Whamcloud - gitweb
e2fsck: improve error messages when the journal sb is corrupt
[tools/e2fsprogs.git] / e2fsck / pass1.c
index faafa7c..799158e 100644 (file)
@@ -56,6 +56,8 @@
 #define _INLINE_ inline
 #endif
 
+#undef DEBUG
+
 static int process_block(ext2_filsys fs, blk64_t       *blocknr,
                         e2_blkcnt_t blockcnt, blk64_t ref_blk,
                         int ref_offset, void *priv_data);
@@ -68,6 +70,7 @@ 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);
@@ -93,13 +96,13 @@ struct process_block_struct {
        struct problem_context *pctx;
        ext2fs_block_bitmap fs_meta_blocks;
        e2fsck_t        ctx;
-       blk64_t         bad_ref;
        region_t        region;
+       struct extent_tree_info eti;
 };
 
 struct process_inode_block {
        ext2_ino_t ino;
-       struct ext2_inode inode;
+       struct ext2_inode_large inode;
 };
 
 struct scan_callback_struct {
@@ -235,7 +238,11 @@ 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) {
@@ -268,7 +275,8 @@ exit_inline:
                        return 0;
        }
        if (len != inode->i_size)
-               return 0;
+               if ((inode->i_flags & EXT4_ENCRYPT_FL) == 0)
+                       return 0;
        return 1;
 }
 
@@ -440,6 +448,21 @@ fix:
                                EXT2_INODE_SIZE(sb), "pass1");
 }
 
+static int check_inode_extra_negative_epoch(__u32 xtime, __u32 extra) {
+       return (xtime & (1 << 31)) != 0 &&
+               (extra & EXT4_EPOCH_MASK) == EXT4_EPOCH_MASK;
+}
+
+#define CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, xtime) \
+       check_inode_extra_negative_epoch(inode->i_##xtime, \
+                                        inode->i_##xtime##_extra)
+
+/* When today's date is earlier than 2242, we assume that atimes,
+ * ctimes, crtimes, and mtimes with years in the range 2310..2378 are
+ * actually pre-1970 dates mis-encoded.
+ */
+#define EXT4_EXTRA_NEGATIVE_DATE_CUTOFF 2 * (1LL << 32)
+
 static void check_inode_extra_space(e2fsck_t ctx, struct problem_context *pctx)
 {
        struct ext2_super_block *sb = ctx->fs->super;
@@ -474,12 +497,42 @@ static void check_inode_extra_space(e2fsck_t ctx, struct problem_context *pctx)
                return;
        }
 
+       /* check if there is no place for an EA header */
+       if (inode->i_extra_isize >= max - sizeof(__u32))
+               return;
+
        eamagic = (__u32 *) (((char *) inode) + EXT2_GOOD_OLD_INODE_SIZE +
                        inode->i_extra_isize);
        if (*eamagic == EXT2_EXT_ATTR_MAGIC) {
                /* it seems inode has an extended attribute(s) in body */
                check_ea_in_inode(ctx, pctx);
        }
+
+       /*
+        * If the inode's extended atime (ctime, crtime, mtime) is stored in
+        * the old, invalid format, repair it.
+        */
+       if (sizeof(time_t) > 4 && ctx->now < EXT4_EXTRA_NEGATIVE_DATE_CUTOFF &&
+           (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, atime) ||
+            CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, ctime) ||
+            CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, crtime) ||
+            CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, mtime))) {
+
+               if (!fix_problem(ctx, PR_1_EA_TIME_OUT_OF_RANGE, pctx))
+                       return;
+
+               if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, atime))
+                       inode->i_atime_extra &= ~EXT4_EPOCH_MASK;
+               if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, ctime))
+                       inode->i_ctime_extra &= ~EXT4_EPOCH_MASK;
+               if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, crtime))
+                       inode->i_crtime_extra &= ~EXT4_EPOCH_MASK;
+               if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, mtime))
+                       inode->i_mtime_extra &= ~EXT4_EPOCH_MASK;
+               e2fsck_write_inode_full(ctx, pctx->ino, pctx->inode,
+                                       EXT2_INODE_SIZE(sb), "pass1");
+       }
+
 }
 
 /*
@@ -532,12 +585,13 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
         * data.  If it's true, we will treat it as a directory.
         */
 
-       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);
+       extent_fs = ext2fs_has_feature_extents(ctx->fs->super);
+       inlinedata_fs = ext2fs_has_feature_inline_data(ctx->fs->super);
        if (inlinedata_fs && (inode->i_flags & EXT4_INLINE_DATA_FL)) {
                size_t size;
+               __u32 dotdot;
+               unsigned int rec_len2;
+               struct ext2_dir_entry de;
 
                if (ext2fs_inline_data_size(ctx->fs, pctx->ino, &size))
                        return;
@@ -547,6 +601,26 @@ static void check_is_really_dir(e2fsck_t ctx, struct problem_context *pctx,
                 */
                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_len2);
+               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_len2 > 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)) {
@@ -799,10 +873,8 @@ static int fix_inline_data_extents_file(e2fsck_t ctx,
        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))
+       if (!ext2fs_has_feature_extents(fs->super) &&
+           !ext2fs_has_feature_inline_data(fs->super))
                return 0;
 
        /* Clear both flags if it's a special file */
@@ -824,8 +896,8 @@ static int fix_inline_data_extents_file(e2fsck_t ctx,
        }
 
        /* 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) -
+       if (inode_size > EXT2_GOOD_OLD_INODE_SIZE)
+               max_inline_ea_size = inode_size -
                                     (EXT2_GOOD_OLD_INODE_SIZE +
                                      ((struct ext2_inode_large *)inode)->i_extra_isize);
        else
@@ -869,6 +941,95 @@ out:
        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);
+       }
+}
+
+/*
+ * Check if the passed ino is one of the used superblock quota inodes.
+ *
+ * Before the quota inodes were journaled, older superblock quota inodes
+ * were just regular files in the filesystem and not reserved inodes.  This
+ * checks if the passed ino is one of the s_*_quota_inum superblock fields,
+ * which may not always be the same as the EXT4_*_QUOTA_INO fields.
+ */
+static int quota_inum_is_super(struct ext2_super_block *sb, ext2_ino_t ino)
+{
+       enum quota_type qtype;
+
+       for (qtype = 0; qtype < MAXQUOTAS; qtype++)
+               if (*quota_sb_inump(sb, qtype) == ino)
+                       return 1;
+
+       return 0;
+}
+
+/*
+ * Check if the passed ino is one of the reserved quota inodes.
+ * This checks if the inode number is one of the reserved EXT4_*_QUOTA_INO
+ * inodes.  These inodes may or may not be in use by the quota feature.
+ */
+static int quota_inum_is_reserved(ext2_filsys fs, ext2_ino_t ino)
+{
+       enum quota_type qtype;
+
+       for (qtype = 0; qtype < MAXQUOTAS; qtype++)
+               if (quota_type2inum(qtype, fs->super) == ino)
+                       return 1;
+
+       return 0;
+}
+
 void e2fsck_pass1(e2fsck_t ctx)
 {
        int     i;
@@ -889,16 +1050,25 @@ void e2fsck_pass1(e2fsck_t ctx)
        unsigned int    save_type;
        int             imagic_fs, extent_fs, inlinedata_fs;
        int             low_dtime_check = 1;
-       int             inode_size;
+       int             inode_size = EXT2_INODE_SIZE(fs->super);
        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);
 
-       if ((fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) &&
+       if (ext2fs_has_feature_dir_index(fs->super) &&
            !(ctx->options & E2F_OPT_NO)) {
                if (ext2fs_u32_list_create(&ctx->dirs_to_hash, 50))
                        ctx->dirs_to_hash = 0;
@@ -919,10 +1089,9 @@ void e2fsck_pass1(e2fsck_t ctx)
        }
 #undef EXT2_BPP
 
-       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);
+       imagic_fs = ext2fs_has_feature_imagic_inodes(sb);
+       extent_fs = ext2fs_has_feature_extents(sb);
+       inlinedata_fs = ext2fs_has_feature_inline_data(sb);
 
        /*
         * Allocate bitmaps structures
@@ -988,7 +1157,6 @@ void e2fsck_pass1(e2fsck_t ctx)
                ctx->flags |= E2F_FLAG_ABORT;
                return;
        }
-       inode_size = EXT2_INODE_SIZE(fs->super);
        inode = (struct ext2_inode *)
                e2fsck_allocate_memory(ctx, inode_size, "scratch inode");
 
@@ -1057,7 +1225,7 @@ void e2fsck_pass1(e2fsck_t ctx)
             fs->super->s_mkfs_time < fs->super->s_inodes_count))
                low_dtime_check = 0;
 
-       if ((fs->super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_MMP) &&
+       if (ext2fs_has_feature_mmp(fs->super) &&
            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,
@@ -1074,9 +1242,11 @@ 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;
+                       goto endit;
                if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE) {
                        /*
                         * If badblocks says badblocks is bad, offer to clear
@@ -1167,12 +1337,11 @@ void e2fsck_pass1(e2fsck_t ctx)
 
                        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;
+                           fix_problem(ctx, PR_1_INLINE_DATA_FEATURE, &pctx)) {
+                               ext2fs_set_feature_inline_data(sb);
                                ext2fs_mark_super_dirty(fs);
                                inlinedata_fs = 1;
-                       } else if (!fix_problem(ctx, PR_1_INLINE_DATA_SET, &pctx)) {
+                       } else if (fix_problem(ctx, PR_1_INLINE_DATA_SET, &pctx)) {
                                e2fsck_clear_inode(ctx, ino, inode, 0, "pass1");
                                /* skip FINISH_INODE_LOOP */
                                continue;
@@ -1222,6 +1391,8 @@ void e2fsck_pass1(e2fsck_t ctx)
                                                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;
@@ -1256,7 +1427,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                        if ((ext2fs_extent_header_verify(inode->i_block,
                                                 sizeof(inode->i_block)) == 0) &&
                            fix_problem(ctx, PR_1_EXTENT_FEATURE, &pctx)) {
-                               sb->s_feature_incompat |= EXT3_FEATURE_INCOMPAT_EXTENTS;
+                               ext2fs_set_feature_extents(sb);
                                ext2fs_mark_super_dirty(fs);
                                extent_fs = 1;
                        } else if (fix_problem(ctx, PR_1_EXTENTS_SET, &pctx)) {
@@ -1407,13 +1578,10 @@ void e2fsck_pass1(e2fsck_t ctx)
                                                        inode_size, "pass1");
                                failed_csum = 0;
                        }
-               } else if ((ino == EXT4_USR_QUOTA_INO) ||
-                          (ino == EXT4_GRP_QUOTA_INO)) {
+               } else if (quota_inum_is_reserved(fs, ino)) {
                        ext2fs_mark_inode_bitmap2(ctx->inode_used_map, ino);
-                       if ((fs->super->s_feature_ro_compat &
-                                       EXT4_FEATURE_RO_COMPAT_QUOTA) &&
-                           ((fs->super->s_usr_quota_inum == ino) ||
-                            (fs->super->s_grp_quota_inum == ino))) {
+                       if (ext2fs_has_feature_quota(fs->super) &&
+                           quota_inum_is_super(fs->super, ino)) {
                                if (!LINUX_S_ISREG(inode->i_mode) &&
                                    fix_problem(ctx, PR_1_QUOTA_BAD_MODE,
                                                        &pctx)) {
@@ -1541,13 +1709,12 @@ 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 & 
-                     EXT4_FEATURE_INCOMPAT_64BIT) &&
+               if ((fs->super->s_creator_os == EXT2_OS_LINUX) &&
+                   !ext2fs_has_feature_64bit(fs->super) &&
                    inode->osd2.linux2.l_i_file_acl_high != 0)
                        mark_inode_bad(ctx, ino);
                if ((fs->super->s_creator_os == EXT2_OS_LINUX) &&
-                   !(fs->super->s_feature_ro_compat &
-                     EXT4_FEATURE_RO_COMPAT_HUGE_FILE) &&
+                   !ext2fs_has_feature_huge_file(fs->super) &&
                    (inode->osd2.linux2.l_i_blocks_hi != 0))
                        mark_inode_bad(ctx, ino);
                if (inode->i_flags & EXT2_IMAGIC_FL) {
@@ -1586,6 +1753,8 @@ void e2fsck_pass1(e2fsck_t ctx)
                        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++;
@@ -1646,7 +1815,8 @@ void e2fsck_pass1(e2fsck_t ctx)
                     inode->i_block[EXT2_TIND_BLOCK] ||
                     ext2fs_file_acl_block(fs, inode))) {
                        inodes_to_process[process_inode_count].ino = ino;
-                       inodes_to_process[process_inode_count].inode = *inode;
+                       inodes_to_process[process_inode_count].inode =
+                                      *(struct ext2_inode_large *)inode;
                        process_inode_count++;
                } else
                        check_blocks(ctx, &pctx, block_buf);
@@ -1736,6 +1906,7 @@ void e2fsck_pass1(e2fsck_t ctx)
                }
                e2fsck_pass1_dupblocks(ctx, block_buf);
        }
+       ctx->flags |= E2F_FLAG_ALLOC_OK;
        ext2fs_free_mem(&inodes_to_process);
 endit:
        e2fsck_use_inode_shortcuts(ctx, 0);
@@ -1755,6 +1926,8 @@ endit:
 
        if ((ctx->flags & E2F_FLAG_SIGNAL_MASK) == 0)
                print_resource_track(ctx, _("Pass 1"), &rtrack, ctx->fs->io);
+       else
+               ctx->invalid_bitmaps++;
 }
 #undef FINISH_INODE_LOOP
 
@@ -1806,7 +1979,8 @@ static void process_inodes(e2fsck_t ctx, char *block_buf)
                      sizeof(struct process_inode_block), process_inode_cmp);
        clear_problem_context(&pctx);
        for (i=0; i < process_inode_count; i++) {
-               pctx.inode = ctx->stashed_inode = &inodes_to_process[i].inode;
+               pctx.inode = ctx->stashed_inode =
+                       (struct ext2_inode *) &inodes_to_process[i].inode;
                pctx.ino = ctx->stashed_ino = inodes_to_process[i].ino;
 
 #if 0
@@ -1844,8 +2018,8 @@ static EXT2_QSORT_TYPE process_inode_cmp(const void *a, const void *b)
                 * inodes, so it's OK to pass NULL to
                 * ext2fs_file_acl_block() here.
                 */
-               ret = ext2fs_file_acl_block(0, &(ib_a->inode)) -
-                       ext2fs_file_acl_block(0, &(ib_b->inode));
+               ret = ext2fs_file_acl_block(0, ext2fs_const_inode(&ib_a->inode)) -
+                       ext2fs_file_acl_block(0, ext2fs_const_inode(&ib_b->inode));
        if (ret == 0)
                ret = ib_a->ino - ib_b->ino;
        return ret;
@@ -1875,6 +2049,23 @@ 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
@@ -2036,7 +2227,7 @@ static int check_ext_attr(e2fsck_t ctx, struct problem_context *pctx,
         * Or if the extended attribute block is an invalid block,
         * then the inode is also corrupted.
         */
-       if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_EXT_ATTR) ||
+       if (!ext2fs_has_feature_xattr(fs->super) ||
            (blk < fs->super->s_first_data_block) ||
            (blk >= ext2fs_blocks_count(fs->super))) {
                mark_inode_bad(ctx, ino);
@@ -2230,7 +2421,7 @@ static int handle_htree(e2fsck_t ctx, struct problem_context *pctx,
 
        if ((!LINUX_S_ISDIR(inode->i_mode) &&
             fix_problem(ctx, PR_1_HTREE_NODIR, pctx)) ||
-           (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) &&
+           (!ext2fs_has_feature_dir_index(fs->super) &&
             fix_problem(ctx, PR_1_HTREE_SET, pctx)))
                return 1;
 
@@ -2370,6 +2561,23 @@ static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
        pctx->errcode = ext2fs_extent_get_info(ehandle, &info);
        if (pctx->errcode)
                return;
+       if (!(ctx->options & E2F_OPT_FIXES_ONLY) &&
+           !pb->eti.force_rebuild) {
+               struct extent_tree_level *etl;
+
+               etl = pb->eti.ext_info + info.curr_level;
+               etl->num_extents += info.num_entries;
+               etl->max_extents += info.max_entries;
+               /*
+                * Implementation wart: Splitting extent blocks when appending
+                * will leave the old block with one free entry.  Therefore
+                * unless the node is totally full, pretend that a non-root
+                * extent block can hold one fewer entry than it actually does,
+                * so that we don't repeatedly rebuild the extent tree.
+                */
+               if (info.curr_level && info.num_entries < info.max_entries)
+                       etl->max_extents--;
+       }
 
        pctx->errcode = ext2fs_extent_get(ehandle, EXT2_EXTENT_FIRST_SIB,
                                          &extent);
@@ -2694,7 +2902,20 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
        ext2_ino_t              ino = pctx->ino;
        errcode_t               retval;
        blk64_t                 eof_lblk;
+       struct ext3_extent_header       *eh;
+
+       /* Check for a proper extent header... */
+       eh = (struct ext3_extent_header *) &inode->i_block[0];
+       retval = ext2fs_extent_header_verify(eh, sizeof(inode->i_block));
+       if (retval) {
+               if (fix_problem(ctx, PR_1_MISSING_EXTENT_HEADER, pctx))
+                       e2fsck_clear_inode(ctx, ino, inode, 0,
+                                          "check_blocks_extents");
+               pctx->errcode = 0;
+               return;
+       }
 
+       /* ...since this function doesn't fail if i_block is zeroed. */
        pctx->errcode = ext2fs_extent_open2(fs, ino, inode, &ehandle);
        if (pctx->errcode) {
                if (fix_problem(ctx, PR_1_READ_EXTENT, pctx))
@@ -2706,11 +2927,27 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
 
        retval = ext2fs_extent_get_info(ehandle, &info);
        if (retval == 0) {
-               if (info.max_depth >= MAX_EXTENT_DEPTH_COUNT)
-                       info.max_depth = MAX_EXTENT_DEPTH_COUNT-1;
-               ctx->extent_depth_count[info.max_depth]++;
+               int max_depth = info.max_depth;
+
+               if (max_depth >= MAX_EXTENT_DEPTH_COUNT)
+                       max_depth = MAX_EXTENT_DEPTH_COUNT-1;
+               ctx->extent_depth_count[max_depth]++;
        }
 
+       /* Check maximum extent depth */
+       pctx->blk = info.max_depth;
+       pctx->blk2 = ext2fs_max_extent_depth(ehandle);
+       if (pctx->blk2 < pctx->blk &&
+           fix_problem(ctx, PR_1_EXTENT_BAD_MAX_DEPTH, pctx))
+               pb->eti.force_rebuild = 1;
+
+       /* Can we collect extent tree level stats? */
+       pctx->blk = MAX_EXTENT_DEPTH_COUNT;
+       if (pctx->blk2 > pctx->blk)
+               fix_problem(ctx, PR_1E_MAX_EXTENT_TREE_DEPTH, pctx);
+       memset(pb->eti.ext_info, 0, sizeof(pb->eti.ext_info));
+       pb->eti.ino = pb->ino;
+
        pb->region = region_create(0, info.max_lblk);
        if (!pb->region) {
                ext2fs_extent_free(ehandle);
@@ -2733,6 +2970,16 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
        region_free(pb->region);
        pb->region = NULL;
        ext2fs_extent_free(ehandle);
+
+       /* Rebuild unless it's a dir and we're rehashing it */
+       if (LINUX_S_ISDIR(inode->i_mode) &&
+           e2fsck_dir_will_be_rehashed(ctx, ino))
+               return;
+
+       if (ctx->options & E2F_OPT_CONVERT_BMAP)
+               e2fsck_rebuild_extents_later(ctx, ino);
+       else
+               e2fsck_should_rebuild_extents(ctx, pctx, &pb->eti, &info);
 }
 
 /*
@@ -2743,18 +2990,44 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
 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->blk = 0;
-               pctx->num = 0;
-               fix_problem(ctx, PR_1_ADD_DBLOCK, pctx);
-               ctx->flags |= E2F_FLAG_ABORT;
+               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;
 }
 
 /*
@@ -2791,26 +3064,12 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        pb.pctx = pctx;
        pb.ctx = ctx;
        pb.inode_modified = 0;
-       pb.bad_ref = 0;
+       pb.eti.force_rebuild = 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++;
-                       }
-               }
-       }
+       extent_fs = ext2fs_has_feature_extents(ctx->fs->super);
+       inlinedata_fs = ext2fs_has_feature_inline_data(ctx->fs->super);
 
        if (check_ext_attr(ctx, pctx, block_buf)) {
                if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
@@ -2855,6 +3114,15 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                                                  "check_blocks");
                        fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
                                    (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
+
+                       if (ctx->options & E2F_OPT_CONVERT_BMAP) {
+#ifdef DEBUG
+                               printf("bmap rebuild ino=%d\n", ino);
+#endif
+                               if (!LINUX_S_ISDIR(inode->i_mode) ||
+                                   !e2fsck_dir_will_be_rehashed(ctx, ino))
+                                       e2fsck_rebuild_extents_later(ctx, ino);
+                       }
                }
        }
        end_problem_latch(ctx, PR_LATCH_BLOCK);
@@ -2882,9 +3150,7 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                        inode->i_flags &= ~EXT2_INDEX_FL;
                        dirty_inode++;
                } else {
-#ifdef ENABLE_HTREE
                        e2fsck_add_dx_dir(ctx, ino, pb.last_block+1);
-#endif
                }
        }
 
@@ -2897,14 +3163,15 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                }
        }
 
-       if (ino == EXT2_ROOT_INO || ino >= EXT2_FIRST_INODE(ctx->fs->super)) {
-               quota_data_add(ctx->qctx, inode, ino,
-                              pb.num_blocks * fs->blocksize);
-               quota_data_inodes(ctx->qctx, inode, ino, +1);
+       if (ino != quota_type2inum(PRJQUOTA, fs->super) &&
+           (ino == EXT2_ROOT_INO || ino >= EXT2_FIRST_INODE(ctx->fs->super))) {
+               quota_data_add(ctx->qctx, (struct ext2_inode_large *) inode,
+                              ino, pb.num_blocks * fs->blocksize);
+               quota_data_inodes(ctx->qctx, (struct ext2_inode_large *) inode,
+                                 ino, +1);
        }
 
-       if (!(fs->super->s_feature_ro_compat &
-             EXT4_FEATURE_RO_COMPAT_HUGE_FILE) ||
+       if (!ext2fs_has_feature_huge_file(fs->super) ||
            !(inode->i_flags & EXT4_HUGE_FILE_FL))
                pb.num_blocks *= (fs->blocksize / 512);
        pb.num_blocks *= EXT2FS_CLUSTER_RATIO(fs);
@@ -2986,11 +3253,11 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
        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)) ||
+            (ext2fs_has_feature_huge_file(fs->super) &&
+             (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;
@@ -3000,6 +3267,22 @@ static void check_blocks(e2fsck_t ctx, struct problem_context *pctx,
                pctx->num = 0;
        }
 
+       /*
+        * The kernel gets mad if we ask it to allocate bigalloc clusters to
+        * a block mapped file, so rebuild it as an extent file.  We can skip
+        * symlinks because they're never rewritten.
+        */
+       if (ext2fs_has_feature_bigalloc(fs->super) &&
+           (LINUX_S_ISREG(inode->i_mode) || LINUX_S_ISDIR(inode->i_mode)) &&
+           ext2fs_inode_data_blocks2(fs, inode) > 0 &&
+           (ino == EXT2_ROOT_INO || ino >= EXT2_FIRST_INO(fs->super)) &&
+           !(inode->i_flags & (EXT4_EXTENTS_FL | EXT4_INLINE_DATA_FL)) &&
+           fix_problem(ctx, PR_1_NO_BIGALLOC_BLOCKMAP_FILES, pctx)) {
+               pctx->errcode = e2fsck_rebuild_extents_later(ctx, ino);
+               if (pctx->errcode)
+                       goto out;
+       }
+
        if (ctx->dirs_to_hash && pb.is_dir &&
            !(ctx->lost_and_found && ctx->lost_and_found == ino) &&
            !(inode->i_flags & EXT2_INDEX_FL) &&
@@ -3084,28 +3367,6 @@ static int process_block(ext2_filsys fs,
        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 ".."
@@ -3134,7 +3395,7 @@ 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 = '?';
@@ -3175,7 +3436,6 @@ static int process_block(ext2_filsys fs,
        if (blockcnt < 0 &&
            p->ino != EXT2_RESIZE_INO &&
            ext2fs_test_block_bitmap2(ctx->block_metadata_map, blk)) {
-               p->bad_ref = blk;
                pctx->blk = blk;
                fix_problem(ctx, PR_1_CRITICAL_METADATA_COLLISION, pctx);
                ctx->flags |= E2F_FLAG_RESTART_LATER;
@@ -3185,13 +3445,23 @@ static int process_block(ext2_filsys fs,
                p->num_illegal_blocks++;
                /*
                 * A bit of subterfuge here -- we're trying to fix a block
-                * mapping, but know that the IND/DIND/TIND block has collided
+                * 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->bad_ref && ref_block == p->bad_ref) {
+               if (p->ino != EXT2_RESIZE_INO &&
+                   ref_block != 0 &&
+                   ext2fs_test_block_bitmap2(ctx->block_metadata_map,
+                                             ref_block)) {
                        *block_nr = 0;
                        return 0;
                }
@@ -3299,11 +3569,6 @@ 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;
 
@@ -3473,8 +3738,7 @@ static void new_table_block(e2fsck_t ctx, blk64_t first_block, dgrp_t group,
         * within the flex_bg, and if that fails then try finding the
         * space anywhere in the filesystem.
         */
-       is_flexbg = EXT2_HAS_INCOMPAT_FEATURE(fs->super,
-                                             EXT4_FEATURE_INCOMPAT_FLEX_BG);
+       is_flexbg = ext2fs_has_feature_flex_bg(fs->super);
        if (is_flexbg) {
                flexbg_size = 1 << fs->super->s_log_groups_per_flex;
                flexbg = group / flexbg_size;
@@ -3739,7 +4003,7 @@ 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;
        }
@@ -3748,6 +4012,26 @@ static errcode_t e2fsck_get_alloc_block(ext2_filsys fs, blk64_t goal,
        return (0);
 }
 
+static errcode_t e2fsck_new_range(ext2_filsys fs, int flags, blk64_t goal,
+                                 blk64_t len, blk64_t *pblk, blk64_t *plen)
+{
+       e2fsck_t ctx = (e2fsck_t) fs->priv_data;
+       errcode_t       retval;
+
+       if (ctx->block_found_map)
+               return ext2fs_new_range(fs, flags, goal, len,
+                                       ctx->block_found_map, pblk, plen);
+
+       if (!fs->block_map) {
+               retval = ext2fs_read_block_bitmap(fs);
+               if (retval)
+                       return retval;
+       }
+
+       return ext2fs_new_range(fs, flags, goal, len, fs->block_map,
+                               pblk, plen);
+}
+
 static void e2fsck_block_alloc_stats(ext2_filsys fs, blk64_t blk, int inuse)
 {
        e2fsck_t ctx = (e2fsck_t) fs->priv_data;
@@ -3767,6 +4051,28 @@ static void e2fsck_block_alloc_stats(ext2_filsys fs, blk64_t blk, int inuse)
        }
 }
 
+static void e2fsck_block_alloc_stats_range(ext2_filsys fs, blk64_t blk,
+                                          blk_t num, 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_bitmap_range2(ctx->block_metadata_map, blk, num))
+               return;
+
+       if (ctx->block_found_map) {
+               if (inuse > 0)
+                       ext2fs_mark_block_bitmap_range2(ctx->block_found_map,
+                                                       blk, num);
+               else
+                       ext2fs_unmark_block_bitmap_range2(ctx->block_found_map,
+                                                       blk, num);
+       }
+}
+
 void e2fsck_use_inode_shortcuts(e2fsck_t ctx, int use_shortcuts)
 {
        ext2_filsys fs = ctx->fs;
@@ -3790,4 +4096,7 @@ 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);
+       ext2fs_set_new_range_callback(ctx->fs, e2fsck_new_range, NULL);
+       ext2fs_set_block_alloc_stats_range_callback(ctx->fs,
+                                       e2fsck_block_alloc_stats_range, NULL);
 }