ufs: ufs_trunc_indirect: infinite cycle
PATCH 1/3 ufs: ufs_trunc_indirect: infinite cycle
Currently, ufs write support have two sets of problems:
work with files and work with directories.
This series of patches should solve the first problem.
This patch similar to
http://lkml.org/lkml/2006/1/17/61
this patch complements it.
The situation the same: in ufs_trunc_(not direct), we read block,
check if count of links to it is equal to one,
if so we finish cycle, if not continue.
Because of "count of links" always >=2 this operation cause infinite cycle and hang up the kernel.
---
fs/ufs/truncate.c +24 -37
@@ -237,19 +237,14 @@ static int ufs_trunc_indirect (struct infor (i =0 ; i < uspi->s_apb; i++)if (*ubh_get_addr32 (ind_ubh,i))break ; -if (i >= uspi->s_apb) { -if (ubh_max_bcount (ind_ubh) !=1 ) { - retry =1 ; - } -else { - tmp =fs32_to_cpu (sb, *p); - *p =0 ; - inode->i_blocks -= uspi->s_nspb; -mark_inode_dirty (inode); -ufs_free_blocks (inode, tmp, uspi->s_fpb); -ubh_bforget (ind_ubh); - ind_ubh =NULL ; - } +if (i >= uspi->s_apb) { + tmp =fs32_to_cpu (sb, *p); + *p =0 ; + inode->i_blocks -= uspi->s_nspb; +mark_inode_dirty (inode); +ufs_free_blocks (inode, tmp, uspi->s_fpb); +ubh_bforget (ind_ubh); + ind_ubh =NULL ; }if (IS_SYNC (inode) && ind_ubh &&ubh_buffer_dirty (ind_ubh)) {ubh_ll_rw_block (SWRITE ,1 , &ind_ubh);
@@ -305,18 +300,14 @@ static int ufs_trunc_dindirect (struct ifor (i =0 ; i < uspi->s_apb; i++)if (*ubh_get_addr32 (dind_bh, i))break ; -if (i >= uspi->s_apb) { -if (ubh_max_bcount (dind_bh) !=1 ) - retry =1 ; -else { - tmp =fs32_to_cpu (sb, *p); - *p =0 ; - inode->i_blocks -= uspi->s_nspb; -mark_inode_dirty (inode); -ufs_free_blocks (inode, tmp, uspi->s_fpb); -ubh_bforget (dind_bh); - dind_bh =NULL ; - } +if (i >= uspi->s_apb) { + tmp =fs32_to_cpu (sb, *p); + *p =0 ; + inode->i_blocks -= uspi->s_nspb; +mark_inode_dirty (inode); +ufs_free_blocks (inode, tmp, uspi->s_fpb); +ubh_bforget (dind_bh); + dind_bh =NULL ; }if (IS_SYNC (inode) && dind_bh &&ubh_buffer_dirty (dind_bh)) {ubh_ll_rw_block (SWRITE ,1 , &dind_bh);
@@ -369,18 +360,14 @@ static int ufs_trunc_tindirect (struct ifor (i =0 ; i < uspi->s_apb; i++)if (*ubh_get_addr32 (tind_bh, i))break ; -if (i >= uspi->s_apb) { -if (ubh_max_bcount (tind_bh) !=1 ) - retry =1 ; -else { - tmp =fs32_to_cpu (sb, *p); - *p =0 ; - inode->i_blocks -= uspi->s_nspb; -mark_inode_dirty (inode); -ufs_free_blocks (inode, tmp, uspi->s_fpb); -ubh_bforget (tind_bh); - tind_bh =NULL ; - } +if (i >= uspi->s_apb) { + tmp =fs32_to_cpu (sb, *p); + *p =0 ; + inode->i_blocks -= uspi->s_nspb; +mark_inode_dirty (inode); +ufs_free_blocks (inode, tmp, uspi->s_fpb); +ubh_bforget (tind_bh); + tind_bh =NULL ; }if (IS_SYNC (inode) && tind_bh &&ubh_buffer_dirty (tind_bh)) {ubh_ll_rw_block (SWRITE ,1 , &tind_bh);
fs/ufs/util.c +0 -12
@@ -139,18 +139,6 @@ void ubh_wait_on_buffer (struct ufs_buffwait_on_buffer (ubh->bh[i]); } -unsigned ubh_max_bcount (struct ufs_buffer_head * ubh) -{ -unsigned i; -unsigned max =0 ; -if (!ubh) -return 0 ; -for ( i =0 ; i < ubh->count; i++ ) -if (atomic_read (&ubh->bh[i]->b_count) > max ) - max =atomic_read (&ubh->bh[i]->b_count); -return max; -} -void ubh_bforget (struct ufs_buffer_head * ubh) {unsigned i;
fs/ufs/util.h +0 -1
@@ -238,7 +238,6 @@ extern void ubh_mark_buffer_dirty (strucextern void ubh_mark_buffer_uptodate (struct ufs_buffer_head *,int );extern void ubh_ll_rw_block (int ,unsigned ,struct ufs_buffer_head **);extern void ubh_wait_on_buffer (struct ufs_buffer_head *); -extern unsigned ubh_max_bcount (struct ufs_buffer_head *);extern void ubh_bforget (struct ufs_buffer_head *);extern int ubh_buffer_dirty (struct ufs_buffer_head *);#define ubh_ubhcpymem (mem,ubh,size) _ubh_ubhcpymem_(uspi,mem,ubh,size)
Signature · 1 line
/Evgeniy
Quoted 7 lines · Evgeniy Dushistov
The situation the same: in
ufs_trunc_(not direct),
we read block,
check if count of links to it is equal to one,
if so we finish cycle, if not continue.
Because of "count of links" always >=2 this operation cause infinite cycle and hang up the kernel.
okay, but do we know what that code which you removed was actually intended to do?
Do you know whether 2.4 kernels exhibit the same bug? If not, we might have accidentally broken UFS at some stage.
Quoted 9 lines · Andrew Morton
Quoted 7 lines · Evgeniy Dushistov
The situation the same: in
ufs_trunc_(not direct),
we read block,
check if count of links to it is equal to one,
if so we finish cycle, if not continue.
Because of "count of links" always >=2 this operation cause infinite cycle and hang up the kernel.
Quoted 3 lines
okay, but do we know what that code which you removed was actually intended to do?
I suppose it wait untill all range of buffers, which used by "inode" will be freed. "vmtruncate" should care about this.
Quoted 1 line
Do you know whether 2.4 kernels exhibit the same bug?
I suppose 2.4 doesn't affected, because of "sb_get_hash_table" (analog of sb_find_get_block in 2.6) use "get_bh" only once, so b_count equal to 1 if nobody used it.
Signature · 1 line
/Evgeniy