Merge branches 'for-linus' and 'for-linus-3.2' of git://git.kernel.org/pub/scm/linux...
authorLinus Torvalds <torvalds@linux-foundation.org>
Fri, 16 Dec 2011 20:15:50 +0000 (12:15 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 16 Dec 2011 20:15:50 +0000 (12:15 -0800)
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  Btrfs: unplug every once and a while
  Btrfs: deal with NULL srv_rsv in the delalloc inode reservation code
  Btrfs: only set cache_generation if we setup the block group
  Btrfs: don't panic if orphan item already exists
  Btrfs: fix leaked space in truncate
  Btrfs: fix how we do delalloc reservations and how we free reservations on error
  Btrfs: deal with enospc from dirtying inodes properly
  Btrfs: fix num_workers_starting bug and other bugs in async thread
  BTRFS: Establish i_ops before calling d_instantiate
  Btrfs: add a cond_resched() into the worker loop
  Btrfs: fix ctime update of on-disk inode
  btrfs: keep orphans for subvolume deletion
  Btrfs: fix inaccurate available space on raid0 profile
  Btrfs: fix wrong disk space information of the files
  Btrfs: fix wrong i_size when truncating a file to a larger size
  Btrfs: fix btrfs_end_bio to deal with write errors to a single mirror

* 'for-linus-3.2' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  btrfs: lower the dirty balance poll interval

13 files changed:
fs/btrfs/async-thread.c
fs/btrfs/async-thread.h
fs/btrfs/ctree.h
fs/btrfs/delayed-inode.c
fs/btrfs/disk-io.c
fs/btrfs/extent-tree.c
fs/btrfs/file.c
fs/btrfs/inode.c
fs/btrfs/ioctl.c
fs/btrfs/relocation.c
fs/btrfs/scrub.c
fs/btrfs/super.c
fs/btrfs/volumes.c

index 7ec1409..cb97174 100644 (file)
@@ -64,6 +64,8 @@ struct btrfs_worker_thread {
        int idle;
 };
 
+static int __btrfs_start_workers(struct btrfs_workers *workers);
+
 /*
  * btrfs_start_workers uses kthread_run, which can block waiting for memory
  * for a very long time.  It will actually throttle on page writeback,
@@ -88,27 +90,10 @@ static void start_new_worker_func(struct btrfs_work *work)
 {
        struct worker_start *start;
        start = container_of(work, struct worker_start, work);
-       btrfs_start_workers(start->queue, 1);
+       __btrfs_start_workers(start->queue);
        kfree(start);
 }
 
-static int start_new_worker(struct btrfs_workers *queue)
-{
-       struct worker_start *start;
-       int ret;
-
-       start = kzalloc(sizeof(*start), GFP_NOFS);
-       if (!start)
-               return -ENOMEM;
-
-       start->work.func = start_new_worker_func;
-       start->queue = queue;
-       ret = btrfs_queue_worker(queue->atomic_worker_start, &start->work);
-       if (ret)
-               kfree(start);
-       return ret;
-}
-
 /*
  * helper function to move a thread onto the idle list after it
  * has finished some requests.
@@ -153,12 +138,20 @@ 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;
+       struct worker_start *start;
        unsigned long flags;
 
        rmb();
        if (!workers->atomic_start_pending)
                return;
 
+       start = kzalloc(sizeof(*start), GFP_NOFS);
+       if (!start)
+               return;
+
+       start->work.func = start_new_worker_func;
+       start->queue = workers;
+
        spin_lock_irqsave(&workers->lock, flags);
        if (!workers->atomic_start_pending)
                goto out;
@@ -170,10 +163,11 @@ static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
 
        workers->num_workers_starting += 1;
        spin_unlock_irqrestore(&workers->lock, flags);
-       start_new_worker(workers);
+       btrfs_queue_worker(workers->atomic_worker_start, &start->work);
        return;
 
 out:
+       kfree(start);
        spin_unlock_irqrestore(&workers->lock, flags);
 }
 
@@ -331,7 +325,7 @@ again:
                        run_ordered_completions(worker->workers, work);
 
                        check_pending_worker_creates(worker);
-
+                       cond_resched();
                }
 
                spin_lock_irq(&worker->lock);
@@ -462,56 +456,55 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
  * starts new worker threads.  This does not enforce the max worker
  * count in case you need to temporarily go past it.
  */
-static int __btrfs_start_workers(struct btrfs_workers *workers,
-                                int num_workers)
+static int __btrfs_start_workers(struct btrfs_workers *workers)
 {
        struct btrfs_worker_thread *worker;
        int ret = 0;
-       int i;
 
-       for (i = 0; i < num_workers; i++) {
-               worker = kzalloc(sizeof(*worker), GFP_NOFS);
-               if (!worker) {
-                       ret = -ENOMEM;
-                       goto fail;
-               }
+       worker = kzalloc(sizeof(*worker), GFP_NOFS);
+       if (!worker) {
+               ret = -ENOMEM;
+               goto fail;
+       }
 
-               INIT_LIST_HEAD(&worker->pending);
-               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;
-               worker->task = kthread_run(worker_loop, worker,
-                                          "btrfs-%s-%d", workers->name,
-                                          workers->num_workers + i);
-               if (IS_ERR(worker->task)) {
-                       ret = PTR_ERR(worker->task);
-                       kfree(worker);
-                       goto fail;
-               }
-               spin_lock_irq(&workers->lock);
-               list_add_tail(&worker->worker_list, &workers->idle_list);
-               worker->idle = 1;
-               workers->num_workers++;
-               workers->num_workers_starting--;
-               WARN_ON(workers->num_workers_starting < 0);
-               spin_unlock_irq(&workers->lock);
+       INIT_LIST_HEAD(&worker->pending);
+       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;
+       worker->task = kthread_run(worker_loop, worker,
+                                  "btrfs-%s-%d", workers->name,
+                                  workers->num_workers + 1);
+       if (IS_ERR(worker->task)) {
+               ret = PTR_ERR(worker->task);
+               kfree(worker);
+               goto fail;
        }
+       spin_lock_irq(&workers->lock);
+       list_add_tail(&worker->worker_list, &workers->idle_list);
+       worker->idle = 1;
+       workers->num_workers++;
+       workers->num_workers_starting--;
+       WARN_ON(workers->num_workers_starting < 0);
+       spin_unlock_irq(&workers->lock);
+
        return 0;
 fail:
-       btrfs_stop_workers(workers);
+       spin_lock_irq(&workers->lock);
+       workers->num_workers_starting--;
+       spin_unlock_irq(&workers->lock);
        return ret;
 }
 
-int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
+int btrfs_start_workers(struct btrfs_workers *workers)
 {
        spin_lock_irq(&workers->lock);
-       workers->num_workers_starting += num_workers;
+       workers->num_workers_starting++;
        spin_unlock_irq(&workers->lock);
-       return __btrfs_start_workers(workers, num_workers);
+       return __btrfs_start_workers(workers);
 }
 
 /*
@@ -568,6 +561,7 @@ static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
        struct btrfs_worker_thread *worker;
        unsigned long flags;
        struct list_head *fallback;
+       int ret;
 
 again:
        spin_lock_irqsave(&workers->lock, flags);
@@ -584,7 +578,9 @@ again:
                        workers->num_workers_starting++;
                        spin_unlock_irqrestore(&workers->lock, flags);
                        /* we're below the limit, start another worker */
-                       __btrfs_start_workers(workers, 1);
+                       ret = __btrfs_start_workers(workers);
+                       if (ret)
+                               goto fallback;
                        goto again;
                }
        }
@@ -665,7 +661,7 @@ void btrfs_set_work_high_prio(struct btrfs_work *work)
 /*
  * places a struct btrfs_work into the pending queue of one of the kthreads
  */
-int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
+void btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 {
        struct btrfs_worker_thread *worker;
        unsigned long flags;
@@ -673,7 +669,7 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 
        /* don't requeue something already on a list */
        if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
-               goto out;
+               return;
 
        worker = find_worker(workers);
        if (workers->ordered) {
@@ -712,7 +708,4 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
        if (wake)
                wake_up_process(worker->task);
        spin_unlock_irqrestore(&worker->lock, flags);
-
-out:
-       return 0;
 }
index 5077746..f34cc31 100644 (file)
@@ -109,8 +109,8 @@ struct btrfs_workers {
        char *name;
 };
 
-int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work);
-int btrfs_start_workers(struct btrfs_workers *workers, int num_workers);
+void btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work);
+int btrfs_start_workers(struct btrfs_workers *workers);
 int btrfs_stop_workers(struct btrfs_workers *workers);
 void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
                        struct btrfs_workers *async_starter);
index 50634ab..6738503 100644 (file)
@@ -2692,7 +2692,8 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
 int btrfs_readpage(struct file *file, struct page *page);
 void btrfs_evict_inode(struct inode *inode);
 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc);
-void btrfs_dirty_inode(struct inode *inode, int flags);
+int btrfs_dirty_inode(struct inode *inode);
+int btrfs_update_time(struct file *file);
 struct inode *btrfs_alloc_inode(struct super_block *sb);
 void btrfs_destroy_inode(struct inode *inode);
 int btrfs_drop_inode(struct inode *inode);
index 5b16357..9c1eccc 100644 (file)
@@ -640,8 +640,8 @@ static int btrfs_delayed_inode_reserve_metadata(
         * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
         * we're accounted for.
         */
-       if (!trans->bytes_reserved &&
-           src_rsv != &root->fs_info->delalloc_block_rsv) {
+       if (!src_rsv || (!trans->bytes_reserved &&
+           src_rsv != &root->fs_info->delalloc_block_rsv)) {
                ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
                /*
                 * Since we're under a transaction reserve_metadata_bytes could
index 632f8f3..f44b392 100644 (file)
@@ -2194,19 +2194,27 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        fs_info->endio_meta_write_workers.idle_thresh = 2;
        fs_info->readahead_workers.idle_thresh = 2;
 
-       btrfs_start_workers(&fs_info->workers, 1);
-       btrfs_start_workers(&fs_info->generic_worker, 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, 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);
-       btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
-       btrfs_start_workers(&fs_info->delayed_workers, 1);
-       btrfs_start_workers(&fs_info->caching_workers, 1);
-       btrfs_start_workers(&fs_info->readahead_workers, 1);
+       /*
+        * btrfs_start_workers can really only fail because of ENOMEM so just
+        * return -ENOMEM if any of these fail.
+        */
+       ret = btrfs_start_workers(&fs_info->workers);
+       ret |= btrfs_start_workers(&fs_info->generic_worker);
+       ret |= btrfs_start_workers(&fs_info->submit_workers);
+       ret |= btrfs_start_workers(&fs_info->delalloc_workers);
+       ret |= btrfs_start_workers(&fs_info->fixup_workers);
+       ret |= btrfs_start_workers(&fs_info->endio_workers);
+       ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
+       ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
+       ret |= btrfs_start_workers(&fs_info->endio_write_workers);
+       ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
+       ret |= btrfs_start_workers(&fs_info->delayed_workers);
+       ret |= btrfs_start_workers(&fs_info->caching_workers);
+       ret |= btrfs_start_workers(&fs_info->readahead_workers);
+       if (ret) {
+               ret = -ENOMEM;
+               goto fail_sb_buffer;
+       }
 
        fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
        fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
index 2ad8136..f5fbe57 100644 (file)
@@ -2822,7 +2822,7 @@ out_free:
        btrfs_release_path(path);
 out:
        spin_lock(&block_group->lock);
-       if (!ret)
+       if (!ret && dcs == BTRFS_DC_SETUP)
                block_group->cache_generation = trans->transid;
        block_group->disk_cache_state = dcs;
        spin_unlock(&block_group->lock);
@@ -4204,12 +4204,17 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
        u64 to_reserve = 0;
+       u64 csum_bytes;
        unsigned nr_extents = 0;
+       int extra_reserve = 0;
        int flush = 1;
        int ret;
 
+       /* Need to be holding the i_mutex here if we aren't free space cache */
        if (btrfs_is_free_space_inode(root, inode))
                flush = 0;
+       else
+               WARN_ON(!mutex_is_locked(&inode->i_mutex));
 
        if (flush && btrfs_transaction_in_commit(root->fs_info))
                schedule_timeout(1);
@@ -4220,11 +4225,9 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
        BTRFS_I(inode)->outstanding_extents++;
 
        if (BTRFS_I(inode)->outstanding_extents >
-           BTRFS_I(inode)->reserved_extents) {
+           BTRFS_I(inode)->reserved_extents)
                nr_extents = BTRFS_I(inode)->outstanding_extents -
                        BTRFS_I(inode)->reserved_extents;
-               BTRFS_I(inode)->reserved_extents += nr_extents;
-       }
 
        /*
         * Add an item to reserve for updating the inode when we complete the
@@ -4232,11 +4235,12 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
         */
        if (!BTRFS_I(inode)->delalloc_meta_reserved) {
                nr_extents++;
-               BTRFS_I(inode)->delalloc_meta_reserved = 1;
+               extra_reserve = 1;
        }
 
        to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
        to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
+       csum_bytes = BTRFS_I(inode)->csum_bytes;
        spin_unlock(&BTRFS_I(inode)->lock);
 
        ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
@@ -4246,22 +4250,35 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
 
                spin_lock(&BTRFS_I(inode)->lock);
                dropped = drop_outstanding_extent(inode);
-               to_free = calc_csum_metadata_size(inode, num_bytes, 0);
-               spin_unlock(&BTRFS_I(inode)->lock);
-               to_free += btrfs_calc_trans_metadata_size(root, dropped);
-
                /*
-                * Somebody could have come in and twiddled with the
-                * reservation, so if we have to free more than we would have
-                * reserved from this reservation go ahead and release those
-                * bytes.
+                * If the inodes csum_bytes is the same as the original
+                * csum_bytes then we know we haven't raced with any free()ers
+                * so we can just reduce our inodes csum bytes and carry on.
+                * Otherwise we have to do the normal free thing to account for
+                * the case that the free side didn't free up its reserve
+                * because of this outstanding reservation.
                 */
-               to_free -= to_reserve;
+               if (BTRFS_I(inode)->csum_bytes == csum_bytes)
+                       calc_csum_metadata_size(inode, num_bytes, 0);
+               else
+                       to_free = calc_csum_metadata_size(inode, num_bytes, 0);
+               spin_unlock(&BTRFS_I(inode)->lock);
+               if (dropped)
+                       to_free += btrfs_calc_trans_metadata_size(root, dropped);
+
                if (to_free)
                        btrfs_block_rsv_release(root, block_rsv, to_free);
                return ret;
        }
 
+       spin_lock(&BTRFS_I(inode)->lock);
+       if (extra_reserve) {
+               BTRFS_I(inode)->delalloc_meta_reserved = 1;
+               nr_extents--;
+       }
+       BTRFS_I(inode)->reserved_extents += nr_extents;
+       spin_unlock(&BTRFS_I(inode)->lock);
+
        block_rsv_add_bytes(block_rsv, to_reserve, 1);
 
        return 0;
index dafdfa0..97fbe93 100644 (file)
@@ -1167,6 +1167,8 @@ static noinline ssize_t __btrfs_buffered_write(struct file *file,
        nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
                     PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
                     (sizeof(struct page *)));
+       nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
+       nrptrs = max(nrptrs, 8);
        pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
        if (!pages)
                return -ENOMEM;
@@ -1387,7 +1389,11 @@ static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
                goto out;
        }
 
-       file_update_time(file);
+       err = btrfs_update_time(file);
+       if (err) {
+               mutex_unlock(&inode->i_mutex);
+               goto out;
+       }
        BTRFS_I(inode)->sequence++;
 
        start_pos = round_down(pos, root->sectorsize);
index 2c984f7..0a6b928 100644 (file)
@@ -38,6 +38,7 @@
 #include <linux/falloc.h>
 #include <linux/slab.h>
 #include <linux/ratelimit.h>
+#include <linux/mount.h>
 #include "compat.h"
 #include "ctree.h"
 #include "disk-io.h"
@@ -2031,7 +2032,7 @@ int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
        /* insert an orphan item to track this unlinked/truncated file */
        if (insert >= 1) {
                ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
-               BUG_ON(ret);
+               BUG_ON(ret && ret != -EEXIST);
        }
 
        /* insert an orphan item to track subvolume contains orphan files */
@@ -2158,6 +2159,38 @@ int btrfs_orphan_cleanup(struct btrfs_root *root)
                if (ret && ret != -ESTALE)
                        goto out;
 
+               if (ret == -ESTALE && root == root->fs_info->tree_root) {
+                       struct btrfs_root *dead_root;
+                       struct btrfs_fs_info *fs_info = root->fs_info;
+                       int is_dead_root = 0;
+
+                       /*
+                        * this is an orphan in the tree root. Currently these
+                        * could come from 2 sources:
+                        *  a) a snapshot deletion in progress
+                        *  b) a free space cache inode
+                        * We need to distinguish those two, as the snapshot
+                        * orphan must not get deleted.
+                        * find_dead_roots already ran before us, so if this
+                        * is a snapshot deletion, we should find the root
+                        * in the dead_roots list
+                        */
+                       spin_lock(&fs_info->trans_lock);
+                       list_for_each_entry(dead_root, &fs_info->dead_roots,
+                                           root_list) {
+                               if (dead_root->root_key.objectid ==
+                                   found_key.objectid) {
+                                       is_dead_root = 1;
+                                       break;
+                               }
+                       }
+                       spin_unlock(&fs_info->trans_lock);
+                       if (is_dead_root) {
+                               /* prevent this orphan from being found again */
+                               key.offset = found_key.objectid - 1;
+                               continue;
+                       }
+               }
                /*
                 * Inode is already gone but the orphan item is still there,
                 * kill the orphan item.
@@ -2191,7 +2224,14 @@ int btrfs_orphan_cleanup(struct btrfs_root *root)
                                continue;
                        }
                        nr_truncate++;
+                       /*
+                        * Need to hold the imutex for reservation purposes, not
+                        * a huge deal here but I have a WARN_ON in
+                        * btrfs_delalloc_reserve_space to catch offenders.
+                        */
+                       mutex_lock(&inode->i_mutex);
                        ret = btrfs_truncate(inode);
+                       mutex_unlock(&inode->i_mutex);
                } else {
                        nr_unlink++;
                }
@@ -3327,7 +3367,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
                        u64 hint_byte = 0;
                        hole_size = last_byte - cur_offset;
 
-                       trans = btrfs_start_transaction(root, 2);
+                       trans = btrfs_start_transaction(root, 3);
                        if (IS_ERR(trans)) {
                                err = PTR_ERR(trans);
                                break;
@@ -3337,6 +3377,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
                                                 cur_offset + hole_size,
                                                 &hint_byte, 1);
                        if (err) {
+                               btrfs_update_inode(trans, root, inode);
                                btrfs_end_transaction(trans, root);
                                break;
                        }
@@ -3346,6 +3387,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
                                        0, hole_size, 0, hole_size,
                                        0, 0, 0);
                        if (err) {
+                               btrfs_update_inode(trans, root, inode);
                                btrfs_end_transaction(trans, root);
                                break;
                        }
@@ -3353,6 +3395,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
                        btrfs_drop_extent_cache(inode, hole_start,
                                        last_byte - 1, 0);
 
+                       btrfs_update_inode(trans, root, inode);
                        btrfs_end_transaction(trans, root);
                }
                free_extent_map(em);
@@ -3370,6 +3413,8 @@ int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
 
 static int btrfs_setsize(struct inode *inode, loff_t newsize)
 {
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct btrfs_trans_handle *trans;
        loff_t oldsize = i_size_read(inode);
        int ret;
 
@@ -3377,16 +3422,19 @@ static int btrfs_setsize(struct inode *inode, loff_t newsize)
                return 0;
 
        if (newsize > oldsize) {
-               i_size_write(inode, newsize);
-               btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
                truncate_pagecache(inode, oldsize, newsize);
                ret = btrfs_cont_expand(inode, oldsize, newsize);
-               if (ret) {
-                       btrfs_setsize(inode, oldsize);
+               if (ret)
                        return ret;
-               }
 
-               mark_inode_dirty(inode);
+               trans = btrfs_start_transaction(root, 1);
+               if (IS_ERR(trans))
+                       return PTR_ERR(trans);
+
+               i_size_write(inode, newsize);
+               btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
+               ret = btrfs_update_inode(trans, root, inode);
+               btrfs_end_transaction_throttle(trans, root);
        } else {
 
                /*
@@ -3426,9 +3474,9 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 
        if (attr->ia_valid) {
                setattr_copy(inode, attr);
-               mark_inode_dirty(inode);
+               err = btrfs_dirty_inode(inode);
 
-               if (attr->ia_valid & ATTR_MODE)
+               if (!err && attr->ia_valid & ATTR_MODE)
                        err = btrfs_acl_chmod(inode);
        }
 
@@ -4204,42 +4252,80 @@ int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
  * FIXME, needs more benchmarking...there are no reasons other than performance
  * to keep or drop this code.
  */
-void btrfs_dirty_inode(struct inode *inode, int flags)
+int btrfs_dirty_inode(struct inode *inode)
 {
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_trans_handle *trans;
        int ret;
 
        if (BTRFS_I(inode)->dummy_inode)
-               return;
+               return 0;
 
        trans = btrfs_join_transaction(root);
-       BUG_ON(IS_ERR(trans));
+       if (IS_ERR(trans))
+               return PTR_ERR(trans);
 
        ret = btrfs_update_inode(trans, root, inode);
        if (ret && ret == -ENOSPC) {
                /* whoops, lets try again with the full transaction */
                btrfs_end_transaction(trans, root);
                trans = btrfs_start_transaction(root, 1);
-               if (IS_ERR(trans)) {
-                       printk_ratelimited(KERN_ERR "btrfs: fail to "
-                                      "dirty  inode %llu error %ld\n",
-                                      (unsigned long long)btrfs_ino(inode),
-                                      PTR_ERR(trans));
-                       return;
-               }
+               if (IS_ERR(trans))
+                       return PTR_ERR(trans);
 
                ret = btrfs_update_inode(trans, root, inode);
-               if (ret) {
-                       printk_ratelimited(KERN_ERR "btrfs: fail to "
-                                      "dirty  inode %llu error %d\n",
-                                      (unsigned long long)btrfs_ino(inode),
-                                      ret);
-               }
        }
        btrfs_end_transaction(trans, root);
        if (BTRFS_I(inode)->delayed_node)
                btrfs_balance_delayed_items(root);
+
+       return ret;
+}
+
+/*
+ * This is a copy of file_update_time.  We need this so we can return error on
+ * ENOSPC for updating the inode in the case of file write and mmap writes.
+ */
+int btrfs_update_time(struct file *file)
+{
+       struct inode *inode = file->f_path.dentry->d_inode;
+       struct timespec now;
+       int ret;
+       enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
+
+       /* First try to exhaust all avenues to not sync */
+       if (IS_NOCMTIME(inode))
+               return 0;
+
+       now = current_fs_time(inode->i_sb);
+       if (!timespec_equal(&inode->i_mtime, &now))
+               sync_it = S_MTIME;
+
+       if (!timespec_equal(&inode->i_ctime, &now))
+               sync_it |= S_CTIME;
+
+       if (IS_I_VERSION(inode))
+               sync_it |= S_VERSION;
+
+       if (!sync_it)
+               return 0;
+
+       /* Finally allowed to write? Takes lock. */
+       if (mnt_want_write_file(file))
+               return 0;
+
+       /* Only change inode inside the lock region */
+       if (sync_it & S_VERSION)
+               inode_inc_iversion(inode);
+       if (sync_it & S_CTIME)
+               inode->i_ctime = now;
+       if (sync_it & S_MTIME)
+               inode->i_mtime = now;
+       ret = btrfs_dirty_inode(inode);
+       if (!ret)
+               mark_inode_dirty_sync(inode);
+       mnt_drop_write(file->f_path.mnt);
+       return ret;
 }
 
 /*
@@ -4555,11 +4641,18 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
                goto out_unlock;
        }
 
+       /*
+       * If the active LSM wants to access the inode during
+       * d_instantiate it needs these. Smack checks to see
+       * if the filesystem supports xattrs by looking at the
+       * ops vector.
+       */
+
+       inode->i_op = &btrfs_special_inode_operations;
        err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
        if (err)
                drop_inode = 1;
        else {
-               inode->i_op = &btrfs_special_inode_operations;
                init_special_inode(inode, inode->i_mode, rdev);
                btrfs_update_inode(trans, root, inode);
        }
@@ -4613,14 +4706,21 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                goto out_unlock;
        }
 
+       /*
+       * If the active LSM wants to access the inode during
+       * d_instantiate it needs these. Smack checks to see
+       * if the filesystem supports xattrs by looking at the
+       * ops vector.
+       */
+       inode->i_fop = &btrfs_file_operations;
+       inode->i_op = &btrfs_file_inode_operations;
+
        err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
        if (err)
                drop_inode = 1;
        else {
                inode->i_mapping->a_ops = &btrfs_aops;
                inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
-               inode->i_fop = &btrfs_file_operations;
-               inode->i_op = &btrfs_file_inode_operations;
                BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
        }
 out_unlock:
@@ -6303,7 +6403,12 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
        u64 page_start;
        u64 page_end;
 
+       /* Need this to keep space reservations serialized */
+       mutex_lock(&inode->i_mutex);
        ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
+       mutex_unlock(&inode->i_mutex);
+       if (!ret)
+               ret = btrfs_update_time(vma->vm_file);
        if (ret) {
                if (ret == -ENOMEM)
                        ret = VM_FAULT_OOM;
@@ -6515,8 +6620,9 @@ static int btrfs_truncate(struct inode *inode)
                        /* Just need the 1 for updating the inode */
                        trans = btrfs_start_transaction(root, 1);
                        if (IS_ERR(trans)) {
-                               err = PTR_ERR(trans);
-                               goto out;
+                               ret = err = PTR_ERR(trans);
+                               trans = NULL;
+                               break;
                        }
                }
 
@@ -7076,14 +7182,21 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
                goto out_unlock;
        }
 
+       /*
+       * If the active LSM wants to access the inode during
+       * d_instantiate it needs these. Smack checks to see
+       * if the filesystem supports xattrs by looking at the
+       * ops vector.
+       */
+       inode->i_fop = &btrfs_file_operations;
+       inode->i_op = &btrfs_file_inode_operations;
+
        err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
        if (err)
                drop_inode = 1;
        else {
                inode->i_mapping->a_ops = &btrfs_aops;
                inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
-               inode->i_fop = &btrfs_file_operations;
-               inode->i_op = &btrfs_file_inode_operations;
                BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
        }
        if (drop_inode)
@@ -7353,6 +7466,7 @@ static const struct inode_operations btrfs_symlink_inode_operations = {
        .follow_link    = page_follow_link_light,
        .put_link       = page_put_link,
        .getattr        = btrfs_getattr,
+       .setattr        = btrfs_setattr,
        .permission     = btrfs_permission,
        .setxattr       = btrfs_setxattr,
        .getxattr       = btrfs_getxattr,
index 72d4616..c04f02c 100644 (file)
@@ -252,11 +252,11 @@ static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
        trans = btrfs_join_transaction(root);
        BUG_ON(IS_ERR(trans));
 
+       btrfs_update_iflags(inode);
+       inode->i_ctime = CURRENT_TIME;
        ret = btrfs_update_inode(trans, root, inode);
        BUG_ON(ret);
 
-       btrfs_update_iflags(inode);
-       inode->i_ctime = CURRENT_TIME;
        btrfs_end_transaction(trans, root);
 
        mnt_drop_write(file->f_path.mnt);
@@ -858,8 +858,10 @@ static int cluster_pages_for_defrag(struct inode *inode,
                return 0;
        file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 
+       mutex_lock(&inode->i_mutex);
        ret = btrfs_delalloc_reserve_space(inode,
                                           num_pages << PAGE_CACHE_SHIFT);
+       mutex_unlock(&inode->i_mutex);
        if (ret)
                return ret;
 again:
index dff29d5..cfb5543 100644 (file)
@@ -2947,7 +2947,9 @@ static int relocate_file_extent_cluster(struct inode *inode,
        index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
        last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
        while (index <= last_index) {
+               mutex_lock(&inode->i_mutex);
                ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
+               mutex_unlock(&inode->i_mutex);
                if (ret)
                        goto out;
 
index c27bcb6..ddf2c90 100644 (file)
@@ -1535,18 +1535,22 @@ static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
 {
        struct btrfs_fs_info *fs_info = root->fs_info;
+       int ret = 0;
 
        mutex_lock(&fs_info->scrub_lock);
        if (fs_info->scrub_workers_refcnt == 0) {
                btrfs_init_workers(&fs_info->scrub_workers, "scrub",
                           fs_info->thread_pool_size, &fs_info->generic_worker);
                fs_info->scrub_workers.idle_thresh = 4;
-               btrfs_start_workers(&fs_info->scrub_workers, 1);
+               ret = btrfs_start_workers(&fs_info->scrub_workers);
+               if (ret)
+                       goto out;
        }
        ++fs_info->scrub_workers_refcnt;
+out:
        mutex_unlock(&fs_info->scrub_lock);
 
-       return 0;
+       return ret;
 }
 
 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
index e28ad4b..200f63b 100644 (file)
@@ -41,6 +41,7 @@
 #include <linux/slab.h>
 #include <linux/cleancache.h>
 #include <linux/mnt_namespace.h>
+#include <linux/ratelimit.h>
 #include "compat.h"
 #include "delayed-inode.h"
 #include "ctree.h"
@@ -1053,7 +1054,7 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
        u64 avail_space;
        u64 used_space;
        u64 min_stripe_size;
-       int min_stripes = 1;
+       int min_stripes = 1, num_stripes = 1;
        int i = 0, nr_devices;
        int ret;
 
@@ -1067,12 +1068,16 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 
        /* calc min stripe number for data space alloction */
        type = btrfs_get_alloc_profile(root, 1);
-       if (type & BTRFS_BLOCK_GROUP_RAID0)
+       if (type & BTRFS_BLOCK_GROUP_RAID0) {
                min_stripes = 2;
-       else if (type & BTRFS_BLOCK_GROUP_RAID1)
+               num_stripes = nr_devices;
+       } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
                min_stripes = 2;
-       else if (type & BTRFS_BLOCK_GROUP_RAID10)
+               num_stripes = 2;
+       } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
                min_stripes = 4;
+               num_stripes = 4;
+       }
 
        if (type & BTRFS_BLOCK_GROUP_DUP)
                min_stripe_size = 2 * BTRFS_STRIPE_LEN;
@@ -1141,13 +1146,16 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
        i = nr_devices - 1;
        avail_space = 0;
        while (nr_devices >= min_stripes) {
+               if (num_stripes > nr_devices)
+                       num_stripes = nr_devices;
+
                if (devices_info[i].max_avail >= min_stripe_size) {
                        int j;
                        u64 alloc_size;
 
-                       avail_space += devices_info[i].max_avail * min_stripes;
+                       avail_space += devices_info[i].max_avail * num_stripes;
                        alloc_size = devices_info[i].max_avail;
-                       for (j = i + 1 - min_stripes; j <= i; j++)
+                       for (j = i + 1 - num_stripes; j <= i; j++)
                                devices_info[j].max_avail -= alloc_size;
                }
                i--;
@@ -1264,6 +1272,16 @@ static int btrfs_unfreeze(struct super_block *sb)
        return 0;
 }
 
+static void btrfs_fs_dirty_inode(struct inode *inode, int flags)
+{
+       int ret;
+
+       ret = btrfs_dirty_inode(inode);
+       if (ret)
+               printk_ratelimited(KERN_ERR "btrfs: fail to dirty inode %Lu "
+                                  "error %d\n", btrfs_ino(inode), ret);
+}
+
 static const struct super_operations btrfs_super_ops = {
        .drop_inode     = btrfs_drop_inode,
        .evict_inode    = btrfs_evict_inode,
@@ -1271,7 +1289,7 @@ static const struct super_operations btrfs_super_ops = {
        .sync_fs        = btrfs_sync_fs,
        .show_options   = btrfs_show_options,
        .write_inode    = btrfs_write_inode,
-       .dirty_inode    = btrfs_dirty_inode,
+       .dirty_inode    = btrfs_fs_dirty_inode,
        .alloc_inode    = btrfs_alloc_inode,
        .destroy_inode  = btrfs_destroy_inode,
        .statfs         = btrfs_statfs,
index 0a8c8f8..f4b839f 100644 (file)
@@ -295,6 +295,12 @@ loop_lock:
                        btrfs_requeue_work(&device->work);
                        goto done;
                }
+               /* unplug every 64 requests just for good measure */
+               if (batch_run % 64 == 0) {
+                       blk_finish_plug(&plug);
+                       blk_start_plug(&plug);
+                       sync_pending = 0;
+               }
        }
 
        cond_resched();
@@ -3258,7 +3264,7 @@ static void btrfs_end_bio(struct bio *bio, int err)
                 */
                if (atomic_read(&bbio->error) > bbio->max_errors) {
                        err = -EIO;
-               } else if (err) {
+               } else {
                        /*
                         * this bio is actually up to date, we didn't
                         * go over the max number of errors