From b7ec40d7845bffca8bb3af2ea3f192d6257bbe21 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Thu, 12 Mar 2009 20:12:45 -0400 Subject: Btrfs: reduce stalls during transaction commit To avoid deadlocks and reduce latencies during some critical operations, some transaction writers are allowed to jump into the running transaction and make it run a little longer, while others sit around and wait for the commit to finish. This is a bit unfair, especially when the callers that jump in do a bunch of IO that makes all the others procs on the box wait. This commit reduces the stalls this produces by pre-reading file extent pointers during btrfs_finish_ordered_io before the transaction is joined. It also tunes the drop_snapshot code to politely wait for transactions that have started writing out their delayed refs to finish. This avoids new delayed refs being flooded into the queue while we're trying to close off the transaction. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 7d4f948bc22a..13a17477c4f4 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -1502,6 +1502,7 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) struct btrfs_trans_handle *trans; struct btrfs_ordered_extent *ordered_extent; struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; + struct btrfs_path *path; int compressed = 0; int ret; @@ -1509,6 +1510,23 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) if (!ret) return 0; + /* + * before we join the transaction, try to do some of our IO. + * This will limit the amount of IO that we have to do with + * the transaction running. We're unlikely to need to do any + * IO if the file extents are new, the disk_i_size checks + * covers the most common case. + */ + if (start < BTRFS_I(inode)->disk_i_size) { + path = btrfs_alloc_path(); + if (path) { + ret = btrfs_lookup_file_extent(NULL, root, path, + inode->i_ino, + start, 0); + btrfs_free_path(path); + } + } + trans = btrfs_join_transaction(root, 1); ordered_extent = btrfs_lookup_ordered_extent(inode, start); -- cgit v1.2.3 From 7f366cfecfc126731f8ac505d72026d691dac79a Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Thu, 12 Mar 2009 20:12:45 -0400 Subject: Btrfs: reduce stack in cow_file_range The fs/btrfs/inode.c code to run delayed allocation during writout needed some stack usage optimization. This is the first pass, it does the check for compression earlier on, which allows us to do the common (no compression) case higher up in the call chain. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 13a17477c4f4..c427011dc453 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -204,7 +204,7 @@ fail: * does the checks required to make sure the data is small enough * to fit as an inline extent. */ -static int cow_file_range_inline(struct btrfs_trans_handle *trans, +static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, u64 start, u64 end, size_t compressed_size, @@ -854,11 +854,6 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page, u64 cur_end; int limit = 10 * 1024 * 1042; - if (!btrfs_test_opt(root, COMPRESS)) { - return cow_file_range(inode, locked_page, start, end, - page_started, nr_written, 1); - } - clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC, 1, 0, GFP_NOFS); while (start < end) { @@ -935,7 +930,8 @@ static noinline int csum_exist_in_range(struct btrfs_root *root, * If no cow copies or snapshots exist, we write directly to the existing * blocks on disk */ -static int run_delalloc_nocow(struct inode *inode, struct page *locked_page, +static noinline int run_delalloc_nocow(struct inode *inode, + struct page *locked_page, u64 start, u64 end, int *page_started, int force, unsigned long *nr_written) { @@ -1133,6 +1129,7 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page, unsigned long *nr_written) { int ret; + struct btrfs_root *root = BTRFS_I(inode)->root; if (btrfs_test_flag(inode, NODATACOW)) ret = run_delalloc_nocow(inode, locked_page, start, end, @@ -1140,10 +1137,12 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page, else if (btrfs_test_flag(inode, PREALLOC)) ret = run_delalloc_nocow(inode, locked_page, start, end, page_started, 0, nr_written); + else if (!btrfs_test_opt(root, COMPRESS)) + ret = cow_file_range(inode, locked_page, start, end, + page_started, nr_written, 1); else ret = cow_file_range_async(inode, locked_page, start, end, page_started, nr_written); - return ret; } -- cgit v1.2.3 From b9473439d3e84d9fc1a0a83faca69cc1b7566341 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 13 Mar 2009 11:00:37 -0400 Subject: Btrfs: leave btree locks spinning more often btrfs_mark_buffer dirty would set dirty bits in the extent_io tree for the buffers it was dirtying. This may require a kmalloc and it was not atomic. So, anyone who called btrfs_mark_buffer_dirty had to set any btree locks they were holding to blocking first. This commit changes dirty tracking for extent buffers to just use a flag in the extent buffer. Now that we have one and only one extent buffer per page, this can be safely done without losing dirty bits along the way. This also introduces a path->leave_spinning flag that callers of btrfs_search_slot can use to indicate they will properly deal with a path returned where all the locks are spinning instead of blocking. Many of the btree search callers now expect spinning paths, resulting in better btree concurrency overall. Signed-off-by: Chris Mason --- fs/btrfs/ctree.c | 19 ++++++++------ fs/btrfs/ctree.h | 12 ++++++--- fs/btrfs/dir-item.c | 3 +++ fs/btrfs/disk-io.c | 67 ++++++++++++++++++++++++++++++++++++++---------- fs/btrfs/disk-io.h | 1 + fs/btrfs/extent-tree.c | 69 ++++++++++++++++++++++++++++++++++++-------------- fs/btrfs/extent_io.c | 51 +++++++------------------------------ fs/btrfs/extent_io.h | 3 +++ fs/btrfs/file-item.c | 7 +++-- fs/btrfs/file.c | 4 +++ fs/btrfs/inode-item.c | 3 +++ fs/btrfs/inode.c | 17 ++++++++++--- fs/btrfs/locking.c | 11 ++++---- fs/btrfs/tree-log.c | 1 - 14 files changed, 172 insertions(+), 96 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index 3764248bdc05..8686a3d2ab3a 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -1684,7 +1684,8 @@ done: * we don't really know what they plan on doing with the path * from here on, so for now just mark it as blocking */ - btrfs_set_path_blocking(p); + if (!p->leave_spinning) + btrfs_set_path_blocking(p); return ret; } @@ -3032,26 +3033,27 @@ int btrfs_split_item(struct btrfs_trans_handle *trans, return -EAGAIN; } + btrfs_set_path_blocking(path); ret = split_leaf(trans, root, &orig_key, path, sizeof(struct btrfs_item), 1); path->keep_locks = 0; BUG_ON(ret); + btrfs_unlock_up_safe(path, 1); + leaf = path->nodes[0]; + BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); + +split: /* * make sure any changes to the path from split_leaf leave it * in a blocking state */ btrfs_set_path_blocking(path); - leaf = path->nodes[0]; - BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); - -split: item = btrfs_item_nr(leaf, path->slots[0]); orig_offset = btrfs_item_offset(leaf, item); item_size = btrfs_item_size(leaf, item); - buf = kmalloc(item_size, GFP_NOFS); read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, path->slots[0]), item_size); @@ -3545,7 +3547,6 @@ setup_items_for_insert(struct btrfs_trans_handle *trans, } btrfs_set_header_nritems(leaf, nritems + nr); - btrfs_mark_buffer_dirty(leaf); ret = 0; if (slot == 0) { @@ -3553,6 +3554,8 @@ setup_items_for_insert(struct btrfs_trans_handle *trans, btrfs_cpu_key_to_disk(&disk_key, cpu_key); ret = fixup_low_keys(trans, root, path, &disk_key, 1); } + btrfs_unlock_up_safe(path, 1); + btrfs_mark_buffer_dirty(leaf); if (btrfs_leaf_free_space(root, leaf) < 0) { btrfs_print_leaf(root, leaf); @@ -3596,7 +3599,6 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, total_data, total_size, nr); out: - btrfs_unlock_up_safe(path, 1); return ret; } @@ -3792,6 +3794,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, slot = path->slots[1]; extent_buffer_get(leaf); + btrfs_set_path_blocking(path); wret = push_leaf_left(trans, root, path, 1, 1); if (wret < 0 && wret != -ENOSPC) ret = wret; diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 08d9f8d15538..4ddce91cf3f9 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -401,15 +401,16 @@ struct btrfs_path { int locks[BTRFS_MAX_LEVEL]; int reada; /* keep some upper locks as we walk down */ - int keep_locks; - int skip_locking; int lowest_level; /* * set by btrfs_split_item, tells search_slot to keep all locks * and to force calls to keep space in the nodes */ - int search_for_split; + unsigned int search_for_split:1; + unsigned int keep_locks:1; + unsigned int skip_locking:1; + unsigned int leave_spinning:1; }; /* @@ -779,6 +780,11 @@ struct btrfs_fs_info { atomic_t throttle_gen; u64 total_pinned; + + /* protected by the delalloc lock, used to keep from writing + * metadata until there is a nice batch + */ + u64 dirty_metadata_bytes; struct list_head dirty_cowonly_roots; struct btrfs_fs_devices *fs_devices; diff --git a/fs/btrfs/dir-item.c b/fs/btrfs/dir-item.c index 926a0b287a7d..1d70236ba00c 100644 --- a/fs/btrfs/dir-item.c +++ b/fs/btrfs/dir-item.c @@ -145,7 +145,10 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root key.objectid = dir; btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY); key.offset = btrfs_name_hash(name, name_len); + path = btrfs_alloc_path(); + path->leave_spinning = 1; + data_size = sizeof(*dir_item) + name_len; dir_item = insert_with_overflow(trans, root, path, &key, data_size, name, name_len); diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 1f1d89b18818..9244cd7313d4 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -668,14 +668,31 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, static int btree_writepage(struct page *page, struct writeback_control *wbc) { struct extent_io_tree *tree; + struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; + struct extent_buffer *eb; + int was_dirty; + tree = &BTRFS_I(page->mapping->host)->io_tree; + if (!(current->flags & PF_MEMALLOC)) { + return extent_write_full_page(tree, page, + btree_get_extent, wbc); + } - if (current->flags & PF_MEMALLOC) { - redirty_page_for_writepage(wbc, page); - unlock_page(page); - return 0; + redirty_page_for_writepage(wbc, page); + eb = btrfs_find_tree_block(root, page_offset(page), + PAGE_CACHE_SIZE); + WARN_ON(!eb); + + was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); + if (!was_dirty) { + spin_lock(&root->fs_info->delalloc_lock); + root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE; + spin_unlock(&root->fs_info->delalloc_lock); } - return extent_write_full_page(tree, page, btree_get_extent, wbc); + free_extent_buffer(eb); + + unlock_page(page); + return 0; } static int btree_writepages(struct address_space *mapping, @@ -684,15 +701,15 @@ static int btree_writepages(struct address_space *mapping, struct extent_io_tree *tree; tree = &BTRFS_I(mapping->host)->io_tree; if (wbc->sync_mode == WB_SYNC_NONE) { + struct btrfs_root *root = BTRFS_I(mapping->host)->root; u64 num_dirty; - u64 start = 0; unsigned long thresh = 32 * 1024 * 1024; if (wbc->for_kupdate) return 0; - num_dirty = count_range_bits(tree, &start, (u64)-1, - thresh, EXTENT_DIRTY); + /* this is a bit racy, but that's ok */ + num_dirty = root->fs_info->dirty_metadata_bytes; if (num_dirty < thresh) return 0; } @@ -859,9 +876,17 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, root->fs_info->running_transaction->transid) { btrfs_assert_tree_locked(buf); - /* ugh, clear_extent_buffer_dirty can be expensive */ - btrfs_set_lock_blocking(buf); + if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { + spin_lock(&root->fs_info->delalloc_lock); + if (root->fs_info->dirty_metadata_bytes >= buf->len) + root->fs_info->dirty_metadata_bytes -= buf->len; + else + WARN_ON(1); + spin_unlock(&root->fs_info->delalloc_lock); + } + /* ugh, clear_extent_buffer_dirty needs to lock the page */ + btrfs_set_lock_blocking(buf); clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf); } @@ -2348,8 +2373,7 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf) struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; u64 transid = btrfs_header_generation(buf); struct inode *btree_inode = root->fs_info->btree_inode; - - btrfs_set_lock_blocking(buf); + int was_dirty; btrfs_assert_tree_locked(buf); if (transid != root->fs_info->generation) { @@ -2360,7 +2384,13 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf) (unsigned long long)root->fs_info->generation); WARN_ON(1); } - set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf); + was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, + buf); + if (!was_dirty) { + spin_lock(&root->fs_info->delalloc_lock); + root->fs_info->dirty_metadata_bytes += buf->len; + spin_unlock(&root->fs_info->delalloc_lock); + } } void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) @@ -2400,6 +2430,7 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) int btree_lock_page_hook(struct page *page) { struct inode *inode = page->mapping->host; + struct btrfs_root *root = BTRFS_I(inode)->root; struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; struct extent_buffer *eb; unsigned long len; @@ -2415,6 +2446,16 @@ int btree_lock_page_hook(struct page *page) btrfs_tree_lock(eb); btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); + + if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { + spin_lock(&root->fs_info->delalloc_lock); + if (root->fs_info->dirty_metadata_bytes >= eb->len) + root->fs_info->dirty_metadata_bytes -= eb->len; + else + WARN_ON(1); + spin_unlock(&root->fs_info->delalloc_lock); + } + btrfs_tree_unlock(eb); free_extent_buffer(eb); out: diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h index 95029db227be..c958ecbc1916 100644 --- a/fs/btrfs/disk-io.h +++ b/fs/btrfs/disk-io.h @@ -72,6 +72,7 @@ int btrfs_insert_dev_radix(struct btrfs_root *root, void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr); int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root); void btrfs_mark_buffer_dirty(struct extent_buffer *buf); +void btrfs_mark_buffer_dirty_nonblocking(struct extent_buffer *buf); int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid); int btrfs_set_buffer_uptodate(struct extent_buffer *buf); int wait_on_tree_block_writeback(struct btrfs_root *root, diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index a421c32c6cfe..8933d15a240f 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -56,9 +56,6 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, int ref_mod); static int update_reserved_extents(struct btrfs_root *root, u64 bytenr, u64 num, int reserve); -static int pin_down_bytes(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, int is_data); static int update_block_group(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, int alloc, @@ -618,6 +615,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, } else { goto out; } + btrfs_unlock_up_safe(path, 1); btrfs_mark_buffer_dirty(path->nodes[0]); out: btrfs_release_path(root, path); @@ -760,6 +758,7 @@ static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans, return -ENOMEM; path->reada = 1; + path->leave_spinning = 1; key.objectid = bytenr; key.type = BTRFS_EXTENT_ITEM_KEY; key.offset = num_bytes; @@ -767,8 +766,10 @@ static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans, /* first find the extent item and update its reference count */ ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, 0, 1); - if (ret < 0) + if (ret < 0) { + btrfs_set_path_blocking(path); return ret; + } if (ret > 0) { WARN_ON(1); @@ -791,11 +792,15 @@ static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans, refs = btrfs_extent_refs(l, item); btrfs_set_extent_refs(l, item, refs + refs_to_add); + btrfs_unlock_up_safe(path, 1); + btrfs_mark_buffer_dirty(path->nodes[0]); btrfs_release_path(root->fs_info->extent_root, path); path->reada = 1; + path->leave_spinning = 1; + /* now insert the actual backref */ ret = insert_extent_backref(trans, root->fs_info->extent_root, path, bytenr, parent, @@ -2050,6 +2055,8 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, clear_extent_dirty(&fs_info->pinned_extents, bytenr, bytenr + num - 1, GFP_NOFS); } + mutex_unlock(&root->fs_info->pinned_mutex); + while (num > 0) { cache = btrfs_lookup_block_group(fs_info, bytenr); BUG_ON(!cache); @@ -2141,8 +2148,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, u64 end; int ret; - mutex_lock(&root->fs_info->pinned_mutex); while (1) { + mutex_lock(&root->fs_info->pinned_mutex); ret = find_first_extent_bit(unpin, 0, &start, &end, EXTENT_DIRTY); if (ret) @@ -2150,14 +2157,11 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, ret = btrfs_discard_extent(root, start, end + 1 - start); + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, start, end + 1 - start, 0); clear_extent_dirty(unpin, start, end, GFP_NOFS); - if (need_resched()) { - mutex_unlock(&root->fs_info->pinned_mutex); - cond_resched(); - mutex_lock(&root->fs_info->pinned_mutex); - } + cond_resched(); } mutex_unlock(&root->fs_info->pinned_mutex); return ret; @@ -2165,7 +2169,9 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, static int pin_down_bytes(struct btrfs_trans_handle *trans, struct btrfs_root *root, - u64 bytenr, u64 num_bytes, int is_data) + struct btrfs_path *path, + u64 bytenr, u64 num_bytes, int is_data, + struct extent_buffer **must_clean) { int err = 0; struct extent_buffer *buf; @@ -2191,15 +2197,16 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans, header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID && header_transid == trans->transid && !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { - clean_tree_block(NULL, root, buf); - btrfs_tree_unlock(buf); - free_extent_buffer(buf); + *must_clean = buf; return 1; } btrfs_tree_unlock(buf); } free_extent_buffer(buf); pinit: + btrfs_set_path_blocking(path); + mutex_lock(&root->fs_info->pinned_mutex); + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); BUG_ON(err < 0); @@ -2236,6 +2243,7 @@ static int __free_extent(struct btrfs_trans_handle *trans, return -ENOMEM; path->reada = 1; + path->leave_spinning = 1; ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent, root_objectid, ref_generation, owner_objectid, 1); @@ -2261,6 +2269,7 @@ static int __free_extent(struct btrfs_trans_handle *trans, refs_to_drop); BUG_ON(ret); btrfs_release_path(extent_root, path); + path->leave_spinning = 1; ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); if (ret) { @@ -2318,6 +2327,7 @@ static int __free_extent(struct btrfs_trans_handle *trans, /* if refs are 0, we need to setup the path for deletion */ if (refs == 0) { btrfs_release_path(extent_root, path); + path->leave_spinning = 1; ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); BUG_ON(ret); @@ -2327,16 +2337,18 @@ static int __free_extent(struct btrfs_trans_handle *trans, if (refs == 0) { u64 super_used; u64 root_used; + struct extent_buffer *must_clean = NULL; if (pin) { - mutex_lock(&root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, root, bytenr, num_bytes, - owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); - mutex_unlock(&root->fs_info->pinned_mutex); + ret = pin_down_bytes(trans, root, path, + bytenr, num_bytes, + owner_objectid >= BTRFS_FIRST_FREE_OBJECTID, + &must_clean); if (ret > 0) mark_free = 1; BUG_ON(ret < 0); } + /* block accounting for super block */ spin_lock(&info->delalloc_lock); super_used = btrfs_super_bytes_used(&info->super_copy); @@ -2348,11 +2360,27 @@ static int __free_extent(struct btrfs_trans_handle *trans, btrfs_set_root_used(&root->root_item, root_used - num_bytes); spin_unlock(&info->delalloc_lock); + + /* + * it is going to be very rare for someone to be waiting + * on the block we're freeing. del_items might need to + * schedule, so rather than get fancy, just force it + * to blocking here + */ + if (must_clean) + btrfs_set_lock_blocking(must_clean); + ret = btrfs_del_items(trans, extent_root, path, path->slots[0], num_to_del); BUG_ON(ret); btrfs_release_path(extent_root, path); + if (must_clean) { + clean_tree_block(NULL, root, must_clean); + btrfs_tree_unlock(must_clean); + free_extent_buffer(must_clean); + } + if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { ret = btrfs_del_csums(trans, root, bytenr, num_bytes); BUG_ON(ret); @@ -2480,8 +2508,9 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID && owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { mutex_lock(&root->fs_info->pinned_mutex); + + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); - mutex_unlock(&root->fs_info->pinned_mutex); update_reserved_extents(root, bytenr, num_bytes, 0); ret = 0; } else { @@ -2931,6 +2960,7 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, path = btrfs_alloc_path(); BUG_ON(!path); + path->leave_spinning = 1; ret = btrfs_insert_empty_items(trans, extent_root, path, keys, sizes, 2); BUG_ON(ret); @@ -5435,6 +5465,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; ret = btrfs_insert_empty_inode(trans, root, path, objectid); if (ret) goto out; diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index ebe6b29e6069..08085af089e2 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -3124,20 +3124,15 @@ void free_extent_buffer(struct extent_buffer *eb) int clear_extent_buffer_dirty(struct extent_io_tree *tree, struct extent_buffer *eb) { - int set; unsigned long i; unsigned long num_pages; struct page *page; - u64 start = eb->start; - u64 end = start + eb->len - 1; - - set = clear_extent_dirty(tree, start, end, GFP_NOFS); num_pages = num_extent_pages(eb->start, eb->len); for (i = 0; i < num_pages; i++) { page = extent_buffer_page(eb, i); - if (!set && !PageDirty(page)) + if (!PageDirty(page)) continue; lock_page(page); @@ -3146,22 +3141,6 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree, else set_page_private(page, EXTENT_PAGE_PRIVATE); - /* - * if we're on the last page or the first page and the - * block isn't aligned on a page boundary, do extra checks - * to make sure we don't clean page that is partially dirty - */ - if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || - ((i == num_pages - 1) && - ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { - start = (u64)page->index << PAGE_CACHE_SHIFT; - end = start + PAGE_CACHE_SIZE - 1; - if (test_range_bit(tree, start, end, - EXTENT_DIRTY, 0)) { - unlock_page(page); - continue; - } - } clear_page_dirty_for_io(page); spin_lock_irq(&page->mapping->tree_lock); if (!PageDirty(page)) { @@ -3187,29 +3166,13 @@ int set_extent_buffer_dirty(struct extent_io_tree *tree, { unsigned long i; unsigned long num_pages; + int was_dirty = 0; + was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); num_pages = num_extent_pages(eb->start, eb->len); - for (i = 0; i < num_pages; i++) { - struct page *page = extent_buffer_page(eb, i); - /* writepage may need to do something special for the - * first page, we have to make sure page->private is - * properly set. releasepage may drop page->private - * on us if the page isn't already dirty. - */ - lock_page(page); - if (i == 0) { - set_page_extent_head(page, eb->len); - } else if (PagePrivate(page) && - page->private != EXTENT_PAGE_PRIVATE) { - set_page_extent_mapped(page); - } + for (i = 0; i < num_pages; i++) __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); - set_extent_dirty(tree, page_offset(page), - page_offset(page) + PAGE_CACHE_SIZE - 1, - GFP_NOFS); - unlock_page(page); - } - return 0; + return was_dirty; } int clear_extent_buffer_uptodate(struct extent_io_tree *tree, @@ -3789,6 +3752,10 @@ int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) ret = 0; goto out; } + if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { + ret = 0; + goto out; + } /* at this point we can safely release the extent buffer */ num_pages = num_extent_pages(eb->start, eb->len); for (i = 0; i < num_pages; i++) diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index 1f9df88afbf6..5bc20abf3f3d 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -25,6 +25,7 @@ /* these are bit numbers for test/set bit */ #define EXTENT_BUFFER_UPTODATE 0 #define EXTENT_BUFFER_BLOCKING 1 +#define EXTENT_BUFFER_DIRTY 2 /* * page->private values. Every page that is controlled by the extent @@ -254,6 +255,8 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree, struct extent_buffer *eb); int set_extent_buffer_dirty(struct extent_io_tree *tree, struct extent_buffer *eb); +int test_extent_buffer_dirty(struct extent_io_tree *tree, + struct extent_buffer *eb); int set_extent_buffer_uptodate(struct extent_io_tree *tree, struct extent_buffer *eb); int clear_extent_buffer_uptodate(struct extent_io_tree *tree, diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c index 964652435fd1..9b99886562d0 100644 --- a/fs/btrfs/file-item.c +++ b/fs/btrfs/file-item.c @@ -52,6 +52,7 @@ int btrfs_insert_file_extent(struct btrfs_trans_handle *trans, file_key.offset = pos; btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY); + path->leave_spinning = 1; ret = btrfs_insert_empty_item(trans, root, path, &file_key, sizeof(*item)); if (ret < 0) @@ -523,6 +524,7 @@ int btrfs_del_csums(struct btrfs_trans_handle *trans, key.offset = end_byte - 1; key.type = BTRFS_EXTENT_CSUM_KEY; + path->leave_spinning = 1; ret = btrfs_search_slot(trans, root, &key, path, -1, 1); if (ret > 0) { if (path->slots[0] == 0) @@ -757,8 +759,10 @@ insert: } else { ins_size = csum_size; } + path->leave_spinning = 1; ret = btrfs_insert_empty_item(trans, root, path, &file_key, ins_size); + path->leave_spinning = 0; if (ret < 0) goto fail_unlock; if (ret != 0) { @@ -776,7 +780,6 @@ found: item_end = (struct btrfs_csum_item *)((unsigned char *)item_end + btrfs_item_size_nr(leaf, path->slots[0])); eb_token = NULL; - cond_resched(); next_sector: if (!eb_token || @@ -817,9 +820,9 @@ next_sector: eb_token = NULL; } btrfs_mark_buffer_dirty(path->nodes[0]); - cond_resched(); if (total_bytes < sums->len) { btrfs_release_path(root, path); + cond_resched(); goto again; } out: diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index c80075497645..f06c275644b7 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -606,6 +606,7 @@ next_slot: btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY); btrfs_release_path(root, path); + path->leave_spinning = 1; ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*extent)); BUG_ON(ret); @@ -639,7 +640,9 @@ next_slot: ram_bytes); btrfs_set_file_extent_type(leaf, extent, found_type); + btrfs_unlock_up_safe(path, 1); btrfs_mark_buffer_dirty(path->nodes[0]); + btrfs_set_lock_blocking(path->nodes[0]); if (disk_bytenr != 0) { ret = btrfs_update_extent_ref(trans, root, @@ -652,6 +655,7 @@ next_slot: BUG_ON(ret); } + path->leave_spinning = 0; btrfs_release_path(root, path); if (disk_bytenr != 0) inode_add_bytes(inode, extent_end - end); diff --git a/fs/btrfs/inode-item.c b/fs/btrfs/inode-item.c index 3d46fa1f29a4..6b627c611808 100644 --- a/fs/btrfs/inode-item.c +++ b/fs/btrfs/inode-item.c @@ -73,6 +73,8 @@ int btrfs_del_inode_ref(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; + ret = btrfs_search_slot(trans, root, &key, path, -1, 1); if (ret > 0) { ret = -ENOENT; @@ -127,6 +129,7 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; ret = btrfs_insert_empty_item(trans, root, path, &key, ins_len); if (ret == -EEXIST) { diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index c427011dc453..b83a45dc717e 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -134,6 +134,7 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; btrfs_set_trans_block_group(trans, inode); key.objectid = inode->i_ino; @@ -167,9 +168,9 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, cur_size = min_t(unsigned long, compressed_size, PAGE_CACHE_SIZE); - kaddr = kmap(cpage); + kaddr = kmap_atomic(cpage, KM_USER0); write_extent_buffer(leaf, kaddr, ptr, cur_size); - kunmap(cpage); + kunmap_atomic(kaddr, KM_USER0); i++; ptr += cur_size; @@ -1452,6 +1453,7 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, path = btrfs_alloc_path(); BUG_ON(!path); + path->leave_spinning = 1; ret = btrfs_drop_extents(trans, root, inode, file_pos, file_pos + num_bytes, file_pos, &hint); BUG_ON(ret); @@ -1474,6 +1476,10 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, btrfs_set_file_extent_compression(leaf, fi, compression); btrfs_set_file_extent_encryption(leaf, fi, encryption); btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); + + btrfs_unlock_up_safe(path, 1); + btrfs_set_lock_blocking(leaf); + btrfs_mark_buffer_dirty(leaf); inode_add_bytes(inode, num_bytes); @@ -1486,8 +1492,8 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, root->root_key.objectid, trans->transid, inode->i_ino, &ins); BUG_ON(ret); - btrfs_free_path(path); + return 0; } @@ -2118,6 +2124,7 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, path = btrfs_alloc_path(); BUG_ON(!path); + path->leave_spinning = 1; ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, 1); if (ret) { @@ -2164,6 +2171,7 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans, goto err; } + path->leave_spinning = 1; di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, name, name_len, -1); if (IS_ERR(di)) { @@ -2515,6 +2523,7 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, key.type = (u8)-1; search_again: + path->leave_spinning = 1; ret = btrfs_search_slot(trans, root, &key, path, -1, 1); if (ret < 0) goto error; @@ -2661,6 +2670,7 @@ delete: break; } if (found_extent) { + btrfs_set_path_blocking(path); ret = btrfs_free_extent(trans, root, extent_start, extent_num_bytes, leaf->start, root_owner, @@ -3466,6 +3476,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, sizes[0] = sizeof(struct btrfs_inode_item); sizes[1] = name_len + sizeof(*ref); + path->leave_spinning = 1; ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); if (ret != 0) goto fail; diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c index 6d8db2f5c38d..a5310c0f41e2 100644 --- a/fs/btrfs/locking.c +++ b/fs/btrfs/locking.c @@ -96,11 +96,12 @@ int btrfs_try_spin_lock(struct extent_buffer *eb) { int i; - spin_nested(eb); - if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) - return 1; - spin_unlock(&eb->lock); - + if (btrfs_spin_on_block(eb)) { + spin_nested(eb); + if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) + return 1; + spin_unlock(&eb->lock); + } /* spin for a bit on the BLOCKING flag */ for (i = 0; i < 2; i++) { cpu_relax(); diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 9c462fbd60fa..a93934fc93bd 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -203,7 +203,6 @@ static int process_one_buffer(struct btrfs_root *log, mutex_lock(&log->fs_info->pinned_mutex); btrfs_update_pinned_extents(log->fs_info->extent_root, eb->start, eb->len, 1); - mutex_unlock(&log->fs_info->pinned_mutex); } if (btrfs_buffer_uptodate(eb, gen)) { -- cgit v1.2.3 From 5d13a98f3bf5afc1113f7db184c627a44659bc29 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 13 Mar 2009 11:41:46 -0400 Subject: Btrfs: readahead checksums during btrfs_finish_ordered_io This reads in blocks in the checksum btree before starting the transaction in btrfs_finish_ordered_io. It makes it much more likely we'll be able to do operations inside the transaction without needing any btree reads, which limits transaction latencies overall. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index b83a45dc717e..9b4faac50c18 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -1497,6 +1497,30 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, return 0; } +/* + * helper function for btrfs_finish_ordered_io, this + * just reads in some of the csum leaves to prime them into ram + * before we start the transaction. It limits the amount of btree + * reads required while inside the transaction. + */ +static noinline void reada_csum(struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_ordered_extent *ordered_extent) +{ + struct btrfs_ordered_sum *sum; + u64 bytenr; + + sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, + list); + bytenr = sum->sums[0].bytenr; + + /* + * we don't care about the results, the point of this search is + * just to get the btree leaves into ram + */ + btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); +} + /* as ordered data IO finishes, this gets called so we can finish * an ordered extent if the range of bytes in the file it covers are * fully written. @@ -1505,7 +1529,7 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) { struct btrfs_root *root = BTRFS_I(inode)->root; struct btrfs_trans_handle *trans; - struct btrfs_ordered_extent *ordered_extent; + struct btrfs_ordered_extent *ordered_extent = NULL; struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; struct btrfs_path *path; int compressed = 0; @@ -1528,13 +1552,20 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino, start, 0); + ordered_extent = btrfs_lookup_ordered_extent(inode, + start); + if (!list_empty(&ordered_extent->list)) { + btrfs_release_path(root, path); + reada_csum(root, path, ordered_extent); + } btrfs_free_path(path); } } trans = btrfs_join_transaction(root, 1); - ordered_extent = btrfs_lookup_ordered_extent(inode, start); + if (!ordered_extent) + ordered_extent = btrfs_lookup_ordered_extent(inode, start); BUG_ON(!ordered_extent); if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) goto nocow; -- cgit v1.2.3 From 12fcfd22fe5bf4fe74710232098bc101af497995 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 24 Mar 2009 10:24:20 -0400 Subject: Btrfs: tree logging unlink/rename fixes The tree logging code allows individual files or directories to be logged without including operations on other files and directories in the FS. It tries to commit the minimal set of changes to disk in order to fsync the single file or directory that was sent to fsync or O_SYNC. The tree logging code was allowing files and directories to be unlinked if they were part of a rename operation where only one directory in the rename was in the fsync log. This patch adds a few new rules to the tree logging. 1) on rename or unlink, if the inode being unlinked isn't in the fsync log, we must force a full commit before doing an fsync of the directory where the unlink was done. The commit isn't done during the unlink, but it is forced the next time we try to log the parent directory. Solution: record transid of last unlink/rename per directory when the directory wasn't already logged. For renames this is only done when renaming to a different directory. mkdir foo/some_dir normal commit rename foo/some_dir foo2/some_dir mkdir foo/some_dir fsync foo/some_dir/some_file The fsync above will unlink the original some_dir without recording it in its new location (foo2). After a crash, some_dir will be gone unless the fsync of some_file forces a full commit 2) we must log any new names for any file or dir that is in the fsync log. This way we make sure not to lose files that are unlinked during the same transaction. 2a) we must log any new names for any file or dir during rename when the directory they are being removed from was logged. 2a is actually the more important variant. Without the extra logging a crash might unlink the old name without recreating the new one 3) after a crash, we must go through any directories with a link count of zero and redo the rm -rf mkdir f1/foo normal commit rm -rf f1/foo fsync(f1) The directory f1 was fully removed from the FS, but fsync was never called on f1, only its parent dir. After a crash the rm -rf must be replayed. This must be able to recurse down the entire directory tree. The inode link count fixup code takes care of the ugly details. Signed-off-by: Chris Mason --- fs/btrfs/btrfs_inode.h | 13 +- fs/btrfs/ctree.h | 7 +- fs/btrfs/extent-tree.c | 2 +- fs/btrfs/file.c | 14 +- fs/btrfs/inode.c | 28 +++- fs/btrfs/tree-log.c | 389 +++++++++++++++++++++++++++++++++++++++---------- fs/btrfs/tree-log.h | 17 ++- 7 files changed, 372 insertions(+), 98 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h index 72677ce2b74f..3af4cfb5654c 100644 --- a/fs/btrfs/btrfs_inode.h +++ b/fs/btrfs/btrfs_inode.h @@ -86,12 +86,6 @@ struct btrfs_inode { */ u64 logged_trans; - /* - * trans that last made a change that should be fully fsync'd. This - * gets reset to zero each time the inode is logged - */ - u64 log_dirty_trans; - /* total number of bytes pending delalloc, used by stat to calc the * real block usage of the file */ @@ -121,6 +115,13 @@ struct btrfs_inode { /* the start of block group preferred for allocations. */ u64 block_group; + /* the fsync log has some corner cases that mean we have to check + * directories to see if any unlinks have been done before + * the directory was logged. See tree-log.c for all the + * details + */ + u64 last_unlink_trans; + struct inode vfs_inode; }; diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 4ddce91cf3f9..2737facbd341 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -695,7 +695,12 @@ struct btrfs_fs_info { u64 generation; u64 last_trans_committed; - u64 last_trans_new_blockgroup; + + /* + * this is updated to the current trans every time a full commit + * is required instead of the faster short fsync log commits + */ + u64 last_trans_log_full_commit; u64 open_ioctl_trans; unsigned long mount_opt; u64 max_extent; diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 8933d15a240f..0c482e0d7c43 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -5897,7 +5897,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, extent_root = root->fs_info->extent_root; - root->fs_info->last_trans_new_blockgroup = trans->transid; + root->fs_info->last_trans_log_full_commit = trans->transid; cache = kzalloc(sizeof(*cache), GFP_NOFS); if (!cache) diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index f06c275644b7..32d10a617613 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -1173,8 +1173,11 @@ out_nolock: ret = btrfs_log_dentry_safe(trans, root, file->f_dentry); if (ret == 0) { - btrfs_sync_log(trans, root); - btrfs_end_transaction(trans, root); + ret = btrfs_sync_log(trans, root); + if (ret == 0) + btrfs_end_transaction(trans, root); + else + btrfs_commit_transaction(trans, root); } else { btrfs_commit_transaction(trans, root); } @@ -1266,8 +1269,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) if (ret > 0) { ret = btrfs_commit_transaction(trans, root); } else { - btrfs_sync_log(trans, root); - ret = btrfs_end_transaction(trans, root); + ret = btrfs_sync_log(trans, root); + if (ret == 0) + ret = btrfs_end_transaction(trans, root); + else + ret = btrfs_commit_transaction(trans, root); } mutex_lock(&dentry->d_inode->i_mutex); out: diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 9b4faac50c18..bffd79faffb5 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2246,8 +2246,6 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans, ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, dir->i_ino); BUG_ON(ret != 0 && ret != -ENOENT); - if (ret != -ENOENT) - BTRFS_I(dir)->log_dirty_trans = trans->transid; ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, index); @@ -2280,6 +2278,9 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry) trans = btrfs_start_transaction(root, 1); btrfs_set_trans_block_group(trans, dir); + + btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); + ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, dentry->d_name.name, dentry->d_name.len); @@ -3042,7 +3043,7 @@ static noinline void init_btrfs_i(struct inode *inode) bi->disk_i_size = 0; bi->flags = 0; bi->index_cnt = (u64)-1; - bi->log_dirty_trans = 0; + bi->last_unlink_trans = 0; extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping, GFP_NOFS); @@ -3786,6 +3787,8 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir, drop_inode = 1; nr = trans->blocks_used; + + btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); btrfs_end_transaction_throttle(trans, root); fail: if (drop_inode) { @@ -4666,6 +4669,15 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, trans = btrfs_start_transaction(root, 1); + /* + * this is an ugly little race, but the rename is required to make + * sure that if we crash, the inode is either at the old name + * or the new one. pinning the log transaction lets us make sure + * we don't allow a log commit to come in after we unlink the + * name but before we add the new name back in. + */ + btrfs_pin_log_trans(root); + btrfs_set_trans_block_group(trans, new_dir); btrfs_inc_nlink(old_dentry->d_inode); @@ -4673,6 +4685,9 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, new_dir->i_ctime = new_dir->i_mtime = ctime; old_inode->i_ctime = ctime; + if (old_dentry->d_parent != new_dentry->d_parent) + btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); + ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, old_dentry->d_name.name, old_dentry->d_name.len); @@ -4704,7 +4719,14 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, if (ret) goto out_fail; + btrfs_log_new_name(trans, old_inode, old_dir, + new_dentry->d_parent); out_fail: + + /* this btrfs_end_log_trans just allows the current + * log-sub transaction to complete + */ + btrfs_end_log_trans(root); btrfs_end_transaction_throttle(trans, root); out_unlock: return ret; diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 405439ca4c45..1b7f04a8f168 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -34,6 +34,49 @@ #define LOG_INODE_ALL 0 #define LOG_INODE_EXISTS 1 +/* + * directory trouble cases + * + * 1) on rename or unlink, if the inode being unlinked isn't in the fsync + * log, we must force a full commit before doing an fsync of the directory + * where the unlink was done. + * ---> record transid of last unlink/rename per directory + * + * mkdir foo/some_dir + * normal commit + * rename foo/some_dir foo2/some_dir + * mkdir foo/some_dir + * fsync foo/some_dir/some_file + * + * The fsync above will unlink the original some_dir without recording + * it in its new location (foo2). After a crash, some_dir will be gone + * unless the fsync of some_file forces a full commit + * + * 2) we must log any new names for any file or dir that is in the fsync + * log. ---> check inode while renaming/linking. + * + * 2a) we must log any new names for any file or dir during rename + * when the directory they are being removed from was logged. + * ---> check inode and old parent dir during rename + * + * 2a is actually the more important variant. With the extra logging + * a crash might unlink the old name without recreating the new one + * + * 3) after a crash, we must go through any directories with a link count + * of zero and redo the rm -rf + * + * mkdir f1/foo + * normal commit + * rm -rf f1/foo + * fsync(f1) + * + * The directory f1 was fully removed from the FS, but fsync was never + * called on f1, only its parent dir. After a crash the rm -rf must + * be replayed. This must be able to recurse down the entire + * directory tree. The inode link count fixup code takes care of the + * ugly details. + */ + /* * stages for the tree walking. The first * stage (0) is to only pin down the blocks we find @@ -47,12 +90,17 @@ #define LOG_WALK_REPLAY_INODES 1 #define LOG_WALK_REPLAY_ALL 2 -static int __btrfs_log_inode(struct btrfs_trans_handle *trans, +static int btrfs_log_inode(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, int inode_only); static int link_to_fixup_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, u64 objectid); +static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_root *log, + struct btrfs_path *path, + u64 dirid, int del_all); /* * tree logging is a special write ahead log used to make sure that @@ -132,11 +180,26 @@ static int join_running_log_trans(struct btrfs_root *root) return ret; } +/* + * This either makes the current running log transaction wait + * until you call btrfs_end_log_trans() or it makes any future + * log transactions wait until you call btrfs_end_log_trans() + */ +int btrfs_pin_log_trans(struct btrfs_root *root) +{ + int ret = -ENOENT; + + mutex_lock(&root->log_mutex); + atomic_inc(&root->log_writers); + mutex_unlock(&root->log_mutex); + return ret; +} + /* * indicate we're done making changes to the log tree * and wake up anyone waiting to do a sync */ -static int end_log_trans(struct btrfs_root *root) +int btrfs_end_log_trans(struct btrfs_root *root) { if (atomic_dec_and_test(&root->log_writers)) { smp_mb(); @@ -602,6 +665,7 @@ static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, ret = link_to_fixup_dir(trans, root, path, location.objectid); BUG_ON(ret); + ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); BUG_ON(ret); kfree(name); @@ -803,6 +867,7 @@ conflict_again: victim_name_len)) { btrfs_inc_nlink(inode); btrfs_release_path(root, path); + ret = btrfs_unlink_inode(trans, root, dir, inode, victim_name, victim_name_len); @@ -921,13 +986,20 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, key.offset--; btrfs_release_path(root, path); } - btrfs_free_path(path); + btrfs_release_path(root, path); if (nlink != inode->i_nlink) { inode->i_nlink = nlink; btrfs_update_inode(trans, root, inode); } BTRFS_I(inode)->index_cnt = (u64)-1; + if (inode->i_nlink == 0 && S_ISDIR(inode->i_mode)) { + ret = replay_dir_deletes(trans, root, NULL, path, + inode->i_ino, 1); + BUG_ON(ret); + } + btrfs_free_path(path); + return 0; } @@ -970,9 +1042,12 @@ static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, iput(inode); - if (key.offset == 0) - break; - key.offset--; + /* + * fixup on a directory may create new entries, + * make sure we always look for the highset possible + * offset + */ + key.offset = (u64)-1; } btrfs_release_path(root, path); return 0; @@ -1312,11 +1387,11 @@ again: read_extent_buffer(eb, name, (unsigned long)(di + 1), name_len); log_di = NULL; - if (dir_key->type == BTRFS_DIR_ITEM_KEY) { + if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { log_di = btrfs_lookup_dir_item(trans, log, log_path, dir_key->objectid, name, name_len, 0); - } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) { + } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { log_di = btrfs_lookup_dir_index_item(trans, log, log_path, dir_key->objectid, @@ -1377,7 +1452,7 @@ static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_root *log, struct btrfs_path *path, - u64 dirid) + u64 dirid, int del_all) { u64 range_start; u64 range_end; @@ -1407,10 +1482,14 @@ again: range_start = 0; range_end = 0; while (1) { - ret = find_dir_range(log, path, dirid, key_type, - &range_start, &range_end); - if (ret != 0) - break; + if (del_all) + range_end = (u64)-1; + else { + ret = find_dir_range(log, path, dirid, key_type, + &range_start, &range_end); + if (ret != 0) + break; + } dir_key.offset = range_start; while (1) { @@ -1436,7 +1515,8 @@ again: break; ret = check_item_in_log(trans, root, log, path, - log_path, dir, &found_key); + log_path, dir, + &found_key); BUG_ON(ret); if (found_key.offset == (u64)-1) break; @@ -1513,7 +1593,7 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, mode = btrfs_inode_mode(eb, inode_item); if (S_ISDIR(mode)) { ret = replay_dir_deletes(wc->trans, - root, log, path, key.objectid); + root, log, path, key.objectid, 0); BUG_ON(ret); } ret = overwrite_item(wc->trans, root, path, @@ -1850,7 +1930,8 @@ static int update_log_root(struct btrfs_trans_handle *trans, return ret; } -static int wait_log_commit(struct btrfs_root *root, unsigned long transid) +static int wait_log_commit(struct btrfs_trans_handle *trans, + struct btrfs_root *root, unsigned long transid) { DEFINE_WAIT(wait); int index = transid % 2; @@ -1864,9 +1945,12 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid) prepare_to_wait(&root->log_commit_wait[index], &wait, TASK_UNINTERRUPTIBLE); mutex_unlock(&root->log_mutex); - if (root->log_transid < transid + 2 && + + if (root->fs_info->last_trans_log_full_commit != + trans->transid && root->log_transid < transid + 2 && atomic_read(&root->log_commit[index])) schedule(); + finish_wait(&root->log_commit_wait[index], &wait); mutex_lock(&root->log_mutex); } while (root->log_transid < transid + 2 && @@ -1874,14 +1958,16 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid) return 0; } -static int wait_for_writer(struct btrfs_root *root) +static int wait_for_writer(struct btrfs_trans_handle *trans, + struct btrfs_root *root) { DEFINE_WAIT(wait); while (atomic_read(&root->log_writers)) { prepare_to_wait(&root->log_writer_wait, &wait, TASK_UNINTERRUPTIBLE); mutex_unlock(&root->log_mutex); - if (atomic_read(&root->log_writers)) + if (root->fs_info->last_trans_log_full_commit != + trans->transid && atomic_read(&root->log_writers)) schedule(); mutex_lock(&root->log_mutex); finish_wait(&root->log_writer_wait, &wait); @@ -1892,7 +1978,14 @@ static int wait_for_writer(struct btrfs_root *root) /* * btrfs_sync_log does sends a given tree log down to the disk and * updates the super blocks to record it. When this call is done, - * you know that any inodes previously logged are safely on disk + * you know that any inodes previously logged are safely on disk only + * if it returns 0. + * + * Any other return value means you need to call btrfs_commit_transaction. + * Some of the edge cases for fsyncing directories that have had unlinks + * or renames done in the past mean that sometimes the only safe + * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, + * that has happened. */ int btrfs_sync_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) @@ -1906,7 +1999,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, mutex_lock(&root->log_mutex); index1 = root->log_transid % 2; if (atomic_read(&root->log_commit[index1])) { - wait_log_commit(root, root->log_transid); + wait_log_commit(trans, root, root->log_transid); mutex_unlock(&root->log_mutex); return 0; } @@ -1914,18 +2007,26 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, /* wait for previous tree log sync to complete */ if (atomic_read(&root->log_commit[(index1 + 1) % 2])) - wait_log_commit(root, root->log_transid - 1); + wait_log_commit(trans, root, root->log_transid - 1); while (1) { unsigned long batch = root->log_batch; mutex_unlock(&root->log_mutex); schedule_timeout_uninterruptible(1); mutex_lock(&root->log_mutex); - wait_for_writer(root); + + wait_for_writer(trans, root); if (batch == root->log_batch) break; } + /* bail out if we need to do a full commit */ + if (root->fs_info->last_trans_log_full_commit == trans->transid) { + ret = -EAGAIN; + mutex_unlock(&root->log_mutex); + goto out; + } + ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); BUG_ON(ret); @@ -1961,16 +2062,29 @@ 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])) { - wait_log_commit(log_root_tree, log_root_tree->log_transid); + wait_log_commit(trans, log_root_tree, + log_root_tree->log_transid); mutex_unlock(&log_root_tree->log_mutex); goto out; } atomic_set(&log_root_tree->log_commit[index2], 1); - if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) - wait_log_commit(log_root_tree, log_root_tree->log_transid - 1); + if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { + wait_log_commit(trans, log_root_tree, + log_root_tree->log_transid - 1); + } + + wait_for_writer(trans, log_root_tree); - wait_for_writer(log_root_tree); + /* + * now that we've moved on to the tree of log tree roots, + * check the full commit flag again + */ + if (root->fs_info->last_trans_log_full_commit == trans->transid) { + mutex_unlock(&log_root_tree->log_mutex); + ret = -EAGAIN; + goto out_wake_log_root; + } ret = btrfs_write_and_wait_marked_extents(log_root_tree, &log_root_tree->dirty_log_pages); @@ -1995,7 +2109,9 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans, * in and cause problems either. */ write_ctree_super(trans, root->fs_info->tree_root, 2); + ret = 0; +out_wake_log_root: atomic_set(&log_root_tree->log_commit[index2], 0); smp_mb(); if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) @@ -2008,7 +2124,8 @@ out: return 0; } -/* * free all the extents used by the tree log. This should be called +/* + * free all the extents used by the tree log. This should be called * at commit time of the full transaction */ int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) @@ -2142,7 +2259,7 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, btrfs_free_path(path); mutex_unlock(&BTRFS_I(dir)->log_mutex); - end_log_trans(root); + btrfs_end_log_trans(root); return 0; } @@ -2169,7 +2286,7 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino, dirid, &index); mutex_unlock(&BTRFS_I(inode)->log_mutex); - end_log_trans(root); + btrfs_end_log_trans(root); return ret; } @@ -2569,7 +2686,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans, * * This handles both files and directories. */ -static int __btrfs_log_inode(struct btrfs_trans_handle *trans, +static int btrfs_log_inode(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, int inode_only) { @@ -2595,28 +2712,17 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans, min_key.offset = 0; max_key.objectid = inode->i_ino; + + /* today the code can only do partial logging of directories */ + if (!S_ISDIR(inode->i_mode)) + inode_only = LOG_INODE_ALL; + if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode)) max_key.type = BTRFS_XATTR_ITEM_KEY; else max_key.type = (u8)-1; max_key.offset = (u64)-1; - /* - * if this inode has already been logged and we're in inode_only - * mode, we don't want to delete the things that have already - * been written to the log. - * - * But, if the inode has been through an inode_only log, - * the logged_trans field is not set. This allows us to catch - * any new names for this inode in the backrefs by logging it - * again - */ - if (inode_only == LOG_INODE_EXISTS && - BTRFS_I(inode)->logged_trans == trans->transid) { - btrfs_free_path(path); - btrfs_free_path(dst_path); - goto out; - } mutex_lock(&BTRFS_I(inode)->log_mutex); /* @@ -2703,7 +2809,6 @@ next_slot: if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { btrfs_release_path(root, path); btrfs_release_path(log, dst_path); - BTRFS_I(inode)->log_dirty_trans = 0; ret = log_directory_changes(trans, root, inode, path, dst_path); BUG_ON(ret); } @@ -2712,19 +2817,58 @@ next_slot: btrfs_free_path(path); btrfs_free_path(dst_path); -out: return 0; } -int btrfs_log_inode(struct btrfs_trans_handle *trans, - struct btrfs_root *root, struct inode *inode, - int inode_only) +/* + * follow the dentry parent pointers up the chain and see if any + * of the directories in it require a full commit before they can + * be logged. Returns zero if nothing special needs to be done or 1 if + * a full commit is required. + */ +static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, + struct inode *inode, + struct dentry *parent, + struct super_block *sb, + u64 last_committed) { - int ret; + int ret = 0; + struct btrfs_root *root; - start_log_trans(trans, root); - ret = __btrfs_log_inode(trans, root, inode, inode_only); - end_log_trans(root); + if (!S_ISDIR(inode->i_mode)) { + if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) + goto out; + inode = parent->d_inode; + } + + while (1) { + BTRFS_I(inode)->logged_trans = trans->transid; + smp_mb(); + + if (BTRFS_I(inode)->last_unlink_trans > last_committed) { + root = BTRFS_I(inode)->root; + + /* + * make sure any commits to the log are forced + * to be full commits + */ + root->fs_info->last_trans_log_full_commit = + trans->transid; + ret = 1; + break; + } + + if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) + break; + + if (parent == sb->s_root) + break; + + parent = parent->d_parent; + inode = parent->d_inode; + + } +out: return ret; } @@ -2734,31 +2878,53 @@ int btrfs_log_inode(struct btrfs_trans_handle *trans, * only logging is done of any parent directories that are older than * the last committed transaction */ -int btrfs_log_dentry(struct btrfs_trans_handle *trans, - struct btrfs_root *root, struct dentry *dentry) +int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, struct inode *inode, + struct dentry *parent, int exists_only) { - int inode_only = LOG_INODE_ALL; + int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; struct super_block *sb; - int ret; + int ret = 0; + u64 last_committed = root->fs_info->last_trans_committed; + + sb = inode->i_sb; + + if (root->fs_info->last_trans_log_full_commit > + root->fs_info->last_trans_committed) { + ret = 1; + goto end_no_trans; + } + + ret = check_parent_dirs_for_sync(trans, inode, parent, + sb, last_committed); + if (ret) + goto end_no_trans; start_log_trans(trans, root); - sb = dentry->d_inode->i_sb; - while (1) { - ret = __btrfs_log_inode(trans, root, dentry->d_inode, - inode_only); - BUG_ON(ret); - inode_only = LOG_INODE_EXISTS; - dentry = dentry->d_parent; - if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb) + ret = btrfs_log_inode(trans, root, inode, inode_only); + BUG_ON(ret); + inode_only = LOG_INODE_EXISTS; + + while (1) { + if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) break; - if (BTRFS_I(dentry->d_inode)->generation <= - root->fs_info->last_trans_committed) + inode = parent->d_inode; + if (BTRFS_I(inode)->generation > + root->fs_info->last_trans_committed) { + ret = btrfs_log_inode(trans, root, inode, inode_only); + BUG_ON(ret); + } + if (parent == sb->s_root) break; + + parent = parent->d_parent; } - end_log_trans(root); - return 0; + ret = 0; + btrfs_end_log_trans(root); +end_no_trans: + return ret; } /* @@ -2770,12 +2936,8 @@ int btrfs_log_dentry(struct btrfs_trans_handle *trans, int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct dentry *dentry) { - u64 gen; - gen = root->fs_info->last_trans_new_blockgroup; - if (gen > root->fs_info->last_trans_committed) - return 1; - else - return btrfs_log_dentry(trans, root, dentry); + return btrfs_log_inode_parent(trans, root, dentry->d_inode, + dentry->d_parent, 0); } /* @@ -2894,3 +3056,74 @@ again: kfree(log_root_tree); return 0; } + +/* + * there are some corner cases where we want to force a full + * commit instead of allowing a directory to be logged. + * + * They revolve around files there were unlinked from the directory, and + * this function updates the parent directory so that a full commit is + * properly done if it is fsync'd later after the unlinks are done. + */ +void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, + struct inode *dir, struct inode *inode, + int for_rename) +{ + /* + * if this directory was already logged any new + * names for this file/dir will get recorded + */ + smp_mb(); + if (BTRFS_I(dir)->logged_trans == trans->transid) + return; + + /* + * if the inode we're about to unlink was logged, + * the log will be properly updated for any new names + */ + if (BTRFS_I(inode)->logged_trans == trans->transid) + return; + + /* + * when renaming files across directories, if the directory + * there we're unlinking from gets fsync'd later on, there's + * no way to find the destination directory later and fsync it + * properly. So, we have to be conservative and force commits + * so the new name gets discovered. + */ + if (for_rename) + goto record; + + /* we can safely do the unlink without any special recording */ + return; + +record: + BTRFS_I(dir)->last_unlink_trans = trans->transid; +} + +/* + * Call this after adding a new name for a file and it will properly + * update the log to reflect the new name. + * + * It will return zero if all goes well, and it will return 1 if a + * full transaction commit is required. + */ +int btrfs_log_new_name(struct btrfs_trans_handle *trans, + struct inode *inode, struct inode *old_dir, + struct dentry *parent) +{ + struct btrfs_root * root = BTRFS_I(inode)->root; + + /* + * if this inode hasn't been logged and directory we're renaming it + * from hasn't been logged, we don't need to log it + */ + if (BTRFS_I(inode)->logged_trans <= + root->fs_info->last_trans_committed && + (!old_dir || BTRFS_I(old_dir)->logged_trans <= + root->fs_info->last_trans_committed)) + return 0; + + return btrfs_log_inode_parent(trans, root, inode, parent, 1); +} + diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h index b9409b32ed02..d09c7609e16b 100644 --- a/fs/btrfs/tree-log.h +++ b/fs/btrfs/tree-log.h @@ -22,14 +22,9 @@ 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); -int btrfs_log_dentry(struct btrfs_trans_handle *trans, - struct btrfs_root *root, struct dentry *dentry); int btrfs_recover_log_trees(struct btrfs_root *tree_root); int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct dentry *dentry); -int btrfs_log_inode(struct btrfs_trans_handle *trans, - struct btrfs_root *root, struct inode *inode, - int inode_only); int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, struct btrfs_root *root, const char *name, int name_len, @@ -38,4 +33,16 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, struct btrfs_root *root, const char *name, int name_len, struct inode *inode, u64 dirid); +int btrfs_join_running_log_trans(struct btrfs_root *root); +int btrfs_end_log_trans(struct btrfs_root *root); +int btrfs_pin_log_trans(struct btrfs_root *root); +int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, struct inode *inode, + struct dentry *parent, int exists_only); +void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, + struct inode *dir, struct inode *inode, + int for_rename); +int btrfs_log_new_name(struct btrfs_trans_handle *trans, + struct inode *inode, struct inode *old_dir, + struct dentry *parent); #endif -- cgit v1.2.3 From 5a3f23d515a2ebf0c750db80579ca57b28cbce6d Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 31 Mar 2009 13:27:11 -0400 Subject: Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason --- fs/btrfs/btrfs_inode.h | 18 ++++++++ fs/btrfs/ctree.h | 35 ++++++++++++++ fs/btrfs/disk-io.c | 2 + fs/btrfs/file.c | 26 +++++++++++ fs/btrfs/inode.c | 81 ++++++++++++++++++++++++++++++--- fs/btrfs/ordered-data.c | 118 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/btrfs/ordered-data.h | 4 ++ fs/btrfs/transaction.c | 11 +++++ 8 files changed, 288 insertions(+), 7 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h index 3af4cfb5654c..b30986f00b9d 100644 --- a/fs/btrfs/btrfs_inode.h +++ b/fs/btrfs/btrfs_inode.h @@ -66,6 +66,12 @@ struct btrfs_inode { */ struct list_head delalloc_inodes; + /* + * list for tracking inodes that must be sent to disk before a + * rename or truncate commit + */ + struct list_head ordered_operations; + /* the space_info for where this inode's data allocations are done */ struct btrfs_space_info *space_info; @@ -122,6 +128,18 @@ struct btrfs_inode { */ u64 last_unlink_trans; + /* + * ordered_data_close is set by truncate when a file that used + * to have good data has been truncated to zero. When it is set + * the btrfs file release call will add this inode to the + * ordered operations list so that we make sure to flush out any + * new data the application may have written before commit. + * + * yes, its silly to have a single bitflag, but we might grow more + * of these. + */ + unsigned ordered_data_close:1; + struct inode vfs_inode; }; diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 2737facbd341..f48905ee5240 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -45,6 +45,13 @@ struct btrfs_ordered_sum; #define BTRFS_MAX_LEVEL 8 +/* + * files bigger than this get some pre-flushing when they are added + * to the ordered operations list. That way we limit the total + * work done by the commit + */ +#define BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT (8 * 1024 * 1024) + /* holds pointers to all of the tree roots */ #define BTRFS_ROOT_TREE_OBJECTID 1ULL @@ -727,6 +734,15 @@ struct btrfs_fs_info { struct mutex volume_mutex; struct mutex tree_reloc_mutex; + /* + * this protects the ordered operations list only while we are + * processing all of the entries on it. This way we make + * sure the commit code doesn't find the list temporarily empty + * because another function happens to be doing non-waiting preflush + * before jumping into the main commit. + */ + struct mutex ordered_operations_mutex; + struct list_head trans_list; struct list_head hashers; struct list_head dead_roots; @@ -741,9 +757,28 @@ struct btrfs_fs_info { * ordered extents */ spinlock_t ordered_extent_lock; + + /* + * all of the data=ordered extents pending writeback + * these can span multiple transactions and basically include + * every dirty data page that isn't from nodatacow + */ struct list_head ordered_extents; + + /* + * all of the inodes that have delalloc bytes. It is possible for + * this list to be empty even when there is still dirty data=ordered + * extents waiting to finish IO. + */ struct list_head delalloc_inodes; + /* + * special rename and truncate targets that must be on disk before + * we're allowed to commit. This is basically the ext3 style + * data=ordered list. + */ + struct list_head ordered_operations; + /* * there is a pool of worker threads for checksumming during writes * and a pool for checksumming after reads. This is because readers diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 9244cd7313d4..1747dfd18654 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1572,6 +1572,7 @@ struct btrfs_root *open_ctree(struct super_block *sb, INIT_LIST_HEAD(&fs_info->dead_roots); INIT_LIST_HEAD(&fs_info->hashers); INIT_LIST_HEAD(&fs_info->delalloc_inodes); + INIT_LIST_HEAD(&fs_info->ordered_operations); spin_lock_init(&fs_info->delalloc_lock); spin_lock_init(&fs_info->new_trans_lock); spin_lock_init(&fs_info->ref_cache_lock); @@ -1643,6 +1644,7 @@ struct btrfs_root *open_ctree(struct super_block *sb, insert_inode_hash(fs_info->btree_inode); mutex_init(&fs_info->trans_mutex); + mutex_init(&fs_info->ordered_operations_mutex); mutex_init(&fs_info->tree_log_mutex); mutex_init(&fs_info->drop_mutex); mutex_init(&fs_info->pinned_mutex); diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index 32d10a617613..9c9fb46ccd08 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -1161,6 +1161,20 @@ out_nolock: page_cache_release(pinned[1]); *ppos = pos; + /* + * we want to make sure fsync finds this change + * but we haven't joined a transaction running right now. + * + * Later on, someone is sure to update the inode and get the + * real transid recorded. + * + * We set last_trans now to the fs_info generation + 1, + * this will either be one more than the running transaction + * or the generation used for the next transaction if there isn't + * one running right now. + */ + BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; + if (num_written > 0 && will_write) { struct btrfs_trans_handle *trans; @@ -1194,6 +1208,18 @@ out_nolock: int btrfs_release_file(struct inode *inode, struct file *filp) { + /* + * ordered_data_close is set by settattr when we are about to truncate + * a file from a non-zero size to a zero size. This tries to + * flush down new bytes that may have been written if the + * application were using truncate to replace a file in place. + */ + if (BTRFS_I(inode)->ordered_data_close) { + BTRFS_I(inode)->ordered_data_close = 0; + btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); + if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) + filemap_flush(inode->i_mapping); + } if (filp->private_data) btrfs_ioctl_trans_end(filp); return 0; diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index bffd79faffb5..1cff528d5b51 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2907,11 +2907,21 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) if (err) return err; - if (S_ISREG(inode->i_mode) && - attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) { - err = btrfs_cont_expand(inode, attr->ia_size); - if (err) - return err; + if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { + if (attr->ia_size > inode->i_size) { + err = btrfs_cont_expand(inode, attr->ia_size); + if (err) + return err; + } else if (inode->i_size > 0 && + attr->ia_size == 0) { + + /* we're truncating a file that used to have good + * data down to zero. Make sure it gets into + * the ordered flush list so that any new writes + * get down to disk quickly. + */ + BTRFS_I(inode)->ordered_data_close = 1; + } } err = inode_setattr(inode, attr); @@ -3050,6 +3060,7 @@ static noinline void init_btrfs_i(struct inode *inode) extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, inode->i_mapping, GFP_NOFS); INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); + INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); mutex_init(&BTRFS_I(inode)->extent_mutex); mutex_init(&BTRFS_I(inode)->log_mutex); @@ -4419,6 +4430,8 @@ again: } ClearPageChecked(page); set_page_dirty(page); + + BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; unlock_extent(io_tree, page_start, page_end, GFP_NOFS); out_unlock: @@ -4444,6 +4457,27 @@ static void btrfs_truncate(struct inode *inode) btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); trans = btrfs_start_transaction(root, 1); + + /* + * setattr is responsible for setting the ordered_data_close flag, + * but that is only tested during the last file release. That + * could happen well after the next commit, leaving a great big + * window where new writes may get lost if someone chooses to write + * to this file after truncating to zero + * + * The inode doesn't have any dirty data here, and so if we commit + * this is a noop. If someone immediately starts writing to the inode + * it is very likely we'll catch some of their writes in this + * transaction, and the commit will find this file on the ordered + * data list with good things to send down. + * + * This is a best effort solution, there is still a window where + * using truncate to replace the contents of the file will + * end up with a zero length file after a crash. + */ + if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) + btrfs_add_ordered_operation(trans, root, inode); + btrfs_set_trans_block_group(trans, inode); btrfs_i_size_write(inode, inode->i_size); @@ -4520,12 +4554,15 @@ struct inode *btrfs_alloc_inode(struct super_block *sb) ei->i_acl = BTRFS_ACL_NOT_CACHED; ei->i_default_acl = BTRFS_ACL_NOT_CACHED; INIT_LIST_HEAD(&ei->i_orphan); + INIT_LIST_HEAD(&ei->ordered_operations); return &ei->vfs_inode; } void btrfs_destroy_inode(struct inode *inode) { struct btrfs_ordered_extent *ordered; + struct btrfs_root *root = BTRFS_I(inode)->root; + WARN_ON(!list_empty(&inode->i_dentry)); WARN_ON(inode->i_data.nrpages); @@ -4536,13 +4573,24 @@ void btrfs_destroy_inode(struct inode *inode) BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED) posix_acl_release(BTRFS_I(inode)->i_default_acl); - spin_lock(&BTRFS_I(inode)->root->list_lock); + /* + * Make sure we're properly removed from the ordered operation + * lists. + */ + smp_mb(); + if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { + spin_lock(&root->fs_info->ordered_extent_lock); + list_del_init(&BTRFS_I(inode)->ordered_operations); + spin_unlock(&root->fs_info->ordered_extent_lock); + } + + spin_lock(&root->list_lock); if (!list_empty(&BTRFS_I(inode)->i_orphan)) { printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" " list\n", inode->i_ino); dump_stack(); } - spin_unlock(&BTRFS_I(inode)->root->list_lock); + spin_unlock(&root->list_lock); while (1) { ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); @@ -4667,8 +4715,27 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, if (ret) goto out_unlock; + /* + * we're using rename to replace one file with another. + * and the replacement file is large. Start IO on it now so + * we don't add too much work to the end of the transaction + */ + if (new_inode && old_inode && S_ISREG(old_inode->i_mode) && + new_inode->i_size && + old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) + filemap_flush(old_inode->i_mapping); + trans = btrfs_start_transaction(root, 1); + /* + * make sure the inode gets flushed if it is replacing + * something. + */ + if (new_inode && new_inode->i_size && + old_inode && S_ISREG(old_inode->i_mode)) { + btrfs_add_ordered_operation(trans, root, old_inode); + } + /* * this is an ugly little race, but the rename is required to make * sure that if we crash, the inode is either at the old name diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index 77c2411a5f0f..53c87b197d70 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -310,6 +310,16 @@ int btrfs_remove_ordered_extent(struct inode *inode, spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); list_del_init(&entry->root_extent_list); + + /* + * we have no more ordered extents for this inode and + * no dirty pages. We can safely remove it from the + * list of ordered extents + */ + if (RB_EMPTY_ROOT(&tree->tree) && + !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { + list_del_init(&BTRFS_I(inode)->ordered_operations); + } spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); mutex_unlock(&tree->mutex); @@ -369,6 +379,68 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) return 0; } +/* + * this is used during transaction commit to write all the inodes + * added to the ordered operation list. These files must be fully on + * disk before the transaction commits. + * + * we have two modes here, one is to just start the IO via filemap_flush + * and the other is to wait for all the io. When we wait, we have an + * extra check to make sure the ordered operation list really is empty + * before we return + */ +int btrfs_run_ordered_operations(struct btrfs_root *root, int wait) +{ + struct btrfs_inode *btrfs_inode; + struct inode *inode; + struct list_head splice; + + INIT_LIST_HEAD(&splice); + + mutex_lock(&root->fs_info->ordered_operations_mutex); + spin_lock(&root->fs_info->ordered_extent_lock); +again: + list_splice_init(&root->fs_info->ordered_operations, &splice); + + while (!list_empty(&splice)) { + btrfs_inode = list_entry(splice.next, struct btrfs_inode, + ordered_operations); + + inode = &btrfs_inode->vfs_inode; + + list_del_init(&btrfs_inode->ordered_operations); + + /* + * the inode may be getting freed (in sys_unlink path). + */ + inode = igrab(inode); + + if (!wait && inode) { + list_add_tail(&BTRFS_I(inode)->ordered_operations, + &root->fs_info->ordered_operations); + } + spin_unlock(&root->fs_info->ordered_extent_lock); + + if (inode) { + if (wait) + btrfs_wait_ordered_range(inode, 0, (u64)-1); + else + filemap_flush(inode->i_mapping); + iput(inode); + } + + cond_resched(); + spin_lock(&root->fs_info->ordered_extent_lock); + } + if (wait && !list_empty(&root->fs_info->ordered_operations)) + goto again; + + spin_unlock(&root->fs_info->ordered_extent_lock); + mutex_unlock(&root->fs_info->ordered_operations_mutex); + + return 0; +} + /* * Used to start IO or wait for a given ordered extent to finish. * @@ -726,3 +798,49 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping, return ret; } + +/* + * add a given inode to the list of inodes that must be fully on + * disk before a transaction commit finishes. + * + * This basically gives us the ext3 style data=ordered mode, and it is mostly + * used to make sure renamed files are fully on disk. + * + * It is a noop if the inode is already fully on disk. + * + * If trans is not null, we'll do a friendly check for a transaction that + * is already flushing things and force the IO down ourselves. + */ +int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct inode *inode) +{ + u64 last_mod; + + last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); + + /* + * if this file hasn't been changed since the last transaction + * commit, we can safely return without doing anything + */ + if (last_mod < root->fs_info->last_trans_committed) + return 0; + + /* + * the transaction is already committing. Just start the IO and + * don't bother with all of this list nonsense + */ + if (trans && root->fs_info->running_transaction->blocked) { + btrfs_wait_ordered_range(inode, 0, (u64)-1); + return 0; + } + + spin_lock(&root->fs_info->ordered_extent_lock); + if (list_empty(&BTRFS_I(inode)->ordered_operations)) { + list_add_tail(&BTRFS_I(inode)->ordered_operations, + &root->fs_info->ordered_operations); + } + spin_unlock(&root->fs_info->ordered_extent_lock); + + return 0; +} diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h index ab66d5e8d6d6..3d31c8827b01 100644 --- a/fs/btrfs/ordered-data.h +++ b/fs/btrfs/ordered-data.h @@ -155,4 +155,8 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping, int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start, loff_t end, int sync_mode); int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only); +int btrfs_run_ordered_operations(struct btrfs_root *root, int wait); +int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct inode *inode); #endif diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c index 9c8f158dd2db..664782c6a2df 100644 --- a/fs/btrfs/transaction.c +++ b/fs/btrfs/transaction.c @@ -975,6 +975,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans, int should_grow = 0; unsigned long now = get_seconds(); + btrfs_run_ordered_operations(root, 0); + /* make a pass through all the delayed refs we have so far * any runnings procs may add more while we are here */ @@ -1056,6 +1058,15 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans, BUG_ON(ret); } + /* + * rename don't use btrfs_join_transaction, so, once we + * set the transaction to blocked above, we aren't going + * to get any new ordered operations. We can safely run + * it here and no for sure that nothing new will be added + * to the list + */ + btrfs_run_ordered_operations(root, 1); + smp_mb(); if (cur_trans->num_writers > 1 || should_grow) schedule_timeout(timeout); -- cgit v1.2.3 From c2ec175c39f62949438354f603f4aa170846aabb Mon Sep 17 00:00:00 2001 From: Nick Piggin Date: Tue, 31 Mar 2009 15:23:21 -0700 Subject: mm: page_mkwrite change prototype to match fault Change the page_mkwrite prototype to take a struct vm_fault, and return VM_FAULT_xxx flags. There should be no functional change. This makes it possible to return much more detailed error information to the VM (and also can provide more information eg. virtual_address to the driver, which might be important in some special cases). This is required for a subsequent fix. And will also make it easier to merge page_mkwrite() with fault() in future. Signed-off-by: Nick Piggin Cc: Chris Mason Cc: Trond Myklebust Cc: Miklos Szeredi Cc: Steven Whitehouse Cc: Mark Fasheh Cc: Joel Becker Cc: Artem Bityutskiy Cc: Felix Blyakher Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- Documentation/filesystems/Locking | 2 +- drivers/video/fb_defio.c | 3 ++- fs/btrfs/ctree.h | 2 +- fs/btrfs/inode.c | 5 ++++- fs/buffer.c | 6 +++++- fs/ext4/ext4.h | 2 +- fs/ext4/inode.c | 5 ++++- fs/fuse/file.c | 3 ++- fs/gfs2/ops_file.c | 5 ++++- fs/nfs/file.c | 5 ++++- fs/ocfs2/mmap.c | 6 ++++-- fs/ubifs/file.c | 9 ++++++--- fs/xfs/linux-2.6/xfs_file.c | 4 ++-- include/linux/buffer_head.h | 2 +- include/linux/mm.h | 3 ++- mm/memory.c | 26 ++++++++++++++++++++++---- 16 files changed, 65 insertions(+), 23 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking index 4e78ce677843..76efe5b71d7d 100644 --- a/Documentation/filesystems/Locking +++ b/Documentation/filesystems/Locking @@ -505,7 +505,7 @@ prototypes: void (*open)(struct vm_area_struct*); void (*close)(struct vm_area_struct*); int (*fault)(struct vm_area_struct*, struct vm_fault *); - int (*page_mkwrite)(struct vm_area_struct *, struct page *); + int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *); int (*access)(struct vm_area_struct *, unsigned long, void*, int, int); locking rules: diff --git a/drivers/video/fb_defio.c b/drivers/video/fb_defio.c index 082026546aee..0a7a6679ee6e 100644 --- a/drivers/video/fb_defio.c +++ b/drivers/video/fb_defio.c @@ -85,8 +85,9 @@ EXPORT_SYMBOL_GPL(fb_deferred_io_fsync); /* vm_ops->page_mkwrite handler */ static int fb_deferred_io_mkwrite(struct vm_area_struct *vma, - struct page *page) + struct vm_fault *vmf) { + struct page *page = vmf->page; struct fb_info *info = vma->vm_private_data; struct fb_deferred_io *fbdefio = info->fbdefio; struct page *cur; diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 5e1d4e30e9d8..7dd1b6d0bf32 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -2060,7 +2060,7 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset, unsigned long btrfs_force_ra(struct address_space *mapping, struct file_ra_state *ra, struct file *file, pgoff_t offset, pgoff_t last_index); -int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page); +int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf); int btrfs_readpage(struct file *file, struct page *page); void btrfs_delete_inode(struct inode *inode); void btrfs_put_inode(struct inode *inode); diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 7d4f948bc22a..ec5423790bbb 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4292,8 +4292,9 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) * beyond EOF, then the page is guaranteed safe against truncation until we * unlock the page. */ -int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) +int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; struct inode *inode = fdentry(vma->vm_file)->d_inode; struct btrfs_root *root = BTRFS_I(inode)->root; struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; @@ -4362,6 +4363,8 @@ again: out_unlock: unlock_page(page); out: + if (ret) + ret = VM_FAULT_SIGBUS; return ret; } diff --git a/fs/buffer.c b/fs/buffer.c index 73abe6d8218c..6d51a3da362c 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -2313,9 +2313,10 @@ int block_commit_write(struct page *page, unsigned from, unsigned to) * unlock the page. */ int -block_page_mkwrite(struct vm_area_struct *vma, struct page *page, +block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, get_block_t get_block) { + struct page *page = vmf->page; struct inode *inode = vma->vm_file->f_path.dentry->d_inode; unsigned long end; loff_t size; @@ -2340,6 +2341,9 @@ block_page_mkwrite(struct vm_area_struct *vma, struct page *page, ret = block_commit_write(page, 0, end); out_unlock: + if (ret) + ret = VM_FAULT_SIGBUS; + unlock_page(page); return ret; } diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h index 6083bb38057b..990c94000924 100644 --- a/fs/ext4/ext4.h +++ b/fs/ext4/ext4.h @@ -1098,7 +1098,7 @@ extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int idxblocks); extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks); extern int ext4_block_truncate_page(handle_t *handle, struct address_space *mapping, loff_t from); -extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page); +extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf); extern qsize_t ext4_get_reserved_space(struct inode *inode); /* ioctl.c */ diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c index 71d3ecd5db79..dd82ff390067 100644 --- a/fs/ext4/inode.c +++ b/fs/ext4/inode.c @@ -5146,8 +5146,9 @@ static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) return !buffer_mapped(bh); } -int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page) +int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; loff_t size; unsigned long len; int ret = -EINVAL; @@ -5199,6 +5200,8 @@ int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page) goto out_unlock; ret = 0; out_unlock: + if (ret) + ret = VM_FAULT_SIGBUS; up_read(&inode->i_alloc_sem); return ret; } diff --git a/fs/fuse/file.c b/fs/fuse/file.c index 821d10f719bd..4e340fedf768 100644 --- a/fs/fuse/file.c +++ b/fs/fuse/file.c @@ -1234,8 +1234,9 @@ static void fuse_vma_close(struct vm_area_struct *vma) * - sync(2) * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER */ -static int fuse_page_mkwrite(struct vm_area_struct *vma, struct page *page) +static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; /* * Don't use page->mapping as it may become NULL from a * concurrent truncate. diff --git a/fs/gfs2/ops_file.c b/fs/gfs2/ops_file.c index 3b9e8de3500b..70b9b8548945 100644 --- a/fs/gfs2/ops_file.c +++ b/fs/gfs2/ops_file.c @@ -337,8 +337,9 @@ static int gfs2_allocate_page_backing(struct page *page) * blocks allocated on disk to back that page. */ -static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page) +static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; struct inode *inode = vma->vm_file->f_path.dentry->d_inode; struct gfs2_inode *ip = GFS2_I(inode); struct gfs2_sbd *sdp = GFS2_SB(inode); @@ -412,6 +413,8 @@ out_unlock: gfs2_glock_dq(&gh); out: gfs2_holder_uninit(&gh); + if (ret) + ret = VM_FAULT_SIGBUS; return ret; } diff --git a/fs/nfs/file.c b/fs/nfs/file.c index 90f292b520d2..cec79392e4ba 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -451,8 +451,9 @@ const struct address_space_operations nfs_file_aops = { .launder_page = nfs_launder_page, }; -static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) +static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; struct file *filp = vma->vm_file; struct dentry *dentry = filp->f_path.dentry; unsigned pagelen; @@ -483,6 +484,8 @@ static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) ret = pagelen; out_unlock: unlock_page(page); + if (ret) + ret = VM_FAULT_SIGBUS; return ret; } diff --git a/fs/ocfs2/mmap.c b/fs/ocfs2/mmap.c index eea1d24713ea..b606496b72ec 100644 --- a/fs/ocfs2/mmap.c +++ b/fs/ocfs2/mmap.c @@ -154,8 +154,9 @@ out: return ret; } -static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page) +static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; struct inode *inode = vma->vm_file->f_path.dentry->d_inode; struct buffer_head *di_bh = NULL; sigset_t blocked, oldset; @@ -196,7 +197,8 @@ out: ret2 = ocfs2_vm_op_unblock_sigs(&oldset); if (ret2 < 0) mlog_errno(ret2); - + if (ret) + ret = VM_FAULT_SIGBUS; return ret; } diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c index 93b6de51f261..0ff89fe71e51 100644 --- a/fs/ubifs/file.c +++ b/fs/ubifs/file.c @@ -1434,8 +1434,9 @@ static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags) * mmap()d file has taken write protection fault and is being made * writable. UBIFS must ensure page is budgeted for. */ -static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) +static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) { + struct page *page = vmf->page; struct inode *inode = vma->vm_file->f_path.dentry->d_inode; struct ubifs_info *c = inode->i_sb->s_fs_info; struct timespec now = ubifs_current_time(inode); @@ -1447,7 +1448,7 @@ static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) ubifs_assert(!(inode->i_sb->s_flags & MS_RDONLY)); if (unlikely(c->ro_media)) - return -EROFS; + return VM_FAULT_SIGBUS; /* -EROFS */ /* * We have not locked @page so far so we may budget for changing the @@ -1480,7 +1481,7 @@ static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) if (err == -ENOSPC) ubifs_warn("out of space for mmapped file " "(inode number %lu)", inode->i_ino); - return err; + return VM_FAULT_SIGBUS; } lock_page(page); @@ -1520,6 +1521,8 @@ static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) out_unlock: unlock_page(page); ubifs_release_budget(c, &req); + if (err) + err = VM_FAULT_SIGBUS; return err; } diff --git a/fs/xfs/linux-2.6/xfs_file.c b/fs/xfs/linux-2.6/xfs_file.c index e14c4e3aea0c..f4e255441574 100644 --- a/fs/xfs/linux-2.6/xfs_file.c +++ b/fs/xfs/linux-2.6/xfs_file.c @@ -234,9 +234,9 @@ xfs_file_mmap( STATIC int xfs_vm_page_mkwrite( struct vm_area_struct *vma, - struct page *page) + struct vm_fault *vmf) { - return block_page_mkwrite(vma, page, xfs_get_blocks); + return block_page_mkwrite(vma, vmf, xfs_get_blocks); } const struct file_operations xfs_file_operations = { diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h index f19fd9045ea0..3d7bcde2e332 100644 --- a/include/linux/buffer_head.h +++ b/include/linux/buffer_head.h @@ -216,7 +216,7 @@ int cont_write_begin(struct file *, struct address_space *, loff_t, get_block_t *, loff_t *); int generic_cont_expand_simple(struct inode *inode, loff_t size); int block_commit_write(struct page *page, unsigned from, unsigned to); -int block_page_mkwrite(struct vm_area_struct *vma, struct page *page, +int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, get_block_t get_block); void block_sync_page(struct page *); sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *); diff --git a/include/linux/mm.h b/include/linux/mm.h index 2223f8dfa568..aeabe953ba4f 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -135,6 +135,7 @@ extern pgprot_t protection_map[16]; #define FAULT_FLAG_WRITE 0x01 /* Fault was a write access */ #define FAULT_FLAG_NONLINEAR 0x02 /* Fault was via a nonlinear mapping */ +#define FAULT_FLAG_MKWRITE 0x04 /* Fault was mkwrite of existing pte */ /* * This interface is used by x86 PAT code to identify a pfn mapping that is @@ -187,7 +188,7 @@ struct vm_operations_struct { /* notification that a previously read-only page is about to become * writable, if an error is returned it will cause a SIGBUS */ - int (*page_mkwrite)(struct vm_area_struct *vma, struct page *page); + int (*page_mkwrite)(struct vm_area_struct *vma, struct vm_fault *vmf); /* called by access_process_vm when get_user_pages() fails, typically * for use by special VMAs that can switch between memory and hardware diff --git a/mm/memory.c b/mm/memory.c index 5b4ad5e4f98d..cf6873e91c6a 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -1945,6 +1945,15 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, * get_user_pages(.write=1, .force=1). */ if (vma->vm_ops && vma->vm_ops->page_mkwrite) { + struct vm_fault vmf; + int tmp; + + vmf.virtual_address = (void __user *)(address & + PAGE_MASK); + vmf.pgoff = old_page->index; + vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; + vmf.page = old_page; + /* * Notify the address space that the page is about to * become writable so that it can prohibit this or wait @@ -1956,8 +1965,12 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, page_cache_get(old_page); pte_unmap_unlock(page_table, ptl); - if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) + tmp = vma->vm_ops->page_mkwrite(vma, &vmf); + if (unlikely(tmp & + (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { + ret = tmp; goto unwritable_page; + } /* * Since we dropped the lock we need to revalidate @@ -2106,7 +2119,7 @@ oom: unwritable_page: page_cache_release(old_page); - return VM_FAULT_SIGBUS; + return ret; } /* @@ -2648,9 +2661,14 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, * to become writable */ if (vma->vm_ops->page_mkwrite) { + int tmp; + unlock_page(page); - if (vma->vm_ops->page_mkwrite(vma, page) < 0) { - ret = VM_FAULT_SIGBUS; + vmf.flags |= FAULT_FLAG_MKWRITE; + tmp = vma->vm_ops->page_mkwrite(vma, &vmf); + if (unlikely(tmp & + (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { + ret = tmp; anon = 1; /* no anon but release vmf.page */ goto out_unlocked; } -- cgit v1.2.3 From 56a76f8275c379ed73c8a43cfa1dfa2f5e9cfa19 Mon Sep 17 00:00:00 2001 From: Nick Piggin Date: Tue, 31 Mar 2009 15:23:23 -0700 Subject: fs: fix page_mkwrite error cases in core code and btrfs page_mkwrite is called with neither the page lock nor the ptl held. This means a page can be concurrently truncated or invalidated out from underneath it. Callers are supposed to prevent truncate races themselves, however previously the only thing they can do in case they hit one is to raise a SIGBUS. A sigbus is wrong for the case that the page has been invalidated or truncated within i_size (eg. hole punched). Callers may also have to perform memory allocations in this path, where again, SIGBUS would be wrong. The previous patch ("mm: page_mkwrite change prototype to match fault") made it possible to properly specify errors. Convert the generic buffer.c code and btrfs to return sane error values (in the case of page removed from pagecache, VM_FAULT_NOPAGE will cause the fault handler to exit without doing anything, and the fault will be retried properly). This fixes core code, and converts btrfs as a template/example. All other filesystems defining their own page_mkwrite should be fixed in a similar manner. Acked-by: Chris Mason Signed-off-by: Nick Piggin Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- fs/btrfs/inode.c | 11 +++++++---- fs/buffer.c | 12 ++++++++---- 2 files changed, 15 insertions(+), 8 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index ec5423790bbb..17e608c4dc70 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4307,10 +4307,15 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) u64 page_end; ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); - if (ret) + if (ret) { + if (ret == -ENOMEM) + ret = VM_FAULT_OOM; + else /* -ENOSPC, -EIO, etc */ + ret = VM_FAULT_SIGBUS; goto out; + } - ret = -EINVAL; + ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ again: lock_page(page); size = i_size_read(inode); @@ -4363,8 +4368,6 @@ again: out_unlock: unlock_page(page); out: - if (ret) - ret = VM_FAULT_SIGBUS; return ret; } diff --git a/fs/buffer.c b/fs/buffer.c index 6d51a3da362c..0c14f8d52ee5 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -2320,7 +2320,7 @@ block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, struct inode *inode = vma->vm_file->f_path.dentry->d_inode; unsigned long end; loff_t size; - int ret = -EINVAL; + int ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ lock_page(page); size = i_size_read(inode); @@ -2340,10 +2340,14 @@ block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, if (!ret) ret = block_commit_write(page, 0, end); -out_unlock: - if (ret) - ret = VM_FAULT_SIGBUS; + if (unlikely(ret)) { + if (ret == -ENOMEM) + ret = VM_FAULT_OOM; + else /* -ENOSPC, -EIO, etc */ + ret = VM_FAULT_SIGBUS; + } +out_unlock: unlock_page(page); return ret; } -- cgit v1.2.3 From 09771430f3b46ee27c69daa7ecad82007568e834 Mon Sep 17 00:00:00 2001 From: Shen Feng Date: Thu, 2 Apr 2009 16:46:06 -0400 Subject: Btrfs: free inode struct when btrfs_new_inode fails btrfs_new_inode doesn't call iput to free the inode when it fails. Signed-off-by: Shen Feng Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 1cff528d5b51..0ecb3fa75dc3 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3481,8 +3481,10 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, if (dir) { ret = btrfs_set_inode_index(dir, index); - if (ret) + if (ret) { + iput(inode); return ERR_PTR(ret); + } } /* * index_cnt is ignored for everything but a dir, @@ -3565,6 +3567,7 @@ fail: if (dir) BTRFS_I(dir)->index_cnt--; btrfs_free_path(path); + iput(inode); return ERR_PTR(ret); } -- cgit v1.2.3