ext4: check journal inode extents more carefully
authorJan Kara <jack@suse.cz>
Tue, 28 Jul 2020 13:04:34 +0000 (15:04 +0200)
committerTheodore Ts'o <tytso@mit.edu>
Fri, 7 Aug 2020 18:12:36 +0000 (14:12 -0400)
Currently, system zones just track ranges of block, that are "important"
fs metadata (bitmaps, group descriptors, journal blocks, etc.). This
however complicates how extent tree (or indirect blocks) can be checked
for inodes that actually track such metadata - currently the journal
inode but arguably we should be treating quota files or resize inode
similarly. We cannot run __ext4_ext_check() on such metadata inodes when
loading their extents as that would immediately trigger the validity
checks and so we just hack around that and special-case the journal
inode. This however leads to a situation that a journal inode which has
extent tree of depth at least one can have invalid extent tree that gets
unnoticed until ext4_cache_extents() crashes.

To overcome this limitation, track inode number each system zone belongs
to (0 is used for zones not belonging to any inode). We can then verify
inode number matches the expected one when verifying extent tree and
thus avoid the false errors. With this there's no need to to
special-case journal inode during extent tree checking anymore so remove
it.

Fixes: 0a944e8a6c66 ("ext4: don't perform block validity checks on the journal inode")
Reported-by: Wolfgang Frisch <wolfgang.frisch@suse.com>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20200728130437.7804-4-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
fs/ext4/block_validity.c
fs/ext4/ext4.h
fs/ext4/extents.c
fs/ext4/indirect.c
fs/ext4/inode.c
fs/ext4/mballoc.c

index b394a50ebbe308d2ff21e162aa52820fdba46098..e830a9d4e10d35c0a1b9e35b9046686b1db2b5ea 100644 (file)
@@ -24,6 +24,7 @@ struct ext4_system_zone {
        struct rb_node  node;
        ext4_fsblk_t    start_blk;
        unsigned int    count;
+       u32             ino;
 };
 
 static struct kmem_cache *ext4_system_zone_cachep;
@@ -45,7 +46,8 @@ void ext4_exit_system_zone(void)
 static inline int can_merge(struct ext4_system_zone *entry1,
                     struct ext4_system_zone *entry2)
 {
-       if ((entry1->start_blk + entry1->count) == entry2->start_blk)
+       if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
+           entry1->ino == entry2->ino)
                return 1;
        return 0;
 }
@@ -66,7 +68,7 @@ static void release_system_zone(struct ext4_system_blocks *system_blks)
  */
 static int add_system_zone(struct ext4_system_blocks *system_blks,
                           ext4_fsblk_t start_blk,
-                          unsigned int count)
+                          unsigned int count, u32 ino)
 {
        struct ext4_system_zone *new_entry, *entry;
        struct rb_node **n = &system_blks->root.rb_node, *node;
@@ -89,6 +91,7 @@ static int add_system_zone(struct ext4_system_blocks *system_blks,
                return -ENOMEM;
        new_entry->start_blk = start_blk;
        new_entry->count = count;
+       new_entry->ino = ino;
        new_node = &new_entry->node;
 
        rb_link_node(new_node, parent, n);
@@ -149,7 +152,7 @@ static void debug_print_tree(struct ext4_sb_info *sbi)
 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
                                     struct ext4_system_blocks *system_blks,
                                     ext4_fsblk_t start_blk,
-                                    unsigned int count)
+                                    unsigned int count, ino_t ino)
 {
        struct ext4_system_zone *entry;
        struct rb_node *n;
@@ -170,7 +173,7 @@ static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
                else if (start_blk >= (entry->start_blk + entry->count))
                        n = n->rb_right;
                else
-                       return 0;
+                       return entry->ino == ino;
        }
        return 1;
 }
@@ -204,19 +207,18 @@ static int ext4_protect_reserved_inode(struct super_block *sb,
                if (n == 0) {
                        i++;
                } else {
-                       if (!ext4_data_block_valid_rcu(sbi, system_blks,
-                                               map.m_pblk, n)) {
-                               err = -EFSCORRUPTED;
-                               __ext4_error(sb, __func__, __LINE__, -err,
-                                            map.m_pblk, "blocks %llu-%llu "
-                                            "from inode %u overlap system zone",
-                                            map.m_pblk,
-                                            map.m_pblk + map.m_len - 1, ino);
+                       err = add_system_zone(system_blks, map.m_pblk, n, ino);
+                       if (err < 0) {
+                               if (err == -EFSCORRUPTED) {
+                                       __ext4_error(sb, __func__, __LINE__,
+                                                    -err, map.m_pblk,
+                                                    "blocks %llu-%llu from inode %u overlap system zone",
+                                                    map.m_pblk,
+                                                    map.m_pblk + map.m_len - 1,
+                                                    ino);
+                               }
                                break;
                        }
-                       err = add_system_zone(system_blks, map.m_pblk, n);
-                       if (err < 0)
-                               break;
                        i += n;
                }
        }
@@ -270,19 +272,19 @@ int ext4_setup_system_zone(struct super_block *sb)
                    ((i < 5) || ((i % flex_size) == 0)))
                        add_system_zone(system_blks,
                                        ext4_group_first_block_no(sb, i),
-                                       ext4_bg_num_gdb(sb, i) + 1);
+                                       ext4_bg_num_gdb(sb, i) + 1, 0);
                gdp = ext4_get_group_desc(sb, i, NULL);
                ret = add_system_zone(system_blks,
-                               ext4_block_bitmap(sb, gdp), 1);
+                               ext4_block_bitmap(sb, gdp), 1, 0);
                if (ret)
                        goto err;
                ret = add_system_zone(system_blks,
-                               ext4_inode_bitmap(sb, gdp), 1);
+                               ext4_inode_bitmap(sb, gdp), 1, 0);
                if (ret)
                        goto err;
                ret = add_system_zone(system_blks,
                                ext4_inode_table(sb, gdp),
-                               sbi->s_itb_per_group);
+                               sbi->s_itb_per_group, 0);
                if (ret)
                        goto err;
        }
@@ -331,7 +333,7 @@ void ext4_release_system_zone(struct super_block *sb)
                call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
 }
 
-int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
+int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
                          unsigned int count)
 {
        struct ext4_system_blocks *system_blks;
@@ -343,9 +345,9 @@ int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
         * mount option.
         */
        rcu_read_lock();
-       system_blks = rcu_dereference(sbi->system_blks);
-       ret = ext4_data_block_valid_rcu(sbi, system_blks, start_blk,
-                                       count);
+       system_blks = rcu_dereference(EXT4_SB(inode->i_sb)->system_blks);
+       ret = ext4_data_block_valid_rcu(EXT4_SB(inode->i_sb), system_blks,
+                                       start_blk, count, inode->i_ino);
        rcu_read_unlock();
        return ret;
 }
@@ -364,8 +366,7 @@ int ext4_check_blockref(const char *function, unsigned int line,
        while (bref < p+max) {
                blk = le32_to_cpu(*bref++);
                if (blk &&
-                   unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
-                                                   blk, 1))) {
+                   unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
                        ext4_error_inode(inode, function, line, blk,
                                         "invalid block");
                        return -EFSCORRUPTED;
index c73b9a1959a9832484c17af11ce039f3c2e574ee..3994f012a9de72b46b901045980ffe8913471d52 100644 (file)
@@ -3397,9 +3397,9 @@ extern void ext4_release_system_zone(struct super_block *sb);
 extern int ext4_setup_system_zone(struct super_block *sb);
 extern int __init ext4_init_system_zone(void);
 extern void ext4_exit_system_zone(void);
-extern int ext4_data_block_valid(struct ext4_sb_info *sbi,
-                                ext4_fsblk_t start_blk,
-                                unsigned int count);
+extern int ext4_inode_block_valid(struct inode *inode,
+                                 ext4_fsblk_t start_blk,
+                                 unsigned int count);
 extern int ext4_check_blockref(const char *, unsigned int,
                               struct inode *, __le32 *, unsigned int);
 
index 4af3f36c8351d39efc699a978b7c031362ada4fc..0eea09aa0f267a3858062d5b19fda728473aa06a 100644 (file)
@@ -340,7 +340,7 @@ static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
         */
        if (lblock + len <= lblock)
                return 0;
-       return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
+       return ext4_inode_block_valid(inode, block, len);
 }
 
 static int ext4_valid_extent_idx(struct inode *inode,
@@ -348,7 +348,7 @@ static int ext4_valid_extent_idx(struct inode *inode,
 {
        ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 
-       return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
+       return ext4_inode_block_valid(inode, block, 1);
 }
 
 static int ext4_valid_extent_entries(struct inode *inode,
@@ -507,14 +507,10 @@ __read_extent_tree_block(const char *function, unsigned int line,
        }
        if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
                return bh;
-       if (!ext4_has_feature_journal(inode->i_sb) ||
-           (inode->i_ino !=
-            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
-               err = __ext4_ext_check(function, line, inode,
-                                      ext_block_hdr(bh), depth, pblk);
-               if (err)
-                       goto errout;
-       }
+       err = __ext4_ext_check(function, line, inode,
+                              ext_block_hdr(bh), depth, pblk);
+       if (err)
+               goto errout;
        set_buffer_verified(bh);
        /*
         * If this is a leaf block, cache all of its entries
index be2b66eb65f7a398329e292454de667befb417f1..4026418257121d5f576f29115d3879bbef3c8bc0 100644 (file)
@@ -858,8 +858,7 @@ static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
        else if (ext4_should_journal_data(inode))
                flags |= EXT4_FREE_BLOCKS_FORGET;
 
-       if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
-                                  count)) {
+       if (!ext4_inode_block_valid(inode, block_to_free, count)) {
                EXT4_ERROR_INODE(inode, "attempt to clear invalid "
                                 "blocks %llu len %lu",
                                 (unsigned long long) block_to_free, count);
@@ -1004,8 +1003,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
                        if (!nr)
                                continue;               /* A hole */
 
-                       if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
-                                                  nr, 1)) {
+                       if (!ext4_inode_block_valid(inode, nr, 1)) {
                                EXT4_ERROR_INODE(inode,
                                                 "invalid indirect mapped "
                                                 "block %lu (level %d)",
index 551a1056870d76f25281f63febec3a1cdda31e6c..0b07576af3bf87d47fef28b69205a45c0b4dd75f 100644 (file)
@@ -394,8 +394,7 @@ static int __check_block_validity(struct inode *inode, const char *func,
            (inode->i_ino ==
             le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
                return 0;
-       if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
-                                  map->m_len)) {
+       if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
                ext4_error_inode(inode, func, line, map->m_pblk,
                                 "lblock %lu mapped to illegal pblock %llu "
                                 "(length %d)", (unsigned long) map->m_lblk,
@@ -4761,7 +4760,7 @@ struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
 
        ret = 0;
        if (ei->i_file_acl &&
-           !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
+           !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
                ext4_error_inode(inode, function, line, 0,
                                 "iget: bad extended attribute block %llu",
                                 ei->i_file_acl);
index 19101941daddd15d5cfc76a6d51bf56537e97954..8b8c84db243a976852fd8fe9f4678b9c62fb4e03 100644 (file)
@@ -3233,7 +3233,7 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
        block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
 
        len = EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
-       if (!ext4_data_block_valid(sbi, block, len)) {
+       if (!ext4_inode_block_valid(ac->ac_inode, block, len)) {
                ext4_error(sb, "Allocating blocks %llu-%llu which overlap "
                           "fs metadata", block, block+len);
                /* File system mounted not to panic on error
@@ -5058,7 +5058,7 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
 
        sbi = EXT4_SB(sb);
        if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
-           !ext4_data_block_valid(sbi, block, count)) {
+           !ext4_inode_block_valid(inode, block, count)) {
                ext4_error(sb, "Freeing blocks not in datazone - "
                           "block = %llu, count = %lu", block, count);
                goto error_return;