This patch fixes the following deadlock bug during the recovery.
INFO: task mount:1322 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
mount D ffffffff81125870 0 1322 1266 0x00000000
ffff8801207e39d8 0000000000000046 ffff88012ab1dee0 0000000000000046
ffff8801207e3a08 ffff880115903f40 ffff8801207e3fd8 ffff8801207e3fd8
ffff8801207e3fd8 ffff880115903f40 ffff8801207e39d8 ffff88012fc94520
Call Trace:
[<ffffffff81125870>] ? __lock_page+0x70/0x70
[<ffffffff816a92d9>] schedule+0x29/0x70
[<ffffffff816a93af>] io_schedule+0x8f/0xd0
[<ffffffff8112587e>] sleep_on_page+0xe/0x20
[<ffffffff816a649a>] __wait_on_bit_lock+0x5a/0xc0
[<ffffffff81125867>] __lock_page+0x67/0x70
[<ffffffff8106c7b0>] ? autoremove_wake_function+0x40/0x40
[<ffffffff81126857>] find_lock_page+0x67/0x80
[<ffffffff8112698f>] find_or_create_page+0x3f/0xb0
[<ffffffffa03901a8>] ? sync_inode_page+0xa8/0xd0 [f2fs]
[<ffffffffa038fdf7>] get_node_page+0x67/0x180 [f2fs]
[<ffffffffa039818b>] recover_fsync_data+0xacb/0xff0 [f2fs]
[<ffffffff816aaa1e>] ? _raw_spin_unlock+0x3e/0x40
[<ffffffffa0389634>] f2fs_fill_super+0x7d4/0x850 [f2fs]
[<ffffffff81184cf9>] mount_bdev+0x1c9/0x210
[<ffffffffa0388e60>] ? validate_superblock+0x180/0x180 [f2fs]
[<ffffffffa0387635>] f2fs_mount+0x15/0x20 [f2fs]
[<ffffffff81185a13>] mount_fs+0x43/0x1b0
[<ffffffff81145ba0>] ? __alloc_percpu+0x10/0x20
[<ffffffff811a0796>] vfs_kern_mount+0x76/0x120
[<ffffffff811a2cb7>] do_mount+0x237/0xa10
[<ffffffff81140b9b>] ? strndup_user+0x5b/0x80
[<ffffffff811a3520>] SyS_mount+0x90/0xe0
[<ffffffff816b3502>] system_call_fastpath+0x16/0x1b
The bug is triggered when check_index_in_prev_nodes tries to get the direct
node page by calling get_node_page.
At this point, if the direct node page is already locked by get_dnode_of_data,
its caller, we got a deadlock condition.
This patch adds additional condition check for the reuse of locked direct node
pages prior to the get_node_page call.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
If we met an error during the dentry recovery, we should not conduct checkpoint.
Otherwise, some errorneous dentry blocks overwrites the existing blocks that
contain the remaining recovery information.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
The allocated page used by the recovery is not on HIGHMEM, so that we don't
need to use kmap/kunmap.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
During the dentry recovery routine, recover_inode() triggers __f2fs_add_link
with its directory inode.
In the following scenario, a bug is captured.
1. dir = f2fs_iget(pino)
2. __f2fs_add_link(dir, name)
3. iput(dir)
-> f2fs_evict_inode() faces with BUG_ON(atomic_read(fi->dirty_dents))
Kernel BUG at ffffffffa01c0676 [verbose debug info unavailable]
[<ffffffffa01c0676>] f2fs_evict_inode+0x276/0x300 [f2fs]
Call Trace:
[<ffffffff8118ea00>] evict+0xb0/0x1b0
[<ffffffff8118f1c5>] iput+0x105/0x190
[<ffffffffa01d2dac>] recover_fsync_data+0x3bc/0x1070 [f2fs]
[<ffffffff81692e8a>] ? io_schedule+0xaa/0xd0
[<ffffffff81690acb>] ? __wait_on_bit_lock+0x7b/0xc0
[<ffffffff8111a0e7>] ? __lock_page+0x67/0x70
[<ffffffff81165e21>] ? kmem_cache_alloc+0x31/0x140
[<ffffffff8118a502>] ? __d_instantiate+0x92/0xf0
[<ffffffff812a949b>] ? security_d_instantiate+0x1b/0x30
[<ffffffff8118a5b4>] ? d_instantiate+0x54/0x70
This means that we should flush all the dentry pages between iget and iput().
But, during the recovery routine, it is unallowed due to consistency, so we
have to wait the whole recovery process.
And then, write_checkpoint flushes all the dirty dentry blocks, and nicely we
can put the stale dir inodes from the dirty_dir_inode_list.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
The reason of using sbi->por_doing is to alleviate data writes during the
recovery.
The find_fsync_dnodes() produces some dirty dentry pages, so we should
cover it too with sbi->por_doing.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
When recovering a journal file with fsync data for files that have
been deleted, don't bail out on recovery.
Signed-off-by: Chris Fries <C.Fries@motorola.com>
Reviewed-by: Russell Knize <rknize2@motorola.com>
Reviewed-by: Jason Hrycay <jason.hrycay@motorola.com>
[Jaegeuk Kim: fit the coding style]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
In the previous version, f2fs uses global locks according to the usage types,
such as directory operations, block allocation, block write, and so on.
Reference the following lock types in f2fs.h.
enum lock_type {
RENAME, /* for renaming operations */
DENTRY_OPS, /* for directory operations */
DATA_WRITE, /* for data write */
DATA_NEW, /* for data allocation */
DATA_TRUNC, /* for data truncate */
NODE_NEW, /* for node allocation */
NODE_TRUNC, /* for node truncate */
NODE_WRITE, /* for node write */
NR_LOCK_TYPE,
};
In that case, we lose the performance under the multi-threading environment,
since every types of operations must be conducted one at a time.
In order to address the problem, let's share the locks globally with a mutex
array regardless of any types.
So, let users grab a mutex and perform their jobs in parallel as much as
possbile.
For this, I propose a new global lock scheme as follows.
0. Data structure
- f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS]
- f2fs_sb_info -> node_write
1. mutex_lock_op(sbi)
- try to get an avaiable lock from the array.
- returns the index of the gottern lock variable.
2. mutex_unlock_op(sbi, index of the lock)
- unlock the given index of the lock.
3. mutex_lock_all(sbi)
- grab all the locks in the array before the checkpoint.
4. mutex_unlock_all(sbi)
- release all the locks in the array after checkpoint.
5. block_operations()
- call mutex_lock_all()
- sync_dirty_dir_inodes()
- grab node_write
- sync_node_pages()
Note that,
the pairs of mutex_lock_op()/mutex_unlock_op() and
mutex_lock_all()/mutex_unlock_all() should be used together.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
We should handle errors during the recovery flow correctly.
For example, if we get -ENOMEM, we should report a mount failure instead of
conducting the remained mount procedure.
Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch reduces redundant locking and unlocking pages during read operations.
In f2fs_readpage, let's use wait_on_page_locked() instead of lock_page.
And then, when we need to modify any data finally, let's lock the page so that
we can avoid lock contention.
[readpage rule]
- The f2fs_readpage returns unlocked page, or released page too in error cases.
- Its caller should handle read error, -EIO, after locking the page, which
indicates read completion.
- Its caller should check PageUptodate after grab_cache_page.
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Previously, f2fs reads several node pages ahead when get_dnode_of_data is called
with RDONLY_NODE flag.
And, this flag is set by the following functions.
- get_data_block_ro
- get_lock_data_page
- do_write_data_page
- truncate_blocks
- truncate_hole
However, this readahead mechanism is initially introduced for the use of
get_data_block_ro to enhance the sequential read performance.
So, let's clarify all the cases with the additional modes as follows.
enum {
ALLOC_NODE, /* allocate a new node page if needed */
LOOKUP_NODE, /* look up a node without readahead */
LOOKUP_NODE_RA, /*
* look up a node with readahead called
* by get_datablock_ro.
*/
}
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
Pull f2fs cleanup patches from Al Viro:
f2fs: get rid of fake on-stack dentries
f2fs: switch init_inode_metadata() to passing parent and name separately
f2fs: switch new_inode_page() from dentry to qstr
f2fs: init_dent_inode() should take qstr
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Conflicts:
fs/f2fs/recovery.c
This patch makes clearer the ambiguous f2fs_gc flow as follows.
1. Remove intermediate checkpoint condition during f2fs_gc
(i.e., should_do_checkpoint() and GC_BLOCKED)
2. Remove unnecessary return values of f2fs_gc because of #1.
(i.e., GC_NODE, GC_OK, etc)
3. Simplify write_checkpoint() because of #2.
4. Clarify the main f2fs_gc flow.
o monitor how many freed sections during one iteration of do_garbage_collect().
o do GC more without checkpoints if we can't get enough free sections.
o do checkpoint once we've got enough free sections through forground GCs.
5. Adopt thread-logging (Slack-Space-Recycle) scheme more aggressively on data
log types. See. get_ssr_segement()
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
1. Background
Previously, if f2fs tries to move data blocks of an *evicting* inode during the
cleaning process, it stops the process incompletely and then restarts the whole
process, since it needs a locked inode to grab victim data pages in its address
space. In order to get a locked inode, iget_locked() by f2fs_iget() is normally
used, but, it waits if the inode is on freeing.
So, here is a deadlock scenario.
1. f2fs_evict_inode() <- inode "A"
2. f2fs_balance_fs()
3. f2fs_gc()
4. gc_data_segment()
5. f2fs_iget() <- inode "A" too!
If step #1 and #5 treat a same inode "A", step #5 would fall into deadlock since
the inode "A" is on freeing. In order to resolve this, f2fs_iget_nowait() which
skips __wait_on_freeing_inode() was introduced in step #5, and stops f2fs_gc()
to complete f2fs_evict_inode().
1. f2fs_evict_inode() <- inode "A"
2. f2fs_balance_fs()
3. f2fs_gc()
4. gc_data_segment()
5. f2fs_iget_nowait() <- inode "A", then stop f2fs_gc() w/ -ENOENT
2. Problem and Solution
In the above scenario, however, f2fs cannot finish f2fs_evict_inode() only if:
o there are not enough free sections, and
o f2fs_gc() tries to move data blocks of the *evicting* inode repeatedly.
So, the final solution is to use f2fs_iget() and remove f2fs_balance_fs() in
f2fs_evict_inode().
The f2fs_evict_inode() actually truncates all the data and node blocks, which
means that it doesn't produce any dirty node pages accordingly.
So, we don't need to do f2fs_balance_fs() in practical.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This is calling list_del() inside a loop which is a problem when we try
move to the next item on the list. I've converted it to use the _safe
version. And also, as a cleanup, I've converted it to use
list_for_each_entry instead of list_for_each.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch resolves Coverity #753102:
>>> No check of the return value of "f2fs_add_link(&dent, inode)".
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
While creating a new entry for addition to the list(orphan inode list
and fsync inode entry list), there is no need to call HEAD initialization
for these entries. So, remove that init part.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Amit Sahrawat <a.sahrawat@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
In function find_fsync_dnodes() - the fsync inodes gets added to the list, but
in one path suppose f2fs_iget results in error, in such case - error gets added
to the fsync inode list.
In next call to recover_data()->get_fsync_inode()
entry = list_entry(this, struct fsync_inode_entry, list);
if (entry->inode->i_ino == ino)
This can result in "invalid access to memory" when it encounters 'error' as
entry in the fsync inode list.
So, add the fsync inode entry to the list only in case of no errors.
And, free the object at that point itself in case of issue.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Amit Sahrawat <a.sahrawat@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
In case f2fs_iget_nowait returns error, it results in truncate_hole being
called with 'error' value as inode pointer. There is no check in truncate_hole
for valid inode, so it could result in crash due "invalid access to memory".
Avoid this by handling error condition properly.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Amit Sahrawat <a.sahrawat@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
As pointed out by Randy Dunlap, this patch removes all usage of "/**" for comment
blocks. Instead, just use "/*".
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This patch should resolve the bugs reported by the sparse tool.
Initial reports were written by "kbuild test robot" managed by fengguang.wu.
In my local machines, I've tested also by running:
> make C=2 CF="-D__CHECK_ENDIAN__"
Accordingly, I've found lots of warnings and bugs related to the endian
conversion. And I've fixed all at this moment.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
This adds roll-forward routines to recover fsynced data.
- F2FS uses basically roll-back model with checkpointing.
- In order to implement fsync(), there are two approaches as follows.
1. A roll-back model with checkpointing at every fsync()
: This is a naive method, but suffers from very low performance.
2. A roll-forward model
: F2FS adopts this model where all the fsynced data should be recovered, which
were written after checkpointing was done. In order to figure out the data,
F2FS keeps a "fsync" mark in direct node blocks. In addition, F2FS remains
the location of next node block in each direct node block for reconstructing
the chain of node blocks during the recovery.
- In order to enhance the performance, F2FS keeps a "dentry" mark also in direct
node blocks. If this is set during the recovery, F2FS replays adding a dentry.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Jaegeuk Kim