From dcc3eb9638c3c927f1597075e851d0a16300a876 Mon Sep 17 00:00:00 2001 From: Nikolay Borisov Date: Thu, 30 Jan 2020 14:59:45 +0200 Subject: [PATCH] btrfs: convert snapshot/nocow exlcusion to drew lock This patch removes all haphazard code implementing nocow writers exclusion from pending snapshot creation and switches to using the drew lock to ensure this invariant still holds. 'Readers' are snapshot creators from create_snapshot and 'writers' are nocow writers from buffered write path or btrfs_setsize. This locking scheme allows for multiple snapshots to happen while any nocow writers are blocked, since writes to page cache in the nocow path will make snapshots inconsistent. So for performance reasons we'd like to have the ability to run multiple concurrent snapshots and also favors readers in this case. And in case there aren't pending snapshots (which will be the majority of the cases) we rely on the percpu's writers counter to avoid cacheline contention. The main gain from using the drew lock is it's now a lot easier to reason about the guarantees of the locking scheme and whether there is some silent breakage lurking. Signed-off-by: Nikolay Borisov Reviewed-by: David Sterba Signed-off-by: David Sterba --- fs/btrfs/ctree.h | 10 +++------- fs/btrfs/disk-io.c | 45 ++++++++++-------------------------------- fs/btrfs/extent-tree.c | 44 ----------------------------------------- fs/btrfs/file.c | 11 +++++------ fs/btrfs/inode.c | 8 ++++---- fs/btrfs/ioctl.c | 10 +++------- 6 files changed, 25 insertions(+), 103 deletions(-) diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index ab8151247b93..db9e872bcc79 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -957,11 +957,6 @@ static inline struct btrfs_fs_info *btrfs_sb(struct super_block *sb) return sb->s_fs_info; } -struct btrfs_subvolume_writers { - struct percpu_counter counter; - wait_queue_head_t wait; -}; - /* * The state of btrfs root */ @@ -1133,8 +1128,9 @@ struct btrfs_root { * root_item_lock. */ int dedupe_in_progress; - struct btrfs_subvolume_writers *subv_writers; - atomic_t will_be_snapshotted; + /* For exclusion of snapshot creation and nocow writes */ + struct btrfs_drew_lock snapshot_lock; + atomic_t snapshot_force_cow; /* For qgroup metadata reserved space */ diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 770d469e1d9c..06819c41e4f4 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1104,32 +1104,6 @@ void btrfs_clean_tree_block(struct extent_buffer *buf) } } -static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) -{ - struct btrfs_subvolume_writers *writers; - int ret; - - writers = kmalloc(sizeof(*writers), GFP_NOFS); - if (!writers) - return ERR_PTR(-ENOMEM); - - ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); - if (ret < 0) { - kfree(writers); - return ERR_PTR(ret); - } - - init_waitqueue_head(&writers->wait); - return writers; -} - -static void -btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) -{ - percpu_counter_destroy(&writers->counter); - kfree(writers); -} - static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, u64 objectid) { @@ -1178,7 +1152,6 @@ static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, atomic_set(&root->log_writers, 0); atomic_set(&root->log_batch, 0); refcount_set(&root->refs, 1); - atomic_set(&root->will_be_snapshotted, 0); atomic_set(&root->snapshot_force_cow, 0); atomic_set(&root->nr_swapfiles, 0); root->log_transid = 0; @@ -1450,7 +1423,7 @@ alloc_fail: static int btrfs_init_fs_root(struct btrfs_root *root) { int ret; - struct btrfs_subvolume_writers *writers; + unsigned int nofs_flag; root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), @@ -1460,12 +1433,15 @@ static int btrfs_init_fs_root(struct btrfs_root *root) goto fail; } - writers = btrfs_alloc_subvolume_writers(); - if (IS_ERR(writers)) { - ret = PTR_ERR(writers); + /* + * We might be called under a transaction (e.g. indirect backref + * resolution) which could deadlock if it triggers memory reclaim + */ + nofs_flag = memalloc_nofs_save(); + ret = btrfs_drew_lock_init(&root->snapshot_lock); + memalloc_nofs_restore(nofs_flag); + if (ret) goto fail; - } - root->subv_writers = writers; if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { set_bit(BTRFS_ROOT_REF_COWS, &root->state); @@ -3961,8 +3937,7 @@ void btrfs_free_fs_root(struct btrfs_root *root) WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); if (root->anon_dev) free_anon_bdev(root->anon_dev); - if (root->subv_writers) - btrfs_free_subvolume_writers(root->subv_writers); + btrfs_drew_lock_destroy(&root->snapshot_lock); free_extent_buffer(root->node); free_extent_buffer(root->commit_root); kfree(root->free_ino_ctl); diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 7eef91d6c2b6..9dcd70cc3ca3 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -5740,47 +5740,3 @@ int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) return bg_ret; return dev_ret; } - -/* - * btrfs_{start,end}_write_no_snapshotting() are similar to - * mnt_{want,drop}_write(), they are used to prevent some tasks from writing - * data into the page cache through nocow before the subvolume is snapshoted, - * but flush the data into disk after the snapshot creation, or to prevent - * operations while snapshotting is ongoing and that cause the snapshot to be - * inconsistent (writes followed by expanding truncates for example). - */ -void btrfs_end_write_no_snapshotting(struct btrfs_root *root) -{ - percpu_counter_dec(&root->subv_writers->counter); - cond_wake_up(&root->subv_writers->wait); -} - -int btrfs_start_write_no_snapshotting(struct btrfs_root *root) -{ - if (atomic_read(&root->will_be_snapshotted)) - return 0; - - percpu_counter_inc(&root->subv_writers->counter); - /* - * Make sure counter is updated before we check for snapshot creation. - */ - smp_mb(); - if (atomic_read(&root->will_be_snapshotted)) { - btrfs_end_write_no_snapshotting(root); - return 0; - } - return 1; -} - -void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) -{ - while (true) { - int ret; - - ret = btrfs_start_write_no_snapshotting(root); - if (ret) - break; - wait_var_event(&root->will_be_snapshotted, - !atomic_read(&root->will_be_snapshotted)); - } -} diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index fd52ad00b6c8..8a974a82be51 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -1553,8 +1553,7 @@ static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos, u64 num_bytes; int ret; - ret = btrfs_start_write_no_snapshotting(root); - if (!ret) + if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) return -EAGAIN; lockstart = round_down(pos, fs_info->sectorsize); @@ -1569,7 +1568,7 @@ static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos, NULL, NULL, NULL); if (ret <= 0) { ret = 0; - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); } else { *write_bytes = min_t(size_t, *write_bytes , num_bytes - pos + lockstart); @@ -1675,7 +1674,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, data_reserved, pos, write_bytes); else - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); break; } @@ -1779,7 +1778,7 @@ again: release_bytes = 0; if (only_release_metadata) - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); if (only_release_metadata && copied > 0) { lockstart = round_down(pos, @@ -1808,7 +1807,7 @@ again: if (release_bytes) { if (only_release_metadata) { - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); btrfs_delalloc_release_metadata(BTRFS_I(inode), release_bytes, true); } else { diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 254f5ea17e40..1e138c83cc6e 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4726,16 +4726,16 @@ static int btrfs_setsize(struct inode *inode, struct iattr *attr) * truncation, it must capture all writes that happened before * this truncation. */ - btrfs_wait_for_snapshot_creation(root); + btrfs_drew_write_lock(&root->snapshot_lock); ret = btrfs_cont_expand(inode, oldsize, newsize); if (ret) { - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); return ret; } trans = btrfs_start_transaction(root, 1); if (IS_ERR(trans)) { - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); return PTR_ERR(trans); } @@ -4743,7 +4743,7 @@ static int btrfs_setsize(struct inode *inode, struct iattr *attr) btrfs_inode_safe_disk_i_size_write(inode, 0); pagecache_isize_extended(inode, oldsize, newsize); ret = btrfs_update_inode(trans, root, inode); - btrfs_end_write_no_snapshotting(root); + btrfs_drew_write_unlock(&root->snapshot_lock); btrfs_end_transaction(trans); } else { diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c index 56bd3ea7fb67..6ded5e346821 100644 --- a/fs/btrfs/ioctl.c +++ b/fs/btrfs/ioctl.c @@ -791,11 +791,7 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir, * possible. This is to avoid later writeback (running dealloc) to * fallback to COW mode and unexpectedly fail with ENOSPC. */ - atomic_inc(&root->will_be_snapshotted); - smp_mb__after_atomic(); - /* wait for no snapshot writes */ - wait_event(root->subv_writers->wait, - percpu_counter_sum(&root->subv_writers->counter) == 0); + btrfs_drew_read_lock(&root->snapshot_lock); ret = btrfs_start_delalloc_snapshot(root); if (ret) @@ -876,8 +872,8 @@ fail: dec_and_free: if (snapshot_force_cow) atomic_dec(&root->snapshot_force_cow); - if (atomic_dec_and_test(&root->will_be_snapshotted)) - wake_up_var(&root->will_be_snapshotted); + btrfs_drew_read_unlock(&root->snapshot_lock); + free_pending: kfree(pending_snapshot->root_item); btrfs_free_path(pending_snapshot->path);