Whamcloud - gitweb
Merge branch 'maint' into next
[tools/e2fsprogs.git] / e2fsck / rehash.c
index 0d218e8..8cc36f2 100644 (file)
@@ -51,6 +51,7 @@
 #include <errno.h>
 #include "e2fsck.h"
 #include "problem.h"
+#include "support/sort_r.h"
 
 /* Schedule a dir to be rebuilt during pass 3A. */
 void e2fsck_rehash_dir_later(e2fsck_t ctx, ext2_ino_t ino)
@@ -101,6 +102,21 @@ struct out_dir {
        ext2_dirhash_t  *hashes;
 };
 
+#define DOTDOT_OFFSET 12
+
+static int is_fake_entry(ext2_filsys fs, int lblk, unsigned int offset)
+{
+       /* Entries in the first block before this value refer to . or .. */
+       if (lblk == 0 && offset <= DOTDOT_OFFSET)
+               return 1;
+       /* Check if this is likely the csum entry */
+       if (ext2fs_has_feature_metadata_csum(fs->super) &&
+           (offset & (fs->blocksize - 1)) ==
+                           fs->blocksize - sizeof(struct ext2_dir_entry_tail))
+               return 1;
+       return 0;
+}
+
 static int fill_dir_block(ext2_filsys fs,
                          blk64_t *block_nr,
                          e2_blkcnt_t blockcnt,
@@ -113,7 +129,7 @@ static int fill_dir_block(ext2_filsys fs,
        struct ext2_dir_entry   *dirent;
        char                    *dir;
        unsigned int            offset, dir_offset, rec_len, name_len;
-       int                     hash_alg, hash_flags;
+       int                     hash_alg, hash_flags, hash_in_entry;
 
        if (blockcnt < 0)
                return 0;
@@ -140,6 +156,7 @@ static int fill_dir_block(ext2_filsys fs,
                        return BLOCK_ABORT;
        }
        hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
+       hash_in_entry = ext4_hash_in_dirent(fd->inode);
        hash_alg = fs->super->s_def_hash_version;
        if ((hash_alg <= EXT2_HASH_TEA) &&
            (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
@@ -147,13 +164,18 @@ static int fill_dir_block(ext2_filsys fs,
        /* While the directory block is "hot", index it. */
        dir_offset = 0;
        while (dir_offset < fs->blocksize) {
+               unsigned int min_rec = EXT2_DIR_ENTRY_HEADER_LEN;
+               int extended = hash_in_entry && !is_fake_entry(fs, blockcnt, dir_offset);
+
+               if (extended)
+                       min_rec += EXT2_DIR_ENTRY_HASH_LEN;
                dirent = (struct ext2_dir_entry *) (dir + dir_offset);
                (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
                name_len = ext2fs_dirent_name_len(dirent);
                if (((dir_offset + rec_len) > fs->blocksize) ||
-                   (rec_len < 8) ||
+                   (rec_len < min_rec) ||
                    ((rec_len % 4) != 0) ||
-                   (name_len + 8 > rec_len)) {
+                   (name_len + min_rec > rec_len)) {
                        fd->err = EXT2_ET_DIR_CORRUPTED;
                        return BLOCK_ABORT;
                }
@@ -187,11 +209,14 @@ static int fill_dir_block(ext2_filsys fs,
                }
                ent = fd->harray + fd->num_array++;
                ent->dir = dirent;
-               fd->dir_size += EXT2_DIR_REC_LEN(name_len);
+               fd->dir_size += ext2fs_dir_rec_len(name_len, extended);
                ent->ino = dirent->inode;
-               if (fd->compress)
+               if (extended) {
+                       ent->hash = EXT2_DIRENT_HASH(dirent);
+                       ent->minor_hash = EXT2_DIRENT_MINOR_HASH(dirent);
+               } else if (fd->compress) {
                        ent->hash = ent->minor_hash = 0;
-               else {
+               else {
                        fd->err = ext2fs_dirhash2(hash_alg,
                                                  dirent->name, name_len,
                                                  fs->encoding, hash_flags,
@@ -214,6 +239,23 @@ static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
        return (he_a->ino - he_b->ino);
 }
 
+struct name_cmp_ctx
+{
+       int casefold;
+       const struct ext2fs_nls_table *tbl;
+};
+
+static int same_name(const struct name_cmp_ctx *cmp_ctx, char *s1,
+                    int len1, char *s2, int len2)
+{
+       if (!cmp_ctx->casefold)
+               return (len1 == len2 && !memcmp(s1, s2, len1));
+       else
+               return !ext2fs_casefold_cmp(cmp_ctx->tbl,
+                                           (unsigned char *) s1, len1,
+                                           (unsigned char *) s2, len2);
+}
+
 /* Used for sorting the hash entry */
 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 {
@@ -240,9 +282,35 @@ static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
        return ret;
 }
 
+static EXT2_QSORT_TYPE name_cf_cmp(const struct name_cmp_ctx *ctx,
+                                  const void *a, const void *b)
+{
+       const struct hash_entry *he_a = (const struct hash_entry *) a;
+       const struct hash_entry *he_b = (const struct hash_entry *) b;
+       unsigned int he_a_len, he_b_len;
+       int ret;
+
+       he_a_len = ext2fs_dirent_name_len(he_a->dir);
+       he_b_len = ext2fs_dirent_name_len(he_b->dir);
+
+       ret = ext2fs_casefold_cmp(ctx->tbl,
+                                 (unsigned char *) he_a->dir->name, he_a_len,
+                                 (unsigned char *) he_b->dir->name, he_b_len);
+       if (ret == 0) {
+               if (he_a_len > he_b_len)
+                       ret = 1;
+               else if (he_a_len < he_b_len)
+                       ret = -1;
+               else
+                       ret = he_b->dir->inode - he_a->dir->inode;
+       }
+       return ret;
+}
+
 /* Used for sorting the hash entry */
-static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
+static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b, void *arg)
 {
+       const struct name_cmp_ctx *ctx = (struct name_cmp_ctx *) arg;
        const struct hash_entry *he_a = (const struct hash_entry *) a;
        const struct hash_entry *he_b = (const struct hash_entry *) b;
        int     ret;
@@ -256,8 +324,12 @@ static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
                        ret = 1;
                else if (he_a->minor_hash < he_b->minor_hash)
                        ret = -1;
-               else
-                       ret = name_cmp(a, b);
+               else {
+                       if (ctx->casefold)
+                               ret = name_cf_cmp(ctx, a, b);
+                       else
+                               ret = name_cmp(a, b);
+               }
        }
        return ret;
 }
@@ -380,7 +452,8 @@ static void mutate_name(char *str, unsigned int *len)
 
 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
                                    ext2_ino_t ino,
-                                   struct fill_dir_struct *fd)
+                                   struct fill_dir_struct *fd,
+                                   const struct name_cmp_ctx *cmp_ctx)
 {
        struct problem_context  pctx;
        struct hash_entry       *ent, *prev;
@@ -403,10 +476,10 @@ static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
                ent = fd->harray + i;
                prev = ent - 1;
                if (!ent->dir->inode ||
-                   (ext2fs_dirent_name_len(ent->dir) !=
-                    ext2fs_dirent_name_len(prev->dir)) ||
-                   memcmp(ent->dir->name, prev->dir->name,
-                            ext2fs_dirent_name_len(ent->dir)))
+                   !same_name(cmp_ctx, ent->dir->name,
+                              ext2fs_dirent_name_len(ent->dir),
+                              prev->dir->name,
+                              ext2fs_dirent_name_len(prev->dir)))
                        continue;
                pctx.dirent = ent->dir;
                if ((ent->dir->inode == prev->dir->inode) &&
@@ -435,10 +508,11 @@ static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
                mutate_name(new_name, &new_len);
                for (j=0; j < fd->num_array; j++) {
                        if ((i==j) ||
-                           (new_len !=
-                            (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
-                           memcmp(new_name, fd->harray[j].dir->name, new_len))
+                           !same_name(cmp_ctx, new_name, new_len,
+                                      fd->harray[j].dir->name,
+                                      ext2fs_dirent_name_len(fd->harray[j].dir))) {
                                continue;
+                       }
                        mutate_name(new_name, &new_len);
 
                        j = -1;
@@ -473,6 +547,8 @@ static errcode_t copy_dir_entries(e2fsck_t ctx,
        ext2_dirhash_t          prev_hash;
        int                     csum_size = 0;
        struct                  ext2_dir_entry_tail *t;
+       int hash_in_entry = ext4_hash_in_dirent(fd->inode);
+       unsigned int min_rec_len = ext2fs_dir_rec_len(1, hash_in_entry);
 
        if (ctx->htree_slack_percentage == 255) {
                profile_get_uint(ctx->profile, "options",
@@ -501,15 +577,16 @@ static errcode_t copy_dir_entries(e2fsck_t ctx,
        prev_rec_len = 0;
        rec_len = 0;
        left = fs->blocksize - csum_size;
-       slack = fd->compress ? 12 :
+       slack = fd->compress ? min_rec_len :
                ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
-       if (slack < 12)
-               slack = 12;
+       if (slack < min_rec_len)
+               slack = min_rec_len;
        for (i = 0; i < fd->num_array; i++) {
                ent = fd->harray + i;
                if (ent->dir->inode == 0)
                        continue;
-               rec_len = EXT2_DIR_REC_LEN(ext2fs_dirent_name_len(ent->dir));
+               rec_len = ext2fs_dir_rec_len(ext2fs_dirent_name_len(ent->dir),
+                                            hash_in_entry);
                if (rec_len > left) {
                        if (left) {
                                left += prev_rec_len;
@@ -546,6 +623,11 @@ static errcode_t copy_dir_entries(e2fsck_t ctx,
                prev_rec_len = rec_len;
                memcpy(dirent->name, ent->dir->name,
                       ext2fs_dirent_name_len(dirent));
+               if (hash_in_entry) {
+                       EXT2_DIRENT_HASHES(dirent)->hash = ext2fs_cpu_to_le32(ent->hash);
+                       EXT2_DIRENT_HASHES(dirent)->minor_hash =
+                                                       ext2fs_cpu_to_le32(ent->minor_hash);
+               }
                offset += rec_len;
                left -= rec_len;
                if (left < slack) {
@@ -570,7 +652,8 @@ static errcode_t copy_dir_entries(e2fsck_t ctx,
 
 
 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
-                                   ext2_ino_t ino, ext2_ino_t parent)
+                                   ext2_ino_t ino, ext2_ino_t parent,
+                                   struct ext2_inode *inode)
 {
        struct ext2_dir_entry           *dir;
        struct ext2_dx_root_info        *root;
@@ -598,7 +681,10 @@ static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 
        root = (struct ext2_dx_root_info *) (buf+24);
        root->reserved_zero = 0;
-       root->hash_version = fs->super->s_def_hash_version;
+       if (ext4_hash_in_dirent(inode))
+               root->hash_version = EXT2_HASH_SIPHASH;
+       else
+               root->hash_version = fs->super->s_def_hash_version;
        root->info_length = 8;
        root->indirect_levels = 0;
        root->unused_flags = 0;
@@ -684,7 +770,8 @@ static int alloc_blocks(ext2_filsys fs,
 static errcode_t calculate_tree(ext2_filsys fs,
                                struct out_dir *outdir,
                                ext2_ino_t ino,
-                               ext2_ino_t parent)
+                               ext2_ino_t parent,
+                               struct ext2_inode *inode)
 {
        struct ext2_dx_root_info        *root_info;
        struct ext2_dx_entry            *root, *int_ent, *dx_ent = 0;
@@ -693,7 +780,7 @@ static errcode_t calculate_tree(ext2_filsys fs,
        int                             i, c1, c2, c3, nblks;
        int                             limit_offset, int_offset, root_offset;
 
-       root_info = set_root_node(fs, outdir->buf, ino, parent);
+       root_info = set_root_node(fs, outdir->buf, ino, parent, inode);
        root_offset = limit_offset = ((char *) root_info - outdir->buf) +
                root_info->info_length;
        root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
@@ -903,6 +990,7 @@ errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
        struct fill_dir_struct  fd = { NULL, NULL, 0, 0, 0, NULL,
                                       0, 0, 0, 0, 0, 0 };
        struct out_dir          outdir = { 0, 0, 0, 0 };
+       struct name_cmp_ctx name_cmp_ctx = {0, NULL};
 
        e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 
@@ -930,6 +1018,11 @@ errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
                fd.compress = 1;
        fd.parent = 0;
 
+       if (fs->encoding && (inode.i_flags & EXT4_CASEFOLD_FL)) {
+               name_cmp_ctx.casefold = 1;
+               name_cmp_ctx.tbl = fs->encoding;
+       }
+
 retry_nohash:
        /* Read in the entire directory into memory */
        retval = ext2fs_block_iterate3(fs, ino, 0, 0,
@@ -958,16 +1051,18 @@ retry_nohash:
        /* Sort the list */
 resort:
        if (fd.compress && fd.num_array > 1)
-               qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
-                     hash_cmp);
+               sort_r_simple(fd.harray+2, fd.num_array-2,
+                             sizeof(struct hash_entry),
+                             hash_cmp, &name_cmp_ctx);
        else
-               qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
-                     hash_cmp);
+               sort_r_simple(fd.harray, fd.num_array,
+                             sizeof(struct hash_entry),
+                             hash_cmp, &name_cmp_ctx);
 
        /*
         * Look for duplicates
         */
-       if (duplicate_search_and_fix(ctx, fs, ino, &fd))
+       if (duplicate_search_and_fix(ctx, fs, ino, &fd, &name_cmp_ctx))
                goto resort;
 
        if (ctx->options & E2F_OPT_NO) {
@@ -992,7 +1087,7 @@ resort:
 
        if (!fd.compress) {
                /* Calculate the interior nodes */
-               retval = calculate_tree(fs, &outdir, ino, fd.parent);
+               retval = calculate_tree(fs, &outdir, ino, fd.parent, fd.inode);
                if (retval)
                        goto errout;
        }