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.
It should unlock 'tcon->tc_lock' before return from cifs_tree_connect().
Fixes: fe67bd563ec2 ("cifs: avoid use of global locks for high contention data")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Reviewed-by: Shyam Prasad N <sprasad@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
During analysis of multichannel perf, it was seen that
the global locks cifs_tcp_ses_lock and GlobalMid_Lock, which
were shared between various data structures were causing a
lot of contention points.
With this change, we're breaking down the use of these locks
by introducing new locks at more granular levels. i.e.
server->srv_lock, ses->ses_lock and tcon->tc_lock to protect
the unprotected fields of server, session and tcon structs;
and server->mid_lock to protect mid related lists and entries
at server level.
Signed-off-by: Shyam Prasad N <sprasad@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Removed remaining warnings related to externs. These warnings
although harmless could be distracting e.g.
fs/cifs/cifsfs.c: note: in included file:
fs/cifs/cifsglob.h:1968:24: warning: symbol 'sesInfoAllocCount' was not declared. Should it be static?
Reviewed-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>
Replace list_for_each() by list_for_each_entr() where appropriate.
Remove no longer used list_head stack variables.
Signed-off-by: Enzo Matsumiya <ematsumiya@suse.de>
Signed-off-by: Steve French <stfrench@microsoft.com>
If the command is SMB2_IOCTL, OutputLength and OutputContext are
optional and can be zero, so return early and skip calculated length
check.
Move the mismatched length message to the end of the check, to avoid
unnecessary logs when the check was not a real miscalculation.
Also change the pr_warn_once() to a pr_warn() so we're sure to get a
log for the real mismatches.
Signed-off-by: Enzo Matsumiya <ematsumiya@suse.de>
Signed-off-by: Steve French <stfrench@microsoft.com>
If we hit the 'index == next_cached' case, we leak a refcount on the
struct page. Fix this by using readahead_folio() which takes care of
the refcount for you.
Fixes: 0174ee9947bd ("cifs: Implement cache I/O by accessing the cache directly")
Cc: David Howells <dhowells@redhat.com>
Cc: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
The build warning:
warning: symbol 'cifs_tcp_ses_lock' was not declared. Should it be static?
can be distracting. Fix two of these.
Reviewed-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>
Remove warnings for five global variables. For example:
fs/cifs/cifsglob.h:1984:24: warning: symbol 'midCount' was not declared. Should it be static?
Also change them from camelCase (e.g. "midCount" to "mid_count")
Reviewed-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>
Variable mnt_sign_enabled is being initialized with a value that
is never read, it is being reassigned later on with a different
value. The initialization is redundant and can be removed.
Cleans up clang scan-build warning:
fs/cifs/cifssmb.c:465:7: warning: Value stored to 'mnt_sign_enabled
during its initialization is never read
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Coverity complains about assigning a pointer based on
value length before checking that value length goes
beyond the end of the SMB. Although this is even more
unlikely as value length is a single byte, and the
pointer is not dereferenced until laterm, it is clearer
to check the lengths first.
Addresses-Coverity: 1467704 ("Speculative execution data leak")
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
The allocated memory didn't free under an error
path in smb2_handle_negotiate().
Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
Cc: stable@vger.kernel.org
Reported-by: zdi-disclosures@trendmicro.com # ZDI-CAN-17815
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Reviewed-by: Hyunchul Lee <hyc.lee@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
After multi-channel connection with windows, Several channels of
session are connected. Among them, if there is a problem in one channel,
Windows connects again after disconnecting the channel. In this process,
the session is released and a kernel oop can occurs while processing
requests to other channels. When the channel is disconnected, if other
channels still exist in the session after deleting the channel from
the channel list in the session, the session should not be released.
Finally, the session will be released after all channels are disconnected.
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Reviewed-by: Hyunchul Lee <hyc.lee@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
ksmbd threads eating masses of cputime when connection is disconnected.
If connection is disconnected, ksmbd thread waits for pending requests
to be processed using schedule_timeout. schedule_timeout() incorrectly
is used, and it is more efficient to use wait_event/wake_up than to check
r_count every time with timeout.
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Reviewed-by: Hyunchul Lee <hyc.lee@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
exfat_err() adds the new line at the end of the message by itself,
hence the passed string shouldn't contain a new line. Drop the
superfluous newline letters in the error messages in a few places that
have been put mistakenly.
Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
The ENAMETOOLONG error message is printed at each time when user tries
to operate with a too long name, and this can flood the kernel logs
easily, as every user can trigger this. Let's downgrade this error
message level to a debug message for suppressing the superfluous
logs.
BugLink: https://bugzilla.suse.com/show_bug.cgi?id=1201725
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Currently the error and info messages handled by exfat_err() and co
are tossed to exfat_msg() function that does nothing but passes the
strings with printk() invocation. Not only that this is more overhead
by the indirect calls, but also this makes harder to extend for the
debug print usage; because of the direct printk() call, you cannot
make it for dynamic debug or without debug like the standard helpers
such as pr_debug() or dev_dbg().
For addressing the problem, this patch replaces exfat_*() macro to
expand to pr_*() directly. Along with it, add the new exfat_debug()
macro that is expanded to pr_debug() (which output can be gracefully
suppressed via dyndbg).
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
NLS_NAME_* are bit flags although they are currently defined as enum;
it's casually working so far (from 0 to 2), but it's error-prone and
may bring a problem when we want to add more flag.
This patch changes the definitions of NLS_NAME_* explicitly being bit
flags.
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
LTP has a test for oversized file path renames and it expects the
return value to be ENAMETOOLONG. However, exfat returns EINVAL
unexpectedly in some cases, hence LTP test fails. The further
investigation indicated that the problem happens only when iocharset
isn't set to utf8.
The difference comes from that, in the case of utf8,
exfat_utf8_to_utf16() returns the error -ENAMETOOLONG directly and
it's treated as the final error code. Meanwhile, on other iocharsets,
exfat_nls_to_ucs2() returns the max path size but it sets
NLS_NAME_OVERLEN to lossy flag instead; the caller side checks only
whether lossy flag is set or not, resulting in always -EINVAL
unconditionally.
This patch aligns the return code for both cases by checking the lossy
flag bit and returning ENAMETOOLONG when NLS_NAME_OVERLEN bit is set.
BugLink: https://bugzilla.suse.com/show_bug.cgi?id=1201725
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Since the timestamps need to be updated, the directory entries
will be updated by mark_inode_dirty() whether or not a new
cluster is allocated for the file or directory, so there is no
need to use __exfat_write_inode() to update the directory entries
when allocating a new cluster for a file or directory.
Signed-off-by: Yuezhang Mo <Yuezhang.Mo@sony.com>
Reviewed-by: Andy Wu <Andy.Wu@sony.com>
Reviewed-by: Aoyama Wataru <wataru.aoyama@sony.com>
Reviewed-by: Daniel Palmer <daniel.palmer@sony.com>
Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
This commit moves updating file attributes and timestamps before
calling __exfat_write_inode(), so that all updates of the inode
had been written by __exfat_write_inode(), mark_inode_dirty() is
unneeded.
Signed-off-by: Yuezhang Mo <Yuezhang.Mo@sony.com>
Reviewed-by: Andy Wu <Andy.Wu@sony.com>
Reviewed-by: Aoyama Wataru <wataru.aoyama@sony.com>
Reviewed-by: Daniel Palmer <daniel.palmer@sony.com>
Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
__exfat_write_inode() is used to update file and stream directory
entries, except for file->start_clu and stream->flags.
This commit moves update file->start_clu and stream->flags to
__exfat_write_inode() and reuse __exfat_write_inode() to update
directory entries.
Signed-off-by: Yuezhang Mo <Yuezhang.Mo@sony.com>
Reviewed-by: Andy Wu <Andy.Wu@sony.com>
Reviewed-by: Aoyama Wataru <wataru.aoyama@sony.com>
Reviewed-by: Daniel Palmer <daniel.palmer@sony.com>
Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
delete extra space and tab in blank line, there is no functional change.
Reported-by: Hacash Robot <hacashRobot@santino.com>
Signed-off-by: Xie Shaowen <studentxswpy@163.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
erofs_readdir update ctx->pos after filling a batch of dentries
and it may cause dir/files duplication for NFS readdirplus which
depends on ctx->pos to fill dir correctly. So update ctx->pos for
every emitted dirent in erofs_fill_dentries to fix it.
Also fix the update of ctx->pos when the initial file position has
exceeded nameoff.
Fixes: 3e917cc305c6 ("erofs: make filesystem exportable")
Signed-off-by: Hongnan Li <hongnan.li@linux.alibaba.com>
Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Link: https://lore.kernel.org/r/20220722082732.30935-1-jefflexu@linux.alibaba.com
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
As Wenqing Liu <wenqingliu0120@gmail.com> reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=216285
RIP: 0010:memcpy_erms+0x6/0x10
f2fs_update_meta_page+0x84/0x570 [f2fs]
change_curseg.constprop.0+0x159/0xbd0 [f2fs]
f2fs_do_replace_block+0x5c7/0x18a0 [f2fs]
f2fs_replace_block+0xeb/0x180 [f2fs]
recover_data+0x1abd/0x6f50 [f2fs]
f2fs_recover_fsync_data+0x12ce/0x3250 [f2fs]
f2fs_fill_super+0x4459/0x6190 [f2fs]
mount_bdev+0x2cf/0x3b0
legacy_get_tree+0xed/0x1d0
vfs_get_tree+0x81/0x2b0
path_mount+0x47e/0x19d0
do_mount+0xce/0xf0
__x64_sys_mount+0x12c/0x1a0
do_syscall_64+0x38/0x90
entry_SYSCALL_64_after_hwframe+0x63/0xcd
The root cause is segment type is invalid, so in f2fs_do_replace_block(),
f2fs accesses f2fs_sm_info::curseg_array with out-of-range segment type,
result in accessing invalid curseg->sum_blk during memcpy in
f2fs_update_meta_page(). Fix this by adding sanity check on segment type
in build_sit_entries().
Reported-by: Wenqing Liu <wenqingliu0120@gmail.com>
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
After commit a7eeb823854c ("f2fs: use bitmap in discard_entry"),
MAX_DISCARD_BLOCKS became obsolete, remove it.
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Dipanjan Das <mail.dipanjan.das@gmail.com> reported, syzkaller
found a f2fs bug as below:
RIP: 0010:f2fs_new_node_page+0x19ac/0x1fc0 fs/f2fs/node.c:1295
Call Trace:
write_all_xattrs fs/f2fs/xattr.c:487 [inline]
__f2fs_setxattr+0xe76/0x2e10 fs/f2fs/xattr.c:743
f2fs_setxattr+0x233/0xab0 fs/f2fs/xattr.c:790
f2fs_xattr_generic_set+0x133/0x170 fs/f2fs/xattr.c:86
__vfs_setxattr+0x115/0x180 fs/xattr.c:182
__vfs_setxattr_noperm+0x125/0x5f0 fs/xattr.c:216
__vfs_setxattr_locked+0x1cf/0x260 fs/xattr.c:277
vfs_setxattr+0x13f/0x330 fs/xattr.c:303
setxattr+0x146/0x160 fs/xattr.c:611
path_setxattr+0x1a7/0x1d0 fs/xattr.c:630
__do_sys_lsetxattr fs/xattr.c:653 [inline]
__se_sys_lsetxattr fs/xattr.c:649 [inline]
__x64_sys_lsetxattr+0xbd/0x150 fs/xattr.c:649
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x46/0xb0
NAT entry and nat bitmap can be inconsistent, e.g. one nid is free
in nat bitmap, and blkaddr in its NAT entry is not NULL_ADDR, it
may trigger BUG_ON() in f2fs_new_node_page(), fix it.
Reported-by: Dipanjan Das <mail.dipanjan.das@gmail.com>
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If the inode has the compress flag, it will fail to use
'chattr -c +m' to remove its compress flag and tag no compress flag.
However, the same command will be successful when executed again,
as shown below:
$ touch foo.txt
$ chattr +c foo.txt
$ chattr -c +m foo.txt
chattr: Invalid argument while setting flags on foo.txt
$ chattr -c +m foo.txt
$ f2fs_io getflags foo.txt
get a flag on foo.txt ret=0, flags=nocompression,inline_data
Fix this by removing some checks in f2fs_setflags_common()
that do not affect the original logic. I go through all the
possible scenarios, and the results are as follows. Bold is
the only thing that has changed.
+---------------+-----------+-----------+----------+
| | file flags |
+ command +-----------+-----------+----------+
| | no flag | compr | nocompr |
+---------------+-----------+-----------+----------+
| chattr +c | compr | compr | -EINVAL |
| chattr -c | no flag | no flag | nocompr |
| chattr +m | nocompr | -EINVAL | nocompr |
| chattr -m | no flag | compr | no flag |
| chattr +c +m | -EINVAL | -EINVAL | -EINVAL |
| chattr +c -m | compr | compr | compr |
| chattr -c +m | nocompr | *nocompr* | nocompr |
| chattr -c -m | no flag | no flag | no flag |
+---------------+-----------+-----------+----------+
Link: https://lore.kernel.org/linux-f2fs-devel/20220621064833.1079383-1-chaoliu719@gmail.com/
Fixes: 4c8ff7095bef ("f2fs: support data compression")
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Chao Liu <liuchao@coolpad.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
introduce the below 4 new sysfs node for atomic write statistics.
- current_atomic_write: the total current atomic write block count,
which is not committed yet.
- peak_atomic_write: the peak value of total current atomic write block
count after boot.
- committed_atomic_block: the accumulated total committed atomic write
block count after boot.
- revoked_atomic_block: the accumulated total revoked atomic write block
count after boot.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
f2fs_gc returns -EINVAL via f2fs_balance_fs when there is enough free
secs after write checkpoint, but with gc_merge enabled, it will cause
the sleep time of gc thread to be set to no_gc_sleep_time even if there
are many dirty segments can be selected.
Signed-off-by: qixiaoyu1 <qixiaoyu1@xiaomi.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
After commit e3b49ea36802 ("f2fs: invalidate META_MAPPING before
IPU/DIO write"), invalidate_mapping_pages() will be called to
avoid race condition in between IPU/DIO and readahead for GC.
However, readahead flow is only used for post_read required inode,
so this patch adds check condition to avoids unnecessary page cache
invalidating for non-post_read inode.
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Files created by truncate(1) have a size but no blocks, so
they can be allowed to enable compression.
Signed-off-by: Chao Liu <liuchao@coolpad.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When converting inode to compressed one via ioctl, it needs to check
inline_data, since inline_data flag and compressed flag are incompatible.
Fixes: 4c8ff7095bef ("f2fs: support data compression")
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
f2fs_copy_page() is a wrapper around two kmap() + one memcpy() from/to
the mapped pages. It unnecessarily duplicates a kernel API and it makes
use of kmap(), which is being deprecated in favor of kmap_local_page().
Two main problems with kmap(): (1) It comes with an overhead as mapping
space is restricted and protected by a global lock for synchronization and
(2) it also requires global TLB invalidation when the kmap’s pool wraps
and it might block when the mapping space is fully utilized until a slot
becomes available.
With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Therefore, its
use in __clone_blkaddrs() is safe and should be preferred.
Delete f2fs_copy_page() and use a plain memcpy_page() in the only one
site calling the removed function. memcpy_page() avoids open coding two
kmap_local_page() + one memcpy() between the two kernel virtual addresses.
Suggested-by: Christoph Hellwig <hch@infradead.org>
Suggested-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Quoted from commit e3b49ea36802 ("f2fs: invalidate META_MAPPING before
IPU/DIO write")
"
Encrypted pages during GC are read and cached in META_MAPPING.
However, due to cached pages in META_MAPPING, there is an issue where
newly written pages are lost by IPU or DIO writes.
Thread A - f2fs_gc() Thread B
/* phase 3 */
down_write(i_gc_rwsem)
ra_data_block() ---- (a)
up_write(i_gc_rwsem)
f2fs_direct_IO() :
- down_read(i_gc_rwsem)
- __blockdev_direct_io()
- get_data_block_dio_write()
- f2fs_dio_submit_bio() ---- (b)
- up_read(i_gc_rwsem)
/* phase 4 */
down_write(i_gc_rwsem)
move_data_block() ---- (c)
up_write(i_gc_rwsem)
(a) In phase 3 of f2fs_gc(), up-to-date page is read from storage and
cached in META_MAPPING.
(b) In thread B, writing new data by IPU or DIO write on same blkaddr as
read in (a). cached page in META_MAPPING become out-dated.
(c) In phase 4 of f2fs_gc(), out-dated page in META_MAPPING is copied to
new blkaddr. In conclusion, the newly written data in (b) is lost.
To address this issue, invalidating pages in META_MAPPING before IPU or
DIO write.
"
In previous commit, we missed to cover extent cache hit case, and passed
wrong value for parameter @end of invalidate_mapping_pages(), fix both
issues.
Fixes: 6aa58d8ad20a ("f2fs: readahead encrypted block during GC")
Fixes: e3b49ea36802 ("f2fs: invalidate META_MAPPING before IPU/DIO write")
Cc: Hyeong-Jun Kim <hj514.kim@samsung.com>
Signed-off-by: Chao Yu <chao.yu@oppo.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds a sysfs entry showing the unusable space in a section
made by zone capacity.
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes counting unusable blocks set by zone capacity when
checking the valid block count in a section.
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In order to simplify the complicated per-zone capacity, let's support
only one capacity for entire zoned device.
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Remove the redundant code and use local variant as the
argument directly. Make it more human-readable.
Signed-off-by: duguowei <duguowei@xiaomi.com>
[Jaegeuk Kim: make code neat]
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Introduce memory mode to supports "normal" and "low" memory modes.
"low" mode is to support low memory devices. Because of the nature of
low memory devices, in this mode, f2fs will try to save memory sometimes
by sacrificing performance. "normal" mode is the default mode and same
as before.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
__d_add() and __d_move() wake up waiters on dentry::d_wait from within
the i_seq_dir write held region. This violates the PREEMPT_RT
constraints as the wake up acquires wait_queue_head::lock which is a
"sleeping" spinlock on RT.
There is no requirement to do so. __d_lookup_unhash() has cleared
DCACHE_PAR_LOOKUP and dentry::d_wait and returned the now unreachable wait
queue head pointer to the caller, so the actual wake up can be postponed
until the i_dir_seq write side critical section is left. The only
requirement is that dentry::lock is held across the whole sequence
including the wake up. The previous commit includes an analysis why this
is considered safe.
Move the wake up past end_dir_add() which leaves the i_dir_seq write side
critical section and enables preemption.
For non RT kernels there is no difference because preemption is still
disabled due to dentry::lock being held, but it shortens the time between
wake up and unlocking dentry::lock, which reduces the contention for the
woken up waiter.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
__d_lookup_done() wakes waiters on dentry->d_wait. On PREEMPT_RT we are
not allowed to do that with preemption disabled, since the wakeup
acquired wait_queue_head::lock, which is a "sleeping" spinlock on RT.
Calling it under dentry->d_lock is not a problem, since that is also a
"sleeping" spinlock on the same configs. Unfortunately, two of its
callers (__d_add() and __d_move()) are holding more than just ->d_lock
and that needs to be dealt with.
The key observation is that wakeup can be moved to any point before
dropping ->d_lock.
As a first step to solve this, move the wake up outside of the
hlist_bl_lock() held section.
This is safe because:
Waiters get inserted into ->d_wait only after they'd taken ->d_lock
and observed DCACHE_PAR_LOOKUP in flags. As long as they are
woken up (and evicted from the queue) between the moment __d_lookup_done()
has removed DCACHE_PAR_LOOKUP and dropping ->d_lock, we are safe,
since the waitqueue ->d_wait points to won't get destroyed without
having __d_lookup_done(dentry) called (under ->d_lock).
->d_wait is set only by d_alloc_parallel() and only in case when
it returns a freshly allocated in-lookup dentry. Whenever that happens,
we are guaranteed that __d_lookup_done() will be called for resulting
dentry (under ->d_lock) before the wq in question gets destroyed.
With two exceptions wq lives in call frame of the caller of
d_alloc_parallel() and we have an explicit d_lookup_done() on the
resulting in-lookup dentry before we leave that frame.
One of those exceptions is nfs_call_unlink(), where wq is embedded into
(dynamically allocated) struct nfs_unlinkdata. It is destroyed in
nfs_async_unlink_release() after an explicit d_lookup_done() on the
dentry wq went into.
Remaining exception is d_add_ci(). There wq is what we'd found in
->d_wait of d_add_ci() argument. Callers of d_add_ci() are two
instances of ->d_lookup() and they must have been given an in-lookup
dentry. Which means that they'd been called by __lookup_slow() or
lookup_open(), with wq in the call frame of one of those.
Result of d_alloc_parallel() in d_add_ci() is fed to
d_splice_alias(), which either returns non-NULL (and d_add_ci() does
d_lookup_done()) or feeds dentry to __d_add() that will do
__d_lookup_done() under ->d_lock. That concludes the analysis.
Let __d_lookup_unhash():
1) Lock the lookup hash and clear DCACHE_PAR_LOOKUP
2) Unhash the dentry
3) Retrieve and clear dentry::d_wait
4) Unlock the hash and return the retrieved waitqueue head pointer
5) Let the caller handle the wake up.
6) Rename __d_lookup_done() to __d_lookup_unhash_wake() to enforce
build failures for OOT code that used __d_lookup_done() and is not
aware of the new return value.
This does not yet solve the PREEMPT_RT problem completely because
preemption is still disabled due to i_dir_seq being held for write. This
will be addressed in subsequent steps.
An alternative solution would be to switch the waitqueue to a simple
waitqueue, but aside of Linus not being a fan of them, moving the wake up
closer to the place where dentry::lock is unlocked reduces lock contention
time for the woken up waiter.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://lkml.kernel.org/r/20220613140712.77932-3-bigeasy@linutronix.de
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
i_dir_seq is a sequence counter with a lock which is represented by the
lowest bit. The writer atomically updates the counter which ensures that it
can be modified by only one writer at a time. This requires preemption to
be disabled across the write side critical section.
On !PREEMPT_RT kernels this is implicit by the caller acquiring
dentry::lock. On PREEMPT_RT kernels spin_lock() does not disable preemption
which means that a preempting writer or reader would live lock. It's
therefore required to disable preemption explicitly.
An alternative solution would be to replace i_dir_seq with a seqlock_t for
PREEMPT_RT, but that comes with its own set of problems due to arbitrary
lock nesting. A pure sequence count with an associated spinlock is not
possible because the locks held by the caller are not necessarily related.
As the critical section is small, disabling preemption is a sensible
solution.
Reported-by: Oleg.Karfich@wago.com
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://lkml.kernel.org/r/20220613140712.77932-2-bigeasy@linutronix.de
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
All callers of d_alloc_parallel() must make sure that resulting
in-lookup dentry (if any) will encounter __d_lookup_done() before
the final dput(). d_add_ci() might end up creating in-lookup
dentries; they are fed to d_splice_alias(), which will normally
make sure they meet __d_lookup_done(). However, it is possible
to end up with d_splice_alias() failing with ERR_PTR(-ELOOP)
without having done so. It takes a corrupted ntfs or case-insensitive
xfs image, but neither should end up with memory corruption...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
