For new uptodate buffers we also need to call write_end_fn() to persist the
uptodate content, similarly as folio_zero_new_buffers() does it.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Shida Zhang <zhangshida@kylinos.cn>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240830053739.3588573-2-zhangshida@kylinos.cn
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Any extending write for ext4 requires the inode to be placed on the
orphan list before the actual write. In addition, the inode can be
actually removed from the orphan list only after all writes are
completed. Otherwise we'd leave allocated blocks beyond i_disksize if we
could not copy all the data into allocated block and e2fsck would
complain.
Currently, direct IO and buffered IO comply with this logic(buffered
IO will truncate all overflow allocated blocks that has not been
written successfully, and direct IO will truncate all allocated blocks
when error occurs). However, dax write break this since dax write will
remove the inode from the orphan list by calling
ext4_handle_inode_extension unconditionally during extending write.
We add a argument to help determine does we do a fully write, and for
the case not fully write, we leave the inode on the orphan list, and the
latter ext4_inode_extension_cleanup will help us truncate the overflow
allocated blocks, and then remove the inode from the orphan list.
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240829110222.126685-1-yangerkun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Commit 985b67cd86 ("ext4: filesystems without casefold feature cannot
be mounted with siphash") properly rejects volumes where
s_def_hash_version is set to DX_HASH_SIPHASH, but the check and the
error message should not look into casefold setup - a filesystem should
never have DX_HASH_SIPHASH as the default hash. Fix it and, since we
are there, move the check to ext4_hash_info_init.
Fixes:985b67cd8639 ("ext4: filesystems without casefold feature cannot
be mounted with siphash")
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
Link: https://patch.msgid.link/87jzg1en6j.fsf_-_@mailhost.krisman.be
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Save an indentation level in ext4_ext_create_new_leaf() by removing
unnecessary 'else'. Besides, the variable 'ee_block' is declared to
avoid line breaks. No functional changes.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240822023545.1994557-26-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The ext4_find_extent() can update the extent path so that it does not have
to allocate and free the path repeatedly, thus reducing the consumption of
memory allocation and freeing in the following functions:
ext4_ext_clear_bb
ext4_ext_replay_set_iblocks
ext4_fc_replay_add_range
ext4_fc_set_bitmaps_and_counters
No functional changes. Note that ext4_find_extent() does not support error
pointers, so in this case set path to NULL first.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-25-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The ext4_find_extent() can update the extent path so it doesn't have to
allocate and free path repeatedly, thus reducing the consumption of memory
allocation and freeing in ext4_swap_extents().
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-24-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in convert_initialized_extent(), the following is
done here:
* Free the extents path when an error is encountered.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-23-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_ext_handle_unwritten_extents(), the
following is done here:
* Free the extents path when an error is encountered.
* The 'allocated' is changed from passing a value to passing an address.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-22-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_ext_convert_to_initialized(), the following
is done here:
* Free the extents path when an error is encountered.
* Its caller needs to update ppath if it uses ppath.
* The 'allocated' is changed from passing a value to passing an address.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-21-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_convert_unwritten_extents_endio(), the
following is done here:
* Free the extents path when an error is encountered.
* Its caller needs to update ppath if it uses ppath.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-20-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_split_convert_extents(), the following is
done here:
* Its caller needs to update ppath if it uses ppath.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-19-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_split_extent(), the following is done here:
* The 'allocated' is changed from passing a value to passing an address.
* Its caller needs to update ppath if it uses ppath.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-18-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_force_split_extent_at(), the following is
done here:
* Free the extents path when an error is encountered.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-17-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_split_extent_at(), the following is done
here:
* Free the extents path when an error is encountered.
* Its caller needs to update ppath if it uses ppath.
* Teach ext4_ext_show_leaf() to skip error pointer.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-16-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_ext_insert_extent(), the following is done
here:
* Free the extents path when an error is encountered.
* Its caller needs to update ppath if it uses ppath.
* Free path when npath is used, free npath when it is not used.
* The got_allocated_blocks label in ext4_ext_map_blocks() does not
update err now, so err is updated to 0 if the err returned by
ext4_ext_search_right() is greater than 0 and is about to enter
got_allocated_blocks.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-15-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
To get rid of the ppath in ext4_ext_create_new_leaf(), the following is
done here:
* Free the extents path when an error is encountered.
* Its caller needs to update ppath if it uses ppath.
No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-14-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
After getting rid of ppath in get_ext_path(), its caller may pass an error
pointer to ext4_free_ext_path(), so it needs to teach ext4_free_ext_path()
and ext4_ext_drop_refs() to skip the error pointer. No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-13-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The use of path and ppath is now very confusing, so to make the code more
readable, pass path between functions uniformly, and get rid of ppath.
Getting rid of ppath in ext4_find_extent() requires its caller to update
ppath. These ppaths will also be dropped later. No functional changes.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-12-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Even though ext4_find_extent() returns an error, ext4_insert_range() still
returns 0. This may confuse the user as to why fallocate returns success,
but the contents of the file are not as expected. So propagate the error
returned by ext4_find_extent() to avoid inconsistencies.
Fixes: 331573febb ("ext4: Add support FALLOC_FL_INSERT_RANGE for fallocate")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-11-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Add ext4_ext_path_brelse() helper function to reduce duplicate code
and ensure that path->p_bh is set to NULL after it is released.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-10-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_ext_try_to_merge_up(), set path[1].p_bh to NULL after it has been
released, otherwise it may be released twice. An example of what triggers
this is as follows:
split2 map split1
|--------|-------|--------|
ext4_ext_map_blocks
ext4_ext_handle_unwritten_extents
ext4_split_convert_extents
// path->p_depth == 0
ext4_split_extent
// 1. do split1
ext4_split_extent_at
|ext4_ext_insert_extent
| ext4_ext_create_new_leaf
| ext4_ext_grow_indepth
| le16_add_cpu(&neh->eh_depth, 1)
| ext4_find_extent
| // return -ENOMEM
|// get error and try zeroout
|path = ext4_find_extent
| path->p_depth = 1
|ext4_ext_try_to_merge
| ext4_ext_try_to_merge_up
| path->p_depth = 0
| brelse(path[1].p_bh) ---> not set to NULL here
|// zeroout success
// 2. update path
ext4_find_extent
// 3. do split2
ext4_split_extent_at
ext4_ext_insert_extent
ext4_ext_create_new_leaf
ext4_ext_grow_indepth
le16_add_cpu(&neh->eh_depth, 1)
ext4_find_extent
path[0].p_bh = NULL;
path->p_depth = 1
read_extent_tree_block ---> return err
// path[1].p_bh is still the old value
ext4_free_ext_path
ext4_ext_drop_refs
// path->p_depth == 1
brelse(path[1].p_bh) ---> brelse a buffer twice
Finally got the following WARRNING when removing the buffer from lru:
============================================
VFS: brelse: Trying to free free buffer
WARNING: CPU: 2 PID: 72 at fs/buffer.c:1241 __brelse+0x58/0x90
CPU: 2 PID: 72 Comm: kworker/u19:1 Not tainted 6.9.0-dirty #716
RIP: 0010:__brelse+0x58/0x90
Call Trace:
<TASK>
__find_get_block+0x6e7/0x810
bdev_getblk+0x2b/0x480
__ext4_get_inode_loc+0x48a/0x1240
ext4_get_inode_loc+0xb2/0x150
ext4_reserve_inode_write+0xb7/0x230
__ext4_mark_inode_dirty+0x144/0x6a0
ext4_ext_insert_extent+0x9c8/0x3230
ext4_ext_map_blocks+0xf45/0x2dc0
ext4_map_blocks+0x724/0x1700
ext4_do_writepages+0x12d6/0x2a70
[...]
============================================
Fixes: ecb94f5fdf ("ext4: collapse a single extent tree block into the inode if possible")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-9-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When calling ext4_force_split_extent_at() in ext4_ext_replay_update_ex(),
the 'ppath' is updated but it is the 'path' that is freed, thus potentially
triggering a double-free in the following process:
ext4_ext_replay_update_ex
ppath = path
ext4_force_split_extent_at(&ppath)
ext4_split_extent_at
ext4_ext_insert_extent
ext4_ext_create_new_leaf
ext4_ext_grow_indepth
ext4_find_extent
if (depth > path[0].p_maxdepth)
kfree(path) ---> path First freed
*orig_path = path = NULL ---> null ppath
kfree(path) ---> path double-free !!!
So drop the unnecessary ppath and use path directly to avoid this problem.
And use ext4_find_extent() directly to update path, avoiding unnecessary
memory allocation and freeing. Also, propagate the error returned by
ext4_find_extent() instead of using strange error codes.
Fixes: 8016e29f43 ("ext4: fast commit recovery path")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-8-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
As Ojaswin mentioned in Link, in ext4_ext_insert_extent(), if the path is
reallocated in ext4_ext_create_new_leaf(), we'll use the stale path and
cause UAF. Below is a sample trace with dummy values:
ext4_ext_insert_extent
path = *ppath = 2000
ext4_ext_create_new_leaf(ppath)
ext4_find_extent(ppath)
path = *ppath = 2000
if (depth > path[0].p_maxdepth)
kfree(path = 2000);
*ppath = path = NULL;
path = kcalloc() = 3000
*ppath = 3000;
return path;
/* here path is still 2000, UAF! */
eh = path[depth].p_hdr
==================================================================
BUG: KASAN: slab-use-after-free in ext4_ext_insert_extent+0x26d4/0x3330
Read of size 8 at addr ffff8881027bf7d0 by task kworker/u36:1/179
CPU: 3 UID: 0 PID: 179 Comm: kworker/u6:1 Not tainted 6.11.0-rc2-dirty #866
Call Trace:
<TASK>
ext4_ext_insert_extent+0x26d4/0x3330
ext4_ext_map_blocks+0xe22/0x2d40
ext4_map_blocks+0x71e/0x1700
ext4_do_writepages+0x1290/0x2800
[...]
Allocated by task 179:
ext4_find_extent+0x81c/0x1f70
ext4_ext_map_blocks+0x146/0x2d40
ext4_map_blocks+0x71e/0x1700
ext4_do_writepages+0x1290/0x2800
ext4_writepages+0x26d/0x4e0
do_writepages+0x175/0x700
[...]
Freed by task 179:
kfree+0xcb/0x240
ext4_find_extent+0x7c0/0x1f70
ext4_ext_insert_extent+0xa26/0x3330
ext4_ext_map_blocks+0xe22/0x2d40
ext4_map_blocks+0x71e/0x1700
ext4_do_writepages+0x1290/0x2800
ext4_writepages+0x26d/0x4e0
do_writepages+0x175/0x700
[...]
==================================================================
So use *ppath to update the path to avoid the above problem.
Reported-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Closes: https://lore.kernel.org/r/ZqyL6rmtwl6N4MWR@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.com
Fixes: 10809df84a ("ext4: teach ext4_ext_find_extent() to realloc path if necessary")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240822023545.1994557-7-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_find_extent(), if the path is not big enough, we free it and set
*orig_path to NULL. But after reallocating and successfully initializing
the path, we don't update *orig_path, in which case the caller gets a
valid path but a NULL ppath, and this may cause a NULL pointer dereference
or a path memory leak. For example:
ext4_split_extent
path = *ppath = 2000
ext4_find_extent
if (depth > path[0].p_maxdepth)
kfree(path = 2000);
*orig_path = path = NULL;
path = kcalloc() = 3000
ext4_split_extent_at(*ppath = NULL)
path = *ppath;
ex = path[depth].p_ext;
// NULL pointer dereference!
==================================================================
BUG: kernel NULL pointer dereference, address: 0000000000000010
CPU: 6 UID: 0 PID: 576 Comm: fsstress Not tainted 6.11.0-rc2-dirty #847
RIP: 0010:ext4_split_extent_at+0x6d/0x560
Call Trace:
<TASK>
ext4_split_extent.isra.0+0xcb/0x1b0
ext4_ext_convert_to_initialized+0x168/0x6c0
ext4_ext_handle_unwritten_extents+0x325/0x4d0
ext4_ext_map_blocks+0x520/0xdb0
ext4_map_blocks+0x2b0/0x690
ext4_iomap_begin+0x20e/0x2c0
[...]
==================================================================
Therefore, *orig_path is updated when the extent lookup succeeds, so that
the caller can safely use path or *ppath.
Fixes: 10809df84a ("ext4: teach ext4_ext_find_extent() to realloc path if necessary")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240822023545.1994557-6-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_find_extent(), path may be freed by error or be reallocated, so
using a previously saved *ppath may have been freed and thus may trigger
use-after-free, as follows:
ext4_split_extent
path = *ppath;
ext4_split_extent_at(ppath)
path = ext4_find_extent(ppath)
ext4_split_extent_at(ppath)
// ext4_find_extent fails to free path
// but zeroout succeeds
ext4_ext_show_leaf(inode, path)
eh = path[depth].p_hdr
// path use-after-free !!!
Similar to ext4_split_extent_at(), we use *ppath directly as an input to
ext4_ext_show_leaf(). Fix a spelling error by the way.
Same problem in ext4_ext_handle_unwritten_extents(). Since 'path' is only
used in ext4_ext_show_leaf(), remove 'path' and use *ppath directly.
This issue is triggered only when EXT_DEBUG is defined and therefore does
not affect functionality.
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-5-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
We hit the following use-after-free:
==================================================================
BUG: KASAN: slab-use-after-free in ext4_split_extent_at+0xba8/0xcc0
Read of size 2 at addr ffff88810548ed08 by task kworker/u20:0/40
CPU: 0 PID: 40 Comm: kworker/u20:0 Not tainted 6.9.0-dirty #724
Call Trace:
<TASK>
kasan_report+0x93/0xc0
ext4_split_extent_at+0xba8/0xcc0
ext4_split_extent.isra.0+0x18f/0x500
ext4_split_convert_extents+0x275/0x750
ext4_ext_handle_unwritten_extents+0x73e/0x1580
ext4_ext_map_blocks+0xe20/0x2dc0
ext4_map_blocks+0x724/0x1700
ext4_do_writepages+0x12d6/0x2a70
[...]
Allocated by task 40:
__kmalloc_noprof+0x1ac/0x480
ext4_find_extent+0xf3b/0x1e70
ext4_ext_map_blocks+0x188/0x2dc0
ext4_map_blocks+0x724/0x1700
ext4_do_writepages+0x12d6/0x2a70
[...]
Freed by task 40:
kfree+0xf1/0x2b0
ext4_find_extent+0xa71/0x1e70
ext4_ext_insert_extent+0xa22/0x3260
ext4_split_extent_at+0x3ef/0xcc0
ext4_split_extent.isra.0+0x18f/0x500
ext4_split_convert_extents+0x275/0x750
ext4_ext_handle_unwritten_extents+0x73e/0x1580
ext4_ext_map_blocks+0xe20/0x2dc0
ext4_map_blocks+0x724/0x1700
ext4_do_writepages+0x12d6/0x2a70
[...]
==================================================================
The flow of issue triggering is as follows:
ext4_split_extent_at
path = *ppath
ext4_ext_insert_extent(ppath)
ext4_ext_create_new_leaf(ppath)
ext4_find_extent(orig_path)
path = *orig_path
read_extent_tree_block
// return -ENOMEM or -EIO
ext4_free_ext_path(path)
kfree(path)
*orig_path = NULL
a. If err is -ENOMEM:
ext4_ext_dirty(path + path->p_depth)
// path use-after-free !!!
b. If err is -EIO and we have EXT_DEBUG defined:
ext4_ext_show_leaf(path)
eh = path[depth].p_hdr
// path also use-after-free !!!
So when trying to zeroout or fix the extent length, call ext4_find_extent()
to update the path.
In addition we use *ppath directly as an ext4_ext_show_leaf() input to
avoid possible use-after-free when EXT_DEBUG is defined, and to avoid
unnecessary path updates.
Fixes: dfe5080939 ("ext4: drop EXT4_EX_NOFREE_ON_ERR from rest of extents handling code")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-4-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_ext_rm_idx() and ext4_ext_correct_indexes(), there is no proper
rollback of already executed updates when updating a level of the extents
path fails, so we may get an inconsistent extents tree, which may trigger
some bad things in errors=continue mode.
Hence clear the verified bit of modified extents buffers if the tree fails
to be updated in ext4_ext_rm_idx() or ext4_ext_correct_indexes(), which
forces the extents buffers to be checked in ext4_valid_extent_entries(),
ensuring that the extents tree is consistent.
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Link: https://lore.kernel.org/r/20230213080514.535568-3-zhanchengbin1@huawei.com/
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-3-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
As suggested by Honza in Link，modify ext4_ext_rm_idx() to leave 'path'
alone and just index it like ext4_ext_correct_indexes() does it. This
facilitates adding error handling later. No functional changes.
Suggested-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/all/20230216130305.nrbtd42tppxhbynn@quack3/
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Tested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
Link: https://patch.msgid.link/20240822023545.1994557-2-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
__ext4_find_entry currently ignores the return of ext4_find_inline_entry,
except for returning the bh or NULL when has_inline_data is 1.
Even though has_inline_data is set to 1 before calling
ext4_find_inline_entry and would only be set to 0 when that function
returns NULL, check for an encoded error return explicitly in order to
exit.
That makes the code more readable, not requiring that one assumes the cases
when has_inline_data is 1.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@igalia.com>
Link: https://patch.msgid.link/20240821152324.3621860-4-cascardo@igalia.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In case of errors when reading an inode from disk or traversing inline
directory entries, return an error-encoded ERR_PTR instead of returning
NULL. ext4_find_inline_entry only caller, __ext4_find_entry already returns
such encoded errors.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@igalia.com>
Link: https://patch.msgid.link/20240821152324.3621860-3-cascardo@igalia.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_search_dir currently returns -1 in case of a failure, while it returns
0 when the name is not found. In such failure cases, it should return an
error code instead.
This becomes even more important when ext4_find_inline_entry returns an
error code as well in the next commit.
-EFSCORRUPTED seems appropriate as such error code as these failures would
be caused by unexpected record lengths and is in line with other instances
of ext4_check_dir_entry failures.
In the case of ext4_dx_find_entry, the current use of ERR_BAD_DX_DIR was
left as is to reduce the risk of regressions.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@igalia.com>
Link: https://patch.msgid.link/20240821152324.3621860-2-cascardo@igalia.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Check buffer_verified in advance to avoid unneeded ext4_get_group_info().
This could be a simple cleanup as compiler may handle this.
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-8-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If gdp from ext4_get_group_desc() is not NULL, then returned group_desc_bh
won't be NULL either. Remove check of group_desc_bh and only check
returned gdp from ext4_get_group_desc() like how other callers do.
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-7-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
There are some little improve:
1. remove repeat code to calculate checksum length of inode bitmap
2. remove unnecessary checksum length calculation if checksum is not
enabled.
3. use more efficient bit shift operation instead of div opreation.
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-6-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If we can't grab any inode, the prvious find_inode_bit() will set ino
to be >= EXT4_INODES_PER_GROUP(sb). So the check of need to repeat
in the same group is not needed.
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-5-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
min_clusters is signed integer and will be converted to unsigned
integer when compared with unsigned number stats.free_clusters.
If min_clusters is negative, it will be converted to a huge unsigned
value in which case all groups may not meet the actual desired free
clusters.
Set negative min_clusters to 0 to avoid unexpected behavior.
Fixes: ac27a0ec11 ("[PATCH] ext4: initial copy of files from ext3")
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-4-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If a group is marked EXT4_GROUP_INFO_IBITMAP_CORRUPT after it's inode
bitmap buffer_head was successfully verified, then __ext4_new_inode()
will get a valid inode_bitmap_bh of a corrupted group from
ext4_read_inode_bitmap() in which case inode_bitmap_bh misses a release.
Hnadle "IS_ERR(inode_bitmap_bh)" and group corruption separately like
how ext4_free_inode() does to avoid buffer_head leak.
Fixes: 9008a58e5d ("ext4: make the bitmap read routines return real error codes")
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-3-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Release inode_bitmap_bh from ext4_read_inode_bitmap() in
ext4_mark_inode_used() to avoid buffer_head leak.
By the way, remove unneeded goto for invalid ino when inode_bitmap_bh
is NULL.
Fixes: 8016e29f43 ("ext4: fast commit recovery path")
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240820132234.2759926-2-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Commit 3d56b8d2c7 ("ext4: Speed up FITRIM by recording flags in
ext4_group_info") speed up fstrim by skipping trim trimmed group. We
also has the chance to clear trimmed once there exists some block free
for this group(mount without discard), and the next trim for this group
will work well too.
For mount with discard, we will issue dicard when we free blocks, so
leave trimmed flag keep alive to skip useless trim trigger from
userspace seems reasonable. But for some case like ext4 build on
dm-thinpool(ext4 blocksize 4K, pool blocksize 128K), discard from ext4
maybe unaligned for dm thinpool, and thinpool will just finish this
discard(see process_discard_bio when begein equals to end) without
actually process discard. For this case, trim from userspace can really
help us to free some thinpool block.
So convert to clear trimmed flag for all case no matter mounted with
discard or not.
Fixes: 3d56b8d2c7 ("ext4: Speed up FITRIM by recording flags in ext4_group_info")
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240817085510.2084444-1-yangerkun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When counting reserved clusters, delayed type is always equal to delonly
type now, hence drop all delonly descriptions in parameters and
comments.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-13-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since we don't add delayed flag in unwritten extents, so there is no
difference between ext4_es_is_delayed() and ext4_es_is_delonly(),
just drop ext4_es_is_delonly().
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-12-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since we don't add delayed flag in unwritten extents, all of the four
extent status types EXTENT_STATUS_WRITTEN, EXTENT_STATUS_UNWRITTEN,
EXTENT_STATUS_DELAYED and EXTENT_STATUS_HOLE are exclusive now, add
assertion when storing pblock before inserting extent into status tree
and add comment to the status definition.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-11-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The blocks map querying logic in ext4_map_blocks() are the same as
ext4_map_query_blocks(), so switch to directly use it.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240813123452.2824659-9-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since we move ext4_da_update_reserve_space() to ext4_es_insert_extent(),
no one uses ext4_es_delayed_clu() and __es_delayed_clu(), just drop
them.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Now that we update data reserved space for delalloc after allocating
new blocks in ext4_{ind|ext}_map_blocks(), and if bigalloc feature is
enabled, we also need to query the extents_status tree to calculate the
exact reserved clusters. This is complicated now and it appears that
it's better to do this job in ext4_es_insert_extent(), because
__es_remove_extent() have already count delalloc blocks when removing
delalloc extents and __revise_pending() return new adding pending count,
we could update the reserved blocks easily in ext4_es_insert_extent().
We direct reduce the reserved cluster count when replacing a delalloc
extent. However, thers are two special cases need to concern about the
quota claiming when doing direct block allocation (e.g. from fallocate).
A),
fallocate a range that covers a delalloc extent but start with
non-delayed allocated blocks, e.g. a hole.
hhhhhhh+ddddddd+ddddddd
^^^^^^^^^^^^^^^^^^^^^^^ fallocate this range
Current ext4_map_blocks() can't always trim the extent since it may
release i_data_sem before calling ext4_map_create_blocks() and raced by
another delayed allocation. Hence the EXT4_GET_BLOCKS_DELALLOC_RESERVE
may not set even when we are replacing a delalloc extent, without this
flag set, the quota has already been claimed by ext4_mb_new_blocks(), so
we should release the quota reservations instead of claim them again.
B),
bigalloc feature is enabled, fallocate a range that contains non-delayed
allocated blocks.
|< one cluster >|
hhhhhhh+hhhhhhh+hhhhhhh+ddddddd
^^^^^^^ fallocate this range
This case is similar to above case, the EXT4_GET_BLOCKS_DELALLOC_RESERVE
flag is also not set.
Hence we should release the quota reservations if we replace a delalloc
extent but without EXT4_GET_BLOCKS_DELALLOC_RESERVE set.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-7-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Just pass the block allocation flag to ext4_es_insert_extent() when we
replacing a current extent after an actually block allocation or extent
status conversion, this flag will be used by later changes.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Let __insert_pending() return 1 after successfully inserting a new
pending cluster, and also let __revise_pending() to return the number of
of newly inserted pendings.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20240813123452.2824659-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, we release delayed allocation reservation when removing
delayed extent from extent status tree (which also happens when
overwriting one extent with another one). When we allocated unwritten
extent under some delayed allocated extent, we don't need the
reservation anymore and hence we don't need to preserve the
EXT4_MAP_DELAYED status bit. Allocating the new extent blocks will
properly release the reservation.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240813123452.2824659-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When doing block allocation, magic EXT4_GET_BLOCKS_DELALLOC_RESERVE
means the allocating range covers a range of delayed allocated clusters,
the blocks and quotas have already been reserved in ext4_da_map_blocks(),
we should update the reserved space and don't need to claim them again.
At the moment, we only set this magic in mpage_map_one_extent() when
allocating a range of delayed allocated clusters in the write back path,
it makes things complicated since we have to notice and deal with the
case of allocating non-delayed allocated clusters separately in
ext4_ext_map_blocks(). For example, it we fallocate some blocks that
have been delayed allocated, free space would be claimed again in
ext4_mb_new_blocks() (this is wrong exactily), and we can't claim quota
space again, we have to release the quota reservations made for that
previously delayed allocated clusters.
Move the position thats set the EXT4_GET_BLOCKS_DELALLOC_RESERVE to
where we actually do block allocation, it could simplify above handling
a lot, it means that we always set this magic once the allocation range
covers delalloc blocks, no need to take care of the allocation path.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240813123452.2824659-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The dax_iomap_rw() does two things in each iteration: map written blocks
and copy user data to blocks. If the process is killed by user(See signal
handling in dax_iomap_iter()), the copied data will be returned and added
on inode size, which means that the length of written extents may exceed
the inode size, then fsck will fail. An example is given as:
dd if=/dev/urandom of=file bs=4M count=1
dax_iomap_rw
iomap_iter // round 1
ext4_iomap_begin
ext4_iomap_alloc // allocate 0~2M extents(written flag)
dax_iomap_iter // copy 2M data
iomap_iter // round 2
iomap_iter_advance
iter->pos += iter->processed // iter->pos = 2M
ext4_iomap_begin
ext4_iomap_alloc // allocate 2~4M extents(written flag)
dax_iomap_iter
fatal_signal_pending
done = iter->pos - iocb->ki_pos // done = 2M
ext4_handle_inode_extension
ext4_update_inode_size // inode size = 2M
fsck reports: Inode 13, i_size is 2097152, should be 4194304. Fix?
Fix the problem by truncating extents if the written length is smaller
than expected.
Fixes: 776722e85d ("ext4: DAX iomap write support")
CC: stable@vger.kernel.org
Link: https://bugzilla.kernel.org/show_bug.cgi?id=219136
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>
Link: https://patch.msgid.link/20240809121532.2105494-1-chengzhihao@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When the filesystem is mounted with errors=remount-ro, we were setting
SB_RDONLY flag to stop all filesystem modifications. We knew this misses
proper locking (sb->s_umount) and does not go through proper filesystem
remount procedure but it has been the way this worked since early ext2
days and it was good enough for catastrophic situation damage
mitigation. Recently, syzbot has found a way (see link) to trigger
warnings in filesystem freezing because the code got confused by
SB_RDONLY changing under its hands. Since these days we set
EXT4_FLAGS_SHUTDOWN on the superblock which is enough to stop all
filesystem modifications, modifying SB_RDONLY shouldn't be needed. So
stop doing that.
Link: https://lore.kernel.org/all/000000000000b90a8e061e21d12f@google.com
Reported-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Link: https://patch.msgid.link/20240805201241.27286-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Add the __counted_by compiler attribute to the flexible array member
inodes to improve access bounds-checking via CONFIG_UBSAN_BOUNDS and
CONFIG_FORTIFY_SOURCE.
Remove the now obsolete comment on the count field.
In ext4_expand_inode_array(), use struct_size() instead of offsetof()
and remove the local variable count. Increment the count field before
adding a new inode to the inodes array.
Compile-tested only.
Signed-off-by: Thorsten Blum <thorsten.blum@toblux.com>
Link: https://patch.msgid.link/20240730220200.410939-3-thorsten.blum@toblux.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Function jbd2_journal_shrink_checkpoint_list() assumes that '0' is not a
valid value for transaction IDs, which is incorrect.
Furthermore, the sbi->s_fc_ineligible_tid handling also makes the same
assumption by being initialised to '0'. Fortunately, the sb flag
EXT4_MF_FC_INELIGIBLE can be used to check whether sbi->s_fc_ineligible_tid
has been previously set instead of comparing it with '0'.
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240724161119.13448-5-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Function ext4_wait_for_tail_page_commit() assumes that '0' is not a valid
value for transaction IDs, which is incorrect. Don't assume that and invoke
jbd2_log_wait_commit() if the journal had a committing transaction instead.
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240724161119.13448-2-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Iterate the folio's list of buffer_heads twice instead of keeping
an array of pointers. This solves a too-large-array-for-stack problem
on architectures with a ridiculoously large PAGE_SIZE and prepares
ext4 to support larger folios.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: https://patch.msgid.link/20240718223005.568869-3-willy@infradead.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Instead of synchronously reading one buffer at a time, submit reads
as we walk the buffers in the first loop, then wait for them in the
second loop. This should be significantly more efficient, particularly
on HDDs, but I have not measured.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: https://patch.msgid.link/20240718223005.568869-2-willy@infradead.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This function is very similar to do_mpage_readpage() and a similar
approach to that taken in commit 12ac5a65cb will work. As in
do_mpage_readpage(), we only use this array for checking block contiguity
and we can do that more efficiently with a little arithmetic.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: https://patch.msgid.link/20240718223005.568869-1-willy@infradead.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The following kernel trace can be triggered with fstest generic/629 when
executed against a filesystem with fast-commit feature enabled:
INFO: trying to register non-static key.
The code is fine but needs lockdep annotation, or maybe
you didn't initialize this object before use?
turning off the locking correctness validator.
CPU: 0 PID: 866 Comm: mount Not tainted 6.10.0+ #11
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.2-3-gd478f380-prebuilt.qemu.org 04/01/2014
Call Trace:
<TASK>
dump_stack_lvl+0x66/0x90
register_lock_class+0x759/0x7d0
__lock_acquire+0x85/0x2630
? __find_get_block+0xb4/0x380
lock_acquire+0xd1/0x2d0
? __ext4_journal_get_write_access+0xd5/0x160
_raw_spin_lock+0x33/0x40
? __ext4_journal_get_write_access+0xd5/0x160
__ext4_journal_get_write_access+0xd5/0x160
ext4_reserve_inode_write+0x61/0xb0
__ext4_mark_inode_dirty+0x79/0x270
? ext4_ext_replay_set_iblocks+0x2f8/0x450
ext4_ext_replay_set_iblocks+0x330/0x450
ext4_fc_replay+0x14c8/0x1540
? jread+0x88/0x2e0
? rcu_is_watching+0x11/0x40
do_one_pass+0x447/0xd00
jbd2_journal_recover+0x139/0x1b0
jbd2_journal_load+0x96/0x390
ext4_load_and_init_journal+0x253/0xd40
ext4_fill_super+0x2cc6/0x3180
...
In the replay path there's an attempt to lock sbi->s_bdev_wb_lock in
function ext4_check_bdev_write_error(). Unfortunately, at this point this
spinlock has not been initialized yet. Moving it's initialization to an
earlier point in __ext4_fill_super() fixes this splat.
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Link: https://patch.msgid.link/20240718094356.7863-1-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
When a full journal commit is on-going, any fast commit has to be enqueued
into a different queue: FC_Q_STAGING instead of FC_Q_MAIN. This enqueueing
is done only once, i.e. if an inode is already queued in a previous fast
commit entry it won't be enqueued again. However, if a full commit starts
_after_ the inode is enqueued into FC_Q_MAIN, the next fast commit needs to
be done into FC_Q_STAGING. And this is not being done in function
ext4_fc_track_template().
This patch fixes the issue by re-enqueuing an inode into the STAGING queue
during the fast commit clean-up callback when doing a full commit. However,
to prevent a race with a fast-commit, the clean-up callback has to be called
with the journal locked.
This bug was found using fstest generic/047. This test creates several 32k
bytes files, sync'ing each of them after it's creation, and then shutting
down the filesystem. Some data may be loss in this operation; for example a
file may have it's size truncated to zero.
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240717172220.14201-1-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Syzbot has found an ODEBUG bug in ext4_fill_super
The del_timer_sync function cancels the s_err_report timer,
which reminds about filesystem errors daily. We should
guarantee the timer is no longer active before kfree(sbi).
When filesystem mounting fails, the flow goes to failed_mount3,
where an error occurs when ext4_stop_mmpd is called, causing
a read I/O failure. This triggers the ext4_handle_error function
that ultimately re-arms the timer,
leaving the s_err_report timer active before kfree(sbi) is called.
Fix the issue by canceling the s_err_report timer after calling ext4_stop_mmpd.
Signed-off-by: Xiaxi Shen <shenxiaxi26@gmail.com>
Reported-and-tested-by: syzbot+59e0101c430934bc9a36@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=59e0101c430934bc9a36
Link: https://patch.msgid.link/20240715043336.98097-1-shenxiaxi26@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Single characters (line breaks) should be put into a sequence.
Thus use the corresponding function “seq_putc”.
This issue was transformed by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Link: https://patch.msgid.link/076974ab-4da3-4176-89dc-0514e020c276@web.de
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Checksum of xattr block is always crc32c(uuid+blknum+xattrblock), see
ext4_xattr_block_csum_set for detail. Remove incorrect comment that
"id = inum if refcount=1".
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://patch.msgid.link/20240606125508.1459893-4-shikemeng@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The macro parameter 'entry' of EXT4_DIRENT_HASH and
EXT4_DIRENT_MINOR_HASH was not used, but rather the variable 'de' was
directly used, which may be a local variable inside a function that
calls the macros. Fortunately, all callers have passed in 'de' so
far, so this bug didn't have an effect.
Signed-off-by: carrion bent <carrionbent@linux.alibaba.com>
Link: https://patch.msgid.link/1717652596-58760-1-git-send-email-carrionbent@linux.alibaba.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Using pahole, we can see that there are some padding holes
in the current ext4_inode_info structure. Adjusting the
layout of ext4_inode_info can reduce these holes,
resulting in the size of the structure decreasing
from 2424 bytes to 2408 bytes.
Signed-off-by: Junchao Sun <sunjunchao2870@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240603131524.324224-1-sunjunchao2870@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For any changes of struct fd representation we need to
turn existing accesses to fields into calls of wrappers.
Accesses to struct fd::flags are very few (3 in linux/file.h,
1 in net/socket.c, 3 in fs/overlayfs/file.c and 3 more in
explicit initializers).
Those can be dealt with in the commit converting to
new layout; accesses to struct fd::file are too many for that.
This commit converts (almost) all of f.file to
fd_file(f). It's not entirely mechanical ('file' is used as
a member name more than just in struct fd) and it does not
even attempt to distinguish the uses in pointer context from
those in boolean context; the latter will be eventually turned
into a separate helper (fd_empty()).
NOTE: mass conversion to fd_empty(), tempting as it
might be, is a bad idea; better do that piecewise in commit
that convert from fdget...() to CLASS(...).
[conflicts in fs/fhandle.c, kernel/bpf/syscall.c, mm/memcontrol.c
caught by git; fs/stat.c one got caught by git grep]
[fs/xattr.c conflict]
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Almost all callers have a folio now, so change __block_write_begin()
to take a folio and remove a call to compound_head().
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Convert all callers from working on a page to working on one page
of a folio (support for working on an entire folio can come later).
Removes a lot of folio->page->folio conversions.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Most callers have a folio, and most implementations operate on a folio,
so remove the conversion from folio->page->folio to fit through this
interface.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
All callers now have a folio, so pass it in instead of converting
from a folio to a page and back to a folio again. Saves a call
to compound_head().
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
feature. Also some performance improvements; in particular, improving
IOPS and throughput on fast devices running Async Direct I/O by up to
20% by optimizing jbd2_transaction_committed().
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEK2m5VNv+CHkogTfJ8vlZVpUNgaMFAmaYiqsACgkQ8vlZVpUN
gaOWpQf/d6Y9WGyjeC1jOc+vIBxLgL+X0kbzYkkjGTSIZ7mZJS9X4NMMEtqayJ4f
1zGobcGENc05l4LVxf3uMbDj1aGlHeI9X4GLGaP5s5NcaAl4HKjQ3aFs3MuiJHPj
Ol2CebXJx+NKt1lkD8PSPGgaTb5zg+SeZifI+OZ1RpkcKmGnkSNa5NkUNAaBh6dl
5LLXTc2p9NcCwAwDAQSiAJCV35bAZpcp6fwLLaPQ6Eok9HxGcJuYXW2Fict4rbtV
mXeogXVIo2bkMcfh6tDchDBrFvORYIA7uBVmaG1LgAMrtEnYxnxnEntD0h6j/bzF
Fl4jjQfd8o2uYto/4eo+iY6Z0haxyQ==
=rcOo
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus-6.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 updates from Ted Ts'o:
"Many cleanups and bug fixes in ext4, especially for the fast commit
feature.
Also some performance improvements; in particular, improving IOPS and
throughput on fast devices running Async Direct I/O by up to 20% by
optimizing jbd2_transaction_committed()"
* tag 'ext4_for_linus-6.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (40 commits)
ext4: make sure the first directory block is not a hole
ext4: check dot and dotdot of dx_root before making dir indexed
ext4: sanity check for NULL pointer after ext4_force_shutdown
jbd2: increase maximum transaction size
jbd2: drop pointless shrinker batch initialization
jbd2: avoid infinite transaction commit loop
jbd2: precompute number of transaction descriptor blocks
jbd2: make jbd2_journal_get_max_txn_bufs() internal
jbd2: avoid mount failed when commit block is partial submitted
ext4: avoid writing unitialized memory to disk in EA inodes
ext4: don't track ranges in fast_commit if inode has inlined data
ext4: fix possible tid_t sequence overflows
ext4: use ext4_update_inode_fsync_trans() helper in inode creation
ext4: add missing MODULE_DESCRIPTION()
jbd2: add missing MODULE_DESCRIPTION()
ext4: use memtostr_pad() for s_volume_name
jbd2: speed up jbd2_transaction_committed()
ext4: make ext4_da_map_blocks() buffer_head unaware
ext4: make ext4_insert_delayed_block() insert multi-blocks
ext4: factor out a helper to check the cluster allocation state
...
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCZpEGjAAKCRCRxhvAZXjc
okXfAP4tFUYszUsSqYdsgy9UvXw3Dr5zOIzQmN++NdjGkbU5fgEAs2ystqEfJgr3
v7XvGbu65CvL4/slNhBZOU4yekGx5Qc=
=C4QD
-----END PGP SIGNATURE-----
Merge tag 'vfs-6.11.mount.api' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Pull vfs mount API updates from Christian Brauner:
- Add a generic helper to parse uid and gid mount options.
Currently we open-code the same logic in various filesystems which is
error prone, especially since the verification of uid and gid mount
options is a sensitive operation in the face of idmappings.
Add a generic helper and convert all filesystems over to it. Make
sure that filesystems that are mountable in unprivileged containers
verify that the specified uid and gid can be represented in the
owning namespace of the filesystem.
- Convert hostfs to the new mount api.
* tag 'vfs-6.11.mount.api' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
fuse: Convert to new uid/gid option parsing helpers
fuse: verify {g,u}id mount options correctly
fat: Convert to new uid/gid option parsing helpers
fat: Convert to new mount api
fat: move debug into fat_mount_options
vboxsf: Convert to new uid/gid option parsing helpers
tracefs: Convert to new uid/gid option parsing helpers
smb: client: Convert to new uid/gid option parsing helpers
tmpfs: Convert to new uid/gid option parsing helpers
ntfs3: Convert to new uid/gid option parsing helpers
isofs: Convert to new uid/gid option parsing helpers
hugetlbfs: Convert to new uid/gid option parsing helpers
ext4: Convert to new uid/gid option parsing helpers
exfat: Convert to new uid/gid option parsing helpers
efivarfs: Convert to new uid/gid option parsing helpers
debugfs: Convert to new uid/gid option parsing helpers
autofs: Convert to new uid/gid option parsing helpers
fs_parse: add uid & gid option option parsing helpers
hostfs: Add const qualifier to host_root in hostfs_fill_super()
hostfs: convert hostfs to use the new mount API
The syzbot constructs a directory that has no dirblock but is non-inline,
i.e. the first directory block is a hole. And no errors are reported when
creating files in this directory in the following flow.
ext4_mknod
...
ext4_add_entry
// Read block 0
ext4_read_dirblock(dir, block, DIRENT)
bh = ext4_bread(NULL, inode, block, 0)
if (!bh && (type == INDEX || type == DIRENT_HTREE))
// The first directory block is a hole
// But type == DIRENT, so no error is reported.
After that, we get a directory block without '.' and '..' but with a valid
dentry. This may cause some code that relies on dot or dotdot (such as
make_indexed_dir()) to crash.
Therefore when ext4_read_dirblock() finds that the first directory block
is a hole report that the filesystem is corrupted and return an error to
avoid loading corrupted data from disk causing something bad.
Reported-by: syzbot+ae688d469e36fb5138d0@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=ae688d469e36fb5138d0
Fixes: 4e19d6b65f ("ext4: allow directory holes")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240702132349.2600605-3-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Syzbot reports a issue as follows:
============================================
BUG: unable to handle page fault for address: ffffed11022e24fe
PGD 23ffee067 P4D 23ffee067 PUD 0
Oops: Oops: 0000 [#1] PREEMPT SMP KASAN PTI
CPU: 0 PID: 5079 Comm: syz-executor306 Not tainted 6.10.0-rc5-g55027e689933 #0
Call Trace:
<TASK>
make_indexed_dir+0xdaf/0x13c0 fs/ext4/namei.c:2341
ext4_add_entry+0x222a/0x25d0 fs/ext4/namei.c:2451
ext4_rename fs/ext4/namei.c:3936 [inline]
ext4_rename2+0x26e5/0x4370 fs/ext4/namei.c:4214
[...]
============================================
The immediate cause of this problem is that there is only one valid dentry
for the block to be split during do_split, so split==0 results in out of
bounds accesses to the map triggering the issue.
do_split
unsigned split
dx_make_map
count = 1
split = count/2 = 0;
continued = hash2 == map[split - 1].hash;
---> map[4294967295]
The maximum length of a filename is 255 and the minimum block size is 1024,
so it is always guaranteed that the number of entries is greater than or
equal to 2 when do_split() is called.
But syzbot's crafted image has no dot and dotdot in dir, and the dentry
distribution in dirblock is as follows:
bus dentry1 hole dentry2 free
|xx--|xx-------------|...............|xx-------------|...............|
0 12 (8+248)=256 268 256 524 (8+256)=264 788 236 1024
So when renaming dentry1 increases its name_len length by 1, neither hole
nor free is sufficient to hold the new dentry, and make_indexed_dir() is
called.
In make_indexed_dir() it is assumed that the first two entries of the
dirblock must be dot and dotdot, so bus and dentry1 are left in dx_root
because they are treated as dot and dotdot, and only dentry2 is moved
to the new leaf block. That's why count is equal to 1.
Therefore add the ext4_check_dx_root() helper function to add more sanity
checks to dot and dotdot before starting the conversion to avoid the above
issue.
Reported-by: syzbot+ae688d469e36fb5138d0@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=ae688d469e36fb5138d0
Fixes: ac27a0ec11 ("[PATCH] ext4: initial copy of files from ext3")
Cc: stable@kernel.org
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240702132349.2600605-2-libaokun@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If the extended attribute size is not a multiple of block size, the last
block in the EA inode will have uninitialized tail which will get
written to disk. We will never expose the data to userspace but still
this is not a good practice so just zero out the tail of the block as it
isn't going to cause a noticeable performance overhead.
Fixes: e50e5129f3 ("ext4: xattr-in-inode support")
Reported-by: syzbot+9c1fe13fcb51574b249b@syzkaller.appspotmail.com
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240613150234.25176-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When fast-commit needs to track ranges, it has to handle inodes that have
inlined data in a different way because ext4_fc_write_inode_data(), in the
actual commit path, will attempt to map the required blocks for the range.
However, inodes that have inlined data will have it's data stored in
inode->i_block and, eventually, in the extended attribute space.
Unfortunately, because fast commit doesn't currently support extended
attributes, the solution is to mark this commit as ineligible.
Link: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1039883
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Tested-by: Ben Hutchings <benh@debian.org>
Fixes: 9725958bb7 ("ext4: fast commit may miss tracking unwritten range during ftruncate")
Link: https://patch.msgid.link/20240618144312.17786-1-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In the fast commit code there are a few places where tid_t variables are
being compared without taking into account the fact that these sequence
numbers may wrap. Fix this issue by using the helper functions tid_gt()
and tid_geq().
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://patch.msgid.link/20240529092030.9557-3-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Call helper function ext4_update_inode_fsync_trans() instead of open
coding it in __ext4_new_inode(). This helper checks both that the handle
is valid *and* that it hasn't been aborted due to some fatal error in the
journalling layer, using is_handle_aborted().
Signed-off-by: Luis Henriques (SUSE) <luis.henriques@linux.dev>
Link: https://patch.msgid.link/20240527161447.21434-1-luis.henriques@linux.dev
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Fix the 'make W=1' warning:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/ext4/ext4-inode-test.o
Signed-off-by: Jeff Johnson <quic_jjohnson@quicinc.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240527-md-fs-ext4-v1-1-07aad5936bb1@quicinc.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After calling the ext4_da_map_blocks(), a delalloc extent state could
be identified through the EXT4_MAP_DELAYED flag in map. So factor out
buffer_head related handles in ext4_da_map_blocks(), make this function
buffer_head unaware and becomes a common helper, and also update the
stale function commtents, preparing for the iomap da write path in the
future.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-11-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Rename ext4_insert_delayed_block() to ext4_insert_delayed_blocks(),
pass length parameter to make it insert multiple delalloc blocks at a
time. For non-bigalloc case, just reserve len blocks and insert delalloc
extent. For bigalloc case, we can ensure that the clusters in the middle
of a extent must be unallocated, we only need to check whether the start
and end clusters are delayed/allocated. We should subtract the space for
the start and/or end block(s) if they are allocated.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-10-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Factor out a common helper ext4_clu_alloc_state(), check whether the
cluster containing a delalloc block to be added has been allocated or
has delalloc reservation, no logic changes.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-9-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Add 'nr_resv' parameter to ext4_da_reserve_space(), which indicates the
number of clusters wants to reserve, make it reserve multiple clusters
at a time.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Rename ext4_es_insert_delayed_block() to ext4_es_insert_delayed_extent()
and pass length parameter to make it insert multiple delalloc blocks at
a time. For the case of bigalloc, split the allocated parameter to
lclu_allocated and end_allocated. lclu_allocated indicates the
allocation state of the cluster which is containing the lblk,
end_allocated indicates the allocation state of the extent end, clusters
in the middle of delay allocated extent must be unallocated.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-7-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The start block of the delalloc extent to be inserted is equal to
map->m_lblk, just drop the duplicate iblock input parameter.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
Link: https://patch.msgid.link/20240517124005.347221-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In ext4_da_map_blocks(), we could find four kind of extents in the
extent status tree: hole, unwritten, written and delayed extent. Now we
only trim the map len if we found an unwritten extent or a written
extent. This is okay now since map->m_len is always set to one and we
always insert one delayed block at a time. But this will become isn't
okay for other two cases if ext4_insert_delayed_block() and
ext4_da_map_blocks() support inserting multiple map->len blocks later.
1. If we found a hole in the extent status tree which es->es_len is
shorter than the length we want to write, we should trim the
map->m_len to prevent adding extra delay more blocks than we
expected. For example, assume we write data [A, C) to a file that
contains a hole extent [A, B) and a written extent [B, D) in the
cache.
A B C D
before da write: ...hhhhhh|wwwwww....
Then we will get extent [A, B), we should trim map->m_len to B-A
before inserting new delalloc blocks, if not, the range [B, C) will
be duplicated.
2. If we found a delayed extent in the extent status tree which
es->es_len is shorter than the length we want to write, we should
trim the map->m_len to es->es_len and return directly since the front
part of this map has been delayed, we can't insert the delalloc
extent that contains the latter part in this round, we should return
the delayed length and the caller should increase the position and
call ext4_da_map_blocks() again. For example, assume we write data
[A, C) to a file that contains a delayed extent [A, B) in the cache.
A B C
before da write: ...dddddd|hhh....
Then we will get delayed extent [A, B), we should also trim map->m_len
to B-A and return, if not, we will incorrectly assume that the write
is complete and won't insert [B, C).
So we need to always trim the map->m_len if the found es->es_len in the
extent status tree is shorter than the map->m_len, prearing for
inserting a extent with multiple delalloc blocks. This patch only does a
pre-fix, the handle is crude and ext4_da_map_blocks() deserve a cleanup,
we will do that later.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The per-inode i_reserved_data_blocks count the reserved delalloc blocks
in a regular file, it should be zero when destroying the file. The
per-fs s_dirtyclusters_counter count all reserved delalloc blocks in a
filesystem, it also should be zero when umounting the filesystem. Now we
have only an error message if the i_reserved_data_blocks is not zero,
which is unable to be simply captured, so add WARN_ON_ONCE to make it
more visable.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_da_map_blocks looks up for any extent entry in the extent status
tree (w/o i_data_sem) and then the looks up for any ondisk extent
mapping (with i_data_sem in read mode).
If it finds a hole in the extent status tree or if it couldn't find any
entry at all, it then takes the i_data_sem in write mode to add a da
entry into the extent status tree. This can actually race with page
mkwrite & fallocate path.
Note that this is ok between
1. ext4 buffered-write path v/s ext4_page_mkwrite(), because of the
folio lock
2. ext4 buffered write path v/s ext4 fallocate because of the inode
lock.
But this can race between ext4_page_mkwrite() & ext4 fallocate path
ext4_page_mkwrite() ext4_fallocate()
block_page_mkwrite()
ext4_da_map_blocks()
//find hole in extent status tree
ext4_alloc_file_blocks()
ext4_map_blocks()
//allocate block and unwritten extent
ext4_insert_delayed_block()
ext4_da_reserve_space()
//reserve one more block
ext4_es_insert_delayed_block()
//drop unwritten extent and add delayed extent by mistake
Then, the delalloc extent is wrong until writeback and the extra
reserved block can't be released any more and it triggers below warning:
EXT4-fs (pmem2): Inode 13 (00000000bbbd4d23): i_reserved_data_blocks(1) not cleared!
Fix the problem by looking up extent status tree again while the
i_data_sem is held in write mode. If it still can't find any entry, then
we insert a new da entry into the extent status tree.
Cc: stable@vger.kernel.org
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20240517124005.347221-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
