From cbc3d65ebcb0c494183d45cf202a53352cbf3871 Mon Sep 17 00:00:00 2001 From: Dave Kleikamp Date: Wed, 27 Jul 2005 09:17:57 -0500 Subject: JFS: Improve sync barrier processing Under heavy load, hot metadata pages are often locked by non-committed transactions, making them difficult to flush to disk. This prevents the sync point from advancing past a transaction that had modified the page. There is a point during the sync barrier processing where all outstanding transactions have been committed to disk, but no new transaction have been allowed to proceed. This is the best time to write the metadata. Signed-off-by: Dave Kleikamp --- fs/jfs/jfs_logmgr.c | 36 +++++++++++++++++++----------------- fs/jfs/jfs_logmgr.h | 2 +- fs/jfs/jfs_txnmgr.c | 10 +++++----- fs/jfs/super.c | 2 +- 4 files changed, 26 insertions(+), 24 deletions(-) (limited to 'fs') diff --git a/fs/jfs/jfs_logmgr.c b/fs/jfs/jfs_logmgr.c index 22815e8..d27bac6 100644 --- a/fs/jfs/jfs_logmgr.c +++ b/fs/jfs/jfs_logmgr.c @@ -191,7 +191,7 @@ static int lbmIOWait(struct lbuf * bp, int flag); static bio_end_io_t lbmIODone; static void lbmStartIO(struct lbuf * bp); static void lmGCwrite(struct jfs_log * log, int cant_block); -static int lmLogSync(struct jfs_log * log, int nosyncwait); +static int lmLogSync(struct jfs_log * log, int hard_sync); @@ -915,19 +915,17 @@ static void lmPostGC(struct lbuf * bp) * if new sync address is available * (normally the case if sync() is executed by back-ground * process). - * if not, explicitly run jfs_blogsync() to initiate - * getting of new sync address. * calculate new value of i_nextsync which determines when * this code is called again. * * PARAMETERS: log - log structure - * nosyncwait - 1 if called asynchronously + * hard_sync - 1 to force all metadata to be written * * RETURN: 0 * * serialization: LOG_LOCK() held on entry/exit */ -static int lmLogSync(struct jfs_log * log, int nosyncwait) +static int lmLogSync(struct jfs_log * log, int hard_sync) { int logsize; int written; /* written since last syncpt */ @@ -941,11 +939,18 @@ static int lmLogSync(struct jfs_log * log, int nosyncwait) unsigned long flags; /* push dirty metapages out to disk */ - list_for_each_entry(sbi, &log->sb_list, log_list) { - filemap_flush(sbi->ipbmap->i_mapping); - filemap_flush(sbi->ipimap->i_mapping); - filemap_flush(sbi->direct_inode->i_mapping); - } + if (hard_sync) + list_for_each_entry(sbi, &log->sb_list, log_list) { + filemap_fdatawrite(sbi->ipbmap->i_mapping); + filemap_fdatawrite(sbi->ipimap->i_mapping); + filemap_fdatawrite(sbi->direct_inode->i_mapping); + } + else + list_for_each_entry(sbi, &log->sb_list, log_list) { + filemap_flush(sbi->ipbmap->i_mapping); + filemap_flush(sbi->ipimap->i_mapping); + filemap_flush(sbi->direct_inode->i_mapping); + } /* * forward syncpt @@ -1021,10 +1026,6 @@ static int lmLogSync(struct jfs_log * log, int nosyncwait) /* next syncpt trigger = written + more */ log->nextsync = written + more; - /* return if lmLogSync() from outside of transaction, e.g., sync() */ - if (nosyncwait) - return lsn; - /* if number of bytes written from last sync point is more * than 1/4 of the log size, stop new transactions from * starting until all current transactions are completed @@ -1049,11 +1050,12 @@ static int lmLogSync(struct jfs_log * log, int nosyncwait) * * FUNCTION: write log SYNCPT record for specified log * - * PARAMETERS: log - log structure + * PARAMETERS: log - log structure + * hard_sync - set to 1 to force metadata to be written */ -void jfs_syncpt(struct jfs_log *log) +void jfs_syncpt(struct jfs_log *log, int hard_sync) { LOG_LOCK(log); - lmLogSync(log, 1); + lmLogSync(log, hard_sync); LOG_UNLOCK(log); } diff --git a/fs/jfs/jfs_logmgr.h b/fs/jfs/jfs_logmgr.h index 747114c..e4978b5 100644 --- a/fs/jfs/jfs_logmgr.h +++ b/fs/jfs/jfs_logmgr.h @@ -510,6 +510,6 @@ extern int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize); extern int lmGroupCommit(struct jfs_log *, struct tblock *); extern int jfsIOWait(void *); extern void jfs_flush_journal(struct jfs_log * log, int wait); -extern void jfs_syncpt(struct jfs_log *log); +extern void jfs_syncpt(struct jfs_log *log, int hard_sync); #endif /* _H_JFS_LOGMGR */ diff --git a/fs/jfs/jfs_txnmgr.c b/fs/jfs/jfs_txnmgr.c index 121c981..3555acf 100644 --- a/fs/jfs/jfs_txnmgr.c +++ b/fs/jfs/jfs_txnmgr.c @@ -552,6 +552,11 @@ void txEnd(tid_t tid) * synchronize with logsync barrier */ if (test_bit(log_SYNCBARRIER, &log->flag)) { + TXN_UNLOCK(); + + /* write dirty metadata & forward log syncpt */ + jfs_syncpt(log, 1); + jfs_info("log barrier off: 0x%x", log->lsn); /* enable new transactions start */ @@ -560,11 +565,6 @@ void txEnd(tid_t tid) /* wakeup all waitors for logsync barrier */ TXN_WAKEUP(&log->syncwait); - TXN_UNLOCK(); - - /* forward log syncpt */ - jfs_syncpt(log); - goto wakeup; } } diff --git a/fs/jfs/super.c b/fs/jfs/super.c index ee32211..c2abdae 100644 --- a/fs/jfs/super.c +++ b/fs/jfs/super.c @@ -531,7 +531,7 @@ static int jfs_sync_fs(struct super_block *sb, int wait) /* log == NULL indicates read-only mount */ if (log) { jfs_flush_journal(log, wait); - jfs_syncpt(log); + jfs_syncpt(log, 0); } return 0; -- cgit v1.1 From 30db1ae8640d3527ca7ac8df4bcbf14ccc6ae9cd Mon Sep 17 00:00:00 2001 From: Dave Kleikamp Date: Mon, 1 Aug 2005 16:54:26 -0500 Subject: JFS: Check for invalid inodes in jfs_delete_inode Some error paths may iput an invalid inode with i_nlink=0. jfs should not try to actually delete such an inode. Signed-off-by: Dave Kleikamp --- fs/jfs/inode.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'fs') diff --git a/fs/jfs/inode.c b/fs/jfs/inode.c index 2137138..767c7ec 100644 --- a/fs/jfs/inode.c +++ b/fs/jfs/inode.c @@ -128,6 +128,10 @@ void jfs_delete_inode(struct inode *inode) { jfs_info("In jfs_delete_inode, inode = 0x%p", inode); + if (is_bad_inode(inode) || + (JFS_IP(inode)->fileset != cpu_to_le32(FILESYSTEM_I))) + return; + if (test_cflag(COMMIT_Freewmap, inode)) jfs_free_zero_link(inode); -- cgit v1.1 From 8a9cd6d676728792aaee31f30015d284acd154a3 Mon Sep 17 00:00:00 2001 From: Dave Kleikamp Date: Wed, 10 Aug 2005 11:14:39 -0500 Subject: JFS: Fix race in txLock TxAnchor.anon_list is protected by jfsTxnLock (TXN_LOCK), but there was a place in txLock() that was removing an entry from the list without holding the spinlock. Signed-off-by: Dave Kleikamp --- fs/jfs/jfs_txnmgr.c | 2 ++ fs/jfs/super.c | 2 ++ 2 files changed, 4 insertions(+) (limited to 'fs') diff --git a/fs/jfs/jfs_txnmgr.c b/fs/jfs/jfs_txnmgr.c index 3555acf..c7a92f9 100644 --- a/fs/jfs/jfs_txnmgr.c +++ b/fs/jfs/jfs_txnmgr.c @@ -657,7 +657,9 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp, /* only anonymous txn. * Remove from anon_list */ + TXN_LOCK(); list_del_init(&jfs_ip->anon_inode_list); + TXN_UNLOCK(); } jfs_ip->atlhead = tlck->next; } else { diff --git a/fs/jfs/super.c b/fs/jfs/super.c index c2abdae..9ff8972 100644 --- a/fs/jfs/super.c +++ b/fs/jfs/super.c @@ -114,6 +114,8 @@ static void jfs_destroy_inode(struct inode *inode) { struct jfs_inode_info *ji = JFS_IP(inode); + BUG_ON(!list_empty(&ji->anon_inode_list)); + spin_lock_irq(&ji->ag_lock); if (ji->active_ag != -1) { struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap; -- cgit v1.1