Fix reiserfs deadlocks with quotas
Jan Kara
reviewercover Fix reiserfs deadlocks with quotas
Hello,
recent user report brought to my attention that quota on reiserfs is
rather easy to deadlock. This is a fallout from BKL conversion. This series should fix the issue (at least fsstress runs now on reiserfs with quotas enabled without any lockdep warnings and reported deadlocks are solved as well). I'll push these patches to Linus soon if noone complains.
Honza
Jan Kara
reviewerPATCH 1/4 reiserfs: Fix lock ordering during remount
When remounting reiserfs dquot_suspend() or dquot_resume() can be called. These functions take dqonoff_mutex which ranks above write lock so we have to drop it before calling into quota code.
--- fs/reiserfs/super.c | 27 ++++++++++++++++++++------- 1 files changed, 20 insertions(+), 7 deletions(-)
fs/reiserfs/super.c +20 -7
@@ -1335,7 +1335,7 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg)kfree (qf_names[i]);#endif err = -EINVAL ; - goto out_err; + goto out_unlock; }#ifdef CONFIG_QUOTA handle_quota_files (s, qf_names, &qfmt);
@@ -1379,7 +1379,7 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg)if (blocks) { err =reiserfs_resize (s, blocks);if (err !=0 ) - goto out_err; + goto out_unlock; }if (*mount_flags &MS_RDONLY ) {
@@ -1389,9 +1389,15 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg)/* it is read-only already */ goto out_ok; +/* +* Drop write lock. Quota will retake it when needed and lock +* ordering requires calling dquot_suspend() without it. +*/ +reiserfs_write_unlock (s); err =dquot_suspend (s,-1 );if (err <0 ) goto out_err; +reiserfs_write_lock (s);/* try to remount file system with read-only permissions */ if (sb_umount_state(rs) == REISERFS_VALID_FS
@@ -1401,7 +1407,7 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg) err =journal_begin (&th, s,10 );if (err) - goto out_err; + goto out_unlock;/* Mounting a rw partition read-only. */ reiserfs_prepare_for_journal (s,SB_BUFFER_WITH_SB (s),1 );
@@ -1416,7 +1422,7 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg)if (reiserfs_is_journal_aborted (journal)) { err = journal->j_errno; - goto out_err; + goto out_unlock; }handle_data_mode (s, mount_options);
@@ -1425,7 +1431,7 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg) s->s_flags &= ~MS_RDONLY ;/* now it is safe to call journal_begin */ err =journal_begin (&th, s,10 );if (err) - goto out_err; + goto out_unlock;/* Mount a partition which is read-only, read-write */ reiserfs_prepare_for_journal (s,SB_BUFFER_WITH_SB (s),1 );
@@ -1442,10 +1448,16 @@ static int reiserfs_remount(struct super_block *s, int *mount_flags, char *arg)SB_JOURNAL (s)->j_must_wait =1 ; err =journal_end (&th, s,10 );if (err) - goto out_err; + goto out_unlock;if (!(*mount_flags &MS_RDONLY )) { +/* +* Drop write lock. Quota will retake it when needed and lock +* ordering requires calling dquot_resume() without it. +*/ +reiserfs_write_unlock (s);dquot_resume (s,-1 ); +reiserfs_write_lock (s);finish_unfinished (s);reiserfs_xattr_init (s, *mount_flags); }
@@ -1455,9 +1467,10 @@ out_ok:reiserfs_write_unlock (s);return 0 ; +out_unlock: +reiserfs_write_unlock (s); out_err:kfree (new_opts); -reiserfs_write_unlock (s);return err; }
Signature · 1 line
1.7.1
Jan Kara
reviewerPATCH 2/4 reiserfs: Protect reiserfs_quota_on() with write lock
In reply to Jan KaraIn reiserfs_quota_on() we do quite some work - for example unpacking tail of a quota file. Thus we have to hold write lock until a moment we call back into the quota code.
--- fs/reiserfs/super.c | 11 ++++++++--- 1 files changed, 8 insertions(+), 3 deletions(-)
fs/reiserfs/super.c +8 -3
@@ -2216,8 +2216,11 @@ static int reiserfs_quota_on(struct super_block *sb, int type, int format_id,struct reiserfs_transaction_handle th;int opt = type ==USRQUOTA ?REISERFS_USRQUOTA :REISERFS_GRPQUOTA ; -if (!(REISERFS_SB (sb)->s_mount_opt & (1 << opt))) -return -EINVAL ; +reiserfs_write_lock (sb); +if (!(REISERFS_SB (sb)->s_mount_opt & (1 << opt))) { + err = -EINVAL ; + goto out; + }/* Quotafile not on the same filesystem? */ if (path->dentry->d_sb != sb) {
@@ -2259,8 +2262,10 @@ static int reiserfs_quota_on(struct super_block *sb, int type, int format_id,if (err) goto out; } - err =dquot_quota_on (sb, type, format_id, path); +reiserfs_write_unlock (sb); +return dquot_quota_on (sb, type, format_id, path); out: +reiserfs_write_unlock (sb);return err; }
Signature · 1 line
1.7.1
Jan Kara
reviewerPATCH 3/4 reiserfs: Protect reiserfs_quota_write() with write lock
In reply to Jan KaraCalls into reiserfs journalling code and reiserfs_get_block() need to be protected with write lock. We remove write lock around calls to high level quota code in the next patch so these paths would suddently become unprotected.
--- fs/reiserfs/super.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)
fs/reiserfs/super.c +4 -0
@@ -2338,7 +2338,9 @@ static ssize_t reiserfs_quota_write(struct super_block *sb, int type, tocopy = sb->s_blocksize - offset < towrite ? sb->s_blocksize - offset : towrite; tmp_bh.b_state =0 ; +reiserfs_write_lock (sb); err =reiserfs_get_block (inode, blk, &tmp_bh,GET_BLOCK_CREATE ); +reiserfs_write_unlock (sb);if (err) goto out;if (offset || tocopy != sb->s_blocksize)
@@ -2354,10 +2356,12 @@ static ssize_t reiserfs_quota_write(struct super_block *sb, int type,flush_dcache_page (bh->b_page);set_buffer_uptodate (bh);unlock_buffer (bh); +reiserfs_write_lock (sb);reiserfs_prepare_for_journal (sb, bh,1 );journal_mark_dirty (current->journal_info, sb, bh);if (!journal_quota)reiserfs_add_ordered_list (inode, bh); +reiserfs_write_unlock (sb);brelse (bh); offset =0 ; towrite -= tocopy;
Signature · 1 line
1.7.1
Jan Kara
reviewerPATCH 4/4 reiserfs: Move quota calls out of write lock
In reply to Jan KaraCalls into highlevel quota code cannot happen under the write lock. These calls take dqio_mutex which ranks above write lock. So drop write lock before calling back into quota code.
--- fs/reiserfs/inode.c | 10 +++++++--- fs/reiserfs/stree.c | 4 ++++ fs/reiserfs/super.c | 18 ++++++++++++++---- 3 files changed, 25 insertions(+), 7 deletions(-)
fs/reiserfs/inode.c +7 -3
@@ -1782,8 +1782,9 @@ int reiserfs_new_inode(struct reiserfs_transaction_handle *th,BUG_ON (!th->t_trans_id); -dquot_initialize (inode); +reiserfs_write_unlock (inode->i_sb); err =dquot_alloc_inode (inode); +reiserfs_write_lock (inode->i_sb);if (err) goto out_end_trans;if (!dir->i_nlink) {
@@ -1979,8 +1980,10 @@ int reiserfs_new_inode(struct reiserfs_transaction_handle *th, out_end_trans:journal_end (th, th->t_super, th->t_blocks_allocated); +reiserfs_write_unlock (inode->i_sb);/* Drop can be outside and it needs more credits so it's better to have it outside */ dquot_drop (inode); +reiserfs_write_lock (inode->i_sb); inode->i_flags |=S_NOQUOTA ;make_bad_inode (inode);
@@ -3103,10 +3106,9 @@ int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)/* must be turned off for recursive notify_change calls */ ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID |ATTR_KILL_SGID ); - depth =reiserfs_write_lock_once (inode->i_sb);if (is_quota_modification (inode, attr))dquot_initialize (inode); - + depth =reiserfs_write_lock_once (inode->i_sb);if (attr->ia_valid &ATTR_SIZE ) {/* version 2 items will be caught by the s_maxbytes check ** done for us in vmtruncate
@@ -3170,7 +3172,9 @@ int reiserfs_setattr(struct dentry *dentry, struct iattr *attr) error =journal_begin (&th, inode->i_sb, jbegin_count);if (error) goto out; +reiserfs_write_unlock_once (inode->i_sb, depth); error =dquot_transfer (inode, attr); + depth =reiserfs_write_lock_once (inode->i_sb);if (error) {journal_end (&th, inode->i_sb, jbegin_count); goto out;
fs/reiserfs/stree.c +4 -0
@@ -1968,7 +1968,9 @@ int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct tree key2type(&(key->on_disk_key)));#endif +reiserfs_write_unlock (inode->i_sb); retval =dquot_alloc_space_nodirty (inode, pasted_size); +reiserfs_write_lock (inode->i_sb);if (retval) {pathrelse (search_path);return retval;
@@ -2061,9 +2063,11 @@ int reiserfs_insert_item(struct reiserfs_transaction_handle *th,"reiserquota insert_item(): allocating %u id=%u type=%c" , quota_bytes, inode->i_uid, head2type(ih));#endif +reiserfs_write_unlock (inode->i_sb);/* We can't dirty inode here. It would be immediately written but * appropriate stat item isn't inserted yet... */ retval =dquot_alloc_space_nodirty (inode, quota_bytes); +reiserfs_write_lock (inode->i_sb);if (retval) {pathrelse (path);return retval;
fs/reiserfs/super.c +14 -4
@@ -298,7 +298,9 @@ static int finish_unfinished(struct super_block *s) retval =remove_save_link_only (s, &save_link_key,0 );continue ; } +reiserfs_write_unlock (s);dquot_initialize (inode); +reiserfs_write_lock (s);if (truncate &&S_ISDIR (inode->i_mode)) {/* We got a truncate request for a dir which is impossible.
@@ -2108,13 +2110,15 @@ static int reiserfs_write_dquot(struct dquot *dquot) REISERFS_QUOTA_TRANS_BLOCKS(dquot->dq_sb));if (ret) goto out; +reiserfs_write_unlock (dquot->dq_sb); ret =dquot_commit (dquot); +reiserfs_write_lock (dquot->dq_sb); err =journal_end (&th, dquot->dq_sb,REISERFS_QUOTA_TRANS_BLOCKS (dquot->dq_sb));if (!ret && err) ret = err; - out: +out:reiserfs_write_unlock (dquot->dq_sb);return ret; }
@@ -2130,13 +2134,15 @@ static int reiserfs_acquire_dquot(struct dquot *dquot) REISERFS_QUOTA_INIT_BLOCKS(dquot->dq_sb));if (ret) goto out; +reiserfs_write_unlock (dquot->dq_sb); ret =dquot_acquire (dquot); +reiserfs_write_lock (dquot->dq_sb); err =journal_end (&th, dquot->dq_sb,REISERFS_QUOTA_INIT_BLOCKS (dquot->dq_sb));if (!ret && err) ret = err; - out: +out:reiserfs_write_unlock (dquot->dq_sb);return ret; }
@@ -2150,19 +2156,21 @@ static int reiserfs_release_dquot(struct dquot *dquot) ret =journal_begin (&th, dquot->dq_sb,REISERFS_QUOTA_DEL_BLOCKS (dquot->dq_sb)); +reiserfs_write_unlock (dquot->dq_sb);if (ret) {/* Release dquot anyway to avoid endless cycle in dqput() */ dquot_release (dquot); goto out; } ret =dquot_release (dquot); +reiserfs_write_lock (dquot->dq_sb); err =journal_end (&th, dquot->dq_sb,REISERFS_QUOTA_DEL_BLOCKS (dquot->dq_sb));if (!ret && err) ret = err; - out:reiserfs_write_unlock (dquot->dq_sb); +out:return ret; }
@@ -2187,11 +2195,13 @@ static int reiserfs_write_info(struct super_block *sb, int type) ret =journal_begin (&th, sb,2 );if (ret) goto out; +reiserfs_write_unlock (sb); ret =dquot_commit_info (sb, type); +reiserfs_write_lock (sb); err =journal_end (&th, sb,2 );if (!ret && err) ret = err; - out: +out:reiserfs_write_unlock (sb);return ret; }
Signature · 1 line
1.7.1