]> wimlib.net Git - wimlib/blobdiff - src/ntfs-3g_capture.c
Reparse point updates
[wimlib] / src / ntfs-3g_capture.c
index 06d46913966d8444bcf8c08b99a99eaa8a871e1c..7940e0eac0661b4946e6315e08c0b9ce7d6c6f6d 100644 (file)
 #include "wimlib/reparse.h"
 #include "wimlib/security.h"
 
-static inline ntfschar *
-attr_record_name(ATTR_RECORD *ar)
+/* A reference-counted NTFS volume than is automatically unmounted when the
+ * reference count reaches 0  */
+struct ntfs_volume_wrapper {
+       ntfs_volume *vol;
+       size_t refcnt;
+};
+
+/* Description of where data is located in an NTFS volume  */
+struct ntfs_location {
+       struct ntfs_volume_wrapper *volume;
+       u64 mft_no;
+       utf16lechar *attr_name;
+       unsigned attr_name_nchars;
+       unsigned attr_type;
+       u64 sort_key;
+};
+
+static struct ntfs_volume_wrapper *
+get_ntfs_volume(struct ntfs_volume_wrapper *volume)
+{
+       volume->refcnt++;
+       return volume;
+}
+
+static void
+put_ntfs_volume(struct ntfs_volume_wrapper *volume)
+{
+       if (--volume->refcnt == 0) {
+               ntfs_umount(volume->vol, FALSE);
+               FREE(volume);
+       }
+}
+
+static inline const ntfschar *
+attr_record_name(const ATTR_RECORD *record)
 {
-       return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset));
+       return (const ntfschar *)
+               ((const u8 *)record + le16_to_cpu(record->name_offset));
 }
 
 static ntfs_attr *
@@ -64,18 +98,18 @@ open_ntfs_attr(ntfs_inode *ni, const struct ntfs_location *loc)
                            loc->attr_name,
                            loc->attr_name_nchars);
        if (!na) {
-               ERROR_WITH_ERRNO("Failed to open attribute of \"%"TS"\" in "
-                                "NTFS volume", loc->path);
+               ERROR_WITH_ERRNO("Failed to open attribute of NTFS inode %"PRIu64,
+                                loc->mft_no);
        }
        return na;
 }
 
 int
 read_ntfs_attribute_prefix(const struct blob_descriptor *blob, u64 size,
-                          consume_data_callback_t cb, void *cb_ctx)
+                          const struct read_blob_callbacks *cbs)
 {
        const struct ntfs_location *loc = blob->ntfs_loc;
-       ntfs_volume *vol = loc->ntfs_vol;
+       ntfs_volume *vol = loc->volume->vol;
        ntfs_inode *ni;
        ntfs_attr *na;
        s64 pos;
@@ -83,9 +117,10 @@ read_ntfs_attribute_prefix(const struct blob_descriptor *blob, u64 size,
        int ret;
        u8 buf[BUFFER_SIZE];
 
-       ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
+       ni = ntfs_inode_open(vol, loc->mft_no);
        if (!ni) {
-               ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
+               ERROR_WITH_ERRNO("Failed to open NTFS inode %"PRIu64,
+                                loc->mft_no);
                ret = WIMLIB_ERR_NTFS_3G;
                goto out;
        }
@@ -101,13 +136,14 @@ read_ntfs_attribute_prefix(const struct blob_descriptor *blob, u64 size,
        while (bytes_remaining) {
                s64 to_read = min(bytes_remaining, sizeof(buf));
                if (ntfs_attr_pread(na, pos, to_read, buf) != to_read) {
-                       ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
+                       ERROR_WITH_ERRNO("Error reading data from NTFS inode "
+                                        "%"PRIu64, loc->mft_no);
                        ret = WIMLIB_ERR_NTFS_3G;
                        goto out_close_ntfs_attr;
                }
                pos += to_read;
                bytes_remaining -= to_read;
-               ret = cb(buf, to_read, cb_ctx);
+               ret = call_consume_chunk(buf, to_read, cbs);
                if (ret)
                        goto out_close_ntfs_attr;
        }
@@ -120,6 +156,41 @@ out:
        return ret;
 }
 
+void
+free_ntfs_location(struct ntfs_location *loc)
+{
+       put_ntfs_volume(loc->volume);
+       FREE(loc->attr_name);
+       FREE(loc);
+}
+
+struct ntfs_location *
+clone_ntfs_location(const struct ntfs_location *loc)
+{
+       struct ntfs_location *new = memdup(loc, sizeof(*loc));
+       if (!new)
+               goto err0;
+       if (loc->attr_name) {
+               new->attr_name = utf16le_dup(loc->attr_name);
+               if (!new->attr_name)
+                       goto err1;
+       }
+       new->volume = get_ntfs_volume(loc->volume);
+       return new;
+
+err1:
+       FREE(new);
+err0:
+       return NULL;
+}
+
+int
+cmp_ntfs_locations(const struct ntfs_location *loc1,
+                  const struct ntfs_location *loc2)
+{
+       return cmp_u64(loc1->sort_key, loc2->sort_key);
+}
+
 static int
 read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
                 u32 *reparse_tag_ret)
@@ -165,138 +236,168 @@ attr_type_to_wimlib_stream_type(ATTR_TYPES type)
        }
 }
 
-/* Load attributes of the specified type from a file in the NTFS volume  */
+/* When sorting blobs located in NTFS volumes for sequential reading, we sort
+ * first by starting LCN of the attribute if available, otherwise no sort order
+ * is defined.  This usually results in better sequential access to the volume.
+ */
+static int
+set_attr_sort_key(ntfs_inode *ni, struct ntfs_location *loc)
+{
+       ntfs_attr *na;
+       runlist_element *rl;
+
+       na = open_ntfs_attr(ni, loc);
+       if (!na)
+               return WIMLIB_ERR_NTFS_3G;
+
+       rl = ntfs_attr_find_vcn(na, 0);
+       if (rl && rl->lcn != LCN_HOLE)
+               loc->sort_key = rl->lcn;
+       else
+               loc->sort_key = 0;
+
+       ntfs_attr_close(na);
+       return 0;
+}
+
+/* Save information about an NTFS attribute (stream) to a WIM inode.  */
 static int
-load_ntfs_attrs_with_type(struct wim_inode *inode,
+scan_ntfs_attr(struct wim_inode *inode,
+              ntfs_inode *ni,
+              const char *path,
+              size_t path_len,
+              struct list_head *unhashed_blobs,
+              struct ntfs_volume_wrapper *volume,
+              ATTR_TYPES type,
+              const ATTR_RECORD *record)
+{
+       const u64 data_size = ntfs_get_attribute_value_length(record);
+       const size_t name_nchars = record->name_length;
+       struct blob_descriptor *blob = NULL;
+       utf16lechar *stream_name = NULL;
+       struct wim_inode_stream *strm;
+       int ret;
+
+       if (unlikely(name_nchars)) {
+               /* Named stream  */
+               stream_name = utf16le_dupz(attr_record_name(record),
+                                          name_nchars * sizeof(ntfschar));
+               if (!stream_name) {
+                       ret = WIMLIB_ERR_NOMEM;
+                       goto out_cleanup;
+               }
+       }
+
+       /* If the stream is non-empty, set up a blob descriptor for it.  */
+       if (data_size != 0) {
+               blob = new_blob_descriptor();
+               if (unlikely(!blob)) {
+                       ret = WIMLIB_ERR_NOMEM;
+                       goto out_cleanup;
+               }
+
+               blob->ntfs_loc = CALLOC(1, sizeof(struct ntfs_location));
+               if (unlikely(!blob->ntfs_loc)) {
+                       ret = WIMLIB_ERR_NOMEM;
+                       goto out_cleanup;
+               }
+
+               blob->blob_location = BLOB_IN_NTFS_VOLUME;
+               blob->size = data_size;
+               blob->ntfs_loc->volume = get_ntfs_volume(volume);
+               blob->ntfs_loc->attr_type = type;
+               blob->ntfs_loc->mft_no = ni->mft_no;
+
+               if (unlikely(name_nchars)) {
+                       blob->ntfs_loc->attr_name = utf16le_dup(stream_name);
+                       if (!blob->ntfs_loc->attr_name) {
+                               ret = WIMLIB_ERR_NOMEM;
+                               goto out_cleanup;
+                       }
+                       blob->ntfs_loc->attr_name_nchars = name_nchars;
+               }
+
+               ret = set_attr_sort_key(ni, blob->ntfs_loc);
+               if (ret)
+                       goto out_cleanup;
+
+               if (unlikely(type == AT_REPARSE_POINT)) {
+                       if (blob->size < REPARSE_DATA_OFFSET) {
+                               ERROR("Reparse data of \"%s\" "
+                                     "is invalid (only %"PRIu64" bytes)!",
+                                     path, data_size);
+                               ret = WIMLIB_ERR_INVALID_REPARSE_DATA;
+                               goto out_cleanup;
+                       }
+                       blob->size -= REPARSE_DATA_OFFSET;
+                       ret = read_reparse_tag(ni, blob->ntfs_loc,
+                                              &inode->i_reparse_tag);
+                       if (ret)
+                               goto out_cleanup;
+               }
+       }
+
+       strm = inode_add_stream(inode,
+                               attr_type_to_wimlib_stream_type(type),
+                               stream_name ? stream_name : NO_STREAM_NAME,
+                               blob);
+       if (unlikely(!strm)) {
+               ret = WIMLIB_ERR_NOMEM;
+               goto out_cleanup;
+       }
+       prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
+       blob = NULL;
+       ret = 0;
+out_cleanup:
+       free_blob_descriptor(blob);
+       FREE(stream_name);
+       return ret;
+}
+
+/* Scan attributes of the specified type from a file in the NTFS volume  */
+static int
+scan_ntfs_attrs_with_type(struct wim_inode *inode,
                          ntfs_inode *ni,
                          char *path,
                          size_t path_len,
                          struct list_head *unhashed_blobs,
-                         ntfs_volume *vol,
+                         struct ntfs_volume_wrapper *volume,
                          ATTR_TYPES type)
 {
        ntfs_attr_search_ctx *actx;
-       struct ntfs_location *ntfs_loc;
        int ret;
-       struct blob_descriptor *blob;
-       utf16lechar *stream_name;
 
-       DEBUG("Loading NTFS attributes from \"%s\"", path);
+       DEBUG("Scanning NTFS attributes from \"%s\"", path);
 
-       /* Get context to search the attributes of the NTFS file. */
        actx = ntfs_attr_get_search_ctx(ni, NULL);
        if (!actx) {
-               ERROR_WITH_ERRNO("Cannot get NTFS attribute search "
+               ERROR_WITH_ERRNO("Failed to get NTFS attribute search "
                                 "context for \"%s\"", path);
                return WIMLIB_ERR_NTFS_3G;
        }
 
-       /* Save each attribute  */
        while (!ntfs_attr_lookup(type, NULL, 0,
                                 CASE_SENSITIVE, 0, NULL, 0, actx))
        {
-               u64 data_size = ntfs_get_attribute_value_length(actx->attr);
-               size_t name_nchars = actx->attr->name_length;
-               struct wim_inode_stream *strm;
-
-               if (name_nchars) {
-                       stream_name = utf16le_dupz(attr_record_name(actx->attr),
-                                                  name_nchars * sizeof(ntfschar));
-                       if (!stream_name) {
-                               ret = WIMLIB_ERR_NOMEM;
-                               goto out_put_actx;
-                       }
-               } else {
-                       stream_name = NULL;
-               }
-
-               if (data_size == 0) {
-                       /* Empty attribute.  No blob is needed. */
-                       blob = NULL;
-                       ntfs_loc = NULL;
-               } else {
-                       ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
-                       if (!ntfs_loc) {
-                               ret = WIMLIB_ERR_NOMEM;
-                               goto out_free_stream_name;
-                       }
-                       ntfs_loc->ntfs_vol = vol;
-                       ntfs_loc->attr_type = type;
-                       ntfs_loc->path = memdup(path, path_len + 1);
-                       if (!ntfs_loc->path) {
-                               ret = WIMLIB_ERR_NOMEM;
-                               goto out_free_ntfs_loc;
-                       }
-                       if (name_nchars) {
-                               ntfs_loc->attr_name = utf16le_dup(stream_name);
-                               if (!ntfs_loc->attr_name) {
-                                       ret = WIMLIB_ERR_NOMEM;
-                                       goto out_free_ntfs_loc;
-                               }
-                               ntfs_loc->attr_name_nchars = name_nchars;
-                       }
-
-                       blob = new_blob_descriptor();
-                       if (!blob) {
-                               ret = WIMLIB_ERR_NOMEM;
-                               goto out_free_ntfs_loc;
-                       }
-                       blob->blob_location = BLOB_IN_NTFS_VOLUME;
-                       blob->ntfs_loc = ntfs_loc;
-                       blob->size = data_size;
-                       ntfs_loc = NULL;
-                       if (type == AT_REPARSE_POINT) {
-                               if (data_size < REPARSE_DATA_OFFSET) {
-                                       ERROR("Reparse data of \"%s\" "
-                                             "is invalid (only %u bytes)!",
-                                             path, (unsigned)data_size);
-                                       ret = WIMLIB_ERR_NTFS_3G;
-                                       goto out_free_blob;
-                               }
-                               blob->size -= REPARSE_DATA_OFFSET;
-                               ret = read_reparse_tag(ni, blob->ntfs_loc,
-                                                      &inode->i_reparse_tag);
-                               if (ret)
-                                       goto out_free_blob;
-                       }
-               }
-
-               strm = inode_add_stream(inode,
-                                       attr_type_to_wimlib_stream_type(type),
-                                       stream_name ? stream_name : NO_STREAM_NAME,
-                                       blob);
-               if (!strm) {
-                       ret = WIMLIB_ERR_NOMEM;
-                       goto out_free_blob;
-               }
-               prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
-
-               FREE(stream_name);
-               stream_name = NULL;
+               ret = scan_ntfs_attr(inode,
+                                    ni,
+                                    path,
+                                    path_len,
+                                    unhashed_blobs,
+                                    volume,
+                                    type,
+                                    actx->attr);
+               if (ret)
+                       goto out_put_actx;
        }
-       if (errno == ENOENT) {
-               ret = 0;
-       } else {
+       if (errno != ENOENT) {
                ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
                ret = WIMLIB_ERR_NTFS_3G;
+               goto out_put_actx;
        }
-       goto out_put_actx;
-out_free_blob:
-       free_blob_descriptor(blob);
-out_free_ntfs_loc:
-       if (ntfs_loc) {
-               FREE(ntfs_loc->path);
-               FREE(ntfs_loc->attr_name);
-               FREE(ntfs_loc);
-       }
-out_free_stream_name:
-       FREE(stream_name);
+       ret = 0;
 out_put_actx:
        ntfs_attr_put_search_ctx(actx);
-       if (ret == 0)
-               DEBUG("Successfully loaded NTFS attributes from \"%s\"", path);
-       else
-               ERROR("Failed to load NTFS attributes from \"%s\"", path);
        return ret;
 }
 
@@ -422,7 +523,7 @@ struct readdir_ctx {
        char *path;
        size_t path_len;
        struct dos_name_map *dos_name_map;
-       ntfs_volume *vol;
+       struct ntfs_volume_wrapper *volume;
        struct capture_params *params;
        int ret;
 };
@@ -433,7 +534,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
                                 char *path,
                                 size_t path_len,
                                 int name_type,
-                                ntfs_volume *ntfs_vol,
+                                struct ntfs_volume_wrapper *volume,
                                 struct capture_params *params);
 
 static int
@@ -479,7 +580,7 @@ wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
 
        /* Open the inode for this directory entry and recursively capture the
         * directory tree rooted at it */
-       ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref);
+       ntfs_inode *ni = ntfs_inode_open(ctx->volume->vol, mref);
        if (!ni) {
                /* XXX This used to be treated as an error, but NTFS-3g seemed
                 * to be unable to read some inodes on a Windows 8 image for
@@ -497,7 +598,7 @@ wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
        child = NULL;
        ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path,
                                               path_len, name_type,
-                                              ctx->vol, ctx->params);
+                                              ctx->volume, ctx->params);
        path_len -= mbs_name_nbytes + 1;
        if (child)
                dentry_add_child(ctx->parent, child);
@@ -517,7 +618,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
                                 char *path,
                                 size_t path_len,
                                 int name_type,
-                                ntfs_volume *vol,
+                                struct ntfs_volume_wrapper *volume,
                                 struct capture_params *params)
 {
        u32 attributes;
@@ -575,22 +676,23 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
        inode->i_attributes       = attributes;
 
        if (attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
-               /* Load the reparse point stream.  */
-               ret = load_ntfs_attrs_with_type(inode, ni, path, path_len,
+               /* Scan the reparse point stream.  */
+               ret = scan_ntfs_attrs_with_type(inode, ni, path, path_len,
                                                params->unhashed_blobs,
-                                               vol, AT_REPARSE_POINT);
+                                               volume, AT_REPARSE_POINT);
                if (ret)
                        goto out;
        }
 
-       /* Load the data streams.
+       /* Scan the data streams.
         *
         * Note: directories should not have an unnamed data stream, but they
         * may have named data streams.  Nondirectories (including reparse
         * points) can have an unnamed data stream as well as named data
         * streams.  */
-       ret = load_ntfs_attrs_with_type(inode, ni, path, path_len,
-                                       params->unhashed_blobs, vol, AT_DATA);
+       ret = scan_ntfs_attrs_with_type(inode, ni, path, path_len,
+                                       params->unhashed_blobs,
+                                       volume, AT_DATA);
        if (ret)
                goto out;
 
@@ -604,7 +706,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
                        .path            = path,
                        .path_len        = path_len,
                        .dos_name_map    = &dos_name_map,
-                       .vol             = vol,
+                       .volume          = volume,
                        .params          = params,
                        .ret             = 0,
                };
@@ -638,7 +740,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
         * only allow capturing an entire volume. */
        if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
            inode_is_symlink(inode))
-               inode->i_not_rpfixed = 0;
+               inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED;
 
        if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
                struct SECURITY_CONTEXT sec_ctx;
@@ -647,7 +749,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
 
                /* Get security descriptor */
                memset(&sec_ctx, 0, sizeof(sec_ctx));
-               sec_ctx.vol = vol;
+               sec_ctx.vol = volume->vol;
 
                errno = 0;
                sd = _sd;
@@ -695,36 +797,31 @@ out:
        return ret;
 }
 
-
-int
-do_ntfs_umount(struct _ntfs_volume *vol)
-{
-       DEBUG("Unmounting NTFS volume");
-       if (ntfs_umount(vol, FALSE))
-               return WIMLIB_ERR_NTFS_3G;
-       else
-               return 0;
-}
-
 int
 build_dentry_tree_ntfs(struct wim_dentry **root_p,
                       const char *device,
                       struct capture_params *params)
 {
+       struct ntfs_volume_wrapper *volume;
        ntfs_volume *vol;
        ntfs_inode *root_ni;
+       char *path;
        int ret;
 
+       volume = MALLOC(sizeof(struct ntfs_volume_wrapper));
+       if (!volume)
+               return WIMLIB_ERR_NOMEM;
+
        DEBUG("Mounting NTFS volume `%s' read-only", device);
 
-/* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
- * NTFS_MNT_RDONLY.
- *
- * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
- * NTFS_MNT_RDONLY is an enumerated constant.  Also, the NTFS-3g headers don't
- * seem to contain any explicit version information.  So we have to rely on a
- * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
* */
+       /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
       * NTFS_MNT_RDONLY.
       *
       * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
+        * NTFS_MNT_RDONLY is an enumerated constant.  Also, the NTFS-3g headers
+        * don't seem to contain any explicit version information.  So we have
+        * to rely on a test done at configure time to detect whether
       * NTFS_MNT_RDONLY should be used.  */
 #ifdef HAVE_NTFS_MNT_RDONLY
        /* NTFS-3g 2013 */
        vol = ntfs_mount(device, NTFS_MNT_RDONLY);
@@ -737,8 +834,13 @@ build_dentry_tree_ntfs(struct wim_dentry **root_p,
        if (!vol) {
                ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
                                 device);
+               FREE(volume);
                return WIMLIB_ERR_NTFS_3G;
        }
+
+       volume->vol = vol;
+       volume->refcnt = 1;
+
        ntfs_open_secure(vol);
 
        /* We don't want to capture the special NTFS files such as $Bitmap.  Not
@@ -752,40 +854,29 @@ build_dentry_tree_ntfs(struct wim_dentry **root_p,
                ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
                                 "`%s'", device);
                ret = WIMLIB_ERR_NTFS_3G;
-               goto out;
+               goto out_put_ntfs_volume;
        }
 
        /* Currently we assume that all the paths fit into this length and there
         * is no check for overflow. */
-       char *path = MALLOC(32768);
+       path = MALLOC(32768);
        if (!path) {
-               ERROR("Could not allocate memory for NTFS pathname");
                ret = WIMLIB_ERR_NOMEM;
-               goto out_cleanup;
+               goto out_close_root_ni;
        }
 
        path[0] = '/';
        path[1] = '\0';
        ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1,
-                                              FILE_NAME_POSIX, vol, params);
-out_cleanup:
+                                              FILE_NAME_POSIX, volume, params);
        FREE(path);
+out_close_root_ni:
        ntfs_inode_close(root_ni);
-out:
+out_put_ntfs_volume:
        ntfs_index_ctx_put(vol->secure_xsii);
        ntfs_index_ctx_put(vol->secure_xsdh);
        ntfs_inode_close(vol->secure_ni);
-
-       if (ret) {
-               if (do_ntfs_umount(vol)) {
-                       ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
-                                        device);
-               }
-       } else {
-               /* We need to leave the NTFS volume mounted so that we can read
-                * the NTFS files again when we are actually writing the WIM */
-               *(ntfs_volume**)params->extra_arg = vol;
-       }
+       put_ntfs_volume(volume);
        return ret;
 }
 #endif /* WITH_NTFS_3G */