IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Previously we would always try to enqueue work even if the filesystem is not
mounted with fscache enabled (or the file has no cookie). In the case of the
filesystem mouned nofsc (but with fscache compiled in) this would lead to a
crash.
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Previous patch that allowed us to cleanup most of the issues with pages marked
as private_2 when calling ceph_readpages. However, there seams to be a case in
the error case clean up in start read that still trigers this from time to
time. I've only seen this one a couple times.
BUG: Bad page state in process petabucket pfn:335b82
page:ffffea000cd6e080 count:0 mapcount:0 mapping: (null) index:0x0
page flags: 0x200000000001000(private_2)
Call Trace:
[<ffffffff81563442>] dump_stack+0x46/0x58
[<ffffffff8112c7f7>] bad_page+0xc7/0x120
[<ffffffff8112cd9e>] free_pages_prepare+0x10e/0x120
[<ffffffff8112e580>] free_hot_cold_page+0x40/0x160
[<ffffffff81132427>] __put_single_page+0x27/0x30
[<ffffffff81132d95>] put_page+0x25/0x40
[<ffffffffa02cb409>] ceph_readpages+0x2e9/0x6f0 [ceph]
[<ffffffff811313cf>] __do_page_cache_readahead+0x1af/0x260
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Signed-off-by: Sage Weil <sage@inktank.com>
In some cases the ceph readapages code code bails without filling all the pages
already marked by fscache. When we return back to readahead code this causes
a BUG.
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Adding support for fscache to the Ceph filesystem. This would bring it to on
par with some of the other network filesystems in Linux (like NFS, AFS, etc...)
In order to mount the filesystem with fscache the 'fsc' mount option must be
passed.
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Following we will begin to add memcg dirty page accounting around
__set_page_dirty_{buffers,nobuffers} in vfs layer, so we'd better use vfs interface to
avoid exporting those details to filesystems.
Since vfs set_page_dirty() should be called under page lock, here we don't need elaborate
codes to handle racy anymore, and two WARN_ON() are added to detect such exceptions.
Thanks very much for Sage and Yan Zheng's coaching!
I tested it in a two server's ceph environment that one is client and the other is
mds/osd/mon, and run the following fsx test from xfstests:
./fsx 1MB -N 50000 -p 10000 -l 1048576
./fsx 10MB -N 50000 -p 10000 -l 10485760
./fsx 100MB -N 50000 -p 10000 -l 104857600
The fsx does lots of mmap-read/mmap-write/truncate operations and the tests completed
successfully without triggering any of WARN_ON.
Signed-off-by: Sha Zhengju <handai.szj@taobao.com>
Reviewed-by: Sage Weil <sage@inktank.com>
For sync_read/write, it may do multi stripe operations.If one of those
met erro, we return the former successed size rather than a error value.
There is a exception for write-operation met -EOLDSNAPC.If this occur,we
retry the whole write again.
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
cephfs . show_layout
>layyout.data_pool: 0
>layout.object_size: 4194304
>layout.stripe_unit: 4194304
>layout.stripe_count: 1
TestA:
>dd if=/dev/urandom of=test bs=1M count=2 oflag=direct
>dd if=/dev/urandom of=test bs=1M count=2 seek=4 oflag=direct
>dd if=test of=/dev/null bs=6M count=1 iflag=direct
The messages from func striped_read are:
ceph: file.c:350 : striped_read 0~6291456 (read 0) got 2097152 HITSTRIPE SHORT
ceph: file.c:350 : striped_read 2097152~4194304 (read 2097152) got 0 HITSTRIPE SHORT
ceph: file.c:381 : zero tail 4194304
ceph: file.c:390 : striped_read returns 6291456
The hole of file is from 2M--4M.But actualy it zero the last 4M include
the last 2M area which isn't a hole.
Using this patch, the messages are:
ceph: file.c:350 : striped_read 0~6291456 (read 0) got 2097152 HITSTRIPE SHORT
ceph: file.c:358 : zero gap 2097152 to 4194304
ceph: file.c:350 : striped_read 4194304~2097152 (read 4194304) got 2097152
ceph: file.c:384 : striped_read returns 6291456
TestB:
>echo majianpeng > test
>dd if=test of=/dev/null bs=2M count=1 iflag=direct
The messages are:
ceph: file.c:350 : striped_read 0~6291456 (read 0) got 11 HITSTRIPE SHORT
ceph: file.c:350 : striped_read 11~6291445 (read 11) got 0 HITSTRIPE SHORT
ceph: file.c:390 : striped_read returns 11
For this case,it did once more striped_read.It's no meaningless.
Using this patch, the message are:
ceph: file.c:350 : striped_read 0~6291456 (read 0) got 11 HITSTRIPE SHORT
ceph: file.c:384 : striped_read returns 11
Big thanks to Yan Zheng for the patch.
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
This patch implements fallocate and punch hole support for Ceph kernel client.
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
ceph_check_caps() requests new max size only when there is Fw cap.
If we call check_max_size() while there is no Fw cap. It updates
i_wanted_max_size and calls ceph_check_caps(), but ceph_check_caps()
does nothing. Later when Fw cap is issued, we call check_max_size()
again. But i_wanted_max_size is equal to 'endoff' at this time, so
check_max_size() doesn't call ceph_check_caps() and we end up with
waiting for the new max size forever.
The fix is duplicate ceph_check_caps()'s "request max size" code in
check_max_size(), and make try_get_cap_refs() wait for the Fw cap
before retry requesting new max size.
This patch also removes the "endoff > (inode->i_size << 1)" check
in check_max_size(). It's useless because there is no corresponding
logic in ceph_check_caps().
Reviewed-by: Sage Weil <sage@inktank.com>
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
I encountered below deadlock when running fsstress
wmtruncate work truncate MDS
--------------- ------------------ --------------------------
lock i_mutex
<- truncate file
lock i_mutex (blocked)
<- revoking Fcb (filelock to MIX)
send request ->
handle request (xlock filelock)
At the initial time, there are some dirty pages in the page cache.
When the kclient receives the truncate message, it reduces inode size
and creates some 'out of i_size' dirty pages. wmtruncate work can't
truncate these dirty pages because it's blocked by the i_mutex. Later
when the kclient receives the cap message that revokes Fcb caps, It
can't flush all dirty pages because writepages() only flushes dirty
pages within the inode size.
When the MDS handles the 'truncate' request from kclient, it waits
for the filelock to become stable. But the filelock is stuck in
unstable state because it can't finish revoking kclient's Fcb caps.
The truncate pagecache locking has already caused lots of trouble
for use. I think it's time simplify it by introducing a new mutex.
We use the new mutex to prevent concurrent truncate_inode_pages().
There is no need to worry about race between buffered write and
truncate_inode_pages(), because our "get caps" mechanism prevents
them from concurrent execution.
Reviewed-by: Sage Weil <sage@inktank.com>
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
The invalidatepage code bails if it encounters a non-zero page offset. The
current logic that does is non-obvious with multiple if statements.
This should be logically and functionally equivalent.
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The early bug checks are moot because the VMA layer ensures those things.
1. It will not call invalidatepage unless PagePrivate (or PagePrivate2) are set
2. It will not call invalidatepage without taking a PageLock first.
3. Guantrees that the inode page is mapped.
Signed-off-by: Milosz Tanski <milosz@adfin.com>
Reviewed-by: Sage Weil <sage@inktank.com>
All of the early exit paths need to drop the mutex; it is only the normal
path through the function that does not. Skip the unlock in that case
with a goto out_unlocked.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Jianpeng Ma <majianpeng@gmail.com>
Only for ceph_sync_write, the osd can return EOLDSNAPC.so move the
related codes after the call ceph_sync_write.
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
remove_session_caps() uses iterate_session_caps() to remove caps,
but iterate_session_caps() skips inodes that are being deleted.
So session->s_nr_caps can be non-zero after iterate_session_caps()
return.
We can fix the issue by waiting until deletions are complete.
__wait_on_freeing_inode() is designed for the job, but it is not
exported, so we use lookup inode function to access it.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Func ceph_calc_ceph_pg maybe failed.So add check for returned value.
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Sending reads and writes through the sync read/write paths bypasses the
page cache, which is not expected or generally a good idea. Removing
the write check is safe as there is a conditional vfs_fsync_range() later
in ceph_aio_write that already checks for the same flag (via
IS_SYNC(inode)).
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
We pass in a u64 value for "len" and then immediately truncate away the
upper 32 bits.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <alex.elder@linaro.org>
The MDS uses caps message to notify clients about deleted inode.
when receiving a such message, invalidate any alias of the inode.
This makes the kernel release the inode ASAP.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
To write data, the writer first acquires the i_mutex, then try getting
caps. The writer may sleep while holding the i_mutex. If the MDS revokes
Fb cap in this case, vmtruncate work can't do its job because i_mutex
is locked. We should wake up the writer and let it truncate the pages.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
To handle "link" request, the MDS need to xlock inode's linklock,
which requires revoking any CAP_LINK_SHARED.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
When register_session() is given an out-of-range argument for mds,
ceph_mdsmap_get_addr() will return a null pointer, which would be given to
ceph_con_open() & be dereferenced, causing a kernel oops. This fixes bug #4685
in the Ceph bug tracker <http://tracker.ceph.com/issues/4685>.
Signed-off-by: Nathaniel Yazdani <n1ght.4nd.d4y@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pull Ceph updates from Sage Weil:
"There is some follow-on RBD cleanup after the last window's code drop,
a series from Yan fixing multi-mds behavior in cephfs, and then a
sprinkling of bug fixes all around. Some warnings, sleeping while
atomic, a null dereference, and cleanups"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client: (36 commits)
libceph: fix invalid unsigned->signed conversion for timespec encoding
libceph: call r_unsafe_callback when unsafe reply is received
ceph: fix race between cap issue and revoke
ceph: fix cap revoke race
ceph: fix pending vmtruncate race
ceph: avoid accessing invalid memory
libceph: Fix NULL pointer dereference in auth client code
ceph: Reconstruct the func ceph_reserve_caps.
ceph: Free mdsc if alloc mdsc->mdsmap failed.
ceph: remove sb_start/end_write in ceph_aio_write.
ceph: avoid meaningless calling ceph_caps_revoking if sync_mode == WB_SYNC_ALL.
ceph: fix sleeping function called from invalid context.
ceph: move inode to proper flushing list when auth MDS changes
rbd: fix a couple warnings
ceph: clear migrate seq when MDS restarts
ceph: check migrate seq before changing auth cap
ceph: fix race between page writeback and truncate
ceph: reset iov_len when discarding cap release messages
ceph: fix cap release race
libceph: fix truncate size calculation
...
If we receive new caps from the auth MDS and the non-auth MDS is
revoking the newly issued caps, we should release the caps from
the non-auth MDS. The scenario is filelock's state changes from
SYNC to LOCK. Non-auth MDS revokes Fc cap, the client gets Fc cap
from the auth MDS at the same time.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
If caps are been revoking by the auth MDS, don't consider them as
issued even they are still issued by non-auth MDS. The non-auth
MDS should also be revoking/exporting these caps, the client just
hasn't received the cap revoke/export message.
The race I encountered is: When caps are exporting to new MDS, the
client receives cap import message and cap revoke message from the
new MDS, then receives cap export message from the old MDS. When
the client receives cap revoke message from the new MDS, the revoking
caps are still issued by the old MDS, so the client does nothing.
Later when the cap export message is received, the client removes
the caps issued by the old MDS. (Another way to fix the race is
calling ceph_check_caps() in handle_cap_export())
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The locking order for pending vmtruncate is wrong, it can lead to
following race:
write wmtruncate work
------------------------ ----------------------
lock i_mutex
check i_truncate_pending check i_truncate_pending
truncate_inode_pages() lock i_mutex (blocked)
copy data to page cache
unlock i_mutex
truncate_inode_pages()
The fix is take i_mutex before calling __ceph_do_pending_vmtruncate()
Fixes: http://tracker.ceph.com/issues/5453
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Drop ignored return value. Fix allocation failure case to not leak.
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Either in vfs_write or io_submit,it call file_start/end_write.
The different between file_start/end_write and sb_start/end_write is
file_ only handle regular file.But i think in ceph_aio_write,it only
for regular file.
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Acked-by: Yan, Zheng <zheng.z.yan@intel.com>
We may receive old request reply from the exporter MDS after receiving
the importer MDS' cap import message.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The client can receive truncate request from MDS at any time.
So the page writeback code need to get i_size, truncate_seq and
truncate_size atomically
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
ceph_encode_inode_release() can race with ceph_open() and release
caps wanted by open files. So it should call __ceph_caps_wanted()
to get the wanted caps.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pull second set of VFS changes from Al Viro:
"Assorted f_pos race fixes, making do_splice_direct() safe to call with
i_mutex on parent, O_TMPFILE support, Jeff's locks.c series,
->d_hash/->d_compare calling conventions changes from Linus, misc
stuff all over the place."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (63 commits)
Document ->tmpfile()
ext4: ->tmpfile() support
vfs: export lseek_execute() to modules
lseek_execute() doesn't need an inode passed to it
block_dev: switch to fixed_size_llseek()
cpqphp_sysfs: switch to fixed_size_llseek()
tile-srom: switch to fixed_size_llseek()
proc_powerpc: switch to fixed_size_llseek()
ubi/cdev: switch to fixed_size_llseek()
pci/proc: switch to fixed_size_llseek()
isapnp: switch to fixed_size_llseek()
lpfc: switch to fixed_size_llseek()
locks: give the blocked_hash its own spinlock
locks: add a new "lm_owner_key" lock operation
locks: turn the blocked_list into a hashtable
locks: convert fl_link to a hlist_node
locks: avoid taking global lock if possible when waking up blocked waiters
locks: protect most of the file_lock handling with i_lock
locks: encapsulate the fl_link list handling
locks: make "added" in __posix_lock_file a bool
...
For those file systems(btrfs/ext4/ocfs2/tmpfs) that support
SEEK_DATA/SEEK_HOLE functions, we end up handling the similar
matter in lseek_execute() to update the current file offset
to the desired offset if it is valid, ceph also does the
simliar things at ceph_llseek().
To reduce the duplications, this patch make lseek_execute()
public accessible so that we can call it directly from the
underlying file systems.
Thanks Dave Chinner for this suggestion.
[AV: call it vfs_setpos(), don't bring the removed 'inode' argument back]
v2->v1:
- Add kernel-doc comments for lseek_execute()
- Call lseek_execute() in ceph->llseek()
Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chris Mason <chris.mason@fusionio.com>
Cc: Josef Bacik <jbacik@fusionio.com>
Cc: Ben Myers <bpm@sgi.com>
Cc: Ted Tso <tytso@mit.edu>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Sage Weil <sage@inktank.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
category, of note is a fix for on-line resizing file systems where the
block size is smaller than the page size (i.e., file systems 1k blocks
on x86, or more interestingly file systems with 4k blocks on Power or
ia64 systems.)
In the cleanup category, the ext4's punch hole implementation was
significantly improved by Lukas Czerner, and now supports bigalloc
file systems. In addition, Jan Kara significantly cleaned up the
write submission code path. We also improved error checking and added
a few sanity checks.
In the optimizations category, two major optimizations deserve
mention. The first is that ext4_writepages() is now used for
nodelalloc and ext3 compatibility mode. This allows writes to be
submitted much more efficiently as a single bio request, instead of
being sent as individual 4k writes into the block layer (which then
relied on the elevator code to coalesce the requests in the block
queue). Secondly, the extent cache shrink mechanism, which was
introduce in 3.9, no longer has a scalability bottleneck caused by the
i_es_lru spinlock. Other optimizations include some changes to reduce
CPU usage and to avoid issuing empty commits unnecessarily.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)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=SwaI
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 update from Ted Ts'o:
"Lots of bug fixes, cleanups and optimizations. In the bug fixes
category, of note is a fix for on-line resizing file systems where the
block size is smaller than the page size (i.e., file systems 1k blocks
on x86, or more interestingly file systems with 4k blocks on Power or
ia64 systems.)
In the cleanup category, the ext4's punch hole implementation was
significantly improved by Lukas Czerner, and now supports bigalloc
file systems. In addition, Jan Kara significantly cleaned up the
write submission code path. We also improved error checking and added
a few sanity checks.
In the optimizations category, two major optimizations deserve
mention. The first is that ext4_writepages() is now used for
nodelalloc and ext3 compatibility mode. This allows writes to be
submitted much more efficiently as a single bio request, instead of
being sent as individual 4k writes into the block layer (which then
relied on the elevator code to coalesce the requests in the block
queue). Secondly, the extent cache shrink mechanism, which was
introduce in 3.9, no longer has a scalability bottleneck caused by the
i_es_lru spinlock. Other optimizations include some changes to reduce
CPU usage and to avoid issuing empty commits unnecessarily."
* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (86 commits)
ext4: optimize starting extent in ext4_ext_rm_leaf()
jbd2: invalidate handle if jbd2_journal_restart() fails
ext4: translate flag bits to strings in tracepoints
ext4: fix up error handling for mpage_map_and_submit_extent()
jbd2: fix theoretical race in jbd2__journal_restart
ext4: only zero partial blocks in ext4_zero_partial_blocks()
ext4: check error return from ext4_write_inline_data_end()
ext4: delete unnecessary C statements
ext3,ext4: don't mess with dir_file->f_pos in htree_dirblock_to_tree()
jbd2: move superblock checksum calculation to jbd2_write_superblock()
ext4: pass inode pointer instead of file pointer to punch hole
ext4: improve free space calculation for inline_data
ext4: reduce object size when !CONFIG_PRINTK
ext4: improve extent cache shrink mechanism to avoid to burn CPU time
ext4: implement error handling of ext4_mb_new_preallocation()
ext4: fix corruption when online resizing a fs with 1K block size
ext4: delete unused variables
ext4: return FIEMAP_EXTENT_UNKNOWN for delalloc extents
jbd2: remove debug dependency on debug_fs and update Kconfig help text
jbd2: use a single printk for jbd_debug()
...
I introduced a new temporary variable "info" instead of
"m->m_info[mds]". Also I reversed the if condition and pulled
everything in one indent level.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Alex Elder <elder@inktank.com>
This patch makes the following improvements to the error handling
in the ceph_mdsmap_decode function:
- Add a NULL check for return value from kcalloc
- Make use of the variable err
Signed-off-by: Emil Goode <emilgoode@gmail.com>
Signed-off-by: Sage Weil <sage@inktank.com>
