X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fresource.c;h=3c573f1c9131d4cfb0af8f1ff8e68b87964366b0;hb=c155e206bf806dd639b57922c3eef0b428c985df;hp=00dc9d0ea6432556e38ac407a77ca47967b01f94;hpb=980dec359010f14f558dc4708a689f450befea50;p=wimlib diff --git a/src/resource.c b/src/resource.c index 00dc9d0e..3c573f1c 100644 --- a/src/resource.c +++ b/src/resource.c @@ -36,6 +36,11 @@ #include #include +#ifdef WITH_NTFS_3G +#include +#include +#include +#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(<e->file_on_disk == <e->staging_file_name); /* Use existing file pointer if available; otherwise open one * temporarily */ if (lte->file_on_disk_fp) { @@ -475,6 +466,7 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], if (!fp) { ERROR_WITH_ERRNO("Failed to open the file " "`%s'", lte->file_on_disk); + return WIMLIB_ERR_OPEN; } } ret = read_uncompressed_resource(fp, offset, size, buf); @@ -489,6 +481,22 @@ 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 +710,11 @@ 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 +740,17 @@ 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 +758,8 @@ 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 +767,15 @@ 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 +783,12 @@ 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 +798,40 @@ 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 +846,12 @@ 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 +861,16 @@ 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 +922,43 @@ 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(<e, out_fp, out_ctype, out_res_entry); + + zero_hash(lte.hash); + ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry); + if (ret != 0) + return ret; + copy_hash(hash, lte.hash); + return 0; } /* @@ -931,7 +1015,8 @@ 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. @@ -1009,41 +1094,44 @@ int write_dentry_resources(struct dentry *dentry, void *wim_p) * * @return: Zero on success, nonzero on failure. */ -int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) +int read_metadata_resource(WIMStruct *w, struct image_metadata *imd) { u8 *buf; int ctype; u32 dentry_offset; int ret; struct dentry *dentry; - struct wim_security_data *sd; struct link_group_table *lgt; const struct lookup_table_entry *metadata_lte; - const struct resource_entry *res_entry; + u64 metadata_len; + u64 metadata_offset; metadata_lte = imd->metadata_lte; - res_entry = &metadata_lte->resource_entry; + metadata_len = wim_resource_size(metadata_lte); + metadata_offset = metadata_lte->resource_entry.offset; DEBUG("Reading metadata resource: length = %"PRIu64", " - "offset = %"PRIu64"", - res_entry->original_size, res_entry->offset); - - if (res_entry->original_size < 8) { - ERROR("Expected at least 8 bytes for the metadata resource"); + "offset = %"PRIu64"", metadata_len, metadata_offset); + + /* There is no way the metadata resource could possibly be less than (8 + * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with + * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root + * dentry. */ + if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) { + ERROR("Expected at least %zu bytes for the metadata resource", + 8 + WIM_DENTRY_DISK_SIZE); return WIMLIB_ERR_INVALID_RESOURCE_SIZE; } /* Allocate memory for the uncompressed metadata resource. */ - buf = MALLOC(res_entry->original_size); + buf = MALLOC(metadata_len); if (!buf) { ERROR("Failed to allocate %"PRIu64" bytes for uncompressed " - "metadata resource", res_entry->original_size); + "metadata resource", metadata_len); return WIMLIB_ERR_NOMEM; } - /* Determine the compression type of the metadata resource. */ - /* Read the metadata resource into memory. (It may be compressed.) */ ret = read_full_wim_resource(metadata_lte, buf); if (ret != 0) @@ -1051,17 +1139,27 @@ int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) DEBUG("Finished reading metadata resource into memory."); - /* The root directory entry starts after security data, on an 8-byte - * aligned address. + /* The root directory entry starts after security data, aligned on an + * 8-byte boundary within the metadata resource. * * The security data starts with a 4-byte integer giving its total - * length. */ + * length, so if we round that up to an 8-byte boundary that gives us + * the offset of the root dentry. + * + * Here we read the security data into a wim_security_data structure, + * and if successful, go ahead and calculate the offset in the metadata + * resource of the root dentry. */ - /* Read the security data into a wim_security_data structure. */ - ret = read_security_data(buf, res_entry->original_size, &sd); + ret = read_security_data(buf, metadata_len, &imd->security_data); if (ret != 0) goto out_free_buf; + get_u32(buf, &dentry_offset); + if (dentry_offset == 0) + dentry_offset = 8; + dentry_offset = (dentry_offset + 7) & ~7; + + /* Allocate memory for the root dentry and read it into memory */ dentry = MALLOC(sizeof(struct dentry)); if (!dentry) { ERROR("Failed to allocate %zu bytes for root dentry", @@ -1069,13 +1167,9 @@ int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) ret = WIMLIB_ERR_NOMEM; goto out_free_security_data; } - - get_u32(buf, &dentry_offset); - if (dentry_offset == 0) - dentry_offset = 8; - dentry_offset = (dentry_offset + 7) & ~7; - ret = read_dentry(buf, res_entry->original_size, dentry_offset, dentry); + ret = read_dentry(buf, metadata_len, dentry_offset, dentry); + /* This is the root dentry, so set its pointers correctly. */ dentry->parent = dentry; dentry->next = dentry; @@ -1083,20 +1177,20 @@ int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) if (ret != 0) goto out_free_dentry_tree; + /* Now read the entire directory entry tree into memory. */ DEBUG("Reading dentry tree"); - /* Now read the entire directory entry tree. */ - ret = read_dentry_tree(buf, res_entry->original_size, dentry); + ret = read_dentry_tree(buf, metadata_len, dentry); if (ret != 0) goto out_free_dentry_tree; - DEBUG("Calculating dentry full paths"); /* Calculate the full paths in the dentry tree. */ + DEBUG("Calculating dentry full paths"); ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL); if (ret != 0) goto out_free_dentry_tree; - DEBUG("Building link group table"); /* Build hash table that maps hard link group IDs to dentry sets */ + DEBUG("Building link group table"); lgt = new_link_group_table(9001); if (!lgt) goto out_free_dentry_tree; @@ -1108,10 +1202,15 @@ int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) ret = link_groups_free_duplicate_data(lgt); if (ret != 0) goto out_free_lgt; + + DEBUG("Running miscellaneous verifications on the dentry tree"); + ret = for_dentry_in_tree(dentry, verify_dentry, w); + if (ret != 0) + goto out_free_lgt; + DEBUG("Done reading image metadata"); imd->lgt = lgt; - imd->security_data = sd; imd->root_dentry = dentry; goto out_free_buf; out_free_lgt: @@ -1119,44 +1218,51 @@ out_free_lgt: out_free_dentry_tree: free_dentry_tree(dentry, NULL); out_free_security_data: - free_security_data(sd); + free_security_data(imd->security_data); + imd->security_data = NULL; out_free_buf: 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]; + const struct wim_security_data *sd; + 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; + sd = wim_security_data(w); + + /* We do not allow the security data pointer to be NULL, although it may + * point to an empty security data with no entries. */ + wimlib_assert(sd); + + /* Offset of first child of the root dentry. It's equal to: + * - The total length of the security data, rounded to the next 8-byte + * boundary, + * - plus the total length of the root dentry, + * - plus 8 bytes for an end-of-directory entry following the root + * dentry (shouldn't really be needed, but just in case...) + */ + subdir_offset = ((sd->total_length + 7) & ~7) + dentry_total_length(root) + 8; - 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; + /* Calculate the subdirectory offsets for the entire dentry tree. */ calculate_subdir_offsets(root, &subdir_offset); - metadata_original_size = subdir_offset; + + /* Total length of the metadata resource (uncompressed) */ + metadata_original_size = subdir_offset + random_tail_len; + + /* Allocate a buffer to contain the uncompressed metadata resource */ buf = MALLOC(metadata_original_size); if (!buf) { ERROR("Failed to allocate %"PRIu64" bytes for " @@ -1164,31 +1270,59 @@ int write_metadata_resource(WIMStruct *w) return WIMLIB_ERR_NOMEM; } + /* Write the security data into the resource buffer */ p = write_security_data(sd, buf); + /* Write the dentry tree into the resource buffer */ 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); + /* + * 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. + */ + randomize_byte_array(p, random_tail_len); + /* We MUST have exactly filled the buffer; otherwise we calculated its + * size incorrectly or wrote the data incorrectly. */ + wimlib_assert(p - buf + random_tail_len == metadata_original_size); + /* Get the lookup table entry for the metadata resource so we can update + * it. */ lte = wim_metadata_lookup_table_entry(w); + /* Write the metadata resource to the output WIM using the proper + * compression type. The lookup table entry for the metadata resource + * is updated. */ ret = write_wim_resource_from_buffer(buf, metadata_original_size, - hash, out, metadata_ctype, - <e->output_resource_entry); + w->out_fp, + wimlib_get_compression_type(w), + <e->output_resource_entry, + lte->hash); + if (ret != 0) + goto out; + /* It's very likely the SHA1 message digest of the metadata resource, so + * re-insert the lookup table entry into the lookup table. */ lookup_table_unlink(w->lookup_table, lte); - copy_hash(lte->hash, hash); lookup_table_insert(w->lookup_table, lte); - lte->out_refcnt++; + + /* We do not allow a metadata resource to be referenced multiple times, + * and the 20 random bytes appended to it should make it extremely + * likely for each metadata resource to be unique, even if the exact + * same image is captured. */ + wimlib_assert(lte->out_refcnt == 0); + lte->out_refcnt = 1; + + /* Make sure that the resource entry is written marked with the metadata + * flag. */ lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA; +out: + /* All the data has been written to the new WIM; no need for the buffer + * anymore */ FREE(buf); - if (ret != 0) - return ret; - - return 0; + return ret; }