From 9042846bc7ae69cc3288d85af6bad16208d93a95 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Tue, 4 Aug 2009 16:56:34 -0400 Subject: [PATCH 01/16] Btrfs: Allow worker threads to exit when idle The Btrfs worker threads don't currently die off after they have been idle for a while, leading to a lot of threads sitting around doing nothing for each mount. Also, they are unable to start atomically (from end_io hanlders). This commit reworks the worker threads so they can be started from end_io handlers (just setting a flag that asks for a thread to be added at a later date) and so they can exit if they have been idle for a long time. Signed-off-by: Chris Mason --- fs/btrfs/async-thread.c | 133 +++++++++++++++++++++++++++++++++------- fs/btrfs/async-thread.h | 9 +++ fs/btrfs/disk-io.c | 22 ++++--- 3 files changed, 132 insertions(+), 32 deletions(-) diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c index 019e8af449a..f10c895224a 100644 --- a/fs/btrfs/async-thread.c +++ b/fs/btrfs/async-thread.c @@ -48,6 +48,9 @@ struct btrfs_worker_thread { /* number of things on the pending list */ atomic_t num_pending; + /* reference counter for this struct */ + atomic_t refs; + unsigned long sequence; /* protects the pending list. */ @@ -93,6 +96,31 @@ static void check_busy_worker(struct btrfs_worker_thread *worker) } } +static void check_pending_worker_creates(struct btrfs_worker_thread *worker) +{ + struct btrfs_workers *workers = worker->workers; + unsigned long flags; + + rmb(); + if (!workers->atomic_start_pending) + return; + + spin_lock_irqsave(&workers->lock, flags); + if (!workers->atomic_start_pending) + goto out; + + workers->atomic_start_pending = 0; + if (workers->num_workers >= workers->max_workers) + goto out; + + spin_unlock_irqrestore(&workers->lock, flags); + btrfs_start_workers(workers, 1); + return; + +out: + spin_unlock_irqrestore(&workers->lock, flags); +} + static noinline int run_ordered_completions(struct btrfs_workers *workers, struct btrfs_work *work) { @@ -140,6 +168,36 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers, return 0; } +static void put_worker(struct btrfs_worker_thread *worker) +{ + if (atomic_dec_and_test(&worker->refs)) + kfree(worker); +} + +static int try_worker_shutdown(struct btrfs_worker_thread *worker) +{ + int freeit = 0; + + spin_lock_irq(&worker->lock); + spin_lock_irq(&worker->workers->lock); + if (worker->workers->num_workers > 1 && + worker->idle && + !worker->working && + !list_empty(&worker->worker_list) && + list_empty(&worker->prio_pending) && + list_empty(&worker->pending)) { + freeit = 1; + list_del_init(&worker->worker_list); + worker->workers->num_workers--; + } + spin_unlock_irq(&worker->workers->lock); + spin_unlock_irq(&worker->lock); + + if (freeit) + put_worker(worker); + return freeit; +} + /* * main loop for servicing work items */ @@ -175,6 +233,8 @@ again_locked: */ run_ordered_completions(worker->workers, work); + check_pending_worker_creates(worker); + spin_lock_irq(&worker->lock); check_idle_worker(worker); } @@ -226,8 +286,13 @@ again_locked: worker->working = 0; spin_unlock_irq(&worker->lock); - if (!kthread_should_stop()) - schedule(); + if (!kthread_should_stop()) { + schedule_timeout(HZ * 120); + if (!worker->working && + try_worker_shutdown(worker)) { + return 0; + } + } } __set_current_state(TASK_RUNNING); } @@ -242,16 +307,30 @@ int btrfs_stop_workers(struct btrfs_workers *workers) { struct list_head *cur; struct btrfs_worker_thread *worker; + int can_stop; + spin_lock_irq(&workers->lock); list_splice_init(&workers->idle_list, &workers->worker_list); while (!list_empty(&workers->worker_list)) { cur = workers->worker_list.next; worker = list_entry(cur, struct btrfs_worker_thread, worker_list); - kthread_stop(worker->task); - list_del(&worker->worker_list); - kfree(worker); + + atomic_inc(&worker->refs); + workers->num_workers -= 1; + if (!list_empty(&worker->worker_list)) { + list_del_init(&worker->worker_list); + put_worker(worker); + can_stop = 1; + } else + can_stop = 0; + spin_unlock_irq(&workers->lock); + if (can_stop) + kthread_stop(worker->task); + spin_lock_irq(&workers->lock); + put_worker(worker); } + spin_unlock_irq(&workers->lock); return 0; } @@ -270,6 +349,8 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max) workers->idle_thresh = 32; workers->name = name; workers->ordered = 0; + workers->atomic_start_pending = 0; + workers->atomic_worker_start = 0; } /* @@ -294,6 +375,7 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) INIT_LIST_HEAD(&worker->worker_list); spin_lock_init(&worker->lock); atomic_set(&worker->num_pending, 0); + atomic_set(&worker->refs, 1); worker->workers = workers; worker->task = kthread_run(worker_loop, worker, "btrfs-%s-%d", workers->name, @@ -303,7 +385,6 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) kfree(worker); goto fail; } - spin_lock_irq(&workers->lock); list_add_tail(&worker->worker_list, &workers->idle_list); worker->idle = 1; @@ -367,6 +448,7 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers) { struct btrfs_worker_thread *worker; unsigned long flags; + struct list_head *fallback; again: spin_lock_irqsave(&workers->lock, flags); @@ -376,19 +458,10 @@ again: if (!worker) { spin_lock_irqsave(&workers->lock, flags); if (workers->num_workers >= workers->max_workers) { - struct list_head *fallback = NULL; - /* - * we have failed to find any workers, just - * return the force one - */ - if (!list_empty(&workers->worker_list)) - fallback = workers->worker_list.next; - if (!list_empty(&workers->idle_list)) - fallback = workers->idle_list.next; - BUG_ON(!fallback); - worker = list_entry(fallback, - struct btrfs_worker_thread, worker_list); - spin_unlock_irqrestore(&workers->lock, flags); + goto fallback; + } else if (workers->atomic_worker_start) { + workers->atomic_start_pending = 1; + goto fallback; } else { spin_unlock_irqrestore(&workers->lock, flags); /* we're below the limit, start another worker */ @@ -397,6 +470,22 @@ again: } } return worker; + +fallback: + fallback = NULL; + /* + * we have failed to find any workers, just + * return the first one we can find. + */ + if (!list_empty(&workers->worker_list)) + fallback = workers->worker_list.next; + if (!list_empty(&workers->idle_list)) + fallback = workers->idle_list.next; + BUG_ON(!fallback); + worker = list_entry(fallback, + struct btrfs_worker_thread, worker_list); + spin_unlock_irqrestore(&workers->lock, flags); + return worker; } /* @@ -435,9 +524,9 @@ int btrfs_requeue_work(struct btrfs_work *work) worker->working = 1; } - spin_unlock_irqrestore(&worker->lock, flags); if (wake) wake_up_process(worker->task); + spin_unlock_irqrestore(&worker->lock, flags); out: return 0; @@ -492,10 +581,10 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work) wake = 1; worker->working = 1; - spin_unlock_irqrestore(&worker->lock, flags); - if (wake) wake_up_process(worker->task); + spin_unlock_irqrestore(&worker->lock, flags); + out: return 0; } diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h index 1b511c109db..a562ad8d83a 100644 --- a/fs/btrfs/async-thread.h +++ b/fs/btrfs/async-thread.h @@ -73,6 +73,15 @@ struct btrfs_workers { /* force completions in the order they were queued */ int ordered; + /* more workers required, but in an interrupt handler */ + int atomic_start_pending; + + /* + * are we allowed to sleep while starting workers or are we required + * to start them at a later time? + */ + int atomic_worker_start; + /* list with all the work threads. The workers on the idle thread * may be actively servicing jobs, but they haven't yet hit the * idle thresh limit above. diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 3cf4cfa575c..20cefc6f22c 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -1682,7 +1682,7 @@ struct btrfs_root *open_ctree(struct super_block *sb, err = -EINVAL; goto fail_iput; } - +printk("thread pool is %d\n", fs_info->thread_pool_size); /* * we need to start all the end_io workers up front because the * queue work function gets called at interrupt time, and so it @@ -1727,20 +1727,22 @@ struct btrfs_root *open_ctree(struct super_block *sb, fs_info->endio_workers.idle_thresh = 4; fs_info->endio_meta_workers.idle_thresh = 4; - fs_info->endio_write_workers.idle_thresh = 64; - fs_info->endio_meta_write_workers.idle_thresh = 64; + fs_info->endio_write_workers.idle_thresh = 2; + fs_info->endio_meta_write_workers.idle_thresh = 2; + + fs_info->endio_workers.atomic_worker_start = 1; + fs_info->endio_meta_workers.atomic_worker_start = 1; + fs_info->endio_write_workers.atomic_worker_start = 1; + fs_info->endio_meta_write_workers.atomic_worker_start = 1; btrfs_start_workers(&fs_info->workers, 1); btrfs_start_workers(&fs_info->submit_workers, 1); btrfs_start_workers(&fs_info->delalloc_workers, 1); btrfs_start_workers(&fs_info->fixup_workers, 1); - btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size); - btrfs_start_workers(&fs_info->endio_meta_workers, - fs_info->thread_pool_size); - btrfs_start_workers(&fs_info->endio_meta_write_workers, - fs_info->thread_pool_size); - btrfs_start_workers(&fs_info->endio_write_workers, - fs_info->thread_pool_size); + btrfs_start_workers(&fs_info->endio_workers, 1); + btrfs_start_workers(&fs_info->endio_meta_workers, 1); + btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); + btrfs_start_workers(&fs_info->endio_write_workers, 1); fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, From 40431d6c1288793a682fc6f5e5b5c9d5cac34608 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 5 Aug 2009 12:57:59 -0400 Subject: [PATCH 02/16] Btrfs: optimize set extent bit The Btrfs set_extent_bit call currently searches the rbtree every time it needs to find more extent_state objects to fill the requested operation. This adds a simple test with rb_next to see if the next object in the tree was adjacent to the one we just found. If so, we skip the search and just use the next object. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 15 +++++++++++++-- fs/btrfs/file.c | 2 -- 2 files changed, 13 insertions(+), 4 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 68260180f58..7e5c5a0749e 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -694,8 +694,8 @@ again: BUG_ON(err == -EEXIST); goto out; } - state = rb_entry(node, struct extent_state, rb_node); +hit_next: last_start = state->start; last_end = state->end; @@ -706,6 +706,7 @@ again: * Just lock what we found and keep going */ if (state->start == start && state->end <= end) { + struct rb_node *next_node; set = state->state & bits; if (set && exclusive) { *failed_start = state->start; @@ -716,7 +717,17 @@ again: merge_state(tree, state); if (last_end == (u64)-1) goto out; + start = last_end + 1; + if (start < end && prealloc && !need_resched()) { + next_node = rb_next(node); + if (next_node) { + state = rb_entry(next_node, struct extent_state, + rb_node); + if (state->start == start) + goto hit_next; + } + } goto search_again; } @@ -852,7 +863,7 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { return set_extent_bit(tree, start, end, - EXTENT_DELALLOC | EXTENT_DIRTY, + EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, 0, NULL, mask); } diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index 7c3cd248d8d..a760d97279a 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -136,8 +136,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, btrfs_set_trans_block_group(trans, inode); hint_byte = 0; - set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS); - /* check for reserved extents on each page, we don't want * to reset the delalloc bit on things that already have * extents reserved. From 4e3f9c5042b43301d70781aee4a164a20878066b Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 5 Aug 2009 16:36:45 -0400 Subject: [PATCH 03/16] Btrfs: keep irqs on more often in the worker threads The btrfs worker thread spinlock was being used both for the queueing of IO and for the processing of ordered events. The ordered events never happen from end_io handlers, and so they don't need to use the _irq version of spinlocks. This adds a dedicated lock to the ordered lists so they don't have to run with irqs off. Signed-off-by: Chris Mason --- fs/btrfs/async-thread.c | 23 +++++++++++++---------- fs/btrfs/async-thread.h | 3 +++ 2 files changed, 16 insertions(+), 10 deletions(-) diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c index f10c895224a..4b4372df3b6 100644 --- a/fs/btrfs/async-thread.c +++ b/fs/btrfs/async-thread.c @@ -124,14 +124,12 @@ out: static noinline int run_ordered_completions(struct btrfs_workers *workers, struct btrfs_work *work) { - unsigned long flags; - if (!workers->ordered) return 0; set_bit(WORK_DONE_BIT, &work->flags); - spin_lock_irqsave(&workers->lock, flags); + spin_lock(&workers->order_lock); while (1) { if (!list_empty(&workers->prio_order_list)) { @@ -154,17 +152,17 @@ static noinline int run_ordered_completions(struct btrfs_workers *workers, if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags)) break; - spin_unlock_irqrestore(&workers->lock, flags); + spin_unlock(&workers->order_lock); work->ordered_func(work); /* now take the lock again and call the freeing code */ - spin_lock_irqsave(&workers->lock, flags); + spin_lock(&workers->order_lock); list_del(&work->order_list); work->ordered_free(work); } - spin_unlock_irqrestore(&workers->lock, flags); + spin_unlock(&workers->order_lock); return 0; } @@ -345,6 +343,7 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max) INIT_LIST_HEAD(&workers->order_list); INIT_LIST_HEAD(&workers->prio_order_list); spin_lock_init(&workers->lock); + spin_lock_init(&workers->order_lock); workers->max_workers = max; workers->idle_thresh = 32; workers->name = name; @@ -374,6 +373,7 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) INIT_LIST_HEAD(&worker->prio_pending); INIT_LIST_HEAD(&worker->worker_list); spin_lock_init(&worker->lock); + atomic_set(&worker->num_pending, 0); atomic_set(&worker->refs, 1); worker->workers = workers; @@ -453,10 +453,8 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers) again: spin_lock_irqsave(&workers->lock, flags); worker = next_worker(workers); - spin_unlock_irqrestore(&workers->lock, flags); if (!worker) { - spin_lock_irqsave(&workers->lock, flags); if (workers->num_workers >= workers->max_workers) { goto fallback; } else if (workers->atomic_worker_start) { @@ -469,6 +467,7 @@ again: goto again; } } + spin_unlock_irqrestore(&workers->lock, flags); return worker; fallback: @@ -552,14 +551,18 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work) worker = find_worker(workers); if (workers->ordered) { - spin_lock_irqsave(&workers->lock, flags); + /* + * you're not allowed to do ordered queues from an + * interrupt handler + */ + spin_lock(&workers->order_lock); if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) { list_add_tail(&work->order_list, &workers->prio_order_list); } else { list_add_tail(&work->order_list, &workers->order_list); } - spin_unlock_irqrestore(&workers->lock, flags); + spin_unlock(&workers->order_lock); } else { INIT_LIST_HEAD(&work->order_list); } diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h index a562ad8d83a..fc089b95ec1 100644 --- a/fs/btrfs/async-thread.h +++ b/fs/btrfs/async-thread.h @@ -99,6 +99,9 @@ struct btrfs_workers { /* lock for finding the next worker thread to queue on */ spinlock_t lock; + /* lock for the ordered lists */ + spinlock_t order_lock; + /* extra name for this worker, used for current->name */ char *name; }; From 4f878e8475a465ddbd951e06a23317303f1b5b30 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 7 Aug 2009 09:27:38 -0400 Subject: [PATCH 04/16] Btrfs: reduce worker thread spin_lock_irq hold times This changes the btrfs worker threads to batch work items into a local list. It allows us to pull work items in large chunks and significantly reduces the number of times we need to take the worker thread spinlock. Signed-off-by: Chris Mason --- fs/btrfs/async-thread.c | 74 +++++++++++++++++++++++++++++++++-------- 1 file changed, 60 insertions(+), 14 deletions(-) diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c index 4b4372df3b6..6ea5cd0a595 100644 --- a/fs/btrfs/async-thread.c +++ b/fs/btrfs/async-thread.c @@ -196,31 +196,73 @@ static int try_worker_shutdown(struct btrfs_worker_thread *worker) return freeit; } +static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker, + struct list_head *prio_head, + struct list_head *head) +{ + struct btrfs_work *work = NULL; + struct list_head *cur = NULL; + + if(!list_empty(prio_head)) + cur = prio_head->next; + + smp_mb(); + if (!list_empty(&worker->prio_pending)) + goto refill; + + if (!list_empty(head)) + cur = head->next; + + if (cur) + goto out; + +refill: + spin_lock_irq(&worker->lock); + list_splice_tail_init(&worker->prio_pending, prio_head); + list_splice_tail_init(&worker->pending, head); + + if (!list_empty(prio_head)) + cur = prio_head->next; + else if (!list_empty(head)) + cur = head->next; + spin_unlock_irq(&worker->lock); + + if (!cur) + goto out_fail; + +out: + work = list_entry(cur, struct btrfs_work, list); + +out_fail: + return work; +} + /* * main loop for servicing work items */ static int worker_loop(void *arg) { struct btrfs_worker_thread *worker = arg; - struct list_head *cur; + struct list_head head; + struct list_head prio_head; struct btrfs_work *work; + + INIT_LIST_HEAD(&head); + INIT_LIST_HEAD(&prio_head); + do { - spin_lock_irq(&worker->lock); -again_locked: +again: while (1) { - if (!list_empty(&worker->prio_pending)) - cur = worker->prio_pending.next; - else if (!list_empty(&worker->pending)) - cur = worker->pending.next; - else + + + work = get_next_work(worker, &prio_head, &head); + if (!work) break; - work = list_entry(cur, struct btrfs_work, list); list_del(&work->list); clear_bit(WORK_QUEUED_BIT, &work->flags); work->worker = worker; - spin_unlock_irq(&worker->lock); work->func(work); @@ -233,9 +275,11 @@ again_locked: check_pending_worker_creates(worker); - spin_lock_irq(&worker->lock); - check_idle_worker(worker); } + + spin_lock_irq(&worker->lock); + check_idle_worker(worker); + if (freezing(current)) { worker->working = 0; spin_unlock_irq(&worker->lock); @@ -274,8 +318,10 @@ again_locked: spin_lock_irq(&worker->lock); set_current_state(TASK_INTERRUPTIBLE); if (!list_empty(&worker->pending) || - !list_empty(&worker->prio_pending)) - goto again_locked; + !list_empty(&worker->prio_pending)) { + spin_unlock_irq(&worker->lock); + goto again; + } /* * this makes sure we get a wakeup when someone From a97adc9fffb1707da4e97f91c801660c6be92aac Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 7 Aug 2009 09:28:20 -0400 Subject: [PATCH 05/16] Btrfs: use larger nr_to_write for larger extents When btrfs fills a large delayed allocation extent, it is a good idea to try and convince the write_cache_pages caller to go ahead and write a good chunk of that extent. The extra IO is basically free because we know it is contiguous. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 7e5c5a0749e..8d7a152a90c 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2135,6 +2135,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, delalloc_end = 0; page_started = 0; if (!epd->extent_locked) { + u64 delalloc_to_write; /* * make sure the wbc mapping index is at least updated * to this page. @@ -2154,6 +2155,14 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, tree->ops->fill_delalloc(inode, page, delalloc_start, delalloc_end, &page_started, &nr_written); + delalloc_to_write = (delalloc_end - + max_t(u64, page_offset(page), + delalloc_start) + 1) >> + PAGE_CACHE_SHIFT; + if (wbc->nr_to_write < delalloc_to_write) { + wbc->nr_to_write = min_t(long, 8192, + delalloc_to_write); + } delalloc_start = delalloc_end + 1; } @@ -2350,7 +2359,6 @@ static int extent_write_cache_pages(struct extent_io_tree *tree, writepage_t writepage, void *data, void (*flush_fn)(void *)) { - struct backing_dev_info *bdi = mapping->backing_dev_info; int ret = 0; int done = 0; struct pagevec pvec; @@ -2425,10 +2433,6 @@ retry: } if (ret || wbc->nr_to_write <= 0) done = 1; - if (wbc->nonblocking && bdi_write_congested(bdi)) { - wbc->encountered_congestion = 1; - done = 1; - } } pagevec_release(&pvec); cond_resched(); From 57fd5a5ff8b48b99e90b22fc143082aba755c6c0 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 7 Aug 2009 09:59:15 -0400 Subject: [PATCH 06/16] Btrfs: tweak congestion backoff The btrfs io submission thread tries to back off congested devices in favor of rotating off to another disk. But, it tries to make sure it submits at least some IO before rotating on (the others may be congested too), and so it has a magic number of requests it tries to write before it hops. This makes the magic number smaller. Testing shows that we're spending too much time on congested devices and leaving the other devices idle. Signed-off-by: Chris Mason --- fs/btrfs/volumes.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 5dbefd11b4a..a7e53773e74 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -276,7 +276,7 @@ loop_lock: * is now congested. Back off and let other work structs * run instead */ - if (pending && bdi_write_congested(bdi) && batch_run > 32 && + if (pending && bdi_write_congested(bdi) && batch_run > 8 && fs_info->fs_devices->open_devices > 1) { struct io_context *ioc; From 890871be854b5f5e43e7ba2475f706209906cc24 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 16:24:52 -0400 Subject: [PATCH 07/16] Btrfs: switch extent_map to a rw lock There are two main users of the extent_map tree. The first is regular file inodes, where it is evenly spread between readers and writers. The second is the chunk allocation tree, which maps blocks from logical addresses to phyiscal ones, and it is 99.99% reads. The mapping tree is a point of lock contention during heavy IO workloads, so this commit switches things to a rw lock. Signed-off-by: Chris Mason --- fs/btrfs/compression.c | 8 ++++---- fs/btrfs/disk-io.c | 14 +++++++------- fs/btrfs/extent-tree.c | 4 ++-- fs/btrfs/extent_io.c | 8 ++++---- fs/btrfs/extent_map.c | 5 +---- fs/btrfs/extent_map.h | 2 +- fs/btrfs/file.c | 8 ++++---- fs/btrfs/inode.c | 24 ++++++++++++------------ fs/btrfs/relocation.c | 4 ++-- fs/btrfs/volumes.c | 40 ++++++++++++++++++++-------------------- 10 files changed, 57 insertions(+), 60 deletions(-) diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c index de1e2fd3208..78451a58f20 100644 --- a/fs/btrfs/compression.c +++ b/fs/btrfs/compression.c @@ -507,10 +507,10 @@ static noinline int add_ra_bio_pages(struct inode *inode, */ set_page_extent_mapped(page); lock_extent(tree, last_offset, end, GFP_NOFS); - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, last_offset, PAGE_CACHE_SIZE); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); if (!em || last_offset < em->start || (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || @@ -594,11 +594,11 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, em_tree = &BTRFS_I(inode)->extent_tree; /* we need the actual starting offset of this extent in the file */ - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, page_offset(bio->bi_io_vec->bv_page), PAGE_CACHE_SIZE); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); compressed_len = em->block_len; cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 20cefc6f22c..b6cfdd9164e 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -121,15 +121,15 @@ static struct extent_map *btree_get_extent(struct inode *inode, struct extent_map *em; int ret; - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, start, len); if (em) { em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); goto out; } - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); em = alloc_extent_map(GFP_NOFS); if (!em) { @@ -142,7 +142,7 @@ static struct extent_map *btree_get_extent(struct inode *inode, em->block_start = 0; em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); if (ret == -EEXIST) { u64 failed_start = em->start; @@ -161,7 +161,7 @@ static struct extent_map *btree_get_extent(struct inode *inode, free_extent_map(em); em = NULL; } - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret) em = ERR_PTR(ret); @@ -1323,9 +1323,9 @@ static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) offset = page_offset(page); em_tree = &BTRFS_I(inode)->extent_tree; - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); if (!em) { __unplug_io_fn(bdi, page); return; diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 72a2b9c28e9..edd86ae9e14 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -5396,9 +5396,9 @@ static noinline int relocate_data_extent(struct inode *reloc_inode, lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); while (1) { int ret; - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret != -EEXIST) { free_extent_map(em); break; diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 8d7a152a90c..41cf1b451b4 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2786,15 +2786,15 @@ int try_release_extent_mapping(struct extent_map_tree *map, u64 len; while (start <= end) { len = end - start + 1; - spin_lock(&map->lock); + write_lock(&map->lock); em = lookup_extent_mapping(map, start, len); if (!em || IS_ERR(em)) { - spin_unlock(&map->lock); + write_unlock(&map->lock); break; } if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || em->start != start) { - spin_unlock(&map->lock); + write_unlock(&map->lock); free_extent_map(em); break; } @@ -2808,7 +2808,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, free_extent_map(em); } start = extent_map_end(em); - spin_unlock(&map->lock); + write_unlock(&map->lock); /* once for us */ free_extent_map(em); diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c index 30c9365861e..72e9fa3c31f 100644 --- a/fs/btrfs/extent_map.c +++ b/fs/btrfs/extent_map.c @@ -36,7 +36,7 @@ void extent_map_exit(void) void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask) { tree->map.rb_node = NULL; - spin_lock_init(&tree->lock); + rwlock_init(&tree->lock); } /** @@ -222,7 +222,6 @@ int add_extent_mapping(struct extent_map_tree *tree, ret = -EEXIST; goto out; } - assert_spin_locked(&tree->lock); rb = tree_insert(&tree->map, em->start, &em->rb_node); if (rb) { ret = -EEXIST; @@ -285,7 +284,6 @@ struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree, struct rb_node *next = NULL; u64 end = range_end(start, len); - assert_spin_locked(&tree->lock); rb_node = __tree_search(&tree->map, start, &prev, &next); if (!rb_node && prev) { em = rb_entry(prev, struct extent_map, rb_node); @@ -331,7 +329,6 @@ int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em) int ret = 0; WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags)); - assert_spin_locked(&tree->lock); rb_erase(&em->rb_node, &tree->map); em->in_tree = 0; return ret; diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h index fb6eeef06bb..6216dfbcf9b 100644 --- a/fs/btrfs/extent_map.h +++ b/fs/btrfs/extent_map.h @@ -31,7 +31,7 @@ struct extent_map { struct extent_map_tree { struct rb_root map; - spinlock_t lock; + rwlock_t lock; }; static inline u64 extent_map_end(struct extent_map *em) diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index a760d97279a..8a9c76aecdf 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -188,15 +188,15 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, if (!split2) split2 = alloc_extent_map(GFP_NOFS); - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, start, len); if (!em) { - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); break; } flags = em->flags; if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (em->start <= start && (!testend || em->start + em->len >= start + len)) { free_extent_map(em); @@ -259,7 +259,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, free_extent_map(split); split = NULL; } - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); /* once for us */ free_extent_map(em); diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 04b53b5ebe5..f1df1171861 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -612,9 +612,9 @@ static noinline int submit_compressed_extents(struct inode *inode, set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); while (1) { - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret != -EEXIST) { free_extent_map(em); break; @@ -748,9 +748,9 @@ static noinline int cow_file_range(struct inode *inode, set_bit(EXTENT_FLAG_PINNED, &em->flags); while (1) { - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret != -EEXIST) { free_extent_map(em); break; @@ -1081,9 +1081,9 @@ out_check: em->bdev = root->fs_info->fs_devices->latest_bdev; set_bit(EXTENT_FLAG_PINNED, &em->flags); while (1) { - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret != -EEXIST) { free_extent_map(em); break; @@ -1670,13 +1670,13 @@ static int btrfs_io_failed_hook(struct bio *failed_bio, failrec->last_mirror = 0; failrec->bio_flags = 0; - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, start, failrec->len); if (em->start > start || em->start + em->len < start) { free_extent_map(em); em = NULL; } - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); if (!em || IS_ERR(em)) { kfree(failrec); @@ -4069,11 +4069,11 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, int compressed; again: - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, start, len); if (em) em->bdev = root->fs_info->fs_devices->latest_bdev; - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); if (em) { if (em->start > start || em->start + em->len <= start) @@ -4264,7 +4264,7 @@ insert: } err = 0; - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); /* it is possible that someone inserted the extent into the tree * while we had the lock dropped. It is also possible that @@ -4304,7 +4304,7 @@ insert: err = 0; } } - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); out: if (path) btrfs_free_path(path); diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c index c04f7f21260..4adab903fc2 100644 --- a/fs/btrfs/relocation.c +++ b/fs/btrfs/relocation.c @@ -2646,9 +2646,9 @@ int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key) lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS); while (1) { int ret; - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); if (ret != -EEXIST) { free_extent_map(em); break; diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index a7e53773e74..d2358c06bbd 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -1749,9 +1749,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root, * step two, delete the device extents and the * chunk tree entries */ - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, chunk_offset, 1); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); BUG_ON(em->start > chunk_offset || em->start + em->len < chunk_offset); @@ -1780,9 +1780,9 @@ static int btrfs_relocate_chunk(struct btrfs_root *root, ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); BUG_ON(ret); - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); remove_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); kfree(map); em->bdev = NULL; @@ -2294,9 +2294,9 @@ again: em->block_len = em->len; em_tree = &extent_root->fs_info->mapping_tree.map_tree; - spin_lock(&em_tree->lock); + write_lock(&em_tree->lock); ret = add_extent_mapping(em_tree, em); - spin_unlock(&em_tree->lock); + write_unlock(&em_tree->lock); BUG_ON(ret); free_extent_map(em); @@ -2491,9 +2491,9 @@ int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) int readonly = 0; int i; - spin_lock(&map_tree->map_tree.lock); + read_lock(&map_tree->map_tree.lock); em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); - spin_unlock(&map_tree->map_tree.lock); + read_unlock(&map_tree->map_tree.lock); if (!em) return 1; @@ -2518,11 +2518,11 @@ void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) struct extent_map *em; while (1) { - spin_lock(&tree->map_tree.lock); + write_lock(&tree->map_tree.lock); em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); if (em) remove_extent_mapping(&tree->map_tree, em); - spin_unlock(&tree->map_tree.lock); + write_unlock(&tree->map_tree.lock); if (!em) break; kfree(em->bdev); @@ -2540,9 +2540,9 @@ int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) struct extent_map_tree *em_tree = &map_tree->map_tree; int ret; - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, logical, len); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); BUG_ON(!em); BUG_ON(em->start > logical || em->start + em->len < logical); @@ -2604,9 +2604,9 @@ again: atomic_set(&multi->error, 0); } - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, logical, *length); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); if (!em && unplug_page) return 0; @@ -2763,9 +2763,9 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, u64 stripe_nr; int i, j, nr = 0; - spin_lock(&em_tree->lock); + read_lock(&em_tree->lock); em = lookup_extent_mapping(em_tree, chunk_start, 1); - spin_unlock(&em_tree->lock); + read_unlock(&em_tree->lock); BUG_ON(!em || em->start != chunk_start); map = (struct map_lookup *)em->bdev; @@ -3053,9 +3053,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, logical = key->offset; length = btrfs_chunk_length(leaf, chunk); - spin_lock(&map_tree->map_tree.lock); + read_lock(&map_tree->map_tree.lock); em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); - spin_unlock(&map_tree->map_tree.lock); + read_unlock(&map_tree->map_tree.lock); /* already mapped? */ if (em && em->start <= logical && em->start + em->len > logical) { @@ -3114,9 +3114,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, map->stripes[i].dev->in_fs_metadata = 1; } - spin_lock(&map_tree->map_tree.lock); + write_lock(&map_tree->map_tree.lock); ret = add_extent_mapping(&map_tree->map_tree, em); - spin_unlock(&map_tree->map_tree.lock); + write_unlock(&map_tree->map_tree.lock); BUG_ON(ret); free_extent_map(em); From e48c465bb366c0169f7908bfe62ae7080874ee7d Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 11 Sep 2009 11:25:02 -0400 Subject: [PATCH 08/16] Btrfs: Fix new state initialization order As the extent state tree is manipulated, there are call backs that are used to take extra actions when different state bits are set or cleared. One example of this is a counter for the total number of delayed allocation bytes in a single inode and in the whole FS. When new states are inserted, this callback is being done before we properly setup the new state. This hasn't caused problems before because the lock bit was always done first, and the existing call backs don't care about the lock bit. This patch makes sure the state is properly setup before using the callback, which is important for later optimizations that do more work without using the lock bit. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 41cf1b451b4..8e168a457a3 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -367,10 +367,10 @@ static int insert_state(struct extent_io_tree *tree, } if (bits & EXTENT_DIRTY) tree->dirty_bytes += end - start + 1; - set_state_cb(tree, state, bits); - state->state |= bits; state->start = start; state->end = end; + set_state_cb(tree, state, bits); + state->state |= bits; node = tree_insert(&tree->state, end, &state->rb_node); if (node) { struct extent_state *found; From 1edbb734b4e010974c41d2859d22a43d04f5f1cf Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 13:24:36 -0400 Subject: [PATCH 09/16] Btrfs: reduce CPU usage in the extent_state tree Btrfs is currently mirroring some of the page state bits into its extent state tree. The goal behind this was to use it in supporting blocksizes other than the page size. But, we don't currently support that, and we're using quite a lot of CPU on the rb tree and its spin lock. This commit starts a series of cleanups to reduce the amount of work done in the extent state tree as part of each IO. This commit: * Adds the ability to lock an extent in the state tree and also set other bits. The idea is to do locking and delalloc in one call * Removes the EXTENT_WRITEBACK and EXTENT_DIRTY bits. Btrfs is using a combination of the page bits and the ordered write code for this instead. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 75 +++++++++++++++----------------------------- fs/btrfs/extent_io.h | 2 ++ fs/btrfs/file.c | 19 ----------- 3 files changed, 28 insertions(+), 68 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 8e168a457a3..7c70613eb72 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -654,25 +654,24 @@ static void set_state_bits(struct extent_io_tree *tree, } /* - * set some bits on a range in the tree. This may require allocations - * or sleeping, so the gfp mask is used to indicate what is allowed. + * set some bits on a range in the tree. This may require allocations or + * sleeping, so the gfp mask is used to indicate what is allowed. * - * If 'exclusive' == 1, this will fail with -EEXIST if some part of the - * range already has the desired bits set. The start of the existing - * range is returned in failed_start in this case. + * If any of the exclusive bits are set, this will fail with -EEXIST if some + * part of the range already has the desired bits set. The start of the + * existing range is returned in failed_start in this case. * - * [start, end] is inclusive - * This takes the tree lock. + * [start, end] is inclusive This takes the tree lock. */ + static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, - int bits, int exclusive, u64 *failed_start, + int bits, int exclusive_bits, u64 *failed_start, gfp_t mask) { struct extent_state *state; struct extent_state *prealloc = NULL; struct rb_node *node; int err = 0; - int set; u64 last_start; u64 last_end; again: @@ -707,8 +706,7 @@ hit_next: */ if (state->start == start && state->end <= end) { struct rb_node *next_node; - set = state->state & bits; - if (set && exclusive) { + if (state->state & exclusive_bits) { *failed_start = state->start; err = -EEXIST; goto out; @@ -748,8 +746,7 @@ hit_next: * desired bit on it. */ if (state->start < start) { - set = state->state & bits; - if (exclusive && set) { + if (state->state & exclusive_bits) { *failed_start = start; err = -EEXIST; goto out; @@ -799,8 +796,7 @@ hit_next: * on the first half */ if (state->start <= end && state->end > end) { - set = state->state & bits; - if (exclusive && set) { + if (state->state & exclusive_bits) { *failed_start = start; err = -EEXIST; goto out; @@ -906,19 +902,6 @@ static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); } -static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end, - gfp_t mask) -{ - return set_extent_bit(tree, start, end, EXTENT_WRITEBACK, - 0, NULL, mask); -} - -static int clear_extent_writeback(struct extent_io_tree *tree, u64 start, - u64 end, gfp_t mask) -{ - return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask); -} - int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) { return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); @@ -928,13 +911,14 @@ int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) * either insert or lock state struct between start and end use mask to tell * us if waiting is desired. */ -int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) +int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, + int bits, gfp_t mask) { int err; u64 failed_start; while (1) { - err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, - &failed_start, mask); + err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, + EXTENT_LOCKED, &failed_start, mask); if (err == -EEXIST && (mask & __GFP_WAIT)) { wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); start = failed_start; @@ -946,6 +930,11 @@ int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) return err; } +int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) +{ + return lock_extent_bits(tree, start, end, 0, mask); +} + int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { @@ -985,7 +974,6 @@ int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) page_cache_release(page); index++; } - set_extent_dirty(tree, start, end, GFP_NOFS); return 0; } @@ -1005,7 +993,6 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) page_cache_release(page); index++; } - set_extent_writeback(tree, start, end, GFP_NOFS); return 0; } @@ -1563,10 +1550,7 @@ static int check_page_locked(struct extent_io_tree *tree, static int check_page_writeback(struct extent_io_tree *tree, struct page *page) { - u64 start = (u64)page->index << PAGE_CACHE_SHIFT; - u64 end = start + PAGE_CACHE_SIZE - 1; - if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0)) - end_page_writeback(page); + end_page_writeback(page); return 0; } @@ -1624,13 +1608,11 @@ static void end_bio_extent_writepage(struct bio *bio, int err) } if (!uptodate) { - clear_extent_uptodate(tree, start, end, GFP_ATOMIC); + clear_extent_uptodate(tree, start, end, GFP_NOFS); ClearPageUptodate(page); SetPageError(page); } - clear_extent_writeback(tree, start, end, GFP_ATOMIC); - if (whole_page) end_page_writeback(page); else @@ -2208,8 +2190,9 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, printk(KERN_ERR "btrfs delalloc bits after lock_extent\n"); if (last_byte <= start) { - clear_extent_dirty(tree, start, page_end, GFP_NOFS); - unlock_extent(tree, start, page_end, GFP_NOFS); + clear_extent_bit(tree, start, page_end, + EXTENT_LOCKED | EXTENT_DIRTY, + 1, 0, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, start, page_end, NULL, 1); @@ -2217,12 +2200,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, goto done; } - set_extent_uptodate(tree, start, page_end, GFP_NOFS); blocksize = inode->i_sb->s_blocksize; while (cur <= end) { if (cur >= last_byte) { - clear_extent_dirty(tree, cur, page_end, GFP_NOFS); unlock_extent(tree, unlock_start, page_end, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, cur, @@ -2255,9 +2236,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, */ if (compressed || block_start == EXTENT_MAP_HOLE || block_start == EXTENT_MAP_INLINE) { - clear_extent_dirty(tree, cur, - cur + iosize - 1, GFP_NOFS); - unlock_extent(tree, unlock_start, cur + iosize - 1, GFP_NOFS); @@ -2291,7 +2269,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, continue; } - clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS); if (tree->ops && tree->ops->writepage_io_hook) { ret = tree->ops->writepage_io_hook(page, cur, cur + iosize - 1); @@ -2619,7 +2596,7 @@ int extent_invalidatepage(struct extent_io_tree *tree, return 0; lock_extent(tree, start, end, GFP_NOFS); - wait_on_extent_writeback(tree, start, end); + wait_on_page_writeback(page); clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 1, 1, GFP_NOFS); diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index 5bc20abf3f3..88d134d01fb 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -142,6 +142,8 @@ int try_release_extent_state(struct extent_map_tree *map, struct extent_io_tree *tree, struct page *page, gfp_t mask); int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); +int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, + int bits, gfp_t mask); int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index 8a9c76aecdf..ef66c3d989b 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -113,8 +113,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, int err = 0; int i; struct inode *inode = fdentry(file)->d_inode; - struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; - u64 hint_byte; u64 num_bytes; u64 start_pos; u64 end_of_last_block; @@ -126,20 +124,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, root->sectorsize - 1) & ~((u64)root->sectorsize - 1); end_of_last_block = start_pos + num_bytes - 1; - - lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS); - trans = btrfs_join_transaction(root, 1); - if (!trans) { - err = -ENOMEM; - goto out_unlock; - } - btrfs_set_trans_block_group(trans, inode); - hint_byte = 0; - - /* check for reserved extents on each page, we don't want - * to reset the delalloc bit on things that already have - * extents reserved. - */ btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block); for (i = 0; i < num_pages; i++) { struct page *p = pages[i]; @@ -154,9 +138,6 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, * at this time. */ } - err = btrfs_end_transaction(trans, root); -out_unlock: - unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS); return err; } From 2c64c53d8d30d43d0670482503a3914dfd3d6d46 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 15:04:12 -0400 Subject: [PATCH 10/16] Btrfs: cache values for locking extents Many of the btrfs extent state tree users follow the same pattern. They lock an extent range in the tree, do some operation and then unlock. This translates to at least 2 rbtree searches, and maybe more if they are doing operations on the extent state tree. A locked extent in the tree isn't going to be merged or changed, and so we can safely return the extent state structure as a cached handle. This changes set_extent_bit to give back a cached handle, and also changes both set_extent_bit and clear_extent_bit to use the cached handle if it is available. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 125 ++++++++++++++++++++++++++++++++----------- fs/btrfs/extent_io.h | 5 +- fs/btrfs/inode.c | 6 +-- 3 files changed, 100 insertions(+), 36 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 7c70613eb72..c7a5e860fe2 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -471,10 +471,14 @@ static int clear_state_bit(struct extent_io_tree *tree, * bits were already set, or zero if none of the bits were already set. */ int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, - int bits, int wake, int delete, gfp_t mask) + int bits, int wake, int delete, + struct extent_state **cached_state, + gfp_t mask) { struct extent_state *state; + struct extent_state *cached; struct extent_state *prealloc = NULL; + struct rb_node *next_node; struct rb_node *node; u64 last_end; int err; @@ -488,6 +492,17 @@ again: } spin_lock(&tree->lock); + if (cached_state) { + cached = *cached_state; + *cached_state = NULL; + if (cached->tree && cached->start == start) { + atomic_dec(&cached->refs); + state = cached; + last_end = state->end; + goto found; + } + free_extent_state(cached); + } /* * this search will find the extents that end after * our range starts @@ -496,6 +511,7 @@ again: if (!node) goto out; state = rb_entry(node, struct extent_state, rb_node); +hit_next: if (state->start > end) goto out; WARN_ON(state->end < start); @@ -555,11 +571,21 @@ again: prealloc = NULL; goto out; } - +found: + if (state->end < end && prealloc && !need_resched()) + next_node = rb_next(&state->rb_node); + else + next_node = NULL; set |= clear_state_bit(tree, state, bits, wake, delete); if (last_end == (u64)-1) goto out; start = last_end + 1; + if (start <= end && next_node) { + state = rb_entry(next_node, struct extent_state, + rb_node); + if (state->start == start) + goto hit_next; + } goto search_again; out: @@ -653,6 +679,17 @@ static void set_state_bits(struct extent_io_tree *tree, state->state |= bits; } +static void cache_state(struct extent_state *state, + struct extent_state **cached_ptr) +{ + if (cached_ptr && !(*cached_ptr)) { + if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { + *cached_ptr = state; + atomic_inc(&state->refs); + } + } +} + /* * set some bits on a range in the tree. This may require allocations or * sleeping, so the gfp mask is used to indicate what is allowed. @@ -666,6 +703,7 @@ static void set_state_bits(struct extent_io_tree *tree, static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, int exclusive_bits, u64 *failed_start, + struct extent_state **cached_state, gfp_t mask) { struct extent_state *state; @@ -712,6 +750,7 @@ hit_next: goto out; } set_state_bits(tree, state, bits); + cache_state(state, cached_state); merge_state(tree, state); if (last_end == (u64)-1) goto out; @@ -758,6 +797,7 @@ hit_next: goto out; if (state->end <= end) { set_state_bits(tree, state, bits); + cache_state(state, cached_state); merge_state(tree, state); if (last_end == (u64)-1) goto out; @@ -782,6 +822,7 @@ hit_next: this_end = last_start - 1; err = insert_state(tree, prealloc, start, this_end, bits); + cache_state(prealloc, cached_state); prealloc = NULL; BUG_ON(err == -EEXIST); if (err) @@ -805,6 +846,7 @@ hit_next: BUG_ON(err == -EEXIST); set_state_bits(tree, prealloc, bits); + cache_state(prealloc, cached_state); merge_state(tree, prealloc); prealloc = NULL; goto out; @@ -833,26 +875,27 @@ int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, - mask); + NULL, mask); } int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask); + return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, NULL, + mask); } int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask) { return set_extent_bit(tree, start, end, bits, 0, NULL, - mask); + NULL, mask); } int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask) { - return clear_extent_bit(tree, start, end, bits, 0, 0, mask); + return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); } int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, @@ -860,46 +903,50 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, { return set_extent_bit(tree, start, end, EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, - 0, NULL, mask); + 0, NULL, NULL, mask); } int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { return clear_extent_bit(tree, start, end, - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); + EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, + NULL, mask); } int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask); + return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, + NULL, mask); } int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, - mask); + NULL, mask); } static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); + return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, + NULL, mask); } int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, - mask); + NULL, mask); } static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); + return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, + NULL, mask); } int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) @@ -912,13 +959,14 @@ int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) * us if waiting is desired. */ int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, - int bits, gfp_t mask) + int bits, struct extent_state **cached_state, gfp_t mask) { int err; u64 failed_start; while (1) { err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, - EXTENT_LOCKED, &failed_start, mask); + EXTENT_LOCKED, &failed_start, + cached_state, mask); if (err == -EEXIST && (mask & __GFP_WAIT)) { wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); start = failed_start; @@ -932,7 +980,7 @@ int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return lock_extent_bits(tree, start, end, 0, mask); + return lock_extent_bits(tree, start, end, 0, NULL, mask); } int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, @@ -941,21 +989,29 @@ int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, int err; u64 failed_start; - err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, - &failed_start, mask); + err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, + &failed_start, NULL, mask); if (err == -EEXIST) { if (failed_start > start) clear_extent_bit(tree, start, failed_start - 1, - EXTENT_LOCKED, 1, 0, mask); + EXTENT_LOCKED, 1, 0, NULL, mask); return 0; } return 1; } +int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, + struct extent_state **cached, gfp_t mask) +{ + return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, + mask); +} + int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { - return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); + return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, + mask); } /* @@ -1323,7 +1379,7 @@ int extent_clear_unlock_delalloc(struct inode *inode, if (clear_delalloc) clear_bits |= EXTENT_DELALLOC; - clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS); + clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); if (!(unlock_pages || clear_dirty || set_writeback || end_writeback)) return 0; @@ -2071,6 +2127,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, u64 iosize; u64 unlock_start; sector_t sector; + struct extent_state *cached_state = NULL; struct extent_map *em; struct block_device *bdev; int ret; @@ -2162,7 +2219,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, goto done_unlocked; } } - lock_extent(tree, start, page_end, GFP_NOFS); + lock_extent_bits(tree, start, page_end, 0, &cached_state, GFP_NOFS); unlock_start = start; @@ -2170,7 +2227,8 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, ret = tree->ops->writepage_start_hook(page, start, page_end); if (ret == -EAGAIN) { - unlock_extent(tree, start, page_end, GFP_NOFS); + unlock_extent_cached(tree, start, page_end, + &cached_state, GFP_NOFS); redirty_page_for_writepage(wbc, page); update_nr_written(page, wbc, nr_written); unlock_page(page); @@ -2192,7 +2250,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, if (last_byte <= start) { clear_extent_bit(tree, start, page_end, EXTENT_LOCKED | EXTENT_DIRTY, - 1, 0, GFP_NOFS); + 1, 0, NULL, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, start, page_end, NULL, 1); @@ -2204,7 +2262,8 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, while (cur <= end) { if (cur >= last_byte) { - unlock_extent(tree, unlock_start, page_end, GFP_NOFS); + unlock_extent_cached(tree, unlock_start, page_end, + &cached_state, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, cur, page_end, NULL, 1); @@ -2236,8 +2295,9 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, */ if (compressed || block_start == EXTENT_MAP_HOLE || block_start == EXTENT_MAP_INLINE) { - unlock_extent(tree, unlock_start, cur + iosize - 1, - GFP_NOFS); + unlock_extent_cached(tree, unlock_start, + cur + iosize - 1, &cached_state, + GFP_NOFS); /* * end_io notification does not happen here for @@ -2307,11 +2367,14 @@ done: end_page_writeback(page); } if (unlock_start <= page_end) - unlock_extent(tree, unlock_start, page_end, GFP_NOFS); + unlock_extent_cached(tree, unlock_start, page_end, + &cached_state, GFP_NOFS); unlock_page(page); done_unlocked: + /* drop our reference on any cached states */ + free_extent_state(cached_state); return 0; } @@ -2599,7 +2662,7 @@ int extent_invalidatepage(struct extent_io_tree *tree, wait_on_page_writeback(page); clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, - 1, 1, GFP_NOFS); + 1, 1, NULL, GFP_NOFS); return 0; } @@ -2693,7 +2756,7 @@ int extent_prepare_write(struct extent_io_tree *tree, */ set_extent_bit(tree, block_start, block_start + iosize - 1, - EXTENT_LOCKED, 0, NULL, GFP_NOFS); + EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS); ret = submit_extent_page(READ, tree, page, sector, iosize, page_offset, em->bdev, NULL, 1, @@ -2740,7 +2803,7 @@ int try_release_extent_state(struct extent_map_tree *map, if ((mask & GFP_NOFS) == GFP_NOFS) mask = GFP_NOFS; clear_extent_bit(tree, start, end, EXTENT_UPTODATE, - 1, 1, mask); + 1, 1, NULL, mask); } return ret; } diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index 88d134d01fb..c8ead2b8c4c 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -143,7 +143,7 @@ int try_release_extent_state(struct extent_map_tree *map, gfp_t mask); int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, - int bits, gfp_t mask); + int bits, struct extent_state **cached, gfp_t mask); int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask); @@ -161,7 +161,8 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask); int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, - int bits, int wake, int delete, gfp_t mask); + int bits, int wake, int delete, struct extent_state **cached, + gfp_t mask); int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask); int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index f1df1171861..e494545c420 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -854,7 +854,7 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page, int limit = 10 * 1024 * 1042; clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | - EXTENT_DELALLOC, 1, 0, GFP_NOFS); + EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS); while (start < end) { async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); async_cow->inode = inode; @@ -4420,7 +4420,7 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) */ clear_extent_bit(tree, page_start, page_end, EXTENT_DIRTY | EXTENT_DELALLOC | - EXTENT_LOCKED, 1, 0, GFP_NOFS); + EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); btrfs_finish_ordered_io(page->mapping->host, page_start, page_end); btrfs_put_ordered_extent(ordered); @@ -4429,7 +4429,7 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_ORDERED, - 1, 1, GFP_NOFS); + 1, 1, NULL, GFP_NOFS); __btrfs_releasepage(page, GFP_NOFS); ClearPageChecked(page); From d5550c6315fe0647b7ac21a6a736bf4a42620eac Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 15:11:07 -0400 Subject: [PATCH 11/16] Btrfs: don't lock bits in the extent tree during writepage At writepage time, we have the page locked and we have the extent_map entry for this extent pinned in the extent_map tree. So, the page can't go away and its mapping can't change. There is no need for the extra extent_state lock bits during writepage. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 21 --------------------- 1 file changed, 21 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index c7a5e860fe2..04fafc3cffc 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2219,16 +2219,10 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, goto done_unlocked; } } - lock_extent_bits(tree, start, page_end, 0, &cached_state, GFP_NOFS); - - unlock_start = start; - if (tree->ops && tree->ops->writepage_start_hook) { ret = tree->ops->writepage_start_hook(page, start, page_end); if (ret == -EAGAIN) { - unlock_extent_cached(tree, start, page_end, - &cached_state, GFP_NOFS); redirty_page_for_writepage(wbc, page); update_nr_written(page, wbc, nr_written); unlock_page(page); @@ -2244,13 +2238,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, update_nr_written(page, wbc, nr_written + 1); end = page_end; - if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) - printk(KERN_ERR "btrfs delalloc bits after lock_extent\n"); - if (last_byte <= start) { - clear_extent_bit(tree, start, page_end, - EXTENT_LOCKED | EXTENT_DIRTY, - 1, 0, NULL, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, start, page_end, NULL, 1); @@ -2262,8 +2250,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, while (cur <= end) { if (cur >= last_byte) { - unlock_extent_cached(tree, unlock_start, page_end, - &cached_state, GFP_NOFS); if (tree->ops && tree->ops->writepage_end_io_hook) tree->ops->writepage_end_io_hook(page, cur, page_end, NULL, 1); @@ -2295,10 +2281,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, */ if (compressed || block_start == EXTENT_MAP_HOLE || block_start == EXTENT_MAP_INLINE) { - unlock_extent_cached(tree, unlock_start, - cur + iosize - 1, &cached_state, - GFP_NOFS); - /* * end_io notification does not happen here for * compressed extents @@ -2366,9 +2348,6 @@ done: set_page_writeback(page); end_page_writeback(page); } - if (unlock_start <= page_end) - unlock_extent_cached(tree, unlock_start, page_end, - &cached_state, GFP_NOFS); unlock_page(page); done_unlocked: From 9655d2982b53fdb38a9e0f2f11315b99b92d66e2 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 15:22:30 -0400 Subject: [PATCH 12/16] Btrfs: use a cached state for extent state operations during delalloc This changes the btrfs code to find delalloc ranges in the extent state tree to use the new state caching code from set/test bit. It reduces one of the biggest causes of rbtree searches in the writeback path. test_range_bit is also modified to take the cached state as a starting point while searching. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 46 +++++++++++++++++++++++++++-------------- fs/btrfs/extent_io.h | 2 +- fs/btrfs/inode.c | 6 +++--- fs/btrfs/ordered-data.c | 8 +++---- fs/btrfs/relocation.c | 2 +- 5 files changed, 40 insertions(+), 24 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 04fafc3cffc..c9a438d374b 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -720,6 +720,13 @@ again: } spin_lock(&tree->lock); + if (cached_state && *cached_state) { + state = *cached_state; + if (state->start == start && state->tree) { + node = &state->rb_node; + goto hit_next; + } + } /* * this search will find all the extents that end after * our range starts. @@ -1286,6 +1293,7 @@ static noinline u64 find_lock_delalloc_range(struct inode *inode, u64 delalloc_start; u64 delalloc_end; u64 found; + struct extent_state *cached_state = NULL; int ret; int loops = 0; @@ -1323,6 +1331,7 @@ again: /* some of the pages are gone, lets avoid looping by * shortening the size of the delalloc range we're searching */ + free_extent_state(cached_state); if (!loops) { unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); max_bytes = PAGE_CACHE_SIZE - offset; @@ -1336,18 +1345,21 @@ again: BUG_ON(ret); /* step three, lock the state bits for the whole range */ - lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); + lock_extent_bits(tree, delalloc_start, delalloc_end, + 0, &cached_state, GFP_NOFS); /* then test to make sure it is all still delalloc */ ret = test_range_bit(tree, delalloc_start, delalloc_end, - EXTENT_DELALLOC, 1); + EXTENT_DELALLOC, 1, cached_state); if (!ret) { - unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); + unlock_extent_cached(tree, delalloc_start, delalloc_end, + &cached_state, GFP_NOFS); __unlock_for_delalloc(inode, locked_page, delalloc_start, delalloc_end); cond_resched(); goto again; } + free_extent_state(cached_state); *start = delalloc_start; *end = delalloc_end; out_failed: @@ -1530,14 +1542,17 @@ out: * range is found set. */ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, - int bits, int filled) + int bits, int filled, struct extent_state *cached) { struct extent_state *state = NULL; struct rb_node *node; int bitset = 0; spin_lock(&tree->lock); - node = tree_search(tree, start); + if (cached && cached->tree && cached->start == start) + node = &cached->rb_node; + else + node = tree_search(tree, start); while (node && start <= end) { state = rb_entry(node, struct extent_state, rb_node); @@ -1580,7 +1595,7 @@ static int check_page_uptodate(struct extent_io_tree *tree, { u64 start = (u64)page->index << PAGE_CACHE_SHIFT; u64 end = start + PAGE_CACHE_SIZE - 1; - if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) + if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) SetPageUptodate(page); return 0; } @@ -1594,7 +1609,7 @@ static int check_page_locked(struct extent_io_tree *tree, { u64 start = (u64)page->index << PAGE_CACHE_SHIFT; u64 end = start + PAGE_CACHE_SIZE - 1; - if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) + if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) unlock_page(page); return 0; } @@ -2032,7 +2047,8 @@ static int __extent_read_full_page(struct extent_io_tree *tree, continue; } /* the get_extent function already copied into the page */ - if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { + if (test_range_bit(tree, cur, cur_end, + EXTENT_UPTODATE, 1, NULL)) { check_page_uptodate(tree, page); unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); cur = cur + iosize; @@ -2305,7 +2321,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc, } /* leave this out until we have a page_mkwrite call */ if (0 && !test_range_bit(tree, cur, cur + iosize - 1, - EXTENT_DIRTY, 0)) { + EXTENT_DIRTY, 0, NULL)) { cur = cur + iosize; pg_offset += iosize; continue; @@ -2721,7 +2737,7 @@ int extent_prepare_write(struct extent_io_tree *tree, !isnew && !PageUptodate(page) && (block_off_end > to || block_off_start < from) && !test_range_bit(tree, block_start, cur_end, - EXTENT_UPTODATE, 1)) { + EXTENT_UPTODATE, 1, NULL)) { u64 sector; u64 extent_offset = block_start - em->start; size_t iosize; @@ -2776,7 +2792,7 @@ int try_release_extent_state(struct extent_map_tree *map, int ret = 1; if (test_range_bit(tree, start, end, - EXTENT_IOBITS | EXTENT_ORDERED, 0)) + EXTENT_IOBITS | EXTENT_ORDERED, 0, NULL)) ret = 0; else { if ((mask & GFP_NOFS) == GFP_NOFS) @@ -2821,7 +2837,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, extent_map_end(em) - 1, EXTENT_LOCKED | EXTENT_WRITEBACK | EXTENT_ORDERED, - 0)) { + 0, NULL)) { remove_extent_mapping(map, em); /* once for the rb tree */ free_extent_map(em); @@ -3237,7 +3253,7 @@ int extent_range_uptodate(struct extent_io_tree *tree, int uptodate; unsigned long index; - ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1); + ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL); if (ret) return 1; while (start <= end) { @@ -3267,7 +3283,7 @@ int extent_buffer_uptodate(struct extent_io_tree *tree, return 1; ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, - EXTENT_UPTODATE, 1); + EXTENT_UPTODATE, 1, NULL); if (ret) return ret; @@ -3303,7 +3319,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree, return 0; if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, - EXTENT_UPTODATE, 1)) { + EXTENT_UPTODATE, 1, NULL)) { return 0; } diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index c8ead2b8c4c..09cd6fa3cc8 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -157,7 +157,7 @@ u64 count_range_bits(struct extent_io_tree *tree, u64 max_bytes, unsigned long bits); int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, - int bits, int filled); + int bits, int filled, struct extent_state *cached_state); int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask); int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index e494545c420..3f8e93de298 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -1376,7 +1376,7 @@ again: /* already ordered? We're done */ if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, - EXTENT_ORDERED, 0)) { + EXTENT_ORDERED, 0, NULL)) { goto out; } @@ -1417,7 +1417,7 @@ static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) int ret; ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, - EXTENT_ORDERED, 0); + EXTENT_ORDERED, 0, NULL); if (ret) return 0; @@ -1795,7 +1795,7 @@ static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, return 0; if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && - test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) { + test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, GFP_NOFS); return 0; diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index d6f0806c682..7f751e462f0 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -262,7 +262,7 @@ int btrfs_dec_test_ordered_pending(struct inode *inode, ret = test_range_bit(io_tree, entry->file_offset, entry->file_offset + entry->len - 1, - EXTENT_ORDERED, 0); + EXTENT_ORDERED, 0, NULL); if (ret == 0) ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); out: @@ -522,7 +522,7 @@ again: end--; } if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, - EXTENT_ORDERED | EXTENT_DELALLOC, 0)) { + EXTENT_ORDERED | EXTENT_DELALLOC, 0, NULL)) { schedule_timeout(1); goto again; } @@ -613,7 +613,7 @@ int btrfs_ordered_update_i_size(struct inode *inode, */ if (test_range_bit(io_tree, disk_i_size, ordered->file_offset + ordered->len - 1, - EXTENT_DELALLOC, 0)) { + EXTENT_DELALLOC, 0, NULL)) { goto out; } /* @@ -664,7 +664,7 @@ int btrfs_ordered_update_i_size(struct inode *inode, */ if (i_size_test > entry_end(ordered) && !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, - EXTENT_DELALLOC, 0)) { + EXTENT_DELALLOC, 0, NULL)) { new_i_size = min_t(u64, i_size_test, i_size_read(inode)); } BTRFS_I(inode)->disk_i_size = new_i_size; diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c index 4adab903fc2..3be16ccc7ee 100644 --- a/fs/btrfs/relocation.c +++ b/fs/btrfs/relocation.c @@ -2180,7 +2180,7 @@ static int tree_block_processed(u64 bytenr, u32 blocksize, struct reloc_control *rc) { if (test_range_bit(&rc->processed_blocks, bytenr, - bytenr + blocksize - 1, EXTENT_DIRTY, 1)) + bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) return 1; return 0; } From 8b62b72b26bcd72082c4a69d179dd906bcc22200 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Wed, 2 Sep 2009 16:53:46 -0400 Subject: [PATCH 13/16] Btrfs: Use PagePrivate2 to track pages in the data=ordered code. Btrfs writes go through delalloc to the data=ordered code. This makes sure that all of the data is on disk before the metadata that references it. The tracking means that we have to make sure each page in an extent is fully written before we add that extent into the on-disk btree. This was done in the past by setting the EXTENT_ORDERED bit for the range of an extent when it was added to the data=ordered code, and then clearing the EXTENT_ORDERED bit in the extent state tree as each page finished IO. One of the reasons we had to do this was because sometimes pages are magically dirtied without page_mkwrite being called. The EXTENT_ORDERED bit is checked at writepage time, and if it isn't there, our page become dirty without going through the proper path. These bit operations make for a number of rbtree searches for each page, and can cause considerable lock contention. This commit switches from the EXTENT_ORDERED bit to use PagePrivate2. As pages go into the ordered code, PagePrivate2 is set on each one. This is a cheap operation because we already have all the pages locked and ready to go. As IO finishes, the PagePrivate2 bit is cleared and the ordered accoutning is updated for each page. At writepage time, if the PagePrivate2 bit is missing, we go into the writepage fixup code to handle improperly dirtied pages. Signed-off-by: Chris Mason --- fs/btrfs/extent_io.c | 29 +++++++++---------------- fs/btrfs/extent_io.h | 9 ++++---- fs/btrfs/inode.c | 47 ++++++++++++++++++++++++++--------------- fs/btrfs/ordered-data.c | 29 +++++++++++++------------ fs/btrfs/ordered-data.h | 3 +++ 5 files changed, 62 insertions(+), 55 deletions(-) diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index c9a438d374b..a102422cd92 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -885,13 +885,6 @@ int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, NULL, mask); } -int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, - gfp_t mask) -{ - return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, NULL, - mask); -} - int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, int bits, gfp_t mask) { @@ -921,13 +914,6 @@ int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, NULL, mask); } -int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, - gfp_t mask) -{ - return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, - NULL, mask); -} - int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) { @@ -1373,7 +1359,8 @@ int extent_clear_unlock_delalloc(struct inode *inode, int clear_unlock, int clear_delalloc, int clear_dirty, int set_writeback, - int end_writeback) + int end_writeback, + int set_private2) { int ret; struct page *pages[16]; @@ -1392,7 +1379,8 @@ int extent_clear_unlock_delalloc(struct inode *inode, clear_bits |= EXTENT_DELALLOC; clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); - if (!(unlock_pages || clear_dirty || set_writeback || end_writeback)) + if (!(unlock_pages || clear_dirty || set_writeback || end_writeback || + set_private2)) return 0; while (nr_pages > 0) { @@ -1400,6 +1388,10 @@ int extent_clear_unlock_delalloc(struct inode *inode, min_t(unsigned long, nr_pages, ARRAY_SIZE(pages)), pages); for (i = 0; i < ret; i++) { + + if (set_private2) + SetPagePrivate2(pages[i]); + if (pages[i] == locked_page) { page_cache_release(pages[i]); continue; @@ -2792,7 +2784,7 @@ int try_release_extent_state(struct extent_map_tree *map, int ret = 1; if (test_range_bit(tree, start, end, - EXTENT_IOBITS | EXTENT_ORDERED, 0, NULL)) + EXTENT_IOBITS, 0, NULL)) ret = 0; else { if ((mask & GFP_NOFS) == GFP_NOFS) @@ -2835,8 +2827,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, } if (!test_range_bit(tree, em->start, extent_map_end(em) - 1, - EXTENT_LOCKED | EXTENT_WRITEBACK | - EXTENT_ORDERED, + EXTENT_LOCKED | EXTENT_WRITEBACK, 0, NULL)) { remove_extent_mapping(map, em); /* once for the rb tree */ diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index 09cd6fa3cc8..14ed16fd862 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -13,10 +13,8 @@ #define EXTENT_DEFRAG (1 << 6) #define EXTENT_DEFRAG_DONE (1 << 7) #define EXTENT_BUFFER_FILLED (1 << 8) -#define EXTENT_ORDERED (1 << 9) -#define EXTENT_ORDERED_METADATA (1 << 10) -#define EXTENT_BOUNDARY (1 << 11) -#define EXTENT_NODATASUM (1 << 12) +#define EXTENT_BOUNDARY (1 << 9) +#define EXTENT_NODATASUM (1 << 10) #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) /* flags for bio submission */ @@ -285,5 +283,6 @@ int extent_clear_unlock_delalloc(struct inode *inode, int clear_unlock, int clear_delalloc, int clear_dirty, int set_writeback, - int end_writeback); + int end_writeback, + int set_private2); #endif diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 3f8e93de298..739a245e25d 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -426,7 +426,7 @@ again: extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, start, end, NULL, 1, 0, - 0, 1, 1, 1); + 0, 1, 1, 1, 0); ret = 0; goto free_pages_out; } @@ -641,7 +641,7 @@ static noinline int submit_compressed_extents(struct inode *inode, async_extent->start, async_extent->start + async_extent->ram_size - 1, - NULL, 1, 1, 0, 1, 1, 0); + NULL, 1, 1, 0, 1, 1, 0, 0); ret = btrfs_submit_compressed_write(inode, async_extent->start, @@ -714,7 +714,7 @@ static noinline int cow_file_range(struct inode *inode, extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, start, end, NULL, 1, 1, - 1, 1, 1, 1); + 1, 1, 1, 1, 0); *nr_written = *nr_written + (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; *page_started = 1; @@ -777,11 +777,14 @@ static noinline int cow_file_range(struct inode *inode, /* we're not doing compressed IO, don't unlock the first * page (which the caller expects to stay locked), don't * clear any dirty bits and don't set any writeback bits + * + * Do set the Private2 bit so we know this page was properly + * setup for writepage */ extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, start, start + ram_size - 1, locked_page, unlock, 1, - 1, 0, 0, 0); + 1, 0, 0, 0, 1); disk_num_bytes -= cur_alloc_size; num_bytes -= cur_alloc_size; alloc_hint = ins.objectid + ins.offset; @@ -1102,7 +1105,7 @@ out_check: extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, cur_offset, cur_offset + num_bytes - 1, - locked_page, 1, 1, 1, 0, 0, 0); + locked_page, 1, 1, 1, 0, 0, 0, 1); cur_offset = extent_end; if (cur_offset > end) break; @@ -1375,10 +1378,8 @@ again: lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); /* already ordered? We're done */ - if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, - EXTENT_ORDERED, 0, NULL)) { + if (PagePrivate2(page)) goto out; - } ordered = btrfs_lookup_ordered_extent(inode, page_start); if (ordered) { @@ -1414,11 +1415,9 @@ static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) struct inode *inode = page->mapping->host; struct btrfs_writepage_fixup *fixup; struct btrfs_root *root = BTRFS_I(inode)->root; - int ret; - ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, - EXTENT_ORDERED, 0, NULL); - if (ret) + /* this page is properly in the ordered list */ + if (TestClearPagePrivate2(page)) return 0; if (PageChecked(page)) @@ -1624,6 +1623,7 @@ nocow: static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, struct extent_state *state, int uptodate) { + ClearPagePrivate2(page); return btrfs_finish_ordered_io(page->mapping->host, start, end); } @@ -4403,13 +4403,21 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) u64 page_start = page_offset(page); u64 page_end = page_start + PAGE_CACHE_SIZE - 1; + + /* + * we have the page locked, so new writeback can't start, + * and the dirty bit won't be cleared while we are here. + * + * Wait for IO on this page so that we can safely clear + * the PagePrivate2 bit and do ordered accounting + */ wait_on_page_writeback(page); + tree = &BTRFS_I(page->mapping->host)->io_tree; if (offset) { btrfs_releasepage(page, GFP_NOFS); return; } - lock_extent(tree, page_start, page_end, GFP_NOFS); ordered = btrfs_lookup_ordered_extent(page->mapping->host, page_offset(page)); @@ -4421,14 +4429,19 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) clear_extent_bit(tree, page_start, page_end, EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); - btrfs_finish_ordered_io(page->mapping->host, - page_start, page_end); + /* + * whoever cleared the private bit is responsible + * for the finish_ordered_io + */ + if (TestClearPagePrivate2(page)) { + btrfs_finish_ordered_io(page->mapping->host, + page_start, page_end); + } btrfs_put_ordered_extent(ordered); lock_extent(tree, page_start, page_end, GFP_NOFS); } clear_extent_bit(tree, page_start, page_end, - EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | - EXTENT_ORDERED, + EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 1, 1, NULL, GFP_NOFS); __btrfs_releasepage(page, GFP_NOFS); diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c index 7f751e462f0..4a9c8c4cec2 100644 --- a/fs/btrfs/ordered-data.c +++ b/fs/btrfs/ordered-data.c @@ -159,8 +159,6 @@ static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, * * len is the length of the extent * - * This also sets the EXTENT_ORDERED bit on the range in the inode. - * * The tree is given a single reference on the ordered extent that was * inserted. */ @@ -181,6 +179,7 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, entry->start = start; entry->len = len; entry->disk_len = disk_len; + entry->bytes_left = len; entry->inode = inode; if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) set_bit(type, &entry->flags); @@ -195,9 +194,6 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, &entry->rb_node); BUG_ON(node); - set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset, - entry_end(entry) - 1, GFP_NOFS); - spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); list_add_tail(&entry->root_extent_list, &BTRFS_I(inode)->root->fs_info->ordered_extents); @@ -241,13 +237,10 @@ int btrfs_dec_test_ordered_pending(struct inode *inode, struct btrfs_ordered_inode_tree *tree; struct rb_node *node; struct btrfs_ordered_extent *entry; - struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; int ret; tree = &BTRFS_I(inode)->ordered_tree; mutex_lock(&tree->mutex); - clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1, - GFP_NOFS); node = tree_search(tree, file_offset); if (!node) { ret = 1; @@ -260,11 +253,16 @@ int btrfs_dec_test_ordered_pending(struct inode *inode, goto out; } - ret = test_range_bit(io_tree, entry->file_offset, - entry->file_offset + entry->len - 1, - EXTENT_ORDERED, 0, NULL); - if (ret == 0) + if (io_size > entry->bytes_left) { + printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", + (unsigned long long)entry->bytes_left, + (unsigned long long)io_size); + } + entry->bytes_left -= io_size; + if (entry->bytes_left == 0) ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); + else + ret = 1; out: mutex_unlock(&tree->mutex); return ret == 0; @@ -476,6 +474,7 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) u64 orig_end; u64 wait_end; struct btrfs_ordered_extent *ordered; + int found; if (start + len < start) { orig_end = INT_LIMIT(loff_t); @@ -502,6 +501,7 @@ again: orig_end >> PAGE_CACHE_SHIFT); end = orig_end; + found = 0; while (1) { ordered = btrfs_lookup_first_ordered_extent(inode, end); if (!ordered) @@ -514,6 +514,7 @@ again: btrfs_put_ordered_extent(ordered); break; } + found++; btrfs_start_ordered_extent(inode, ordered, 1); end = ordered->file_offset; btrfs_put_ordered_extent(ordered); @@ -521,8 +522,8 @@ again: break; end--; } - if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, - EXTENT_ORDERED | EXTENT_DELALLOC, 0, NULL)) { + if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, + EXTENT_DELALLOC, 0, NULL)) { schedule_timeout(1); goto again; } diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h index 3d31c8827b0..993a7ea45c7 100644 --- a/fs/btrfs/ordered-data.h +++ b/fs/btrfs/ordered-data.h @@ -85,6 +85,9 @@ struct btrfs_ordered_extent { /* extent length on disk */ u64 disk_len; + /* number of bytes that still need writing */ + u64 bytes_left; + /* flags (described above) */ unsigned long flags; From a1ed835e1ab5795f91b198d08c43e2f56848dcf3 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 11 Sep 2009 12:27:37 -0400 Subject: [PATCH 14/16] Btrfs: Fix extent replacment race Data COW means that whenever we write to a file, we replace any old extent pointers with new ones. There was a window where a readpage might find the old extent pointers on disk and cache them in the extent_map tree in ram in the middle of a given write replacing them. Even though both the readpage and the write had their respective bytes in the file locked, the extent readpage inserts may cover more bytes than it had locked down. This commit closes the race by keeping the new extent pinned in the extent map tree until after the on-disk btree is properly setup with the new extent pointers. Signed-off-by: Chris Mason --- fs/btrfs/ctree.h | 2 +- fs/btrfs/extent_map.c | 50 +++++++++++++++++++++++++++++++++++++++++++ fs/btrfs/extent_map.h | 1 + fs/btrfs/file.c | 8 ++++--- fs/btrfs/inode.c | 25 +++++++++++++++++----- fs/btrfs/ioctl.c | 5 ++--- fs/btrfs/tree-log.c | 2 +- 7 files changed, 80 insertions(+), 13 deletions(-) diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 38eeb6c49c8..1ceab8b4d6d 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -2292,7 +2292,7 @@ extern struct file_operations btrfs_file_operations; int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, u64 start, u64 end, u64 locked_end, - u64 inline_limit, u64 *hint_block); + u64 inline_limit, u64 *hint_block, int drop_cache); int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, u64 start, u64 end); diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c index 72e9fa3c31f..5bc7a0d325e 100644 --- a/fs/btrfs/extent_map.c +++ b/fs/btrfs/extent_map.c @@ -198,6 +198,56 @@ static int mergable_maps(struct extent_map *prev, struct extent_map *next) return 0; } +int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len) +{ + int ret = 0; + struct extent_map *merge = NULL; + struct rb_node *rb; + struct extent_map *em; + + write_lock(&tree->lock); + em = lookup_extent_mapping(tree, start, len); + + WARN_ON(em->start != start || !em); + + if (!em) + goto out; + + clear_bit(EXTENT_FLAG_PINNED, &em->flags); + + if (em->start != 0) { + rb = rb_prev(&em->rb_node); + if (rb) + merge = rb_entry(rb, struct extent_map, rb_node); + if (rb && mergable_maps(merge, em)) { + em->start = merge->start; + em->len += merge->len; + em->block_len += merge->block_len; + em->block_start = merge->block_start; + merge->in_tree = 0; + rb_erase(&merge->rb_node, &tree->map); + free_extent_map(merge); + } + } + + rb = rb_next(&em->rb_node); + if (rb) + merge = rb_entry(rb, struct extent_map, rb_node); + if (rb && mergable_maps(em, merge)) { + em->len += merge->len; + em->block_len += merge->len; + rb_erase(&merge->rb_node, &tree->map); + merge->in_tree = 0; + free_extent_map(merge); + } + + free_extent_map(em); +out: + write_unlock(&tree->lock); + return ret; + +} + /** * add_extent_mapping - add new extent map to the extent tree * @tree: tree to insert new map in diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h index 6216dfbcf9b..d3d442f4bbb 100644 --- a/fs/btrfs/extent_map.h +++ b/fs/btrfs/extent_map.h @@ -59,4 +59,5 @@ struct extent_map *alloc_extent_map(gfp_t mask); void free_extent_map(struct extent_map *em); int __init extent_map_init(void); void extent_map_exit(void); +int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len); #endif diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index ef66c3d989b..4123db9d514 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c @@ -177,10 +177,10 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, } flags = em->flags; if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { - write_unlock(&em_tree->lock); if (em->start <= start && (!testend || em->start + em->len >= start + len)) { free_extent_map(em); + write_unlock(&em_tree->lock); break; } if (start < em->start) { @@ -190,6 +190,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, start = em->start + em->len; } free_extent_map(em); + write_unlock(&em_tree->lock); continue; } compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); @@ -269,7 +270,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct inode *inode, u64 start, u64 end, u64 locked_end, - u64 inline_limit, u64 *hint_byte) + u64 inline_limit, u64 *hint_byte, int drop_cache) { u64 extent_end = 0; u64 search_start = start; @@ -294,7 +295,8 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans, int ret; inline_limit = 0; - btrfs_drop_extent_cache(inode, start, end - 1, 0); + if (drop_cache) + btrfs_drop_extent_cache(inode, start, end - 1, 0); path = btrfs_alloc_path(); if (!path) diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 739a245e25d..233fe6f2612 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -232,7 +232,8 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, } ret = btrfs_drop_extents(trans, root, inode, start, - aligned_end, aligned_end, start, &hint_byte); + aligned_end, aligned_end, start, + &hint_byte, 1); BUG_ON(ret); if (isize > actual_end) @@ -241,7 +242,7 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, inline_len, compressed_size, compressed_pages); BUG_ON(ret); - btrfs_drop_extent_cache(inode, start, aligned_end, 0); + btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); return 0; } @@ -1455,9 +1456,19 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, BUG_ON(!path); path->leave_spinning = 1; + + /* + * we may be replacing one extent in the tree with another. + * The new extent is pinned in the extent map, and we don't want + * to drop it from the cache until it is completely in the btree. + * + * So, tell btrfs_drop_extents to leave this extent in the cache. + * the caller is expected to unpin it and allow it to be merged + * with the others. + */ ret = btrfs_drop_extents(trans, root, inode, file_pos, file_pos + num_bytes, locked_end, - file_pos, &hint); + file_pos, &hint, 0); BUG_ON(ret); ins.objectid = inode->i_ino; @@ -1485,7 +1496,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, btrfs_mark_buffer_dirty(leaf); inode_add_bytes(inode, num_bytes); - btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0); ins.objectid = disk_bytenr; ins.offset = disk_num_bytes; @@ -1596,6 +1606,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) ordered_extent->len, compressed, 0, 0, BTRFS_FILE_EXTENT_REG); + unpin_extent_cache(&BTRFS_I(inode)->extent_tree, + ordered_extent->file_offset, + ordered_extent->len); BUG_ON(ret); } unlock_extent(io_tree, ordered_extent->file_offset, @@ -2940,7 +2953,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size) cur_offset, cur_offset + hole_size, block_end, - cur_offset, &hint_byte); + cur_offset, &hint_byte, 1); if (err) break; err = btrfs_insert_file_extent(trans, root, @@ -5086,6 +5099,8 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans, 0, 0, 0, BTRFS_FILE_EXTENT_PREALLOC); BUG_ON(ret); + btrfs_drop_extent_cache(inode, cur_offset, + cur_offset + ins.offset -1, 0); num_bytes -= ins.offset; cur_offset += ins.offset; alloc_hint = ins.objectid + ins.offset; diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c index 9f4db848db1..e2d8e90259b 100644 --- a/fs/btrfs/ioctl.c +++ b/fs/btrfs/ioctl.c @@ -597,9 +597,8 @@ again: clear_page_dirty_for_io(page); btrfs_set_extent_delalloc(inode, page_start, page_end); - - unlock_extent(io_tree, page_start, page_end, GFP_NOFS); set_page_dirty(page); + unlock_extent(io_tree, page_start, page_end, GFP_NOFS); unlock_page(page); page_cache_release(page); balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1); @@ -977,7 +976,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, /* punch hole in destination first */ btrfs_drop_extents(trans, root, inode, off, off + len, - off + len, 0, &hint_byte); + off + len, 0, &hint_byte, 1); /* clone data */ key.objectid = src->i_ino; diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index d91b0de7c50..8661a7381b3 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -534,7 +534,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans, saved_nbytes = inode_get_bytes(inode); /* drop any overlapping extents */ ret = btrfs_drop_extents(trans, root, inode, - start, extent_end, extent_end, start, &alloc_hint); + start, extent_end, extent_end, start, &alloc_hint, 1); BUG_ON(ret); if (found_type == BTRFS_FILE_EXTENT_REG || From 50a9b214bc6c052caa05a210ebfc1bdf0d7085b2 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 11 Sep 2009 12:33:12 -0400 Subject: [PATCH 15/16] Btrfs: fix btrfs page_mkwrite to return locked page This closes a whole where the page may be written before the page_mkwrite caller has a chance to dirty it (thanks to Nick Piggin) Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index 233fe6f2612..c846482e798 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4552,11 +4552,14 @@ again: } ClearPageChecked(page); set_page_dirty(page); + SetPageUptodate(page); BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; unlock_extent(io_tree, page_start, page_end, GFP_NOFS); out_unlock: + if (!ret) + return VM_FAULT_LOCKED; unlock_page(page); out: return ret; From 93c82d575055f1bd0277acae6f966bebafd80dd5 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Fri, 11 Sep 2009 12:36:29 -0400 Subject: [PATCH 16/16] Btrfs: zero page past end of inline file items When btrfs_get_extent is reading inline file items for readpage, it needs to copy the inline extent into the page. If the inline extent doesn't cover all of the page, that means there is a hole in the file, or that our file is smaller than one page. readpage does zeroing for the case where the file is smaller than one page, but nobody is currently zeroing for the case where there is a hole after the inline item. This commit changes btrfs_get_extent to zero fill the page past the end of the inline item. Signed-off-by: Chris Mason --- fs/btrfs/inode.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index c846482e798..88f9df7bfda 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -4233,6 +4233,11 @@ again: map = kmap(page); read_extent_buffer(leaf, map + pg_offset, ptr, copy_size); + if (pg_offset + copy_size < PAGE_CACHE_SIZE) { + memset(map + pg_offset + copy_size, 0, + PAGE_CACHE_SIZE - pg_offset - + copy_size); + } kunmap(page); } flush_dcache_page(page);