From bb721703aa551e98dc5c7fb259cf90343408baf2 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 29 Jan 2013 18:44:12 -0500 Subject: Btrfs: reduce CPU contention while waiting for delayed extent operations We batch up operations to the extent allocation tree, which allows us to deal with the recursive nature of using the extent allocation tree to allocate extents to the extent allocation tree. It also provides a mechanism to sort and collect extent operations, which makes it much more efficient to record extents that are close together. The delayed extent operations must all be finished before the running transaction commits, so we have code to make sure and run a few of the batched operations when closing our transaction handles. This creates a great deal of contention for the locks in the delayed extent operation tree, and also contention for the lock on the extent allocation tree itself. All the extra contention just slows down the operations and doesn't get things done any faster. This commit changes things to use a wait queue instead. As procs want to run the delayed operations, one of them races in and gets permission to hit the tree, and the others step back and wait for progress to be made. Signed-off-by: Chris Mason --- fs/btrfs/delayed-ref.h | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'fs/btrfs/delayed-ref.h') diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h index c9d703693df0..23bdeb8502a6 100644 --- a/fs/btrfs/delayed-ref.h +++ b/fs/btrfs/delayed-ref.h @@ -131,6 +131,15 @@ struct btrfs_delayed_ref_root { /* total number of head nodes ready for processing */ unsigned long num_heads_ready; + /* + * bumped when someone is making progress on the delayed + * refs, so that other procs know they are just adding to + * contention intead of helping + */ + atomic_t procs_running_refs; + atomic_t ref_seq; + wait_queue_head_t wait; + /* * set when the tree is flushing before a transaction commit, * used by the throttling code to decide if new updates need -- cgit v1.2.3 From 78a6184a3ff9041280ee56273c01e5679a831b39 Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Wed, 21 Nov 2012 02:21:28 +0000 Subject: Btrfs: use slabs for delayed reference allocation The delayed reference allocation is in the fast path of the IO, so use slabs to improve the speed of the allocation. And besides that, it can do check for leaked objects when the module is removed. Signed-off-by: Miao Xie --- fs/btrfs/delayed-ref.c | 74 ++++++++++++++++++++++++++++++++++++++++++-------- fs/btrfs/delayed-ref.h | 37 ++++++++++++++++++++++++- fs/btrfs/disk-io.c | 2 +- fs/btrfs/extent-tree.c | 14 +++++----- fs/btrfs/super.c | 9 +++++- 5 files changed, 115 insertions(+), 21 deletions(-) (limited to 'fs/btrfs/delayed-ref.h') diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c index ae9411773397..455894f1ca3b 100644 --- a/fs/btrfs/delayed-ref.c +++ b/fs/btrfs/delayed-ref.c @@ -23,6 +23,10 @@ #include "delayed-ref.h" #include "transaction.h" +struct kmem_cache *btrfs_delayed_ref_head_cachep; +struct kmem_cache *btrfs_delayed_tree_ref_cachep; +struct kmem_cache *btrfs_delayed_data_ref_cachep; +struct kmem_cache *btrfs_delayed_extent_op_cachep; /* * delayed back reference update tracking. For subvolume trees * we queue up extent allocations and backref maintenance for @@ -511,7 +515,7 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing, ref->extent_op->flags_to_set; existing_ref->extent_op->update_flags = 1; } - kfree(ref->extent_op); + btrfs_free_delayed_extent_op(ref->extent_op); } } /* @@ -592,7 +596,7 @@ static noinline void add_delayed_ref_head(struct btrfs_fs_info *fs_info, * we've updated the existing ref, free the newly * allocated ref */ - kfree(head_ref); + kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); } else { delayed_refs->num_heads++; delayed_refs->num_heads_ready++; @@ -653,7 +657,7 @@ static noinline void add_delayed_tree_ref(struct btrfs_fs_info *fs_info, * we've updated the existing ref, free the newly * allocated ref */ - kfree(full_ref); + kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref); } else { delayed_refs->num_entries++; trans->delayed_ref_updates++; @@ -714,7 +718,7 @@ static noinline void add_delayed_data_ref(struct btrfs_fs_info *fs_info, * we've updated the existing ref, free the newly * allocated ref */ - kfree(full_ref); + kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref); } else { delayed_refs->num_entries++; trans->delayed_ref_updates++; @@ -738,13 +742,13 @@ int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, struct btrfs_delayed_ref_root *delayed_refs; BUG_ON(extent_op && extent_op->is_data); - ref = kmalloc(sizeof(*ref), GFP_NOFS); + ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); if (!ref) return -ENOMEM; - head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); + head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); if (!head_ref) { - kfree(ref); + kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); return -ENOMEM; } @@ -786,13 +790,13 @@ int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, struct btrfs_delayed_ref_root *delayed_refs; BUG_ON(extent_op && !extent_op->is_data); - ref = kmalloc(sizeof(*ref), GFP_NOFS); + ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); if (!ref) return -ENOMEM; - head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); + head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); if (!head_ref) { - kfree(ref); + kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); return -ENOMEM; } @@ -826,7 +830,7 @@ int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, struct btrfs_delayed_ref_head *head_ref; struct btrfs_delayed_ref_root *delayed_refs; - head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); + head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); if (!head_ref) return -ENOMEM; @@ -860,3 +864,51 @@ btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr) return btrfs_delayed_node_to_head(ref); return NULL; } + +void btrfs_delayed_ref_exit(void) +{ + if (btrfs_delayed_ref_head_cachep) + kmem_cache_destroy(btrfs_delayed_ref_head_cachep); + if (btrfs_delayed_tree_ref_cachep) + kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); + if (btrfs_delayed_data_ref_cachep) + kmem_cache_destroy(btrfs_delayed_data_ref_cachep); + if (btrfs_delayed_extent_op_cachep) + kmem_cache_destroy(btrfs_delayed_extent_op_cachep); +} + +int btrfs_delayed_ref_init(void) +{ + btrfs_delayed_ref_head_cachep = kmem_cache_create( + "btrfs_delayed_ref_head", + sizeof(struct btrfs_delayed_ref_head), 0, + SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); + if (!btrfs_delayed_ref_head_cachep) + goto fail; + + btrfs_delayed_tree_ref_cachep = kmem_cache_create( + "btrfs_delayed_tree_ref", + sizeof(struct btrfs_delayed_tree_ref), 0, + SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); + if (!btrfs_delayed_tree_ref_cachep) + goto fail; + + btrfs_delayed_data_ref_cachep = kmem_cache_create( + "btrfs_delayed_data_ref", + sizeof(struct btrfs_delayed_data_ref), 0, + SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); + if (!btrfs_delayed_data_ref_cachep) + goto fail; + + btrfs_delayed_extent_op_cachep = kmem_cache_create( + "btrfs_delayed_extent_op", + sizeof(struct btrfs_delayed_extent_op), 0, + SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); + if (!btrfs_delayed_extent_op_cachep) + goto fail; + + return 0; +fail: + btrfs_delayed_ref_exit(); + return -ENOMEM; +} diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h index c9d703693df0..fe50392cdf76 100644 --- a/fs/btrfs/delayed-ref.h +++ b/fs/btrfs/delayed-ref.h @@ -141,12 +141,47 @@ struct btrfs_delayed_ref_root { u64 run_delayed_start; }; +extern struct kmem_cache *btrfs_delayed_ref_head_cachep; +extern struct kmem_cache *btrfs_delayed_tree_ref_cachep; +extern struct kmem_cache *btrfs_delayed_data_ref_cachep; +extern struct kmem_cache *btrfs_delayed_extent_op_cachep; + +int btrfs_delayed_ref_init(void); +void btrfs_delayed_ref_exit(void); + +static inline struct btrfs_delayed_extent_op * +btrfs_alloc_delayed_extent_op(void) +{ + return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS); +} + +static inline void +btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op) +{ + if (op) + kmem_cache_free(btrfs_delayed_extent_op_cachep, op); +} + static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref) { WARN_ON(atomic_read(&ref->refs) == 0); if (atomic_dec_and_test(&ref->refs)) { WARN_ON(ref->in_tree); - kfree(ref); + switch (ref->type) { + case BTRFS_TREE_BLOCK_REF_KEY: + case BTRFS_SHARED_BLOCK_REF_KEY: + kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); + break; + case BTRFS_EXTENT_DATA_REF_KEY: + case BTRFS_SHARED_DATA_REF_KEY: + kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); + break; + case 0: + kmem_cache_free(btrfs_delayed_ref_head_cachep, ref); + break; + default: + BUG(); + } } } diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 65f03670a952..4438aac4947f 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -3614,7 +3614,7 @@ int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, continue; } - kfree(head->extent_op); + btrfs_free_delayed_extent_op(head->extent_op); delayed_refs->num_heads--; if (list_empty(&head->cluster)) delayed_refs->num_heads_ready--; diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 61da9d0bb805..9bd87f0d73d0 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -2285,7 +2285,7 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, ref = &locked_ref->node; if (extent_op && must_insert_reserved) { - kfree(extent_op); + btrfs_free_delayed_extent_op(extent_op); extent_op = NULL; } @@ -2294,7 +2294,7 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, ret = run_delayed_extent_op(trans, root, ref, extent_op); - kfree(extent_op); + btrfs_free_delayed_extent_op(extent_op); if (ret) { list_del_init(&locked_ref->cluster); @@ -2338,7 +2338,7 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, must_insert_reserved); btrfs_put_delayed_ref(ref); - kfree(extent_op); + btrfs_free_delayed_extent_op(extent_op); count++; if (ret) { @@ -2586,7 +2586,7 @@ int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, struct btrfs_delayed_extent_op *extent_op; int ret; - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); + extent_op = btrfs_alloc_delayed_extent_op(); if (!extent_op) return -ENOMEM; @@ -2598,7 +2598,7 @@ int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, num_bytes, extent_op); if (ret) - kfree(extent_op); + btrfs_free_delayed_extent_op(extent_op); return ret; } @@ -5330,7 +5330,7 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, if (head->extent_op) { if (!head->must_insert_reserved) goto out; - kfree(head->extent_op); + btrfs_free_delayed_extent_op(head->extent_op); head->extent_op = NULL; } @@ -6400,7 +6400,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { struct btrfs_delayed_extent_op *extent_op; - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); + extent_op = btrfs_alloc_delayed_extent_op(); BUG_ON(!extent_op); /* -ENOMEM */ if (key) memcpy(&extent_op->key, key, sizeof(extent_op->key)); diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c index d8982e9601d3..67b373bf3ff9 100644 --- a/fs/btrfs/super.c +++ b/fs/btrfs/super.c @@ -1684,10 +1684,14 @@ static int __init init_btrfs_fs(void) if (err) goto free_delayed_inode; - err = btrfs_interface_init(); + err = btrfs_delayed_ref_init(); if (err) goto free_auto_defrag; + err = btrfs_interface_init(); + if (err) + goto free_delayed_ref; + err = register_filesystem(&btrfs_fs_type); if (err) goto unregister_ioctl; @@ -1699,6 +1703,8 @@ static int __init init_btrfs_fs(void) unregister_ioctl: btrfs_interface_exit(); +free_delayed_ref: + btrfs_delayed_ref_exit(); free_auto_defrag: btrfs_auto_defrag_exit(); free_delayed_inode: @@ -1720,6 +1726,7 @@ free_compress: static void __exit exit_btrfs_fs(void) { btrfs_destroy_cachep(); + btrfs_delayed_ref_exit(); btrfs_auto_defrag_exit(); btrfs_delayed_inode_exit(); ordered_data_exit(); -- cgit v1.2.3 From 093486c453a55230ccdad4b48863b872fe68c46e Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Wed, 19 Dec 2012 08:10:10 +0000 Subject: Btrfs: make delayed ref lock logic more readable Locking and unlocking delayed ref mutex are in the different functions, and the name of lock functions is not uniform, so the readability is not so good, this patch optimizes the lock logic and makes it more readable. Signed-off-by: Miao Xie Signed-off-by: Josef Bacik --- fs/btrfs/delayed-ref.c | 8 ++++++++ fs/btrfs/delayed-ref.h | 6 ++++++ fs/btrfs/extent-tree.c | 42 ++++++++++++++++++++++++------------------ 3 files changed, 38 insertions(+), 18 deletions(-) (limited to 'fs/btrfs/delayed-ref.h') diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c index 455894f1ca3b..b7a0641ead77 100644 --- a/fs/btrfs/delayed-ref.c +++ b/fs/btrfs/delayed-ref.c @@ -426,6 +426,14 @@ again: return 1; } +void btrfs_release_ref_cluster(struct list_head *cluster) +{ + struct list_head *pos, *q; + + list_for_each_safe(pos, q, cluster) + list_del_init(pos); +} + /* * helper function to update an extent delayed ref in the * rbtree. existing and update must both have the same diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h index fe50392cdf76..7939149f8f27 100644 --- a/fs/btrfs/delayed-ref.h +++ b/fs/btrfs/delayed-ref.h @@ -211,8 +211,14 @@ struct btrfs_delayed_ref_head * btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr); int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, struct btrfs_delayed_ref_head *head); +static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head) +{ + mutex_unlock(&head->mutex); +} + int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans, struct list_head *cluster, u64 search_start); +void btrfs_release_ref_cluster(struct list_head *cluster); int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, struct btrfs_delayed_ref_root *delayed_refs, diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 9bd87f0d73d0..b4cb8186035f 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -2143,7 +2143,6 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans, node->num_bytes); } } - mutex_unlock(&head->mutex); return ret; } @@ -2258,7 +2257,7 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, * process of being added. Don't run this ref yet. */ list_del_init(&locked_ref->cluster); - mutex_unlock(&locked_ref->mutex); + btrfs_delayed_ref_unlock(locked_ref); locked_ref = NULL; delayed_refs->num_heads_ready++; spin_unlock(&delayed_refs->lock); @@ -2297,25 +2296,22 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, btrfs_free_delayed_extent_op(extent_op); if (ret) { - list_del_init(&locked_ref->cluster); - mutex_unlock(&locked_ref->mutex); - - printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret); + printk(KERN_DEBUG + "btrfs: run_delayed_extent_op " + "returned %d\n", ret); spin_lock(&delayed_refs->lock); + btrfs_delayed_ref_unlock(locked_ref); return ret; } goto next; } - - list_del_init(&locked_ref->cluster); - locked_ref = NULL; } ref->in_tree = 0; rb_erase(&ref->rb_node, &delayed_refs->root); delayed_refs->num_entries--; - if (locked_ref) { + if (!btrfs_delayed_ref_is_head(ref)) { /* * when we play the delayed ref, also correct the * ref_mod on head @@ -2337,20 +2333,29 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, ret = run_one_delayed_ref(trans, root, ref, extent_op, must_insert_reserved); - btrfs_put_delayed_ref(ref); btrfs_free_delayed_extent_op(extent_op); - count++; - if (ret) { - if (locked_ref) { - list_del_init(&locked_ref->cluster); - mutex_unlock(&locked_ref->mutex); - } - printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret); + btrfs_delayed_ref_unlock(locked_ref); + btrfs_put_delayed_ref(ref); + printk(KERN_DEBUG + "btrfs: run_one_delayed_ref returned %d\n", ret); spin_lock(&delayed_refs->lock); return ret; } + /* + * If this node is a head, that means all the refs in this head + * have been dealt with, and we will pick the next head to deal + * with, so we must unlock the head and drop it from the cluster + * list before we release it. + */ + if (btrfs_delayed_ref_is_head(ref)) { + list_del_init(&locked_ref->cluster); + btrfs_delayed_ref_unlock(locked_ref); + locked_ref = NULL; + } + btrfs_put_delayed_ref(ref); + count++; next: cond_resched(); spin_lock(&delayed_refs->lock); @@ -2500,6 +2505,7 @@ again: ret = run_clustered_refs(trans, root, &cluster); if (ret < 0) { + btrfs_release_ref_cluster(&cluster); spin_unlock(&delayed_refs->lock); btrfs_abort_transaction(trans, root, ret); return ret; -- cgit v1.2.3