ploop_copy_send: wait for send_thread to finishThis should fix a race condition between the main process and send_thread.
The problem is that the main process doesn't wait for send_thread to
finish its work, incorrectly assuming that if it can take the mutex lock
again, the send_thread has already finished. But there might be a situation
when send_thread() still waits in pthread_cond_wait() (with the mutex
lock released), while async_send ...
ploop_copy_send: use while in pthread_cond_waitIt's a common practice and a good thing to recheck
the condition after being woken up.
Reported-by: Maxim Patlasov <mpatlasov@parallels.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop mount: add check for host mount optsAs pointed out in bug #3030, ext4 fs that hosts ploop images
should not be mounted with data=writeback as it increases
chances of broken ploop.
This patch adds a check for such mount option -- if it is found
then a warning message is printed once, suggesting to not use it.
https://bugzilla.openvz.org/3030
Reported-by: Devon <devon.b@virtualcomplete.com>
Signed-off-by: Kir Kolyshkin <kir@open...
send_thread(): move mutex lock/unlock out of do/whileAs suggested by Maxim, there is no point to release a mutex
and immediately take it again.
Reported-by: Maxim Patlasov <mpatlasov@parallels.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
async_send(): call pthread_cond_signal() with mutex lockedFrom pthread_conf_signal man page:
> The pthread_cond_broadcast() or pthread_cond_signal() functions may be
> called by a thread whether or not it currently owns the mutex that
> threads calling pthread_cond_wait() or pthread_cond_timedwait() have
> associated with the condition variable during their waits; however, if
> predictable scheduling behavior is required, then that mutex s...
ploop_switch_snapshot_ex(): don't show error on ENOENT rmA failure to remove old top delta is not fatal, plus if
ENOENT is returned it doesn't make sense to even show
the error.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
tune_fs(): fix check for non-implemented ioctlThe check was obviously broken -- errno values are positive.
This is a fix to commit 5a0ab39.
Reference: ploop@1044468
Reported-by: Corrado Fiore <lists@corradofiore.it>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_check(): use reopen_rw()1. reopen_rw(): check the result of close(), reassign new fd
2. modify check_and_repair_sparse to use *fd not fd
3. use reopen_rw() in ploop_check()
Reference: ploop@1037067
[style] check.c: rearrange code to remove function declarationmove up:
static int reopen_rw(const char *image, int *fd);
static int fill_hole(const char *image, int *fd, off_t start, off_t end, int *log, int repair);
static int check_and_repair_sparse(const char *image, int fd, __u64 cluster, int flags);
Reference: ploop@1037054
make_fs(): add tunefs -e remount-roAs suggested in bug #3025, it is good to remount ploop inner filesystem
read-only as soon as kernel sees it is corrupted, in order to prevent
any more damage.
https://bugzilla.openvz.org/3025
Reported-by: Pavel Odintsov <pavel.odintsov@gmail.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_check(): don't pass -1 to closeThis fixes a complain from Coverity Scan:
> CID 1228664 (#1 of 1): Improper use of negative value (NEGATIVE_RETURNS)
> negative_returns: fd is passed to a parameter that cannot be negative.
While at it, use close_safe() to close the old fd, so we can
detect and report an error.
Reference: ploop@1036954
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
check_and_repair_sparse(): reopen image rw if neededAfter the commit r1036784, we can be in the situation when image
is opened as read-only but check_and_repair_sparse() needs to
write to it.
As we don't know beforehand whether we will need to write to an image
file or not, let's fix it in a somewhat kludgy way -- as soon as we
found out we need to write, check opened fd mode and reopen the image
read/write if necessary. Note this is only perfo...
fill_hole: use *fdReplace fd argument with pointer to it.
This is a preparation for the next commit, to make it easier to read.
Reference: ploop@1036951
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
check_and_repair_sparse: use fill_hole ret codeSince fill_hole returns SYSEXIT_* values it makes sense to return
those explicitly.
Reference: ploop@1036950
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
fill_hole(): no need to bzero static bufIn C, statically-allocated objects without an explicit initializer
are initialized to zero, so there is no need for explicit bzero().
While at it, mark buf as const to make sure it is kept as is (all
zeroes).
Reference: ploop@1036949
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_check(): open image RO, reopen RW laterMost of the time writing to the image file is not required, so
let's optimize for this case and open the file read only. Later,
if an image has dirty flag set, of CHECK_DETAILED is set, reopen
the file in read/write mode.
This also helps with filesystems on which opening the file read/write
is more expensive than opening it read-only.
Reference: PSBM-26499
References: ploop@1036784, ploop@103...
ploop_create_{image,dd}: use ploop_di_add_image ret valueAs ploop_di_add_image is always returning proper SYSEXIT_* code,
let's just use it directly, saving a couple lines of code.
This also fixes a few cases of returning SYSEXIT_MALLOC instead
of SYSEXIT_PARAM which ploop_di_add_image() returns sometimes.
Reference: ploop@1031842
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
fsutils.c: quiet fsckfsck output can broke tools parser. There is no quiet option
for fsck so just redirect stdout to /dev/null
# fsck -p /dev/ploop26995p1
fsck from util-linux-ng 2.17.2
/dev/ploop26995p1: clean, 12/655360 files, 83119/2620928 blocks
[kir@: errors from fsck will still be shown on stderr]
Reference: PSBM-26738
Reference: ploop@1008262
ploop_get_max_size: improve description, move to cWe do not have descriptions of our public functions in libploop.h
file, so let's not start doing it. So far, the best documentation
for such functions is their code itself, so let's keep the description
close to the code.
This is an addition to the previous commit 71be8d5
Cc: Nikolay Shirokovskiy <nshirokovskiy@parallels.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_get_max_size(): add[kir@: add dynload.h entry]
Signed-off-by: Nikolay Shirokovskiy <nshirokovskiy@parallels.com>
Acked-by: Igor Sukhih <igor@parallels.com>
Reference: PSBM-26570 - ""ploop convert" fails to convert the container due to huge limit for "diskinodes""
Reference: ploop@1028131
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
De-introduce SYSEXIT_INUSEThis was introduced by the previous commit 22ff44b (aka ploop@1025736)
for the sole purpose of reporting the "in-use flag is set" condition
from get_image_param_offline(), which, in turn, takes it from
errno value explicitly set to EBUSY by open_delta().
This condition is then checked in the only place --
do_create_snapshot().
Note that SYSEXIT_* codes are part of public API, but this
new valu...
create_snapshot: clear stale top delta inuse flagSometimes after the node crash the top delta in-use flag is left set, so
snapshot creation returns an error.
Handle this particular case by calling check_delta to validate top delta
and drop inuse flag on create snapshot.
Reference: PSBM-27694 "Failed to create the snapshot: Operation failed. %1 after node crash"
References: ploop@1025736, ploop@1026188
Signed-off-by: Kir Kolyshkin <kir@openv...
ploop info: add -d to show ploop deviceReference: PSBM-26990
Signed-off-by: Nikolay Shirokovskiy <nshirokovskiy@parallels.com>
Acked-by: Igor Sukhih <igor@parallels.com>
References: ploop@1025015, ploop@1026043
tune_fs: fix error message, assorted nitpicksA few nitpicks after commit 5a0ab39 (ploop@1022894),
in the order of importance;
1. Fix an error message (s/for size/to/, obviously a copy-paste error)
2. Pass errno to ploop_err() to know why ioctl failed
3. An empty line between variable declarations and code
4. Some more empty lines to improve code readability.
5. A comma in a comment.
Reference: ploop@1025878
Signed-off-by: Kir Kolysh...
tune_fs(): use EXT4_IOC_SET_RSV_BLOCKS if availableWe have to update reservation block counts according to balloon size.
One can do that directly via tune2fs which is not good for mounted fs,
and should use this ioctl instead
EXT4_IOC_SET_RSV_BLOCKS is available since 042stab092.1. In case it's
not there, fall back to old implementation.
Signed-off-by: dmonakhov@openvz.org
Committed-by: igor@openvz.org
Reference: PSBM-25929 "Error in __ploop_...
ploop_delete_snapshot: fix for temp snapshotIn case snapshot guid supplied is one of temp snapshot, we first
implicitly delete it in merge_temporary_snapshots(), then try to
delete it again explicitly in do_delete_snapshot() and we fail.
To fix, move merge_temporary_snapshots() to after do_delete_snapshot.
Reference: PSBM-27327 "Error in do_delete_snapshot (snapshot.c:61): Can't find snapshot by uuid"
Reference: ploop@1021028, ploop@10...
reset_top_delta(): rm irrelevant messageApparently a copy-paste leftover.
This is an addition to commit 86f3119 (ploop@1019626)
Reference: ploop@1020175
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_switch_snapshot_ex: fixed case with PLOOP_SNAP_SKIP_TOPDELTA_CREATEThe PLOOP_SNAP_SKIP_TOPDELTA_CREATE used in rollback, move logic to separate function
Reference: ploop@1019626
do_delete_snapshot(): fix a warningFix and improve a warning message about inability to remove a snapshot
with >1 references to it, and marking it as temp.
This is an addition to commit ploop@1016663 aka 32711a4.
Reference: ploop@1016859
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_delete_snapshot: deny delete top delta in case parent snapshot is temporaryCreating temporary snapshot and deleting top delta broke DiskDescriptor.xml:
ploop tsnapshot -u {aff55492-7320-4d72-aedd-c1e303a22cc1} -c test DiskDescriptor.xml
ploop snapshot-delete -u {5fbaabe3-6958-40ff-92a7-860e329aab41} DiskDescriptor.xml
ploop mount DiskDescriptor.xml
Error in validate_disk_descriptor (xml.c:327): Validation of
/vz/igor/Virtuozzo/current/ploop/DiskDescriptor.xml f...
BUGFIX Can not delete snapshot with multiple refsThe direct way to delete snapshot with > 1 references is
to make a copy of a snapshot image and merge with the copy
* Delete A
A -> B => A -> B => B
\-> C A -> C A -> C
Let's handle in another way, mark as temporary and merge
when it will be possible
Reference: PSBM-27154 "Can not delete snapshot of Container - double referenced"
Reference: ploop@1016663
ploop_switch_snapshot_ex(): don't return 0 if guid_old is badCode checks that param->guid_old is (1) valid and (2) non-existent,
and prints and error if any of those checks fails, but returns 0
instead of an error.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_resize_image(): handle get_mount_dir() errorget_mount_dir() return value is not boolean, but rather a 3-state.
It doesn't make sense to proceed if it returned an error, as it's
probably fatal.
While at it, document get_mount_dir() return values.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
git-svn-id: https://svn.parallels.com/parallels/vzlin/utils/current/ploop@1015021 3f79cd19-dd19-0410-83c5-8b167e6f9dce
Signed-off-by: Kir Kolyshki...
ploop_resize_image: auto-mount fsProblem: unable to resize ploop if it is running but fs is not mounted
Solution: automount file system before performing online resize
Reference: PSBM-26764
Reference: ploop@1013361
ploop_store_diskdescriptor(): fix error msgAfter commit b519e4c, we call a different function, so
fix the error message accordingly.
@noreview
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_store_diskdescriptor: fix typo, show errno1 s/Faled/Failed/ to fix a typo.
2 s/Failed/Can't/ to use a verb we tend to use more often.
3 Pass errno to ploop_err() to show why exactly we can't.
This is an addition to commit b519e4c
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_store_diskdescriptor: fsync DiskDescriptor.xmlProblem: DiskDescriptor.xml is empty after host's hard reboot
Solution: call fsync() before rename
Reference: PSBM-27028
Reference: ploop@1013886
ploop copy: fix open_mount_point err handlingBefore this patch:
# ploop -vvvv copy -s /dev/ploop47147 -d file
[ 0.000337] Error in open_mount_point (pcopy.c:357):
Can't open : No such file or directory
Apparently, ploop_get_mnt_by_dev() can return 1 in case there is
no error, but mount point can't be found, therefore we should check
for non-zero, not just < 0.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_copy_send(): add check for feedback_fdIf feedback_fd is specified, check that it is a pipe or a socket.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_copy_receive(): improve fd checksUse is_fd_pipe() to check both input fd and feedback fd.
Also, print fd in error message.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Kir Kolyshkin
