Fix compilation with gcc 7GCC 7.0 adds a new warning (which becomes an error with -Werror)
telling us something like this:
> CC ploop.o
> ploop.c: In function ‘plooptool_list’:
> ploop.c:1153:37: warning: ‘%s’ directive output may be truncated
> writing up to 255 bytes into a region of size 59 [-Wformat-truncation=]
> snprintf(dev, sizeof(dev), "/dev/%s", de->d_name);
> ...
Sync ploop_if.h with the latest kernel versionIntroduced new API for "push" backup scheme. This scheme have a number of advantages over
traditional "pull" one:
* No temporary delta which one should care of
* No I/O doplication (store new data to temporary place and move after backup is done)
PSBM-45001
Create .statfs and DiskDescriptor.xml.lck with 644 mode (v2) #PSBM-48352For openstack we need to copy disk directory under "stack" user (not root)
But in disk directory there are files without "read" permission for all user
-rw-r--r-- 1 root root DiskDescriptor.xml
-rw------- 1 root root DiskDescriptor.xml.lck
-rw-rw-r-- 1 stack libvirtd root.hds
-rw------- 1 root root .statfs
v2: use fchmod to set mode
Signed-off-by: Igor...
Fix close of 0 descriptor in case of error in grow_lower_delta. #PSBM-47329In case of node backup with CTs command line lost connection with dispatcher.
It was caused by close 0 fd in case of error in grow_lower_delta before delta was open.
'ploop balloon discard --stat' does not work on not mounted image #PSBM-53998Add automount logic if ploop image is not mounted
Signed-off-by: Igor Sukhih <igor@virtuozzo.com>
defrag_complete: make sure cmd buf is long enoughEmploy BUILD_BUG_ON to make sure our buffer is not shorter
than it should be. No runtime overhead!
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete(): don't use %ms in fscanfUnfortunately, on a RHEL5 system (which we still support!) we have
pretty ancient gcc-4.1.2 and glibc-2.5, and they don't support %ms
in fscanf, spitting the following error:
cc1: warnings being treated as errors
balloon.c: In function 'defrag_complete':
balloon.c:1210: warning: unknown conversion type character 'm' in format
balloon.c:1210: warning: too many arguments for format
In fact, it'...
discard/defrag: take into account image size change caused by defragDefragmentation may increase size of ploop image. In that case,
'ploop-discard --defrag --to-free X' command will not give us expected
result, because delta added by defrag won't be taken into account.
This patch calculates such change of image size and adds it to 'to_free'
parameter so it we will discarded later.
https://jira.sw.ru/browse/PSBM-42548
https://jira.sw.ru/browse/PSBM-45364
Signe...
ploop discard operation cancelled on fs defragmentation error #PSBM-46198fs defragmentation is optimization and should not fail the discard operation
Signed-off-by: Igor Sukhih <igor@parallels.com>
Unable to create ploop image > 16TB without filesystem inside #PSBM-45162Restriction 16T valid only for images with ext4 inside.
Lets remove 16T restriction for images withousd file system
Signed-off-by: Igor Sukhih <igor@parallels.com>
Lock and read disk descriptor before accessing it in ploop_get_*_delta_fname() (PSBM-45496)get_{top,base}_delta_guid() operate on ploop_disk_images_data structure, which
must be already initialized prior to calling these functions.
Call ploop_lock_dd() before calling get_{top,base}_delta_guid().
[kir@: this patch also fixes an error message in get_delta_fname()]
Signed-off-by: Igor Bazhitov <ibazhitov@virtuozzo.com>
api to get base delta file name #PSBM-46238ploop_get_base_delta_fname()
[kir@: add v1.15 to include/dynload.h]
Signed-off-by: Igor Sukhih <igor@parallels.com>
do_mntn_merge(): fix the check for same file1. It's better to use ploop_fname_cmp() as the file names
could differ.
2. Make the error more correct and descriptive.
3. Fix the error code -- this is a problem with dd.xml.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_mntn_merge: compile fixSince commit aeed3c3 ploop_di_merge_image() does not return
the delta file name, so let's use find_image_by_guid() to get it.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Repair DiskDescriptot.xml on process PLOOP_MNTN_MERGE stage #PSBM-33414Online merge of top delta performed in the kernel and
if ioctl(PLOOP_IOC_MERGE) was terminated, ploop device remaons in the
PLOOP_MNTN_MERGE state.
We heve complete on-going operation logic. if we detect PLOOP_MNTN_MERGE
we do complete this operation in the kernel but do not update dd.xml
This patch introduce functionlity to update dd.xml after merge completion.
NB: the problem of inconsisten...
ploop_complete_running_operation(): add 'struct ploop_disk_images_data' argument* rename ploop_complete_running_operation/complete_running_operation
* add extra arg to complete_running_operation()
* deprecate ploop_complete_running_operation()
[kir@: remove PL_EXT from function proto]
Signed-off-by: Igor Sukhih <igor@parallels.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Remove temp mount point after ploop resizeReferences:
#PSBM-42493
https://src.openvz.org/projects/OVZ/repos/ploop/commits/746e5937c232b8ceae5c2a50b23ec22a7a007caa
https://src.openvz.org/projects/OVZ/repos/ploop/commits/0b751692117f869bf82e0c1d57276492684c38ed
https://src.openvz.org/projects/OVZ/repos/ploop/commits/c51df35e1cedc5472b59d06f9ac3cb61577d5aa5
https://src.openvz.org/projects/OVZ/repos/ploop/commits/652d778da87b9d73c602a...
ploop_mount_fs: don't mount ro when finding balloon inodeWhen mounting a ploop-backed filesystem, we did the following:
1. Mount with MS_RDONLY flag.
2. stat() the balloon file to figure out its inode number
3. Remount with intended flags (usually read-write) plus the
added balloon_ino=XXX option.
Unfortunately, this way is not working well with applications
that live in MS_SLAVE mount namespace. A slave mount namespace
is a namespace to which a...
Fix compile with gcc-5.3.0gcc-5.3.0 on Alpine Linux complains:
ploop.h: In function 'ploop_ioff_to_sec':
ploop.h:213:1: error: control reaches end of non-void function
[-Werror=return-type]
This is apparently a bug in compiler, but let it go by adding bogus
"return 0" statement.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Fix compile with musl wrt gnu_dev_*Functions gnu_dev_{minor,major,makedev} are GNU extensions.
Standard functions minor(), major() and makedev() should be used.
This fixes compilation on Alpine Linux.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Fix compile with musl wrt offsetofoffsetof() is defined in stddef.h, which is probably included
from some other includes in GNU libc, but not in musl libc, so
we get a number of errors like this:
balloon_util.c:45:12: error: implicit declaration of function 'offsetof'
[-Werror=implicit-function-declaration]
To fix, add the proper #include.
This fixes compilation for Alpine Linux.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
gensym.sh: busybox sed compatibilityApparently busybox sed does not understand something that GNU sed
was fine with (say, ^ inside \(...\) pair).
Simplify sed expression in gen_h(), making it compatible with
sed from busybox suite.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Resize ploop image mounted inside mnt namespace #PSBM-39492If ploop image mounted inside CT it not possible to use
host e2resizefs to resize ext4 (mount is not visible from the node).
Workaround is mount device from host.
[kir@: whitespace cleanup, make mount_fs() static]
Signed-off-by: Igor Sukhih <igor@parallels.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete(): report stale pidfileIn case the process which pid is saved to pidfile is a different
process than the one we ran, or it doesn't exist, do warn about
the stale pidfile.
v2: check again the file exists before reporting
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete(): report stale pidfileIn case the process which pid is saved to pidfile is a different
process than the one we ran, or it doesn't exist, do warn about
the stale pidfile.
v2: check again the file exists before reporting
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete(): check for cmdlineInstead of checking if the running process executable file on disk
is the same as the one that we ran, it's slightly better to check
that the executable path is the same. This is because the file might
be different, for example after defrag rpm upgrade, but its name
and path stays the same.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete(): report errorsDon't silently exit in case of errors -- an error (except for
ENOENT after open() which is fine) might be an indication of
something going wrong, so it's better if we report it.
v2: remove access(), check for ENOENT after open().
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
defrag_complete: fix a typoThis is a pretty interesting case of missing ; after return,
leading to ploop_log() statement after it to be executed
conditionally. The only issue in this case, though, is that
the message is only printed in the error case, rather than
in the normal case as it should be.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
balloon.c: no need to include libgen.hlibgen.h only provides POSIX version of basename()
and dirname(). Neither is used from balloon.c.
In fact it can be dangerous to include libgen.h,
as GNU basename() can be replaced with POSIX basename()
and we might rely on GNU version. As neither is used
from balloon.c it's not a problem this time.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop defragmentation support #PSBM-30163ext4 defragmentation is part of 'ploop balloon discar' action
If defrag functionality requested the /usr/sbin/e4defrag2 utility is called.
To terminate ploop defragmentation the pid of running e4defrag2 is stored at
'/var/lock/ploop/ploopN.defrag.pid' and will used to terminate defragmentation
process
Signed-off-by: Igor Sukhih <igor@parallels.com>
copy_delta(): avoid double-freeAs pointed out by Coverity (CID 1258241), when we are reallocating
buf and the second p_memalign() fails, free(buf) is called twice.
Make sure to set buf to NULL after free() to avoid it.
While at it, there's no need to check buf for NULL before calling
free(), as free(NULL) is totally valid.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_merge_snapshot_by_guid(): remove merge_modeGet rid of PLOOP_MERGE_WITH_CHILD|PLOOP_MERGE_WITH_PARENT logic.
Now guid is snapshot guid which is to be merged to its parent.
Based on patch by Igor Sukhih:
https://src.openvz.org/projects/OVZ/repos/ploop/commits/b07589212937e277ab49aa27c26a02cdd23ada82
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_delete_snapshot(): don't use PLOOP_MERGE_WITH_CHILDThis is to simplify the logic in ploop_merge_snapshot_by_guid().
Based on patch by Igor Sukhih, see
https://src.openvz.org/projects/OVZ/repos/ploop/commits/ec8d2cd3ef103e00b10711cb67524f8236f14815
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
make_fs(): drop lazy_journal_initCommit d496ba8 introduced somewhat complex retry procedure, and commits
b7d4766 and 4501da4 further improved it.
This patch reverts it all, instead just removing lazy_journal_init option
(which is not understood by older mke2fs as pointed out in #OVZ-6539).
https://bugs.openvz.org/browse/OVZ-6539
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
mount_image: return SYSEXIT_NOSNAP error, #PSBM-38423If snapshot does not exist provide specific error to caller.
Will be used to not fail mount snapshot operation
Signed-off-by: Igor Sukhih <igor@parallels.com>
ploop_get_devs: fix NULL pointer dereference0 0x00007f4a8d0786bb in ploop_get_devs (di=0x13690d0,
out=out@entry=0x7fff8710a2f8) at ploop.c:1205
1205 return ploop_get_dev_by_delta(di->images[0]->file, NULL, NULL,
out);
ploop_open_dd() does not load all metadata so it needed to call
ploop_lock_dd or ploop_read_dd() to update di->images
Reference: PSBM-40203
Signed-off-by: Igor Sukhih <igor@parallels.com>
make_fs(): fix a commentA comment mistakenly says about lazy_itable_init. It should be
lazy_journal_init. Fix accordingly.
While at it, break a particularly long line.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Kir Kolyshkin
