From 7eaebe7d503c3ef240ac7b3efc5433fe647c0298 Mon Sep 17 00:00:00 2001 From: Huang Weiyi Date: Wed, 21 Jan 2009 10:49:16 -0500 Subject: Btrfs: removed unused #include 's Removed unused #include 's in btrfs Signed-off-by: Huang Weiyi Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 1 - 1 file changed, 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 8adfe059ab4..44dbd550c4b 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -34,7 +34,6 @@ #include #include #include -#include #include #include #include -- cgit v1.2.3-70-g09d2 From c6e308713a47527f88a277ee95b7c5d1db80f77b Mon Sep 17 00:00:00 2001 From: Qinghuang Feng Date: Wed, 21 Jan 2009 10:59:08 -0500 Subject: Btrfs: simplify iteration codes Merge list_for_each* and list_entry to list_for_each_entry* Signed-off-by: Qinghuang Feng Signed-off-by: Chris Mason --- fs/btrfs/disk-io.c | 15 ++++----------- fs/btrfs/extent-tree.c | 8 ++------ fs/btrfs/inode.c | 5 ++--- fs/btrfs/ordered-data.c | 4 +--- fs/btrfs/transaction.c | 4 +--- fs/btrfs/volumes.c | 35 +++++++++-------------------------- 6 files changed, 19 insertions(+), 52 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 0d8ccd625ba..26a18779e84 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1135,7 +1135,6 @@ static int btrfs_congested_fn(void *congested_data, int bdi_bits) { struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; int ret = 0; - struct list_head *cur; struct btrfs_device *device; struct backing_dev_info *bdi; #if 0 @@ -1143,8 +1142,7 @@ static int btrfs_congested_fn(void *congested_data, int bdi_bits) btrfs_congested_async(info, 0)) return 1; #endif - list_for_each(cur, &info->fs_devices->devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, &info->fs_devices->devices, dev_list) { if (!device->bdev) continue; bdi = blk_get_backing_dev_info(device->bdev); @@ -1162,13 +1160,11 @@ static int btrfs_congested_fn(void *congested_data, int bdi_bits) */ static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page) { - struct list_head *cur; struct btrfs_device *device; struct btrfs_fs_info *info; info = (struct btrfs_fs_info *)bdi->unplug_io_data; - list_for_each(cur, &info->fs_devices->devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, &info->fs_devices->devices, dev_list) { if (!device->bdev) continue; @@ -1994,7 +1990,6 @@ static int write_dev_supers(struct btrfs_device *device, int write_all_supers(struct btrfs_root *root, int max_mirrors) { - struct list_head *cur; struct list_head *head = &root->fs_info->fs_devices->devices; struct btrfs_device *dev; struct btrfs_super_block *sb; @@ -2010,8 +2005,7 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors) sb = &root->fs_info->super_for_commit; dev_item = &sb->dev_item; - list_for_each(cur, head) { - dev = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(dev, head, dev_list) { if (!dev->bdev) { total_errors++; continue; @@ -2044,8 +2038,7 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors) } total_errors = 0; - list_for_each(cur, head) { - dev = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(dev, head, dev_list) { if (!dev->bdev) continue; if (!dev->in_fs_metadata || !dev->writeable) diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index cdc961e7556..a4e36c38b81 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -325,10 +325,8 @@ static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, u64 flags) { struct list_head *head = &info->space_info; - struct list_head *cur; struct btrfs_space_info *found; - list_for_each(cur, head) { - found = list_entry(cur, struct btrfs_space_info, list); + list_for_each_entry(found, head, list) { if (found->flags == flags) return found; } @@ -3013,7 +3011,6 @@ loop_check: static void dump_space_info(struct btrfs_space_info *info, u64 bytes) { struct btrfs_block_group_cache *cache; - struct list_head *l; printk(KERN_INFO "space_info has %llu free, is %sfull\n", (unsigned long long)(info->total_bytes - info->bytes_used - @@ -3021,8 +3018,7 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes) (info->full) ? "" : "not "); down_read(&info->groups_sem); - list_for_each(l, &info->block_groups) { - cache = list_entry(l, struct btrfs_block_group_cache, list); + list_for_each_entry(cache, &info->block_groups, list) { spin_lock(&cache->lock); printk(KERN_INFO "block group %llu has %llu bytes, %llu used " "%llu pinned %llu reserved\n", diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 44dbd550c4b..45cf03ee1bc 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -1323,12 +1323,11 @@ static noinline int add_pending_csums(struct btrfs_trans_handle *trans, struct inode *inode, u64 file_offset, struct list_head *list) { - struct list_head *cur; struct btrfs_ordered_sum *sum; btrfs_set_trans_block_group(trans, inode); - list_for_each(cur, list) { - sum = list_entry(cur, struct btrfs_ordered_sum, list); + + list_for_each_entry(sum, list, list) { btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root->fs_info->csum_root, sum); } diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index a2094017027..77c2411a5f0 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -613,7 +613,6 @@ int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, struct btrfs_sector_sum *sector_sums; struct btrfs_ordered_extent *ordered; struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; - struct list_head *cur; unsigned long num_sectors; unsigned long i; u32 sectorsize = BTRFS_I(inode)->root->sectorsize; @@ -624,8 +623,7 @@ int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, return 1; mutex_lock(&tree->mutex); - list_for_each_prev(cur, &ordered->list) { - ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list); + list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { if (disk_bytenr >= ordered_sum->bytenr) { num_sectors = ordered_sum->len / sectorsize; sector_sums = ordered_sum->sums; diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c index 8a08f944334..919172de5c9 100644 --- a/fs/btrfs/transaction.c +++ b/fs/btrfs/transaction.c @@ -852,11 +852,9 @@ static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, { struct btrfs_pending_snapshot *pending; struct list_head *head = &trans->transaction->pending_snapshots; - struct list_head *cur; int ret; - list_for_each(cur, head) { - pending = list_entry(cur, struct btrfs_pending_snapshot, list); + list_for_each_entry(pending, head, list) { ret = create_pending_snapshot(trans, fs_info, pending); BUG_ON(ret); } diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 397c8db1bc2..fd0bedb07a6 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -103,10 +103,8 @@ static noinline struct btrfs_device *__find_device(struct list_head *head, u64 devid, u8 *uuid) { struct btrfs_device *dev; - struct list_head *cur; - list_for_each(cur, head) { - dev = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(dev, head, dev_list) { if (dev->devid == devid && (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { return dev; @@ -117,11 +115,9 @@ static noinline struct btrfs_device *__find_device(struct list_head *head, static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) { - struct list_head *cur; struct btrfs_fs_devices *fs_devices; - list_for_each(cur, &fs_uuids) { - fs_devices = list_entry(cur, struct btrfs_fs_devices, list); + list_for_each_entry(fs_devices, &fs_uuids, list) { if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) return fs_devices; } @@ -344,14 +340,11 @@ error: int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) { - struct list_head *tmp; - struct list_head *cur; - struct btrfs_device *device; + struct btrfs_device *device, *next; mutex_lock(&uuid_mutex); again: - list_for_each_safe(cur, tmp, &fs_devices->devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { if (device->in_fs_metadata) continue; @@ -382,14 +375,12 @@ again: static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) { - struct list_head *cur; struct btrfs_device *device; if (--fs_devices->opened > 0) return 0; - list_for_each(cur, &fs_devices->devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, &fs_devices->devices, dev_list) { if (device->bdev) { close_bdev_exclusive(device->bdev, device->mode); fs_devices->open_devices--; @@ -438,7 +429,6 @@ static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, { struct block_device *bdev; struct list_head *head = &fs_devices->devices; - struct list_head *cur; struct btrfs_device *device; struct block_device *latest_bdev = NULL; struct buffer_head *bh; @@ -449,8 +439,7 @@ static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int seeding = 1; int ret = 0; - list_for_each(cur, head) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, head, dev_list) { if (device->bdev) continue; if (!device->name) @@ -1016,14 +1005,12 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path) } if (strcmp(device_path, "missing") == 0) { - struct list_head *cur; struct list_head *devices; struct btrfs_device *tmp; device = NULL; devices = &root->fs_info->fs_devices->devices; - list_for_each(cur, devices) { - tmp = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(tmp, devices, dev_list) { if (tmp->in_fs_metadata && !tmp->bdev) { device = tmp; break; @@ -1279,7 +1266,6 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path) struct btrfs_trans_handle *trans; struct btrfs_device *device; struct block_device *bdev; - struct list_head *cur; struct list_head *devices; struct super_block *sb = root->fs_info->sb; u64 total_bytes; @@ -1303,8 +1289,7 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path) mutex_lock(&root->fs_info->volume_mutex); devices = &root->fs_info->fs_devices->devices; - list_for_each(cur, devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, devices, dev_list) { if (device->bdev == bdev) { ret = -EEXIST; goto error; @@ -1703,7 +1688,6 @@ static u64 div_factor(u64 num, int factor) int btrfs_balance(struct btrfs_root *dev_root) { int ret; - struct list_head *cur; struct list_head *devices = &dev_root->fs_info->fs_devices->devices; struct btrfs_device *device; u64 old_size; @@ -1722,8 +1706,7 @@ int btrfs_balance(struct btrfs_root *dev_root) dev_root = dev_root->fs_info->dev_root; /* step one make some room on all the devices */ - list_for_each(cur, devices) { - device = list_entry(cur, struct btrfs_device, dev_list); + list_for_each_entry(device, devices, dev_list) { old_size = device->total_bytes; size_to_free = div_factor(old_size, 1); size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); -- cgit v1.2.3-70-g09d2 From 35054394c4b3cecd52577c2662c84da1f3e73525 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 21 Jan 2009 13:11:13 -0500 Subject: Btrfs: stop providing a bmap operation to avoid swapfile corruptions Swapfiles use bmap to build a list of extents belonging to the file, and they assume these extents won't change over the life of the file. They also use resulting list to do IO directly to the block device. This causes problems for btrfs in a few ways: btrfs returns logical block numbers through bmap, and these are not suitable for IO. They might translate to different devices, raid etc. COW means that file block mappings are going to change frequently. Using swapfiles on btrfs will lead to corruption, so we're avoiding the problem for now by dropping bmap support entirely. A later commit will add fiemap support for people that really want to know how a file is laid out. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 45cf03ee1bc..2e25d698bab 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4156,11 +4156,6 @@ static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, return -EINVAL; } -static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock) -{ - return extent_bmap(mapping, iblock, btrfs_get_extent); -} - int btrfs_readpage(struct file *file, struct page *page) { struct extent_io_tree *tree; @@ -4985,13 +4980,24 @@ static struct extent_io_ops btrfs_extent_io_ops = { .clear_bit_hook = btrfs_clear_bit_hook, }; +/* + * btrfs doesn't support the bmap operation because swapfiles + * use bmap to make a mapping of extents in the file. They assume + * these extents won't change over the life of the file and they + * use the bmap result to do IO directly to the drive. + * + * the btrfs bmap call would return logical addresses that aren't + * suitable for IO and they also will change frequently as COW + * operations happen. So, swapfile + btrfs == corruption. + * + * For now we're avoiding this by dropping bmap. + */ static struct address_space_operations btrfs_aops = { .readpage = btrfs_readpage, .writepage = btrfs_writepage, .writepages = btrfs_writepages, .readpages = btrfs_readpages, .sync_page = block_sync_page, - .bmap = btrfs_bmap, .direct_IO = btrfs_direct_IO, .invalidatepage = btrfs_invalidatepage, .releasepage = btrfs_releasepage, -- cgit v1.2.3-70-g09d2 From 1506fcc8189cdd4b95e06df7845a09f18b4526a6 Mon Sep 17 00:00:00 2001 From: Yehuda Sadeh Date: Wed, 21 Jan 2009 14:39:14 -0500 Subject: Btrfs: fiemap support Now that bmap support is gone, this is the only way to get extent mappings for userland. These are still not valid for IO, but they can tell us if a file has holes or how much fragmentation there is. Signed-off-by: Yehuda Sadeh --- fs/btrfs/extent_io.c | 92 ++++++++++++++++++++++++++++++++++++++++++++++++++++ fs/btrfs/extent_io.h | 2 ++ fs/btrfs/inode.c | 7 ++++ 3 files changed, 101 insertions(+) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index c9446d4840e..a3b0676403f 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2854,6 +2854,98 @@ out: return sector; } +int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, + __u64 start, __u64 len, get_extent_t *get_extent) +{ + int ret; + u64 off = start; + u64 max = start + len; + u32 flags = 0; + u64 disko = 0; + struct extent_map *em = NULL; + int end = 0; + u64 em_start = 0, em_len = 0; + unsigned long emflags; + ret = 0; + + if (len == 0) + return -EINVAL; + + lock_extent(&BTRFS_I(inode)->io_tree, start, start + len, + GFP_NOFS); + em = get_extent(inode, NULL, 0, off, max - off, 0); + if (!em) + goto out; + if (IS_ERR(em)) { + ret = PTR_ERR(em); + goto out; + } + while (!end) { + off = em->start + em->len; + if (off >= max) + end = 1; + + em_start = em->start; + em_len = em->len; + + disko = 0; + flags = 0; + + switch (em->block_start) { + case EXTENT_MAP_LAST_BYTE: + end = 1; + flags |= FIEMAP_EXTENT_LAST; + break; + case EXTENT_MAP_HOLE: + flags |= FIEMAP_EXTENT_UNWRITTEN; + break; + case EXTENT_MAP_INLINE: + flags |= (FIEMAP_EXTENT_DATA_INLINE | + FIEMAP_EXTENT_NOT_ALIGNED); + break; + case EXTENT_MAP_DELALLOC: + flags |= (FIEMAP_EXTENT_DELALLOC | + FIEMAP_EXTENT_UNKNOWN); + break; + default: + disko = em->block_start; + break; + } + if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) + flags |= FIEMAP_EXTENT_ENCODED; + + emflags = em->flags; + free_extent_map(em); + em = NULL; + + if (!end) { + em = get_extent(inode, NULL, 0, off, max - off, 0); + if (!em) + goto out; + if (IS_ERR(em)) { + ret = PTR_ERR(em); + goto out; + } + emflags = em->flags; + } + if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) { + flags |= FIEMAP_EXTENT_LAST; + end = 1; + } + + ret = fiemap_fill_next_extent(fieinfo, em_start, disko, + em_len, flags); + if (ret) + goto out_free; + } +out_free: + free_extent_map(em); +out: + unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len, + GFP_NOFS); + return ret; +} + static inline struct page *extent_buffer_page(struct extent_buffer *eb, unsigned long i) { diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index c5b483a7913..e80c6d96b31 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -193,6 +193,8 @@ int extent_commit_write(struct extent_io_tree *tree, unsigned from, unsigned to); sector_t extent_bmap(struct address_space *mapping, sector_t iblock, get_extent_t *get_extent); +int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, + __u64 start, __u64 len, get_extent_t *get_extent); int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end); int set_state_private(struct extent_io_tree *tree, u64 start, u64 private); int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private); diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 2e25d698bab..288c2cdc754 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4156,6 +4156,12 @@ static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, return -EINVAL; } +static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, + __u64 start, __u64 len) +{ + return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); +} + int btrfs_readpage(struct file *file, struct page *page) { struct extent_io_tree *tree; @@ -5021,6 +5027,7 @@ static struct inode_operations btrfs_file_inode_operations = { .removexattr = btrfs_removexattr, .permission = btrfs_permission, .fallocate = btrfs_fallocate, + .fiemap = btrfs_fiemap, }; static struct inode_operations btrfs_special_inode_operations = { .getattr = btrfs_getattr, -- cgit v1.2.3-70-g09d2 From 89f135d8b53bcccafd91a075366d2704ba257cf3 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 28 Jan 2009 15:34:27 -0500 Subject: Btrfs: fix readdir on 32 bit machines After btrfs_readdir has gone through all the directory items, it sets the directory f_pos to the largest possible int. This way applications that mix readdir with creating new files don't end up in an endless loop finding the new directory items as they go. It was a workaround for a bug in git, but the assumption was that if git could make this looping mistake than it would be a common problem. The largest possible int chosen was INT_LIMIT(typeof(file->f_pos), and it is possible for that to be a larger number than 32 bit glibc expects to come out of readdir. This patches switches that to INT_LIMIT(off_t), which should keep applications happy on 32 and 64 bit machines. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 288c2cdc754..2bb65e9b144 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3263,7 +3263,7 @@ skip: /* Reached end of directory/root. Bump pos past the last item. */ if (key_type == BTRFS_DIR_INDEX_KEY) - filp->f_pos = INT_LIMIT(typeof(filp->f_pos)); + filp->f_pos = INT_LIMIT(off_t); else filp->f_pos++; nopos: -- cgit v1.2.3-70-g09d2 From 0279b4cd86685b5eea467c1b74ce94f0add2c0a3 Mon Sep 17 00:00:00 2001 From: Jim Owens Date: Wed, 4 Feb 2009 09:29:13 -0500 Subject: Btrfs: selinux support Add call to LSM security initialization and save resulting security xattr for new inodes. Add xattr support to symlink inode ops. Set inode->i_op for existing special files. Signed-off-by: jim owens --- fs/btrfs/inode.c | 23 +++++++++++++++++++---- fs/btrfs/xattr.c | 32 ++++++++++++++++++++++++++++++++ fs/btrfs/xattr.h | 2 ++ 3 files changed, 53 insertions(+), 4 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 2bb65e9b144..4a79e1c5ebd 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -90,6 +90,16 @@ static noinline int cow_file_range(struct inode *inode, u64 start, u64 end, int *page_started, unsigned long *nr_written, int unlock); +static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) +{ + int err; + + err = btrfs_init_acl(inode, dir); + if (!err) + err = btrfs_xattr_security_init(inode, dir); + return err; +} + /* * a very lame attempt at stopping writes when the FS is 85% full. There * are countless ways this is incorrect, but it is better than nothing. @@ -2037,6 +2047,7 @@ void btrfs_read_locked_inode(struct inode *inode) inode->i_mapping->backing_dev_info = &root->fs_info->bdi; break; default: + inode->i_op = &btrfs_special_inode_operations; init_special_inode(inode, inode->i_mode, rdev); break; } @@ -3584,7 +3595,7 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry, if (IS_ERR(inode)) goto out_unlock; - err = btrfs_init_acl(inode, dir); + err = btrfs_init_inode_security(inode, dir); if (err) { drop_inode = 1; goto out_unlock; @@ -3647,7 +3658,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry, if (IS_ERR(inode)) goto out_unlock; - err = btrfs_init_acl(inode, dir); + err = btrfs_init_inode_security(inode, dir); if (err) { drop_inode = 1; goto out_unlock; @@ -3770,7 +3781,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) drop_on_err = 1; - err = btrfs_init_acl(inode, dir); + err = btrfs_init_inode_security(inode, dir); if (err) goto out_fail; @@ -4732,7 +4743,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry, if (IS_ERR(inode)) goto out_unlock; - err = btrfs_init_acl(inode, dir); + err = btrfs_init_inode_security(inode, dir); if (err) { drop_inode = 1; goto out_unlock; @@ -5043,4 +5054,8 @@ static struct inode_operations btrfs_symlink_inode_operations = { .follow_link = page_follow_link_light, .put_link = page_put_link, .permission = btrfs_permission, + .setxattr = btrfs_setxattr, + .getxattr = btrfs_getxattr, + .listxattr = btrfs_listxattr, + .removexattr = btrfs_removexattr, }; diff --git a/fs/btrfs/xattr.c b/fs/btrfs/xattr.c index b4fa5f4b6ad..312b9435e9f 100644 --- a/fs/btrfs/xattr.c +++ b/fs/btrfs/xattr.c @@ -21,6 +21,7 @@ #include #include #include +#include #include "ctree.h" #include "btrfs_inode.h" #include "transaction.h" @@ -330,3 +331,34 @@ int btrfs_removexattr(struct dentry *dentry, const char *name) return -EOPNOTSUPP; return __btrfs_setxattr(dentry->d_inode, name, NULL, 0, XATTR_REPLACE); } + +int btrfs_xattr_security_init(struct inode *inode, struct inode *dir) +{ + int err; + size_t len; + void *value; + char *suffix; + char *name; + + err = security_inode_init_security(inode, dir, &suffix, &value, &len); + if (err) { + if (err == -EOPNOTSUPP) + return 0; + return err; + } + + name = kmalloc(XATTR_SECURITY_PREFIX_LEN + strlen(suffix) + 1, + GFP_NOFS); + if (!name) { + err = -ENOMEM; + } else { + strcpy(name, XATTR_SECURITY_PREFIX); + strcpy(name + XATTR_SECURITY_PREFIX_LEN, suffix); + err = __btrfs_setxattr(inode, name, value, len, 0); + kfree(name); + } + + kfree(suffix); + kfree(value); + return err; +} diff --git a/fs/btrfs/xattr.h b/fs/btrfs/xattr.h index 5b1d08f8e68..c71e9c3cf3f 100644 --- a/fs/btrfs/xattr.h +++ b/fs/btrfs/xattr.h @@ -36,4 +36,6 @@ extern int btrfs_setxattr(struct dentry *dentry, const char *name, const void *value, size_t size, int flags); extern int btrfs_removexattr(struct dentry *dentry, const char *name); +extern int btrfs_xattr_security_init(struct inode *inode, struct inode *dir); + #endif /* __XATTR__ */ -- cgit v1.2.3-70-g09d2 From b4ce94de9b4d64e8ab3cf155d13653c666e22b9b Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 4 Feb 2009 09:25:08 -0500 Subject: Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason --- fs/btrfs/ctree.c | 234 +++++++++++++++++++++++++++++++++++++++++++++++-- fs/btrfs/ctree.h | 4 + fs/btrfs/disk-io.c | 10 ++- fs/btrfs/extent-tree.c | 5 ++ fs/btrfs/extent_io.c | 18 ++-- fs/btrfs/extent_io.h | 16 +++- fs/btrfs/inode.c | 3 + fs/btrfs/locking.c | 208 +++++++++++++++++++++++++++++++++++++++---- fs/btrfs/locking.h | 6 ++ fs/btrfs/tree-defrag.c | 1 + fs/btrfs/tree-log.c | 4 + 11 files changed, 470 insertions(+), 39 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index 3b6e35aafc9..3af777357ac 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -54,6 +54,31 @@ struct btrfs_path *btrfs_alloc_path(void) return path; } +/* + * set all locked nodes in the path to blocking locks. This should + * be done before scheduling + */ +noinline void btrfs_set_path_blocking(struct btrfs_path *p) +{ + int i; + for (i = 0; i < BTRFS_MAX_LEVEL; i++) { + if (p->nodes[i] && p->locks[i]) + btrfs_set_lock_blocking(p->nodes[i]); + } +} + +/* + * reset all the locked nodes in the patch to spinning locks. + */ +noinline void btrfs_clear_path_blocking(struct btrfs_path *p) +{ + int i; + for (i = 0; i < BTRFS_MAX_LEVEL; i++) { + if (p->nodes[i] && p->locks[i]) + btrfs_clear_lock_blocking(p->nodes[i]); + } +} + /* this also releases the path */ void btrfs_free_path(struct btrfs_path *p) { @@ -272,6 +297,8 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, if (IS_ERR(cow)) return PTR_ERR(cow); + /* cow is set to blocking by btrfs_init_new_buffer */ + copy_extent_buffer(cow, buf, 0, 0, cow->len); btrfs_set_header_bytenr(cow, cow->start); btrfs_set_header_generation(cow, trans->transid); @@ -397,6 +424,11 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, } search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); + + if (parent) + btrfs_set_lock_blocking(parent); + btrfs_set_lock_blocking(buf); + ret = __btrfs_cow_block(trans, root, buf, parent, parent_slot, cow_ret, search_start, 0, prealloc_dest); @@ -502,6 +534,8 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans, if (parent_nritems == 1) return 0; + btrfs_set_lock_blocking(parent); + for (i = start_slot; i < end_slot; i++) { int close = 1; @@ -562,6 +596,7 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans, search_start = last_block; btrfs_tree_lock(cur); + btrfs_set_lock_blocking(cur); err = __btrfs_cow_block(trans, root, cur, parent, i, &cur, search_start, min(16 * blocksize, @@ -860,6 +895,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans, return 0; mid = path->nodes[level]; + WARN_ON(!path->locks[level]); WARN_ON(btrfs_header_generation(mid) != trans->transid); @@ -882,6 +918,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans, /* promote the child to a root */ child = read_node_slot(root, mid, 0); btrfs_tree_lock(child); + btrfs_set_lock_blocking(child); BUG_ON(!child); ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0); BUG_ON(ret); @@ -898,6 +935,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans, add_root_to_dirty_list(root); btrfs_tree_unlock(child); + path->locks[level] = 0; path->nodes[level] = NULL; clean_tree_block(trans, root, mid); @@ -922,6 +960,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans, left = read_node_slot(root, parent, pslot - 1); if (left) { btrfs_tree_lock(left); + btrfs_set_lock_blocking(left); wret = btrfs_cow_block(trans, root, left, parent, pslot - 1, &left, 0); if (wret) { @@ -932,6 +971,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans, right = read_node_slot(root, parent, pslot + 1); if (right) { btrfs_tree_lock(right); + btrfs_set_lock_blocking(right); wret = btrfs_cow_block(trans, root, right, parent, pslot + 1, &right, 0); if (wret) { @@ -1107,6 +1147,8 @@ static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, u32 left_nr; btrfs_tree_lock(left); + btrfs_set_lock_blocking(left); + left_nr = btrfs_header_nritems(left); if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { wret = 1; @@ -1153,7 +1195,10 @@ static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, */ if (right) { u32 right_nr; + btrfs_tree_lock(right); + btrfs_set_lock_blocking(right); + right_nr = btrfs_header_nritems(right); if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { wret = 1; @@ -1264,6 +1309,68 @@ static noinline void reada_for_search(struct btrfs_root *root, } } +/* + * returns -EAGAIN if it had to drop the path, or zero if everything was in + * cache + */ +static noinline int reada_for_balance(struct btrfs_root *root, + struct btrfs_path *path, int level) +{ + int slot; + int nritems; + struct extent_buffer *parent; + struct extent_buffer *eb; + u64 gen; + u64 block1 = 0; + u64 block2 = 0; + int ret = 0; + int blocksize; + + parent = path->nodes[level - 1]; + if (!parent) + return 0; + + nritems = btrfs_header_nritems(parent); + slot = path->slots[level]; + blocksize = btrfs_level_size(root, level); + + if (slot > 0) { + block1 = btrfs_node_blockptr(parent, slot - 1); + gen = btrfs_node_ptr_generation(parent, slot - 1); + eb = btrfs_find_tree_block(root, block1, blocksize); + if (eb && btrfs_buffer_uptodate(eb, gen)) + block1 = 0; + free_extent_buffer(eb); + } + if (slot < nritems) { + block2 = btrfs_node_blockptr(parent, slot + 1); + gen = btrfs_node_ptr_generation(parent, slot + 1); + eb = btrfs_find_tree_block(root, block2, blocksize); + if (eb && btrfs_buffer_uptodate(eb, gen)) + block2 = 0; + free_extent_buffer(eb); + } + if (block1 || block2) { + ret = -EAGAIN; + btrfs_release_path(root, path); + if (block1) + readahead_tree_block(root, block1, blocksize, 0); + if (block2) + readahead_tree_block(root, block2, blocksize, 0); + + if (block1) { + eb = read_tree_block(root, block1, blocksize, 0); + free_extent_buffer(eb); + } + if (block1) { + eb = read_tree_block(root, block2, blocksize, 0); + free_extent_buffer(eb); + } + } + return ret; +} + + /* * when we walk down the tree, it is usually safe to unlock the higher layers * in the tree. The exceptions are when our path goes through slot 0, because @@ -1314,6 +1421,32 @@ static noinline void unlock_up(struct btrfs_path *path, int level, } } +/* + * This releases any locks held in the path starting at level and + * going all the way up to the root. + * + * btrfs_search_slot will keep the lock held on higher nodes in a few + * corner cases, such as COW of the block at slot zero in the node. This + * ignores those rules, and it should only be called when there are no + * more updates to be done higher up in the tree. + */ +noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) +{ + int i; + + if (path->keep_locks || path->lowest_level) + return; + + for (i = level; i < BTRFS_MAX_LEVEL; i++) { + if (!path->nodes[i]) + break; + if (!path->locks[i]) + break; + btrfs_tree_unlock(path->nodes[i]); + path->locks[i] = 0; + } +} + /* * look for key in the tree. path is filled in with nodes along the way * if key is found, we return zero and you can find the item in the leaf @@ -1385,6 +1518,7 @@ again: */ if (prealloc_block.objectid && prealloc_block.offset != b->len) { + btrfs_set_path_blocking(p); btrfs_free_reserved_extent(root, prealloc_block.objectid, prealloc_block.offset); @@ -1409,6 +1543,8 @@ again: goto again; } + btrfs_set_path_blocking(p); + wret = btrfs_cow_block(trans, root, b, p->nodes[level + 1], p->slots[level + 1], @@ -1430,6 +1566,22 @@ cow_done: if (!p->skip_locking) p->locks[level] = 1; + btrfs_clear_path_blocking(p); + + /* + * we have a lock on b and as long as we aren't changing + * the tree, there is no way to for the items in b to change. + * It is safe to drop the lock on our parent before we + * go through the expensive btree search on b. + * + * If cow is true, then we might be changing slot zero, + * which may require changing the parent. So, we can't + * drop the lock until after we know which slot we're + * operating on. + */ + if (!cow) + btrfs_unlock_up_safe(p, level + 1); + ret = check_block(root, p, level); if (ret) { ret = -1; @@ -1437,6 +1589,7 @@ cow_done: } ret = bin_search(b, key, level, &slot); + if (level != 0) { if (ret && slot > 0) slot -= 1; @@ -1444,7 +1597,16 @@ cow_done: if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { - int sret = split_node(trans, root, p, level); + int sret; + + sret = reada_for_balance(root, p, level); + if (sret) + goto again; + + btrfs_set_path_blocking(p); + sret = split_node(trans, root, p, level); + btrfs_clear_path_blocking(p); + BUG_ON(sret > 0); if (sret) { ret = sret; @@ -1453,8 +1615,16 @@ cow_done: b = p->nodes[level]; slot = p->slots[level]; } else if (ins_len < 0) { - int sret = balance_level(trans, root, p, - level); + int sret; + + sret = reada_for_balance(root, p, level); + if (sret) + goto again; + + btrfs_set_path_blocking(p); + sret = balance_level(trans, root, p, level); + btrfs_clear_path_blocking(p); + if (sret) { ret = sret; goto done; @@ -1488,7 +1658,7 @@ cow_done: * of the btree by dropping locks before * we read. */ - if (level > 1) { + if (level > 0) { btrfs_release_path(NULL, p); if (tmp) free_extent_buffer(tmp); @@ -1503,6 +1673,7 @@ cow_done: free_extent_buffer(tmp); goto again; } else { + btrfs_set_path_blocking(p); if (tmp) free_extent_buffer(tmp); if (should_reada) @@ -1512,14 +1683,29 @@ cow_done: b = read_node_slot(root, b, slot); } } - if (!p->skip_locking) - btrfs_tree_lock(b); + if (!p->skip_locking) { + int lret; + + btrfs_clear_path_blocking(p); + lret = btrfs_try_spin_lock(b); + + if (!lret) { + btrfs_set_path_blocking(p); + btrfs_tree_lock(b); + btrfs_clear_path_blocking(p); + } + } } else { p->slots[level] = slot; if (ins_len > 0 && btrfs_leaf_free_space(root, b) < ins_len) { - int sret = split_leaf(trans, root, key, + int sret; + + btrfs_set_path_blocking(p); + sret = split_leaf(trans, root, key, p, ins_len, ret == 0); + btrfs_clear_path_blocking(p); + BUG_ON(sret > 0); if (sret) { ret = sret; @@ -1533,12 +1719,16 @@ cow_done: } ret = 1; 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 (prealloc_block.objectid) { btrfs_free_reserved_extent(root, prealloc_block.objectid, prealloc_block.offset); } - return ret; } @@ -1562,6 +1752,8 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans, ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0); BUG_ON(ret); + btrfs_set_lock_blocking(eb); + parent = eb; while (1) { level = btrfs_header_level(parent); @@ -1586,6 +1778,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans, eb = read_tree_block(root, bytenr, blocksize, generation); btrfs_tree_lock(eb); + btrfs_set_lock_blocking(eb); } /* @@ -1610,6 +1803,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans, eb = read_tree_block(root, bytenr, blocksize, generation); btrfs_tree_lock(eb); + btrfs_set_lock_blocking(eb); } ret = btrfs_cow_block(trans, root, eb, parent, slot, @@ -2156,6 +2350,8 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root right = read_node_slot(root, upper, slot + 1); btrfs_tree_lock(right); + btrfs_set_lock_blocking(right); + free_space = btrfs_leaf_free_space(root, right); if (free_space < data_size) goto out_unlock; @@ -2351,6 +2547,8 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root left = read_node_slot(root, path->nodes[1], slot - 1); btrfs_tree_lock(left); + btrfs_set_lock_blocking(left); + free_space = btrfs_leaf_free_space(root, left); if (free_space < data_size) { ret = 1; @@ -2809,6 +3007,12 @@ int btrfs_split_item(struct btrfs_trans_handle *trans, path->keep_locks = 0; BUG_ON(ret); + /* + * 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)); @@ -3338,6 +3542,7 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, BUG(); } out: + btrfs_unlock_up_safe(path, 1); return ret; } @@ -3705,12 +3910,14 @@ find_next_key: */ if (slot >= nritems) { path->slots[level] = slot; + btrfs_set_path_blocking(path); sret = btrfs_find_next_key(root, path, min_key, level, cache_only, min_trans); if (sret == 0) { btrfs_release_path(root, path); goto again; } else { + btrfs_clear_path_blocking(path); goto out; } } @@ -3722,16 +3929,20 @@ find_next_key: unlock_up(path, level, 1); goto out; } + btrfs_set_path_blocking(path); cur = read_node_slot(root, cur, slot); btrfs_tree_lock(cur); + path->locks[level - 1] = 1; path->nodes[level - 1] = cur; unlock_up(path, level, 1); + btrfs_clear_path_blocking(path); } out: if (ret == 0) memcpy(min_key, &found_key, sizeof(found_key)); + btrfs_set_path_blocking(path); return ret; } @@ -3827,6 +4038,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) if (ret < 0) return ret; + btrfs_set_path_blocking(path); nritems = btrfs_header_nritems(path->nodes[0]); /* * by releasing the path above we dropped all our locks. A balance @@ -3857,6 +4069,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) free_extent_buffer(next); } + /* the path was set to blocking above */ if (level == 1 && (path->locks[1] || path->skip_locking) && path->reada) reada_for_search(root, path, level, slot, 0); @@ -3865,6 +4078,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) if (!path->skip_locking) { WARN_ON(!btrfs_tree_locked(c)); btrfs_tree_lock(next); + btrfs_set_lock_blocking(next); } break; } @@ -3881,12 +4095,15 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) path->locks[level] = 1; if (!level) break; + + btrfs_set_path_blocking(path); if (level == 1 && path->locks[1] && path->reada) reada_for_search(root, path, level, slot, 0); next = read_node_slot(root, next, 0); if (!path->skip_locking) { WARN_ON(!btrfs_tree_locked(path->nodes[level])); btrfs_tree_lock(next); + btrfs_set_lock_blocking(next); } } done: @@ -3911,6 +4128,7 @@ int btrfs_previous_item(struct btrfs_root *root, while (1) { if (path->slots[0] == 0) { + btrfs_set_path_blocking(path); ret = btrfs_prev_leaf(root, path); if (ret != 0) return ret; diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index f2b8d26b047..531db112c8b 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -1835,6 +1835,10 @@ void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p); struct btrfs_path *btrfs_alloc_path(void); void btrfs_free_path(struct btrfs_path *p); void btrfs_init_path(struct btrfs_path *p); +void btrfs_set_path_blocking(struct btrfs_path *p); +void btrfs_clear_path_blocking(struct btrfs_path *p); +void btrfs_unlock_up_safe(struct btrfs_path *p, int level); + int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, int slot, int nr); int btrfs_del_leaf(struct btrfs_trans_handle *trans, diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 549271607c1..5aebddd7119 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -799,7 +799,7 @@ struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); if (ret == 0) - buf->flags |= EXTENT_UPTODATE; + set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); else WARN_ON(1); return buf; @@ -813,6 +813,10 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, if (btrfs_header_generation(buf) == root->fs_info->running_transaction->transid) { WARN_ON(!btrfs_tree_locked(buf)); + + /* ugh, clear_extent_buffer_dirty can be expensive */ + btrfs_set_lock_blocking(buf); + clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf); } @@ -2311,6 +2315,8 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf) u64 transid = btrfs_header_generation(buf); struct inode *btree_inode = root->fs_info->btree_inode; + btrfs_set_lock_blocking(buf); + WARN_ON(!btrfs_tree_locked(buf)); if (transid != root->fs_info->generation) { printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " @@ -2353,7 +2359,7 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) int ret; ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); if (ret == 0) - buf->flags |= EXTENT_UPTODATE; + set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); return ret; } diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 7a22f2e6ec4..ed1e25d7248 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -3407,7 +3407,10 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, btrfs_set_header_generation(buf, trans->transid); btrfs_tree_lock(buf); clean_tree_block(trans, root, buf); + + btrfs_set_lock_blocking(buf); btrfs_set_buffer_uptodate(buf); + if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { set_extent_dirty(&root->dirty_log_pages, buf->start, buf->start + buf->len - 1, GFP_NOFS); @@ -3416,6 +3419,7 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, buf->start + buf->len - 1, GFP_NOFS); } trans->blocks_used++; + /* this returns a buffer locked for blocking */ return buf; } @@ -3752,6 +3756,7 @@ static noinline int walk_down_subtree(struct btrfs_trans_handle *trans, next = read_tree_block(root, bytenr, blocksize, ptr_gen); btrfs_tree_lock(next); + btrfs_set_lock_blocking(next); ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, &refs); diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 2ea7f052722..dd5df53e045 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2990,7 +2990,9 @@ static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, eb = kmem_cache_zalloc(extent_buffer_cache, mask); eb->start = start; eb->len = len; - mutex_init(&eb->mutex); + spin_lock_init(&eb->lock); + init_waitqueue_head(&eb->lock_wq); + #if LEAK_DEBUG spin_lock_irqsave(&leak_lock, flags); list_add(&eb->leak_list, &buffers); @@ -3071,8 +3073,7 @@ struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, unlock_page(p); } if (uptodate) - eb->flags |= EXTENT_UPTODATE; - eb->flags |= EXTENT_BUFFER_FILLED; + set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); spin_lock(&tree->buffer_lock); exists = buffer_tree_insert(tree, start, &eb->rb_node); @@ -3226,7 +3227,7 @@ int clear_extent_buffer_uptodate(struct extent_io_tree *tree, unsigned long num_pages; num_pages = num_extent_pages(eb->start, eb->len); - eb->flags &= ~EXTENT_UPTODATE; + clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, GFP_NOFS); @@ -3297,7 +3298,7 @@ int extent_buffer_uptodate(struct extent_io_tree *tree, struct page *page; int pg_uptodate = 1; - if (eb->flags & EXTENT_UPTODATE) + if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) return 1; ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, @@ -3333,7 +3334,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree, struct bio *bio = NULL; unsigned long bio_flags = 0; - if (eb->flags & EXTENT_UPTODATE) + if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) return 0; if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, @@ -3364,7 +3365,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree, } if (all_uptodate) { if (start_i == 0) - eb->flags |= EXTENT_UPTODATE; + set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); goto unlock_exit; } @@ -3400,7 +3401,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree, } if (!ret) - eb->flags |= EXTENT_UPTODATE; + set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); return ret; unlock_exit: @@ -3497,7 +3498,6 @@ int map_extent_buffer(struct extent_buffer *eb, unsigned long start, unmap_extent_buffer(eb, eb->map_token, km); eb->map_token = NULL; save = 1; - WARN_ON(!mutex_is_locked(&eb->mutex)); } err = map_private_extent_buffer(eb, start, min_len, token, map, map_start, map_len, km); diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index e80c6d96b31..1f9df88afbf 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -22,6 +22,10 @@ /* flags for bio submission */ #define EXTENT_BIO_COMPRESSED 1 +/* these are bit numbers for test/set bit */ +#define EXTENT_BUFFER_UPTODATE 0 +#define EXTENT_BUFFER_BLOCKING 1 + /* * page->private values. Every page that is controlled by the extent * map has page->private set to one. @@ -95,11 +99,19 @@ struct extent_buffer { unsigned long map_start; unsigned long map_len; struct page *first_page; + unsigned long bflags; atomic_t refs; - int flags; struct list_head leak_list; struct rb_node rb_node; - struct mutex mutex; + + /* the spinlock is used to protect most operations */ + spinlock_t lock; + + /* + * when we keep the lock held while blocking, waiters go onto + * the wq + */ + wait_queue_head_t lock_wq; }; struct extent_map_tree; diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 4a79e1c5ebd..ebd7d6c37df 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -50,6 +50,7 @@ #include "tree-log.h" #include "ref-cache.h" #include "compression.h" +#include "locking.h" struct btrfs_iget_args { u64 ino; @@ -2021,6 +2022,7 @@ void btrfs_read_locked_inode(struct inode *inode) BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); alloc_group_block = btrfs_inode_block_group(leaf, inode_item); + BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, alloc_group_block, 0); btrfs_free_path(path); @@ -2117,6 +2119,7 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, goto failed; } + btrfs_unlock_up_safe(path, 1); leaf = path->nodes[0]; inode_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c index 39bae7761db..68fd9ccf180 100644 --- a/fs/btrfs/locking.c +++ b/fs/btrfs/locking.c @@ -26,45 +26,215 @@ #include "locking.h" /* - * locks the per buffer mutex in an extent buffer. This uses adaptive locks - * and the spin is not tuned very extensively. The spinning does make a big - * difference in almost every workload, but spinning for the right amount of - * time needs some help. - * - * In general, we want to spin as long as the lock holder is doing btree - * searches, and we should give up if they are in more expensive code. + * btrfs_header_level() isn't free, so don't call it when lockdep isn't + * on */ +#ifdef CONFIG_DEBUG_LOCK_ALLOC +static inline void spin_nested(struct extent_buffer *eb) +{ + spin_lock_nested(&eb->lock, BTRFS_MAX_LEVEL - btrfs_header_level(eb)); +} +#else +static inline void spin_nested(struct extent_buffer *eb) +{ + spin_lock(&eb->lock); +} +#endif -int btrfs_tree_lock(struct extent_buffer *eb) +/* + * Setting a lock to blocking will drop the spinlock and set the + * flag that forces other procs who want the lock to wait. After + * this you can safely schedule with the lock held. + */ +void btrfs_set_lock_blocking(struct extent_buffer *eb) { - int i; + if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) { + set_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags); + spin_unlock(&eb->lock); + } + /* exit with the spin lock released and the bit set */ +} - if (mutex_trylock(&eb->mutex)) - return 0; +/* + * clearing the blocking flag will take the spinlock again. + * After this you can't safely schedule + */ +void btrfs_clear_lock_blocking(struct extent_buffer *eb) +{ + if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) { + spin_nested(eb); + clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags); + smp_mb__after_clear_bit(); + } + /* exit with the spin lock held */ +} + +/* + * unfortunately, many of the places that currently set a lock to blocking + * don't end up blocking for every long, and often they don't block + * at all. For a dbench 50 run, if we don't spin one the blocking bit + * at all, the context switch rate can jump up to 400,000/sec or more. + * + * So, we're still stuck with this crummy spin on the blocking bit, + * at least until the most common causes of the short blocks + * can be dealt with. + */ +static int btrfs_spin_on_block(struct extent_buffer *eb) +{ + int i; for (i = 0; i < 512; i++) { cpu_relax(); - if (mutex_trylock(&eb->mutex)) + if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) + return 1; + if (need_resched()) + break; + } + return 0; +} + +/* + * This is somewhat different from trylock. It will take the + * spinlock but if it finds the lock is set to blocking, it will + * return without the lock held. + * + * returns 1 if it was able to take the lock and zero otherwise + * + * After this call, scheduling is not safe without first calling + * btrfs_set_lock_blocking() + */ +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); + + /* spin for a bit on the BLOCKING flag */ + for (i = 0; i < 2; i++) { + if (!btrfs_spin_on_block(eb)) + break; + + spin_nested(eb); + if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) + return 1; + spin_unlock(&eb->lock); + } + return 0; +} + +/* + * the autoremove wake function will return 0 if it tried to wake up + * a process that was already awake, which means that process won't + * count as an exclusive wakeup. The waitq code will continue waking + * procs until it finds one that was actually sleeping. + * + * For btrfs, this isn't quite what we want. We want a single proc + * to be notified that the lock is ready for taking. If that proc + * already happen to be awake, great, it will loop around and try for + * the lock. + * + * So, btrfs_wake_function always returns 1, even when the proc that we + * tried to wake up was already awake. + */ +static int btrfs_wake_function(wait_queue_t *wait, unsigned mode, + int sync, void *key) +{ + autoremove_wake_function(wait, mode, sync, key); + return 1; +} + +/* + * returns with the extent buffer spinlocked. + * + * This will spin and/or wait as required to take the lock, and then + * return with the spinlock held. + * + * After this call, scheduling is not safe without first calling + * btrfs_set_lock_blocking() + */ +int btrfs_tree_lock(struct extent_buffer *eb) +{ + DEFINE_WAIT(wait); + wait.func = btrfs_wake_function; + + while(1) { + spin_nested(eb); + + /* nobody is blocking, exit with the spinlock held */ + if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) return 0; + + /* + * we have the spinlock, but the real owner is blocking. + * wait for them + */ + spin_unlock(&eb->lock); + + /* + * spin for a bit, and if the blocking flag goes away, + * loop around + */ + if (btrfs_spin_on_block(eb)) + continue; + + prepare_to_wait_exclusive(&eb->lock_wq, &wait, + TASK_UNINTERRUPTIBLE); + + if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) + schedule(); + + finish_wait(&eb->lock_wq, &wait); } - cpu_relax(); - mutex_lock_nested(&eb->mutex, BTRFS_MAX_LEVEL - btrfs_header_level(eb)); return 0; } +/* + * Very quick trylock, this does not spin or schedule. It returns + * 1 with the spinlock held if it was able to take the lock, or it + * returns zero if it was unable to take the lock. + * + * After this call, scheduling is not safe without first calling + * btrfs_set_lock_blocking() + */ int btrfs_try_tree_lock(struct extent_buffer *eb) { - return mutex_trylock(&eb->mutex); + if (spin_trylock(&eb->lock)) { + if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) { + /* + * we've got the spinlock, but the real owner is + * blocking. Drop the spinlock and return failure + */ + spin_unlock(&eb->lock); + return 0; + } + return 1; + } + /* someone else has the spinlock giveup */ + return 0; } int btrfs_tree_unlock(struct extent_buffer *eb) { - mutex_unlock(&eb->mutex); + /* + * if we were a blocking owner, we don't have the spinlock held + * just clear the bit and look for waiters + */ + if (test_and_clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) + smp_mb__after_clear_bit(); + else + spin_unlock(&eb->lock); + + if (waitqueue_active(&eb->lock_wq)) + wake_up(&eb->lock_wq); return 0; } int btrfs_tree_locked(struct extent_buffer *eb) { - return mutex_is_locked(&eb->mutex); + return test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags) || + spin_is_locked(&eb->lock); } /* @@ -75,12 +245,14 @@ int btrfs_path_lock_waiting(struct btrfs_path *path, int level) { int i; struct extent_buffer *eb; + for (i = level; i <= level + 1 && i < BTRFS_MAX_LEVEL; i++) { eb = path->nodes[i]; if (!eb) break; smp_mb(); - if (!list_empty(&eb->mutex.wait_list)) + if (spin_is_contended(&eb->lock) || + waitqueue_active(&eb->lock_wq)) return 1; } return 0; diff --git a/fs/btrfs/locking.h b/fs/btrfs/locking.h index bc1faef1251..d92e707f587 100644 --- a/fs/btrfs/locking.h +++ b/fs/btrfs/locking.h @@ -22,6 +22,12 @@ int btrfs_tree_lock(struct extent_buffer *eb); int btrfs_tree_unlock(struct extent_buffer *eb); int btrfs_tree_locked(struct extent_buffer *eb); + int btrfs_try_tree_lock(struct extent_buffer *eb); +int btrfs_try_spin_lock(struct extent_buffer *eb); + int btrfs_path_lock_waiting(struct btrfs_path *path, int level); + +void btrfs_set_lock_blocking(struct extent_buffer *eb); +void btrfs_clear_lock_blocking(struct extent_buffer *eb); #endif diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c index 3e8358c3616..98d25fa4570 100644 --- a/fs/btrfs/tree-defrag.c +++ b/fs/btrfs/tree-defrag.c @@ -74,6 +74,7 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans, u32 nritems; root_node = btrfs_lock_root_node(root); + btrfs_set_lock_blocking(root_node); nritems = btrfs_header_nritems(root_node); root->defrag_max.objectid = 0; /* from above we know this is not a leaf */ diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 4f26f3ed0c8..20794290256 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -1615,6 +1615,7 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, btrfs_tree_lock(next); clean_tree_block(trans, root, next); + btrfs_set_lock_blocking(next); btrfs_wait_tree_block_writeback(next); btrfs_tree_unlock(next); @@ -1661,6 +1662,7 @@ static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, next = path->nodes[*level]; btrfs_tree_lock(next); clean_tree_block(trans, root, next); + btrfs_set_lock_blocking(next); btrfs_wait_tree_block_writeback(next); btrfs_tree_unlock(next); @@ -1718,6 +1720,7 @@ static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, btrfs_tree_lock(next); clean_tree_block(trans, root, next); + btrfs_set_lock_blocking(next); btrfs_wait_tree_block_writeback(next); btrfs_tree_unlock(next); @@ -1790,6 +1793,7 @@ static int walk_log_tree(struct btrfs_trans_handle *trans, btrfs_tree_lock(next); clean_tree_block(trans, log, next); + btrfs_set_lock_blocking(next); btrfs_wait_tree_block_writeback(next); btrfs_tree_unlock(next); -- cgit v1.2.3-70-g09d2 From bd56b30205bc09da0beb80d4ba3d4c7309792da5 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 4 Feb 2009 09:27:02 -0500 Subject: Btrfs: Make btrfs_drop_snapshot work in larger and more efficient chunks Every transaction in btrfs creates a new snapshot, and then schedules the snapshot from the last transaction for deletion. Snapshot deletion works by walking down the btree and dropping the reference counts on each btree block during the walk. If if a given leaf or node has a reference count greater than one, the reference count is decremented and the subtree pointed to by that node is ignored. If the reference count is one, walking continues down into that node or leaf, and the references of everything it points to are decremented. The old code would try to work in small pieces, walking down the tree until it found the lowest leaf or node to free and then returning. This was very friendly to the rest of the FS because it didn't have a huge impact on other operations. But it wouldn't always keep up with the rate that new commits added new snapshots for deletion, and it wasn't very optimal for the extent allocation tree because it wasn't finding leaves that were close together on disk and processing them at the same time. This changes things to walk down to a level 1 node and then process it in bulk. All the leaf pointers are sorted and the leaves are dropped in order based on their extent number. The extent allocation tree and commit code are now fast enough for this kind of bulk processing to work without slowing the rest of the FS down. Overall it does less IO and is better able to keep up with snapshot deletions under high load. Signed-off-by: Chris Mason --- fs/btrfs/extent-tree.c | 306 ++++++++++++++++++++++++++++++++++++++++++------- fs/btrfs/inode.c | 2 + fs/btrfs/ref-cache.c | 1 + fs/btrfs/ref-cache.h | 1 - 4 files changed, 265 insertions(+), 45 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index ed1e25d7248..1d3e9262a9d 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -1533,6 +1533,11 @@ out: * struct refsort is used to match byte number to slot in the btree block. * we sort based on the byte number and then use the slot to actually * find the item. + * + * struct refsort is smaller than strcut btrfs_item and smaller than + * struct btrfs_key_ptr. Since we're currently limited to the page size + * for a btree block, there's no way for a kmalloc of refsorts for a + * single node to be bigger than a page. */ struct refsort { u64 bytenr; @@ -3457,36 +3462,73 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, { u64 leaf_owner; u64 leaf_generation; + struct refsort *sorted; struct btrfs_key key; struct btrfs_file_extent_item *fi; int i; int nritems; int ret; + int refi = 0; + int slot; BUG_ON(!btrfs_is_leaf(leaf)); nritems = btrfs_header_nritems(leaf); leaf_owner = btrfs_header_owner(leaf); leaf_generation = btrfs_header_generation(leaf); + sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); + /* we do this loop twice. The first time we build a list + * of the extents we have a reference on, then we sort the list + * by bytenr. The second time around we actually do the + * extent freeing. + */ for (i = 0; i < nritems; i++) { u64 disk_bytenr; cond_resched(); btrfs_item_key_to_cpu(leaf, &key, i); + + /* only extents have references, skip everything else */ if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) continue; + fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); + + /* inline extents live in the btree, they don't have refs */ if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) continue; - /* - * FIXME make sure to insert a trans record that - * repeats the snapshot del on crash - */ + disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); + + /* holes don't have refs */ if (disk_bytenr == 0) continue; + sorted[refi].bytenr = disk_bytenr; + sorted[refi].slot = i; + refi++; + } + + if (refi == 0) + goto out; + + sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); + + for (i = 0; i < refi; i++) { + u64 disk_bytenr; + + disk_bytenr = sorted[i].bytenr; + slot = sorted[i].slot; + + cond_resched(); + + btrfs_item_key_to_cpu(leaf, &key, slot); + if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) + continue; + + fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); + ret = __btrfs_free_extent(trans, root, disk_bytenr, btrfs_file_extent_disk_num_bytes(leaf, fi), leaf->start, leaf_owner, leaf_generation, @@ -3497,6 +3539,8 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, wake_up(&root->fs_info->transaction_throttle); cond_resched(); } +out: + kfree(sorted); return 0; } @@ -3506,9 +3550,25 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, { int i; int ret; - struct btrfs_extent_info *info = ref->extents; + struct btrfs_extent_info *info; + struct refsort *sorted; + + if (ref->nritems == 0) + return 0; + sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS); + for (i = 0; i < ref->nritems; i++) { + sorted[i].bytenr = ref->extents[i].bytenr; + sorted[i].slot = i; + } + sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL); + + /* + * the items in the ref were sorted when the ref was inserted + * into the ref cache, so this is already in order + */ for (i = 0; i < ref->nritems; i++) { + info = ref->extents + sorted[i].slot; ret = __btrfs_free_extent(trans, root, info->bytenr, info->num_bytes, ref->bytenr, ref->owner, ref->generation, @@ -3565,6 +3625,152 @@ static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, return ret; } +/* + * this is used while deleting old snapshots, and it drops the refs + * on a whole subtree starting from a level 1 node. + * + * The idea is to sort all the leaf pointers, and then drop the + * ref on all the leaves in order. Most of the time the leaves + * will have ref cache entries, so no leaf IOs will be required to + * find the extents they have references on. + * + * For each leaf, any references it has are also dropped in order + * + * This ends up dropping the references in something close to optimal + * order for reading and modifying the extent allocation tree. + */ +static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path) +{ + u64 bytenr; + u64 root_owner; + u64 root_gen; + struct extent_buffer *eb = path->nodes[1]; + struct extent_buffer *leaf; + struct btrfs_leaf_ref *ref; + struct refsort *sorted = NULL; + int nritems = btrfs_header_nritems(eb); + int ret; + int i; + int refi = 0; + int slot = path->slots[1]; + u32 blocksize = btrfs_level_size(root, 0); + u32 refs; + + if (nritems == 0) + goto out; + + root_owner = btrfs_header_owner(eb); + root_gen = btrfs_header_generation(eb); + sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); + + /* + * step one, sort all the leaf pointers so we don't scribble + * randomly into the extent allocation tree + */ + for (i = slot; i < nritems; i++) { + sorted[refi].bytenr = btrfs_node_blockptr(eb, i); + sorted[refi].slot = i; + refi++; + } + + /* + * nritems won't be zero, but if we're picking up drop_snapshot + * after a crash, slot might be > 0, so double check things + * just in case. + */ + if (refi == 0) + goto out; + + sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); + + /* + * the first loop frees everything the leaves point to + */ + for (i = 0; i < refi; i++) { + u64 ptr_gen; + + bytenr = sorted[i].bytenr; + + /* + * check the reference count on this leaf. If it is > 1 + * we just decrement it below and don't update any + * of the refs the leaf points to. + */ + ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); + BUG_ON(ret); + if (refs != 1) + continue; + + ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); + + /* + * the leaf only had one reference, which means the + * only thing pointing to this leaf is the snapshot + * we're deleting. It isn't possible for the reference + * count to increase again later + * + * The reference cache is checked for the leaf, + * and if found we'll be able to drop any refs held by + * the leaf without needing to read it in. + */ + ref = btrfs_lookup_leaf_ref(root, bytenr); + if (ref && ref->generation != ptr_gen) { + btrfs_free_leaf_ref(root, ref); + ref = NULL; + } + if (ref) { + ret = cache_drop_leaf_ref(trans, root, ref); + BUG_ON(ret); + btrfs_remove_leaf_ref(root, ref); + btrfs_free_leaf_ref(root, ref); + } else { + /* + * the leaf wasn't in the reference cache, so + * we have to read it. + */ + leaf = read_tree_block(root, bytenr, blocksize, + ptr_gen); + ret = btrfs_drop_leaf_ref(trans, root, leaf); + BUG_ON(ret); + free_extent_buffer(leaf); + } + atomic_inc(&root->fs_info->throttle_gen); + wake_up(&root->fs_info->transaction_throttle); + cond_resched(); + } + + /* + * run through the loop again to free the refs on the leaves. + * This is faster than doing it in the loop above because + * the leaves are likely to be clustered together. We end up + * working in nice chunks on the extent allocation tree. + */ + for (i = 0; i < refi; i++) { + bytenr = sorted[i].bytenr; + ret = __btrfs_free_extent(trans, root, bytenr, + blocksize, eb->start, + root_owner, root_gen, 0, 1); + BUG_ON(ret); + + atomic_inc(&root->fs_info->throttle_gen); + wake_up(&root->fs_info->transaction_throttle); + cond_resched(); + } +out: + kfree(sorted); + + /* + * update the path to show we've processed the entire level 1 + * node. This will get saved into the root's drop_snapshot_progress + * field so these drops are not repeated again if this transaction + * commits. + */ + path->slots[1] = nritems; + return 0; +} + /* * helper function for drop_snapshot, this walks down the tree dropping ref * counts as it goes. @@ -3580,7 +3786,6 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, struct extent_buffer *next; struct extent_buffer *cur; struct extent_buffer *parent; - struct btrfs_leaf_ref *ref; u32 blocksize; int ret; u32 refs; @@ -3607,17 +3812,46 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, if (path->slots[*level] >= btrfs_header_nritems(cur)) break; + + /* the new code goes down to level 1 and does all the + * leaves pointed to that node in bulk. So, this check + * for level 0 will always be false. + * + * But, the disk format allows the drop_snapshot_progress + * field in the root to leave things in a state where + * a leaf will need cleaning up here. If someone crashes + * with the old code and then boots with the new code, + * we might find a leaf here. + */ if (*level == 0) { ret = btrfs_drop_leaf_ref(trans, root, cur); BUG_ON(ret); break; } + + /* + * once we get to level one, process the whole node + * at once, including everything below it. + */ + if (*level == 1) { + ret = drop_level_one_refs(trans, root, path); + BUG_ON(ret); + break; + } + bytenr = btrfs_node_blockptr(cur, path->slots[*level]); ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); blocksize = btrfs_level_size(root, *level - 1); ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); BUG_ON(ret); + + /* + * if there is more than one reference, we don't need + * to read that node to drop any references it has. We + * just drop the ref we hold on that node and move on to the + * next slot in this level. + */ if (refs != 1) { parent = path->nodes[*level]; root_owner = btrfs_header_owner(parent); @@ -3636,46 +3870,12 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, continue; } + /* - * at this point, we have a single ref, and since the - * only place referencing this extent is a dead root - * the reference count should never go higher. - * So, we don't need to check it again + * we need to keep freeing things in the next level down. + * read the block and loop around to process it */ - if (*level == 1) { - ref = btrfs_lookup_leaf_ref(root, bytenr); - if (ref && ref->generation != ptr_gen) { - btrfs_free_leaf_ref(root, ref); - ref = NULL; - } - if (ref) { - ret = cache_drop_leaf_ref(trans, root, ref); - BUG_ON(ret); - btrfs_remove_leaf_ref(root, ref); - btrfs_free_leaf_ref(root, ref); - *level = 0; - break; - } - } - next = btrfs_find_tree_block(root, bytenr, blocksize); - if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { - free_extent_buffer(next); - - next = read_tree_block(root, bytenr, blocksize, - ptr_gen); - cond_resched(); -#if 0 - /* - * this is a debugging check and can go away - * the ref should never go all the way down to 1 - * at this point - */ - ret = lookup_extent_ref(NULL, root, bytenr, blocksize, - &refs); - BUG_ON(ret); - WARN_ON(refs != 1); -#endif - } + next = read_tree_block(root, bytenr, blocksize, ptr_gen); WARN_ON(*level <= 0); if (path->nodes[*level-1]) free_extent_buffer(path->nodes[*level-1]); @@ -3700,11 +3900,16 @@ out: root_owner = btrfs_header_owner(parent); root_gen = btrfs_header_generation(parent); + /* + * cleanup and free the reference on the last node + * we processed + */ ret = __btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, root_owner, root_gen, *level, 1); free_extent_buffer(path->nodes[*level]); path->nodes[*level] = NULL; + *level += 1; BUG_ON(ret); @@ -3824,6 +4029,13 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans, if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { struct extent_buffer *node; struct btrfs_disk_key disk_key; + + /* + * there is more work to do in this level. + * Update the drop_progress marker to reflect + * the work we've done so far, and then bump + * the slot number + */ node = path->nodes[i]; path->slots[i]++; *level = i; @@ -3835,6 +4047,11 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans, return 0; } else { struct extent_buffer *parent; + + /* + * this whole node is done, free our reference + * on it and go up one level + */ if (path->nodes[*level] == root->node) parent = path->nodes[*level]; else @@ -4849,6 +5066,7 @@ int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, ref->bytenr = buf->start; ref->owner = btrfs_header_owner(buf); ref->generation = btrfs_header_generation(buf); + ret = btrfs_add_leaf_ref(root, ref, 0); WARN_ON(ret); btrfs_free_leaf_ref(root, ref); diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index ebd7d6c37df..95ea58cb306 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2441,6 +2441,8 @@ next_node: ref->generation = leaf_gen; ref->nritems = 0; + btrfs_sort_leaf_ref(ref); + ret = btrfs_add_leaf_ref(root, ref, 0); WARN_ON(ret); btrfs_free_leaf_ref(root, ref); diff --git a/fs/btrfs/ref-cache.c b/fs/btrfs/ref-cache.c index 6f0acc4c9ea..d0cc62bccb9 100644 --- a/fs/btrfs/ref-cache.c +++ b/fs/btrfs/ref-cache.c @@ -17,6 +17,7 @@ */ #include +#include #include "ctree.h" #include "ref-cache.h" #include "transaction.h" diff --git a/fs/btrfs/ref-cache.h b/fs/btrfs/ref-cache.h index 16f3183d7c5..bc283ad2db7 100644 --- a/fs/btrfs/ref-cache.h +++ b/fs/btrfs/ref-cache.h @@ -73,5 +73,4 @@ int btrfs_add_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref, int btrfs_remove_leaf_refs(struct btrfs_root *root, u64 max_root_gen, int shared); int btrfs_remove_leaf_ref(struct btrfs_root *root, struct btrfs_leaf_ref *ref); - #endif -- cgit v1.2.3-70-g09d2 From 8c087b5183adab186a298f2d6ed39aefdcae413c Mon Sep 17 00:00:00 2001 From: Chris Ball Date: Wed, 4 Feb 2009 09:29:54 -0500 Subject: Btrfs: Handle SGID bit when creating inodes Before this patch, new files/dirs would ignore the SGID bit on their parent directory and always be owned by the creating user's uid/gid. Signed-off-by: Chris Ball Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 95ea58cb306..5792816d4fc 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3472,7 +3472,14 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, root->highest_inode = objectid; inode->i_uid = current_fsuid(); - inode->i_gid = current_fsgid(); + + if (dir->i_mode & S_ISGID) { + inode->i_gid = dir->i_gid; + if (S_ISDIR(mode)) + mode |= S_ISGID; + } else + inode->i_gid = current_fsgid(); + inode->i_mode = mode; inode->i_ino = objectid; inode_set_bytes(inode, 0); -- cgit v1.2.3-70-g09d2 From f03d9301f15fb69cdf1eb59d53c9fb72f68ecccc Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 4 Feb 2009 09:31:06 -0500 Subject: Btrfs: Don't try to compress pages past i_size The compression code had some checks to make sure we were only compressing bytes inside of i_size, but it wasn't catching every case. To make things worse, some incorrect math about the number of bytes remaining would make it try to compress more pages than the file really had. The fix used here is to fall back to the non-compression code in this case, which does all the proper cleanup of delalloc and other accounting. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 5792816d4fc..9b43a6f303b 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -360,6 +360,19 @@ again: nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); + /* + * we don't want to send crud past the end of i_size through + * compression, that's just a waste of CPU time. So, if the + * end of the file is before the start of our current + * requested range of bytes, we bail out to the uncompressed + * cleanup code that can deal with all of this. + * + * It isn't really the fastest way to fix things, but this is a + * very uncommon corner. + */ + if (actual_end <= start) + goto cleanup_and_bail_uncompressed; + total_compressed = actual_end - start; /* we want to make sure that amount of ram required to uncompress @@ -504,6 +517,7 @@ again: goto again; } } else { +cleanup_and_bail_uncompressed: /* * No compression, but we still need to write the pages in * the file we've been given so far. redirty the locked -- cgit v1.2.3-70-g09d2 From 06d9a8d7c24fe22836bf0b0f82db59d6f98e271e Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 4 Feb 2009 09:30:58 -0500 Subject: Btrfs: Change btrfs_truncate_inode_items to stop when it hits the inode btrfs_truncate_inode_items is setup to stop doing btree searches when it has finished removing the items for the inode. It used to detect the end of the inode by looking for an objectid that didn't match the one we were searching for. But, this would result in an extra search through the btree, which adds extra balancing and cow costs to the operation. This commit adds a check to see if we found the inode item, which means we can stop searching early. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 9b43a6f303b..ddb0f0ecda6 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2504,7 +2504,7 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, struct btrfs_path *path; struct btrfs_key key; struct btrfs_key found_key; - u32 found_type; + u32 found_type = (u8)-1; struct extent_buffer *leaf; struct btrfs_file_extent_item *fi; u64 extent_start = 0; @@ -2691,6 +2691,8 @@ next: if (pending_del_nr) goto del_pending; btrfs_release_path(root, path); + if (found_type == BTRFS_INODE_ITEM_KEY) + break; goto search_again; } @@ -2707,6 +2709,8 @@ del_pending: BUG_ON(ret); pending_del_nr = 0; btrfs_release_path(root, path); + if (found_type == BTRFS_INODE_ITEM_KEY) + break; goto search_again; } } -- cgit v1.2.3-70-g09d2 From 42f15d77df8a7e8a2feb15041d5d30710ee7f951 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 6 Feb 2009 11:35:57 -0500 Subject: Btrfs: Make sure dir is non-null before doing S_ISGID checks The S_ISGID check in btrfs_new_inode caused an oops during subvol creation because sometimes the dir is null. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index ddb0f0ecda6..8f0706210a4 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -3491,7 +3491,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, inode->i_uid = current_fsuid(); - if (dir->i_mode & S_ISGID) { + if (dir && (dir->i_mode & S_ISGID)) { inode->i_gid = dir->i_gid; if (S_ISDIR(mode)) mode |= S_ISGID; -- cgit v1.2.3-70-g09d2 From b335b0034e252e79ec2e9c6697f5d663c4627bec Mon Sep 17 00:00:00 2001 From: Yan Zheng Date: Thu, 12 Feb 2009 10:06:04 -0500 Subject: Btrfs: Avoid using __GFP_HIGHMEM with slab allocator btrfs_releasepage may call kmem_cache_alloc indirectly, and provide same GFP flags it gets to kmem_cache_alloc. So it's possible to use __GFP_HIGHMEM with the slab allocator. Signed-off-by: Yan Zheng --- fs/btrfs/inode.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 8f0706210a4..638bcb5e49f 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4263,7 +4263,7 @@ static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) { if (PageWriteback(page) || PageDirty(page)) return 0; - return __btrfs_releasepage(page, gfp_flags); + return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); } static void btrfs_invalidatepage(struct page *page, unsigned long offset) -- cgit v1.2.3-70-g09d2 From e00f7308658622fbd483cb0d9fe41165bf9050d0 Mon Sep 17 00:00:00 2001 From: Jeff Mahoney Date: Thu, 12 Feb 2009 14:11:25 -0500 Subject: Btrfs: remove btrfs_init_path btrfs_init_path was initially used when the path objects were on the stack. Now all the work is done by btrfs_alloc_path and btrfs_init_path isn't required. This patch removes it, and just uses kmem_cache_zalloc to zero out the object. Signed-off-by: Chris Mason --- fs/btrfs/ctree.c | 11 ++--------- fs/btrfs/ctree.h | 1 - fs/btrfs/inode-map.c | 1 - fs/btrfs/inode.c | 2 -- 4 files changed, 2 insertions(+), 13 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index 6674692f702..c8f4c540cc2 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -38,19 +38,12 @@ static int balance_node_right(struct btrfs_trans_handle *trans, static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, int level, int slot); -inline void btrfs_init_path(struct btrfs_path *p) -{ - memset(p, 0, sizeof(*p)); -} - struct btrfs_path *btrfs_alloc_path(void) { struct btrfs_path *path; - path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS); - if (path) { - btrfs_init_path(path); + path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); + if (path) path->reada = 1; - } return path; } diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 531db112c8b..3f7a8058df2 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -1834,7 +1834,6 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans, void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p); struct btrfs_path *btrfs_alloc_path(void); void btrfs_free_path(struct btrfs_path *p); -void btrfs_init_path(struct btrfs_path *p); void btrfs_set_path_blocking(struct btrfs_path *p); void btrfs_clear_path_blocking(struct btrfs_path *p); void btrfs_unlock_up_safe(struct btrfs_path *p, int level); diff --git a/fs/btrfs/inode-map.c b/fs/btrfs/inode-map.c index 2aa79873eb4..cc7334d833c 100644 --- a/fs/btrfs/inode-map.c +++ b/fs/btrfs/inode-map.c @@ -84,7 +84,6 @@ int btrfs_find_free_objectid(struct btrfs_trans_handle *trans, search_key.type = 0; search_key.offset = 0; - btrfs_init_path(path); start_found = 0; ret = btrfs_search_slot(trans, root, &search_key, path, 0, 0); if (ret < 0) diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 638bcb5e49f..3cee77ae03c 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2531,8 +2531,6 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, key.offset = (u64)-1; key.type = (u8)-1; - btrfs_init_path(path); - search_again: ret = btrfs_search_slot(trans, root, &key, path, -1, 1); if (ret < 0) -- cgit v1.2.3-70-g09d2 From 6a63209fc02d5483371f07e4913ee8abad608051 Mon Sep 17 00:00:00 2001 From: Josef Bacik Date: Fri, 20 Feb 2009 11:00:09 -0500 Subject: Btrfs: add better -ENOSPC handling This is a step in the direction of better -ENOSPC handling. Instead of checking the global bytes counter we check the space_info bytes counters to make sure we have enough space. If we don't we go ahead and try to allocate a new chunk, and then if that fails we return -ENOSPC. This patch adds two counters to btrfs_space_info, bytes_delalloc and bytes_may_use. bytes_delalloc account for extents we've actually setup for delalloc and will be allocated at some point down the line. bytes_may_use is to keep track of how many bytes we may use for delalloc at some point. When we actually set the extent_bit for the delalloc bytes we subtract the reserved bytes from the bytes_may_use counter. This keeps us from not actually being able to allocate space for any delalloc bytes. Signed-off-by: Josef Bacik --- fs/btrfs/btrfs_inode.h | 8 ++ fs/btrfs/ctree.h | 40 ++++++--- fs/btrfs/extent-tree.c | 215 +++++++++++++++++++++++++++++++++++++++++++++---- fs/btrfs/file.c | 16 +++- fs/btrfs/inode.c | 62 ++++---------- fs/btrfs/ioctl.c | 6 +- 6 files changed, 271 insertions(+), 76 deletions(-) (limited to 'fs/btrfs/inode.c') diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h index a8c9693b75a..72677ce2b74 100644 --- a/fs/btrfs/btrfs_inode.h +++ b/fs/btrfs/btrfs_inode.h @@ -66,6 +66,9 @@ struct btrfs_inode { */ struct list_head delalloc_inodes; + /* the space_info for where this inode's data allocations are done */ + struct btrfs_space_info *space_info; + /* full 64 bit generation number, struct vfs_inode doesn't have a big * enough field for this. */ @@ -94,6 +97,11 @@ struct btrfs_inode { */ u64 delalloc_bytes; + /* total number of bytes that may be used for this inode for + * delalloc + */ + u64 reserved_bytes; + /* * the size of the file stored in the metadata on disk. data=ordered * means the in-memory i_size might be larger than the size on disk diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 766b31ae318..82491ba8fa4 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -596,13 +596,27 @@ struct btrfs_block_group_item { struct btrfs_space_info { u64 flags; - u64 total_bytes; - u64 bytes_used; - u64 bytes_pinned; - u64 bytes_reserved; - u64 bytes_readonly; - int full; - int force_alloc; + + u64 total_bytes; /* total bytes in the space */ + u64 bytes_used; /* total bytes used on disk */ + u64 bytes_pinned; /* total bytes pinned, will be freed when the + transaction finishes */ + u64 bytes_reserved; /* total bytes the allocator has reserved for + current allocations */ + u64 bytes_readonly; /* total bytes that are read only */ + + /* delalloc accounting */ + u64 bytes_delalloc; /* number of bytes reserved for allocation, + this space is not necessarily reserved yet + by the allocator */ + u64 bytes_may_use; /* number of bytes that may be used for + delalloc */ + + int full; /* indicates that we cannot allocate any more + chunks for this space */ + int force_alloc; /* set if we need to force a chunk alloc for + this space */ + struct list_head list; /* for block groups in our same type */ @@ -1782,6 +1796,16 @@ int btrfs_add_dead_reloc_root(struct btrfs_root *root); int btrfs_cleanup_reloc_trees(struct btrfs_root *root); int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len); u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags); +void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *ionde); +int btrfs_check_metadata_free_space(struct btrfs_root *root); +int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes); +void btrfs_free_reserved_data_space(struct btrfs_root *root, + struct inode *inode, u64 bytes); +void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, + u64 bytes); +void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes); /* ctree.c */ int btrfs_previous_item(struct btrfs_root *root, struct btrfs_path *path, u64 min_objectid, @@ -2027,8 +2051,6 @@ 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_check_free_space(struct btrfs_root *root, u64 num_required, - int for_del); int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page); int btrfs_readpage(struct file *file, struct page *page); void btrfs_delete_inode(struct inode *inode); diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 0a5d796c9f7..e11875e97c2 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -60,6 +60,10 @@ static int update_block_group(struct btrfs_trans_handle *trans, u64 bytenr, u64 num_bytes, int alloc, int mark_free); +static int do_chunk_alloc(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, u64 alloc_bytes, + u64 flags, int force); + static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) { return (cache->flags & bits) == bits; @@ -1909,6 +1913,7 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags, found->bytes_pinned = 0; found->bytes_reserved = 0; found->bytes_readonly = 0; + found->bytes_delalloc = 0; found->full = 0; found->force_alloc = 0; *space_info = found; @@ -1972,6 +1977,196 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) return flags; } +static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) +{ + struct btrfs_fs_info *info = root->fs_info; + u64 alloc_profile; + + if (data) { + alloc_profile = info->avail_data_alloc_bits & + info->data_alloc_profile; + data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; + } else if (root == root->fs_info->chunk_root) { + alloc_profile = info->avail_system_alloc_bits & + info->system_alloc_profile; + data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; + } else { + alloc_profile = info->avail_metadata_alloc_bits & + info->metadata_alloc_profile; + data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; + } + + return btrfs_reduce_alloc_profile(root, data); +} + +void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) +{ + u64 alloc_target; + + alloc_target = btrfs_get_alloc_profile(root, 1); + BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, + alloc_target); +} + +/* + * for now this just makes sure we have at least 5% of our metadata space free + * for use. + */ +int btrfs_check_metadata_free_space(struct btrfs_root *root) +{ + struct btrfs_fs_info *info = root->fs_info; + struct btrfs_space_info *meta_sinfo; + u64 alloc_target, thresh; + + /* get the space info for where the metadata will live */ + alloc_target = btrfs_get_alloc_profile(root, 0); + meta_sinfo = __find_space_info(info, alloc_target); + + /* + * if the metadata area isn't maxed out then there is no sense in + * checking how much is used, since we can always allocate a new chunk + */ + if (!meta_sinfo->full) + return 0; + + spin_lock(&meta_sinfo->lock); + thresh = meta_sinfo->total_bytes * 95; + + do_div(thresh, 100); + + if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + + meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { + spin_unlock(&meta_sinfo->lock); + return -ENOSPC; + } + spin_unlock(&meta_sinfo->lock); + + return 0; +} + +/* + * This will check the space that the inode allocates from to make sure we have + * enough space for bytes. + */ +int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + int ret = 0; + + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); + + data_sinfo = BTRFS_I(inode)->space_info; +again: + /* make sure we have enough space to handle the data first */ + spin_lock(&data_sinfo->lock); + if (data_sinfo->total_bytes - data_sinfo->bytes_used - + data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - + data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - + data_sinfo->bytes_may_use < bytes) { + /* + * if we don't have enough free bytes in this space then we need + * to alloc a new chunk. + */ + if (!data_sinfo->full) { + u64 alloc_target; + struct btrfs_trans_handle *trans; + + data_sinfo->force_alloc = 1; + spin_unlock(&data_sinfo->lock); + + alloc_target = btrfs_get_alloc_profile(root, 1); + trans = btrfs_start_transaction(root, 1); + if (!trans) + return -ENOMEM; + + ret = do_chunk_alloc(trans, root->fs_info->extent_root, + bytes + 2 * 1024 * 1024, + alloc_target, 0); + btrfs_end_transaction(trans, root); + if (ret) + return ret; + goto again; + } + spin_unlock(&data_sinfo->lock); + printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" + ", %llu bytes_used, %llu bytes_reserved, " + "%llu bytes_pinned, %llu bytes_readonly, %llu may use" + "%llu total\n", bytes, data_sinfo->bytes_delalloc, + data_sinfo->bytes_used, data_sinfo->bytes_reserved, + data_sinfo->bytes_pinned, data_sinfo->bytes_readonly, + data_sinfo->bytes_may_use, data_sinfo->total_bytes); + return -ENOSPC; + } + data_sinfo->bytes_may_use += bytes; + BTRFS_I(inode)->reserved_bytes += bytes; + spin_unlock(&data_sinfo->lock); + + return btrfs_check_metadata_free_space(root); +} + +/* + * if there was an error for whatever reason after calling + * btrfs_check_data_free_space, call this so we can cleanup the counters. + */ +void btrfs_free_reserved_data_space(struct btrfs_root *root, + struct inode *inode, u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); + + data_sinfo = BTRFS_I(inode)->space_info; + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; + spin_unlock(&data_sinfo->lock); +} + +/* called when we are adding a delalloc extent to the inode's io_tree */ +void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + + /* get the space info for where this inode will be storing its data */ + data_sinfo = BTRFS_I(inode)->space_info; + + /* make sure we have enough space to handle the data first */ + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_delalloc += bytes; + + /* + * we are adding a delalloc extent without calling + * btrfs_check_data_free_space first. This happens on a weird + * writepage condition, but shouldn't hurt our accounting + */ + if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { + data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; + BTRFS_I(inode)->reserved_bytes = 0; + } else { + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; + } + + spin_unlock(&data_sinfo->lock); +} + +/* called when we are clearing an delalloc extent from the inode's io_tree */ +void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *info; + + info = BTRFS_I(inode)->space_info; + + spin_lock(&info->lock); + info->bytes_delalloc -= bytes; + spin_unlock(&info->lock); +} + static int do_chunk_alloc(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root, u64 alloc_bytes, u64 flags, int force) @@ -3105,6 +3300,10 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes) (unsigned long long)(info->total_bytes - info->bytes_used - info->bytes_pinned - info->bytes_reserved), (info->full) ? "" : "not "); + printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," + " may_use=%llu, used=%llu\n", info->total_bytes, + info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use, + info->bytes_used); down_read(&info->groups_sem); list_for_each_entry(cache, &info->block_groups, list) { @@ -3131,24 +3330,10 @@ static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, { int ret; u64 search_start = 0; - u64 alloc_profile; struct btrfs_fs_info *info = root->fs_info; - if (data) { - alloc_profile = info->avail_data_alloc_bits & - info->data_alloc_profile; - data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; - } else if (root == root->fs_info->chunk_root) { - alloc_profile = info->avail_system_alloc_bits & - info->system_alloc_profile; - data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; - } else { - alloc_profile = info->avail_metadata_alloc_bits & - info->metadata_alloc_profile; - data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; - } + data = btrfs_get_alloc_profile(root, data); again: - data = btrfs_reduce_alloc_profile(root, data); /* * the only place that sets empty_size is btrfs_realloc_node, which * is not called recursively on allocations diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index 872f104576e..dc78954861b 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -1091,19 +1091,24 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf, WARN_ON(num_pages > nrptrs); memset(pages, 0, sizeof(struct page *) * nrptrs); - ret = btrfs_check_free_space(root, write_bytes, 0); + ret = btrfs_check_data_free_space(root, inode, write_bytes); if (ret) goto out; ret = prepare_pages(root, file, pages, num_pages, pos, first_index, last_index, write_bytes); - if (ret) + if (ret) { + btrfs_free_reserved_data_space(root, inode, + write_bytes); goto out; + } ret = btrfs_copy_from_user(pos, num_pages, write_bytes, pages, buf); if (ret) { + btrfs_free_reserved_data_space(root, inode, + write_bytes); btrfs_drop_pages(pages, num_pages); goto out; } @@ -1111,8 +1116,11 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf, ret = dirty_and_release_pages(NULL, root, file, pages, num_pages, pos, write_bytes); btrfs_drop_pages(pages, num_pages); - if (ret) + if (ret) { + btrfs_free_reserved_data_space(root, inode, + write_bytes); goto out; + } if (will_write) { btrfs_fdatawrite_range(inode->i_mapping, pos, @@ -1136,6 +1144,8 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf, } out: mutex_unlock(&inode->i_mutex); + if (ret) + err = ret; out_nolock: kfree(pages); diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 3cee77ae03c..7d4f948bc22 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -101,34 +101,6 @@ static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) return err; } -/* - * a very lame attempt at stopping writes when the FS is 85% full. There - * are countless ways this is incorrect, but it is better than nothing. - */ -int btrfs_check_free_space(struct btrfs_root *root, u64 num_required, - int for_del) -{ - u64 total; - u64 used; - u64 thresh; - int ret = 0; - - spin_lock(&root->fs_info->delalloc_lock); - total = btrfs_super_total_bytes(&root->fs_info->super_copy); - used = btrfs_super_bytes_used(&root->fs_info->super_copy); - if (for_del) - thresh = total * 90; - else - thresh = total * 85; - - do_div(thresh, 100); - - if (used + root->fs_info->delalloc_bytes + num_required > thresh) - ret = -ENOSPC; - spin_unlock(&root->fs_info->delalloc_lock); - return ret; -} - /* * this does all the hard work for inserting an inline extent into * the btree. The caller should have done a btrfs_drop_extents so that @@ -1190,6 +1162,7 @@ static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, */ if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { struct btrfs_root *root = BTRFS_I(inode)->root; + btrfs_delalloc_reserve_space(root, inode, end - start + 1); spin_lock(&root->fs_info->delalloc_lock); BTRFS_I(inode)->delalloc_bytes += end - start + 1; root->fs_info->delalloc_bytes += end - start + 1; @@ -1223,9 +1196,12 @@ static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, (unsigned long long)end - start + 1, (unsigned long long) root->fs_info->delalloc_bytes); + btrfs_delalloc_free_space(root, inode, (u64)-1); root->fs_info->delalloc_bytes = 0; BTRFS_I(inode)->delalloc_bytes = 0; } else { + btrfs_delalloc_free_space(root, inode, + end - start + 1); root->fs_info->delalloc_bytes -= end - start + 1; BTRFS_I(inode)->delalloc_bytes -= end - start + 1; } @@ -2245,10 +2221,6 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry) root = BTRFS_I(dir)->root; - ret = btrfs_check_free_space(root, 1, 1); - if (ret) - goto fail; - trans = btrfs_start_transaction(root, 1); btrfs_set_trans_block_group(trans, dir); @@ -2261,7 +2233,6 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry) nr = trans->blocks_used; btrfs_end_transaction_throttle(trans, root); -fail: btrfs_btree_balance_dirty(root, nr); return ret; } @@ -2284,10 +2255,6 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) return -ENOTEMPTY; } - ret = btrfs_check_free_space(root, 1, 1); - if (ret) - goto fail; - trans = btrfs_start_transaction(root, 1); btrfs_set_trans_block_group(trans, dir); @@ -2304,7 +2271,6 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) fail_trans: nr = trans->blocks_used; ret = btrfs_end_transaction_throttle(trans, root); -fail: btrfs_btree_balance_dirty(root, nr); if (ret && !err) @@ -2818,7 +2784,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size) if (size <= hole_start) return 0; - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) return err; @@ -3014,6 +2980,7 @@ static noinline void init_btrfs_i(struct inode *inode) bi->last_trans = 0; bi->logged_trans = 0; bi->delalloc_bytes = 0; + bi->reserved_bytes = 0; bi->disk_i_size = 0; bi->flags = 0; bi->index_cnt = (u64)-1; @@ -3035,6 +3002,7 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p) inode->i_ino = args->ino; init_btrfs_i(inode); BTRFS_I(inode)->root = args->root; + btrfs_set_inode_space_info(args->root, inode); return 0; } @@ -3455,6 +3423,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, BTRFS_I(inode)->index_cnt = 2; BTRFS_I(inode)->root = root; BTRFS_I(inode)->generation = trans->transid; + btrfs_set_inode_space_info(root, inode); if (mode & S_IFDIR) owner = 0; @@ -3602,7 +3571,7 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry, if (!new_valid_dev(rdev)) return -EINVAL; - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) goto fail; @@ -3665,7 +3634,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry, u64 objectid; u64 index = 0; - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) goto fail; trans = btrfs_start_transaction(root, 1); @@ -3733,7 +3702,7 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir, return -ENOENT; btrfs_inc_nlink(inode); - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) goto fail; err = btrfs_set_inode_index(dir, &index); @@ -3779,7 +3748,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) u64 index = 0; unsigned long nr = 1; - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) goto out_unlock; @@ -4336,7 +4305,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) u64 page_start; u64 page_end; - ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0); + ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); if (ret) goto out; @@ -4349,6 +4318,7 @@ again: if ((page->mapping != inode->i_mapping) || (page_start >= size)) { + btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); /* page got truncated out from underneath us */ goto out_unlock; } @@ -4631,7 +4601,7 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) return -EXDEV; - ret = btrfs_check_free_space(root, 1, 0); + ret = btrfs_check_metadata_free_space(root); if (ret) goto out_unlock; @@ -4749,7 +4719,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry, if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) return -ENAMETOOLONG; - err = btrfs_check_free_space(root, 1, 0); + err = btrfs_check_metadata_free_space(root); if (err) goto out_fail; diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c index 988fdc8b49e..bca729fc80c 100644 --- a/fs/btrfs/ioctl.c +++ b/fs/btrfs/ioctl.c @@ -70,7 +70,7 @@ static noinline int create_subvol(struct btrfs_root *root, u64 index = 0; unsigned long nr = 1; - ret = btrfs_check_free_space(root, 1, 0); + ret = btrfs_check_metadata_free_space(root); if (ret) goto fail_commit; @@ -203,7 +203,7 @@ static int create_snapshot(struct btrfs_root *root, struct dentry *dentry, if (!root->ref_cows) return -EINVAL; - ret = btrfs_check_free_space(root, 1, 0); + ret = btrfs_check_metadata_free_space(root); if (ret) goto fail_unlock; @@ -374,7 +374,7 @@ static int btrfs_defrag_file(struct file *file) unsigned long i; int ret; - ret = btrfs_check_free_space(root, inode->i_size, 0); + ret = btrfs_check_data_free_space(root, inode, inode->i_size); if (ret) return -ENOSPC; -- cgit v1.2.3-70-g09d2