]> git.baikalelectronics.ru Git - kernel.git/commitdiff
btrfs: relax block-group-tree feature dependency checks
authorQu Wenruo <wqu@suse.com>
Mon, 12 Sep 2022 05:44:37 +0000 (13:44 +0800)
committerDavid Sterba <dsterba@suse.com>
Mon, 26 Sep 2022 10:28:07 +0000 (12:28 +0200)
[BUG]
When one user did a wrong attempt to clear block group tree, which can
not be done through mount option, by using "-o clear_cache,space_cache=v2",
it will cause the following error on a fs with block-group-tree feature:

  BTRFS info (device dm-1): force clearing of disk cache
  BTRFS info (device dm-1): using free space tree
  BTRFS info (device dm-1): clearing free space tree
  BTRFS info (device dm-1): clearing compat-ro feature flag for FREE_SPACE_TREE (0x1)
  BTRFS info (device dm-1): clearing compat-ro feature flag for FREE_SPACE_TREE_VALID (0x2)
  BTRFS error (device dm-1): block-group-tree feature requires fres-space-tree and no-holes
  BTRFS error (device dm-1): super block corruption detected before writing it to disk
  BTRFS: error (device dm-1) in write_all_supers:4318: errno=-117 Filesystem corrupted (unexpected superblock corruption detected)
  BTRFS warning (device dm-1: state E): Skipping commit of aborted transaction.

[CAUSE]
Although the dependency for block-group-tree feature is just an
artificial one (to reduce test matrix), we put the dependency check into
btrfs_validate_super().

This is too strict, and during space cache clearing, we will have a
window where free space tree is cleared, and we need to commit the super
block.

In that window, we had block group tree without v2 cache, and triggered
the artificial dependency check.

This is not necessary at all, especially for such a soft dependency.

[FIX]
Introduce a new helper, btrfs_check_features(), to do all the runtime
limitation checks, including:

- Unsupported incompat flags check

- Unsupported compat RO flags check

- Setting missing incompat flags

- Artificial feature dependency checks
  Currently only block group tree will rely on this.

- Subpage runtime check for v1 cache

With this helper, we can move quite some checks from
open_ctree()/btrfs_remount() into it, and just call it after
btrfs_parse_options().

Now "-o clear_cache,space_cache=v2" will not trigger the above error
anymore.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ edit messages ]
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/disk-io.c
fs/btrfs/disk-io.h
fs/btrfs/super.c

index 2132dec3c87c00ba54161b83ccbc56761a2e9468..a2da9313c6947c5111c38d2eaa73f14bf5075e9f 100644 (file)
@@ -3293,6 +3293,112 @@ out:
        return ret;
 }
 
+/*
+ * Do various sanity and dependency checks of different features.
+ *
+ * This is the place for less strict checks (like for subpage or artificial
+ * feature dependencies).
+ *
+ * For strict checks or possible corruption detection, see
+ * btrfs_validate_super().
+ *
+ * This should be called after btrfs_parse_options(), as some mount options
+ * (space cache related) can modify on-disk format like free space tree and
+ * screw up certain feature dependencies.
+ */
+int btrfs_check_features(struct btrfs_fs_info *fs_info, struct super_block *sb)
+{
+       struct btrfs_super_block *disk_super = fs_info->super_copy;
+       u64 incompat = btrfs_super_incompat_flags(disk_super);
+       const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
+       const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
+
+       if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
+               btrfs_err(fs_info,
+               "cannot mount because of unknown incompat features (0x%llx)",
+                   incompat);
+               return -EINVAL;
+       }
+
+       /* Runtime limitation for mixed block groups. */
+       if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
+           (fs_info->sectorsize != fs_info->nodesize)) {
+               btrfs_err(fs_info,
+"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
+                       fs_info->nodesize, fs_info->sectorsize);
+               return -EINVAL;
+       }
+
+       /* Mixed backref is an always-enabled feature. */
+       incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
+
+       /* Set compression related flags just in case. */
+       if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
+               incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
+       else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
+               incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
+
+       /*
+        * An ancient flag, which should really be marked deprecated.
+        * Such runtime limitation doesn't really need a incompat flag.
+        */
+       if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
+               incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
+
+       if (compat_ro_unsupp && !sb_rdonly(sb)) {
+               btrfs_err(fs_info,
+       "cannot mount read-write because of unknown compat_ro features (0x%llx)",
+                      compat_ro);
+               return -EINVAL;
+       }
+
+       /*
+        * We have unsupported RO compat features, although RO mounted, we
+        * should not cause any metadata writes, including log replay.
+        * Or we could screw up whatever the new feature requires.
+        */
+       if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
+           !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
+               btrfs_err(fs_info,
+"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
+                         compat_ro);
+               return -EINVAL;
+       }
+
+       /*
+        * Artificial limitations for block group tree, to force
+        * block-group-tree to rely on no-holes and free-space-tree.
+        */
+       if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
+           (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
+            !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
+               btrfs_err(fs_info,
+"block-group-tree feature requires no-holes and free-space-tree features");
+               return -EINVAL;
+       }
+
+       /*
+        * Subpage runtime limitation on v1 cache.
+        *
+        * V1 space cache still has some hard codeed PAGE_SIZE usage, while
+        * we're already defaulting to v2 cache, no need to bother v1 as it's
+        * going to be deprecated anyway.
+        */
+       if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
+               btrfs_warn(fs_info,
+       "v1 space cache is not supported for page size %lu with sectorsize %u",
+                          PAGE_SIZE, fs_info->sectorsize);
+               return -EINVAL;
+       }
+
+       /* This can be called by remount, we need to protect the super block. */
+       spin_lock(&fs_info->super_lock);
+       btrfs_set_super_incompat_flags(disk_super, incompat);
+       spin_unlock(&fs_info->super_lock);
+
+       return 0;
+}
+
 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
                      char *options)
 {
@@ -3442,72 +3548,12 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
                goto fail_alloc;
        }
 
-       features = btrfs_super_incompat_flags(disk_super) &
-               ~BTRFS_FEATURE_INCOMPAT_SUPP;
-       if (features) {
-               btrfs_err(fs_info,
-                   "cannot mount because of unsupported optional features (0x%llx)",
-                   features);
-               err = -EINVAL;
-               goto fail_alloc;
-       }
-
-       features = btrfs_super_incompat_flags(disk_super);
-       features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
-       if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
-               features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
-       else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
-               features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
-
-       /*
-        * Flag our filesystem as having big metadata blocks if they are bigger
-        * than the page size.
-        */
-       if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
-               features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
-
-       /*
-        * mixed block groups end up with duplicate but slightly offset
-        * extent buffers for the same range.  It leads to corruptions
-        */
-       if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
-           (sectorsize != nodesize)) {
-               btrfs_err(fs_info,
-"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
-                       nodesize, sectorsize);
-               goto fail_alloc;
-       }
-
-       /*
-        * Needn't use the lock because there is no other task which will
-        * update the flag.
-        */
-       btrfs_set_super_incompat_flags(disk_super, features);
-
-       features = btrfs_super_compat_ro_flags(disk_super) &
-               ~BTRFS_FEATURE_COMPAT_RO_SUPP;
-       if (!sb_rdonly(sb) && features) {
-               btrfs_err(fs_info,
-       "cannot mount read-write because of unsupported optional features (0x%llx)",
-                      features);
-               err = -EINVAL;
-               goto fail_alloc;
-       }
-       /*
-        * We have unsupported RO compat features, although RO mounted, we
-        * should not cause any metadata write, including log replay.
-        * Or we could screw up whatever the new feature requires.
-        */
-       if (unlikely(features && btrfs_super_log_root(disk_super) &&
-                    !btrfs_test_opt(fs_info, NOLOGREPLAY))) {
-               btrfs_err(fs_info,
-"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
-                         features);
-               err = -EINVAL;
+       ret = btrfs_check_features(fs_info, sb);
+       if (ret < 0) {
+               err = ret;
                goto fail_alloc;
        }
 
-
        if (sectorsize < PAGE_SIZE) {
                struct btrfs_subpage_info *subpage_info;
 
index 7e545ec09a10b759ad5f446a747de9594dacf069..c67c15d4d20be0a5f52843821a138bf58ad01f2a 100644 (file)
@@ -48,6 +48,7 @@ int __cold open_ctree(struct super_block *sb,
 void __cold close_ctree(struct btrfs_fs_info *fs_info);
 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
                         struct btrfs_super_block *sb, int mirror_num);
+int btrfs_check_features(struct btrfs_fs_info *fs_info, struct super_block *sb);
 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors);
 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev);
 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
index be7df8d1d5b8fd1b465549fb275c4741fb793ac7..9be4fd2db0f44e01b6e4a16157acc66b1871ad8c 100644 (file)
@@ -2014,14 +2014,10 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data)
        if (ret)
                goto restore;
 
-       /* V1 cache is not supported for subpage mount. */
-       if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
-               btrfs_warn(fs_info,
-       "v1 space cache is not supported for page size %lu with sectorsize %u",
-                          PAGE_SIZE, fs_info->sectorsize);
-               ret = -EINVAL;
+       ret = btrfs_check_features(fs_info, sb);
+       if (ret < 0)
                goto restore;
-       }
+
        btrfs_remount_begin(fs_info, old_opts, *flags);
        btrfs_resize_thread_pool(fs_info,
                fs_info->thread_pool_size, old_thread_pool_size);
@@ -2117,15 +2113,6 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data)
                        ret = -EINVAL;
                        goto restore;
                }
-               if (btrfs_super_compat_ro_flags(fs_info->super_copy) &
-                   ~BTRFS_FEATURE_COMPAT_RO_SUPP) {
-                       btrfs_err(fs_info,
-               "can not remount read-write due to unsupported optional flags 0x%llx",
-                               btrfs_super_compat_ro_flags(fs_info->super_copy) &
-                               ~BTRFS_FEATURE_COMPAT_RO_SUPP);
-                       ret = -EINVAL;
-                       goto restore;
-               }
                if (fs_info->fs_devices->rw_devices == 0) {
                        ret = -EACCES;
                        goto restore;