This release, we rework the implementation of creating new encrypted
files in order to fix some deadlocks and prepare for adding fscrypt
support to CephFS, which Jeff Layton is working on.
We also export a symbol in preparation for the above-mentioned CephFS
support and also for ext4/f2fs encrypt+casefold support.
Finally, there are a few other small cleanups.
As usual, all these patches have been in linux-next with no reported
issues, and I've tested them with xfstests.
-----BEGIN PGP SIGNATURE-----
iIoEABYIADIWIQSacvsUNc7UX4ntmEPzXCl4vpKOKwUCX4SD7xQcZWJpZ2dlcnNA
Z29vZ2xlLmNvbQAKCRDzXCl4vpKOKy/AAP92oOybTcuahmvAtHqZP9jAFPJrbI3r
6QLpMFtWznJoOQEAogaWsavtOIBx9afdOfRNj0zdoBIjpXgyMuzR10Ou2gE=
=B/Mj
-----END PGP SIGNATURE-----
Merge tag 'fscrypt-for-linus' of git://git.kernel.org/pub/scm/fs/fscrypt/fscrypt
Pull fscrypt updates from Eric Biggers:
"This release, we rework the implementation of creating new encrypted
files in order to fix some deadlocks and prepare for adding fscrypt
support to CephFS, which Jeff Layton is working on.
We also export a symbol in preparation for the above-mentioned CephFS
support and also for ext4/f2fs encrypt+casefold support.
Finally, there are a few other small cleanups.
As usual, all these patches have been in linux-next with no reported
issues, and I've tested them with xfstests"
* tag 'fscrypt-for-linus' of git://git.kernel.org/pub/scm/fs/fscrypt/fscrypt:
fscrypt: export fscrypt_d_revalidate()
fscrypt: rename DCACHE_ENCRYPTED_NAME to DCACHE_NOKEY_NAME
fscrypt: don't call no-key names "ciphertext names"
fscrypt: use sha256() instead of open coding
fscrypt: make fscrypt_set_test_dummy_encryption() take a 'const char *'
fscrypt: handle test_dummy_encryption in more logical way
fscrypt: move fscrypt_prepare_symlink() out-of-line
fscrypt: make "#define fscrypt_policy" user-only
fscrypt: stop pretending that key setup is nofs-safe
fscrypt: require that fscrypt_encrypt_symlink() already has key
fscrypt: remove fscrypt_inherit_context()
fscrypt: adjust logging for in-creation inodes
ubifs: use fscrypt_prepare_new_inode() and fscrypt_set_context()
f2fs: use fscrypt_prepare_new_inode() and fscrypt_set_context()
ext4: use fscrypt_prepare_new_inode() and fscrypt_set_context()
ext4: factor out ext4_xattr_credits_for_new_inode()
fscrypt: add fscrypt_prepare_new_inode() and fscrypt_set_context()
fscrypt: restrict IV_INO_LBLK_32 to ino_bits <= 32
fscrypt: drop unused inode argument from fscrypt_fname_alloc_buffer
If compressed inode has inconsistent fields on i_compress_algorithm,
i_compr_blocks and i_log_cluster_size, we missed to set SBI_NEED_FSCK
to notice fsck to repair the inode, fix it.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
syzbot reported:
general protection fault, probably for non-canonical address 0xdffffc0000000001: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
CPU: 0 PID: 6860 Comm: syz-executor835 Not tainted 5.9.0-rc8-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:utf8_casefold+0x43/0x1b0 fs/unicode/utf8-core.c:107
[...]
Call Trace:
f2fs_init_casefolded_name fs/f2fs/dir.c:85 [inline]
__f2fs_setup_filename fs/f2fs/dir.c:118 [inline]
f2fs_prepare_lookup+0x3bf/0x640 fs/f2fs/dir.c:163
f2fs_lookup+0x10d/0x920 fs/f2fs/namei.c:494
__lookup_hash+0x115/0x240 fs/namei.c:1445
filename_create+0x14b/0x630 fs/namei.c:3467
user_path_create fs/namei.c:3524 [inline]
do_mkdirat+0x56/0x310 fs/namei.c:3664
do_syscall_64+0x31/0x70 arch/x86/entry/common.c:46
entry_SYSCALL_64_after_hwframe+0x44/0xa9
[...]
The problem is that an inode has F2FS_CASEFOLD_FL set, but the
filesystem doesn't have the casefold feature flag set, and therefore
super_block::s_encoding is NULL.
Fix this by making sanity_check_inode() reject inodes that have
F2FS_CASEFOLD_FL when the filesystem doesn't have the casefold feature.
Reported-by: syzbot+05139c4039d0679e19ff@syzkaller.appspotmail.com
Fixes: 2c2eb7a300cd ("f2fs: Support case-insensitive file name lookups")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In 32bit system, 64-bits key breaks memory alignment.
This fixes the commit "f2fs: support 64-bits key in f2fs rb-tree node entry".
Reported-by: Nicolas Chauvet <kwizart@gmail.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Compressed inode and normal inode has different layout, so we should
disallow enabling compress on non-empty file to avoid race condition
during inode .i_addr array parsing and updating.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: Fix missing condition]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Add two slab caches: "f2fs_cic_entry" and "f2fs_dic_entry" for memory
allocation of compress_io_ctx and decompress_io_ctx structure.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As syzbot reported:
BUG: KASAN: slab-out-of-bounds in init_min_max_mtime fs/f2fs/segment.c:4710 [inline]
BUG: KASAN: slab-out-of-bounds in f2fs_build_segment_manager+0x9302/0xa6d0 fs/f2fs/segment.c:4792
Read of size 8 at addr ffff8880a1b934a8 by task syz-executor682/6878
CPU: 1 PID: 6878 Comm: syz-executor682 Not tainted 5.9.0-rc6-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x198/0x1fd lib/dump_stack.c:118
print_address_description.constprop.0.cold+0xae/0x497 mm/kasan/report.c:383
__kasan_report mm/kasan/report.c:513 [inline]
kasan_report.cold+0x1f/0x37 mm/kasan/report.c:530
init_min_max_mtime fs/f2fs/segment.c:4710 [inline]
f2fs_build_segment_manager+0x9302/0xa6d0 fs/f2fs/segment.c:4792
f2fs_fill_super+0x381a/0x6e80 fs/f2fs/super.c:3633
mount_bdev+0x32e/0x3f0 fs/super.c:1417
legacy_get_tree+0x105/0x220 fs/fs_context.c:592
vfs_get_tree+0x89/0x2f0 fs/super.c:1547
do_new_mount fs/namespace.c:2875 [inline]
path_mount+0x1387/0x20a0 fs/namespace.c:3192
do_mount fs/namespace.c:3205 [inline]
__do_sys_mount fs/namespace.c:3413 [inline]
__se_sys_mount fs/namespace.c:3390 [inline]
__x64_sys_mount+0x27f/0x300 fs/namespace.c:3390
do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
entry_SYSCALL_64_after_hwframe+0x44/0xa9
The root cause is: if segs_per_sec is larger than one, and segment count
in last section is less than segs_per_sec, we will suffer out-of-boundary
memory access on sit_i->sentries[] in init_min_max_mtime().
Fix this by adding sanity check among segment count, section count and
segs_per_sec value in sanity_check_raw_super().
Reported-by: syzbot+481a3ffab50fed41dcc0@syzkaller.appspotmail.com
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As syzbot reported:
Call Trace:
__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x21c/0x280 lib/dump_stack.c:118
kmsan_report+0xf7/0x1e0 mm/kmsan/kmsan_report.c:122
__msan_warning+0x58/0xa0 mm/kmsan/kmsan_instr.c:219
f2fs_lookup+0xe05/0x1a80 fs/f2fs/namei.c:503
lookup_open fs/namei.c:3082 [inline]
open_last_lookups fs/namei.c:3177 [inline]
path_openat+0x2729/0x6a90 fs/namei.c:3365
do_filp_open+0x2b8/0x710 fs/namei.c:3395
do_sys_openat2+0xa88/0x1140 fs/open.c:1168
do_sys_open fs/open.c:1184 [inline]
__do_compat_sys_openat fs/open.c:1242 [inline]
__se_compat_sys_openat+0x2a4/0x310 fs/open.c:1240
__ia32_compat_sys_openat+0x56/0x70 fs/open.c:1240
do_syscall_32_irqs_on arch/x86/entry/common.c:80 [inline]
__do_fast_syscall_32+0x129/0x180 arch/x86/entry/common.c:139
do_fast_syscall_32+0x6a/0xc0 arch/x86/entry/common.c:162
do_SYSENTER_32+0x73/0x90 arch/x86/entry/common.c:205
entry_SYSENTER_compat_after_hwframe+0x4d/0x5c
In f2fs_lookup(), @res_page could be used before being initialized,
because in __f2fs_find_entry(), once F2FS_I(dir)->i_current_depth was
been fuzzed to zero, then @res_page will never be initialized, causing
this kmsan warning, relocating @res_page initialization place to fix
this bug.
Reported-by: syzbot+0eac6f0bbd558fd866d7@syzkaller.appspotmail.com
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We can relocate @res_page assignment in find_in_block() to
its caller, so unneeded parameter could be removed for cleanup.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Meta area is not included in section_count computation.
So the minimum number of total_sections is 1 meanwhile it cannot be
greater than segment_count_main.
The minimum number of meta segments is 8 (SB + 2 (CP + SIT + NAT) + SSA).
Signed-off-by: Wang Xiaojun <wangxiaojun11@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
A NULL will not be return by __bitmap_ptr here.
Remove the unused check.
Signed-off-by: Wang Xiaojun <wangxiaojun11@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Relocate blkzoned feature check into parse_options() like
other feature check.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The type of SM_I(sbi)->reserved_segments is unsigned int,
so change the return value to unsigned int.
The type cast can be removed in reserved_sections as a result.
Signed-off-by: Xiaojun Wang <wangxiaojun11@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Currently we're using the term "ciphertext name" ambiguously because it
can mean either the actual ciphertext filename, or the encoded filename
that is shown when an encrypted directory is listed without its key.
The latter we're now usually calling the "no-key name"; and while it's
derived from the ciphertext name, it's not the same thing.
To avoid this ambiguity, rename fscrypt_name::is_ciphertext_name to
fscrypt_name::is_nokey_name, and update comments that say "ciphertext
name" (or "encrypted name") to say "no-key name" instead when warranted.
Link: https://lore.kernel.org/r/20200924042624.98439-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
fscrypt_set_test_dummy_encryption() requires that the optional argument
to the test_dummy_encryption mount option be specified as a substring_t.
That doesn't work well with filesystems that use the new mount API,
since the new way of parsing mount options doesn't use substring_t.
Make it take the argument as a 'const char *' instead.
Instead of moving the match_strdup() into the callers in ext4 and f2fs,
make them just use arg->from directly. Since the pattern is
"test_dummy_encryption=%s", the argument will be null-terminated.
Acked-by: Jeff Layton <jlayton@kernel.org>
Link: https://lore.kernel.org/r/20200917041136.178600-14-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
The behavior of the test_dummy_encryption mount option is that when a
new file (or directory or symlink) is created in an unencrypted
directory, it's automatically encrypted using a dummy encryption policy.
That's it; in particular, the encryption (or lack thereof) of existing
files (or directories or symlinks) doesn't change.
Unfortunately the implementation of test_dummy_encryption is a bit weird
and confusing. When test_dummy_encryption is enabled and a file is
being created in an unencrypted directory, we set up an encryption key
(->i_crypt_info) for the directory. This isn't actually used to do any
encryption, however, since the directory is still unencrypted! Instead,
->i_crypt_info is only used for inheriting the encryption policy.
One consequence of this is that the filesystem ends up providing a
"dummy context" (policy + nonce) instead of a "dummy policy". In
commit ed318a6cc0b6 ("fscrypt: support test_dummy_encryption=v2"), I
mistakenly thought this was required. However, actually the nonce only
ends up being used to derive a key that is never used.
Another consequence of this implementation is that it allows for
'inode->i_crypt_info != NULL && !IS_ENCRYPTED(inode)', which is an edge
case that can be forgotten about. For example, currently
FS_IOC_GET_ENCRYPTION_POLICY on an unencrypted directory may return the
dummy encryption policy when the filesystem is mounted with
test_dummy_encryption. That seems like the wrong thing to do, since
again, the directory itself is not actually encrypted.
Therefore, switch to a more logical and maintainable implementation
where the dummy encryption policy inheritance is done without setting up
keys for unencrypted directories. This involves:
- Adding a function fscrypt_policy_to_inherit() which returns the
encryption policy to inherit from a directory. This can be a real
policy, a dummy policy, or no policy.
- Replacing struct fscrypt_dummy_context, ->get_dummy_context(), etc.
with struct fscrypt_dummy_policy, ->get_dummy_policy(), etc.
- Making fscrypt_fname_encrypted_size() take an fscrypt_policy instead
of an inode.
Acked-by: Jaegeuk Kim <jaegeuk@kernel.org>
Acked-by: Jeff Layton <jlayton@kernel.org>
Link: https://lore.kernel.org/r/20200917041136.178600-13-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Convert f2fs to use the new functions fscrypt_prepare_new_inode() and
fscrypt_set_context(). This avoids calling
fscrypt_get_encryption_info() from under f2fs_lock_op(), which can
deadlock because fscrypt_get_encryption_info() isn't GFP_NOFS-safe.
For more details about this problem, see the earlier patch
"fscrypt: add fscrypt_prepare_new_inode() and fscrypt_set_context()".
This also fixes a f2fs-specific deadlock when the filesystem is mounted
with '-o test_dummy_encryption' and a file is created in an unencrypted
directory other than the root directory:
INFO: task touch:207 blocked for more than 30 seconds.
Not tainted 5.9.0-rc4-00099-g729e3d0919844 #2
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:touch state:D stack: 0 pid: 207 ppid: 167 flags:0x00000000
Call Trace:
[...]
lock_page include/linux/pagemap.h:548 [inline]
pagecache_get_page+0x25e/0x310 mm/filemap.c:1682
find_or_create_page include/linux/pagemap.h:348 [inline]
grab_cache_page include/linux/pagemap.h:424 [inline]
f2fs_grab_cache_page fs/f2fs/f2fs.h:2395 [inline]
f2fs_grab_cache_page fs/f2fs/f2fs.h:2373 [inline]
__get_node_page.part.0+0x39/0x2d0 fs/f2fs/node.c:1350
__get_node_page fs/f2fs/node.c:35 [inline]
f2fs_get_node_page+0x2e/0x60 fs/f2fs/node.c:1399
read_inline_xattr+0x88/0x140 fs/f2fs/xattr.c:288
lookup_all_xattrs+0x1f9/0x2c0 fs/f2fs/xattr.c:344
f2fs_getxattr+0x9b/0x160 fs/f2fs/xattr.c:532
f2fs_get_context+0x1e/0x20 fs/f2fs/super.c:2460
fscrypt_get_encryption_info+0x9b/0x450 fs/crypto/keysetup.c:472
fscrypt_inherit_context+0x2f/0xb0 fs/crypto/policy.c:640
f2fs_init_inode_metadata+0xab/0x340 fs/f2fs/dir.c:540
f2fs_add_inline_entry+0x145/0x390 fs/f2fs/inline.c:621
f2fs_add_dentry+0x31/0x80 fs/f2fs/dir.c:757
f2fs_do_add_link+0xcd/0x130 fs/f2fs/dir.c:798
f2fs_add_link fs/f2fs/f2fs.h:3234 [inline]
f2fs_create+0x104/0x290 fs/f2fs/namei.c:344
lookup_open.isra.0+0x2de/0x500 fs/namei.c:3103
open_last_lookups+0xa9/0x340 fs/namei.c:3177
path_openat+0x8f/0x1b0 fs/namei.c:3365
do_filp_open+0x87/0x130 fs/namei.c:3395
do_sys_openat2+0x96/0x150 fs/open.c:1168
[...]
That happened because f2fs_add_inline_entry() locks the directory
inode's page in order to add the dentry, then f2fs_get_context() tries
to lock it recursively in order to read the encryption xattr. This
problem is specific to "test_dummy_encryption" because normally the
directory's fscrypt_info would be set up prior to
f2fs_add_inline_entry() in order to encrypt the new filename.
Regardless, the new design fixes this test_dummy_encryption deadlock as
well as potential deadlocks with fs reclaim, by setting up any needed
fscrypt_info structs prior to taking so many locks.
The test_dummy_encryption deadlock was reported by Daniel Rosenberg.
Reported-by: Daniel Rosenberg <drosen@google.com>
Acked-by: Jaegeuk Kim <jaegeuk@kernel.org>
Link: https://lore.kernel.org/r/20200917041136.178600-5-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
After commit 0b6d4ca04a86 ("f2fs: don't return vmalloc() memory from
f2fs_kmalloc()"), f2fs_k{m,z}alloc() will not return vmalloc()'ed
memory, so clean up to use kfree() instead of kvfree() to free
vmalloc()'ed memory.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The returned integer is not required anywhere. So we need to change
the return value to bool type.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
writepages() can be concurrently invoked for the same file by different
threads such as a thread fsyncing the file and a kworker kernel thread.
So, changing i_compr_blocks without protection is racy and we need to
protect it by changing it with atomic type value. Plus, we don't need
a 64bit value for i_compr_blocks, so just we will use a atomic value,
not atomic64.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
to keep consistent with behavior when passing compress mount option
to kernel w/o compression feature, so that mount may not fail on
such condition.
Reported-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As 5kft <5kft@5kft.org> reported:
kworker/u9:3: page allocation failure: order:9, mode:0x40c40(GFP_NOFS|__GFP_COMP), nodemask=(null),cpuset=/,mems_allowed=0
CPU: 3 PID: 8168 Comm: kworker/u9:3 Tainted: G C 5.8.3-sunxi #trunk
Hardware name: Allwinner sun8i Family
Workqueue: f2fs_post_read_wq f2fs_post_read_work
[<c010d6d5>] (unwind_backtrace) from [<c0109a55>] (show_stack+0x11/0x14)
[<c0109a55>] (show_stack) from [<c056d489>] (dump_stack+0x75/0x84)
[<c056d489>] (dump_stack) from [<c0243b53>] (warn_alloc+0xa3/0x104)
[<c0243b53>] (warn_alloc) from [<c024473b>] (__alloc_pages_nodemask+0xb87/0xc40)
[<c024473b>] (__alloc_pages_nodemask) from [<c02267c5>] (kmalloc_order+0x19/0x38)
[<c02267c5>] (kmalloc_order) from [<c02267fd>] (kmalloc_order_trace+0x19/0x90)
[<c02267fd>] (kmalloc_order_trace) from [<c047c665>] (zstd_init_decompress_ctx+0x21/0x88)
[<c047c665>] (zstd_init_decompress_ctx) from [<c047e9cf>] (f2fs_decompress_pages+0x97/0x228)
[<c047e9cf>] (f2fs_decompress_pages) from [<c045d0ab>] (__read_end_io+0xfb/0x130)
[<c045d0ab>] (__read_end_io) from [<c045d141>] (f2fs_post_read_work+0x61/0x84)
[<c045d141>] (f2fs_post_read_work) from [<c0130b2f>] (process_one_work+0x15f/0x3b0)
[<c0130b2f>] (process_one_work) from [<c0130e7b>] (worker_thread+0xfb/0x3e0)
[<c0130e7b>] (worker_thread) from [<c0135c3b>] (kthread+0xeb/0x10c)
[<c0135c3b>] (kthread) from [<c0100159>]
zstd may allocate large size memory for {,de}compression, it may cause
file copy failure on low-end device which has very few memory.
For decompression, let's just allocate proper size memory based on current
file's cluster size instead of max cluster size.
Reported-by: 5kft <5kft@5kft.org>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Current compr_blocks of superblock info is not 64bit value. We are
accumulating each i_compr_blocks count of inodes to this value and
those are 64bit values. So, need to change this to 64bit value.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Need to add block address range check to compressed file case and
avoid calling get_data_block_bmap() for compressed file.
Signed-off-by: Daeho Jeong <daehojeong@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When the move range ioctl is used, check the input and output position and
ensure that it is a non-negative value. Without this check
f2fs_get_dnode_of_data may hit a memmory bug.
Signed-off-by: Dan Robertson <dan@dlrobertson.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Miss to update APP_DIRECT_IO/APP_DIRECT_READ_IO when receiving async DIO.
For example: fio -filename=/data/test.0 -bs=1m -ioengine=libaio -direct=1
-name=fill -size=10m -numjobs=1 -iodepth=32 -rw=write
Signed-off-by: Jack Qiu <jack.qiu@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Instead of finding the first dirty page and then seeing if it matches
the index of a block that is NEW_ADDR, delay the lookup of the dirty
bit until we've actually found a block that's NEW_ADDR.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
There are several issues in current background GC algorithm:
- valid blocks is one of key factors during cost overhead calculation,
so if segment has less valid block, however even its age is young or
it locates hot segment, CB algorithm will still choose the segment as
victim, it's not appropriate.
- GCed data/node will go to existing logs, no matter in-there datas'
update frequency is the same or not, it may mix hot and cold data
again.
- GC alloctor mainly use LFS type segment, it will cost free segment
more quickly.
This patch introduces a new algorithm named age threshold based
garbage collection to solve above issues, there are three steps
mainly:
1. select a source victim:
- set an age threshold, and select candidates beased threshold:
e.g.
0 means youngest, 100 means oldest, if we set age threshold to 80
then select dirty segments which has age in range of [80, 100] as
candiddates;
- set candidate_ratio threshold, and select candidates based the
ratio, so that we can shrink candidates to those oldest segments;
- select target segment with fewest valid blocks in order to
migrate blocks with minimum cost;
2. select a target victim:
- select candidates beased age threshold;
- set candidate_radius threshold, search candidates whose age is
around source victims, searching radius should less than the
radius threshold.
- select target segment with most valid blocks in order to avoid
migrating current target segment.
3. merge valid blocks from source victim into target victim with
SSR alloctor.
Test steps:
- create 160 dirty segments:
* half of them have 128 valid blocks per segment
* left of them have 384 valid blocks per segment
- run background GC
Benefit: GC count and block movement count both decrease obviously:
- Before:
- Valid: 86
- Dirty: 1
- Prefree: 11
- Free: 6001 (6001)
GC calls: 162 (BG: 220)
- data segments : 160 (160)
- node segments : 2 (2)
Try to move 41454 blocks (BG: 41454)
- data blocks : 40960 (40960)
- node blocks : 494 (494)
IPU: 0 blocks
SSR: 0 blocks in 0 segments
LFS: 41364 blocks in 81 segments
- After:
- Valid: 87
- Dirty: 0
- Prefree: 4
- Free: 6008 (6008)
GC calls: 75 (BG: 76)
- data segments : 74 (74)
- node segments : 1 (1)
Try to move 12813 blocks (BG: 12813)
- data blocks : 12544 (12544)
- node blocks : 269 (269)
IPU: 0 blocks
SSR: 12032 blocks in 77 segments
LFS: 855 blocks in 2 segments
Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: fix a bug along with pinfile in-mem segment & clean up]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This switches f2fs over to the generic support provided in
the previous patch.
Since casefolded dentries behave the same in ext4 and f2fs, we decrease
the maintenance burden by unifying them, and any optimizations will
immediately apply to both.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
refcount_t type variable should never be less than one, so it's a
little bit hard to understand when we use it to indicate pending
compressed page count, let's change to use atomic_t for better
readability.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes below compile warning reported by LKP
(kernel test robot)
cppcheck warnings: (new ones prefixed by >>)
>> fs/f2fs/file.c:761:9: warning: Identical condition 'err', second condition is always false [identicalConditionAfterEarlyExit]
return err;
^
fs/f2fs/file.c:753:6: note: first condition
if (err)
^
fs/f2fs/file.c:761:9: note: second condition
return err;
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
then, we can add specified entry into rb-tree with 64-bits segment time
as key.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Don't let f2fs inner GC ruins original aging degree of segment.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, once we update one block in segment, we will update mtime of
segment to last time, making aged segment becoming freshest, result in
that GC with cost benefit algorithm missing such segment, So this patch
changes to record mtime as average block updating time instead of last
updating time.
It's not needed to reset mtime for prefree segment, as se->valid_blocks
is zero, then old se->mtime won't take any weight with below calculation:
se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
se->valid_blocks + 1);
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previous implementation of aligned pinfile allocation will:
- allocate new segment on cold data log no matter whether last used
segment is partially used or not, it makes IOs more random;
- force concurrent cold data/GCed IO going into warm data area, it
can make a bad effect on hot/cold data separation;
In this patch, we introduce a new type of log named 'inmem curseg',
the differents from normal curseg is:
- it reuses existed segment type (CURSEG_XXX_NODE/DATA);
- it only exists in memory, its segno, blkofs, summary will not b
persisted into checkpoint area;
With this new feature, we can enhance scalability of log, special
allocators can be created for purposes:
- pure lfs allocator for aligned pinfile allocation or file
defragmentation
- pure ssr allocator for later feature
So that, let's update aligned pinfile allocation to use this new
inmem curseg fwk.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Since DUMMY_WRITTEN_PAGE and ATOMIC_WRITTEN_PAGE have already been
converted as unsigned long type, we don't need do type casting again.
Signed-off-by: Xiaojun Wang <wangxiaojun11@huawei.com>
Reported-by: Jack Qiu <jack.qiu@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
NVMe Zoned Namespace devices can have zone-capacity less than zone-size.
Zone-capacity indicates the maximum number of sectors that are usable in
a zone beginning from the first sector of the zone. This makes the sectors
sectors after the zone-capacity till zone-size to be unusable.
This patch set tracks zone-size and zone-capacity in zoned devices and
calculate the usable blocks per segment and usable segments per section.
If zone-capacity is less than zone-size mark only those segments which
start before zone-capacity as free segments. All segments at and beyond
zone-capacity are treated as permanently used segments. In cases where
zone-capacity does not align with segment size the last segment will start
before zone-capacity and end beyond the zone-capacity of the zone. For
such spanning segments only sectors within the zone-capacity are used.
During writes and GC manage the usable segments in a section and usable
blocks per segment. Segments which are beyond zone-capacity are never
allocated, and do not need to be garbage collected, only the segments
which are before zone-capacity needs to garbage collected.
For spanning segments based on the number of usable blocks in that
segment, write to blocks only up to zone-capacity.
Zone-capacity is device specific and cannot be configured by the user.
Since NVMe ZNS device zones are sequentially write only, a block device
with conventional zones or any normal block device is needed along with
the ZNS device for the metadata operations of F2fs.
A typical nvme-cli output of a zoned device shows zone start and capacity
and write pointer as below:
SLBA: 0x0 WP: 0x0 Cap: 0x18800 State: EMPTY Type: SEQWRITE_REQ
SLBA: 0x20000 WP: 0x20000 Cap: 0x18800 State: EMPTY Type: SEQWRITE_REQ
SLBA: 0x40000 WP: 0x40000 Cap: 0x18800 State: EMPTY Type: SEQWRITE_REQ
Here zone size is 64MB, capacity is 49MB, WP is at zone start as the zones
are in EMPTY state. For each zone, only zone start + 49MB is usable area,
any lba/sector after 49MB cannot be read or written to, the drive will fail
any attempts to read/write. So, the second zone starts at 64MB and is
usable till 113MB (64 + 49) and the range between 113 and 128MB is
again unusable. The next zone starts at 128MB, and so on.
Signed-off-by: Aravind Ramesh <aravind.ramesh@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Reading past end of file returns EOF for aligned reads but -EINVAL for
unaligned reads on f2fs. While documentation is not strict about this
corner case, most filesystem returns EOF on this case, like iomap
filesystems. This patch consolidates the behavior for f2fs, by making
it return EOF(0).
it can be verified by a read loop on a file that does a partial read
before EOF (A file that doesn't end at an aligned address). The
following code fails on an unaligned file on f2fs, but not on
btrfs, ext4, and xfs.
while (done < total) {
ssize_t delta = pread(fd, buf + done, total - done, off + done);
if (!delta)
break;
...
}
It is arguable whether filesystems should actually return EOF or
-EINVAL, but since iomap filesystems support it, and so does the
original DIO code, it seems reasonable to consolidate on that.
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If the sbi->ckpt->next_free_nid is not NAT block aligned and if there
are free nids in that NAT block between the start of the block and
next_free_nid, then those free nids will not be scanned in scan_nat_page().
This results into mismatch between nm_i->available_nids and the sum of
nm_i->free_nid_count of all NAT blocks scanned. And nm_i->available_nids
will always be greater than the sum of free nids in all the blocks.
Under this condition, if we use all the currently scanned free nids,
then it will loop forever in f2fs_alloc_nid() as nm_i->available_nids
is still not zero but nm_i->free_nid_count of that partially scanned
NAT block is zero.
Fix this to align the nm_i->next_scan_nid to the first nid of the
corresponding NAT block.
Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Commit da52f8ade40b ("f2fs: get the right gc victim section when section
has several segments") added code to count blocks of each section using
variables with type 'unsigned short', which has 2 bytes size in many
systems. However, the counts can be larger than the 2 bytes range and
type conversion results in wrong values. Especially when the f2fs
sections have blocks as many as USHRT_MAX + 1, the count is handled as 0.
This triggers eternal loop in init_dirty_segmap() at mount system call.
Fix this by changing the type of the variables to block_t.
Fixes: da52f8ade40b ("f2fs: get the right gc victim section when section has several segments")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
