From ff782e0a131c7f669445c07fe5c7ba91e043b7ed Mon Sep 17 00:00:00 2001 From: Josef Bacik Date: Thu, 8 Oct 2009 15:30:04 -0400 Subject: Btrfs: optimize fsync for the single writer case This patch optimizes the tree logging stuff so it doesn't always wait 1 jiffie for new people to join the logging transaction if there is only ever 1 writer. This helps a little bit with latency where we have something like RPM where it will fdatasync every file it writes, and so waiting the 1 jiffie for every fdatasync really starts to add up. Signed-off-by: Josef Bacik Signed-off-by: Chris Mason --- fs/btrfs/tree-log.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/tree-log.c') diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 4d7d9abef42..78f6254ac2d 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -137,11 +137,20 @@ static int start_log_trans(struct btrfs_trans_handle *trans, mutex_lock(&root->log_mutex); if (root->log_root) { + if (!root->log_start_pid) { + root->log_start_pid = current->pid; + root->log_multiple_pids = false; + } else if (root->log_start_pid != current->pid) { + root->log_multiple_pids = true; + } + root->log_batch++; atomic_inc(&root->log_writers); mutex_unlock(&root->log_mutex); return 0; } + root->log_multiple_pids = false; + root->log_start_pid = current->pid; mutex_lock(&root->fs_info->tree_log_mutex); if (!root->fs_info->log_root_tree) { ret = btrfs_init_log_root_tree(trans, root->fs_info); @@ -1985,7 +1994,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, if (atomic_read(&root->log_commit[(index1 + 1) % 2])) wait_log_commit(trans, root, root->log_transid - 1); - while (1) { + while (root->log_multiple_pids) { unsigned long batch = root->log_batch; mutex_unlock(&root->log_mutex); schedule_timeout_uninterruptible(1); @@ -2011,6 +2020,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, root->log_batch = 0; root->log_transid++; log->log_transid = root->log_transid; + root->log_start_pid = 0; smp_mb(); /* * log tree has been flushed to disk, new modifications of -- cgit v1.2.3-70-g09d2 From 4722607db6a78bd7748c51fa4c8d7371da797254 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 13 Oct 2009 12:55:09 -0400 Subject: Btrfs: only write one super copy during fsync During a tree-log commit for fsync, we've been writing at least two copies of the super block and forcing them to disk. The other filesystems write only one, and this change brings us on par with them. A full transaction commit will write all the super copies, so we still have redundant info written on a regular basis. Signed-off-by: Chris Mason --- fs/btrfs/tree-log.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/btrfs/tree-log.c') diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 78f6254ac2d..6d9ec285644 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -2092,7 +2092,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, * the running transaction open, so a full commit can't hop * in and cause problems either. */ - write_ctree_super(trans, root->fs_info->tree_root, 2); + write_ctree_super(trans, root->fs_info->tree_root, 1); ret = 0; out_wake_log_root: -- cgit v1.2.3-70-g09d2 From 257c62e1bce03e5b9f3f069fd52ad73a56de71fd Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 13 Oct 2009 13:21:08 -0400 Subject: Btrfs: avoid tree log commit when there are no changes rpm has a habit of running fdatasync when the file hasn't changed. We already detect if a file hasn't been changed in the current transaction but it might have been sent to the tree-log in this transaction and not changed since the last call to fsync. In this case, we want to avoid a tree log sync, which includes a number of synchronous writes and barriers. This commit extends the existing tracking of the last transaction to change a file to also track the last sub-transaction. The end result is that rpm -ivh and -Uvh are roughly twice as fast, and on par with ext3. Signed-off-by: Chris Mason --- fs/btrfs/btrfs_inode.h | 6 ++++++ fs/btrfs/ctree.h | 1 + fs/btrfs/disk-io.c | 2 ++ fs/btrfs/file.c | 41 ++++++++++++++++++++++++++--------------- fs/btrfs/inode.c | 6 +++++- fs/btrfs/transaction.h | 1 + fs/btrfs/tree-log.c | 27 +++++++++++++++++++++++++++ fs/btrfs/tree-log.h | 3 +++ 8 files changed, 71 insertions(+), 16 deletions(-) (limited to 'fs/btrfs/tree-log.c') diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h index c71abec0ab9..f6783a42f01 100644 --- a/fs/btrfs/btrfs_inode.h +++ b/fs/btrfs/btrfs_inode.h @@ -86,6 +86,12 @@ struct btrfs_inode { * transid of the trans_handle that last modified this inode */ u64 last_trans; + + /* + * log transid when this inode was last modified + */ + u64 last_sub_trans; + /* * transid that last logged this inode */ diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 36a19cd43e0..d0cede5ff25 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -1009,6 +1009,7 @@ struct btrfs_root { atomic_t log_writers; atomic_t log_commit[2]; unsigned long log_transid; + unsigned long last_log_commit; unsigned long log_batch; pid_t log_start_pid; bool log_multiple_pids; diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index ac8927bdc33..d4132aad9ea 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -919,6 +919,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, atomic_set(&root->log_writers, 0); root->log_batch = 0; root->log_transid = 0; + root->last_log_commit = 0; extent_io_tree_init(&root->dirty_log_pages, fs_info->btree_inode->i_mapping, GFP_NOFS); @@ -1089,6 +1090,7 @@ int btrfs_add_log_tree(struct btrfs_trans_handle *trans, WARN_ON(root->log_root); root->log_root = log_root; root->log_transid = 0; + root->last_log_commit = 0; return 0; } diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index 53fb1c997f0..4599113ed72 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -1087,8 +1087,10 @@ out_nolock: btrfs_end_transaction(trans, root); else btrfs_commit_transaction(trans, root); - } else { + } else if (ret != BTRFS_NO_LOG_SYNC) { btrfs_commit_transaction(trans, root); + } else { + btrfs_end_transaction(trans, root); } } if (file->f_flags & O_DIRECT) { @@ -1138,6 +1140,13 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) int ret = 0; struct btrfs_trans_handle *trans; + + /* we wait first, since the writeback may change the inode */ + root->log_batch++; + /* the VFS called filemap_fdatawrite for us */ + btrfs_wait_ordered_range(inode, 0, (u64)-1); + root->log_batch++; + /* * check the transaction that last modified this inode * and see if its already been committed @@ -1145,6 +1154,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) if (!BTRFS_I(inode)->last_trans) goto out; + /* + * if the last transaction that changed this file was before + * the current transaction, we can bail out now without any + * syncing + */ mutex_lock(&root->fs_info->trans_mutex); if (BTRFS_I(inode)->last_trans <= root->fs_info->last_trans_committed) { @@ -1154,13 +1168,6 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) } mutex_unlock(&root->fs_info->trans_mutex); - root->log_batch++; - filemap_fdatawrite(inode->i_mapping); - btrfs_wait_ordered_range(inode, 0, (u64)-1); - root->log_batch++; - - if (datasync && !(inode->i_state & I_DIRTY_PAGES)) - goto out; /* * ok we haven't committed the transaction yet, lets do a commit */ @@ -1189,14 +1196,18 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) */ mutex_unlock(&dentry->d_inode->i_mutex); - if (ret > 0) { - ret = btrfs_commit_transaction(trans, root); - } else { - ret = btrfs_sync_log(trans, root); - if (ret == 0) - ret = btrfs_end_transaction(trans, root); - else + if (ret != BTRFS_NO_LOG_SYNC) { + if (ret > 0) { ret = btrfs_commit_transaction(trans, root); + } else { + ret = btrfs_sync_log(trans, root); + if (ret == 0) + ret = btrfs_end_transaction(trans, root); + else + ret = btrfs_commit_transaction(trans, root); + } + } else { + ret = btrfs_end_transaction(trans, root); } mutex_lock(&dentry->d_inode->i_mutex); out: diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index ef399a7794f..5b9567caba0 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3480,6 +3480,7 @@ static noinline void init_btrfs_i(struct inode *inode) bi->generation = 0; bi->sequence = 0; bi->last_trans = 0; + bi->last_sub_trans = 0; bi->logged_trans = 0; bi->delalloc_bytes = 0; bi->reserved_bytes = 0; @@ -4980,7 +4981,9 @@ again: set_page_dirty(page); SetPageUptodate(page); - BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; + BTRFS_I(inode)->last_trans = root->fs_info->generation; + BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; + unlock_extent(io_tree, page_start, page_end, GFP_NOFS); out_unlock: @@ -5100,6 +5103,7 @@ struct inode *btrfs_alloc_inode(struct super_block *sb) if (!ei) return NULL; ei->last_trans = 0; + ei->last_sub_trans = 0; ei->logged_trans = 0; ei->outstanding_extents = 0; ei->reserved_extents = 0; diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h index 663c6740491..f68cbbe61e5 100644 --- a/fs/btrfs/transaction.h +++ b/fs/btrfs/transaction.h @@ -79,6 +79,7 @@ static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans, struct inode *inode) { BTRFS_I(inode)->last_trans = trans->transaction->transid; + BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; } int btrfs_end_transaction(struct btrfs_trans_handle *trans, diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 6d9ec285644..0a1bde26896 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -1980,6 +1980,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, int ret; struct btrfs_root *log = root->log_root; struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; + u64 log_transid = 0; mutex_lock(&root->log_mutex); index1 = root->log_transid % 2; @@ -2018,6 +2019,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, btrfs_set_root_node(&log->root_item, log->node); root->log_batch = 0; + log_transid = root->log_transid; root->log_transid++; log->log_transid = root->log_transid; root->log_start_pid = 0; @@ -2095,6 +2097,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, write_ctree_super(trans, root->fs_info->tree_root, 1); ret = 0; + mutex_lock(&root->log_mutex); + if (root->last_log_commit < log_transid) + root->last_log_commit = log_transid; + mutex_unlock(&root->log_mutex); + out_wake_log_root: atomic_set(&log_root_tree->log_commit[index2], 0); smp_mb(); @@ -2862,6 +2869,21 @@ out: return ret; } +static int inode_in_log(struct btrfs_trans_handle *trans, + struct inode *inode) +{ + struct btrfs_root *root = BTRFS_I(inode)->root; + int ret = 0; + + mutex_lock(&root->log_mutex); + if (BTRFS_I(inode)->logged_trans == trans->transid && + BTRFS_I(inode)->last_sub_trans <= root->last_log_commit) + ret = 1; + mutex_unlock(&root->log_mutex); + return ret; +} + + /* * helper function around btrfs_log_inode to make sure newly created * parent directories also end up in the log. A minimal inode and backref @@ -2901,6 +2923,11 @@ int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, if (ret) goto end_no_trans; + if (inode_in_log(trans, inode)) { + ret = BTRFS_NO_LOG_SYNC; + goto end_no_trans; + } + start_log_trans(trans, root); ret = btrfs_log_inode(trans, root, inode, inode_only); diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h index d09c7609e16..0776eacb508 100644 --- a/fs/btrfs/tree-log.h +++ b/fs/btrfs/tree-log.h @@ -19,6 +19,9 @@ #ifndef __TREE_LOG_ #define __TREE_LOG_ +/* return value for btrfs_log_dentry_safe that means we don't need to log it at all */ +#define BTRFS_NO_LOG_SYNC 256 + int btrfs_sync_log(struct btrfs_trans_handle *trans, struct btrfs_root *root); int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root); -- cgit v1.2.3-70-g09d2 From 690587d109ffe19d6743e4cc80c18b0906b7f9ff Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 13 Oct 2009 13:29:19 -0400 Subject: Btrfs: streamline tree-log btree block writeout Syncing the tree log is a 3 phase operation. 1) write and wait for all the tree log blocks for a given root. 2) write and wait for all the tree log blocks for the tree of tree log roots. 3) write and wait for the super blocks (barriers here) This isn't as efficient as it could be because there is no requirement to wait for the blocks from step one to hit the disk before we start writing the blocks from step two. This commit changes the sequence so that we don't start waiting until all the tree blocks from both steps one and two have been sent to disk. We do this by breaking up btrfs_write_wait_marked_extents into two functions, which is trivial because it was already broken up into two parts. Signed-off-by: Chris Mason --- fs/btrfs/transaction.c | 45 ++++++++++++++++++++++++++++++++++++++++++--- fs/btrfs/transaction.h | 4 ++++ fs/btrfs/tree-log.c | 8 +++++++- 3 files changed, 53 insertions(+), 4 deletions(-) (limited to 'fs/btrfs/tree-log.c') diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c index 0b8f36d4400..bca82a4ca8e 100644 --- a/fs/btrfs/transaction.c +++ b/fs/btrfs/transaction.c @@ -344,10 +344,10 @@ int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, /* * when btree blocks are allocated, they have some corresponding bits set for * them in one of two extent_io trees. This is used to make sure all of - * those extents are on disk for transaction or log commit + * those extents are sent to disk but does not wait on them */ -int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, - struct extent_io_tree *dirty_pages) +int btrfs_write_marked_extents(struct btrfs_root *root, + struct extent_io_tree *dirty_pages) { int ret; int err = 0; @@ -394,6 +394,29 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, page_cache_release(page); } } + if (err) + werr = err; + return werr; +} + +/* + * when btree blocks are allocated, they have some corresponding bits set for + * them in one of two extent_io trees. This is used to make sure all of + * those extents are on disk for transaction or log commit. We wait + * on all the pages and clear them from the dirty pages state tree + */ +int btrfs_wait_marked_extents(struct btrfs_root *root, + struct extent_io_tree *dirty_pages) +{ + int ret; + int err = 0; + int werr = 0; + struct page *page; + struct inode *btree_inode = root->fs_info->btree_inode; + u64 start = 0; + u64 end; + unsigned long index; + while (1) { ret = find_first_extent_bit(dirty_pages, 0, &start, &end, EXTENT_DIRTY); @@ -424,6 +447,22 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, return werr; } +/* + * when btree blocks are allocated, they have some corresponding bits set for + * them in one of two extent_io trees. This is used to make sure all of + * those extents are on disk for transaction or log commit + */ +int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, + struct extent_io_tree *dirty_pages) +{ + int ret; + int ret2; + + ret = btrfs_write_marked_extents(root, dirty_pages); + ret2 = btrfs_wait_marked_extents(root, dirty_pages); + return ret || ret2; +} + int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, struct btrfs_root *root) { diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h index f68cbbe61e5..d4e3e7a6938 100644 --- a/fs/btrfs/transaction.h +++ b/fs/btrfs/transaction.h @@ -108,5 +108,9 @@ int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, struct btrfs_root *root); int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, struct extent_io_tree *dirty_pages); +int btrfs_write_marked_extents(struct btrfs_root *root, + struct extent_io_tree *dirty_pages); +int btrfs_wait_marked_extents(struct btrfs_root *root, + struct extent_io_tree *dirty_pages); int btrfs_transaction_in_commit(struct btrfs_fs_info *info); #endif diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 0a1bde26896..4aff766d171 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -2013,7 +2013,10 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, goto out; } - ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); + /* we start IO on all the marked extents here, but we don't actually + * wait for them until later. + */ + ret = btrfs_write_marked_extents(log, &log->dirty_log_pages); BUG_ON(ret); btrfs_set_root_node(&log->root_item, log->node); @@ -2048,6 +2051,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, index2 = log_root_tree->log_transid % 2; if (atomic_read(&log_root_tree->log_commit[index2])) { + btrfs_wait_marked_extents(log, &log->dirty_log_pages); wait_log_commit(trans, log_root_tree, log_root_tree->log_transid); mutex_unlock(&log_root_tree->log_mutex); @@ -2067,6 +2071,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, * check the full commit flag again */ if (root->fs_info->last_trans_log_full_commit == trans->transid) { + btrfs_wait_marked_extents(log, &log->dirty_log_pages); mutex_unlock(&log_root_tree->log_mutex); ret = -EAGAIN; goto out_wake_log_root; @@ -2075,6 +2080,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, ret = btrfs_write_and_wait_marked_extents(log_root_tree, &log_root_tree->dirty_log_pages); BUG_ON(ret); + btrfs_wait_marked_extents(log, &log->dirty_log_pages); btrfs_set_super_log_root(&root->fs_info->super_for_commit, log_root_tree->node->start); -- cgit v1.2.3-70-g09d2 From 86df7eb921a009515285e7171363fa57dd2d7d31 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Wed, 14 Oct 2009 09:24:59 -0400 Subject: Btrfs: properly wait log writers during log sync A recently fsync optimization make btrfs_sync_log skip calling wait_for_writer in the single log writer case. This is incorrect since the writer count can also be increased by btrfs_pin_log. Signed-off-by: Yan Zheng Signed-off-by: Chris Mason --- fs/btrfs/tree-log.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) (limited to 'fs/btrfs/tree-log.c') diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 4aff766d171..f51bf13125c 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -1995,12 +1995,13 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, if (atomic_read(&root->log_commit[(index1 + 1) % 2])) wait_log_commit(trans, root, root->log_transid - 1); - while (root->log_multiple_pids) { + while (1) { unsigned long batch = root->log_batch; - mutex_unlock(&root->log_mutex); - schedule_timeout_uninterruptible(1); - mutex_lock(&root->log_mutex); - + if (root->log_multiple_pids) { + mutex_unlock(&root->log_mutex); + schedule_timeout_uninterruptible(1); + mutex_lock(&root->log_mutex); + } wait_for_writer(trans, root); if (batch == root->log_batch) break; -- cgit v1.2.3-70-g09d2