linux/fs
Boris Burkov 807d9023e7 btrfs: fix ssd_spread overallocation
If the ssd_spread mount option is enabled, then we run the so called
clustered allocator for data block groups. In practice, this results in
creating a btrfs_free_cluster which caches a block_group and borrows its
free extents for allocation.

Since the introduction of allocation size classes in 6.1, there has been
a bug in the interaction between that feature and ssd_spread.
find_free_extent() has a number of nested loops. The loop going over the
allocation stages, stored in ffe_ctl->loop and managed by
find_free_extent_update_loop(), the loop over the raid levels, and the
loop over all the block_groups in a space_info. The size class feature
relies on the block_group loop to ensure it gets a chance to see a
block_group of a given size class.  However, the clustered allocator
uses the cached cluster block_group and breaks that loop. Each call to
do_allocation() will really just go back to the same cached block_group.
Normally, this is OK, as the allocation either succeeds and we don't
want to loop any more or it fails, and we clear the cluster and return
its space to the block_group.

But with size classes, the allocation can succeed, then later fail,
outside of do_allocation() due to size class mismatch. That latter
failure is not properly handled due to the highly complex multi loop
logic. The result is a painful loop where we continue to allocate the
same num_bytes from the cluster in a tight loop until it fails and
releases the cluster and lets us try a new block_group. But by then, we
have skipped great swaths of the available block_groups and are likely
to fail to allocate, looping the outer loop. In pathological cases like
the reproducer below, the cached block_group is often the very last one,
in which case we don't perform this tight bg loop but instead rip
through the ffe stages to LOOP_CHUNK_ALLOC and allocate a chunk, which
is now the last one, and we enter the tight inner loop until an
allocation failure. Then allocation succeeds on the final block_group
and if the next allocation is a size mismatch, the exact same thing
happens again.

Triggering this is as easy as mounting with -o ssd_spread and then
running:

  mount -o ssd_spread $dev $mnt
  dd if=/dev/zero of=$mnt/big bs=16M count=1 &>/dev/null
  dd if=/dev/zero of=$mnt/med bs=4M count=1 &>/dev/null
  sync

if you do the two writes + sync in a loop, you can force btrfs to spin
an excessive amount on semi-successful clustered allocations, before
ultimately failing and advancing to the stage where we force a chunk
allocation. This results in 2G of data allocated per iteration, despite
only using ~20M of data. By using a small size classed extent, the inner
loop takes longer and we can spin for longer.

The simplest, shortest term fix to unbreak this is to make the clustered
allocator size_class aware in the dumbest way, where it fails on size
class mismatch. This may hinder the operation of the clustered
allocator, but better hindered than completely broken and terribly
overallocating.

Further re-design improvements are also in the works.

Fixes: 52bb7a2166 ("btrfs: introduce size class to block group allocator")
CC: stable@vger.kernel.org # 6.1+
Reported-by: David Sterba <dsterba@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
2025-07-22 01:12:52 +02:00
..
9p vfs-6.16-rc1.netfs 2025-06-02 15:04:06 -07:00
adfs Merge patch series "adfs, affs, befs, hfs, hfsplus: convert to new mount api" 2024-10-08 14:41:53 +02:00
affs vfs-6.15-rc1.async.dir 2025-03-24 10:47:14 -07:00
afs vfs-6.16-rc1.netfs 2025-06-02 15:04:06 -07:00
autofs VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
bcachefs bcachefs: Fix bch2_maybe_casefold() when CONFIG_UTF8=n 2025-07-16 17:32:33 -04:00
befs befs: convert befs to use the new mount api 2024-09-18 11:44:43 +02:00
bfs bfs: convert bfs to use the new mount api 2025-04-07 09:36:20 +02:00
btrfs btrfs: fix ssd_spread overallocation 2025-07-22 01:12:52 +02:00
cachefiles cachefiles: Fix the incorrect return value in __cachefiles_write() 2025-07-10 09:40:17 +02:00
ceph A one-liner that leads to a startling (but also very much rational) 2025-06-06 17:56:19 -07:00
coda Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
configfs - The 3 patch series "hung_task: extend blocking task stacktrace dump to 2025-05-31 19:12:53 -07:00
cramfs
crypto fscrypt: add support for hardware-wrapped keys 2025-04-08 19:32:11 -07:00
debugfs VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
devpts devpts: Fix type for uid and gid params 2025-04-07 15:22:12 +02:00
dlm treewide, timers: Rename from_timer() to timer_container_of() 2025-06-08 09:07:37 +02:00
ecryptfs VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
efivarfs efivarfs: Fix memory leak of efivarfs_fs_info in fs_context error paths 2025-07-18 12:00:20 +02:00
efs efs: fix the efs new mount api implementation 2024-10-15 15:58:36 +02:00
erofs erofs: fix large fragment handling 2025-07-12 04:02:44 +08:00
exfat exfat: do not clear volume dirty flag during sync 2025-05-26 20:25:23 +09:00
exportfs readdir: supply dir_context.count as readdir buffer size hint 2025-05-29 12:31:23 +02:00
ext2 ext2: Deprecate DAX 2025-04-29 13:08:20 +02:00
ext4 treewide, timers: Rename from_timer() to timer_container_of() 2025-06-08 09:07:37 +02:00
f2fs f2fs: fix to zero post-eof page 2025-06-18 21:35:29 +00:00
fat Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
freevxfs freevxfs: Replace one-element array with flexible array member 2024-11-06 10:42:06 +01:00
fuse vfs-6.16-rc5.fixes 2025-07-04 09:06:49 -07:00
gfs2 gfs2: Don't clear sb->s_fs_info in gfs2_sys_fs_add 2025-05-30 19:20:20 +02:00
hfs Revert "hfs{plus}: add deprecation warning" 2025-04-19 22:48:59 +02:00
hfsplus hfsplus: use bdev_rw_virt in hfsplus_submit_bio 2025-05-07 07:31:08 -06:00
hostfs Updates for UML for this cycle, notably: 2025-04-02 12:25:03 -07:00
hpfs Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
hugetlbfs - The 6 patch series "Enable strict percpu address space checks" from 2025-04-01 09:29:18 -07:00
iomap iomap: avoid unnecessary ifs_set_range_uptodate() with locks 2025-07-11 11:42:44 +02:00
isofs isofs: Verify inode mode when loading from disk 2025-07-11 11:39:31 +02:00
jbd2 treewide, timers: Rename from_timer() to timer_container_of() 2025-06-08 09:07:37 +02:00
jffs2 jffs2: check jffs2_prealloc_raw_node_refs() result in few other places 2025-05-22 20:54:38 +02:00
jfs - The 11 patch series "Add folio_mk_pte()" from Matthew Wilcox 2025-05-31 15:44:16 -07:00
kernfs Driver core changes for 6.16-rc1 2025-05-29 09:11:39 -07:00
lockd sysctl: Fixes nsm_local_state bounds 2025-03-10 09:11:13 -04:00
minix Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
netfs netfs: Fix race between cache write completion and ALL_QUEUED being set 2025-07-14 11:05:02 +02:00
nfs NFSv4/flexfiles: Fix handling of NFS level errors in I/O 2025-06-26 13:46:44 -04:00
nfs_common nfs_localio: change nfsd_file_put_local() to take a pointer to __rcu pointer 2025-05-28 17:17:14 -04:00
nfsd nfsd-6.16 fixes: 2025-06-21 09:20:15 -07:00
nilfs2 treewide, timers: Rename from_timer() to timer_container_of() 2025-06-08 09:07:37 +02:00
nls move asm/unaligned.h to linux/unaligned.h 2024-10-02 17:23:23 -04:00
notify fix a leak in fcntl_dirnotify() 2025-07-14 10:13:31 +02:00
ntfs3 - The 2 patch series "zram: support algorithm-specific parameters" from 2025-06-02 16:00:26 -07:00
ocfs2 treewide, timers: Rename from_timer() to timer_container_of() 2025-06-08 09:07:37 +02:00
omfs omfs: convert to new mount API 2025-04-28 10:54:39 +02:00
openpromfs
orangefs orangefs: Convert to use the new mount API 2025-05-28 12:05:30 -07:00
overlayfs vfs-6.16-rc3.fixes 2025-06-16 08:18:43 -07:00
proc mm: fix the inaccurate memory statistics issue for users 2025-07-09 21:07:55 -07:00
pstore treewide: Switch/rename to timer_delete[_sync]() 2025-04-05 10:30:12 +02:00
qnx4
qnx6 fs/qnx6: Fix building with GCC 15 2024-12-03 10:40:36 +01:00
quota VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
ramfs Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
resctrl x86,fs/resctrl: Remove inappropriate references to cacheinfo in the resctrl subsystem 2025-06-16 21:06:12 +02:00
romfs romfs: fix romfs_read_folio() 2024-08-21 22:32:58 +02:00
smb Fix SMB311 posix special file creation to servers which do not advertise reparse support 2025-07-18 12:12:02 -05:00
squashfs squashfs: add optional full compressed block caching 2025-05-27 19:40:33 -07:00
sysfs sysfs: constify attribute_group::bin_attrs 2025-04-15 18:46:10 +02:00
tests
tracefs VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
ubifs This pull request contains the following fixes for JFFS2 and UBIFS: 2025-06-07 07:24:07 -07:00
udf udf: Make sure i_lenExtents is uptodate on inode eviction 2025-05-07 12:04:07 +02:00
ufs ufs: convert ufs to the new mount API 2025-05-14 22:40:55 -04:00
unicode unicode: kunit: change tests filename and path 2025-02-12 14:00:11 -08:00
vboxsf vboxsf: Convert to writepages 2025-04-07 09:36:48 +02:00
verity Revert "fsverity: relax build time dependency on CRYPTO_SHA256" 2025-02-17 11:34:15 -08:00
xfs xfs: don't allocate the xfs_extent_busy structure for zoned RTGs 2025-07-18 17:42:31 +02:00
zonefs zonefs: use bdev_rw_virt in zonefs_read_super 2025-05-07 07:31:07 -06:00
aio.c fs: aio: initialize .ki_write_stream of read-write request 2025-05-07 08:00:11 -06:00
anon_inodes.c fs: export anon_inode_make_secure_inode() and fix secretmem LSM bypass 2025-06-23 12:41:17 +02:00
attr.c fs: handle delegated timestamps in setattr_copy_mgtime 2024-10-10 10:20:51 +02:00
backing-file.c tree-wide: s/revert_creds_light()/revert_creds()/g 2024-12-02 11:25:09 +01:00
bad_inode.c Change inode_operations.mkdir to return struct dentry * 2025-02-27 20:00:17 +01:00
binfmt_elf_fdpic.c binfmt_elf_fdpic: fix variable set but not used warning 2025-03-07 20:07:33 -08:00
binfmt_elf.c vfs-6.16-rc1.misc 2025-05-26 09:02:39 -07:00
binfmt_flat.c binfmt_flat: Fix integer overflow bug on 32 bit systems 2025-01-10 08:49:05 -08:00
binfmt_misc.c VFS: rename lookup_one_len family to lookup_noperm and remove permission check 2025-04-08 11:24:36 +02:00
binfmt_script.c
bpf_fs_kfuncs.c bpf: fs/xattr: Add BPF kfuncs to set and remove xattrs 2025-02-13 19:35:32 -08:00
buffer.c vfs-6.16-rc1.writepage 2025-05-26 08:23:09 -07:00
char_dev.c fs: Reorganize kerneldoc parameter names 2024-10-22 11:16:57 +02:00
compat_binfmt_elf.c binfmt_elf: Wire up AT_HWCAP3 at AT_HWCAP4 2024-10-17 18:38:49 +01:00
coredump.c coredump: validate socket name as it is written 2025-05-21 13:59:12 +02:00
d_path.c
dax.c vfs-6.16-rc2.fixes 2025-06-02 12:49:16 -07:00
dcache.c vfs-6.16-rc1.misc 2025-05-26 09:02:39 -07:00
direct-io.c fs/direct-io: Remove linux/prefetch.h include 2024-08-19 13:45:02 +02:00
drop_caches.c fs: drop_caches: move sysctl to fs/drop_caches.c 2025-02-07 16:53:04 +01:00
eventfd.c make use of anon_inode_getfile_fmode() 2025-02-21 10:25:31 +01:00
eventpoll.c Revert "eventpoll: Fix priority inversion problem" 2025-07-11 17:10:32 -07:00
exec.c anon_inode: rework assertions 2025-07-02 14:41:39 +02:00
fcntl.c fs: get rid of __FMODE_NONOTIFY kludge 2024-12-09 11:34:29 +01:00
fhandle.c exportfs: add permission method 2024-12-17 09:16:11 +01:00
file_table.c fs: Make file-nr output the total allocated file handles 2025-04-21 10:27:58 +02:00
file.c fs: drop assert in file_seek_cur_needs_f_lock 2025-06-16 09:59:24 +02:00
filesystems.c fs/filesystems: Fix potential unsigned integer underflow in fs_name() 2025-04-14 13:05:59 +02:00
fs_context.c fs/fs_context: Mark an unlikely if condition with unlikely() in vfs_parse_monolithic_sep() 2025-04-14 13:05:59 +02:00
fs_parser.c fs/fs_parse: Remove unused and problematic validate_constant_table() 2025-04-21 10:27:59 +02:00
fs_pin.c
fs_struct.c
fs_types.c
fs-writeback.c fs: fs-writeback: move sysctl to fs/fs-writeback.c 2025-02-07 16:53:04 +01:00
fsopen.c fs: support O_PATH fds with FSCONFIG_SET_FD 2025-02-12 10:02:10 +01:00
init.c VFS: Change vfs_mkdir() to return the dentry. 2025-03-05 11:52:50 +01:00
inode.c fs: call inode_sb_list_add() outside of inode hash lock 2025-03-20 13:06:51 +01:00
internal.h vfs-6.16-rc1.super 2025-05-26 09:33:44 -07:00
ioctl.c vfs-6.16-rc1.super 2025-05-26 09:33:44 -07:00
Kconfig fs/resctrl: Add boiler plate for external resctrl code 2025-05-16 11:05:40 +02:00
Kconfig.binfmt
kernel_read_file.c fdget(), trivial conversions 2024-11-03 01:28:06 -05:00
libfs.c anon_inode: rework assertions 2025-07-02 14:41:39 +02:00
locks.c treewide: const qualify ctl_tables where applicable 2025-01-28 13:48:37 +01:00
Makefile fs/resctrl: Add boiler plate for external resctrl code 2025-05-16 11:05:40 +02:00
mbcache.c
mnt_idmapping.c statmount: allow to retrieve idmappings 2025-02-12 12:12:27 +01:00
mount.h Don't propagate mounts into detached trees 2025-05-26 17:35:32 -04:00
mpage.c fs: use writeback_iter directly in mpage_writepages 2025-05-09 12:37:48 +02:00
namei.c vfs-6.16-rc5.fixes 2025-07-04 09:06:49 -07:00
namespace.c userns and mnt_idmap leak in open_tree_attr(2) 2025-06-24 10:25:04 -04:00
nsfs.c vfs-6.15-rc1.nsfs 2025-03-24 11:38:12 -07:00
open.c fs/open: make do_truncate() killable 2025-05-15 12:03:12 +02:00
pidfs.c coredump: fix PIDFD_INFO_COREDUMP ioctl check 2025-07-07 13:26:03 +02:00
pipe.c sort.h: hoist cmp_int() into generic header file 2025-05-11 17:54:12 -07:00
pnode.c Don't propagate mounts into detached trees 2025-05-26 17:35:32 -04:00
pnode.h replace collect_mounts()/drop_collected_mounts() with a safer variant 2025-06-23 14:01:49 -04:00
posix_acl.c acl: Annotate struct posix_acl with __counted_by() 2024-10-22 11:16:59 +02:00
proc_namespace.c ->mnt_devname is never NULL 2025-05-23 14:20:44 +02:00
read_write.c fs/read_write: make default_llseek() killable 2025-05-15 12:03:12 +02:00
readdir.c readdir: supply dir_context.count as readdir buffer size hint 2025-05-29 12:31:23 +02:00
remap_range.c convert vfs_dedupe_file_range(). 2024-11-03 01:28:07 -05:00
select.c select: core_sys_select add unlikely branch hint on return path 2025-04-21 10:27:58 +02:00
seq_file.c fs: Reorganize kerneldoc parameter names 2024-10-22 11:16:57 +02:00
signalfd.c make use of anon_inode_getfile_fmode() 2025-02-21 10:25:31 +01:00
splice.c splice: remove duplicate noinline from pipe_clear_nowait 2025-04-25 12:11:56 +02:00
stack.c
stat.c xfs: New code for 6.16 2025-05-26 12:56:01 -07:00
statfs.c fdget_raw() users: switch to CLASS(fd_raw) 2024-11-03 01:28:06 -05:00
super.c fs: unlock the superblock during iterate_supers_type 2025-06-12 14:27:39 +02:00
sync.c fdget(), trivial conversions 2024-11-03 01:28:06 -05:00
sysctls.c treewide: const qualify ctl_tables where applicable 2025-01-28 13:48:37 +01:00
timerfd.c A treewide hrtimer timer cleanup 2025-03-25 10:54:15 -07:00
userfaultfd.c mm/userfaultfd: fix uninitialized output field for -EAGAIN race 2025-05-07 23:39:39 -07:00
utimes.c fdget(), more trivial conversions 2024-11-03 01:28:06 -05:00
xattr.c fs/xattr.c: fix simple_xattr_list() 2025-06-06 10:00:17 +02:00