]> wimlib.net Git - wimlib/blobdiff - src/resource.c
NTFS capture updates
[wimlib] / src / resource.c
index 00dc9d0ea6432556e38ac407a77ca47967b01f94..73cea71c2de7b264671e5c4629650176d6478814 100644 (file)
 #include <errno.h>
 #include <alloca.h>
 
+#ifdef WITH_NTFS_3G
+#include <ntfs-3g/attrib.h>
+#include <ntfs-3g/inode.h>
+#include <ntfs-3g/dir.h>
+#endif
 
 /* 
  * Reads all or part of a compressed resource into an in-memory buffer.
@@ -431,18 +436,9 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
                wimlib_assert(lte->wim->fp);
                ctype = wim_resource_compression_type(lte);
 
-               /* XXX This check should be moved elsewhere */
-               if (ctype == WIM_COMPRESSION_TYPE_NONE &&
-                    lte->resource_entry.original_size !=
-                     lte->resource_entry.size) {
-                       ERROR("WIM resource at offset %"PRIu64", size %"PRIu64
-                             "has an original size of %"PRIu64", but is "
-                             "uncompressed",
-                             lte->resource_entry.offset,
-                             lte->resource_entry.size,
-                             lte->resource_entry.original_size);
-                       return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
-               }
+               wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
+                             (lte->resource_entry.original_size ==
+                              lte->resource_entry.size));
 
                if (raw || ctype == WIM_COMPRESSION_TYPE_NONE)
                        return read_uncompressed_resource(lte->wim->fp,
@@ -456,16 +452,11 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
                                                        ctype, size, offset, buf);
                break;
        case RESOURCE_IN_STAGING_FILE:
-               /* The WIM FUSE implementation needs to handle multiple open
-                * file descriptors per lookup table entry so it does not
-                * currently work with this function. */
-               wimlib_assert(lte->staging_file_name);
-               wimlib_assert(0);
-               break;
        case RESOURCE_IN_FILE_ON_DISK:
                /* The resource is in some file on the external filesystem and
                 * needs to be read uncompressed */
                wimlib_assert(lte->file_on_disk);
+               wimlib_assert(&lte->file_on_disk == &lte->staging_file_name);
                /* Use existing file pointer if available; otherwise open one
                 * temporarily */
                if (lte->file_on_disk_fp) {
@@ -489,6 +480,22 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
                memcpy(buf, lte->attached_buffer + offset, size);
                return 0;
                break;
+#ifdef WITH_NTFS_3G
+       case RESOURCE_IN_NTFS_VOLUME:
+               if (lte->attr) {
+                       if (ntfs_attr_pread(lte->attr, offset, size, buf) == size) {
+                               return 0;
+                       } else {
+                               ERROR_WITH_ERRNO("Error reading NTFS attribute "
+                                                "at `%s'",
+                                                lte->ntfs_loc->path_utf8);
+                               return WIMLIB_ERR_NTFS_3G;
+                       }
+               } else {
+                       wimlib_assert(0);
+               }
+               break;
+#endif
        default:
                assert(0);
        }
@@ -702,6 +709,11 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
  * Writes a WIM resource to a FILE * opened for writing.  The resource may be
  * written uncompressed or compressed depending on the @out_ctype parameter.
  *
+ * If by chance the resource compresses to more than the original size (this may
+ * happen with random data or files than are pre-compressed), the resource is
+ * instead written uncompressed (and this is reflected in the @out_res_entry by
+ * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
+ *
  * @lte:       The lookup table entry for the WIM resource.
  * @out_fp:    The FILE * to write the resource to.
  * @out_ctype:  The compression type of the resource to write.  Note: if this is
@@ -727,10 +739,17 @@ static int write_wim_resource(struct lookup_table_entry *lte,
        struct chunk_table *chunk_tab = NULL;
        bool raw;
        off_t file_offset;
+#ifdef WITH_NTFS_3G
+       ntfs_inode *ni;
+#endif
 
+       /* Original size of the resource */
        original_size = wim_resource_size(lte);
+
+       /* Compressed size of the resource (as it exists now) */
        old_compressed_size = wim_resource_compressed_size(lte);
 
+       /* Current offset in output file */
        file_offset = ftello(out_fp);
        if (file_offset == -1) {
                ERROR_WITH_ERRNO("Failed to get offset in output "
@@ -738,6 +757,8 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                return WIMLIB_ERR_WRITE;
        }
        
+       /* Are the compression types the same?  If so, do a raw copy (copy
+        * without decompressing and recompressing the data). */
        raw = (wim_resource_compression_type(lte) == out_ctype
               && out_ctype != WIM_COMPRESSION_TYPE_NONE);
        if (raw)
@@ -745,11 +766,15 @@ static int write_wim_resource(struct lookup_table_entry *lte,
        else
                bytes_remaining = original_size;
 
+       /* Empty resource; nothing needs to be done, so just return success. */
        if (bytes_remaining == 0)
                return 0;
 
+       /* Buffer for reading chunks for the resource */
        char buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
 
+       /* If we are writing a compressed resource and not doing a raw copy, we
+        * need to initialize the chunk table */
        if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
                ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
                                                   &chunk_tab);
@@ -757,11 +782,12 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                        goto out;
        }
 
+       /* If the WIM resource is in an external file, open a FILE * to it so we
+        * don't have to open a temporary one in read_wim_resource() for each
+        * chunk. */
        if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
             && !lte->file_on_disk_fp)
        {
-               /* The WIM resource is in an external file; open a FILE * to it
-                * so we don't have to open a temporary one on every read. */
                wimlib_assert(lte->file_on_disk);
                lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
                if (!lte->file_on_disk_fp) {
@@ -771,10 +797,40 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                        goto out;
                }
        }
+#ifdef WITH_NTFS_3G
+       else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
+                  && !lte->attr)
+       {
+               struct ntfs_location *loc = lte->ntfs_loc;
+               wimlib_assert(loc);
+               ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
+               if (!ni) {
+                       ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
+                                        "volume", loc->path_utf8);
+               }
+               lte->attr = ntfs_attr_open(ni,
+                                          loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
+                                          (ntfschar*)loc->stream_name_utf16,
+                                          loc->stream_name_utf16_num_chars);
+               if (!lte->attr) {
+                       ntfs_inode_close(ni);
+                       ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
+                                        "NTFS volume", loc->path_utf8);
+               }
+       }
+#endif
+
+       /* If we aren't doing a raw copy, we will compute the SHA1 message
+        * digest of the resource as we read it, and verify it's the same as the
+        * hash given in the lookup table entry once we've finished reading the
+        * resource. */
        SHA_CTX ctx;
        if (!raw)
                sha1_init(&ctx);
 
+       /* While there are still bytes remaining in the WIM resource, read a
+        * chunk of the resource, update SHA1, then write that chunk using the
+        * desired compression type. */
        do {
                u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
                ret = read_wim_resource(lte, buf, to_read, offset, raw);
@@ -789,6 +845,12 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                bytes_remaining -= to_read;
                offset += to_read;
        } while (bytes_remaining);
+
+       /* If writing a compressed resource and not doing a raw copy, write the
+        * chunk table, and finish_wim_resource_chunk_tab() will provide the
+        * compressed size of the resource we wrote.  Otherwise, the compressed
+        * size of the written resource is the same as the compressed size of
+        * the existing resource. */
        if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
                ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
                                                    &new_compressed_size);
@@ -798,13 +860,16 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                new_compressed_size = old_compressed_size;
        }
 
+       /* Verify SHA1 message digest of the resource, unless we are doing a raw
+        * write (in which case we never even saw the uncompressed data).  Or,
+        * if the hash we had before is all 0's, just re-set it to be the new
+        * hash. */
        if (!raw) {
-               /* Verify SHA1 message digest of the resource, unless we are
-                * doing a raw write (in which case we may have never even seen
-                * the uncompressed data)  */
                u8 md[SHA1_HASH_SIZE];
                sha1_final(md, &ctx);
-               if (!hashes_equal(md, lte->hash)) {
+               if (is_zero_hash(lte->hash)) {
+                       copy_hash(lte->hash, md);
+               } else if (!hashes_equal(md, lte->hash)) {
                        ERROR("WIM resource has incorrect hash!");
                        if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
                                ERROR("We were reading it from `%s'; maybe it changed "
@@ -856,25 +921,43 @@ out_fclose:
                fclose(lte->file_on_disk_fp);
                lte->file_on_disk_fp = NULL;
        }
+#ifdef WITH_NTFS_3G
+       else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
+                && lte->attr) {
+               ntfs_attr_close(lte->attr);
+               ntfs_inode_close(ni);
+       }
+#endif
 out:
        FREE(chunk_tab);
        return ret;
 }
 
+/* Like write_wim_resource(), but the resource is specified by a buffer of
+ * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
+ * the buffer to @hash.  */
 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
-                                         u8 buf_hash[SHA1_HASH_SIZE],
                                          FILE *out_fp, int out_ctype,
-                                         struct resource_entry *out_res_entry)
+                                         struct resource_entry *out_res_entry,
+                                         u8 hash[SHA1_HASH_SIZE])
 {
+       /* Set up a temporary lookup table entry that we provide to
+        * write_wim_resource(). */
        struct lookup_table_entry lte;
+       int ret;
        lte.resource_entry.flags         = 0;
        lte.resource_entry.original_size = buf_size;
        lte.resource_entry.size          = buf_size;
        lte.resource_entry.offset        = 0;
        lte.resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
        lte.attached_buffer              = (u8*)buf;
-       copy_hash(lte.hash, buf_hash);
-       return write_wim_resource(&lte, out_fp, out_ctype, out_res_entry);
+
+       zero_hash(lte.hash);
+       ret = write_wim_resource(&lte, out_fp, out_ctype, out_res_entry);
+       if (ret != 0)
+               return ret;
+       copy_hash(hash, lte.hash);
+       return 0;
 }
 
 /* 
@@ -931,7 +1014,8 @@ int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd
 
 /* 
  * Copies the file resource specified by the lookup table entry @lte from the
- * input WIM the output WIM.
+ * input WIM to the output WIM that has its FILE * given by
+ * ((WIMStruct*)wim)->out_fp.
  *
  * The output_resource_entry, out_refcnt, and part_number fields of @lte are
  * updated.
@@ -1125,38 +1209,35 @@ out_free_buf:
        return ret;
 }
 
-/* Write the metadata resource for the current image. */
+/* Write the metadata resource for the current WIM image. */
 int write_metadata_resource(WIMStruct *w)
 {
-       FILE *out;
        u8 *buf;
        u8 *p;
        int ret;
        u64 subdir_offset;
        struct dentry *root;
-       struct lookup_table_entry *lte;
-       off_t metadata_offset;
+       struct lookup_table_entry *lte, *duplicate_lte;
        u64 metadata_original_size;
-       u64 metadata_compressed_size;
-       int metadata_ctype;
-       u8  hash[SHA1_HASH_SIZE];
+
+       /* 
+        * We append 20 random bytes to the metadata resource so that we don't
+        * have identical metadata resources if we happen to append exactly the
+        * same image twice without any changes in timestamps.  If this were to
+        * happen, it would cause confusion about the number and order of images
+        * in the WIM.
+        */
+       const unsigned random_tail_len = 20;
 
        DEBUG("Writing metadata resource for image %d", w->current_image);
 
-       out = w->out_fp;
        root = wim_root_dentry(w);
-       metadata_ctype = wimlib_get_compression_type(w);
-       metadata_offset = ftello(out);
-       if (metadata_offset == -1)
-               return WIMLIB_ERR_WRITE;
 
-       struct wim_security_data *sd = wim_security_data(w);
-       if (sd)
-               subdir_offset = sd->total_length + root->length + 8;
-       else
-               subdir_offset = 8 + root->length + 8;
+       const struct wim_security_data *sd = wim_security_data(w);
+       wimlib_assert(sd);
+       subdir_offset = sd->total_length + root->length + 8;
        calculate_subdir_offsets(root, &subdir_offset);
-       metadata_original_size = subdir_offset;
+       metadata_original_size = subdir_offset + random_tail_len;
        buf = MALLOC(metadata_original_size);
        if (!buf) {
                ERROR("Failed to allocate %"PRIu64" bytes for "
@@ -1168,27 +1249,22 @@ int write_metadata_resource(WIMStruct *w)
 
        DEBUG("Writing dentry tree.");
        p = write_dentry_tree(root, p);
-
-       /* Like file resources, the lookup table entry for a metadata resource
-        * uses for the hash code a SHA1 message digest of its uncompressed
-        * contents. */
-       sha1_buffer(buf, metadata_original_size, hash);
-
+       randomize_byte_array(p, random_tail_len);
+       wimlib_assert(p - buf + random_tail_len == metadata_original_size);
 
        lte = wim_metadata_lookup_table_entry(w);
 
        ret = write_wim_resource_from_buffer(buf, metadata_original_size,
-                                            hash, out, metadata_ctype,
-                                            &lte->output_resource_entry);
-
+                                            w->out_fp,
+                                            wimlib_get_compression_type(w),
+                                            &lte->output_resource_entry,
+                                            lte->hash);
        lookup_table_unlink(w->lookup_table, lte);
-       copy_hash(lte->hash, hash);
        lookup_table_insert(w->lookup_table, lte);
-       lte->out_refcnt++;
+       wimlib_assert(lte->out_refcnt == 0);
+       lte->out_refcnt = 1;
        lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
+out:
        FREE(buf);
-       if (ret != 0)
-               return ret;
-
-       return 0;
+       return ret;
 }