All the code around struct xfs_bmap_free basically implements a
deferred operation framework through which we can roll transactions
(to unlock buffers and avoid violating lock order rules) while
managing all the necessary log redo items. Previously we only used
this code to free extents after some sort of mapping operation, but
with the advent of rmap and reflink, we suddenly need to do more than
that.
With that in mind, xfs_bmap_free really becomes a deferred ops control
structure. Rename the structure and move the deferred ops into their
own file to avoid further bloating of the bmap code.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Refactor the btree_change_owner function into a more generic apparatus
which visits all blocks in a btree. We'll use this in a subsequent
patch for counting btree blocks for AG reservations.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Create a function to enable querying of btree records mapping to a
range of keys. This will be used in subsequent patches to allow
querying the reverse mapping btree to find the extents mapped to a
range of physical blocks, though the generic code can be used for
any range query.
The overlapped query range function needs to use the btree get_block
helper because the root block could be an inode, in which case
bc_bufs[nlevels-1] will be NULL.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
On a filesystem with both reflink and reverse mapping enabled, it's
possible to have multiple rmap records referring to the same blocks on
disk. When overlapping intervals are possible, querying a classic
btree to find all records intersecting a given interval is inefficient
because we cannot use the left side of the search interval to filter
out non-matching records the same way that we can use the existing
btree key to filter out records coming after the right side of the
search interval. This will become important once we want to use the
rmap btree to rebuild BMBTs, or implement the (future) fsmap ioctl.
(For the non-overlapping case, we can perform such queries trivially
by starting at the left side of the interval and walking the tree
until we pass the right side.)
Therefore, extend the btree code to come closer to supporting
intervals as a first-class record attribute. This involves widening
the btree node's key space to store both the lowest key reachable via
the node pointer (as the btree does now) and the highest key reachable
via the same pointer and teaching the btree modifying functions to
keep the highest-key records up to date.
This behavior can be turned on via a new btree ops flag so that btrees
that cannot store overlapping intervals don't pay the overhead costs
in terms of extra code and disk format changes.
When we're deleting a record in a btree that supports overlapped
interval records and the deletion results in two btree blocks being
joined, we defer updating the high/low keys until after all possible
joining (at higher levels in the tree) have finished. At this point,
the btree pointers at all levels have been updated to remove the empty
blocks and we can update the low and high keys.
When we're doing this, we must be careful to update the keys of all
node pointers up to the root instead of stopping at the first set of
keys that don't need updating. This is because it's possible for a
single deletion to cause joining of multiple levels of tree, and so
we need to update everything going back to the root.
The diff_two_keys functions return < 0, 0, or > 0 if key1 is less than,
equal to, or greater than key2, respectively. This is consistent
with the rest of the kernel and the C library.
In btree_updkeys(), we need to evaluate the force_all parameter before
running the key diff to avoid reading uninitialized memory when we're
forcing a key update. This happens when we've allocated an empty slot
at level N + 1 to point to a new block at level N and we're in the
process of filling out the new keys.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Merge yet more updates from Andrew Morton:
- the rest of ocfs2
- various hotfixes, mainly MM
- quite a bit of misc stuff - drivers, fork, exec, signals, etc.
- printk updates
- firmware
- checkpatch
- nilfs2
- more kexec stuff than usual
- rapidio updates
- w1 things
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (111 commits)
ipc: delete "nr_ipc_ns"
kcov: allow more fine-grained coverage instrumentation
init/Kconfig: add clarification for out-of-tree modules
config: add android config fragments
init/Kconfig: ban CONFIG_LOCALVERSION_AUTO with allmodconfig
relay: add global mode support for buffer-only channels
init: allow blacklisting of module_init functions
w1:omap_hdq: fix regression
w1: add helper macro module_w1_family
w1: remove need for ida and use PLATFORM_DEVID_AUTO
rapidio/switches: add driver for IDT gen3 switches
powerpc/fsl_rio: apply changes for RIO spec rev 3
rapidio: modify for rev.3 specification changes
rapidio: change inbound window size type to u64
rapidio/idt_gen2: fix locking warning
rapidio: fix error handling in mbox request/release functions
rapidio/tsi721_dma: advance queue processing from transfer submit call
rapidio/tsi721: add messaging mbox selector parameter
rapidio/tsi721: add PCIe MRRS override parameter
rapidio/tsi721_dma: add channel mask and queue size parameters
...
Add some function pointers to bc_ops to get the btree keys for
leaf and node blocks, and to update parent keys of a block.
Convert the _btree_updkey calls to use our new pointer, and
modify the tree shape changing code to call the appropriate
get_*_keys pointer instead of _btree_copy_keys because the
overlapping btree has to calculate high key values.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When a btree block has to be split, we pass the new block's ptr from
xfs_btree_split() back to xfs_btree_insert() via a pointer parameter;
however, we pass the block's key through the cursor's record. It is a
little weird to "initialize" a record from a key since the non-key
attributes will have garbage values.
When we go to add support for interval queries, we have to be able to
pass the lowest and highest keys accessible via a pointer. There's no
clean way to pass this back through the cursor's record field.
Therefore, pass the key directly back to xfs_btree_insert() the same
way that we pass the btree_ptr.
As a bonus, we no longer need init_rec_from_key and can drop it from the
codebase.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
If we make the inode root block of a btree unfull by expanding the
root, we must set *stat to 1 to signal success, rather than leaving
it uninitialized.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When we're deleting realtime extents, we need to lock the summary
inode in case we need to update the summary info to prevent an assert
on the rsumip inode lock on a debug kernel. While we're at it, fix
the locking annotations so that we avoid triggering lockdep warnings.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Apparently cris doesn't require structure stride to align with the
largest type in the struct, so list[0] isn't at offset 4 like it is
everywhere else. Fix this... insofar as existing XFSes on cris are
screwed.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When we're iterating inode xattrs by handle, we have to copy the
cursor back to userspace so that a subsequent invocation actually
retrieves subsequent contents.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This allows OrangeFS to utilize the dcache and adds an in kernel
attribute cache. We previously used the user side client for this
purpose.
We see a modest performance increase on small file operations. For
example, without the cache, compiling coreutils takes about 17 minutes.
With the patch and a 50 millisecond timeout for dcache_timeout_msecs and
getattr_timeout_msecs (the default), compiling coreutils takes about
6 minutes 20 seconds. On the same hardware, compiling coreutils on an
xfs filesystem takes 90 seconds. We see similar improvements with mdtest
and a test involving writing, reading, and deleting a large number of
small files.
Interested parties can review more data at the following URL.
https://docs.google.com/spreadsheets/d/1v4aUeppKexIbRMz_Yn9k4eaM3uy2KCaPoe_93YKWOtA/pubhtml
The eventual goal of this is to allow getdents to turn into a
readdirplus to the OrangeFS server. The cache will be filled then, which
should provide a performance benefit to the common case of readdir
followed by getattr on each entry (i.e. ls -l).
This also fixes a bug. When orangefs_inode_permission was added, it did
not collect i_size from the OrangeFS server, since this presses an
unnecessary load on the OrangeFS server. However, it left a case where
i_size is never initialized. Then running an executable could fail.
With this patch, size is always collected to be inserted into the cache.
Thus the bug disappears. If this patch is not accepted during this merge
window, we will send a one-line band-aid for this bug instead.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=rGmf
-----END PGP SIGNATURE-----
Merge tag 'for-linus-v4.8' of git://github.com/martinbrandenburg/linux
Pull orangefs update from Martin Brandenburg:
"Kernel side caching and executable bugfix
This allows OrangeFS to utilize the dcache and adds an in kernel
attribute cache. We previously used the user side client for this
purpose.
We see a modest performance increase on small file operations. For
example, without the cache, compiling coreutils takes about 17
minutes. With the patch and a 50 millisecond timeout for
dcache_timeout_msecs and getattr_timeout_msecs (the default),
compiling coreutils takes about 6 minutes 20 seconds. On the same
hardware, compiling coreutils on an xfs filesystem takes 90 seconds.
We see similar improvements with mdtest and a test involving writing,
reading, and deleting a large number of small files.
Interested parties can review more data at the following URL.
https://docs.google.com/spreadsheets/d/1v4aUeppKexIbRMz_Yn9k4eaM3uy2KCaPoe_93YKWOtA/pubhtml
The eventual goal of this is to allow getdents to turn into a
readdirplus to the OrangeFS server. The cache will be filled then,
which should provide a performance benefit to the common case of
readdir followed by getattr on each entry (i.e. ls -l).
This also fixes a bug. When orangefs_inode_permission was added, it
did not collect i_size from the OrangeFS server, since this presses an
unnecessary load on the OrangeFS server. However, it left a case
where i_size is never initialized. Then running an executable could
fail.
With this patch, size is always collected to be inserted into the
cache. Thus the bug disappears. If this patch is not accepted during
this merge window, we will send a one-line band-aid for this bug
instead"
* tag 'for-linus-v4.8' of git://github.com/martinbrandenburg/linux:
Orangefs: update orangefs.txt
orangefs: Account for jiffies wraparound.
orangefs: Change default dcache and getattr timeout to 50 msec.
orangefs: Allow dcache and getattr cache time to be configured.
orangefs: Cache getattr results.
orangefs: Use d_time to avoid excessive lookups
* RADOS namespace support in libceph and CephFS (Zheng Yan and myself).
The stopgaps added in 4.5 to deny access to inodes in namespaces are
removed and CEPH_FEATURE_FS_FILE_LAYOUT_V2 feature bit is now fully
supported.
* A large rework of the MDS cap flushing code (Zheng Yan).
* Handle some of ->d_revalidate() in RCU mode (Jeff Layton). We were
overly pessimistic before, bailing at the first sight of LOOKUP_RCU.
On top of that we've got a few CephFS bug fixes, a couple of cleanups
and Arnd's workaround for a weird genksyms issue.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAABCAAGBQJXoKLJAAoJEEp/3jgCEfOLDTUIAIcctpKUiNBokc95mQaXYl34
j7lPIaD0/Ur7JPt4nMdtlywYJYSVV2c+SglHztj/+fv0G4bWbLVEFRruh9SwKIci
PzttcmycIAqSn1f5gBZwyQbGuffd/F0EnBj7fFjcukt01i3s1ZQ7t4XtLGtAV0Ts
aIfFtx9SqWig57Z1OZqNgnhnOoh6IqNbic3FL5Hvdl5N5pFbBcQho6Vzoa5O1osH
URG6RmCcO4nykfSoxiivE7UZ+CImsXHkRD7rupBuIjqjZ8wvmZqQF5qxnkb9Dw2F
IkNhrHkTSIiv4EsNPLAETTnFSozrL1nEykKr2FBW+ti8nxNcav+8FgVapqLvFIw=
=gQ0/
-----END PGP SIGNATURE-----
Merge tag 'ceph-for-4.8-rc1' of git://github.com/ceph/ceph-client
Pull Ceph updates from Ilya Dryomov:
"The highlights are:
- RADOS namespace support in libceph and CephFS (Zheng Yan and
myself). The stopgaps added in 4.5 to deny access to inodes in
namespaces are removed and CEPH_FEATURE_FS_FILE_LAYOUT_V2 feature
bit is now fully supported
- A large rework of the MDS cap flushing code (Zheng Yan)
- Handle some of ->d_revalidate() in RCU mode (Jeff Layton). We were
overly pessimistic before, bailing at the first sight of LOOKUP_RCU
On top of that we've got a few CephFS bug fixes, a couple of cleanups
and Arnd's workaround for a weird genksyms issue"
* tag 'ceph-for-4.8-rc1' of git://github.com/ceph/ceph-client: (34 commits)
ceph: fix symbol versioning for ceph_monc_do_statfs
ceph: Correctly return NXIO errors from ceph_llseek
ceph: Mark the file cache as unreclaimable
ceph: optimize cap flush waiting
ceph: cleanup ceph_flush_snaps()
ceph: kick cap flushes before sending other cap message
ceph: introduce an inode flag to indicates if snapflush is needed
ceph: avoid sending duplicated cap flush message
ceph: unify cap flush and snapcap flush
ceph: use list instead of rbtree to track cap flushes
ceph: update types of some local varibles
ceph: include 'follows' of pending snapflush in cap reconnect message
ceph: update cap reconnect message to version 3
ceph: mount non-default filesystem by name
libceph: fsmap.user subscription support
ceph: handle LOOKUP_RCU in ceph_d_revalidate
ceph: allow dentry_lease_is_valid to work under RCU walk
ceph: clear d_fsinfo pointer under d_lock
ceph: remove ceph_mdsc_lease_release
ceph: don't use ->d_time
...
new_insert_key only makes any sense when it's associated with a
new_insert_ptr, which is initialized to NULL and changed to a
buffer_head when we also initialize new_insert_key. We can key off of
that to avoid the uninitialized warning.
Link: http://lkml.kernel.org/r/5eca5ffb-2155-8df2-b4a2-f162f105efed@suse.com
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Jan Kara <jack@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The header file "include/linux/nilfs2_fs.h" is composed of parts for
ioctl and disk format, and both are intended to be shared with user
space programs.
This moves them to the uapi directory "include/uapi/linux" splitting the
file to "nilfs2_api.h" and "nilfs2_ondisk.h". The following minor
changes are accompanied by this migration:
- nilfs_direct_node struct in nilfs2/direct.h is converged to
nilfs2_ondisk.h because it's an on-disk structure.
- inline functions nilfs_rec_len_from_disk() and
nilfs_rec_len_to_disk() are moved to nilfs2/dir.c.
Link: http://lkml.kernel.org/r/1465825507-3407-4-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Variables ns_seg_seq, ns_segnum, ns_nextnum, ns_pseg_offset, ns_cno,
ns_ctime, ns_nongc_ctime, and ns_ndirtyblks, are protected by
ns_segctor_sem, but ns_sem is wrongly used by the nilfs sysfs code when
reading these variables. This fixes the misuse and clarifies which
semaphore protects them in the comment of the_nilfs struct.
Link: http://lkml.kernel.org/r/1465825507-3407-2-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move parser of snapshot mount option to a separate function
nilfs_parse_snapshot_option(), replace simple_strtoull() with
kstrtoull() to avoid checkpatch.pl warning "WARNING: simple_strtoull is
obsolete, use kstrtoull instead", and refine the error message of the
parser.
Link: http://lkml.kernel.org/r/1464875891-5443-9-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use cond_resched() instead of yield() in the loop of
nilfs_transaction_lock() since the usage corresponds to the "be nice for
others" case that the comment of yield() says.
This removes the following checkpatch.pl warning:
"WARNING: Using yield() is generally wrong. See yield() kernel-doc
(sched/core.c)"
Link: http://lkml.kernel.org/r/1464875891-5443-8-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When nilfs returned -EIO as an error code, it's not always clear if it
came from the underlying block device or not. This will mend the issue
by having low level I/O routines of nilfs output an error message when
they detected an I/O error.
Link: http://lkml.kernel.org/r/1464875891-5443-7-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use nilfs_msg() to output warning messages and get rid of
nilfs_warning() function. This also removes function names from the
messages unless we embed them explicitly in format strings. Instead,
some messages are revised to clarify the context.
[arnd@arndb.de: avoid warning about unused variables]
Link: http://lkml.kernel.org/r/20160615201945.3348205-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/1464875891-5443-6-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Replace most use of printk() in nilfs2 implementation with nilfs_msg(),
and reduce the following checkpatch.pl warning:
"WARNING: Prefer [subsystem eg: netdev]_crit([subsystem]dev, ...
then dev_crit(dev, ... then pr_crit(... to printk(KERN_CRIT ..."
This patch also fixes a minor checkpatch warning "WARNING: quoted string
split across lines" that often accompanies the prior warning, and amends
message format as needed.
Link: http://lkml.kernel.org/r/1464875891-5443-5-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Insert a back pointer to super block instance in nilfs object so that
functions of nilfs2 easily refer to the super block instance. This
simplifies replacement of printk() in the successive change.
Link: http://lkml.kernel.org/r/1464875891-5443-4-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Define an own output routine to replace bare use of printk() function.
The output routine is implemented with a macro and a helper function,
which are named nilfs_msg() and __nilfs_msg(), respectively.
__nilfs_msg() formats a message like "NILFS (<device-name>): <message>",
prefixing it with a given log level, and terminates the statement with a
newline. The "device-name" is optional to make it available in early
stages; it will be omitted if a NULL pointer is passed to super block
instance argument. nilfs_msg() wraps __nilfs_msg() and is removed if
CONFIG_PRINTK is not set.
Link: http://lkml.kernel.org/r/1464875891-5443-3-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Simplify nilfs_error(), an output function used to report critical
issues in file system. This renames the original nilfs_error() function
to __nilfs_error() and redefines it as a macro to hide its function name
argument within the macro.
Every call site of nilfs_error() is changed to strip __func__ argument
except nilfs_bmap_convert_error(); nilfs_bmap_convert_error() directly
calls __nilfs_error() because it inherits caller's function name.
Link: http://lkml.kernel.org/r/1464875891-5443-2-git-send-email-konishi.ryusuke@lab.ntt.co.jp
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since the -Wincompatible-pointer-types is reported as error, alpha
doesn't build anymore. Let's fix it in a minimal way.
fs/binfmt_em86.c:73:35: error: passing argument 2 of `copy_strings_kernel' from incompatible pointer type [-Werror=incompatible-pointer-types]
retval = copy_strings_kernel(1, &i_arg, bprm);
^ ^
fs/binfmt_em86.c:77:34: error: passing argument 2 of `copy_strings_kernel' from incompatible pointer type [-Werror=incompatible-pointer-types]
retval = copy_strings_kernel(1, &i_name, bprm);
^
Link: http://lkml.kernel.org/r/1469525978-23359-1-git-send-email-wagi@monom.org
Signed-off-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A double-bug exists in the bss calculation code, where an overflow can
happen in the "last_bss - elf_bss" calculation, but vm_brk internally
aligns the argument, underflowing it, wrapping back around safe. We
shouldn't depend on these bugs staying in sync, so this cleans up the
bss padding handling to avoid the overflow.
This moves the bss padzero() before the last_bss > elf_bss case, since
the zero-filling of the ELF_PAGE should have nothing to do with the
relationship of last_bss and elf_bss: any trailing portion should be
zeroed, and a zero size is already handled by padzero().
Then it handles the math on elf_bss vs last_bss correctly. These need
to both be ELF_PAGE aligned to get the comparison correct, since that's
the expected granularity of the mappings. Since elf_bss already had
alignment-based padding happen in padzero(), the "start" of the new
vm_brk() should be moved forward as done in the original code. However,
since the "end" of the vm_brk() area will already become PAGE_ALIGNed in
vm_brk() then last_bss should get aligned here to avoid hiding it as a
side-effect.
Additionally makes a cosmetic change to the initial last_bss calculation
so it's easier to read in comparison to the load_addr calculation above
it (i.e. the only difference is p_filesz vs p_memsz).
Link: http://lkml.kernel.org/r/1468014494-25291-2-git-send-email-keescook@chromium.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Reported-by: Hector Marco-Gisbert <hecmargi@upv.es>
Cc: Ismael Ripoll Ripoll <iripoll@upv.es>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some systems are memory constrained but they need to load very large
firmwares. The firmware subsystem allows drivers to request this
firmware be loaded from the filesystem, but this requires that the
entire firmware be loaded into kernel memory first before it's provided
to the driver. This can lead to a situation where we map the firmware
twice, once to load the firmware into kernel memory and once to copy the
firmware into the final resting place.
This creates needless memory pressure and delays loading because we have
to copy from kernel memory to somewhere else. Let's add a
request_firmware_into_buf() API that allows drivers to request firmware
be loaded directly into a pre-allocated buffer. This skips the
intermediate step of allocating a buffer in kernel memory to hold the
firmware image while it's read from the filesystem. It also requires
that drivers know how much memory they'll require before requesting the
firmware and negates any benefits of firmware caching because the
firmware layer doesn't manage the buffer lifetime.
For a 16MB buffer, about half the time is spent performing a memcpy from
the buffer to the final resting place. I see loading times go from
0.081171 seconds to 0.047696 seconds after applying this patch. Plus
the vmalloc pressure is reduced.
This is based on a patch from Vikram Mulukutla on codeaurora.org:
https://www.codeaurora.org/cgit/quic/la/kernel/msm-3.18/commit/drivers/base/firmware_class.c?h=rel/msm-3.18&id=0a328c5f6cd999f5c591f172216835636f39bcb5
Link: http://lkml.kernel.org/r/20160607164741.31849-4-stephen.boyd@linaro.org
Signed-off-by: Stephen Boyd <stephen.boyd@linaro.org>
Cc: Mimi Zohar <zohar@linux.vnet.ibm.com>
Cc: Vikram Mulukutla <markivx@codeaurora.org>
Cc: Mark Brown <broonie@kernel.org>
Cc: Ming Lei <ming.lei@canonical.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Suppress a bunch of warnings of the form:
fs/proc/task_mmu.c: In function 'show_smap_vma_flags':
fs/proc/task_mmu.c:635:22: warning: initialized field overwritten [-Wt override-init]
[ilog2(VM_READ)] = "rd",
^~~~
fs/proc/task_mmu.c:635:22: note: (near initialization for 'mnemonics[0]')
They happen because of the way we intentionally build the table, so
silence the warning when building with 'make W=1'.
Link: http://lkml.kernel.org/r/8727.1470022083@turing-police.cc.vt.edu
Signed-off-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
/proc/stat shows (among lots of other things) the current boottime (i.e.
number of seconds since boot). While a 32-bit number is sufficient for
this particular case, we want to get rid of the 'struct timespec'
suffers from a 32-bit overflow in 2038.
This changes the code to use a struct timespec64, which is known to be
safe in all cases.
Link: http://lkml.kernel.org/r/20160617201247.2292101-1-arnd@arndb.de
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was needed before to ensure that ->signal != 0 and do_each_thread()
is safe, see commit b95c35e76b29b ("oom: fix the unsafe usage of
badness() in proc_oom_score()") for details.
Today tsk->signal can't go away and for_each_thread(tsk) is always safe.
Link: http://lkml.kernel.org/r/20160608211921.GA15508@redhat.com
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Radix trees may be used not only for storing page cache pages, so
unconditionally accounting radix tree nodes to the current memory cgroup
is bad: if a radix tree node is used for storing data shared among
different cgroups we risk pinning dead memory cgroups forever.
So let's only account radix tree nodes if it was explicitly requested by
passing __GFP_ACCOUNT to INIT_RADIX_TREE. Currently, we only want to
account page cache entries, so mark mapping->page_tree so.
Fixes: 58e698af4c63 ("radix-tree: account radix_tree_node to memory cgroup")
Link: http://lkml.kernel.org/r/1470057188-7864-1-git-send-email-vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org> [4.6+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found a dlm-blocked situation caused by continuous breakdown of
recovery masters described below. To solve this problem, we should
purge recovery lock once detecting recovery master goes down.
N3 N2 N1(reco master)
go down
pick up recovery lock and
begin recoverying for N2
go down
pick up recovery
lock failed, then
purge it:
dlm_purge_lockres
->DROPPING_REF is set
send deref to N1 failed,
recovery lock is not purged
find N1 go down, begin
recoverying for N1, but
blocked in dlm_do_recovery
as DROPPING_REF is set:
dlm_do_recovery
->dlm_pick_recovery_master
->dlmlock
->dlm_get_lock_resource
->__dlm_wait_on_lockres_flags(tmpres,
DLM_LOCK_RES_DROPPING_REF);
Fixes: 8c0343968163 ("ocfs2/dlm: clear DROPPING_REF flag when the master goes down")
Link: http://lkml.kernel.org/r/578453AF.8030404@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found a BUG situation that lockres is migrated during deref described
below. To solve the BUG, we could purge lockres directly when other
node says I did not have a ref. Additionally, we'd better purge lockres
if master goes down, as no one will response deref done.
Node 1 Node 2(old master) Node3(new master)
dlm_purge_lockres
send deref to N2
leave domain
migrate lockres to N3
finish migration
send do assert
master to N1
receive do assert msg
form N3, but can not
find lockres because
DROPPING_REF is set,
so the owner is still
N2.
receive deref from N1
and response -EINVAL
because lockres is migrated
BUG when receive -EINVAL
in dlm_drop_lockres_ref
Fixes: 842b90b62461d ("ocfs2/dlm: return in progress if master can not clear the refmap bit right now")
Link: http://lkml.kernel.org/r/57845103.3070406@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found a BUG situation in which DLM_LOCK_RES_DROPPING_REF is cleared
unexpected that described below. To solve the bug, we disable the
BUG_ON and purge lockres in dlm_do_local_recovery_cleanup.
Node 1 Node 2(master)
dlm_purge_lockres
dlm_deref_lockres_handler
DLM_LOCK_RES_SETREF_INPROG is set
response DLM_DEREF_RESPONSE_INPROG
receive DLM_DEREF_RESPONSE_INPROG
stop puring in dlm_purge_lockres
and wait for DLM_DEREF_RESPONSE_DONE
dispatch dlm_deref_lockres_worker
response DLM_DEREF_RESPONSE_DONE
receive DLM_DEREF_RESPONSE_DONE and
prepare to purge lockres
Node 2 goes down
find Node2 down and do local
clean up for Node2:
dlm_do_local_recovery_cleanup
-> clear DLM_LOCK_RES_DROPPING_REF
when purging lockres, BUG_ON happens
because DLM_LOCK_RES_DROPPING_REF is clear:
dlm_deref_lockres_done_handler
->BUG_ON(!(res->state & DLM_LOCK_RES_DROPPING_REF));
[akpm@linux-foundation.org: fix duplicated write to `ret']
Fixes: 60d663cb5273 ("ocfs2/dlm: add DEREF_DONE message")
Link: http://lkml.kernel.org/r/57845055.9080702@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The testcase "mmaptruncate" in ocfs2 test suite always fails with ENOSPC
error on small volume (say less than 10G). This testcase repeatedly
performs "extend" and "truncate" on a file. Continuously, it truncates
the file to 1/2 of the size, and then extends to 100% of the size. The
main bitmap will quickly run out of space because the "truncate" code
prevent truncate log from being flushed by
ocfs2_schedule_truncate_log_flush(osb, 1), while truncate log may have
cached lots of clusters.
So retry to allocate after flushing truncate log when ENOSPC is
returned. And we cannot reuse the deleted blocks before the transaction
committed. Fortunately, we already have a function to do this -
ocfs2_try_to_free_truncate_log(). Just need to remove the "static"
modifier and put it into the right place.
The "unlock"/"lock" code isn't elegant, but there seems to be no better
option.
[zren@suse.com: locking fix]
Link: http://lkml.kernel.org/r/1468031546-4797-1-git-send-email-zren@suse.com
Link: http://lkml.kernel.org/r/1466586469-5541-1-git-send-email-zren@suse.com
Signed-off-by: Eric Ren <zren@suse.com>
Reviewed-by: Gang He <ghe@suse.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We encountered a bug from the customer, the user did a fsck.ocfs2 on the
file system and exited unusually, the lockspace (with LVB size = 32) was
left in the kernel space, next, the user mounted this file system, the
kernel module did not create a new lockspace (LVB size = 64) via calling
dlm_new_lockspace() function in mounting stage, just used the existing
lockspace, created by the user space tool, this would lead the user was
not able to mount this file system from the other nodes, with the error
message like:
dlm: 032F5......: config mismatch: 64,0 nodeid 177127961: 32,0
(mount.ocfs2,26981,46):ocfs2_dlm_init:2995 ERROR: status = -71
ocfs2_mount_volume:1881 ERROR: status = -71
ocfs2_fill_super:1236 ERROR: status = -71
The user found it very difficult to find the root cause, then, we
brought out this patch to relieve such problem.
First, we add one more flag in calling dlm_new_lockspace() function, to
make sure the lockspace is created by kernel module itself, and this
change will not affect the backward compatibility.
Second, the obvious error message is reported in the kernel log, let the
user be more easy to find the root cause.
This patch will be used to insure the dlm lockspace is created by kernel
module when mounting a ocfs2 file system. There are two ways to create
a lockspace, from user space and kernel space, but the same name
lockspaces probably have different lvblen lengths/flags.
To avoid this mix using, we add one more flag DLM_LSFL_NEWEXCL, it will
make sure the dlm lockspace is created by kernel module when mounting.
Secondly, if a user space program (ocfs2-tools) is running on a file
system, the user tries to mount this file system in the cluster, DLM
module will return a -EEXIST or -EPROTO errno, we should give the user a
obvious error message, then, the user can let that user space tool exit
before mounting the file system again.
Link: http://lkml.kernel.org/r/1463731940-13044-2-git-send-email-ghe@suse.com
Signed-off-by: Gang He <ghe@suse.com>
Reviewed-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The userspace component attempts to do this, but this will prevent
us from even needing to go into userspace to satisfy certain getattr
requests.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
With commit 56f23fdbb600 ("Btrfs: fix file/data loss caused by fsync after
rename and new inode") we got simple fix for a functional issue when the
following sequence of actions is done:
at transaction N
create file A at directory D
at transaction N + M (where M >= 1)
move/rename existing file A from directory D to directory E
create a new file named A at directory D
fsync the new file
power fail
The solution was to simply detect such scenario and fallback to a full
transaction commit when we detect it. However this turned out to had a
significant impact on throughput (and a bit on latency too) for benchmarks
using the dbench tool, which simulates real workloads from smbd (Samba)
servers. For example on a test vm (with a debug kernel):
Unpatched:
Throughput 19.1572 MB/sec 32 clients 32 procs max_latency=1005.229 ms
Patched:
Throughput 23.7015 MB/sec 32 clients 32 procs max_latency=809.206 ms
The patched results (this patch is applied) are similar to the results of
a kernel with the commit 56f23fdbb600 ("Btrfs: fix file/data loss caused
by fsync after rename and new inode") reverted.
This change avoids the fallback to a transaction commit and instead makes
sure all the names of the conflicting inode (the one that had a name in a
past transaction that matches the name of the new file in the same parent
directory) are logged so that at log replay time we don't lose neither the
new file nor the old file, and the old file gets the name it was renamed
to.
This also ends up avoiding a full transaction commit for a similar case
that involves an unlink instead of a rename of the old file:
at transaction N
create file A at directory D
at transaction N + M (where M >= 1)
remove file A
create a new file named A at directory D
fsync the new file
power fail
Signed-off-by: Filipe Manana <fdmanana@suse.com>
When we attempt to read an inode from disk, we end up always returning an
-ESTALE error to the caller regardless of the actual failure reason, which
can be an out of memory problem (when allocating a path), some error found
when reading from the fs/subvolume btree (like a genuine IO error) or the
inode does not exists. So lets start returning the real error code to the
callers so that they don't treat all -ESTALE errors as meaning that the
inode does not exists (such as during orphan cleanup). This will also be
needed for a subsequent patch in the same series dealing with a special
fsync case.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
When doing an incremental send, if we find a new/modified/deleted extent,
reference or xattr without having previously processed the corresponding
inode item we end up exexuting a BUG_ON(). This is because whenever an
extent, xattr or reference is added, modified or deleted, we always expect
to have the corresponding inode item updated. However there are situations
where this will not happen due to transient -ENOMEM or -ENOSPC errors when
doing delayed inode updates.
For example, when punching holes we can succeed in deleting and modifying
(shrinking) extents but later fail to do the delayed inode update. So after
such failure we close our transaction handle and right after a snapshot of
the fs/subvol tree can be made and used later for a send operation. The
same thing can happen during truncate, link, unlink, and xattr related
operations.
So instead of executing a BUG_ON, make send return an -EIO error and print
an informative error message do dmesg/syslog.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
The caller of send_utimes() is supposed to be sure that the inode number
it passes to this function does actually exists in the send snapshot.
However due to logic/algorithm bugs (such as the one fixed by the patch
titled "Btrfs: send, fix invalid leaf accesses due to incorrect utimes
operations"), this might not be the case and when that happens it makes
send_utimes() access use an unrelated leaf item as the target inode item
or access beyond a leaf's boundaries (when the leaf is full and
path->slots[0] matches the number of items in the leaf).
So if the call to btrfs_search_slot() done by send_utimes() does not find
the inode item, just make sure send_utimes() returns -ENOENT and does not
silently accesses unrelated leaf items or does invalid leaf accesses, also
allowing us to easialy and deterministically catch such algorithmic/logic
bugs.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
During an incremental send, if we have delayed rename operations for inodes
that were children of directories which were removed in the send snapshot,
we can end up accessing incorrect items in a leaf or accessing beyond the
last item of the leaf due to issuing utimes operations for the removed
inodes. Consider the following example:
Parent snapshot:
. (ino 256)
|--- a/ (ino 257)
| |--- c/ (ino 262)
|
|--- b/ (ino 258)
| |--- d/ (ino 263)
|
|--- del/ (ino 261)
|--- x/ (ino 259)
|--- y/ (ino 260)
Send snapshot:
. (ino 256)
|--- a/ (ino 257)
|
|--- b/ (ino 258)
|
|--- c/ (ino 262)
| |--- y/ (ino 260)
|
|--- d/ (ino 263)
|--- x/ (ino 259)
1) When processing inodes 259 and 260, we end up delaying their rename
operations because their parents, inodes 263 and 262 respectively, were
not yet processed and therefore not yet renamed;
2) When processing inode 262, its rename operation is issued and right
after the rename operation for inode 260 is issued. However right after
issuing the rename operation for inode 260, at send.c:apply_dir_move(),
we issue utimes operations for all current and past parents of inode
260. This means we try to send a utimes operation for its old parent,
inode 261 (deleted in the send snapshot), which does not cause any
immediate and deterministic failure, because when the target inode is
not found in the send snapshot, the send.c:send_utimes() function
ignores it and uses the leaf region pointed to by path->slots[0],
which can be any unrelated item (belonging to other inode) or it can
be a region outside the leaf boundaries, if the leaf is full and
path->slots[0] matches the number of items in the leaf. So we end
up either successfully sending a utimes operation, which is fine
and irrelevant because the old parent (inode 261) will end up being
deleted later, or we end up doing an invalid memory access tha
crashes the kernel.
So fix this by making apply_dir_move() issue utimes operations only for
parents that still exist in the send snapshot. In a separate patch we
will make send_utimes() return an error (-ENOENT) if the given inode
does not exists in the send snapshot.
Signed-off-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
[Rewrote change log to be more detailed and better organized]
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Under certain situations, when doing an incremental send, we can end up
not freeing orphan_dir_info structures as soon as they are no longer
needed. Instead we end up freeing them only after finishing the send
stream, which causes a warning to be emitted:
[282735.229200] ------------[ cut here ]------------
[282735.229968] WARNING: CPU: 9 PID: 10588 at fs/btrfs/send.c:6298 btrfs_ioctl_send+0xe2f/0xe51 [btrfs]
[282735.231282] Modules linked in: btrfs crc32c_generic xor raid6_pq acpi_cpufreq tpm_tis ppdev tpm parport_pc psmouse parport sg pcspkr i2c_piix4 i2c_core evdev processor serio_raw button loop autofs4 ext4 crc16 jbd2 mbcache sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy [last unloaded: btrfs]
[282735.237130] CPU: 9 PID: 10588 Comm: btrfs Tainted: G W 4.6.0-rc7-btrfs-next-31+ #1
[282735.239309] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[282735.240160] 0000000000000000 ffff880224273ca8 ffffffff8126b42c 0000000000000000
[282735.240160] 0000000000000000 ffff880224273ce8 ffffffff81052b14 0000189a24273ac8
[282735.240160] ffff8802210c9800 0000000000000000 0000000000000001 0000000000000000
[282735.240160] Call Trace:
[282735.240160] [<ffffffff8126b42c>] dump_stack+0x67/0x90
[282735.240160] [<ffffffff81052b14>] __warn+0xc2/0xdd
[282735.240160] [<ffffffff81052beb>] warn_slowpath_null+0x1d/0x1f
[282735.240160] [<ffffffffa03c99d5>] btrfs_ioctl_send+0xe2f/0xe51 [btrfs]
[282735.240160] [<ffffffffa0398358>] btrfs_ioctl+0x14f/0x1f81 [btrfs]
[282735.240160] [<ffffffff8108e456>] ? arch_local_irq_save+0x9/0xc
[282735.240160] [<ffffffff8118da05>] vfs_ioctl+0x18/0x34
[282735.240160] [<ffffffff8118e00c>] do_vfs_ioctl+0x550/0x5be
[282735.240160] [<ffffffff81196f0c>] ? __fget+0x6b/0x77
[282735.240160] [<ffffffff81196fa1>] ? __fget_light+0x62/0x71
[282735.240160] [<ffffffff8118e0d1>] SyS_ioctl+0x57/0x79
[282735.240160] [<ffffffff8149e025>] entry_SYSCALL_64_fastpath+0x18/0xa8
[282735.240160] [<ffffffff81100c6b>] ? time_hardirqs_off+0x9/0x14
[282735.240160] [<ffffffff8108e87d>] ? trace_hardirqs_off_caller+0x1f/0xaa
[282735.256343] ---[ end trace a4539270c8056f93 ]---
Consider the following example:
Parent snapshot:
. (ino 256)
|--- a/ (ino 257)
| |--- c/ (ino 260)
|
|--- del/ (ino 259)
|--- tmp/ (ino 258)
|--- x/ (ino 261)
|--- y/ (ino 262)
Send snapshot:
. (ino 256)
|--- a/ (ino 257)
| |--- x/ (ino 261)
| |--- y/ (ino 262)
|
|--- c/ (ino 260)
|--- tmp/ (ino 258)
1) When processing inode 258, we end up delaying its rename operation
because it has an ancestor (in the send snapshot) that has a higher
inode number (inode 260) which was also renamed in the send snapshot,
therefore we delay the rename of inode 258 so that it happens after
inode 260 is renamed;
2) When processing inode 259, we end up delaying its deletion (rmdir
operation) because it has a child inode (258) that has its rename
operation delayed. At this point we allocate an orphan_dir_info
structure and tag inode 258 so that we later attempt to see if we
can delete (rmdir) inode 259 once inode 258 is renamed;
3) When we process inode 260, after renaming it we finally do the rename
operation for inode 258. Once we issue the rename operation for inode
258 we notice that this inode was tagged so that we attempt to see
if at this point we can delete (rmdir) inode 259. But at this point
we can not still delete inode 259 because it has 2 children, inodes
261 and 262, that were not yet processed and therefore not yet
moved (renamed) away from inode 259. We end up not freeing the
orphan_dir_info structure allocated in step 2;
4) We process inodes 261 and 262, and once we move/rename inode 262
we issue the rmdir operation for inode 260;
5) We finish the send stream and notice that red black tree that
contains orphan_dir_info structures is not empty, so we emit
a warning and then free any orphan_dir_structures left.
So fix this by freeing an orphan_dir_info structure once we try to
apply a pending rename operation if we can not delete yet the tagged
directory.
A test case for fstests follows soon.
Signed-off-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
[Modified changelog to be more detailed and easier to understand]
Under certain situations, an incremental send operation can contain
a rmdir operation that will make the receiving end fail when attempting
to execute it, because the target directory is not yet empty.
Consider the following example:
Parent snapshot:
. (ino 256)
|--- a/ (ino 257)
| |--- c/ (ino 260)
|
|--- del/ (ino 259)
|--- tmp/ (ino 258)
|--- x/ (ino 261)
Send snapshot:
. (ino 256)
|--- a/ (ino 257)
| |--- x/ (ino 261)
|
|--- c/ (ino 260)
|--- tmp/ (ino 258)
1) When processing inode 258, we delay its rename operation because inode
260 is its new parent in the send snapshot and it was not yet renamed
(since 260 > 258, that is, beyond the current progress);
2) When processing inode 259, we realize we can not yet send an rmdir
operation (against inode 259) because inode 258 was still not yet
renamed/moved away from inode 259. Therefore we update data structures
so that after inode 258 is renamed, we try again to see if we can
finally send an rmdir operation for inode 259;
3) When we process inode 260, we send a rename operation for it followed
by a rename operation for inode 258. Once we send the rename operation
for inode 258 we then check if we can finally issue an rmdir for its
previous parent, inode 259, by calling the can_rmdir() function with
a value of sctx->cur_ino + 1 (260 + 1 = 261) for its "progress"
argument. This makes can_rmdir() return true (value 1) because even
though there's still a child inode of inode 259 that was not yet
renamed/moved, which is inode 261, the given value of progress (261)
is not lower then 261 (that is, not lower than the inode number of
some child of inode 259). So we end up sending a rmdir operation for
inode 259 before its child inode 261 is processed and renamed.
So fix this by passing the correct progress value to the call to
can_rmdir() from within apply_dir_move() (where we issue delayed rename
operations), which should match stcx->cur_ino (the number of the inode
currently being processed) and not sctx->cur_ino + 1.
A test case for fstests follows soon.
Signed-off-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
[Rewrote change log to be more detailed, clear and well formatted]
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Example scenario:
Parent snapshot:
. (ino 277)
|---- tmp/ (ino 278)
|---- pre/ (ino 280)
| |---- wait_dir/ (ino 281)
|
|---- desc/ (ino 282)
|---- ance/ (ino 283)
| |---- below_ance/ (ino 279)
|
|---- other_dir/ (ino 284)
Send snapshot:
. (ino 277)
|---- tmp/ (ino 278)
|---- other_dir/ (ino 284)
|---- below_ance/ (ino 279)
| |---- pre/ (ino 280)
|
|---- wait_dir/ (ino 281)
|---- desc/ (ino 282)
|---- ance/ (ino 283)
While computing the send stream the following steps happen:
1) While processing inode 279 we end up delaying its rename operation
because its new parent in the send snapshot, inode 284, was not
yet processed and therefore not yet renamed;
2) Later when processing inode 280 we end up renaming it immediately to
"ance/below_once/pre" and not delay its rename operation because its
new parent (inode 279 in the send snapshot) has its rename operation
delayed and inode 280 is not an encestor of inode 279 (its parent in
the send snapshot) in the parent snapshot;
3) When processing inode 281 we end up delaying its rename operation
because its new parent in the send snapshot, inode 284, was not yet
processed and therefore not yet renamed;
4) When processing inode 282 we do not delay its rename operation because
its parent in the send snapshot, inode 281, already has its own rename
operation delayed and our current inode (282) is not an ancestor of
inode 281 in the parent snapshot. Therefore inode 282 is renamed to
"ance/below_ance/pre/wait_dir";
5) When processing inode 283 we realize that we can rename it because one
of its ancestors in the send snapshot, inode 281, has its rename
operation delayed and inode 283 is not an ancestor of inode 281 in the
parent snapshot. So a rename operation to rename inode 283 to
"ance/below_ance/pre/wait_dir/desc/ance" is issued. This path is
invalid due to a missing path building loop that was undetected by
the incremental send implementation, as inode 283 ends up getting
included twice in the path (once with its path in the parent snapshot).
Therefore its rename operation must wait before the ancestor inode 284
is renamed.
Fix this by not terminating the rename dependency checks when we find an
ancestor, in the send snapshot, that has its rename operation delayed. So
that we continue doing the same checks if the current inode is not an
ancestor, in the parent snapshot, of an ancestor in the send snapshot we
are processing in the loop.
The problem and reproducer were reported by Robbie Ko, as part of a patch
titled "Btrfs: incremental send, avoid ancestor rename to descendant".
However the fix was unnecessarily complicated and can be addressed with
much less code and effort.
Reported-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
