tail
and manage that in isolation, then do the bulk of the request
on an aligned boundary. The fallback when the driver returns
-ENOTSUP for an unaligned request is to silently ignore that
portion of the discard request; but for devices that can pass
the partial request all the way down to hardware, this can
result in the hardware coalescing requests and discarding
aligned pages after all.
Reported by: Peter Lieven <pl@kamp.de>
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-11-17 21:13:58 +01:00
|
|
|
|
|
|
|
if (head) {
|
|
|
|
/* Make small requests to get to alignment boundaries. */
|
2017-06-09 12:18:08 +02:00
|
|
|
num = MIN(bytes, align - head);
|
block: Pass unaligned discard requests to drivers
Discard is advisory, so rounding the requests to alignment
boundaries is never semantically wrong from the data that
the guest sees. But at least the Dell Equallogic iSCSI SANs
has an interesting property that its advertised discard
alignment is 15M, yet documents that discarding a sequence
of 1M slices will eventually result in the 15M page being
marked as discarded, and it is possible to observe which
pages have been discarded.
Between commits 9f1963b and b8d0a980, we converted the block
layer to a byte-based interface that ultimately ignores any
unaligned head or tail based on the driver's advertised
discard granularity, which means that qemu 2.7 refuses to
pass any discard request smaller than 15M down to the Dell
Equallogic hardware. This is a slight regression in behavior
compared to earlier qemu, where a guest executing discards
in power-of-2 chunks used to be able to get every page
discarded, but is now left with various pages still allocated
because the guest requests did not align with the hardware's
15M pages.
Since the SCSI specification says nothing about a minimum
discard granularity, and only documents the preferred
alignment, it is best if the block layer gives the driver
every bit of information about discard requests, rather than
rounding it to alignment boundaries early.
Rework the block layer discard algorithm to mirror the write
zero algorithm: always peel off any unaligned head or tail
and manage that in isolation, then do the bulk of the request
on an aligned boundary. The fallback when the driver returns
-ENOTSUP for an unaligned request is to silently ignore that
portion of the discard request; but for devices that can pass
the partial request all the way down to hardware, this can
result in the hardware coalescing requests and discarding
aligned pages after all.
Reported by: Peter Lieven <pl@kamp.de>
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-11-17 21:13:58 +01:00
|
|
|
if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) {
|
|
|
|
num %= bs->bl.request_alignment;
|
|
|
|
}
|
|
|
|
head = (head + num) % align;
|
|
|
|
assert(num < max_pdiscard);
|
|
|
|
} else if (tail) {
|
|
|
|
if (num > align) {
|
|
|
|
/* Shorten the request to the last aligned cluster. */
|
|
|
|
num -= tail;
|
|
|
|
} else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) &&
|
|
|
|
tail > bs->bl.request_alignment) {
|
|
|
|
tail %= bs->bl.request_alignment;
|
|
|
|
num -= tail;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/* limit request size */
|
|
|
|
if (num > max_pdiscard) {
|
|
|
|
num = max_pdiscard;
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2017-11-10 21:31:09 +01:00
|
|
|
if (!bs->drv) {
|
|
|
|
ret = -ENOMEDIUM;
|
|
|
|
goto out;
|
|
|
|
}
|
2016-07-16 01:22:58 +02:00
|
|
|
if (bs->drv->bdrv_co_pdiscard) {
|
|
|
|
ret = bs->drv->bdrv_co_pdiscard(bs, offset, num);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
|
|
|
BlockAIOCB *acb;
|
|
|
|
CoroutineIOCompletion co = {
|
|
|
|
.coroutine = qemu_coroutine_self(),
|
|
|
|
};
|
|
|
|
|
2016-07-16 01:22:57 +02:00
|
|
|
acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num,
|
|
|
|
bdrv_co_io_em_complete, &co);
|
2015-04-28 15:27:52 +02:00
|
|
|
if (acb == NULL) {
|
2015-11-09 11:16:48 +01:00
|
|
|
ret = -EIO;
|
|
|
|
goto out;
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
|
|
|
qemu_coroutine_yield();
|
|
|
|
ret = co.ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (ret && ret != -ENOTSUP) {
|
2015-11-09 11:16:48 +01:00
|
|
|
goto out;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2016-07-16 01:22:50 +02:00
|
|
|
offset += num;
|
2017-06-09 12:18:08 +02:00
|
|
|
bytes -= num;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
2015-11-09 11:16:48 +01:00
|
|
|
ret = 0;
|
|
|
|
out:
|
2017-06-05 14:39:01 +02:00
|
|
|
atomic_inc(&bs->write_gen);
|
2017-09-25 16:55:25 +02:00
|
|
|
bdrv_set_dirty(bs, req.offset, req.bytes);
|
2015-11-09 11:16:48 +01:00
|
|
|
tracked_request_end(&req);
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2015-11-09 11:16:48 +01:00
|
|
|
return ret;
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
2017-06-09 12:18:08 +02:00
|
|
|
int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
Coroutine *co;
|
|
|
|
DiscardCo rwco = {
|
|
|
|
.bs = bs,
|
2016-07-16 01:22:51 +02:00
|
|
|
.offset = offset,
|
2017-06-09 12:18:08 +02:00
|
|
|
.bytes = bytes,
|
2015-04-28 15:27:52 +02:00
|
|
|
.ret = NOT_DONE,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (qemu_in_coroutine()) {
|
|
|
|
/* Fast-path if already in coroutine context */
|
2016-07-16 01:22:51 +02:00
|
|
|
bdrv_pdiscard_co_entry(&rwco);
|
2015-04-28 15:27:52 +02:00
|
|
|
} else {
|
2016-07-16 01:22:51 +02:00
|
|
|
co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco);
|
block: Use bdrv_coroutine_enter to start I/O coroutines
BDRV_POLL_WHILE waits for the started I/O by releasing bs's ctx then polling
the main context, which relies on the yielded coroutine continuing on bs->ctx
before notifying qemu_aio_context with bdrv_wakeup().
Thus, using qemu_coroutine_enter to start I/O is wrong because if the coroutine
is entered from main loop, co->ctx will be qemu_aio_context, as a result of the
"release, poll, acquire" loop of BDRV_POLL_WHILE, race conditions happen when
both main thread and the iothread access the same BDS:
main loop iothread
-----------------------------------------------------------------------
blockdev_snapshot
aio_context_acquire(bs->ctx)
virtio_scsi_data_plane_handle_cmd
bdrv_drained_begin(bs->ctx)
bdrv_flush(bs)
bdrv_co_flush(bs) aio_context_acquire(bs->ctx).enter
...
qemu_coroutine_yield(co)
BDRV_POLL_WHILE()
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx).return
...
aio_co_wake(co)
aio_poll(qemu_aio_context) ...
co_schedule_bh_cb() ...
qemu_coroutine_enter(co) ...
/* (A) bdrv_co_flush(bs) /* (B) I/O on bs */
continues... */
aio_context_release(bs->ctx)
aio_context_acquire(bs->ctx)
Note that in above case, bdrv_drained_begin() doesn't do the "release,
poll, acquire" in BDRV_POLL_WHILE, because bs->in_flight == 0.
Fix this by using bdrv_coroutine_enter and enter coroutine in the right
context.
iotests 109 output is updated because the coroutine reenter flow during
mirror job complete is different (now through co_queue_wakeup, instead
of the unconditional qemu_coroutine_switch before), making the end job
len different.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
2017-04-10 14:20:17 +02:00
|
|
|
bdrv_coroutine_enter(bs, co);
|
2016-10-27 12:48:55 +02:00
|
|
|
BDRV_POLL_WHILE(bs, rwco.ret == NOT_DONE);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return rwco.ret;
|
|
|
|
}
|
|
|
|
|
2016-10-20 12:56:14 +02:00
|
|
|
int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf)
|
2015-04-28 15:27:52 +02:00
|
|
|
{
|
|
|
|
BlockDriver *drv = bs->drv;
|
2015-11-09 11:16:51 +01:00
|
|
|
CoroutineIOCompletion co = {
|
|
|
|
.coroutine = qemu_coroutine_self(),
|
|
|
|
};
|
|
|
|
BlockAIOCB *acb;
|
2015-04-28 15:27:52 +02:00
|
|
|
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_inc_in_flight(bs);
|
2016-10-20 15:07:27 +02:00
|
|
|
if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) {
|
2015-11-09 11:16:51 +01:00
|
|
|
co.ret = -ENOTSUP;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-10-20 15:07:27 +02:00
|
|
|
if (drv->bdrv_co_ioctl) {
|
|
|
|
co.ret = drv->bdrv_co_ioctl(bs, req, buf);
|
|
|
|
} else {
|
|
|
|
acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
|
|
|
|
if (!acb) {
|
|
|
|
co.ret = -ENOTSUP;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
qemu_coroutine_yield();
|
2015-11-09 11:16:51 +01:00
|
|
|
}
|
|
|
|
out:
|
2016-10-27 12:48:52 +02:00
|
|
|
bdrv_dec_in_flight(bs);
|
2015-11-09 11:16:51 +01:00
|
|
|
return co.ret;
|
|
|
|
}
|
|
|
|
|
2015-04-28 15:27:52 +02:00
|
|
|
void *qemu_blockalign(BlockDriverState *bs, size_t size)
|
|
|
|
{
|
|
|
|
return qemu_memalign(bdrv_opt_mem_align(bs), size);
|
|
|
|
}
|
|
|
|
|
|
|
|
void *qemu_blockalign0(BlockDriverState *bs, size_t size)
|
|
|
|
{
|
|
|
|
return memset(qemu_blockalign(bs, size), 0, size);
|
|
|
|
}
|
|
|
|
|
|
|
|
void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
|
|
|
|
{
|
|
|
|
size_t align = bdrv_opt_mem_align(bs);
|
|
|
|
|
|
|
|
/* Ensure that NULL is never returned on success */
|
|
|
|
assert(align > 0);
|
|
|
|
if (size == 0) {
|
|
|
|
size = align;
|
|
|
|
}
|
|
|
|
|
|
|
|
return qemu_try_memalign(align, size);
|
|
|
|
}
|
|
|
|
|
|
|
|
void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
|
|
|
|
{
|
|
|
|
void *mem = qemu_try_blockalign(bs, size);
|
|
|
|
|
|
|
|
if (mem) {
|
|
|
|
memset(mem, 0, size);
|
|
|
|
}
|
|
|
|
|
|
|
|
return mem;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check if all memory in this vector is sector aligned.
|
|
|
|
*/
|
|
|
|
bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
|
|
|
|
{
|
|
|
|
int i;
|
2015-05-12 16:30:55 +02:00
|
|
|
size_t alignment = bdrv_min_mem_align(bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
|
|
|
|
for (i = 0; i < qiov->niov; i++) {
|
|
|
|
if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (qiov->iov[i].iov_len % alignment) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_add_before_write_notifier(BlockDriverState *bs,
|
|
|
|
NotifierWithReturn *notifier)
|
|
|
|
{
|
|
|
|
notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_io_plug(BlockDriverState *bs)
|
|
|
|
{
|
2016-04-07 18:33:34 +02:00
|
|
|
BdrvChild *child;
|
|
|
|
|
|
|
|
QLIST_FOREACH(child, &bs->children, next) {
|
|
|
|
bdrv_io_plug(child->bs);
|
|
|
|
}
|
|
|
|
|
2017-06-05 14:38:55 +02:00
|
|
|
if (atomic_fetch_inc(&bs->io_plugged) == 0) {
|
2016-04-07 18:33:34 +02:00
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
if (drv && drv->bdrv_io_plug) {
|
|
|
|
drv->bdrv_io_plug(bs);
|
|
|
|
}
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_io_unplug(BlockDriverState *bs)
|
|
|
|
{
|
2016-04-07 18:33:34 +02:00
|
|
|
BdrvChild *child;
|
|
|
|
|
|
|
|
assert(bs->io_plugged);
|
2017-06-05 14:38:55 +02:00
|
|
|
if (atomic_fetch_dec(&bs->io_plugged) == 1) {
|
2016-04-07 18:33:34 +02:00
|
|
|
BlockDriver *drv = bs->drv;
|
|
|
|
if (drv && drv->bdrv_io_unplug) {
|
|
|
|
drv->bdrv_io_unplug(bs);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
QLIST_FOREACH(child, &bs->children, next) {
|
|
|
|
bdrv_io_unplug(child->bs);
|
2015-04-28 15:27:52 +02:00
|
|
|
}
|
|
|
|
}
|
2018-01-16 07:08:56 +01:00
|
|
|
|
|
|
|
void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size)
|
|
|
|
{
|
|
|
|
BdrvChild *child;
|
|
|
|
|
|
|
|
if (bs->drv && bs->drv->bdrv_register_buf) {
|
|
|
|
bs->drv->bdrv_register_buf(bs, host, size);
|
|
|
|
}
|
|
|
|
QLIST_FOREACH(child, &bs->children, next) {
|
|
|
|
bdrv_register_buf(child->bs, host, size);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void bdrv_unregister_buf(BlockDriverState *bs, void *host)
|
|
|
|
{
|
|
|
|
BdrvChild *child;
|
|
|
|
|
|
|
|
if (bs->drv && bs->drv->bdrv_unregister_buf) {
|
|
|
|
bs->drv->bdrv_unregister_buf(bs, host);
|
|
|
|
}
|
|
|
|
QLIST_FOREACH(child, &bs->children, next) {
|
|
|
|
bdrv_unregister_buf(child->bs, host);
|
|
|
|
}
|
|
|
|
}
|
2018-06-01 11:26:39 +02:00
|
|
|
|
|
|
|
static int coroutine_fn bdrv_co_copy_range_internal(BdrvChild *src,
|
|
|
|
uint64_t src_offset,
|
|
|
|
BdrvChild *dst,
|
|
|
|
uint64_t dst_offset,
|
|
|
|
uint64_t bytes,
|
|
|
|
BdrvRequestFlags flags,
|
|
|
|
bool recurse_src)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!src || !dst || !src->bs || !dst->bs) {
|
|
|
|
return -ENOMEDIUM;
|
|
|
|
}
|
|
|
|
ret = bdrv_check_byte_request(src->bs, src_offset, bytes);
|
|
|
|
if (ret) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes);
|
|
|
|
if (ret) {
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
if (flags & BDRV_REQ_ZERO_WRITE) {
|
|
|
|
return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!src->bs->drv->bdrv_co_copy_range_from
|
|
|
|
|| !dst->bs->drv->bdrv_co_copy_range_to
|
|
|
|
|| src->bs->encrypted || dst->bs->encrypted) {
|
|
|
|
return -ENOTSUP;
|
|
|
|
}
|
|
|
|
if (recurse_src) {
|
|
|
|
return src->bs->drv->bdrv_co_copy_range_from(src->bs,
|
|
|
|
src, src_offset,
|
|
|
|
dst, dst_offset,
|
|
|
|
bytes, flags);
|
|
|
|
} else {
|
|
|
|
return dst->bs->drv->bdrv_co_copy_range_to(dst->bs,
|
|
|
|
src, src_offset,
|
|
|
|
dst, dst_offset,
|
|
|
|
bytes, flags);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Copy range from @src to @dst.
|
|
|
|
*
|
|
|
|
* See the comment of bdrv_co_copy_range for the parameter and return value
|
|
|
|
* semantics. */
|
|
|
|
int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset,
|
|
|
|
BdrvChild *dst, uint64_t dst_offset,
|
|
|
|
uint64_t bytes, BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
|
|
|
|
bytes, flags, true);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Copy range from @src to @dst.
|
|
|
|
*
|
|
|
|
* See the comment of bdrv_co_copy_range for the parameter and return value
|
|
|
|
* semantics. */
|
|
|
|
int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset,
|
|
|
|
BdrvChild *dst, uint64_t dst_offset,
|
|
|
|
uint64_t bytes, BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
|
|
|
|
bytes, flags, false);
|
|
|
|
}
|
|
|
|
|
|
|
|
int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
|
|
|
|
BdrvChild *dst, uint64_t dst_offset,
|
|
|
|
uint64_t bytes, BdrvRequestFlags flags)
|
|
|
|
{
|
|
|
|
BdrvTrackedRequest src_req, dst_req;
|
|
|
|
BlockDriverState *src_bs = src->bs;
|
|
|
|
BlockDriverState *dst_bs = dst->bs;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
bdrv_inc_in_flight(src_bs);
|
|
|
|
bdrv_inc_in_flight(dst_bs);
|
|
|
|
tracked_request_begin(&src_req, src_bs, src_offset,
|
|
|
|
bytes, BDRV_TRACKED_READ);
|
|
|
|
tracked_request_begin(&dst_req, dst_bs, dst_offset,
|
|
|
|
bytes, BDRV_TRACKED_WRITE);
|
|
|
|
|
|
|
|
wait_serialising_requests(&src_req);
|
|
|
|
wait_serialising_requests(&dst_req);
|
|
|
|
ret = bdrv_co_copy_range_from(src, src_offset,
|
|
|
|
dst, dst_offset,
|
|
|
|
bytes, flags);
|
|
|
|
|
|
|
|
tracked_request_end(&src_req);
|
|
|
|
tracked_request_end(&dst_req);
|
|
|
|
bdrv_dec_in_flight(src_bs);
|
|
|
|
bdrv_dec_in_flight(dst_bs);
|
|
|
|
return ret;
|
|
|
|
}
|