jffs2: Fix lock acquisition order bug in jffs2_write_begin
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH RESEND 1/1 jffs2: Fix lock acquisition order bug in jffs2_write_begin
jffs2_write_begin() first acquires the page lock, then f->sem. This causes an AB-BA deadlock with jffs2_garbage_collect_live(), which first acquires f->sem, then the page lock:
jffs2_garbage_collect_live
mutex_lock(&f->sem) (A)
jffs2_garbage_collect_dnode
jffs2_gc_fetch_page
read_cache_page_async
do_read_cache_page
lock_page(page) (B)jffs2_write_begin
grab_cache_page_write_begin
find_lock_page
lock_page(page) (B)
mutex_lock(&f->sem) (A)We fix this by restructuring jffs2_write_begin() to take f->sem before the page lock. However, we make sure that f->sem is not held when calling jffs2_reserve_space(), as this is not permitted by the locking rules.
The deadlock above was observed multiple times on an SoC with a dual ARMv7 (Cortex-A9), running the long-term 3.4.11 kernel; it occurred when using scp to copy files from a host system to the ARM target system. The fix was heavily tested on the same target system.
If the patch is accepted, please get it also pushed to 3.4; it applies cleanly both to linux-mtd.git and the current linux-3.4 tree.
fs/jffs2/file.c +21 -18
@@ -138,33 +138,39 @@struct page *pg;struct inode *inode = mapping->host;struct jffs2_inode_info *f =JFFS2_INODE_INFO (inode); +struct jffs2_sb_info *c =JFFS2_SB_INFO (inode->i_sb); +struct jffs2_raw_inode ri; +uint32_t alloc_len =0 ;pgoff_t index = pos >>PAGE_CACHE_SHIFT ;uint32_t pageofs = index <<PAGE_CACHE_SHIFT ;int ret =0 ; +jffs2_dbg (1 ,"%s()\n" , __func__); + +if (pageofs > inode->i_size) { + ret =jffs2_reserve_space (c,sizeof (ri), &alloc_len, +ALLOC_NORMAL ,JFFS2_SUMMARY_INODE_SIZE ); +if (ret) +return ret; + } + +mutex_lock (&f->sem); pg =grab_cache_page_write_begin (mapping, index, flags); -if (!pg) +if (!pg) { +if (alloc_len) +jffs2_complete_reservation (c); +mutex_unlock (&f->sem);return -ENOMEM ; + } *pagep = pg; -jffs2_dbg (1 ,"%s()\n" , __func__); - -if (pageofs > inode->i_size) { +if (alloc_len) {/* Make new hole frag from old EOF to new page */ -struct jffs2_sb_info *c =JFFS2_SB_INFO (inode->i_sb); -struct jffs2_raw_inode ri;struct jffs2_full_dnode *fn; -uint32_t alloc_len;jffs2_dbg (1 ,"Writing new hole frag 0x%x-0x%x between current EOF and new page\n" , (unsigned int )inode->i_size, pageofs); - ret =jffs2_reserve_space (c,sizeof (ri), &alloc_len, -ALLOC_NORMAL ,JFFS2_SUMMARY_INODE_SIZE ); -if (ret) - goto out_page; - -mutex_lock (&f->sem);memset (&ri,0 ,sizeof (ri)); ri.magic =cpu_to_je16 (JFFS2_MAGIC_BITMASK );
@@ -191,7 +197,6 @@if (IS_ERR (fn)) { ret =PTR_ERR (fn);jffs2_complete_reservation (c); -mutex_unlock (&f->sem); goto out_page; } ret =jffs2_add_full_dnode_to_inode (c, f, fn);
@@ -206,12 +211,10 @@jffs2_mark_node_obsolete (c, fn->raw);jffs2_free_full_dnode (fn);jffs2_complete_reservation (c); -mutex_unlock (&f->sem); goto out_page; }jffs2_complete_reservation (c); inode->i_size = pageofs; -mutex_unlock (&f->sem); }/*
@@ -220,18 +223,18 @@* case of a short-copy. */ if (!PageUptodate (pg)) { -mutex_lock (&f->sem); ret =jffs2_do_readpage_nolock (inode, pg); -mutex_unlock (&f->sem);if (ret) goto out_page; } +mutex_unlock (&f->sem);jffs2_dbg (1 ,"end write_begin(). pg->flags %lx\n" , pg->flags);return ret; out_page:unlock_page (pg);page_cache_release (pg); +mutex_unlock (&f->sem);return ret; }
Quoted 17 lines
jffs2_write_begin()first acquires the page lock, thenf->sem. This causes an AB-BA deadlock withjffs2_garbage_collect_live(), which first acquiresf->sem, then the page lock:
jffs2_garbage_collect_livemutex_lock(&f->sem) (A) jffs2_garbage_collect_dnode jffs2_gc_fetch_page read_cache_page_async do_read_cache_page lock_page(page) (B)
jffs2_write_begingrab_cache_page_write_begin find_lock_page lock_page(page) (B) mutex_lock(&f->sem) (A)
Quoted 8 lines
We fix this by restructuring
jffs2_write_begin()to takef->sembefore the page lock. However, we make sure thatf->semis not held when callingjffs2_reserve_space(), as this is not permitted by the locking rules.The deadlock above was observed multiple times on an SoC with a dual ARMv7 (Cortex-A9), running the long-term 3.4.11 kernel; it occurred when using scp to copy files from a host system to the ARM target system. The fix was heavily tested on the same target system.
If the patch is accepted, please get it also pushed to 3.4; it applies cleanly both to linux-mtd.git and the current linux-3.4 tree.
Cc Joakim Tjernlund <[email protected]>Signed-off-by Thomas Betker <[email protected]>
David, are you happy with this?
Jocke
Quoted 1 line · Thomas Betker
jffs2_write_begin()first acquires the page lock, thenf->sem. This causes an AB-BA deadlock withjffs2_garbage_collect_live(), which first acquiresf->sem, then the page lock:
This patch does not apply to l2-mtd.git, which is based on 3.7-rc1.
$:~/git/l2-mtd$ git apply --check ~/tmp/thomas.mbox error: patch failed: fs/jffs2/file.c:138 error: fs/jffs2/file.c: patch does not apply
Could you please wrap commit message lines to 80 characters when you re-send?
Also, would you please add Cc: [email protected] [vX.Y+], where vX.Y is the starting version for which this patch is relevant, e.g., v3.0.
Thanks!
Signature · 2 lines
Best Regards, Artem Bityutskiy
Hello Artem,
I have mailed theupdated patch to the list as "[PATCH v3] ...".
Quoted 5 lines
This patch does not apply to l2-mtd.git, which is based on 3.7-rc1.
$:~/git/l2-mtd$ git apply --check ~/tmp/thomas.mbox error: patch failed: fs/jffs2/file.c:138 error: fs/jffs2/file.c: patch does not apply
Fixed, using another mailer ...
Quoted 2 lines
Could you please wrap commit message lines to 80 characters when you re-send?
Fixed -- sorry, this should not have happened.
Quoted 3 lines
Also, would you please add Cc: [email protected] [vX.Y+], where vX.Y is the starting version for which this patch is relevant, e.g., v3.0.
Fixed, using [v3.4+]. (The patch may also be relevant for earlier kernel versions, but I don't have the means to test them.)
Thanks,
Thomas