]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
Btrfs: recheck bio against block device when we map the bio
authorJosef Bacik <jbacik@fusionio.com>
Fri, 19 Oct 2012 20:50:56 +0000 (16:50 -0400)
committerJosef Bacik <jbacik@fusionio.com>
Tue, 11 Dec 2012 18:31:32 +0000 (13:31 -0500)
Alex reported a problem where we were writing between chunks on a rbd
device.  The thing is we do bio_add_page using logical offsets, but the
physical offset may be different.  So when we map the bio now check to see
if the bio is still ok with the physical offset, and if it is not split the
bio up and redo the bio_add_page with the physical sector.  This fixes the
problem for Alex and doesn't affect performance in the normal case.  Thanks,

Reported-and-tested-by: Alex Elder <elder@inktank.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
fs/btrfs/volumes.c

index a8adf26864735a4eae99c0ce89dba398f728ed4f..eaaf0bf52791afdb3c8d951a598b02f4efe10c09 100644 (file)
@@ -4217,6 +4217,113 @@ static noinline void schedule_bio(struct btrfs_root *root,
                                   &device->work);
 }
 
+static int bio_size_ok(struct block_device *bdev, struct bio *bio,
+                      sector_t sector)
+{
+       struct bio_vec *prev;
+       struct request_queue *q = bdev_get_queue(bdev);
+       unsigned short max_sectors = queue_max_sectors(q);
+       struct bvec_merge_data bvm = {
+               .bi_bdev = bdev,
+               .bi_sector = sector,
+               .bi_rw = bio->bi_rw,
+       };
+
+       if (bio->bi_vcnt == 0) {
+               WARN_ON(1);
+               return 1;
+       }
+
+       prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
+       if ((bio->bi_size >> 9) > max_sectors)
+               return 0;
+
+       if (!q->merge_bvec_fn)
+               return 1;
+
+       bvm.bi_size = bio->bi_size - prev->bv_len;
+       if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
+               return 0;
+       return 1;
+}
+
+static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
+                             struct bio *bio, u64 physical, int dev_nr,
+                             int rw, int async)
+{
+       struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
+
+       bio->bi_private = bbio;
+       bio->bi_private = merge_stripe_index_into_bio_private(
+                       bio->bi_private, (unsigned int)dev_nr);
+       bio->bi_end_io = btrfs_end_bio;
+       bio->bi_sector = physical >> 9;
+#ifdef DEBUG
+       {
+               struct rcu_string *name;
+
+               rcu_read_lock();
+               name = rcu_dereference(dev->name);
+               pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
+                        "(%s id %llu), size=%u\n", rw,
+                        (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
+                        name->str, dev->devid, bio->bi_size);
+               rcu_read_unlock();
+       }
+#endif
+       bio->bi_bdev = dev->bdev;
+       if (async)
+               schedule_bio(root, dev, rw, bio);
+       else
+               btrfsic_submit_bio(rw, bio);
+}
+
+static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
+                             struct bio *first_bio, struct btrfs_device *dev,
+                             int dev_nr, int rw, int async)
+{
+       struct bio_vec *bvec = first_bio->bi_io_vec;
+       struct bio *bio;
+       int nr_vecs = bio_get_nr_vecs(dev->bdev);
+       u64 physical = bbio->stripes[dev_nr].physical;
+
+again:
+       bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
+       if (!bio)
+               return -ENOMEM;
+
+       while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
+               if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
+                                bvec->bv_offset) < bvec->bv_len) {
+                       u64 len = bio->bi_size;
+
+                       atomic_inc(&bbio->stripes_pending);
+                       submit_stripe_bio(root, bbio, bio, physical, dev_nr,
+                                         rw, async);
+                       physical += len;
+                       goto again;
+               }
+               bvec++;
+       }
+
+       submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
+       return 0;
+}
+
+static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
+{
+       atomic_inc(&bbio->error);
+       if (atomic_dec_and_test(&bbio->stripes_pending)) {
+               bio->bi_private = bbio->private;
+               bio->bi_end_io = bbio->end_io;
+               bio->bi_bdev = (struct block_device *)
+                       (unsigned long)bbio->mirror_num;
+               bio->bi_sector = logical >> 9;
+               kfree(bbio);
+               bio_endio(bio, -EIO);
+       }
+}
+
 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
                  int mirror_num, int async_submit)
 {
@@ -4255,40 +4362,36 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
        atomic_set(&bbio->stripes_pending, bbio->num_stripes);
 
        while (dev_nr < total_devs) {
+               dev = bbio->stripes[dev_nr].dev;
+               if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
+                       bbio_error(bbio, first_bio, logical);
+                       dev_nr++;
+                       continue;
+               }
+
+               /*
+                * Check and see if we're ok with this bio based on it's size
+                * and offset with the given device.
+                */
+               if (!bio_size_ok(dev->bdev, first_bio,
+                                bbio->stripes[dev_nr].physical >> 9)) {
+                       ret = breakup_stripe_bio(root, bbio, first_bio, dev,
+                                                dev_nr, rw, async_submit);
+                       BUG_ON(ret);
+                       dev_nr++;
+                       continue;
+               }
+
                if (dev_nr < total_devs - 1) {
                        bio = bio_clone(first_bio, GFP_NOFS);
                        BUG_ON(!bio); /* -ENOMEM */
                } else {
                        bio = first_bio;
                }
-               bio->bi_private = bbio;
-               bio->bi_private = merge_stripe_index_into_bio_private(
-                               bio->bi_private, (unsigned int)dev_nr);
-               bio->bi_end_io = btrfs_end_bio;
-               bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
-               dev = bbio->stripes[dev_nr].dev;
-               if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
-#ifdef DEBUG
-                       struct rcu_string *name;
-
-                       rcu_read_lock();
-                       name = rcu_dereference(dev->name);
-                       pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
-                                "(%s id %llu), size=%u\n", rw,
-                                (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
-                                name->str, dev->devid, bio->bi_size);
-                       rcu_read_unlock();
-#endif
-                       bio->bi_bdev = dev->bdev;
-                       if (async_submit)
-                               schedule_bio(root, dev, rw, bio);
-                       else
-                               btrfsic_submit_bio(rw, bio);
-               } else {
-                       bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
-                       bio->bi_sector = logical >> 9;
-                       bio_endio(bio, -EIO);
-               }
+
+               submit_stripe_bio(root, bbio, bio,
+                                 bbio->stripes[dev_nr].physical, dev_nr, rw,
+                                 async_submit);
                dev_nr++;
        }
        return 0;