btrfs: subpage: make readahead work properly
authorQu Wenruo <wqu@suse.com>
Mon, 15 Mar 2021 05:39:15 +0000 (13:39 +0800)
committerDavid Sterba <dsterba@suse.com>
Tue, 16 Mar 2021 10:06:21 +0000 (11:06 +0100)
In readahead infrastructure, we are using a lot of hard coded PAGE_SHIFT
while we're not doing anything specific to PAGE_SIZE.

One of the most affected part is the radix tree operation of
btrfs_fs_info::reada_tree.

If using PAGE_SHIFT, subpage metadata readahead is broken and does no
help reading metadata ahead.

Fix the problem by using btrfs_fs_info::sectorsize_bits so that
readahead could work for subpage.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/reada.c

index 20fd4aa..06713a8 100644 (file)
@@ -209,7 +209,7 @@ int btree_readahead_hook(struct extent_buffer *eb, int err)
        /* find extent */
        spin_lock(&fs_info->reada_lock);
        re = radix_tree_lookup(&fs_info->reada_tree,
-                              eb->start >> PAGE_SHIFT);
+                              eb->start >> fs_info->sectorsize_bits);
        if (re)
                re->refcnt++;
        spin_unlock(&fs_info->reada_lock);
@@ -240,7 +240,7 @@ static struct reada_zone *reada_find_zone(struct btrfs_device *dev, u64 logical,
        zone = NULL;
        spin_lock(&fs_info->reada_lock);
        ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
-                                    logical >> PAGE_SHIFT, 1);
+                                    logical >> fs_info->sectorsize_bits, 1);
        if (ret == 1 && logical >= zone->start && logical <= zone->end) {
                kref_get(&zone->refcnt);
                spin_unlock(&fs_info->reada_lock);
@@ -283,13 +283,13 @@ static struct reada_zone *reada_find_zone(struct btrfs_device *dev, u64 logical,
 
        spin_lock(&fs_info->reada_lock);
        ret = radix_tree_insert(&dev->reada_zones,
-                               (unsigned long)(zone->end >> PAGE_SHIFT),
-                               zone);
+                       (unsigned long)(zone->end >> fs_info->sectorsize_bits),
+                       zone);
 
        if (ret == -EEXIST) {
                kfree(zone);
                ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
-                                            logical >> PAGE_SHIFT, 1);
+                                       logical >> fs_info->sectorsize_bits, 1);
                if (ret == 1 && logical >= zone->start && logical <= zone->end)
                        kref_get(&zone->refcnt);
                else
@@ -315,7 +315,7 @@ static struct reada_extent *reada_find_extent(struct btrfs_fs_info *fs_info,
        u64 length;
        int real_stripes;
        int nzones = 0;
-       unsigned long index = logical >> PAGE_SHIFT;
+       unsigned long index = logical >> fs_info->sectorsize_bits;
        int dev_replace_is_ongoing;
        int have_zone = 0;
 
@@ -497,7 +497,7 @@ static void reada_extent_put(struct btrfs_fs_info *fs_info,
                             struct reada_extent *re)
 {
        int i;
-       unsigned long index = re->logical >> PAGE_SHIFT;
+       unsigned long index = re->logical >> fs_info->sectorsize_bits;
 
        spin_lock(&fs_info->reada_lock);
        if (--re->refcnt) {
@@ -538,11 +538,12 @@ static void reada_extent_put(struct btrfs_fs_info *fs_info,
 static void reada_zone_release(struct kref *kref)
 {
        struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);
+       struct btrfs_fs_info *fs_info = zone->device->fs_info;
 
-       lockdep_assert_held(&zone->device->fs_info->reada_lock);
+       lockdep_assert_held(&fs_info->reada_lock);
 
        radix_tree_delete(&zone->device->reada_zones,
-                         zone->end >> PAGE_SHIFT);
+                         zone->end >> fs_info->sectorsize_bits);
 
        kfree(zone);
 }
@@ -593,7 +594,7 @@ static int reada_add_block(struct reada_control *rc, u64 logical,
 static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
 {
        int i;
-       unsigned long index = zone->end >> PAGE_SHIFT;
+       unsigned long index = zone->end >> zone->device->fs_info->sectorsize_bits;
 
        for (i = 0; i < zone->ndevs; ++i) {
                struct reada_zone *peer;
@@ -628,7 +629,7 @@ static int reada_pick_zone(struct btrfs_device *dev)
                                             (void **)&zone, index, 1);
                if (ret == 0)
                        break;
-               index = (zone->end >> PAGE_SHIFT) + 1;
+               index = (zone->end >> dev->fs_info->sectorsize_bits) + 1;
                if (zone->locked) {
                        if (zone->elems > top_locked_elems) {
                                top_locked_elems = zone->elems;
@@ -709,7 +710,7 @@ static int reada_start_machine_dev(struct btrfs_device *dev)
         * plugging to speed things up
         */
        ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
-                                    dev->reada_next >> PAGE_SHIFT, 1);
+                               dev->reada_next >> fs_info->sectorsize_bits, 1);
        if (ret == 0 || re->logical > dev->reada_curr_zone->end) {
                ret = reada_pick_zone(dev);
                if (!ret) {
@@ -718,7 +719,7 @@ static int reada_start_machine_dev(struct btrfs_device *dev)
                }
                re = NULL;
                ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
-                                       dev->reada_next >> PAGE_SHIFT, 1);
+                               dev->reada_next >> fs_info->sectorsize_bits, 1);
        }
        if (ret == 0) {
                spin_unlock(&fs_info->reada_lock);
@@ -885,7 +886,7 @@ static void dump_devs(struct btrfs_fs_info *fs_info, int all)
                                pr_cont(" curr off %llu",
                                        device->reada_next - zone->start);
                        pr_cont("\n");
-                       index = (zone->end >> PAGE_SHIFT) + 1;
+                       index = (zone->end >> fs_info->sectorsize_bits) + 1;
                }
                cnt = 0;
                index = 0;
@@ -910,7 +911,7 @@ static void dump_devs(struct btrfs_fs_info *fs_info, int all)
                                }
                        }
                        pr_cont("\n");
-                       index = (re->logical >> PAGE_SHIFT) + 1;
+                       index = (re->logical >> fs_info->sectorsize_bits) + 1;
                        if (++cnt > 15)
                                break;
                }
@@ -926,7 +927,7 @@ static void dump_devs(struct btrfs_fs_info *fs_info, int all)
                if (ret == 0)
                        break;
                if (!re->scheduled) {
-                       index = (re->logical >> PAGE_SHIFT) + 1;
+                       index = (re->logical >> fs_info->sectorsize_bits) + 1;
                        continue;
                }
                pr_debug("re: logical %llu size %u list empty %d scheduled %d",
@@ -942,7 +943,7 @@ static void dump_devs(struct btrfs_fs_info *fs_info, int all)
                        }
                }
                pr_cont("\n");
-               index = (re->logical >> PAGE_SHIFT) + 1;
+               index = (re->logical >> fs_info->sectorsize_bits) + 1;
        }
        spin_unlock(&fs_info->reada_lock);
 }