Whamcloud - gitweb
e2fsck: read-ahead metadata during passes 1, 2, and 4
[tools/e2fsprogs.git] / e2fsck / pass1.c
index 8279ce5..b13be54 100644 (file)
@@ -68,6 +68,7 @@ static void mark_table_blocks(e2fsck_t ctx);
 static void alloc_bb_map(e2fsck_t ctx);
 static void alloc_imagic_map(e2fsck_t ctx);
 static void mark_inode_bad(e2fsck_t ctx, ino_t ino);
+static void add_encrypted_dir(e2fsck_t ctx, ino_t ino);
 static void handle_fs_bad_blocks(e2fsck_t ctx);
 static void process_inodes(e2fsck_t ctx, char *block_buf);
 static EXT2_QSORT_TYPE process_inode_cmp(const void *a, const void *b);
@@ -80,7 +81,8 @@ static void adjust_extattr_refcount(e2fsck_t ctx, ext2_refcount_t refcount,
 struct process_block_struct {
        ext2_ino_t      ino;
        unsigned        is_dir:1, is_reg:1, clear:1, suppress:1,
-                               fragmented:1, compressed:1, bbcheck:1;
+                               fragmented:1, compressed:1, bbcheck:1,
+                               inode_modified:1;
        blk64_t         num_blocks;
        blk64_t         max_blocks;
        e2_blkcnt_t     last_block;
@@ -92,6 +94,7 @@ struct process_block_struct {
        struct problem_context *pctx;
        ext2fs_block_bitmap fs_meta_blocks;
        e2fsck_t        ctx;
+       region_t        region;
 };
 
 struct process_inode_block {
@@ -175,12 +178,15 @@ int e2fsck_pass1_check_symlink(ext2_filsys fs, ext2_ino_t ino,
        ext2_extent_handle_t    handle;
        struct ext2_extent_info info;
        struct ext2fs_extent    extent;
+       int encrypted = 0;
 
        if ((inode->i_size_high || inode->i_size == 0) ||
            (inode->i_flags & EXT2_INDEX_FL))
                return 0;
 
        if (inode->i_flags & EXT4_EXTENTS_FL) {
+               if (inode->i_flags & EXT4_INLINE_DATA_FL)
+                       return 0;
                if (inode->i_size > fs->blocksize)
                        return 0;
                if (ext2fs_extent_open2(fs, ino, inode, &handle))
@@ -202,8 +208,21 @@ int e2fsck_pass1_check_symlink(ext2_filsys fs, ext2_ino_t ino,
                return i;
        }
 
+       if (inode->i_flags & EXT4_INLINE_DATA_FL) {
+               size_t inline_size;
+
+               if (ext2fs_inline_data_size(fs, ino, &inline_size))
+                       return 0;
+               if (inode->i_size != inline_size)
+                       return 0;
+
+               return 1;
+       }
+
        blocks = ext2fs_inode_data_blocks2(fs, inode);
        if (blocks) {
+               if (inode->i_flags & EXT4_INLINE_DATA_FL)
+                       return 0;
                if ((inode->i_size >= fs->blocksize) ||
                    (blocks != fs->blocksize >> 9) ||
                    (inode->i_block[0] < fs->super->s_first_data_block) ||
@@ -217,9 +236,34 @@ int e2fsck_pass1_check_symlink(ext2_filsys fs, ext2_ino_t ino,
                if (io_channel_read_blk64(fs->io, inode->i_block[0], 1, buf))
                        return 0;
 
-               len = strnlen(buf, fs->blocksize);
+               if (inode->i_flags & EXT4_ENCRYPT_FL) {
+                       len = ext2fs_le32_to_cpu(*((__u32 *)buf)) + 4;
+               } else {
+                       len = strnlen(buf, fs->blocksize);
+               }
                if (len == fs->blocksize)
                        return 0;
+       } else if (inode->i_flags & EXT4_INLINE_DATA_FL) {
+               char *inline_buf = NULL;
+               size_t inline_sz = 0;
+
+               if (ext2fs_inline_data_size(fs, ino, &inline_sz))
+                       return 0;
+               if (inode->i_size != inline_sz)
+                       return 0;
+               if (ext2fs_get_mem(inline_sz + 1, &inline_buf))
+                       return 0;
+               i = 0;
+               if (ext2fs_inline_data_get(fs, ino, inode, inline_buf, NULL))
+                       goto exit_inline;
+               inline_buf[inline_sz] = 0;
+               len = strnlen(inline_buf, inline_sz);
+               if (len != inline_sz)
+                       goto exit_inline;
+               i = 1;
+exit_inline:
+               ext2fs_free_mem(&inline_buf);
+               return i;
        } else {
                if (inode->i_size >= sizeof(inode->i_block))
                        return 0;
@@ -229,11 +273,30 @@ int e2fsck_pass1_check_symlink(ext2_filsys fs, ext2_ino_t ino,
                        return 0;
        }
        if (len != inode->i_size)
-               return 0;
+               if ((inode->i_flags & EXT4_ENCRYPT_FL) == 0)
+                       return 0;
        return 1;
 }
 
 /*
+ * If the extents or inlinedata flags are set on the inode, offer to clear 'em.
+ */
+#define BAD_SPECIAL_FLAGS (EXT4_EXTENTS_FL | EXT4_INLINE_DATA_FL)
+static void check_extents_inlinedata(e2fsck_t ctx,
+                                    struct problem_context *pctx)
+{
+       if (!(pctx->inode->i_flags & BAD_SPECIAL_FLAGS))
+               return;
+
+       if (!fix_problem(ctx, PR_1_SPECIAL_EXTENTS_IDATA, pctx))
+               return;
+
+       pctx->inode->i_flags &= ~BAD_SPECIAL_FLAGS;
+       e2fsck_write_inode(ctx, pctx->ino, pctx->inode, "pass1");
+}
+#undef BAD_SPECIAL_FLAGS
+
+/*
  * If the immutable (or append-only) flag is set on the inode, offer
  * to clear it.
  */
@@ -264,8 +327,7 @@ static void check_size(e2fsck_t ctx, struct problem_context *pctx)
        if (!fix_problem(ctx, PR_1_SET_NONZSIZE, pctx))
                return;
 
-       inode->i_size = 0;
-       inode->i_size_high = 0;
+       ext2fs_inode_size_set(ctx->fs, inode, 0);
        e2fsck_write_inode(ctx, pctx->ino, pctx->inode, "pass1");
 }
 
@@ -274,15 +336,17 @@ static void check_ea_in_inode(e2fsck_t ctx, struct problem_context *pctx)
        struct ext2_super_block *sb = ctx->fs->super;
        struct ext2_inode_large *inode;
        struct ext2_ext_attr_entry *entry;
-       char *start;
+       char *start, *header;
        unsigned int storage_size, remain;
-       int problem = 0;
+       problem_t problem = 0;
+       region_t region = 0;
 
        inode = (struct ext2_inode_large *) pctx->inode;
        storage_size = EXT2_INODE_SIZE(ctx->fs->super) - EXT2_GOOD_OLD_INODE_SIZE -
                inode->i_extra_isize;
-       start = ((char *) inode) + EXT2_GOOD_OLD_INODE_SIZE +
-               inode->i_extra_isize + sizeof(__u32);
+       header = ((char *) inode) + EXT2_GOOD_OLD_INODE_SIZE +
+                inode->i_extra_isize;
+       start = header + sizeof(__u32);
        entry = (struct ext2_ext_attr_entry *) start;
 
        /* scan all entry's headers first */
@@ -290,9 +354,28 @@ static void check_ea_in_inode(e2fsck_t ctx, struct problem_context *pctx)
        /* take finish entry 0UL into account */
        remain = storage_size - sizeof(__u32);
 
-       while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
+       region = region_create(0, storage_size);
+       if (!region) {
+               fix_problem(ctx, PR_1_EA_ALLOC_REGION_ABORT, pctx);
+               problem = 0;
+               ctx->flags |= E2F_FLAG_ABORT;
+               return;
+       }
+       if (region_allocate(region, 0, sizeof(__u32))) {
+               problem = PR_1_INODE_EA_ALLOC_COLLISION;
+               goto fix;
+       }
+
+       while (remain >= sizeof(struct ext2_ext_attr_entry) &&
+              !EXT2_EXT_IS_LAST_ENTRY(entry)) {
                __u32 hash;
 
+               if (region_allocate(region, (char *)entry - (char *)header,
+                                   EXT2_EXT_ATTR_LEN(entry->e_name_len))) {
+                       problem = PR_1_INODE_EA_ALLOC_COLLISION;
+                       goto fix;
+               }
+
                /* header eats this space */
                remain -= sizeof(struct ext2_ext_attr_entry);
 
@@ -320,6 +403,13 @@ static void check_ea_in_inode(e2fsck_t ctx, struct problem_context *pctx)
                        goto fix;
                }
 
+               if (entry->e_value_size &&
+                   region_allocate(region, sizeof(__u32) + entry->e_value_offs,
+                                   EXT2_EXT_ATTR_SIZE(entry->e_value_size))) {
+                       problem = PR_1_INODE_EA_ALLOC_COLLISION;
+                       goto fix;
+               }
+
                hash = ext2fs_ext_attr_hash_entry(entry,
                                                  start + entry->e_value_offs);
 
@@ -334,7 +424,15 @@ static void check_ea_in_inode(e2fsck_t ctx, struct problem_context *pctx)
 
                entry = EXT2_EXT_ATTR_NEXT(entry);
        }
+
+       if (region_allocate(region, (char *)entry - (char *)header,
+                           sizeof(__u32))) {
+               problem = PR_1_INODE_EA_ALLOC_COLLISION;
+               goto fix;
+       }
 fix:
+       if (region)
+               region_free(region);
        /*
         * it seems like a corruption. it's very unlikely we could repair
         * EA(s) in automatic fashion -bzzz
@@ -343,7 +441,7 @@ fix:
                return;
 
        /* simply remove all possible EA(s) */
-       *((__u32 *)start) = 0UL;
+       *((__u32 *)header) = 0UL;
        e2fsck_write_inode_full(ctx, pctx->ino, pctx->inode,
                                EXT2_INODE_SIZE(sb), "pass1");
 }
@@ -407,6 +505,7 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
        blk64_t                 blk;
        unsigned int            i, rec_len, not_device = 0;
        int                     extent_fs;
+       int                     inlinedata_fs;
 
        /*
         * If the mode looks OK, we believe it.  If the first block in
@@ -434,11 +533,52 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
         * For extent mapped files, we don't do any sanity checking:
         * just try to get the phys block of logical block 0 and run
         * with it.
+        *
+        * For inline data files, we just try to get the size of inline
+        * data.  If it's true, we will treat it as a directory.
         */
 
        extent_fs = (ctx->fs->super->s_feature_incompat &
                     EXT3_FEATURE_INCOMPAT_EXTENTS);
-       if (extent_fs && (inode->i_flags & EXT4_EXTENTS_FL)) {
+       inlinedata_fs = (ctx->fs->super->s_feature_incompat &
+                        EXT4_FEATURE_INCOMPAT_INLINE_DATA);
+       if (inlinedata_fs && (inode->i_flags & EXT4_INLINE_DATA_FL)) {
+               size_t size;
+               __u32 dotdot;
+               unsigned int rec_len;
+               struct ext2_dir_entry de;
+
+               if (ext2fs_inline_data_size(ctx->fs, pctx->ino, &size))
+                       return;
+               /*
+                * If the size isn't a multiple of 4, it's probably not a
+                * directory??
+                */
+               if (size & 3)
+                       return;
+               /*
+                * If the first 10 bytes don't look like a directory entry,
+                * it's probably not a directory.
+                */
+               memcpy(&dotdot, inode->i_block, sizeof(dotdot));
+               memcpy(&de, ((char *)inode->i_block) + EXT4_INLINE_DATA_DOTDOT_SIZE,
+                      EXT2_DIR_REC_LEN(0));
+               dotdot = ext2fs_le32_to_cpu(dotdot);
+               de.inode = ext2fs_le32_to_cpu(de.inode);
+               de.rec_len = ext2fs_le16_to_cpu(de.rec_len);
+               ext2fs_get_rec_len(ctx->fs, &de, &rec_len);
+               if (dotdot >= ctx->fs->super->s_inodes_count ||
+                   (dotdot < EXT2_FIRST_INO(ctx->fs->super) &&
+                    dotdot != EXT2_ROOT_INO) ||
+                   de.inode >= ctx->fs->super->s_inodes_count ||
+                   (de.inode < EXT2_FIRST_INO(ctx->fs->super) &&
+                    de.inode != 0) ||
+                   rec_len > EXT4_MIN_INLINE_DATA_SIZE -
+                             EXT4_INLINE_DATA_DOTDOT_SIZE)
+                       return;
+               /* device files never have a "system.data" entry */
+               goto isdir;
+       } else if (extent_fs && (inode->i_flags & EXT4_EXTENTS_FL)) {
                /* extent mapped */
                if  (ext2fs_bmap2(ctx->fs, pctx->ino, inode, 0, 0, 0, 0,
                                 &blk))
@@ -473,7 +613,7 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
 
        /* read the first block */
        ehandler_operation(_("reading directory block"));
-       retval = ext2fs_read_dir_block3(ctx->fs, blk, buf, 0);
+       retval = ext2fs_read_dir_block4(ctx->fs, blk, buf, 0, pctx->ino);
        ehandler_operation(0);
        if (retval)
                return;
@@ -482,7 +622,7 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
        retval = ext2fs_get_rec_len(ctx->fs, dirent, &rec_len);
        if (retval)
                return;
-       if (((dirent->name_len & 0xFF) != 1) ||
+       if ((ext2fs_dirent_name_len(dirent) != 1) ||
            (dirent->name[0] != '.') ||
            (dirent->inode != pctx->ino) ||
            (rec_len < 12) ||
@@ -494,13 +634,14 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
        retval = ext2fs_get_rec_len(ctx->fs, dirent, &rec_len);
        if (retval)
                return;
-       if (((dirent->name_len & 0xFF) != 2) ||
+       if ((ext2fs_dirent_name_len(dirent) != 2) ||
            (dirent->name[0] != '.') ||
            (dirent->name[1] != '.') ||
            (rec_len < 12) ||
            (rec_len % 4))
                return;
 
+isdir:
        if (fix_problem(ctx, PR_1_TREAT_AS_DIRECTORY, pctx)) {
                inode->i_mode = (inode->i_mode & 07777) | LINUX_S_IFDIR;
                e2fsck_write_inode_full(ctx, pctx->ino, inode,
@@ -509,8 +650,8 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
        }
 }
 
-extern void e2fsck_setup_tdb_icount(e2fsck_t ctx, int flags,
-                                   ext2_icount_t *ret)
+void e2fsck_setup_tdb_icount(e2fsck_t ctx, int flags,
+                            ext2_icount_t *ret)
 {
        unsigned int            threshold;
        ext2_ino_t              num_dirs;
@@ -540,15 +681,286 @@ extern void e2fsck_setup_tdb_icount(e2fsck_t ctx, int flags,
                *ret = 0;
 }
 
+static errcode_t recheck_bad_inode_checksum(ext2_filsys fs, ext2_ino_t ino,
+                                           e2fsck_t ctx,
+                                           struct problem_context *pctx)
+{
+       errcode_t retval;
+       struct ext2_inode_large inode;
+
+       /*
+        * Reread inode.  If we don't see checksum error, then this inode
+        * has been fixed elsewhere.
+        */
+       ctx->stashed_ino = 0;
+       retval = ext2fs_read_inode_full(fs, ino, (struct ext2_inode *)&inode,
+                                       sizeof(inode));
+       if (retval && retval != EXT2_ET_INODE_CSUM_INVALID)
+               return retval;
+       if (!retval)
+               return 0;
+
+       /*
+        * Checksum still doesn't match.  That implies that the inode passes
+        * all the sanity checks, so maybe the checksum is simply corrupt.
+        * See if the user will go for fixing that.
+        */
+       if (!fix_problem(ctx, PR_1_INODE_ONLY_CSUM_INVALID, pctx))
+               return 0;
+
+       retval = ext2fs_write_inode_full(fs, ino, (struct ext2_inode *)&inode,
+                                        sizeof(inode));
+       return retval;
+}
+
+static void reserve_block_for_root_repair(e2fsck_t ctx)
+{
+       blk64_t         blk = 0;
+       errcode_t       err;
+       ext2_filsys     fs = ctx->fs;
+
+       ctx->root_repair_block = 0;
+       if (ext2fs_test_inode_bitmap2(ctx->inode_used_map, EXT2_ROOT_INO))
+               return;
+
+       err = ext2fs_new_block2(fs, 0, ctx->block_found_map, &blk);
+       if (err)
+               return;
+       ext2fs_mark_block_bitmap2(ctx->block_found_map, blk);
+       ctx->root_repair_block = blk;
+}
+
+static void reserve_block_for_lnf_repair(e2fsck_t ctx)
+{
+       blk64_t         blk = 0;
+       errcode_t       err;
+       ext2_filsys     fs = ctx->fs;
+       static const char name[] = "lost+found";
+       ext2_ino_t      ino;
+
+       ctx->lnf_repair_block = 0;
+       if (!ext2fs_lookup(fs, EXT2_ROOT_INO, name, sizeof(name)-1, 0, &ino))
+               return;
+
+       err = ext2fs_new_block2(fs, 0, ctx->block_found_map, &blk);
+       if (err)
+               return;
+       ext2fs_mark_block_bitmap2(ctx->block_found_map, blk);
+       ctx->lnf_repair_block = blk;
+}
+
+static errcode_t get_inline_data_ea_size(ext2_filsys fs, ext2_ino_t ino,
+                                        size_t *sz)
+{
+       void *p;
+       struct ext2_xattr_handle *handle;
+       errcode_t retval;
+
+       retval = ext2fs_xattrs_open(fs, ino, &handle);
+       if (retval)
+               return retval;
+
+       retval = ext2fs_xattrs_read(handle);
+       if (retval)
+               goto err;
+
+       retval = ext2fs_xattr_get(handle, "system.data", &p, sz);
+       if (retval)
+               goto err;
+       ext2fs_free_mem(&p);
+err:
+       (void) ext2fs_xattrs_close(&handle);
+       return retval;
+}
+
+static void finish_processing_inode(e2fsck_t ctx, ext2_ino_t ino,
+                                   struct problem_context *pctx,
+                                   int failed_csum)
+{
+       if (!failed_csum)
+               return;
+
+       /*
+        * If the inode failed the checksum and the user didn't
+        * clear the inode, test the checksum again -- if it still
+        * fails, ask the user if the checksum should be corrected.
+        */
+       pctx->errcode = recheck_bad_inode_checksum(ctx->fs, ino, ctx, pctx);
+       if (pctx->errcode)
+               ctx->flags |= E2F_FLAG_ABORT;
+}
+#define FINISH_INODE_LOOP(ctx, ino, pctx, failed_csum) \
+       do { \
+               finish_processing_inode((ctx), (ino), (pctx), (failed_csum)); \
+               if ((ctx)->flags & E2F_FLAG_ABORT) \
+                       return; \
+       } while (0)
+
+static int could_be_block_map(ext2_filsys fs, struct ext2_inode *inode)
+{
+       __u32 x;
+       int i;
+
+       for (i = 0; i < EXT2_N_BLOCKS; i++) {
+               x = inode->i_block[i];
+#ifdef WORDS_BIGENDIAN
+               x = ext2fs_swab32(x);
+#endif
+               if (x >= ext2fs_blocks_count(fs->super))
+                       return 0;
+       }
+
+       return 1;
+}
+
+/*
+ * Figure out what to do with an inode that has both extents and inline data
+ * inode flags set.  Returns -1 if we decide to erase the inode, 0 otherwise.
+ */
+static int fix_inline_data_extents_file(e2fsck_t ctx,
+                                       ext2_ino_t ino,
+                                       struct ext2_inode *inode,
+                                       int inode_size,
+                                       struct problem_context *pctx)
+{
+       size_t max_inline_ea_size;
+       ext2_filsys fs = ctx->fs;
+       int dirty = 0;
+
+       /* Both feature flags not set?  Just run the regular checks */
+       if (!EXT2_HAS_INCOMPAT_FEATURE(fs->super,
+                                      EXT3_FEATURE_INCOMPAT_EXTENTS) &&
+           !EXT2_HAS_INCOMPAT_FEATURE(fs->super,
+                                      EXT4_FEATURE_INCOMPAT_INLINE_DATA))
+               return 0;
+
+       /* Clear both flags if it's a special file */
+       if (LINUX_S_ISCHR(inode->i_mode) ||
+           LINUX_S_ISBLK(inode->i_mode) ||
+           LINUX_S_ISFIFO(inode->i_mode) ||
+           LINUX_S_ISSOCK(inode->i_mode)) {
+               check_extents_inlinedata(ctx, pctx);
+               return 0;
+       }
+
+       /* If it looks like an extent tree, try to clear inlinedata */
+       if (ext2fs_extent_header_verify(inode->i_block,
+                                sizeof(inode->i_block)) == 0 &&
+           fix_problem(ctx, PR_1_CLEAR_INLINE_DATA_FOR_EXTENT, pctx)) {
+               inode->i_flags &= ~EXT4_INLINE_DATA_FL;
+               dirty = 1;
+               goto out;
+       }
+
+       /* If it looks short enough to be inline data, try to clear extents */
+       if (EXT2_INODE_SIZE(fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
+               max_inline_ea_size = EXT2_INODE_SIZE(fs->super) -
+                                    (EXT2_GOOD_OLD_INODE_SIZE +
+                                     ((struct ext2_inode_large *)inode)->i_extra_isize);
+       else
+               max_inline_ea_size = 0;
+       if (EXT2_I_SIZE(inode) <
+           EXT4_MIN_INLINE_DATA_SIZE + max_inline_ea_size &&
+           fix_problem(ctx, PR_1_CLEAR_EXTENT_FOR_INLINE_DATA, pctx)) {
+               inode->i_flags &= ~EXT4_EXTENTS_FL;
+               dirty = 1;
+               goto out;
+       }
+
+       /*
+        * Too big for inline data, but no evidence of extent tree -
+        * maybe it's a block map file?  If the mappings all look valid?
+        */
+       if (could_be_block_map(fs, inode) &&
+           fix_problem(ctx, PR_1_CLEAR_EXTENT_INLINE_DATA_FLAGS, pctx)) {
+#ifdef WORDS_BIGENDIAN
+               int i;
+
+               for (i = 0; i < EXT2_N_BLOCKS; i++)
+                       inode->i_block[i] = ext2fs_swab32(inode->i_block[i]);
+#endif
+
+               inode->i_flags &= ~(EXT4_EXTENTS_FL | EXT4_INLINE_DATA_FL);
+               dirty = 1;
+               goto out;
+       }
+
+       /* Oh well, just clear the busted inode. */
+       if (fix_problem(ctx, PR_1_CLEAR_EXTENT_INLINE_DATA_INODE, pctx)) {
+               e2fsck_clear_inode(ctx, ino, inode, 0, "pass1");
+               return -1;
+       }
+
+out:
+       if (dirty)
+               e2fsck_write_inode(ctx, ino, inode, "pass1");
+
+       return 0;
+}
+
+static void pass1_readahead(e2fsck_t ctx, dgrp_t *group, ext2_ino_t *next_ino)
+{
+       ext2_ino_t inodes_in_group = 0, inodes_per_block, inodes_per_buffer;
+       dgrp_t start = *group, grp;
+       blk64_t blocks_to_read = 0;
+       errcode_t err = EXT2_ET_INVALID_ARGUMENT;
+
+       if (ctx->readahead_kb == 0)
+               goto out;
+
+       /* Keep iterating groups until we have enough to readahead */
+       inodes_per_block = EXT2_INODES_PER_BLOCK(ctx->fs->super);
+       for (grp = start; grp < ctx->fs->group_desc_count; grp++) {
+               if (ext2fs_bg_flags_test(ctx->fs, grp, EXT2_BG_INODE_UNINIT))
+                       continue;
+               inodes_in_group = ctx->fs->super->s_inodes_per_group -
+                                       ext2fs_bg_itable_unused(ctx->fs, grp);
+               blocks_to_read += (inodes_in_group + inodes_per_block - 1) /
+                                       inodes_per_block;
+               if (blocks_to_read * ctx->fs->blocksize >
+                   ctx->readahead_kb * 1024)
+                       break;
+       }
+
+       err = e2fsck_readahead(ctx->fs, E2FSCK_READA_ITABLE, start,
+                              grp - start + 1);
+       if (err == EAGAIN) {
+               ctx->readahead_kb /= 2;
+               err = 0;
+       }
+
+out:
+       if (err) {
+               /* Error; disable itable readahead */
+               *group = ctx->fs->group_desc_count;
+               *next_ino = ctx->fs->super->s_inodes_count;
+       } else {
+               /*
+                * Don't do more readahead until we've reached the first inode
+                * of the last inode scan buffer block for the last group.
+                */
+               *group = grp + 1;
+               inodes_per_buffer = (ctx->inode_buffer_blocks ?
+                                    ctx->inode_buffer_blocks :
+                                    EXT2_INODE_SCAN_DEFAULT_BUFFER_BLOCKS) *
+                                   ctx->fs->blocksize /
+                                   EXT2_INODE_SIZE(ctx->fs->super);
+               inodes_in_group--;
+               *next_ino = inodes_in_group -
+                           (inodes_in_group % inodes_per_buffer) + 1 +
+                           (grp * ctx->fs->super->s_inodes_per_group);
+       }
+}
+
 void e2fsck_pass1(e2fsck_t ctx)
 {
        int     i;
        __u64   max_sizes;
        ext2_filsys fs = ctx->fs;
        ext2_ino_t      ino = 0;
-       struct ext2_inode *inode;
-       ext2_inode_scan scan;
-       char            *block_buf;
+       struct ext2_inode *inode = NULL;
+       ext2_inode_scan scan = NULL;
+       char            *block_buf = NULL;
 #ifdef RESOURCE_TRACK
        struct resource_track   rtrack;
 #endif
@@ -558,13 +970,23 @@ void e2fsck_pass1(e2fsck_t ctx)
        struct ext2_super_block *sb = ctx->fs->super;
        const char      *old_op;
        unsigned int    save_type;
-       int             imagic_fs, extent_fs;
-       int             busted_fs_time = 0;
+       int             imagic_fs, extent_fs, inlinedata_fs;
+       int             low_dtime_check = 1;
        int             inode_size;
+       int             failed_csum = 0;
+       ext2_ino_t      ino_threshold = 0;
+       dgrp_t          ra_group = 0;
 
        init_resource_track(&rtrack, ctx->fs->io);
        clear_problem_context(&pctx);
 
+       /* If we can do readahead, figure out how many groups to pull in. */
+       if (!e2fsck_can_readahead(ctx->fs))
+               ctx->readahead_kb = 0;
+       else if (ctx->readahead_kb == ~0ULL)
+               ctx->readahead_kb = e2fsck_guess_readahead(ctx->fs);
+       pass1_readahead(ctx, &ra_group, &ino_threshold);
+
        if (!(ctx->options & E2F_OPT_PREEN))
                fix_problem(ctx, PR_1_PASS_HEADER, &pctx);
 
@@ -591,6 +1013,8 @@ void e2fsck_pass1(e2fsck_t ctx)
 
        imagic_fs = (sb->s_feature_compat & EXT2_FEATURE_COMPAT_IMAGIC_INODES);
        extent_fs = (sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS);
+       inlinedata_fs = (sb->s_feature_incompat &
+                       EXT4_FEATURE_INCOMPAT_INLINE_DATA);
 
        /*
         * Allocate bitmaps structures
@@ -633,6 +1057,15 @@ void e2fsck_pass1(e2fsck_t ctx)
                ctx->flags |= E2F_FLAG_ABORT;
                return;
        }
+       pctx.errcode = e2fsck_allocate_block_bitmap(fs,
+                       _("metadata block map"), EXT2FS_BMAP64_RBTREE,
+                       "block_metadata_map", &ctx->block_metadata_map);
+       if (pctx.errcode) {
+               pctx.num = 1;
+               fix_problem(ctx, PR_1_ALLOCATE_BBITMAP_ERROR, &pctx);
+               ctx->flags |= E2F_FLAG_ABORT;
+               return;
+       }
        e2fsck_setup_tdb_icount(ctx, 0, &ctx->inode_link_info);
        if (!ctx->inode_link_info) {
                e2fsck_set_bitmap_type(fs, EXT2FS_BMAP64_RBTREE,
@@ -662,8 +1095,7 @@ void e2fsck_pass1(e2fsck_t ctx)
        if (pctx.errcode) {
                fix_problem(ctx, PR_1_ALLOCATE_DBCOUNT, &pctx);
                ctx->flags |= E2F_FLAG_ABORT;
-               ext2fs_free_mem(&inode);
-               return;
+               goto endit;
        }
 
        /*
@@ -686,12 +1118,13 @@ void e2fsck_pass1(e2fsck_t ctx)
        if (pctx.errcode) {
                fix_problem(ctx, PR_1_CONVERT_SUBCLUSTER, &pctx);
                ctx->flags |= E2F_FLAG_ABORT;
-               ext2fs_free_mem(&inode);
-               return;
+               goto endit;
        }
        block_buf = (char *) e2fsck_allocate_memory(ctx, fs->blocksize * 3,
                                                    "block interate buffer");
-       e2fsck_use_inode_shortcuts(ctx, 1);
+       if (EXT2_INODE_SIZE(fs->super) == EXT2_GOOD_OLD_INODE_SIZE)
+               e2fsck_use_inode_shortcuts(ctx, 1);
+       e2fsck_intercept_block_allocations(ctx);
        old_op = ehandler_operation(_("opening inode scan"));
        pctx.errcode = ext2fs_open_inode_scan(fs, ctx->inode_buffer_blocks,
                                              &scan);
@@ -699,28 +1132,32 @@ void e2fsck_pass1(e2fsck_t ctx)
        if (pctx.errcode) {
                fix_problem(ctx, PR_1_ISCAN_ERROR, &pctx);
                ctx->flags |= E2F_FLAG_ABORT;
-               ext2fs_free_mem(&block_buf);
-               ext2fs_free_mem(&inode);
-               return;
+               goto endit;
        }
-       ext2fs_inode_scan_flags(scan, EXT2_SF_SKIP_MISSING_ITABLE, 0);
+       ext2fs_inode_scan_flags(scan, EXT2_SF_SKIP_MISSING_ITABLE |
+                                     EXT2_SF_WARN_GARBAGE_INODES, 0);
        ctx->stashed_inode = inode;
        scan_struct.ctx = ctx;
        scan_struct.block_buf = block_buf;
        ext2fs_set_inode_callback(scan, scan_callback, &scan_struct);
-       if (ctx->progress)
-               if ((ctx->progress)(ctx, 1, 0, ctx->fs->group_desc_count))
-                       return;
+       if (ctx->progress && ((ctx->progress)(ctx, 1, 0,
+                                             ctx->fs->group_desc_count)))
+               goto endit;
        if ((fs->super->s_wtime < fs->super->s_inodes_count) ||
-           (fs->super->s_mtime < fs->super->s_inodes_count))
-               busted_fs_time = 1;
+           (fs->super->s_mtime < fs->super->s_inodes_count) ||
+           (fs->super->s_mkfs_time &&
+            fs->super->s_mkfs_time < fs->super->s_inodes_count))
+               low_dtime_check = 0;
 
        if ((fs->super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_MMP) &&
-           !(fs->super->s_mmp_block <= fs->super->s_first_data_block ||
-             fs->super->s_mmp_block >= fs->super->s_blocks_count))
+           fs->super->s_mmp_block > fs->super->s_first_data_block &&
+           fs->super->s_mmp_block < ext2fs_blocks_count(fs->super))
                ext2fs_mark_block_bitmap2(ctx->block_found_map,
                                          fs->super->s_mmp_block);
 
+       /* Set up ctx->lost_and_found if possible */
+       (void) e2fsck_get_lost_and_found(ctx, 0);
+
        while (1) {
                if (ino % (fs->super->s_inodes_per_group * 4) == 1) {
                        if (e2fsck_mmp_update(fs))
@@ -729,26 +1166,71 @@ void e2fsck_pass1(e2fsck_t ctx)
                old_op = ehandler_operation(_("getting next inode from scan"));
                pctx.errcode = ext2fs_get_next_inode_full(scan, &ino,
                                                          inode, inode_size);
+               if (ino > ino_threshold)
+                       pass1_readahead(ctx, &ra_group, &ino_threshold);
                ehandler_operation(old_op);
                if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
                        return;
                if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE) {
+                       /*
+                        * If badblocks says badblocks is bad, offer to clear
+                        * the list, update the in-core bb list, and restart
+                        * the inode scan.
+                        */
+                       if (ino == EXT2_BAD_INO &&
+                           fix_problem(ctx, PR_1_BADBLOCKS_IN_BADBLOCKS,
+                                       &pctx)) {
+                               errcode_t err;
+
+                               e2fsck_clear_inode(ctx, ino, inode, 0, "pass1");
+                               ext2fs_badblocks_list_free(ctx->fs->badblocks);
+                               ctx->fs->badblocks = NULL;
+                               err = ext2fs_read_bb_inode(ctx->fs,
+                                                       &ctx->fs->badblocks);
+                               if (err) {
+                                       fix_problem(ctx, PR_1_ISCAN_ERROR,
+                                                   &pctx);
+                                       ctx->flags |= E2F_FLAG_ABORT;
+                                       goto endit;
+                               }
+                               err = ext2fs_inode_scan_goto_blockgroup(scan,
+                                                                       0);
+                               if (err) {
+                                       fix_problem(ctx, PR_1_ISCAN_ERROR,
+                                                   &pctx);
+                                       ctx->flags |= E2F_FLAG_ABORT;
+                                       goto endit;
+                               }
+                               continue;
+                       }
                        if (!ctx->inode_bb_map)
                                alloc_bb_map(ctx);
                        ext2fs_mark_inode_bitmap2(ctx->inode_bb_map, ino);
                        ext2fs_mark_inode_bitmap2(ctx->inode_used_map, ino);
                        continue;
                }
-               if (pctx.errcode) {
+               if (pctx.errcode &&
+                   pctx.errcode != EXT2_ET_INODE_CSUM_INVALID &&
+                   pctx.errcode != EXT2_ET_INODE_IS_GARBAGE) {
                        fix_problem(ctx, PR_1_ISCAN_ERROR, &pctx);
                        ctx->flags |= E2F_FLAG_ABORT;
-                       return;
+                       goto endit;
                }
                if (!ino)
                        break;
                pctx.ino = ino;
                pctx.inode = inode;
                ctx->stashed_ino = ino;
+
+               /* Clear trashed inode? */
+               if (pctx.errcode == EXT2_ET_INODE_IS_GARBAGE &&
+                   inode->i_links_count > 0 &&
+                   fix_problem(ctx, PR_1_INODE_IS_GARBAGE, &pctx)) {
+                       pctx.errcode = 0;
+                       e2fsck_clear_inode(ctx, ino, inode, 0, "pass1");
+               }
+               failed_csum = pctx.errcode != 0;
+
                if (inode->i_links_count) {
                        pctx.errcode = ext2fs_icount_store(ctx->inode_link_info,
                                           ino, inode->i_links_count);
@@ -756,7 +1238,96 @@ void e2fsck_pass1(e2fsck_t ctx)
                                pctx.num = inode->i_links_count;
                                fix_problem(ctx, PR_1_ICOUNT_STORE, &pctx);
                                ctx->flags |= E2F_FLAG_ABORT;
-                               return;
+                               goto endit;
+                       }
+               }
+
+               /* Conflicting inlinedata/extents inode flags? */
+               if ((inode->i_flags & EXT4_INLINE_DATA_FL) &&
+                   (inode->i_flags & EXT4_EXTENTS_FL)) {
+                       int res = fix_inline_data_extents_file(ctx, ino, inode,
+                                                              inode_size,
+                                                              &pctx);
+                       if (res < 0) {
+                               /* skip FINISH_INODE_LOOP */
+                               continue;
+                       }
+               }
+
+               /* Test for incorrect inline_data flags settings. */
+               if ((inode->i_flags & EXT4_INLINE_DATA_FL) && !inlinedata_fs &&
+                   (ino >= EXT2_FIRST_INODE(fs->super))) {
+                       size_t size = 0;
+
+                       pctx.errcode = ext2fs_inline_data_size(fs, ino, &size);
+                       if (!pctx.errcode && size &&
+                           fix_problem(ctx, PR_1_INLINE_DATA_FEATURE, &pctx)) {
+                               sb->s_feature_incompat |=
+                                       EXT4_FEATURE_INCOMPAT_INLINE_DATA;
+                               ext2fs_mark_super_dirty(fs);
+                               inlinedata_fs = 1;
+                       } else if (fix_problem(ctx, PR_1_INLINE_DATA_SET, &pctx)) {
+                               e2fsck_clear_inode(ctx, ino, inode, 0, "pass1");
+                               /* skip FINISH_INODE_LOOP */
+                               continue;
+                       }
+               }
+
+               /* Test for inline data flag but no attr */
+               if ((inode->i_flags & EXT4_INLINE_DATA_FL) && inlinedata_fs &&
+                   EXT2_I_SIZE(inode) > EXT4_MIN_INLINE_DATA_SIZE &&
+                   (ino >= EXT2_FIRST_INODE(fs->super))) {
+                       size_t size = 0;
+                       errcode_t err;
+                       int flags;
+
+                       flags = fs->flags;
+                       if (failed_csum)
+                               fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
+                       err = get_inline_data_ea_size(fs, ino, &size);
+                       fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
+                                   (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
+
+                       switch (err) {
+                       case 0:
+                               /* Everything is awesome... */
+                               break;
+                       case EXT2_ET_BAD_EA_BLOCK_NUM:
+                       case EXT2_ET_BAD_EA_HASH:
+                       case EXT2_ET_BAD_EA_HEADER:
+                       case EXT2_ET_EA_BAD_NAME_LEN:
+                       case EXT2_ET_EA_BAD_VALUE_SIZE:
+                       case EXT2_ET_EA_KEY_NOT_FOUND:
+                       case EXT2_ET_EA_NO_SPACE:
+                       case EXT2_ET_MISSING_EA_FEATURE:
+                       case EXT2_ET_INLINE_DATA_CANT_ITERATE:
+                       case EXT2_ET_INLINE_DATA_NO_BLOCK:
+                       case EXT2_ET_INLINE_DATA_NO_SPACE:
+                       case EXT2_ET_NO_INLINE_DATA:
+                       case EXT2_ET_EXT_ATTR_CSUM_INVALID:
+                       case EXT2_ET_EA_BAD_VALUE_OFFSET:
+                               /* broken EA or no system.data EA; truncate */
+                               if (fix_problem(ctx, PR_1_INLINE_DATA_NO_ATTR,
+                                               &pctx)) {
+                                       err = ext2fs_inode_size_set(fs, inode,
+                                                       sizeof(inode->i_block));
+                                       if (err) {
+                                               pctx.errcode = err;
+                                               ctx->flags |= E2F_FLAG_ABORT;
+                                               goto endit;
+                                       }
+                                       if (LINUX_S_ISLNK(inode->i_mode))
+                                               inode->i_flags &= ~EXT4_INLINE_DATA_FL;
+                                       e2fsck_write_inode(ctx, ino, inode,
+                                                          "pass1");
+                                       failed_csum = 0;
+                               }
+                               break;
+                       default:
+                               /* Some other kind of non-xattr error? */
+                               pctx.errcode = err;
+                               ctx->flags |= E2F_FLAG_ABORT;
+                               goto endit;
                        }
                }
 
@@ -790,6 +1361,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                if (ino == EXT2_BAD_INO)
                                        ext2fs_mark_inode_bitmap2(ctx->inode_used_map,
                                                                 ino);
+                               /* skip FINISH_INODE_LOOP */
                                continue;
                        }
                }
@@ -826,18 +1398,22 @@ void e2fsck_pass1(e2fsck_t ctx)
                                       sizeof(inode->i_block));
 #endif
                                e2fsck_write_inode(ctx, ino, inode, "pass1");
+                               failed_csum = 0;
                        }
                }
 
                if (ino == EXT2_BAD_INO) {
                        struct process_block_struct pb;
 
-                       if ((inode->i_mode || inode->i_uid || inode->i_gid ||
-                            inode->i_links_count || inode->i_file_acl) &&
+                       if ((failed_csum || inode->i_mode || inode->i_uid ||
+                            inode->i_gid || inode->i_links_count ||
+                            (inode->i_flags & EXT4_INLINE_DATA_FL) ||
+                            inode->i_file_acl) &&
                            fix_problem(ctx, PR_1_INVALID_BAD_INODE, &pctx)) {
                                memset(inode, 0, sizeof(struct ext2_inode));
                                e2fsck_write_inode(ctx, ino, inode,
                                                   "clear bad inode");
+                               failed_csum = 0;
                        }
 
                        pctx.errcode = ext2fs_copy_bitmap(ctx->block_found_map,
@@ -846,7 +1422,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                pctx.num = 4;
                                fix_problem(ctx, PR_1_ALLOCATE_BBITMAP_ERROR, &pctx);
                                ctx->flags |= E2F_FLAG_ABORT;
-                               return;
+                               goto endit;
                        }
                        pb.ino = EXT2_BAD_INO;
                        pb.num_blocks = pb.last_block = 0;
@@ -863,15 +1439,16 @@ void e2fsck_pass1(e2fsck_t ctx)
                        if (pctx.errcode) {
                                fix_problem(ctx, PR_1_BLOCK_ITERATE, &pctx);
                                ctx->flags |= E2F_FLAG_ABORT;
-                               return;
+                               goto endit;
                        }
                        if (pb.bbcheck)
                                if (!fix_problem(ctx, PR_1_BBINODE_BAD_METABLOCK_PROMPT, &pctx)) {
                                ctx->flags |= E2F_FLAG_ABORT;
-                               return;
+                               goto endit;
                        }
                        ext2fs_mark_inode_bitmap2(ctx->inode_used_map, ino);
                        clear_problem_context(&pctx);
+                       FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                        continue;
                } else if (ino == EXT2_ROOT_INO) {
                        /*
@@ -897,6 +1474,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_dtime = 0;
                                        e2fsck_write_inode(ctx, ino, inode,
                                                           "pass1");
+                                       failed_csum = 0;
                                }
                        }
                } else if (ino == EXT2_JOURNAL_INO) {
@@ -908,8 +1486,10 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_mode = LINUX_S_IFREG;
                                        e2fsck_write_inode(ctx, ino, inode,
                                                           "pass1");
+                                       failed_csum = 0;
                                }
                                check_blocks(ctx, &pctx, block_buf);
+                               FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                                continue;
                        }
                        if ((inode->i_links_count ||
@@ -921,6 +1501,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                                    ino, 0);
                                e2fsck_write_inode_full(ctx, ino, inode,
                                                        inode_size, "pass1");
+                               failed_csum = 0;
                        }
                } else if ((ino == EXT4_USR_QUOTA_INO) ||
                           (ino == EXT4_GRP_QUOTA_INO)) {
@@ -935,8 +1516,10 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_mode = LINUX_S_IFREG;
                                        e2fsck_write_inode(ctx, ino, inode,
                                                        "pass1");
+                                       failed_csum = 0;
                                }
                                check_blocks(ctx, &pctx, block_buf);
+                               FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                                continue;
                        }
                        if ((inode->i_links_count ||
@@ -948,9 +1531,10 @@ void e2fsck_pass1(e2fsck_t ctx)
                                                    ino, 0);
                                e2fsck_write_inode_full(ctx, ino, inode,
                                                        inode_size, "pass1");
+                               failed_csum = 0;
                        }
                } else if (ino < EXT2_FIRST_INODE(fs->super)) {
-                       int     problem = 0;
+                       problem_t problem = 0;
 
                        ext2fs_mark_inode_bitmap2(ctx->inode_used_map, ino);
                        if (ino == EXT2_BOOT_LOADER_INO) {
@@ -969,9 +1553,11 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_mode = 0;
                                        e2fsck_write_inode(ctx, ino, inode,
                                                           "pass1");
+                                       failed_csum = 0;
                                }
                        }
                        check_blocks(ctx, &pctx, block_buf);
+                       FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                        continue;
                }
 
@@ -992,13 +1578,14 @@ void e2fsck_pass1(e2fsck_t ctx)
                 * than nothing.  The right answer is that there
                 * shouldn't be any bugs in the orphan list handling.  :-)
                 */
-               if (inode->i_dtime && !busted_fs_time &&
+               if (inode->i_dtime && low_dtime_check &&
                    inode->i_dtime < ctx->fs->super->s_inodes_count) {
                        if (fix_problem(ctx, PR_1_LOW_DTIME, &pctx)) {
                                inode->i_dtime = inode->i_links_count ?
                                        0 : ctx->now;
                                e2fsck_write_inode(ctx, ino, inode,
                                                   "pass1");
+                               failed_csum = 0;
                        }
                }
 
@@ -1013,8 +1600,10 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_dtime = ctx->now;
                                        e2fsck_write_inode(ctx, ino, inode,
                                                           "pass1");
+                                       failed_csum = 0;
                                }
                        }
+                       FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                        continue;
                }
                /*
@@ -1031,6 +1620,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                        if (fix_problem(ctx, PR_1_SET_DTIME, &pctx)) {
                                inode->i_dtime = 0;
                                e2fsck_write_inode(ctx, ino, inode, "pass1");
+                               failed_csum = 0;
                        }
                }
 
@@ -1047,7 +1637,8 @@ void e2fsck_pass1(e2fsck_t ctx)
                if (inode->i_faddr || frag || fsize ||
                    (LINUX_S_ISDIR(inode->i_mode) && inode->i_dir_acl))
                        mark_inode_bad(ctx, ino);
-               if (!(fs->super->s_feature_incompat & 
+               if ((fs->super->s_creator_os == EXT2_OS_LINUX) &&
+                   !(fs->super->s_feature_incompat &
                      EXT4_FEATURE_INCOMPAT_64BIT) &&
                    inode->osd2.linux2.l_i_file_acl_high != 0)
                        mark_inode_bad(ctx, ino);
@@ -1067,6 +1658,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                        inode->i_flags &= ~EXT2_IMAGIC_FL;
                                        e2fsck_write_inode(ctx, ino,
                                                           inode, "pass1");
+                                       failed_csum = 0;
                                }
                        }
                }
@@ -1084,22 +1676,27 @@ void e2fsck_pass1(e2fsck_t ctx)
                    fix_problem(ctx, PR_1_FAST_SYMLINK_EXTENT_FL, &pctx)) {
                        inode->i_flags &= ~EXT4_EXTENTS_FL;
                        e2fsck_write_inode(ctx, ino, inode, "pass1");
+                       failed_csum = 0;
                }
 
                if (LINUX_S_ISDIR(inode->i_mode)) {
                        ext2fs_mark_inode_bitmap2(ctx->inode_dir_map, ino);
                        e2fsck_add_dir_info(ctx, ino, 0);
                        ctx->fs_directory_count++;
+                       if (inode->i_flags & EXT4_ENCRYPT_FL)
+                               add_encrypted_dir(ctx, ino);
                } else if (LINUX_S_ISREG (inode->i_mode)) {
                        ext2fs_mark_inode_bitmap2(ctx->inode_reg_map, ino);
                        ctx->fs_regular_count++;
                } else if (LINUX_S_ISCHR (inode->i_mode) &&
                           e2fsck_pass1_check_device_inode(fs, inode)) {
+                       check_extents_inlinedata(ctx, &pctx);
                        check_immutable(ctx, &pctx);
                        check_size(ctx, &pctx);
                        ctx->fs_chardev_count++;
                } else if (LINUX_S_ISBLK (inode->i_mode) &&
                           e2fsck_pass1_check_device_inode(fs, inode)) {
+                       check_extents_inlinedata(ctx, &pctx);
                        check_immutable(ctx, &pctx);
                        check_size(ctx, &pctx);
                        ctx->fs_blockdev_count++;
@@ -1108,25 +1705,32 @@ void e2fsck_pass1(e2fsck_t ctx)
                                                      block_buf)) {
                        check_immutable(ctx, &pctx);
                        ctx->fs_symlinks_count++;
-                       if (ext2fs_inode_data_blocks(fs, inode) == 0) {
+                       if (inode->i_flags & EXT4_INLINE_DATA_FL) {
+                               FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
+                               continue;
+                       } else if (ext2fs_inode_data_blocks(fs, inode) == 0) {
                                ctx->fs_fast_symlinks_count++;
                                check_blocks(ctx, &pctx, block_buf);
+                               FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
                                continue;
                        }
                }
                else if (LINUX_S_ISFIFO (inode->i_mode) &&
                         e2fsck_pass1_check_device_inode(fs, inode)) {
+                       check_extents_inlinedata(ctx, &pctx);
                        check_immutable(ctx, &pctx);
                        check_size(ctx, &pctx);
                        ctx->fs_fifo_count++;
                } else if ((LINUX_S_ISSOCK (inode->i_mode)) &&
                           e2fsck_pass1_check_device_inode(fs, inode)) {
+                       check_extents_inlinedata(ctx, &pctx);
                        check_immutable(ctx, &pctx);
                        check_size(ctx, &pctx);
                        ctx->fs_sockets_count++;
                } else
                        mark_inode_bad(ctx, ino);
-               if (!(inode->i_flags & EXT4_EXTENTS_FL)) {
+               if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
+                   !(inode->i_flags & EXT4_INLINE_DATA_FL)) {
                        if (inode->i_block[EXT2_IND_BLOCK])
                                ctx->fs_ind_count++;
                        if (inode->i_block[EXT2_DIND_BLOCK])
@@ -1135,6 +1739,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                                ctx->fs_tind_count++;
                }
                if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
+                   !(inode->i_flags & EXT4_INLINE_DATA_FL) &&
                    (inode->i_block[EXT2_IND_BLOCK] ||
                     inode->i_block[EXT2_DIND_BLOCK] ||
                     inode->i_block[EXT2_TIND_BLOCK] ||
@@ -1145,18 +1750,24 @@ void e2fsck_pass1(e2fsck_t ctx)
                } else
                        check_blocks(ctx, &pctx, block_buf);
 
+               FINISH_INODE_LOOP(ctx, ino, &pctx, failed_csum);
+
                if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
-                       return;
+                       goto endit;
 
                if (process_inode_count >= ctx->process_inode_size) {
                        process_inodes(ctx, block_buf);
 
                        if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
-                               return;
+                               goto endit;
                }
        }
        process_inodes(ctx, block_buf);
        ext2fs_close_inode_scan(scan);
+       scan = NULL;
+
+       reserve_block_for_root_repair(ctx);
+       reserve_block_for_lnf_repair(ctx);
 
        /*
         * If any extended attribute blocks' reference counts need to
@@ -1185,17 +1796,13 @@ void e2fsck_pass1(e2fsck_t ctx)
        }
 
        if (ctx->flags & E2F_FLAG_RESIZE_INODE) {
-               ext2fs_block_bitmap save_bmap;
-
-               save_bmap = fs->block_map;
-               fs->block_map = ctx->block_found_map;
                clear_problem_context(&pctx);
                pctx.errcode = ext2fs_create_resize_inode(fs);
                if (pctx.errcode) {
                        if (!fix_problem(ctx, PR_1_RESIZE_INODE_CREATE,
                                         &pctx)) {
                                ctx->flags |= E2F_FLAG_ABORT;
-                               return;
+                               goto endit;
                        }
                        pctx.errcode = 0;
                }
@@ -1206,7 +1813,6 @@ void e2fsck_pass1(e2fsck_t ctx)
                        e2fsck_write_inode(ctx, EXT2_RESIZE_INO, inode,
                                           "recreate inode");
                }
-               fs->block_map = save_bmap;
                ctx->flags &= ~E2F_FLAG_RESIZE_INODE;
        }
 
@@ -1233,11 +1839,23 @@ void e2fsck_pass1(e2fsck_t ctx)
 endit:
        e2fsck_use_inode_shortcuts(ctx, 0);
 
-       ext2fs_free_mem(&block_buf);
-       ext2fs_free_mem(&inode);
+       if (scan)
+               ext2fs_close_inode_scan(scan);
+       if (block_buf)
+               ext2fs_free_mem(&block_buf);
+       if (inode)
+               ext2fs_free_mem(&inode);
+
+       /*
+        * The l+f inode may have been cleared, so zap it now and
+        * later passes will recalculate it if necessary
+        */
+       ctx->lost_and_found = 0;
 
-       print_resource_track(ctx, _("Pass 1"), &rtrack, ctx->fs->io);
+       if ((ctx->flags & E2F_FLAG_SIGNAL_MASK) == 0)
+               print_resource_track(ctx, _("Pass 1"), &rtrack, ctx->fs->io);
 }
+#undef FINISH_INODE_LOOP
 
 /*
  * When the inode_scan routines call this callback at the end of the
@@ -1356,6 +1974,23 @@ static void mark_inode_bad(e2fsck_t ctx, ino_t ino)
        ext2fs_mark_inode_bitmap2(ctx->inode_bad_map, ino);
 }
 
+static void add_encrypted_dir(e2fsck_t ctx, ino_t ino)
+{
+       struct          problem_context pctx;
+
+       if (!ctx->encrypted_dirs) {
+               pctx.errcode = ext2fs_u32_list_create(&ctx->encrypted_dirs, 0);
+               if (pctx.errcode)
+                       goto error;
+       }
+       pctx.errcode = ext2fs_u32_list_add(ctx->encrypted_dirs, ino);
+       if (pctx.errcode == 0)
+               return;
+error:
+       fix_problem(ctx, PR_1_ALLOCATE_ENCRYPTED_DIRLIST, &pctx);
+       /* Should never get here */
+       ctx->flags |= E2F_FLAG_ABORT;
+}
 
 /*
  * This procedure will allocate the inode "bb" (badblock) map table
@@ -1465,7 +2100,8 @@ static void adjust_extattr_refcount(e2fsck_t ctx, ext2_refcount_t refcount,
                if ((blk = ea_refcount_intr_next(refcount, &count)) == 0)
                        break;
                pctx.blk = blk;
-               pctx.errcode = ext2fs_read_ext_attr2(fs, blk, block_buf);
+               pctx.errcode = ext2fs_read_ext_attr3(fs, blk, block_buf,
+                                                    pctx.ino);
                if (pctx.errcode) {
                        fix_problem(ctx, PR_1_EXTATTR_READ_ABORT, &pctx);
                        return;
@@ -1476,8 +2112,9 @@ static void adjust_extattr_refcount(e2fsck_t ctx, ext2_refcount_t refcount,
                pctx.num = should_be;
                if (fix_problem(ctx, PR_1_EXTATTR_REFCOUNT, &pctx)) {
                        header->h_refcount = should_be;
-                       pctx.errcode = ext2fs_write_ext_attr2(fs, blk,
-                                                            block_buf);
+                       pctx.errcode = ext2fs_write_ext_attr3(fs, blk,
+                                                            block_buf,
+                                                            pctx.ino);
                        if (pctx.errcode) {
                                fix_problem(ctx, PR_1_EXTATTR_WRITE_ABORT,
                                            &pctx);
@@ -1502,6 +2139,7 @@ static int check_ext_attr(e2fsck_t ctx, struct problem_context *pctx,
        struct ext2_ext_attr_entry *entry;
        int             count;
        region_t        region = 0;
+       int             failed_csum = 0;
 
        blk = ext2fs_file_acl_block(fs, inode);
        if (blk == 0)
@@ -1575,9 +2213,18 @@ static int check_ext_attr(e2fsck_t ctx, struct problem_context *pctx,
         * validate it
         */
        pctx->blk = blk;
-       pctx->errcode = ext2fs_read_ext_attr2(fs, blk, block_buf);
-       if (pctx->errcode && fix_problem(ctx, PR_1_READ_EA_BLOCK, pctx))
+       pctx->errcode = ext2fs_read_ext_attr3(fs, blk, block_buf, pctx->ino);
+       if (pctx->errcode == EXT2_ET_EXT_ATTR_CSUM_INVALID) {
+               pctx->errcode = 0;
+               failed_csum = 1;
+       } else if (pctx->errcode == EXT2_ET_BAD_EA_HEADER)
+               pctx->errcode = 0;
+
+       if (pctx->errcode &&
+           fix_problem(ctx, PR_1_READ_EA_BLOCK, pctx)) {
+               pctx->errcode = 0;
                goto clear_extattr;
+       }
        header = (struct ext2_ext_attr_header *) block_buf;
        pctx->blk = ext2fs_file_acl_block(fs, inode);
        if (((ctx->ext_attr_ver == 1) &&
@@ -1593,6 +2240,9 @@ static int check_ext_attr(e2fsck_t ctx, struct problem_context *pctx,
                        goto clear_extattr;
        }
 
+       if (pctx->errcode && fix_problem(ctx, PR_1_READ_EA_BLOCK, pctx))
+               goto clear_extattr;
+
        region = region_create(0, fs->blocksize);
        if (!region) {
                fix_problem(ctx, PR_1_EA_ALLOC_REGION_ABORT, pctx);
@@ -1657,6 +2307,18 @@ static int check_ext_attr(e2fsck_t ctx, struct problem_context *pctx,
        }
        region_free(region);
 
+       /*
+        * We only get here if there was no other errors that were fixed.
+        * If there was a checksum fail, ask to correct it.
+        */
+       if (failed_csum &&
+           fix_problem(ctx, PR_1_EA_BLOCK_ONLY_CSUM_INVALID, pctx)) {
+               pctx->errcode = ext2fs_write_ext_attr3(fs, blk, block_buf,
+                                                      pctx->ino);
+               if (pctx->errcode)
+                       return 0;
+       }
+
        count = header->h_refcount - 1;
        if (count)
                ea_refcount_store(ctx->refcount, blk, count);
@@ -1739,6 +2401,16 @@ void e2fsck_clear_inode(e2fsck_t ctx, ext2_ino_t ino,
        ext2fs_icount_store(ctx->inode_link_info, ino, 0);
        inode->i_dtime = ctx->now;
 
+       /*
+        * If a special inode has such rotten block mappings that we
+        * want to clear the whole inode, be sure to actually zap
+        * the block maps because i_links_count isn't checked for
+        * special inodes, and we'll end up right back here the next
+        * time we run fsck.
+        */
+       if (ino < EXT2_FIRST_INODE(ctx->fs->super))
+               memset(inode->i_block, 0, sizeof(inode->i_block));
+
        ext2fs_unmark_inode_bitmap2(ctx->inode_dir_map, ino);
        ext2fs_unmark_inode_bitmap2(ctx->inode_used_map, ino);
        if (ctx->inode_reg_map)
@@ -1758,19 +2430,58 @@ void e2fsck_clear_inode(e2fsck_t ctx, ext2_ino_t ino,
        e2fsck_write_inode(ctx, ino, inode, source);
 }
 
+/*
+ * Use the multiple-blocks reclamation code to fix alignment problems in
+ * a bigalloc filesystem.  We want a logical cluster to map to *only* one
+ * physical cluster, and we want the block offsets within that cluster to
+ * line up.
+ */
+static int has_unaligned_cluster_map(e2fsck_t ctx,
+                                    blk64_t last_pblk, e2_blkcnt_t last_lblk,
+                                    blk64_t pblk, blk64_t lblk)
+{
+       blk64_t cluster_mask;
+
+       if (!ctx->fs->cluster_ratio_bits)
+               return 0;
+       cluster_mask = EXT2FS_CLUSTER_MASK(ctx->fs);
+
+       /*
+        * If the block in the logical cluster doesn't align with the block in
+        * the physical cluster...
+        */
+       if ((lblk & cluster_mask) != (pblk & cluster_mask))
+               return 1;
+
+       /*
+        * If we cross a physical cluster boundary within a logical cluster...
+        */
+       if (last_pblk && (lblk & cluster_mask) != 0 &&
+           EXT2FS_B2C(ctx->fs, lblk) == EXT2FS_B2C(ctx->fs, last_lblk) &&
+           EXT2FS_B2C(ctx->fs, pblk) != EXT2FS_B2C(ctx->fs, last_pblk))
+               return 1;
+
+       return 0;
+}
+
 static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
                             struct process_block_struct *pb,
                             blk64_t start_block, blk64_t end_block,
                             blk64_t eof_block,
-                            ext2_extent_handle_t ehandle)
+                            ext2_extent_handle_t ehandle,
+                            int try_repairs)
 {
        struct ext2fs_extent    extent;
        blk64_t                 blk, last_lblk;
        e2_blkcnt_t             blockcnt;
        unsigned int            i;
        int                     is_dir, is_leaf;
-       errcode_t               problem;
+       problem_t               problem;
        struct ext2_extent_info info;
+       int                     failed_csum = 0;
+
+       if (pctx->errcode == EXT2_ET_EXTENT_CSUM_INVALID)
+               failed_csum = 1;
 
        pctx->errcode = ext2fs_extent_get_info(ehandle, &info);
        if (pctx->errcode)
@@ -1778,12 +2489,19 @@ static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
 
        pctx->errcode = ext2fs_extent_get(ehandle, EXT2_EXTENT_FIRST_SIB,
                                          &extent);
-       while (!pctx->errcode && info.num_entries-- > 0) {
+       while ((pctx->errcode == 0 ||
+               pctx->errcode == EXT2_ET_EXTENT_CSUM_INVALID) &&
+              info.num_entries-- > 0) {
                is_leaf = extent.e_flags & EXT2_EXTENT_FLAGS_LEAF;
                is_dir = LINUX_S_ISDIR(pctx->inode->i_mode);
                last_lblk = extent.e_lblk + extent.e_len - 1;
 
                problem = 0;
+               pctx->blk = extent.e_pblk;
+               pctx->blk2 = extent.e_lblk;
+               pctx->num = extent.e_len;
+               pctx->blkcount = extent.e_lblk + extent.e_len;
+
                if (extent.e_pblk == 0 ||
                    extent.e_pblk < ctx->fs->super->s_first_data_block ||
                    extent.e_pblk >= ext2fs_blocks_count(ctx->fs->super))
@@ -1805,21 +2523,63 @@ static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
                          (1 << (21 - ctx->fs->super->s_log_block_size))))
                        problem = PR_1_TOOBIG_DIR;
 
-               if (problem) {
+               if (is_leaf && problem == 0 && extent.e_len > 0 &&
+                   region_allocate(pb->region, extent.e_lblk, extent.e_len))
+                       problem = PR_1_EXTENT_COLLISION;
+
+               /*
+                * Uninitialized blocks in a directory?  Clear the flag and
+                * we'll interpret the blocks later.
+                */
+               if (try_repairs && is_dir && problem == 0 &&
+                   (extent.e_flags & EXT2_EXTENT_FLAGS_UNINIT) &&
+                   fix_problem(ctx, PR_1_UNINIT_DBLOCK, pctx)) {
+                       extent.e_flags &= ~EXT2_EXTENT_FLAGS_UNINIT;
+                       pb->inode_modified = 1;
+                       pctx->errcode = ext2fs_extent_replace(ehandle, 0,
+                                                             &extent);
+                       if (pctx->errcode)
+                               return;
+                       failed_csum = 0;
+               }
+
+               if (try_repairs && problem) {
 report_problem:
-                       pctx->blk = extent.e_pblk;
-                       pctx->blk2 = extent.e_lblk;
-                       pctx->num = extent.e_len;
-                       pctx->blkcount = extent.e_lblk + extent.e_len;
                        if (fix_problem(ctx, problem, pctx)) {
+                               if (ctx->invalid_bitmaps) {
+                                       /*
+                                        * If fsck knows the bitmaps are bad,
+                                        * skip to the next extent and
+                                        * try to clear this extent again
+                                        * after fixing the bitmaps, by
+                                        * restarting fsck.
+                                        */
+                                       pctx->errcode = ext2fs_extent_get(
+                                                         ehandle,
+                                                         EXT2_EXTENT_NEXT_SIB,
+                                                         &extent);
+                                       ctx->flags |= E2F_FLAG_RESTART_LATER;
+                                       if (pctx->errcode ==
+                                                   EXT2_ET_NO_CURRENT_NODE) {
+                                               pctx->errcode = 0;
+                                               break;
+                                       }
+                                       continue;
+                               }
                                e2fsck_read_bitmaps(ctx);
+                               pb->inode_modified = 1;
                                pctx->errcode =
                                        ext2fs_extent_delete(ehandle, 0);
                                if (pctx->errcode) {
                                        pctx->str = "ext2fs_extent_delete";
                                        return;
                                }
-                               ext2fs_extent_fix_parents(ehandle);
+                               pctx->errcode = ext2fs_extent_fix_parents(ehandle);
+                               if (pctx->errcode &&
+                                   pctx->errcode != EXT2_ET_NO_CURRENT_NODE) {
+                                       pctx->str = "ext2fs_extent_fix_parents";
+                                       return;
+                               }
                                pctx->errcode = ext2fs_extent_get(ehandle,
                                                                  EXT2_EXTENT_CURRENT,
                                                                  &extent);
@@ -1827,6 +2587,7 @@ report_problem:
                                        pctx->errcode = 0;
                                        break;
                                }
+                               failed_csum = 0;
                                continue;
                        }
                        goto next;
@@ -1834,13 +2595,33 @@ report_problem:
 
                if (!is_leaf) {
                        blk64_t lblk = extent.e_lblk;
+                       int next_try_repairs = 1;
 
                        blk = extent.e_pblk;
+
+                       /*
+                        * If this lower extent block collides with critical
+                        * metadata, don't try to repair the damage.  Pass 1b
+                        * will reallocate the block; then we can try again.
+                        */
+                       if (pb->ino != EXT2_RESIZE_INO &&
+                           ext2fs_test_block_bitmap2(ctx->block_metadata_map,
+                                                     extent.e_pblk)) {
+                               next_try_repairs = 0;
+                               pctx->blk = blk;
+                               fix_problem(ctx,
+                                           PR_1_CRITICAL_METADATA_COLLISION,
+                                           pctx);
+                               ctx->flags |= E2F_FLAG_RESTART_LATER;
+                       }
                        pctx->errcode = ext2fs_extent_get(ehandle,
                                                  EXT2_EXTENT_DOWN, &extent);
-                       if (pctx->errcode) {
+                       if (pctx->errcode &&
+                           pctx->errcode != EXT2_ET_EXTENT_CSUM_INVALID) {
                                pctx->str = "EXT2_EXTENT_DOWN";
                                problem = PR_1_EXTENT_HEADER_INVALID;
+                               if (!next_try_repairs)
+                                       return;
                                if (pctx->errcode == EXT2_ET_EXTENT_HEADER_BAD)
                                        goto report_problem;
                                return;
@@ -1854,11 +2635,19 @@ report_problem:
                                pctx->blk2 = extent.e_lblk;
                                pctx->num = e_info.curr_level - 1;
                                problem = PR_1_EXTENT_INDEX_START_INVALID;
-                               if (fix_problem(ctx, problem, pctx))
-                                       ext2fs_extent_fix_parents(ehandle);
+                               if (fix_problem(ctx, problem, pctx)) {
+                                       pb->inode_modified = 1;
+                                       pctx->errcode =
+                                               ext2fs_extent_fix_parents(ehandle);
+                                       if (pctx->errcode) {
+                                               pctx->str = "ext2fs_extent_fix_parents";
+                                               return;
+                                       }
+                               }
                        }
                        scan_extent_node(ctx, pctx, pb, extent.e_lblk,
-                                        last_lblk, eof_block, ehandle);
+                                        last_lblk, eof_block, ehandle,
+                                        next_try_repairs);
                        if (pctx->errcode)
                                return;
                        pctx->errcode = ext2fs_extent_get(ehandle,
@@ -1893,6 +2682,46 @@ report_problem:
                        }
                        pb->fragmented = 1;
                }
+               /*
+                * If we notice a gap in the logical block mappings of an
+                * extent-mapped directory, offer to close the hole by
+                * moving the logical block down, otherwise we'll go mad in
+                * pass 3 allocating empty directory blocks to fill the hole.
+                */
+               if (try_repairs && is_dir &&
+                   pb->last_block + 1 < (e2_blkcnt_t)extent.e_lblk) {
+                       blk64_t new_lblk;
+
+                       new_lblk = pb->last_block + 1;
+                       if (EXT2FS_CLUSTER_RATIO(ctx->fs) > 1)
+                               new_lblk = ((new_lblk +
+                                            EXT2FS_CLUSTER_RATIO(ctx->fs) - 1) &
+                                           ~EXT2FS_CLUSTER_MASK(ctx->fs)) |
+                                          (extent.e_pblk &
+                                           EXT2FS_CLUSTER_MASK(ctx->fs));
+                       pctx->blk = extent.e_lblk;
+                       pctx->blk2 = new_lblk;
+                       if (fix_problem(ctx, PR_1_COLLAPSE_DBLOCK, pctx)) {
+                               extent.e_lblk = new_lblk;
+                               pb->inode_modified = 1;
+                               pctx->errcode = ext2fs_extent_replace(ehandle,
+                                                               0, &extent);
+                               if (pctx->errcode) {
+                                       pctx->errcode = 0;
+                                       goto alloc_later;
+                               }
+                               pctx->errcode = ext2fs_extent_fix_parents(ehandle);
+                               if (pctx->errcode)
+                                       goto failed_add_dir_block;
+                               pctx->errcode = ext2fs_extent_goto(ehandle,
+                                                               extent.e_lblk);
+                               if (pctx->errcode)
+                                       goto failed_add_dir_block;
+                               last_lblk = extent.e_lblk + extent.e_len - 1;
+                               failed_csum = 0;
+                       }
+               }
+alloc_later:
                while (is_dir && (++pb->last_db_block <
                                  (e2_blkcnt_t) extent.e_lblk)) {
                        pctx->errcode = ext2fs_add_dir_block2(ctx->fs->dblist,
@@ -1920,7 +2749,16 @@ report_problem:
                                mark_block_used(ctx, blk);
                                pb->num_blocks++;
                        }
-
+                       if (has_unaligned_cluster_map(ctx, pb->previous_block,
+                                                     pb->last_block, blk,
+                                                     blockcnt)) {
+                               pctx->blk = blockcnt;
+                               pctx->blk2 = blk;
+                               fix_problem(ctx, PR_1_MISALIGNED_CLUSTER, pctx);
+                               mark_block_used(ctx, blk);
+                               mark_block_used(ctx, blk);
+                       }
+                       pb->last_block = blockcnt;
                        pb->previous_block = blk;
 
                        if (is_dir) {
@@ -1948,6 +2786,16 @@ report_problem:
                                                  EXT2_EXTENT_NEXT_SIB,
                                                  &extent);
        }
+
+       /* Failed csum but passes checks?  Ask to fix checksum. */
+       if (failed_csum &&
+           fix_problem(ctx, PR_1_EXTENT_ONLY_CSUM_INVALID, pctx)) {
+               pb->inode_modified = 1;
+               pctx->errcode = ext2fs_extent_replace(ehandle, 0, &extent);
+               if (pctx->errcode)
+                       return;
+       }
+
        if (pctx->errcode == EXT2_ET_EXTENT_NO_NEXT)
                pctx->errcode = 0;
 }
@@ -1979,9 +2827,17 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
                ctx->extent_depth_count[info.max_depth]++;
        }
 
+       pb->region = region_create(0, info.max_lblk);
+       if (!pb->region) {
+               ext2fs_extent_free(ehandle);
+               fix_problem(ctx, PR_1_EXTENT_ALLOC_REGION_ABORT, pctx);
+               ctx->flags |= E2F_FLAG_ABORT;
+               return;
+       }
+
        eof_lblk = ((EXT2_I_SIZE(inode) + fs->blocksize - 1) >>
                EXT2_BLOCK_SIZE_BITS(fs->super)) - 1;
-       scan_extent_node(ctx, pctx, pb, 0, 0, eof_lblk, ehandle);
+       scan_extent_node(ctx, pctx, pb, 0, 0, eof_lblk, ehandle, 1);
        if (pctx->errcode &&
            fix_problem(ctx, PR_1_EXTENT_ITERATE_FAILURE, pctx)) {
                pb->num_blocks = 0;
@@ -1990,10 +2846,60 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
                                   "check_blocks_extents");
                pctx->errcode = 0;
        }
+       region_free(pb->region);
+       pb->region = NULL;
        ext2fs_extent_free(ehandle);
 }
 
 /*
+ * In fact we don't need to check blocks for an inode with inline data
+ * because this inode doesn't have any blocks.  In this function all
+ * we need to do is add this inode into dblist when it is a directory.
+ */
+static void check_blocks_inline_data(e2fsck_t ctx, struct problem_context *pctx,
+                                    struct process_block_struct *pb)
+{
+       int     flags;
+       size_t  inline_data_size = 0;
+
+       if (!pb->is_dir) {
+               pctx->errcode = 0;
+               return;
+       }
+
+       /* Process the dirents in i_block[] as the "first" block. */
+       pctx->errcode = ext2fs_add_dir_block2(ctx->fs->dblist, pb->ino, 0, 0);
+       if (pctx->errcode)
+               goto err;
+
+       /* Process the dirents in the EA as a "second" block. */
+       flags = ctx->fs->flags;
+       ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
+       pctx->errcode = ext2fs_inline_data_size(ctx->fs, pb->ino,
+                                               &inline_data_size);
+       ctx->fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
+                        (ctx->fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
+       if (pctx->errcode) {
+               pctx->errcode = 0;
+               return;
+       }
+
+       if (inline_data_size <= EXT4_MIN_INLINE_DATA_SIZE)
+               return;
+
+       pctx->errcode = ext2fs_add_dir_block2(ctx->fs->dblist, pb->ino, 0, 1);
+       if (pctx->errcode)
+               goto err;
+
+       return;
+err:
+       pctx->blk = 0;
+       pctx->num = 0;
+       fix_problem(ctx, PR_1_ADD_DBLOCK, pctx);
+       ctx->flags |= E2F_FLAG_ABORT;
+}
+
+/*
  * This subroutine is called on each inode to account for all of the
  * blocks used by that inode.
  */
@@ -2004,9 +2910,10 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        struct process_block_struct pb;
        ext2_ino_t      ino = pctx->ino;
        struct ext2_inode *inode = pctx->inode;
-       int             bad_size = 0;
+       unsigned        bad_size = 0;
        int             dirty_inode = 0;
        int             extent_fs;
+       int             inlinedata_fs;
        __u64           size;
 
        pb.ino = ino;
@@ -2025,35 +2932,39 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        pb.inode = inode;
        pb.pctx = pctx;
        pb.ctx = ctx;
+       pb.inode_modified = 0;
        pctx->ino = ino;
        pctx->errcode = 0;
 
        extent_fs = (ctx->fs->super->s_feature_incompat &
                      EXT3_FEATURE_INCOMPAT_EXTENTS);
+       inlinedata_fs = (ctx->fs->super->s_feature_incompat &
+                        EXT4_FEATURE_INCOMPAT_INLINE_DATA);
 
-       if (inode->i_flags & EXT2_COMPRBLK_FL) {
-               if (fs->super->s_feature_incompat &
-                   EXT2_FEATURE_INCOMPAT_COMPRESSION)
-                       pb.compressed = 1;
-               else {
-                       if (fix_problem(ctx, PR_1_COMPR_SET, pctx)) {
-                               inode->i_flags &= ~EXT2_COMPRBLK_FL;
-                               dirty_inode++;
-                       }
-               }
-       }
-
-       if (ext2fs_file_acl_block(fs, inode) &&
-           check_ext_attr(ctx, pctx, block_buf)) {
+       if (check_ext_attr(ctx, pctx, block_buf)) {
                if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
                        goto out;
                pb.num_blocks++;
        }
 
-       if (ext2fs_inode_has_valid_blocks2(fs, inode)) {
+       if (inlinedata_fs && (inode->i_flags & EXT4_INLINE_DATA_FL))
+               check_blocks_inline_data(ctx, pctx, &pb);
+       else if (ext2fs_inode_has_valid_blocks2(fs, inode)) {
                if (extent_fs && (inode->i_flags & EXT4_EXTENTS_FL))
                        check_blocks_extents(ctx, pctx, &pb);
                else {
+                       int flags;
+                       /*
+                        * If we've modified the inode, write it out before
+                        * iterate() tries to use it.
+                        */
+                       if (dirty_inode) {
+                               e2fsck_write_inode(ctx, ino, inode,
+                                                  "check_blocks");
+                               dirty_inode = 0;
+                       }
+                       flags = fs->flags;
+                       fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
                        pctx->errcode = ext2fs_block_iterate3(fs, ino,
                                                pb.is_dir ? BLOCK_FLAG_HOLE : 0,
                                                block_buf, process_block, &pb);
@@ -2062,6 +2973,17 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                         * files.
                         */
                        pb.last_init_lblock = pb.last_block;
+                       /*
+                        * If iterate() changed a block mapping, we have to
+                        * re-read the inode.  If we decide to clear the
+                        * inode after clearing some stuff, we'll re-write the
+                        * bad mappings into the inode!
+                        */
+                       if (pb.inode_modified)
+                               e2fsck_read_inode(ctx, ino, inode,
+                                                 "check_blocks");
+                       fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
+                                   (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
                }
        }
        end_problem_latch(ctx, PR_LATCH_BLOCK);
@@ -2095,7 +3017,8 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                }
        }
 
-       if (!pb.num_blocks && pb.is_dir) {
+       if (!pb.num_blocks && pb.is_dir &&
+           !(inode->i_flags & EXT4_INLINE_DATA_FL)) {
                if (fix_problem(ctx, PR_1_ZERO_LENGTH_DIR, pctx)) {
                        e2fsck_clear_inode(ctx, ino, inode, 0, "check_blocks");
                        ctx->fs_directory_count--;
@@ -2121,7 +3044,25 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
 #endif
        if (pb.is_dir) {
                int nblock = inode->i_size >> EXT2_BLOCK_SIZE_BITS(fs->super);
-               if (inode->i_size & (fs->blocksize - 1))
+               if (inode->i_flags & EXT4_INLINE_DATA_FL) {
+                       int flags;
+                       size_t size;
+                       errcode_t err;
+
+                       size = 0;
+                       flags = ctx->fs->flags;
+                       ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
+                       err = ext2fs_inline_data_size(ctx->fs, pctx->ino,
+                                                     &size);
+                       ctx->fs->flags = (flags &
+                                         EXT2_FLAG_IGNORE_CSUM_ERRORS) |
+                                        (ctx->fs->flags &
+                                         ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
+                       if (err || size != inode->i_size) {
+                               bad_size = 7;
+                               pctx->num = size;
+                       }
+               } else if (inode->i_size & (fs->blocksize - 1))
                        bad_size = 5;
                else if (nblock > (pb.last_block + 1))
                        bad_size = 1;
@@ -2153,23 +3094,33 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        }
        /* i_size for symlinks is checked elsewhere */
        if (bad_size && !LINUX_S_ISLNK(inode->i_mode)) {
-               pctx->num = (pb.last_block+1) * fs->blocksize;
+               /* Did inline_data set pctx->num earlier? */
+               if (bad_size != 7)
+                       pctx->num = (pb.last_block + 1) * fs->blocksize;
                pctx->group = bad_size;
                if (fix_problem(ctx, PR_1_BAD_I_SIZE, pctx)) {
-                       inode->i_size = pctx->num;
-                       if (!LINUX_S_ISDIR(inode->i_mode))
-                               inode->i_size_high = pctx->num >> 32;
+                       if (LINUX_S_ISDIR(inode->i_mode))
+                               pctx->num &= 0xFFFFFFFFULL;
+                       ext2fs_inode_size_set(fs, inode, pctx->num);
+                       if (EXT2_I_SIZE(inode) == 0 &&
+                           (inode->i_flags & EXT4_INLINE_DATA_FL)) {
+                               memset(inode->i_block, 0,
+                                      sizeof(inode->i_block));
+                               inode->i_flags &= ~EXT4_INLINE_DATA_FL;
+                       }
                        dirty_inode++;
                }
                pctx->num = 0;
        }
-       if (LINUX_S_ISREG(inode->i_mode) && EXT2_I_SIZE(inode) >= 0x80000000UL)
+       if (LINUX_S_ISREG(inode->i_mode) &&
+           ext2fs_needs_large_file_feature(EXT2_I_SIZE(inode)))
                ctx->large_files++;
-       if ((pb.num_blocks != ext2fs_inode_i_blocks(fs, inode)) ||
-           ((fs->super->s_feature_ro_compat &
-             EXT4_FEATURE_RO_COMPAT_HUGE_FILE) &&
-            (inode->i_flags & EXT4_HUGE_FILE_FL) &&
-            (inode->osd2.linux2.l_i_blocks_hi != 0))) {
+       if ((fs->super->s_creator_os == EXT2_OS_LINUX) &&
+           ((pb.num_blocks != ext2fs_inode_i_blocks(fs, inode)) ||
+            ((fs->super->s_feature_ro_compat &
+              EXT4_FEATURE_RO_COMPAT_HUGE_FILE) &&
+             (inode->i_flags & EXT4_HUGE_FILE_FL) &&
+             (inode->osd2.linux2.l_i_blocks_hi != 0)))) {
                pctx->num = pb.num_blocks;
                if (fix_problem(ctx, PR_1_BAD_I_BLOCKS, pctx)) {
                        inode->i_blocks = pb.num_blocks;
@@ -2180,9 +3131,10 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        }
 
        if (ctx->dirs_to_hash && pb.is_dir &&
+           !(ctx->lost_and_found && ctx->lost_and_found == ino) &&
            !(inode->i_flags & EXT2_INDEX_FL) &&
            ((inode->i_size / fs->blocksize) >= 3))
-               ext2fs_u32_list_add(ctx->dirs_to_hash, ino);
+               e2fsck_rehash_dir_later(ctx, ino);
 
 out:
        if (dirty_inode)
@@ -2255,33 +3207,26 @@ static int process_block(ext2_filsys fs,
        struct problem_context *pctx;
        blk64_t blk = *block_nr;
        int     ret_code = 0;
-       int     problem = 0;
+       problem_t       problem = 0;
        e2fsck_t        ctx;
 
        p = (struct process_block_struct *) priv_data;
        pctx = p->pctx;
        ctx = p->ctx;
 
-       if (p->compressed && (blk == EXT2FS_COMPRESSED_BLKADDR)) {
-               /* todo: Check that the comprblk_fl is high, that the
-                  blkaddr pattern looks right (all non-holes up to
-                  first EXT2FS_COMPRESSED_BLKADDR, then all
-                  EXT2FS_COMPRESSED_BLKADDR up to end of cluster),
-                  that the feature_incompat bit is high, and that the
-                  inode is a regular file.  If we're doing a "full
-                  check" (a concept introduced to e2fsck by e2compr,
-                  meaning that we look at data blocks as well as
-                  metadata) then call some library routine that
-                  checks the compressed data.  I'll have to think
-                  about this, because one particularly important
-                  problem to be able to fix is to recalculate the
-                  cluster size if necessary.  I think that perhaps
-                  we'd better do most/all e2compr-specific checks
-                  separately, after the non-e2compr checks.  If not
-                  doing a full check, it may be useful to test that
-                  the personality is linux; e.g. if it isn't then
-                  perhaps this really is just an illegal block. */
-               return 0;
+       /*
+        * For a directory, add logical block zero for processing even if it's
+        * not mapped or we'll be perennially stuck with broken "." and ".."
+        * entries.
+        */
+       if (p->is_dir && blockcnt == 0 && blk == 0) {
+               pctx->errcode = ext2fs_add_dir_block2(fs->dblist, p->ino, 0, 0);
+               if (pctx->errcode) {
+                       pctx->blk = blk;
+                       pctx->num = blockcnt;
+                       goto failed_add_dir_block;
+               }
+               p->last_db_block++;
        }
 
        if (blk == 0)
@@ -2297,7 +3242,7 @@ static int process_block(ext2_filsys fs,
         * file be contiguous.  (Which can never be true for really
         * big files that are greater than a block group.)
         */
-       if (!HOLE_BLKADDR(p->previous_block) && p->ino != EXT2_RESIZE_INO) {
+       if (p->previous_block && p->ino != EXT2_RESIZE_INO) {
                if (p->previous_block+1 != blk) {
                        if (ctx->options & E2F_OPT_FRAGCHECK) {
                                char type = '?';
@@ -2329,8 +3274,44 @@ static int process_block(ext2_filsys fs,
            blk >= ext2fs_blocks_count(fs->super))
                problem = PR_1_ILLEGAL_BLOCK_NUM;
 
+       /*
+        * If this IND/DIND/TIND block is squatting atop some critical metadata
+        * (group descriptors, superblock, bitmap, inode table), any write to
+        * "fix" mapping problems will destroy the metadata.  We'll let pass 1b
+        * fix that and restart fsck.
+        */
+       if (blockcnt < 0 &&
+           p->ino != EXT2_RESIZE_INO &&
+           ext2fs_test_block_bitmap2(ctx->block_metadata_map, blk)) {
+               pctx->blk = blk;
+               fix_problem(ctx, PR_1_CRITICAL_METADATA_COLLISION, pctx);
+               ctx->flags |= E2F_FLAG_RESTART_LATER;
+       }
+
        if (problem) {
                p->num_illegal_blocks++;
+               /*
+                * A bit of subterfuge here -- we're trying to fix a block
+                * mapping, but the IND/DIND/TIND block could have collided
+                * with some critical metadata.  So, fix the in-core mapping so
+                * iterate won't go insane, but return 0 instead of
+                * BLOCK_CHANGED so that it won't write the remapping out to
+                * our multiply linked block.
+                *
+                * Even if we previously determined that an *IND block
+                * conflicts with critical metadata, we must still try to
+                * iterate the *IND block as if it is an *IND block to find and
+                * mark the blocks it points to.  Better to be overly cautious
+                * with the used_blocks map so that we don't move the *IND
+                * block to a block that's really in use!
+                */
+               if (p->ino != EXT2_RESIZE_INO &&
+                   ref_block != 0 &&
+                   ext2fs_test_block_bitmap2(ctx->block_metadata_map,
+                                             ref_block)) {
+                       *block_nr = 0;
+                       return 0;
+               }
                if (!p->suppress && (p->num_illegal_blocks % 12) == 0) {
                        if (fix_problem(ctx, PR_1_TOO_MANY_BAD_BLOCKS, pctx)) {
                                p->clear = 1;
@@ -2347,6 +3328,18 @@ static int process_block(ext2_filsys fs,
                if (fix_problem(ctx, problem, pctx)) {
                        blk = *block_nr = 0;
                        ret_code = BLOCK_CHANGED;
+                       p->inode_modified = 1;
+                       /*
+                        * If the directory block is too big and is beyond the
+                        * end of the FS, don't bother trying to add it for
+                        * processing -- the kernel would never have created a
+                        * directory this large, and we risk an ENOMEM abort.
+                        * In any case, the toobig handler for extent-based
+                        * directories also doesn't feed toobig blocks to
+                        * pass 2.
+                        */
+                       if (problem == PR_1_TOOBIG_DIR)
+                               return ret_code;
                        goto mark_dir;
                } else
                        return 0;
@@ -2371,6 +3364,13 @@ static int process_block(ext2_filsys fs,
                     ((unsigned) blockcnt & EXT2FS_CLUSTER_MASK(ctx->fs)))) {
                mark_block_used(ctx, blk);
                p->num_blocks++;
+       } else if (has_unaligned_cluster_map(ctx, p->previous_block,
+                                            p->last_block, blk, blockcnt)) {
+               pctx->blk = blockcnt;
+               pctx->blk2 = blk;
+               fix_problem(ctx, PR_1_MISALIGNED_CLUSTER, pctx);
+               mark_block_used(ctx, blk);
+               mark_block_used(ctx, blk);
        }
        if (blockcnt >= 0)
                p->last_block = blockcnt;
@@ -2416,11 +3416,6 @@ static int process_bad_block(ext2_filsys fs,
        struct problem_context *pctx;
        e2fsck_t        ctx;
 
-       /*
-        * Note: This function processes blocks for the bad blocks
-        * inode, which is never compressed.  So we don't use HOLE_BLKADDR().
-        */
-
        if (!blk)
                return 0;
 
@@ -2566,14 +3561,16 @@ static int process_bad_block(ext2_filsys fs,
        return 0;
 }
 
-static void new_table_block(e2fsck_t ctx, blk_t first_block, int group,
+static void new_table_block(e2fsck_t ctx, blk64_t first_block, dgrp_t group,
                            const char *name, int num, blk64_t *new_block)
 {
        ext2_filsys fs = ctx->fs;
        dgrp_t          last_grp;
        blk64_t         old_block = *new_block;
        blk64_t         last_block;
-       int             i, is_flexbg, flexbg, flexbg_size;
+       dgrp_t          flexbg;
+       unsigned        flexbg_size;
+       int             i, is_flexbg;
        char            *buf;
        struct problem_context  pctx;
 
@@ -2596,8 +3593,8 @@ static void new_table_block(e2fsck_t ctx, blk_t first_block, int group,
                first_block = ext2fs_group_first_block2(fs,
                                                        flexbg_size * flexbg);
                last_grp = group | (flexbg_size - 1);
-               if (last_grp > fs->group_desc_count)
-                       last_grp = fs->group_desc_count;
+               if (last_grp >= fs->group_desc_count)
+                       last_grp = fs->group_desc_count - 1;
                last_block = ext2fs_group_last_block2(fs, last_grp);
        } else
                last_block = ext2fs_group_last_block2(fs, group);
@@ -2637,12 +3634,15 @@ static void new_table_block(e2fsck_t ctx, blk_t first_block, int group,
                                   old_block + i, 1, buf);
                        if (pctx.errcode)
                                fix_problem(ctx, PR_1_RELOC_READ_ERR, &pctx);
-               } else
-                       memset(buf, 0, fs->blocksize);
+                       pctx.blk = (*new_block) + i;
+                       pctx.errcode = io_channel_write_blk64(fs->io, pctx.blk,
+                                                             1, buf);
+               } else {
+                       pctx.blk = (*new_block) + i;
+                       pctx.errcode = ext2fs_zero_blocks2(fs, pctx.blk, 1,
+                                                          NULL, NULL);
+               }
 
-               pctx.blk = (*new_block) + i;
-               pctx.errcode = io_channel_write_blk64(fs->io, pctx.blk,
-                                             1, buf);
                if (pctx.errcode)
                        fix_problem(ctx, PR_1_RELOC_WRITE_ERR, &pctx);
        }
@@ -2708,6 +3708,7 @@ static void mark_table_blocks(e2fsck_t ctx)
                pctx.group = i;
 
                ext2fs_reserve_super_and_bgd(fs, i, ctx->block_found_map);
+               ext2fs_reserve_super_and_bgd(fs, i, ctx->block_metadata_map);
 
                /*
                 * Mark the blocks used for the inode table
@@ -2726,8 +3727,10 @@ static void mark_table_blocks(e2fsck_t ctx)
                                                ctx->invalid_bitmaps++;
                                        }
                                } else {
-                                   ext2fs_mark_block_bitmap2(ctx->block_found_map,
-                                                            b);
+                                   ext2fs_mark_block_bitmap2(
+                                               ctx->block_found_map, b);
+                                   ext2fs_mark_block_bitmap2(
+                                               ctx->block_metadata_map, b);
                                }
                        }
                }
@@ -2746,8 +3749,9 @@ static void mark_table_blocks(e2fsck_t ctx)
                        } else {
                            ext2fs_mark_block_bitmap2(ctx->block_found_map,
                                     ext2fs_block_bitmap_loc(fs, i));
-                   }
-
+                           ext2fs_mark_block_bitmap2(ctx->block_metadata_map,
+                                    ext2fs_block_bitmap_loc(fs, i));
+                       }
                }
                /*
                 * Mark block used for the inode bitmap
@@ -2761,6 +3765,8 @@ static void mark_table_blocks(e2fsck_t ctx)
                                        ctx->invalid_bitmaps++;
                                }
                        } else {
+                           ext2fs_mark_block_bitmap2(ctx->block_metadata_map,
+                                    ext2fs_inode_bitmap_loc(fs, i));
                            ext2fs_mark_block_bitmap2(ctx->block_found_map,
                                     ext2fs_inode_bitmap_loc(fs, i));
                        }
@@ -2845,7 +3851,7 @@ static errcode_t e2fsck_get_alloc_block(ext2_filsys fs, blk64_t goal,
                                return retval;
                }
 
-               retval = ext2fs_new_block2(fs, goal, 0, &new_block);
+               retval = ext2fs_new_block2(fs, goal, fs->block_map, &new_block);
                if (retval)
                        return retval;
        }
@@ -2858,6 +3864,13 @@ static void e2fsck_block_alloc_stats(ext2_filsys fs, blk64_t blk, int inuse)
 {
        e2fsck_t ctx = (e2fsck_t) fs->priv_data;
 
+       /* Never free a critical metadata block */
+       if (ctx->block_found_map &&
+           ctx->block_metadata_map &&
+           inuse < 0 &&
+           ext2fs_test_block_bitmap2(ctx->block_metadata_map, blk))
+               return;
+
        if (ctx->block_found_map) {
                if (inuse > 0)
                        ext2fs_mark_block_bitmap2(ctx->block_found_map, blk);
@@ -2876,11 +3889,6 @@ void e2fsck_use_inode_shortcuts(e2fsck_t ctx, int use_shortcuts)
                fs->read_inode = pass1_read_inode;
                fs->write_inode = pass1_write_inode;
                ctx->stashed_ino = 0;
-               ext2fs_set_alloc_block_callback(fs, e2fsck_get_alloc_block,
-                                               0);
-               ext2fs_set_block_alloc_stats_callback(fs,
-                                                     e2fsck_block_alloc_stats,
-                                                     0);
        } else {
                fs->get_blocks = 0;
                fs->check_directory = 0;
@@ -2888,3 +3896,10 @@ void e2fsck_use_inode_shortcuts(e2fsck_t ctx, int use_shortcuts)
                fs->write_inode = 0;
        }
 }
+
+void e2fsck_intercept_block_allocations(e2fsck_t ctx)
+{
+       ext2fs_set_alloc_block_callback(ctx->fs, e2fsck_get_alloc_block, 0);
+       ext2fs_set_block_alloc_stats_callback(ctx->fs,
+                                               e2fsck_block_alloc_stats, 0);
+}