We have seen isolated cases (very few, I might add) of o2hb not detecting all
live nodes on startup. One plausible reasoning for it is that other node had
a hb io delay at the same time. The live threshold set at 2 (as low as it can
be) could be increased to ameliorate the situation.
But increasing the threshold directly affects mount time. Currently it takes
around 5 secs to mount a volume in o2cb cluster with local heartbeat. Increasing
the threshold will make mounts even slower. As the issue itself is rare, we have
left things as they are for the local heartbeat mode.
However we can improve the situation for global heartbeat mode as in that mode,
we start the heartbeat much before the mount.
This patch doubles the live threshold for the start of the first region in
global heartbeat mode.
Addresses internal Oracle bug#10635585.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Acked-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
Patch fixes a bug in the o2dlm protocol negotiation in that it is using
the builtin version rather than the negotiated version during the domain
join. This causes join errors when a node having kernel >= 2.6.37 joins
a cluster with nodes having kernels < 2.6.37.
This only affects the o2cb cluster stack.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Reported-by: Jacek Stepniewski <Jacek.Stepniewski@agora.pl>
Acked-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
In the case of removing a partial extent record which covers a hole, current
punching-hole logic will try to remove more than the length of whole extent
record, which leads to the failure of following assert(fs/ocfs2/alloc.c):
5507 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
This patch tries to skip existing hole at the last attempt of removing a partial
extent record, what's more, it also adds some necessary comments for better
understanding of punching-hole codes.
Signed-off-by: Tristan Ye <tristan.ye@oracle.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
CLANG found that there is a path that has data_ac uninitialized,
this place
2917 /* This gets us the dx_root */
2918 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2919 if (ret) {
3
Taking true branch
2920 mlog_errno(ret);
2921 goto out;
4
Control jumps to line 3168
2922 }
Goes to the out: label without data_ac being initialized.
Ciao, Marcus
Signed-Off-By: Marcus Meissner <meissner@suse.de>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
When re-mounting from R/O mode to R/W mode and the LEB count in the superblock
is not up-to date, because for the underlying UBI volume became larger, we
re-write the superblock. We allocate RAM for these purposes, but never free it.
So this is a memory leak, although very rare one.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Cc: stable@kernel.org
This patch fixes a problem with the following symptoms:
UBIFS: deferred recovery completed
UBIFS error (pid 15676): dbg_check_synced_i_size: ui_size is 11481088, synced_i_size is 11459081, but inode is clean
UBIFS error (pid 15676): dbg_check_synced_i_size: i_ino 128, i_mode 0x81a4, i_size 11481088
It happens when additional debugging checks are enabled and we are recovering
from a power cut. When we fixup corrupted inode size during recovery, we change
them in-place and we change ui_size as well, but not synced_i_size, which
causes this failure. This patch makes sure we change both fields and fixes the
issue.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When the debugging self-checks are enabled, we go trough whole file-system
after mount and check/validate every single node referred to by the index.
This is implemented by the 'dbg_check_filesystem()' function. However, this
function fails if we mount "unclean" file-system, i.e., if we mount the
file-system after a power cut. It fails with the following symptoms:
UBIFS DBG (pid 8171): ubifs_recover_size: ino 937 size 3309925 -> 3317760
UBIFS: recovery deferred
UBIFS error (pid 8171): check_leaf: data node at LEB 1000:0 is not within inode size 3309925
The reason of failure is that recovery fixed up the inode size in memory, but
not on the flash so far. So the value on the flash is incorrect so far,
and would be corrected when we re-mount R/W. But 'check_leaf()' ignores
this fact and tries to validate the size of the on-flash inode, which is
incorrect, so it fails.
This patch teaches the checking code to look at the VFS inode cache first,
and if there is the inode in question, use that inode instead of the inode
on the flash media. This fixes the issue.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
In 'ubifs_recover_size()' we have an "if (!e->inode && c->ro_mount)" statement.
But if 'c->ro_mount' is true, then '!e->inode' must always be true as well. So
we can remove the unnecessary '!e->inode' test and put an
'ubifs_assert(!e->inode)' instead.
This patch also removes an extra trailing white-space in a debugging print,
as well as adds few empty lines to 'ubifs_recover_size()' to make it a bit more
readable.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When recovering the inode size, one of the debugging messages was printed
incorrecly, this patches fixes it.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This commits refactors and cleans up 'ubifs_rcvry_gc_commit()' which was quite
untidy, also removes the commentary which was not 100% correct.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Split the 'ubifs_rcvry_gc_commit()' function and introduce a 'grab_empty_leb()'
heler. This cleans 'ubifs_rcvry_gc_commit()' a little and makes it a bit less
of spagetti.
Also, add a commentary which explains why it is crucial to first search for an
empty LEB and then run commit.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When UBIFS is in the failure mode (used for power cut emulation testing) we for
some reasons do not dump the stack in many places, e.g., in assertions.
Probably at early days we had too many of them and disabled this to make the
development easier, but then never enabled. Nowadays I sometimes observe
assertion failures during power cut testing, but the useful stackdump is not
printed, which is bad. This patch makes UBIFS always print the stackdump when
debugging is enabled.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
If we fail to recover the gc_lnum we just return an error and it then
it is difficult to figure out why this happened. This patch adds useful
debugging information which should make it easier to debug the failure.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This patch removes a piece of code in 'ubifs_rcvry_gc_commit()' which is never
executed. We call 'ubifs_find_dirty_leb()' function with min_space =
wbuf->offs, so if it returns us an LEB, it is guaranteed to have at lease
'wbuf->offs' bytes of free+dirty space. So we can remove the subsequent code
which deals with "returned LEB has less than 'wbuf->offs' bytes of free+dirty
space". This simplifies 'ubifs_rcvry_gc_commit()' a little.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
We have duplicated code in 'ubifs_garbage_collect()' and
'ubifs_rcvry_gc_commit()', which is about handling the special case of free
LEB. In both cases we just want to garbage-collect the LEB using
'ubifs_garbage_collect_leb()'.
This patch teaches 'ubifs_garbage_collect_leb()' to handle free LEB's so that
the caller does not have to do this and the duplicated code is removed.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Remove the following commentary from 'ubifs_file_mmap()':
/* 'generic_file_mmap()' takes care of NOMMU case */
I do not understand what it means, and I could not find anything relater to
NOMMU in 'generic_file_mmap()'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This patch is a tiny improvement which removes few bytes of code.
UBIFS debugfs files are non-seekable and the file position is ignored,
so do not increase it in the write handler.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
The 'dbg_dump_lprop()' is trying to detect journal head LEBs when printing,
so it looks at the write-buffers. However, if we are in R/O mode, we
de-allocate the write-buffers, so 'dbg_dump_lprop()' oopses. This patch fixes
the issue.
Note, this patch is not critical, it is only about the debugging code path, and
it is unlikely that anyone but UBIFS developers would ever hit this issue.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
We have our own flags indicating R/O mode, and c->ro_mode is equivalent
to MS_RDONLY. Let's be consistent and use UBIFS flags everywhere.
This patch is just a minor cleanup.
Additionally, add a comment that we are surprised with VFS behavior -
as a reminder to look at this some day.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When the debugging failure emulation is enabled and UBIFS decides to
emulate an I/O error, it uses EIO error code. In which case UBIFS
switches into R/O mode later on. The for the user-space is that when
a failure is emulated, the file-system sometimes returns EIO and
sometimes EROFS. This makes it more difficult to implement user-space
tests for the failure mode. Let's be consistent and return EROFS in
all the cases.
This patch is an improvement for the debugging code and does not affect
the functionality at all.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This is just a tiny clean-up patch. The variable name for empty address
space operations is "empty_aops". Let's use consistent names for empty
inode and file operations: "empty_iops" and "empty_fops", instead of
inconsistent "none_inode_operations" and "none_file_operations".
Artem: re-write the commit message.
Signed-off-by: Sedat Dilek <sedat.dilek@gmail.com>
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Try to improve UBIFS testing coverage by randomly picking LEBs to
store in lsave, rather than picking them optimally. Create a debugging
version of 'populate_lsave()' for these purposes and enable it when
general debugging self-checks are enabled.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
UBIFS can force itself to use the 'in-the-gaps' commit method - the last resort
method which is normally invoced very very rarely. Currently this "force
int-the-gaps" debugging feature is a separate test mode. But it is a bit saner
to make it to be the "general" self-test check instead.
This patch is just a clean-up which should make the debugging code look a bit
nicer and easier to use - we have way too many debugging options.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This patch improves the 'dbg_check_space_info()' function which checks
whether the amount of space before re-mounting and after re-mounting
is the same (remounting from R/O to R/W modes and vice-versa).
The problem is that 'dbg_check_space_info()' does not save the budgeting
information before re-mounting, so when an error is reported, we do not
know why the amount of free space changed.
This patches makes the following changes:
1. Teaches 'dbg_dump_budg()' function to accept a 'struct ubifs_budg_info'
argument and print out the this argument. This way we may ask it to
print any saved budgeting info, no only the current one.
2. Accordingly changes all the callers of 'dbg_dump_budg()' to comply with
the changed interface.
3. Introduce a 'saved_bi' (saved budgeting info) field to
'struct ubifs_debug_info' and save the budgeting info before re-mounting
there.
4. Change 'dbg_check_space_info()' and make it print both old and new
budgeting information.
5. Additionally, save 'c->igx_gc_cnt' and print it if and error happens. This
value contributes to the amount of free space, so we have to print it.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Re-arrange the budget dump and make sure we first dump all
the 'struct ubifs_budg_info' fields, and then the other information.
Additionally, print the 'uncommitted_idx' variable.
This change is required for to the following dumping function
enhancement where it will be possible to dump saved
'struct ubifs_budg_info' objects, not only the current one.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
The current 'dbg_dump_budg()' calling convention is that the
'c->space_lock' spinlock is held. However, none of the callers
actually use it from contects which have 'c->space_lock' locked,
so all callers have to explicitely lock and unlock the spinlock.
This is not very sensible convention. This patch changes it and
makes 'dbg_dump_budg()' lock the spinlock instead of imposing this
to the callers. This simplifies the code a little.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This patch separates out all the budgeting-related information
from 'struct ubifs_info' to 'struct ubifs_budg_info'. This way the
code looks a bit cleaner. However, the main driver for this is
that we want to save budgeting information and print it later,
so a separate data structure for this is helpful.
This patch is a preparation for the further debugging output
improvements.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
There was an attempt to standartize various "__attribute__" and
other macros in order to have potentially portable and more
consistent code, see commit 82ddcb0405.
Note, that commit refers Rober Love's blog post, but the URL
is broken, the valid URL is:
http://blog.rlove.org/2005/10/with-little-help-from-your-compiler.html
Moreover, nowadays checkpatch.pl warns about using
__attribute__((packed)):
"WARNING: __packed is preferred over __attribute__((packed))"
It is not a big deal for UBIFS to use __packed, so let's do it.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Fix several minor stylistic issues:
* lines longer than 80 characters
* space before closing parenthesis ')'
* spaces in the indentations
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Turn the debufs files UBIFS maintains into non-seekable. Indeed, none
of them is supposed to be seek'ed.
Do this by making the '.lseek()' handler to be 'no_llseek()' and by
using 'nonseekable_open()' in the '.open()' operation.
This does mean an API break but this debugging API is only used by a couple
of test scripts which do not rely in the 'llseek()' operation.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
In a multi device setup, the chunk allocator currently always allocates
chunks on the devices in the same order. This leads to a very uneven
distribution, especially with RAID1 or RAID10 and an uneven number of
devices.
This patch always sorts the devices before allocating, and allocates the
stripes on the devices with the most available space, as long as there
is enough space available. In a low space situation, it first tries to
maximize striping.
The patch also simplifies the allocator and reduces the checks for
corner cases.
The simplification is done by several means. First, it defines the
properties of each RAID type upfront. These properties are used afterwards
instead of differentiating cases in several places.
Second, the old allocator defined a minimum stripe size for each block
group type, tried to find a large enough chunk, and if this fails just
allocates a smaller one. This is now done in one step. The largest possible
chunk (up to max_chunk_size) is searched and allocated.
Because we now have only one pass, the allocation of the map (struct
map_lookup) is moved down to the point where the number of stripes is
already known. This way we avoid reallocation of the map.
We still avoid allocating stripes that are not a multiple of STRIPE_SIZE.
currently alloc_start is disregarded if the requested
chunk size is bigger than (device size - alloc_start),
but smaller than the device size.
The only situation where I see this could have made sense
was when a chunk equal the size of the device has been
requested. This was possible as the allocator failed to
take alloc_start into account when calculating the request
chunk size. As this gets fixed by this patch, the workaround
is not necessary anymore.
Now that there are no longer any exceptions to the normal inode
creation code path, we can move the parts of the locking code
which were duplicated in mkdir/mknod/create/symlink into the
inode create function.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
This moves the symlink specific parts of inode creation
into the function where we initialise the rest of the
dinode. As a result we have one less place where we need
to look up the inode's buffer.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
This moves the initialisation of the directory into the inode
creation functions to avoid having to duplicate the lookup
of the inode's buffer.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
btrfs scrub - make fixups sync, don't reuse fixup bios
Fixups are already sync for csum failures, this patch makes them sync
for EIO case as well.
Fixups are now sharing pages with the parent sbio - instead of
allocating a separate page to do a fixup we grab the page from the sbio
buffer.
Fixup bios are no longer reused.
struct fixup is no longer needed, instead pass [sbio pointer, index].
Originally this was added to look at the possibility of sharing the code
between drive swap and scrub, but it actually fixes a serious bug in
scrub code where errors that could be corrected were ignored and
reported as uncorrectable.
btrfs scrub - restore bios properly after media errors
The current code reallocates a bio after a media error. This is a
temporary measure introduced in v3 after a serious problem related to
bio reuse was found in v2 of scrub patchset.
Basically we did not reset bv_offset and bv_len fields of the bio_vec
structure. They are changed in case I/O error happens, for example, at
offset 512 or 1024 into the page. Also bi_flags field wasn't properly
setup before reusing the bio.
Signed-off-by: Arne Jansen <sensille@gmx.net>
adds ioctls necessary to start and cancel scrubs, to get current
progress and to get info about devices to be scrubbed.
Note that the scrub is done per-device and that the ioctl only
returns after the scrub for this devices is finished or has been
canceled.
Signed-off-by: Arne Jansen <sensille@gmx.net>
This adds an initial implementation for scrub. It works quite
straightforward. The usermode issues an ioctl for each device in the
fs. For each device, it enumerates the allocated device chunks. For
each chunk, the contained extents are enumerated and the data checksums
fetched. The extents are read sequentially and the checksums verified.
If an error occurs (checksum or EIO), a good copy is searched for. If
one is found, the bad copy will be rewritten.
All enumerations happen from the commit roots. During a transaction
commit, the scrubs get paused and afterwards continue from the new
roots.
This commit is based on the series originally posted to linux-btrfs
with some improvements that resulted from comments from David Sterba,
Ilya Dryomov and Jan Schmidt.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Currently, writebacks may end up recursing back into the filesystem due to
GFP_KERNEL direct reclaims in the pnfs subsystem.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
ceph: do not use i_wrbuffer_ref as refcount for Fb cap
ceph: fix list_add in ceph_put_snap_realm
ceph: print debug message before put mds session
Prevents an infinite loop as list was never emptied.
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Free the slot and resend the RPC with new session <slot#,seq#>.
For nfs4_async_handle_error, return -EAGAIN and set the task->tk_status to 0
to restart the async rpc in the rpc_restart_call_prepare state which resets
the slot.
For nfs4_handle_exception, retrying a call that uses nfs4_call_sync will
reset the slot via nfs41_call_sync_prepare.
For open/close/lock/locku/delegreturn/layoutcommit/unlink/rename/write
cachethis is true, so these operations will not trigger an
NFS4ERR_RETRY_UNCACHED_REP.
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
We increments i_wrbuffer_ref when taking the Fb cap. This breaks
the dirty page accounting and causes looping in
__ceph_do_pending_vmtruncate, and ceph client hangs.
This bug can be reproduced occasionally by running blogbench.
Add a new field i_wb_ref to inode and dedicate it to Fb reference
counting.
Signed-off-by: Henry C Chang <henry.cy.chang@gmail.com>
Signed-off-by: Sage Weil <sage@newdream.net>
The mds session, s, could be freed during ceph_put_mds_session.
Move dout before ceph_put_mds_session.
Signed-off-by: Henry C Chang <henry.cy.chang@gmail.com>
Signed-off-by: Sage Weil <sage@newdream.net>
