Btrfs: don't take the chunk_mutex/dev_list mutex in statfs V2
authorJosef Bacik <jbacik@fb.com>
Mon, 3 Nov 2014 13:56:50 +0000 (08:56 -0500)
committerChris Mason <clm@fb.com>
Fri, 21 Nov 2014 01:20:05 +0000 (17:20 -0800)
Our gluster boxes get several thousand statfs() calls per second, which begins
to suck hardcore with all of the lock contention on the chunk mutex and dev list
mutex.  We don't really need to hold these things, if we have transient
weirdness with statfs() because of the chunk allocator we don't care, so remove
this locking.

We still need the dev_list lock if you mount with -o alloc_start however, which
is a good argument for nuking that thing from orbit, but that's a patch for
another day.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/super.c

index 54bd91ece35b6c7a1f5b2cdce9e7be8c80d99e6f..dc337d14ef5cc12e3c0c9ac73a084a7a9202b0c1 100644 (file)
@@ -1644,8 +1644,20 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
        int i = 0, nr_devices;
        int ret;
 
+       /*
+        * We aren't under the device list lock, so this is racey-ish, but good
+        * enough for our purposes.
+        */
        nr_devices = fs_info->fs_devices->open_devices;
-       BUG_ON(!nr_devices);
+       if (!nr_devices) {
+               smp_mb();
+               nr_devices = fs_info->fs_devices->open_devices;
+               ASSERT(nr_devices);
+               if (!nr_devices) {
+                       *free_bytes = 0;
+                       return 0;
+               }
+       }
 
        devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
                               GFP_NOFS);
@@ -1670,11 +1682,17 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
        else
                min_stripe_size = BTRFS_STRIPE_LEN;
 
-       list_for_each_entry(device, &fs_devices->devices, dev_list) {
+       if (fs_info->alloc_start)
+               mutex_lock(&fs_devices->device_list_mutex);
+       rcu_read_lock();
+       list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
                if (!device->in_fs_metadata || !device->bdev ||
                    device->is_tgtdev_for_dev_replace)
                        continue;
 
+               if (i >= nr_devices)
+                       break;
+
                avail_space = device->total_bytes - device->bytes_used;
 
                /* align with stripe_len */
@@ -1689,24 +1707,32 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
                skip_space = 1024 * 1024;
 
                /* user can set the offset in fs_info->alloc_start. */
-               if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
-                   device->total_bytes)
+               if (fs_info->alloc_start &&
+                   fs_info->alloc_start + BTRFS_STRIPE_LEN <=
+                   device->total_bytes) {
+                       rcu_read_unlock();
                        skip_space = max(fs_info->alloc_start, skip_space);
 
-               /*
-                * btrfs can not use the free space in [0, skip_space - 1],
-                * we must subtract it from the total. In order to implement
-                * it, we account the used space in this range first.
-                */
-               ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
-                                                    &used_space);
-               if (ret) {
-                       kfree(devices_info);
-                       return ret;
-               }
+                       /*
+                        * btrfs can not use the free space in
+                        * [0, skip_space - 1], we must subtract it from the
+                        * total. In order to implement it, we account the used
+                        * space in this range first.
+                        */
+                       ret = btrfs_account_dev_extents_size(device, 0,
+                                                            skip_space - 1,
+                                                            &used_space);
+                       if (ret) {
+                               kfree(devices_info);
+                               mutex_unlock(&fs_devices->device_list_mutex);
+                               return ret;
+                       }
 
-               /* calc the free space in [0, skip_space - 1] */
-               skip_space -= used_space;
+                       rcu_read_lock();
+
+                       /* calc the free space in [0, skip_space - 1] */
+                       skip_space -= used_space;
+               }
 
                /*
                 * we can use the free space in [0, skip_space - 1], subtract
@@ -1725,6 +1751,9 @@ static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 
                i++;
        }
+       rcu_read_unlock();
+       if (fs_info->alloc_start)
+               mutex_unlock(&fs_devices->device_list_mutex);
 
        nr_devices = i;
 
@@ -1787,8 +1816,6 @@ static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
         * holding chunk_muext to avoid allocating new chunks, holding
         * device_list_mutex to avoid the device being removed
         */
-       mutex_lock(&fs_info->fs_devices->device_list_mutex);
-       mutex_lock(&fs_info->chunk_mutex);
        rcu_read_lock();
        list_for_each_entry_rcu(found, head, list) {
                if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
@@ -1826,15 +1853,10 @@ static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 
        buf->f_bavail = total_free_data;
        ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
-       if (ret) {
-               mutex_unlock(&fs_info->chunk_mutex);
-               mutex_unlock(&fs_info->fs_devices->device_list_mutex);
+       if (ret)
                return ret;
-       }
        buf->f_bavail += div_u64(total_free_data, factor);
        buf->f_bavail = buf->f_bavail >> bits;
-       mutex_unlock(&fs_info->chunk_mutex);
-       mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 
        buf->f_type = BTRFS_SUPER_MAGIC;
        buf->f_bsize = dentry->d_sb->s_blocksize;