X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fblob_table.c;h=e1fb6adac77c3c46603468c9e1be058d450270f7;hp=bcccae8ba9ddb1ab96ef362cdab176756dcb375f;hb=3d9eb49da969c4ac2c8579db7baf96ae3fd04e1c;hpb=f8223526610a6ed9410ed6ce2e704ef15ab8cade diff --git a/src/blob_table.c b/src/blob_table.c index bcccae8b..e1fb6ada 100644 --- a/src/blob_table.c +++ b/src/blob_table.c @@ -34,6 +34,7 @@ #include /* for unlink() */ #include "wimlib/assert.h" +#include "wimlib/bitops.h" #include "wimlib/blob_table.h" #include "wimlib/encoding.h" #include "wimlib/endianness.h" @@ -49,15 +50,25 @@ struct blob_table { struct hlist_head *array; size_t num_blobs; - size_t capacity; + size_t mask; /* capacity - 1; capacity is a power of 2 */ }; +static size_t +next_power_of_2(size_t n) +{ + if (n <= 1) + return 1; + return (size_t)1 << (1 + flsw(n - 1)); +} + struct blob_table * new_blob_table(size_t capacity) { struct blob_table *table; struct hlist_head *array; + capacity = next_power_of_2(capacity); + table = MALLOC(sizeof(struct blob_table)); if (table == NULL) goto oom; @@ -69,7 +80,7 @@ new_blob_table(size_t capacity) } table->num_blobs = 0; - table->capacity = capacity; + table->mask = capacity - 1; table->array = array; return table; @@ -99,7 +110,7 @@ free_blob_table(struct blob_table *table) struct blob_descriptor * new_blob_descriptor(void) { - BUILD_BUG_ON(BLOB_NONEXISTENT != 0); + STATIC_ASSERT(BLOB_NONEXISTENT == 0); return CALLOC(1, sizeof(struct blob_descriptor)); } @@ -124,8 +135,8 @@ clone_blob_descriptor(const struct blob_descriptor *old) #endif #ifdef WITH_FUSE case BLOB_IN_STAGING_FILE: - BUILD_BUG_ON((void*)&old->file_on_disk != - (void*)&old->staging_file_name); + STATIC_ASSERT((void*)&old->file_on_disk == + (void*)&old->staging_file_name); #endif new->file_on_disk = TSTRDUP(old->file_on_disk); if (new->file_on_disk == NULL) @@ -138,26 +149,11 @@ clone_blob_descriptor(const struct blob_descriptor *old) break; #ifdef WITH_NTFS_3G case BLOB_IN_NTFS_VOLUME: - if (old->ntfs_loc) { - new->ntfs_loc = memdup(old->ntfs_loc, - sizeof(struct ntfs_location)); - if (new->ntfs_loc == NULL) - goto out_free; - new->ntfs_loc->path = STRDUP(old->ntfs_loc->path); - new->ntfs_loc->attr_name = NULL; - if (new->ntfs_loc->path == NULL) - goto out_free; - if (new->ntfs_loc->attr_name_nchars != 0) { - new->ntfs_loc->attr_name = - utf16le_dup(old->ntfs_loc->attr_name); - if (new->ntfs_loc->attr_name == NULL) - goto out_free; - } - } + new->ntfs_loc = clone_ntfs_location(old->ntfs_loc); + if (!new->ntfs_loc) + goto out_free; break; #endif - default: - break; } return new; @@ -166,15 +162,22 @@ out_free: return NULL; } -static void +/* Release a blob descriptor from its location, if any, and set its new location + * to BLOB_NONEXISTENT. */ +void blob_release_location(struct blob_descriptor *blob) { switch (blob->blob_location) { - case BLOB_IN_WIM: + case BLOB_IN_WIM: { + struct wim_resource_descriptor *rdesc = blob->rdesc; + list_del(&blob->rdesc_node); - if (list_empty(&blob->rdesc->blob_list)) - FREE(blob->rdesc); + if (list_empty(&rdesc->blob_list)) { + wim_decrement_refcnt(rdesc->wim); + FREE(rdesc); + } break; + } case BLOB_IN_FILE_ON_DISK: #ifdef __WIN32__ case BLOB_IN_WINNT_FILE_ON_DISK: @@ -182,26 +185,22 @@ blob_release_location(struct blob_descriptor *blob) #endif #ifdef WITH_FUSE case BLOB_IN_STAGING_FILE: - BUILD_BUG_ON((void*)&blob->file_on_disk != - (void*)&blob->staging_file_name); + STATIC_ASSERT((void*)&blob->file_on_disk == + (void*)&blob->staging_file_name); #endif case BLOB_IN_ATTACHED_BUFFER: - BUILD_BUG_ON((void*)&blob->file_on_disk != - (void*)&blob->attached_buffer); + STATIC_ASSERT((void*)&blob->file_on_disk == + (void*)&blob->attached_buffer); FREE(blob->file_on_disk); break; #ifdef WITH_NTFS_3G case BLOB_IN_NTFS_VOLUME: - if (blob->ntfs_loc) { - FREE(blob->ntfs_loc->path); - FREE(blob->ntfs_loc->attr_name); - FREE(blob->ntfs_loc); - } + if (blob->ntfs_loc) + free_ntfs_location(blob->ntfs_loc); break; #endif - default: - break; } + blob->blob_location = BLOB_NONEXISTENT; } void @@ -307,7 +306,7 @@ blob_decrement_num_opened_fds(struct blob_descriptor *blob) static void blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob) { - size_t i = blob->hash_short % table->capacity; + size_t i = blob->hash_short & table->mask; hlist_add_head(&blob->hash_list, &table->array[i]); } @@ -321,21 +320,18 @@ enlarge_blob_table(struct blob_table *table) struct hlist_node *tmp; size_t i; - old_capacity = table->capacity; + old_capacity = table->mask + 1; new_capacity = old_capacity * 2; new_array = CALLOC(new_capacity, sizeof(struct hlist_head)); if (new_array == NULL) return; old_array = table->array; table->array = new_array; - table->capacity = new_capacity; + table->mask = new_capacity - 1; - for (i = 0; i < old_capacity; i++) { - hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) { - hlist_del(&blob->hash_list); + for (i = 0; i < old_capacity; i++) + hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) blob_table_insert_raw(table, blob); - } - } FREE(old_array); } @@ -344,7 +340,7 @@ void blob_table_insert(struct blob_table *table, struct blob_descriptor *blob) { blob_table_insert_raw(table, blob); - if (++table->num_blobs > table->capacity) + if (table->num_blobs++ > table->mask) enlarge_blob_table(table); } @@ -367,7 +363,7 @@ lookup_blob(const struct blob_table *table, const u8 *hash) size_t i; struct blob_descriptor *blob; - i = load_size_t_unaligned(hash) % table->capacity; + i = load_size_t_unaligned(hash) & table->mask; hlist_for_each_entry(blob, &table->array[i], hash_list) if (hashes_equal(hash, blob->hash)) return blob; @@ -384,7 +380,7 @@ for_blob_in_table(struct blob_table *table, struct hlist_node *tmp; int ret; - for (size_t i = 0; i < table->capacity; i++) { + for (size_t i = 0; i <= table->mask; i++) { hlist_for_each_entry_safe(blob, tmp, &table->array[i], hash_list) { @@ -415,7 +411,12 @@ cmp_blobs_by_sequential_order(const void *p1, const void *p2) v = (int)blob1->blob_location - (int)blob2->blob_location; - /* Different resource locations? */ + /* Different locations? Note: "unsafe compaction mode" requires that + * blobs in WIMs sort before all others. For the logic here to ensure + * this, BLOB_IN_WIM must have the lowest value among all defined + * blob_locations. Statically verify that the enum values haven't + * changed. */ + STATIC_ASSERT(BLOB_NONEXISTENT == 0 && BLOB_IN_WIM == 1); if (v) return v; @@ -424,22 +425,39 @@ cmp_blobs_by_sequential_order(const void *p1, const void *p2) wim1 = blob1->rdesc->wim; wim2 = blob2->rdesc->wim; - /* Different (possibly split) WIMs? */ + /* Different WIM files? */ if (wim1 != wim2) { - v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN); + + /* Resources from the WIM file currently being compacted + * (if any) must always sort first. */ + v = (int)wim2->being_compacted - (int)wim1->being_compacted; if (v) return v; + + /* Different split WIMs? */ + v = cmp_guids(wim1->hdr.guid, wim2->hdr.guid); + if (v) + return v; + + /* Different part numbers in the same split WIM? */ + v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number; + if (v) + return v; + + /* Probably two WIMStructs for the same on-disk file. + * Just sort by pointer. */ + return wim1 < wim2 ? -1 : 1; } - /* Different part numbers in the same WIM? */ - v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number; - if (v) - return v; + /* Same WIM file */ + /* Sort by increasing resource offset */ if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim) return cmp_u64(blob1->rdesc->offset_in_wim, blob2->rdesc->offset_in_wim); + /* The blobs are in the same solid resource. Sort by increasing + * offset in the resource. */ return cmp_u64(blob1->offset_in_res, blob2->offset_in_res); case BLOB_IN_FILE_ON_DISK: @@ -449,13 +467,17 @@ cmp_blobs_by_sequential_order(const void *p1, const void *p2) #ifdef __WIN32__ case BLOB_IN_WINNT_FILE_ON_DISK: case BLOB_WIN32_ENCRYPTED: + /* Windows: compare by starting LCN (logical cluster number) */ + v = cmp_u64(blob1->sort_key, blob2->sort_key); + if (v) + return v; #endif /* Compare files by path: just a heuristic that will place files * in the same directory next to each other. */ return tstrcmp(blob1->file_on_disk, blob2->file_on_disk); #ifdef WITH_NTFS_3G case BLOB_IN_NTFS_VOLUME: - return tstrcmp(blob1->ntfs_loc->path, blob2->ntfs_loc->path); + return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc); #endif default: /* No additional sorting order defined for this resource @@ -627,7 +649,7 @@ do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs, rdesc = rdescs[i]; - wim_res_hdr_to_desc(&reshdr, wim, rdesc); + wim_reshdr_to_desc(&reshdr, wim, rdesc); /* For solid resources, the uncompressed size, compression type, * and chunk size are stored in the resource itself, not in the @@ -646,22 +668,12 @@ do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs, /* Compression format numbers must be the same as in * WIMGAPI to be compatible here. */ - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_NONE == 0); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3); rdesc->compression_type = le32_to_cpu(hdr.compression_format); - rdesc->chunk_size = le32_to_cpu(hdr.chunk_size); - - DEBUG("Solid resource %zu/%zu: %"PRIu64" => %"PRIu64" " - "(%"TS"/%"PRIu32") @ +%"PRIu64"", - i + 1, num_rdescs, - rdesc->uncompressed_size, - rdesc->size_in_wim, - wimlib_get_compression_type_string(rdesc->compression_type), - rdesc->chunk_size, - rdesc->offset_in_wim); } return 0; } @@ -703,6 +715,8 @@ load_solid_info(WIMStruct *wim, if (ret) goto out_free_rdescs; + wim->refcnt += num_rdescs; + *rdescs_ret = rdescs; *num_rdescs_ret = num_rdescs; return 0; @@ -743,9 +757,12 @@ static void free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs) { if (rdescs) { - for (size_t i = 0; i < num_rdescs; i++) - if (list_empty(&rdescs[i]->blob_list)) + for (size_t i = 0; i < num_rdescs; i++) { + if (list_empty(&rdescs[i]->blob_list)) { + rdescs[i]->wim->refcnt--; FREE(rdescs[i]); + } + } FREE(rdescs); } } @@ -872,13 +889,12 @@ read_blob_table(WIMStruct *wim) struct blob_table *table = NULL; struct blob_descriptor *cur_blob = NULL; size_t num_duplicate_blobs = 0; + size_t num_empty_blobs = 0; size_t num_wrong_part_blobs = 0; u32 image_index = 0; struct wim_resource_descriptor **cur_solid_rdescs = NULL; size_t cur_num_solid_rdescs = 0; - DEBUG("Reading blob table."); - /* Calculate the number of entries in the blob table. */ num_entries = wim->hdr.blob_table_reshdr.uncompressed_size / sizeof(struct blob_descriptor_disk); @@ -890,7 +906,7 @@ read_blob_table(WIMStruct *wim) /* Allocate a hash table to map SHA-1 message digests into blob * descriptors. This is the in-memory "blob table". */ - table = new_blob_table(num_entries * 2 + 1); + table = new_blob_table(num_entries); if (!table) goto oom; @@ -905,13 +921,6 @@ read_blob_table(WIMStruct *wim) /* Get the resource header */ get_wim_reshdr(&disk_entry->reshdr, &reshdr); - DEBUG("reshdr: size_in_wim=%"PRIu64", " - "uncompressed_size=%"PRIu64", " - "offset_in_wim=%"PRIu64", " - "flags=0x%02x", - reshdr.size_in_wim, reshdr.uncompressed_size, - reshdr.offset_in_wim, reshdr.flags); - /* Ignore SOLID flag if it isn't supposed to be used in this WIM * version. */ if (wim->hdr.wim_version == WIM_VERSION_DEFAULT) @@ -969,27 +978,13 @@ read_blob_table(WIMStruct *wim) goto out; } - /* How to handle an uncompressed resource with its - * uncompressed size different from its compressed size? - * - * Based on a simple test, WIMGAPI seems to handle this - * as follows: - * - * if (size_in_wim > uncompressed_size) { - * Ignore uncompressed_size; use size_in_wim - * instead. - * } else { - * Honor uncompressed_size, but treat the part of - * the file data above size_in_wim as all zeros. - * } - * - * So we will do the same. */ - if (unlikely(!(reshdr.flags & - WIM_RESHDR_FLAG_COMPRESSED) && - (reshdr.size_in_wim > - reshdr.uncompressed_size))) + if (unlikely(!(reshdr.flags & WIM_RESHDR_FLAG_COMPRESSED) && + (reshdr.size_in_wim != reshdr.uncompressed_size))) { - reshdr.uncompressed_size = reshdr.size_in_wim; + ERROR("Uncompressed resource has " + "size_in_wim != uncompressed_size"); + ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY; + goto out; } /* Set up a resource descriptor for this blob. */ @@ -998,20 +993,25 @@ read_blob_table(WIMStruct *wim) if (!rdesc) goto oom; - wim_res_hdr_to_desc(&reshdr, wim, rdesc); - - blob_set_is_located_in_nonsolid_wim_resource(cur_blob, rdesc); + wim_reshdr_to_desc_and_blob(&reshdr, wim, rdesc, cur_blob); + wim->refcnt++; } /* cur_blob is now a blob bound to a resource. */ /* Ignore entries with all zeroes in the hash field. */ - if (is_zero_hash(cur_blob->hash)) + if (unlikely(is_zero_hash(cur_blob->hash))) goto free_cur_blob_and_continue; + /* Verify that the blob has nonzero size. */ + if (unlikely(cur_blob->size == 0)) { + num_empty_blobs++; + goto free_cur_blob_and_continue; + } + /* Verify that the part number matches that of the underlying * WIM file. */ - if (part_number != wim->hdr.part_number) { + if (unlikely(part_number != wim->hdr.part_number)) { num_wrong_part_blobs++; goto free_cur_blob_and_continue; } @@ -1038,6 +1038,13 @@ read_blob_table(WIMStruct *wim) goto out; } + if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) { + ERROR("Image metadata in solid resources " + "is unsupported."); + ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY; + goto out; + } + if (wim->hdr.part_number != 1) { WARNING("Ignoring metadata resource found in a " "non-first part of the split WIM"); @@ -1059,11 +1066,6 @@ read_blob_table(WIMStruct *wim) * this overrides the actual locations of the metadata * resources themselves in the WIM file as well as any * information written in the XML data. */ - DEBUG("Found metadata resource for image %"PRIu32" at " - "offset %"PRIu64".", - image_index + 1, - reshdr.offset_in_wim); - wim->image_metadata[image_index++]->metadata_blob = cur_blob; } else { /* Blob table entry for a non-metadata blob. */ @@ -1107,12 +1109,14 @@ read_blob_table(WIMStruct *wim) if (num_duplicate_blobs > 0) WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs); + if (num_empty_blobs > 0) + WARNING("Ignoring %zu empty blobs", num_empty_blobs); + if (num_wrong_part_blobs > 0) { WARNING("Ignoring %zu blobs with wrong part number", num_wrong_part_blobs); } - DEBUG("Done reading blob table."); wim->blob_table = table; ret = 0; goto out_free_buf; @@ -1172,9 +1176,6 @@ write_blob_table_from_blob_list(struct list_head *blob_list, } } - DEBUG("Writing WIM blob table (size=%zu, offset=%"PRIu64")", - table_size, out_fd->offset); - table_buf = MALLOC(table_size); if (table_buf == NULL) { ERROR("Failed to allocate %zu bytes for temporary blob table", @@ -1233,7 +1234,6 @@ write_blob_table_from_blob_list(struct list_head *blob_list, NULL, write_resource_flags); FREE(table_buf); - DEBUG("ret=%d", ret); return ret; } @@ -1357,7 +1357,8 @@ blob_to_wimlib_resource_entry(const struct blob_descriptor *blob, wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0; wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0; } - copy_hash(wentry->sha1_hash, blob->hash); + if (!blob->unhashed) + copy_hash(wentry->sha1_hash, blob->hash); wentry->reference_count = blob->refcnt; wentry->is_metadata = blob->is_metadata; } @@ -1393,10 +1394,17 @@ wimlib_iterate_lookup_table(WIMStruct *wim, int flags, if (wim_has_metadata(wim)) { int ret; for (int i = 0; i < wim->hdr.image_count; i++) { - ret = do_iterate_blob(wim->image_metadata[i]->metadata_blob, - &ctx); + struct blob_descriptor *blob; + struct wim_image_metadata *imd = wim->image_metadata[i]; + + ret = do_iterate_blob(imd->metadata_blob, &ctx); if (ret) return ret; + image_for_each_unhashed_blob(blob, imd) { + ret = do_iterate_blob(blob, &ctx); + if (ret) + return ret; + } } } return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);