Do not store back to the exact memory from which we just loaded.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Do not set parallel_cpus if there is only one cpu instantiated.
This will allow tcg to use serial code to implement atomics.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The fallback inline expansion for vectorized absolute value,
when the host doesn't support such an insn was flawed.
E.g. when a vector of bytes has all elements negative, mask
will be 0xffff_ffff_ffff_ffff. Subtracting mask only adds 1
to the low element instead of all elements becase -mask is 1
and not 0x0101_0101_0101_0101.
Signed-off-by: Stephen Long <steplong@quicinc.com>
Message-Id: <20200813161818.190-1-steplong@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This has no functional change.
The current function structure is:
inline QEMU_ALWAYSINLINE
store_memop() {
switch () {
...
default:
qemu_build_not_reached();
}
}
inline QEMU_ALWAYSINLINE
store_helper() {
...
if (span_two_pages_or_io) {
...
helper_ret_stb_mmu();
}
store_memop();
}
helper_ret_stb_mmu() {
store_helper();
}
Whereas GCC will generate an error at compile-time when an always_inline
function is not inlined, Clang does not. Nor does Clang prioritize the
inlining of always_inline functions. Both of these are arguably bugs.
Both `store_memop` and `store_helper` need to be inlined and allow
constant propogations to eliminate the `qemu_build_not_reached` call.
However, if the compiler instead chooses to inline helper_ret_stb_mmu
into store_helper, then store_helper is now self-recursive and the
compiler is no longer able to propagate the constant in the same way.
This does not produce at current QEMU head, but was reproducible
at v4.2.0 with `clang-10 -O2 -fexperimental-new-pass-manager`.
The inline recursion problem can be fixed solely by marking
helper_ret_stb_mmu as noinline, so the compiler does not make an
incorrect decision about which functions to inline.
In addition, extract store_helper_unaligned as a noinline subroutine
that can be shared by all of the helpers. This saves about 6k code
size in an optimized x86_64 build.
Reported-by: Shu-Chun Weng <scw@google.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
* Two build system fixes to fix some failures the CI
* One m68k QOMification patch
* Some trivial qtest patches
* Some small improvements for the Gitlab CI
-----BEGIN PGP SIGNATURE-----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=kPTJ
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2020-09-03' into staging
* Cirrus-CI improvements and fixes (compile with -Werror & fix for 1h problem)
* Two build system fixes to fix some failures the CI
* One m68k QOMification patch
* Some trivial qtest patches
* Some small improvements for the Gitlab CI
# gpg: Signature made Thu 03 Sep 2020 12:04:32 BST
# gpg: using RSA key 27B88847EEE0250118F3EAB92ED9D774FE702DB5
# gpg: issuer "thuth@redhat.com"
# gpg: Good signature from "Thomas Huth <th.huth@gmx.de>" [full]
# gpg: aka "Thomas Huth <thuth@redhat.com>" [full]
# gpg: aka "Thomas Huth <huth@tuxfamily.org>" [full]
# gpg: aka "Thomas Huth <th.huth@posteo.de>" [unknown]
# Primary key fingerprint: 27B8 8847 EEE0 2501 18F3 EAB9 2ED9 D774 FE70 2DB5
* remotes/huth-gitlab/tags/pull-request-2020-09-03:
gitlab-ci.yml: Set artifacts expiration time
gitlab-ci.yml: Run check-qtest and check-unit at the end of the fuzzer job
gitlab/travis: Rework the disabled features tests
libqtest: Rename qmp_assert_error_class() to qmp_expect_error_and_unref()
tests/qtest/ipmi-kcs: Fix assert side-effect
tests/qtest/tpm: Declare input buffers const and static
tests/qtest/ahci: Improve error handling (NEGATIVE_RETURNS)
hw/m68k: QOMify the mcf5206 system integration module
configure: Add system = 'linux' for meson when cross-compiling
meson: fix keymaps without qemu-keymap
cirrus.yml: Split FreeBSD job into two parts
cirrus.yml: Update the macOS jobs to Catalina
cirrus.yml: Compile macOS with -Werror
cirrus.yml: Compile FreeBSD with -Werror
configure: Fix atomic64 test for --enable-werror on macOS
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
* New Supermicro X11 BMC machine (Erik)
* Fixed valid access size on AST2400 SCU
* Improved robustness of the ftgmac100 model.
* New flash models in m25p80 (Igor)
* Fixed reset sequence of SDHCI/eMMC controllers
* Improved support of the AST2600 SDMC (Joel)
* Couple of SMC cleanups
-----BEGIN PGP SIGNATURE-----
iQIzBAABCAAdFiEEoPZlSPBIlev+awtgUaNDx8/77KEFAl9OQPgACgkQUaNDx8/7
7KHzsRAAmXw6963D3wIuE2Nzb1G5Zvn6nup3AsF5Xs1IZU/cLqNijiz220KslFtQ
y8KrTO/eyBmAsEjrg1f6bWwCTZsouKq/2vWPtmTx3eU4HgeJdPbkln7E1YGmMfBR
T4WJU6mNqkWfFT3WAW3IbB4qCoH3l0DRkgawYPWbdJmTs5CBtXOYCT14TijDVWQ5
p8S4QjTtfRPwG9csHJ1W93t8jadTzderefkN6Zcmf9y6iOCif6SVDFvF769hzg6e
Pzp3xxRV3ewxhSLrGdCK+fQk/IcPaLVUnh+mM3mGLk2rDQoomFXBpaz1N94rw43s
lGuIyLkUGiHbgONmlZMXj03WWQbgGqjYpDWme1rAKJSX6CRJRixucejsRFTG5Evx
odgY1MGNrdg0K8L0O1SQEx7O+URZZO68WrtrMTwLbOHErE7pWAR+h5RqzclwMr3v
0hwQxDeNjhDBj+nUwoPUjXsgfVafzeywFfKuMymnygGog5hFSWiqAFIqyxj+u6YI
HUG8kMHdLqzAgX1NWAomn2cxUEc4Q2wxDlzUgvjcvBwa6HZD+3nrjMRStHTmeVy5
yPKWmRanXH6xIUJoRd2dMEU6SrwGjmjfnKAbG3vgxJ6B5sk4BrfKOFeaCF9M2zP6
ZePWf6XrsPQY7aZgQRTexmXK83jqn73DOkavI2pM9s/6Ts61mdc=
=ZQHA
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/legoater/tags/pull-aspeed-20200901' into staging
Various fixes of Aspeed machines :
* New Supermicro X11 BMC machine (Erik)
* Fixed valid access size on AST2400 SCU
* Improved robustness of the ftgmac100 model.
* New flash models in m25p80 (Igor)
* Fixed reset sequence of SDHCI/eMMC controllers
* Improved support of the AST2600 SDMC (Joel)
* Couple of SMC cleanups
# gpg: Signature made Tue 01 Sep 2020 13:39:20 BST
# gpg: using RSA key A0F66548F04895EBFE6B0B6051A343C7CFFBECA1
# gpg: Good signature from "Cédric Le Goater <clg@kaod.org>" [undefined]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: A0F6 6548 F048 95EB FE6B 0B60 51A3 43C7 CFFB ECA1
* remotes/legoater/tags/pull-aspeed-20200901:
hw: add a number of SPI-flash's of m25p80 family
arm: aspeed: add strap define `25HZ` of AST2500
aspeed/smc: Open AHB window of the second chip of the AST2600 FMC controller
aspeed/sdmc: Simplify calculation of RAM bits
aspeed/sdmc: Allow writes to unprotected registers
aspeed/sdmc: Perform memory training
ftgmac100: Improve software reset
ftgmac100: Fix integer overflow in ftgmac100_do_tx()
ftgmac100: Check for invalid len and address before doing a DMA transfer
ftgmac100: Change interrupt status when a DMA error occurs
ftgmac100: Fix interrupt status "Packet moved to RX FIFO"
ftgmac100: Fix interrupt status "Packet transmitted on ethernet"
ftgmac100: Fix registers that can be read
aspeed/sdhci: Fix reset sequence
aspeed/smc: Fix max_slaves of the legacy SMC device
aspeed/smc: Fix MemoryRegionOps definition
hw/arm/aspeed: Add board model for Supermicro X11 BMC
aspeed/scu: Fix valid access size on AST2400
m25p80: Add support for n25q512ax3
m25p80: Return the JEDEC ID twice for mx25l25635e
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The default expiration time for artifacts seems to be very high (30 days?).
Since we only need the artifacts to pass the binaries from one stage to
the next one, we can decrease the expiration time to avoid to spam the
file server too much. Two days should be enough in case someone still wants
to have a look after the pipeline finished.
Message-Id: <20200806161546.15325-1-thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The fuzzer job finishes quite early, so we can run the unit tests and
qtests with -fsanitize=address here without extending the total test time.
Message-Id: <20200831153228.229185-1-thuth@redhat.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Let's focus on the gitlab-ci when testing the compilation with disabled
features, thus add more switches there (and while we're at it, sort them
also alphabetically). This should cover the test from the Travis CI now,
too, so that we can remove the now-redundant job from the Travis CI.
Message-Id: <20200806155306.13717-1-thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
qmp_assert_error_class() does more than just assert: it also unrefs
the @rsp argument. Rename to qmp_expect_error_and_unref() to reduce
confusion.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200902115733.1229537-1-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Fix assert side-effect reported by Coverity:
/qemu/tests/qtest/ipmi-kcs-test.c: 84 in kcs_wait_obf()
83 while (IPMI_KCS_CMDREG_GET_OBF() == 0) {
>>> CID 1432368: Incorrect expression (ASSERT_SIDE_EFFECT)
>>> Argument "--count" of g_assert() has a side effect. The containing function might work differently in a non-debug build.
84 g_assert(--count != 0);
Reported-by: Coverity (CID 1432368)
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200902080801.160652-2-philmd@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The functions using these arrays expect a "const unsigned char *"
argument, it is safe to declare these as 'static const'.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200902080909.161034-1-philmd@redhat.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Fix an error handling issue reported by Coverity:
/qemu/tests/qtest/ahci-test.c: 1452 in prepare_iso()
1444 int fd = mkstemp(cdrom_path);
>>> CID 1432375: Error handling issues (NEGATIVE_RETURNS)
>>> "fd" is passed to a parameter that cannot be negative.
1452 ret = write(fd, patt, size);
Reported-by: Coverity (CID 1432375)
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200902080552.159806-1-philmd@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The mcf5206 system integration module should be a proper device.
Let's finally QOMify it.
Signed-off-by: Thomas Huth <huth@tuxfamily.org>
Message-Id: <20200819065201.4045-1-huth@tuxfamily.org>
Meson needs the "system = xyz" line when cross-compiling. We are already
adding a "system = 'windows'" for the MinGW cross-compilation case here,
so let's add a "system = 'linux'" now for Linux hosts, too.
Message-Id: <20200823111757.72002-2-thuth@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
In case the qemu-keymap tool generating them is neither installed on the
system nor built from sources (due to xkbcommon not being available)
qemu will not find the keymaps when started directly from the build
tree,
This happens because commit ddcf607fa3 ("meson: drop keymaps symlink")
removed the symlink to the source tree, and the special handling for
install doesn't help in case we do not install qemu.
Lets fix that by simply copying over the file from the source tree as
fallback.
Reported-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Message-Id: <20200827102617.14448-1-kraxel@redhat.com>
[thuth: Rebased, changed "config_host['qemu_datadir']" to "qemu_datadir",
added Gerd's UNLINK fix to configure script]
Signed-off-by: Thomas Huth <thuth@redhat.com>
The FreeBSD jobs currently hit the 1h time limit in the Cirrus-CI.
We have to split the build targets here to make sure that the job
finishes in time again. According to the Cirrus-CI docs and some
tests that I did, it also seems like the total amount of CPUs that
can be used for FreeBSD jobs is limited to 8, so each job now only
gets 4 CPUs. That increases the compilation time of each job a little
bit, but it still seems to be better to run two jobs with 4 CPUs each
in parallel than to run two jobs with 8 CPUs sequentially.
Message-Id: <20200831154405.229706-1-thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
When looking at the CI jobs on cirrus-ci.com, it seems like the mojave-based
images have been decomissioned a while ago already, since apparently all our
jobs get automatically upgraded to catalina. So let's update our YML script
accordingly to avoid confusion.
Reviewed-by: Ed Maste <emaste@freebsd.org>
Message-Id: <20200728074405.13118-5-thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Compiler warnings currently go unnoticed in our macOS builds, since -Werror
is only enabled for Linux and MinGW builds by default. So let's enable them
here now, too.
Unfortunately, the sasl header is marked as deprecated in the macOS headers
and thus generates a lot of deprecation warnings. Thus we have to also use
-Wno-error=deprecated-declarations to be able to compile the code here.
Message-Id: <20200728074405.13118-4-thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Compiler warnings currently go unnoticed in our FreeBSD builds, since
-Werror is only enabled for Linux and MinGW builds by default. So let's
enable them here now, too.
Reviewed-by: Ed Maste <emaste@freebsd.org>
Message-Id: <20200728074405.13118-3-thuth@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
When using --enable-werror for the macOS builders in the Cirrus-CI,
the atomic64 test is currently failing, and config.log shows a bunch
of error messages like this:
config-temp/qemu-conf.c:6:7: error: implicit declaration of function
'__atomic_load_8' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
y = __atomic_load_8(&x, 0);
^
config-temp/qemu-conf.c:6:7: error: this function declaration is not a
prototype [-Werror,-Wstrict-prototypes]
Seems like these __atomic_*_8 functions are available in one of the
libraries there, so that the test links and passes there when not
using --enable-werror. But there does not seem to be a valid prototype
for them in any of the header files, so that the test fails when using
--enable-werror.
Fix it by using the "official" built-in functions instead (see e.g.
https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html).
We are not using the *_8 variants in QEMU anyway.
Suggested-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Christian Schoenebeck <qemu_oss@crudebyte.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200728074405.13118-2-thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Commit 480324ec8d "docs/qdev-device-use: Clean up the sentences
related to -usbdevice" deleted the information on syntax that no
longer works. Unfortunately, the resulting text suggests you can
configure USB block devices with -drive, and USB network devices with
-net, which is misleading.
Instead of rephrasing the now misleading text, I'm putting the
information back, and just make clear it's about old versions of QEMU.
Cc: Thomas Huth <thuth@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200806081147.3123652-4-armbru@redhat.com>
Commit e2ae6159de "virtio-serial: report frontend connection state via
monitor" neglected to document the new event is rate-limited. Fix
that.
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200806081147.3123652-3-armbru@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Commit 4d8bb958fa0..231aaf3a821 integrated the contents of
docs/qmp-events.txt into QAPI schema doc comments. It left dangling
references to qmp-events.txt behind. Fix to point to the QEMU QMP
reference manual generated from the QAPI schema.
Add a similar reference for commands.
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200806081147.3123652-2-armbru@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
path, prop = "type".rsplit('/', 1) sets path to "", which doesn't
work. Correct to "/".
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200723142738.1868568-4-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200723142738.1868568-3-armbru@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Commit c7b942d7f8 "scripts/qmp: Fix shebang and imports" messed with
it for reasons I don't quite understand. I do understand how it fails
now: it neglects to import sys. Fix that.
It now fails because it expects an old version of module fuse. That's
next.
Fixes: c7b942d7f8
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200723142738.1868568-2-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
On a 'qemu-discuss' thread[1], Kevin identifies that the current doc
blurb for @blockdev-add is stale:
This is actually a documentation bug. @id doesn't exist,
blockdev-add never creates a BlockBackend. This was different in the
very first versions of the patches to add blockdev-add and we
probably just forgot to update the documentation after removing it.
So remove the stale bits.
And the requirement for 'node-name' is already mentioned in the
documentation of @BlockdevOptions:
[...]
# @node-name: the node name of the new node (Since 2.0).
# This option is required on the top level of blockdev-add.
# Valid node names start with an alphabetic character and may
# contain only alphanumeric characters, '-', '.' and '_'. Their
# maximum length is 31 characters.
[...]
[1] https://lists.nongnu.org/archive/html/qemu-discuss/2020-07/msg00071.html
-- equivalent to "-drive if=ide,id=disk0....."
Fixes: be4b67bc7d ("blockdev: Allow creation of BDS trees without BB")
Signed-off-by: Kashyap Chamarthy <kchamart@redhat.com>
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200805100158.1239390-1-kchamart@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Currently QAPI generates a type and function for free'ing it:
typedef struct QCryptoBlockCreateOptions QCryptoBlockCreateOptions;
void qapi_free_QCryptoBlockCreateOptions(QCryptoBlockCreateOptions *obj);
This is used in the traditional manner:
QCryptoBlockCreateOptions *opts = NULL;
opts = g_new0(QCryptoBlockCreateOptions, 1);
....do stuff with opts...
qapi_free_QCryptoBlockCreateOptions(opts);
Since bumping the min glib to 2.48, QEMU has incrementally adopted the
use of g_auto/g_autoptr. This allows the compiler to run a function to
free a variable when it goes out of scope, the benefit being the
compiler can guarantee it is freed in all possible code ptahs.
This benefit is applicable to QAPI types too, and given the seriously
long method names for some qapi_free_XXXX() functions, is much less
typing. This change thus makes the code generator emit:
G_DEFINE_AUTOPTR_CLEANUP_FUNC(QCryptoBlockCreateOptions,
qapi_free_QCryptoBlockCreateOptions)
The above code example now becomes
g_autoptr(QCryptoBlockCreateOptions) opts = NULL;
opts = g_new0(QCryptoBlockCreateOptions, 1);
....do stuff with opts...
Note, if the local pointer needs to live beyond the scope holding the
variable, then g_steal_pointer can be used. This is useful to return the
pointer to the caller in the success codepath, while letting it be freed
in all error codepaths.
return g_steal_pointer(&opts);
The crypto/block.h header needs updating to avoid symbol clash now that
the g_autoptr support is a standard QAPI feature.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200723153845.2934357-1-berrange@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
This patch implements functionality for following ioctls:
BTRFS_IOC_SCRUB - Starting a btrfs filesystem scrub
Start a btrfs filesystem scrub. The third ioctls argument
is a pointer to a following type:
struct btrfs_ioctl_scrub_args {
__u64 devid; /* in */
__u64 start; /* in */
__u64 end; /* in */
__u64 flags; /* in */
struct btrfs_scrub_progress progress; /* out */
/* pad to 1k */
__u64 unused[(1024-32-sizeof(struct btrfs_scrub_progress))/8];
};
Before calling this ioctl, field 'devid' should be filled
with value that represents the device id of the btrfs filesystem
for which the scrub is to be started.
BTRFS_IOC_SCRUB_CANCEL - Canceling scrub of a btrfs filesystem
Cancel a btrfs filesystem scrub if it is running. The third
ioctls argument is ignored.
BTRFS_IOC_SCRUB_PROGRESS - Getting status of a running scrub
Read the status of a running btrfs filesystem scrub. The third
ioctls argument is a pointer to the above mentioned
'struct btrfs_ioctl_scrub_args'. Similarly as with 'BTRFS_IOC_SCRUB',
the 'devid' field should be filled with value that represents the
id of the btrfs device for which the scrub has started. The status
of a running scrub is returned in the field 'progress' which is
of type 'struct btrfs_scrub_progress' and its definition can be
found at:
https://elixir.bootlin.com/linux/latest/source/include/uapi/linux/btrfs.h#L150
Implementation nots:
Ioctls in this patch use type 'struct btrfs_ioctl_scrub_args' as their
third argument. That is the reason why an aproppriate thunk type
definition is added in file 'syscall_types.h'.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-9-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality for following ioctls:
BTRFS_IOC_QUOTA_CTL - Enabling/Disabling quota support
Enable or disable quota support for a btrfs filesystem. Quota
support is enabled or disabled using the ioctls third argument
which represents a pointer to a following type:
struct btrfs_ioctl_quota_ctl_args {
__u64 cmd;
__u64 status;
};
Before calling this ioctl, the 'cmd' field should be filled
with one of the values 'BTRFS_QUOTA_CTL_ENABLE' (enabling quota)
'BTRFS_QUOTA_CTL_DISABLE' (disabling quota).
BTRFS_IOC_QGROUP_CREATE - Creating/Removing a subvolume quota group
Create or remove a subvolume quota group. The subvolume quota
group is created or removed using the ioctl's third argument which
represents a pointer to a following type:
struct btrfs_ioctl_qgroup_create_args {
__u64 create;
__u64 qgroupid;
};
Before calling this ioctl, the 'create' field should be filled
with the aproppriate value depending on if the user wants to
create or remove a quota group (0 for removing, everything else
for creating). Also, the 'qgroupid' field should be filled with
the value for the quota group id that is to be created.
BTRFS_IOC_QGROUP_ASSIGN - Asigning or removing a quota group as child group
Asign or remove a quota group as child quota group of another
group in the btrfs filesystem. The asignment is done using the
ioctl's third argument which represents a pointert to a following type:
struct btrfs_ioctl_qgroup_assign_args {
__u64 assign;
__u64 src;
__u64 dst;
};
Before calling this ioctl, the 'assign' field should be filled with
the aproppriate value depending on if the user wants to asign or remove
a quota group as a child quota group of another group (0 for removing,
everythin else for asigning). Also, the 'src' and 'dst' fields should
be filled with the aproppriate quota group id values depending on which
quota group needs to asigned or removed as child quota group of another
group ('src' gets asigned or removed as child group of 'dst').
BTRFS_IOC_QGROUP_LIMIT - Limiting the size of a quota group
Limit the size of a quota group. The size of the quota group is limited
with the ioctls third argument which represents a pointer to a following
type:
struct btrfs_ioctl_qgroup_limit_args {
__u64 qgroupid;
struct btrfs_qgroup_limit lim;
};
Before calling this ioctl, the 'qgroup' id field should be filled with
aproppriate value of the quota group id for which the size is to be
limited. The second field is of following type:
struct btrfs_qgroup_limit {
__u64 flags;
__u64 max_rfer;
__u64 max_excl;
__u64 rsv_rfer;
__u64 rsv_excl;
};
The 'max_rfer' field should be filled with the size to which the quota
group should be limited. The 'flags' field can be used for passing
additional options and can have values which can be found on:
https://elixir.bootlin.com/linux/latest/source/include/uapi/linux/btrfs.h#L67
BTRFS_IOC_QUOTA_RESCAN_STATUS - Checking status of running rescan operation
Check status of a running rescan operation. The status is checked using
the ioctl's third argument which represents a pointer to a following type:
struct btrfs_ioctl_quota_rescan_args {
__u64 flags;
__u64 progress;
__u64 reserved[6];
};
If there is a rescan operation running, 'flags' field is set to 1, and
'progress' field is set to aproppriate value which represents the progress
of the operation.
BTRFS_IOC_QUOTA_RESCAN - Starting a rescan operation
Start ar rescan operation to Trash all quota groups and scan the metadata
again with the current config. Before calling this ioctl,
BTRFS_IOC_QUOTA_RESCAN_STATUS sould be run to check if there is already a
rescan operation runing. After that ioctl call, the received
'struct btrfs_ioctl_quota_rescan_args' should be than passed as this ioctls
third argument.
BTRFS_IOC_QUOTA_RESCAN_WAIT - Waiting for a rescan operation to finish
Wait until a rescan operation is finished (if there is a rescan operation
running). The third ioctls argument is ignored.
Implementation notes:
Almost all of the ioctls in this patch use structure types as third arguments.
That is the reason why aproppriate thunk definitions were added in file
'syscall_types.h'.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-8-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality for following ioctl:
BTRFS_IOC_DEFAULT_SUBVOL - Setting a default subvolume
Set a default subvolume for a btrfs filesystem. The third
ioctl's argument is a '__u64' (unsigned long long) which
represents the id of a subvolume that is to be set as
the default.
BTRFS_IOC_GET_SUBVOL_ROOTREF - Getting tree and directory id of subvolumes
Read tree and directory id of subvolumes from a btrfs
filesystem. The tree and directory id's are returned in the
ioctl's third argument which represents a pointer to a
following type:
struct btrfs_ioctl_get_subvol_rootref_args {
/* in/out, minimum id of rootref's treeid to be searched */
__u64 min_treeid;
/* out */
struct {
__u64 treeid;
__u64 dirid;
} rootref[BTRFS_MAX_ROOTREF_BUFFER_NUM];
/* out, number of found items */
__u8 num_items;
__u8 align[7];
};
Before calling this ioctl, 'min_treeid' field should be filled
with value that represent the minimum value for the tree id.
Implementation notes:
Ioctl BTRFS_IOC_GET_SUBVOL_ROOTREF uses the above mentioned structure
type as third argument. That is the reason why a aproppriate thunk
structure definition is added in file 'syscall_types.h'.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-7-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality of following ioctls:
BTRFS_IOC_INO_LOOKUP - Reading tree root id and path
Read tree root id and path for a given file or directory.
The name and tree root id are returned in an ioctl's third
argument that represents a pointer to a following type:
struct btrfs_ioctl_ino_lookup_args {
__u64 treeid;
__u64 objectid;
char name[BTRFS_INO_LOOKUP_PATH_MAX];
};
Before calling this ioctl, field 'objectid' should be filled
with the object id value for which the tree id and path are
to be read. Value 'BTRFS_FIRST_FREE_OBJECTID' represents the
object id for the first available btrfs object (directory or
file).
BTRFS_IOC_INO_PATHS - Reading paths to all files
Read path to all files with a certain inode number. The paths
are returned in the ioctl's third argument which represents
a pointer to a following type:
struct btrfs_ioctl_ino_path_args {
__u64 inum; /* in */
__u64 size; /* in */
__u64 reserved[4];
/* struct btrfs_data_container *fspath; out */
__u64 fspath; /* out */
};
Before calling this ioctl, the 'inum' and 'size' field should
be filled with the aproppriate inode number and size of the
directory where file paths should be looked for. For now, the
paths are returned in an '__u64' (unsigned long long) value
'fspath'.
BTRFS_IOC_LOGICAL_INO - Reading inode numbers
Read inode numbers for files on a certain logical adress. The
inode numbers are returned in the ioctl's third argument which
represents a pointer to a following type:
struct btrfs_ioctl_logical_ino_args {
__u64 logical; /* in */
__u64 size; /* in */
__u64 reserved[3]; /* must be 0 for now */
__u64 flags; /* in, v2 only */
/* struct btrfs_data_container *inodes; out */
__u64 inodes;
};
Before calling this ioctl, the 'logical' and 'size' field should
be filled with the aproppriate logical adress and size of where
the inode numbers of files should be looked for. For now, the
inode numbers are returned in an '__u64' (unsigned long long)
value 'inodes'.
BTRFS_IOC_LOGICAL_INO_V2 - Reading inode numbers
Same as the above mentioned ioctl except that it allows passing
a flags 'BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET'.
BTRFS_IOC_INO_LOOKUP_USER - Reading subvolume name and path
Read name and path of a subvolume. The tree root id and
path are read in an ioctl's third argument which represents a
pointer to a following type:
struct btrfs_ioctl_ino_lookup_user_args {
/* in, inode number containing the subvolume of 'subvolid' */
__u64 dirid;
/* in */
__u64 treeid;
/* out, name of the subvolume of 'treeid' */
char name[BTRFS_VOL_NAME_MAX + 1];
/*
* out, constructed path from the directory with which the ioctl is
* called to dirid
*/
char path[BTRFS_INO_LOOKUP_USER_PATH_MAX];
};
Before calling this ioctl, the 'dirid' and 'treeid' field should
be filled with aproppriate values which represent the inode number
of the directory that contains the subvolume and treeid of the
subvolume.
Implementation notes:
All of the ioctls in this patch use structure types as third arguments.
That is the reason why aproppriate thunk definitions were added in file
'syscall_types.h'.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-6-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality for following ioctls:
BTRFS_IOC_GET_FEATURES - Getting feature flags
Read feature flags for a btrfs filesystem. The feature flags
are returned inside the ioctl's third argument which represents
a pointer to a following structure type:
struct btrfs_ioctl_feature_flags {
__u64 compat_flags;
__u64 compat_ro_flags;
__u64 incompat_flags;
};
All of the structure field represent bit masks that can be composed
of values which can be found on:
https://elixir.bootlin.com/linux/latest/source/fs/btrfs/ctree.h#L282
BTRFS_IOC_SET_FEATURES - Setting feature flags
Set and clear feature flags for a btrfs filesystem. The feature flags
are set using the ioctl's third argument which represents a
'struct btrfs_ioctl_feature_flags[2]' array. The first element of the
array represent flags which are to be cleared and the second element of
the array represent flags which are to be set. The second element has the
priority over the first, which means that if there are matching flags
in the elements, they will be set in the filesystem. If the flag values
in the third argument aren't correctly set to be composed of the available
predefined flag values, errno ENOPERM ("Operation not permitted") is returned.
BTRFS_IOC_GET_SUPPORTED_FEATURES - Getting supported feature flags
Read supported feature flags for a btrfs filesystem. The supported
feature flags are read using the ioctl's third argument which represents
a 'struct btrfs_ioctl_feature_flags[3]' array. The first element of this
array represents all of the supported flags in the btrfs filesystem.
The second element represents flags that can be safely set and third element
represent flags that can be safely clearead.
Implementation notes:
All of the implemented ioctls use 'struct btrfs_ioctl_feature_flags' as
third argument. That is the reason why a corresponding defintion was added
in file 'linux-user/syscall_types.h'.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-5-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality for following ioctls:
BTRFS_IOC_SCAN_DEV - Scanning device for a btrfs filesystem
Scan a device for a btrfs filesystem. The device that is to
be scanned is passed in the ioctl's third argument which
represents a pointer to a 'struct ioc_vol_args' (which was
mentioned in a previous patch). Before calling this ioctl,
the name field of this structure should be filled with the
aproppriate name value which represents a path for the device.
If the device contains a btrfs filesystem, the ioctl returns 0,
otherwise a negative value is returned.
BTRFS_IOC_ADD_DEV - Adding a device to a btrfs filesystem
Add a device to a btrfs filesystem. The device that is to be
added is passed in the ioctl's third argument which represents
a pointer to a 'struct ioc_vol_args' (which was mentioned in
a previous patch). Before calling this ioctl, the name field of
this structure should be filled with the aproppriate name value
which represents a path for the device.
BTRFS_IOC_RM_DEV - Removing a device from a btrfs filesystem
Remove a device from a btrfs filesystem. The device that is to be
removed is passed in the ioctl's third argument which represents
a pointer to a 'struct ioc_vol_args' (which was mentioned in
a previous patch). Before calling this ioctl, the name field of
this structure should be filled with the aproppriate name value
which represents a path for the device.
BTRFS_IOC_DEV_INFO - Getting information about a device
Obtain information for device in a btrfs filesystem. The information
is gathered in the ioctl's third argument which represents a pointer
to a following structure type:
struct btrfs_ioctl_dev_info_args {
__u64 devid; /* in/out */
__u8 uuid[BTRFS_UUID_SIZE]; /* in/out */
__u64 bytes_used; /* out */
__u64 total_bytes; /* out */
__u64 unused[379]; /* pad to 4k */
__u8 path[BTRFS_DEVICE_PATH_NAME_MAX]; /* out */
};
Before calling this ioctl, field "devid" should be set with the id value
for the device for which the information is to be obtained. If this field
is not aproppriately set, the errno ENODEV ("No such device") is returned.
BTRFS_IOC_GET_DEV_STATS - Getting device statistics
Obtain stats informatin for device in a btrfs filesystem. The information
is gathered in the ioctl's third argument which represents a pointer to
a following structure type:
struct btrfs_ioctl_get_dev_stats {
__u64 devid; /* in */
__u64 nr_items; /* in/out */
__u64 flags; /* in/out */
/* out values: */
__u64 values[BTRFS_DEV_STAT_VALUES_MAX];
/*
* This pads the struct to 1032 bytes. It was originally meant to pad to
* 1024 bytes, but when adding the flags field, the padding calculation
* was not adjusted.
*/
__u64 unused[128 - 2 - BTRFS_DEV_STAT_VALUES_MAX];
};
Before calling this ioctl, field "devid" should be set with the id value
for the device for which the information is to be obtained. If this field
is not aproppriately set, the errno ENODEV ("No such device") is returned.
BTRFS_IOC_FORGET_DEV - Remove unmounted devices
Search and remove all stale devices (devices which are not mounted).
The third ioctl argument is a pointer to a 'struct btrfs_ioctl_vol_args'.
The ioctl call will release all unmounted devices which match the path
which is specified in the "name" field of the structure. If an empty
path ("") is specified, all unmounted devices will be released.
Implementation notes:
Ioctls BTRFS_IOC_DEV_INFO and BTRFS_IOC_GET_DEV_STATS use types
'struct btrfs_ioctl_dev_info_args' and ' struct btrfs_ioctl_get_dev_stats'
as third argument types. That is the reason why corresponding structure
definitions were added in file 'linux-user/syscall_types.h'.
Since the thunk type for 'struct ioc_vol_args' was already added in a
previous patch, the rest of the implementation was straightforward.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-4-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality for following ioctls:
BTRFS_IOC_SNAP_CREATE - Creating a subvolume snapshot
Create a snapshot of a btrfs subvolume. The snapshot is created using the
ioctl's third argument that is a pointer to a 'struct btrfs_ioctl_vol_args'
(which was mentioned in the previous patch). Before calling this ioctl,
the fields of the structure should be filled with aproppriate values for
the file descriptor and path of the subvolume for which the snapshot is to
be created.
BTRFS_IOC_SNAP_DESTROY - Removing a subvolume snapshot
Delete a snapshot of a btrfs subvolume. The snapshot is deleted using the
ioctl's third argument that is a pointer to a 'struct btrfs_ioctl_vol_args'
(which was mentioned in the previous patch). Before calling this ioctl,
the fields of the structure should be filled with aproppriate values for
the file descriptor and path of the subvolume for which the snapshot is to
be deleted.
Implementation notes:
Since the thunk type 'struct btrfs_ioctl_vol_args' is defined in the
previous patch, the implementation for these ioctls was straightforward.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-3-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This patch implements functionality of following ioctls:
BTRFS_IOC_SUBVOL_CREATE - Creating a btrfs subvolume
Create a btrfs subvolume. The subvolume is created using the ioctl's
third argument which represents a pointer to a following structure
type:
struct btrfs_ioctl_vol_args {
__s64 fd;
char name[BTRFS_PATH_NAME_MAX + 1];
};
Before calling this ioctl, the fields of this structure should be filled
with aproppriate values. The fd field represents the file descriptor
value of the subvolume and the name field represents the subvolume
path.
BTRFS_IOC_SUBVOL_GETFLAGS - Getting subvolume flags
Read the flags of the btrfs subvolume. The flags are read using
the ioctl's third argument that is a pointer of __u64 (unsigned long).
The third argument represents a bit mask that can be composed of following
values:
BTRFS_SUBVOL_RDONLY (1ULL << 1)
BTRFS_SUBVOL_QGROUP_INHERIT (1ULL << 2)
BTRFS_DEVICE_SPEC_BY_ID (1ULL << 3)
BTRFS_SUBVOL_SPEC_BY_ID (1ULL << 4)
BTRFS_IOC_SUBVOL_SETFLAGS - Setting subvolume flags
Set the flags of the btrfs subvolume. The flags are set using the
ioctl's third argument that is a pointer of __u64 (unsigned long).
The third argument represents a bit mask that can be composed of same
values as in the case of previous ioctl (BTRFS_IOC_SUBVOL_GETFLAGS).
BTRFS_IOC_SUBVOL_GETINFO - Getting subvolume information
Read information about the subvolume. The subvolume information is
returned in the ioctl's third argument which represents a pointer to
a following structure type:
struct btrfs_ioctl_get_subvol_info_args {
/* Id of this subvolume */
__u64 treeid;
/* Name of this subvolume, used to get the real name at mount point */
char name[BTRFS_VOL_NAME_MAX + 1];
/*
* Id of the subvolume which contains this subvolume.
* Zero for top-level subvolume or a deleted subvolume.
*/
__u64 parent_id;
/*
* Inode number of the directory which contains this subvolume.
* Zero for top-level subvolume or a deleted subvolume
*/
__u64 dirid;
/* Latest transaction id of this subvolume */
__u64 generation;
/* Flags of this subvolume */
__u64 flags;
/* UUID of this subvolume */
__u8 uuid[BTRFS_UUID_SIZE];
/*
* UUID of the subvolume of which this subvolume is a snapshot.
* All zero for a non-snapshot subvolume.
*/
__u8 parent_uuid[BTRFS_UUID_SIZE];
/*
* UUID of the subvolume from which this subvolume was received.
* All zero for non-received subvolume.
*/
__u8 received_uuid[BTRFS_UUID_SIZE];
/* Transaction id indicating when change/create/send/receive happened */
__u64 ctransid;
__u64 otransid;
__u64 stransid;
__u64 rtransid;
/* Time corresponding to c/o/s/rtransid */
struct btrfs_ioctl_timespec ctime;
struct btrfs_ioctl_timespec otime;
struct btrfs_ioctl_timespec stime;
struct btrfs_ioctl_timespec rtime;
/* Must be zero */
__u64 reserved[8];
};
All of the fields of this structure are filled after the ioctl call.
Implementation notes:
Ioctls BTRFS_IOC_SUBVOL_CREATE and BTRFS_IOC_SUBVOL_GETINFO have structure
types as third arguments. That is the reason why a corresponding definition
are added in file 'linux-user/syscall_types.h'.
The line '#include <linux/btrfs.h>' is added in file 'linux-user/syscall.c' to
recognise preprocessor definitions for these ioctls. Since the file "linux/btrfs.h"
was added in the kernel version 3.9, it is enwrapped in an #ifdef statement
with parameter CONFIG_BTRFS which is defined in 'configure' if the
header file is present.
Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200823195014.116226-2-Filip.Bozuta@syrmia.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
MK_ARRAY(type,size) is used to fill the field_types buffer, and if the
"size" parameter is an enum type, clang [-Werror,-Wenum-conversion] reports
an error when it is assigned to field_types which is also an enum, argtypes.
To avoid that, convert "size" to "int" in MK_ARRAY(). "int" is the type
used for the size evaluation in thunk_type_size().
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20200902125752.1033524-1-laurent@vivier.eu>
Disabling these parts are sufficient to get the qemu-nbd program
compiling in a Windows build.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200825103850.119911-4-berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
The termsig_handler function is used by the client thread handling the
host NBD device connection to do a graceful shutdown. IOW, if we have
disabled NBD device support at compile time, we don't need the SIGTERM
handler. This fixes a build issue for Windows.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200825103850.119911-3-berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Any tool that uses sockets needs to call socket_init() in order to work
on the Windows platform.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20200825103850.119911-2-berrange@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
This makes nbd's connection_co yield during reconnects, so that
reconnect doesn't block the main thread. This is very important in
case of an unavailable nbd server host: connect() call may take a long
time, blocking the main thread (and due to reconnect, it will hang
again and again with small gaps of working time during pauses between
connection attempts).
Realization notes:
- We don't want to implement non-blocking connect() over non-blocking
socket, because getaddrinfo() doesn't have portable non-blocking
realization anyway, so let's just use a thread for both getaddrinfo()
and connect().
- We can't use qio_channel_socket_connect_async (which behaves
similarly and starts a thread to execute connect() call), as it's relying
on someone iterating main loop (g_main_loop_run() or something like
this), which is not always the case.
- We can't use thread_pool_submit_co API, as thread pool waits for all
threads to finish (but we don't want to wait for blocking reconnect
attempt on shutdown.
So, we just create the thread by hand. Some additional difficulties
are:
- We want our connect to avoid blocking drained sections and aio context
switches. To achieve this, we make it possible to "cancel" synchronous
wait for the connect (which is a coroutine yield actually), still,
the thread continues in background, and if successful, its result may be
reused on next reconnect attempt.
- We don't want to wait for reconnect on shutdown, so there is
CONNECT_THREAD_RUNNING_DETACHED thread state, which means that the block
layer is no longer interested in a result, and thread should close new
connected socket on finish and free the state.
How to reproduce the bug, fixed with this commit:
1. Create an image on node1:
qemu-img create -f qcow2 xx 100M
2. Start NBD server on node1:
qemu-nbd xx
3. Start vm with second nbd disk on node2, like this:
./x86_64-softmmu/qemu-system-x86_64 -nodefaults -drive \
file=/work/images/cent7.qcow2 -drive file=nbd+tcp://192.168.100.2 \
-vnc :0 -qmp stdio -m 2G -enable-kvm -vga std
4. Access the vm through vnc (or some other way?), and check that NBD
drive works:
dd if=/dev/sdb of=/dev/null bs=1M count=10
- the command should succeed.
5. Now, let's trigger nbd-reconnect loop in Qemu process. For this:
5.1 Kill NBD server on node1
5.2 run "dd if=/dev/sdb of=/dev/null bs=1M count=10" in the guest
again. The command should fail and a lot of error messages about
failing disk may appear as well.
Now NBD client driver in Qemu tries to reconnect.
Still, VM works well.
6. Make node1 unavailable on NBD port, so connect() from node2 will
last for a long time:
On node1 (Note, that 10809 is just a default NBD port):
sudo iptables -A INPUT -p tcp --dport 10809 -j DROP
After some time the guest hangs, and you may check in gdb that Qemu
hangs in connect() call, issued from the main thread. This is the
BUG.
7. Don't forget to drop iptables rule from your node1:
sudo iptables -D INPUT -p tcp --dport 10809 -j DROP
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200812145237.4396-1-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: minor wording and formatting tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
The error message has changed recently, breaking the test. Fix it.
Fixes: a2b333c018
("block: nbd: Fix convert qcow2 compressed to nbd")
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200811080830.289136-1-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
As of the patch to flush qemu-img's "Formatting" message before the
error message, 059 has been broken for vmdk. Fix it.
Fixes: 4e2f441878
("qemu-img: Flush stdout before before potential stderr messages")
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200811084150.326377-1-mreitz@redhat.com>
Reviewed-by: Eric blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Both VirtioPCIBusClass and VirtioCcwBusClass are typedefs of
VirtioBusClass, but set .class_size in the TypeInfo anyway
to be safe if that changes in the future.
Reported-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Message-Id: <20200824122051.99432-1-cohuck@redhat.com>