From 26c7f8bb32e4a32001d409f1693e0df016270ed5 Mon Sep 17 00:00:00 2001 From: Eric Biggers Date: Mon, 23 Dec 2013 02:14:37 -0600 Subject: [PATCH] Rewrite of write_stream_list(), and writing packed resources --- Makefile.am | 4 + include/wimlib.h | 8 +- include/wimlib/header.h | 48 +- include/wimlib/list.h | 23 + include/wimlib/lookup_table.h | 117 +- include/wimlib/ntfs_3g.h | 3 +- include/wimlib/resource.h | 50 +- include/wimlib/util.h | 11 + include/wimlib/write.h | 3 +- programs/imagex.c | 35 +- src/extract.c | 10 +- src/lookup_table.c | 327 ++-- src/ntfs-3g_capture.c | 7 +- src/resource.c | 453 ++--- src/split.c | 9 +- src/wim.c | 8 +- src/write.c | 3114 +++++++++++++++------------------ src/xml.c | 2 +- 18 files changed, 1972 insertions(+), 2260 deletions(-) diff --git a/Makefile.am b/Makefile.am index 223f884e..7b20b58a 100644 --- a/Makefile.am +++ b/Makefile.am @@ -21,6 +21,9 @@ libwim_la_SOURCES = \ src/add_image.c \ src/capture_common.c \ src/compress.c \ + src/compress_chunk.c \ + src/compress_parallel.c \ + src/compress_serial.c \ src/decompress.c \ src/delete_image.c \ src/dentry.c \ @@ -67,6 +70,7 @@ libwim_la_SOURCES = \ include/wimlib/capture.h \ include/wimlib/compiler.h \ include/wimlib/compress.h \ + include/wimlib/compress_chunks.h \ include/wimlib/decompress.h \ include/wimlib/dentry.h \ include/wimlib/encoding.h \ diff --git a/include/wimlib.h b/include/wimlib.h index 930422a8..efc770cf 100644 --- a/include/wimlib.h +++ b/include/wimlib.h @@ -987,7 +987,7 @@ struct wimlib_resource_entry { */ uint32_t is_missing : 1; - uint32_t is_partial : 1; + uint32_t is_packed_streams : 1; uint32_t reserved_flags : 26; @@ -1513,6 +1513,12 @@ typedef int (*wimlib_iterate_lookup_table_callback_t)(const struct wimlib_resour #define WIMLIB_WRITE_FLAG_RESERVED 0x00000800 +/** TODO */ +#define WIMLIB_WRITE_FLAG_PACK_STREAMS 0x00001000 + +/** TODO */ +#define WIMLIB_WRITE_FLAG_NO_PACK_STREAMS 0x00002000 + /** @} */ /** @ingroup G_general * @{ */ diff --git a/include/wimlib/header.h b/include/wimlib/header.h index 5f26f7c6..d67811c1 100644 --- a/include/wimlib/header.h +++ b/include/wimlib/header.h @@ -49,63 +49,65 @@ /* On-disk format of the WIM header. */ struct wim_header_disk { - /* Magic characters "MSWIM\0\0\0" */ + /* +0x00: Magic characters "MSWIM\0\0\0" */ le64 magic; - /* Size of the WIM header, in bytes; WIM_HEADER_DISK_SIZE expected - * (currently the only supported value). */ + /* +0x08: Size of the WIM header, in bytes; WIM_HEADER_DISK_SIZE + * expected (currently the only supported value). */ u32 hdr_size; - /* Version of the WIM file + /* +0x0c: Version of the WIM file * TODO */ u32 wim_version; - /* Flags for the WIM file (WIM_HDR_FLAG_*) */ + /* +0x10: Flags for the WIM file (WIM_HDR_FLAG_*) */ u32 wim_flags; - /* Chunk size for compressed resources in the WIM, or 0 if the WIM is - * uncompressed. */ + /* +0x14: Chunk size for compressed resources in the WIM, or 0 if the + * WIM is uncompressed. */ u32 chunk_size; - /* Globally unique identifier for the WIM file. Basically a bunch of - * random bytes. */ + /* +0x18: Globally unique identifier for the WIM file. Basically a + * bunch of random bytes. */ u8 guid[WIM_GID_LEN]; - /* Number of this WIM part in the split WIM file, indexed from 1, or 1 - * if the WIM is not split. */ + /* +0x28: Number of this WIM part in the split WIM file, indexed from 1, + * or 1 if the WIM is not split. */ u16 part_number; - /* Total number of parts of the split WIM file, or 1 if the WIM is not - * split. */ + /* +0x2a: Total number of parts of the split WIM file, or 1 if the WIM + * is not split. */ u16 total_parts; - /* Number of images in the WIM. */ + /* +0x2c: Number of images in the WIM. */ u32 image_count; - /* Location and size of the WIM's lookup table. */ + /* +0x30: Location and size of the WIM's lookup table. */ struct wim_reshdr_disk lookup_table_reshdr; - /* Location and size of the WIM's XML data. */ + /* +0x48: Location and size of the WIM's XML data. */ struct wim_reshdr_disk xml_data_reshdr; - /* Location and size of metadata resource for the bootable image of the - * WIM, or all zeroes if no image is bootable. */ + /* +0x60: Location and size of metadata resource for the bootable image + * of the WIM, or all zeroes if no image is bootable. */ struct wim_reshdr_disk boot_metadata_reshdr; - /* 1-based index of the bootable image of the WIM, or 0 if no image is - * bootable. */ + /* +0x78: 1-based index of the bootable image of the WIM, or 0 if no + * image is bootable. */ u32 boot_idx; - /* Location and size of the WIM's integrity table, or all zeroes if the - * WIM has no integrity table. + /* +0x7c: Location and size of the WIM's integrity table, or all zeroes + * if the WIM has no integrity table. * * Note the integrity_table_reshdr here is 4-byte aligned even though * it would ordinarily be 8-byte aligned--- hence, the _packed_attribute * on the `struct wim_header_disk' is essential. */ struct wim_reshdr_disk integrity_table_reshdr; - /* Unused bytes. */ + /* +0x94: Unused bytes. */ u8 unused[60]; + + /* +0xd0 (208) */ } _packed_attribute; diff --git a/include/wimlib/list.h b/include/wimlib/list.h index d6bd55c1..4ac2d0b4 100644 --- a/include/wimlib/list.h +++ b/include/wimlib/list.h @@ -86,6 +86,29 @@ static inline void list_add_tail(struct list_head *new, struct list_head *head) __list_add(new, head->prev, head); } +/** + * list_replace - replace old entry by new one + * @old : the element to be replaced + * @new : the new element to insert + * + * If @old was empty, it will be overwritten. + */ +static inline void list_replace(struct list_head *old, + struct list_head *new) +{ + new->next = old->next; + new->next->prev = new; + new->prev = old->prev; + new->prev->next = new; +} + +static inline void list_replace_init(struct list_head *old, + struct list_head *new) +{ + list_replace(old, new); + INIT_LIST_HEAD(old); +} + /* * Delete a list entry by making the prev/next entries * point to each other. diff --git a/include/wimlib/lookup_table.h b/include/wimlib/lookup_table.h index f0cfb768..6a934e49 100644 --- a/include/wimlib/lookup_table.h +++ b/include/wimlib/lookup_table.h @@ -118,39 +118,24 @@ struct wim_lookup_table_entry { u64 size; /* Stream flags (WIM_RESHDR_FLAG_*). */ - u16 flags : 8; + u32 flags : 8; /* One of the `enum resource_location' values documented above. */ - u16 resource_location : 5; - - /* 1 if this stream is a unique size (only set while writing streams). */ - u16 unique_size : 1; + u32 resource_location : 5; /* 1 if this stream has not had a SHA1 message digest calculated for it * yet */ - u16 unhashed : 1; - - u16 deferred : 1; + u32 unhashed : 1; - u16 no_progress : 1; + /* Temoorary files used for writing; set as documented for + * prepare_stream_list_for_write(). */ + u32 unique_size : 1; + u32 will_be_in_output_wim : 1; /* Set to 1 when a metadata entry has its checksum changed; in such * cases the hash is no longer valid to verify the data if the metadata * resource is read again. */ - u16 dont_check_metadata_hash : 1; - - /* Only used during WIM write. Normal value is 0 (resource not - * filtered). */ - u16 filtered : 2; -#define FILTERED_SAME_WIM 0x1 /* Resource already in same WIM */ -#define FILTERED_EXTERNAL_WIM 0x2 /* Resource already in external WIM */ - - /* (On-disk field) - * Number of times this lookup table entry is referenced by dentries. - * Unfortunately, this field is not always set correctly in Microsoft's - * WIMs, so we have no choice but to fix it if more references to the - * lookup table entry are found than stated here. */ - u32 refcnt; + u32 dont_check_metadata_hash : 1; union { /* (On-disk field) SHA1 message digest of the stream referenced @@ -176,12 +161,26 @@ struct wim_lookup_table_entry { }; }; + /* Number of times this lookup table entry is referenced by dentries in + * the WIM. */ + u32 refcnt; + + /* Actual reference count to this stream (only used while verifying an + * image). */ + u32 real_refcnt; + /* When a WIM file is written, out_refcnt starts at 0 and is incremented * whenever the stream pointed to by this lookup table entry needs to be * written. The stream only need to be written when out_refcnt is * nonzero, since otherwise it is not referenced by any dentries. */ u32 out_refcnt; +#ifdef WITH_FUSE + /* Number of times this stream has been opened (used only during + * mounting) */ + u16 num_opened_fds; +#endif + /* Pointers to somewhere where the stream is actually located. See the * comments for the @resource_location field above. */ union { @@ -199,48 +198,38 @@ struct wim_lookup_table_entry { #endif }; - /* Actual reference count to this stream (only used while - * verifying an image). */ - u32 real_refcnt; - - union { - #ifdef WITH_FUSE - /* Number of times this stream has been opened (used only during - * mounting) */ - u16 num_opened_fds; - #endif + /* Links together streams that share the same underlying WIM resource. + * The head is wim_resource_spec.stream_list. */ + struct list_head rspec_node; - /* This field is used for the special hardlink or symlink image - * extraction mode. In these mode, all identical files are linked - * together, and @extracted_file will be set to the filename of the - * first extracted file containing this stream. */ - tchar *extracted_file; - }; + /* This field is used for the special hardlink or symlink image + * extraction mode. In these mode, all identical files are linked + * together, and @extracted_file will be set to the filename of the + * first extracted file containing this stream. */ + tchar *extracted_file; /* Temporary fields */ union { /* Used temporarily during WIM file writing */ struct { - struct hlist_node hash_list_2; - + union { + struct hlist_node hash_list_2; + struct { + u64 out_res_offset_in_wim; + u64 out_res_size_in_wim; + u64 out_res_uncompressed_size; + }; + }; /* Links streams being written to the WIM. */ struct list_head write_streams_list; - }; - /* Used temporarily during WIM file writing (after above) */ - struct { - struct list_head msg_list; - struct list_head being_compressed_list; + struct wim_reshdr out_reshdr; }; - /* When a WIM file is written, @output_reshdr is filled in with - * the resource header for the output WIM. */ - struct wim_reshdr out_reshdr; - /* Used temporarily during extraction */ union { /* out_refcnt tracks number of slots filled */ - struct wim_dentry *inline_lte_dentries[4]; + struct wim_dentry *inline_lte_dentries[8]; struct { struct wim_dentry **lte_dentries; unsigned long alloc_lte_dentries; @@ -250,7 +239,7 @@ struct wim_lookup_table_entry { /* Temporary list fields */ union { - /* Links streams when writing lookup table. */ + /* Links streams for writing lookup table. */ struct list_head lookup_table_list; /* Links streams being extracted. */ @@ -260,11 +249,9 @@ struct wim_lookup_table_entry { struct list_head export_stream_list; }; - /* Links streams that are still unhashed after being been added - * to a WIM. */ + /* Links streams that are still unhashed after being been added to a + * WIM. */ struct list_head unhashed_list; - - struct list_head wim_resource_list; }; static inline bool @@ -294,9 +281,12 @@ extern int read_wim_lookup_table(WIMStruct *wim); extern int -write_wim_lookup_table(WIMStruct *wim, int image, int write_flags, - struct wim_reshdr *out_reshdr, - struct list_head *stream_list_override); +write_wim_lookup_table_from_stream_list(struct list_head *stream_list, + struct filedes *out_fd, + u16 part_number, + struct wim_reshdr *out_reshdr, + int write_resource_flags, + struct wimlib_lzx_context **comp_ctx); extern void free_lookup_table(struct wim_lookup_table *table); @@ -336,6 +326,11 @@ for_lookup_table_entry(struct wim_lookup_table *table, int (*visitor)(struct wim_lookup_table_entry *, void *), void *arg); +extern int +sort_stream_list(struct list_head *stream_list, + size_t list_head_offset, + int (*compar)(const void *, const void*)); + extern int sort_stream_list_by_sequential_order(struct list_head *stream_list, size_t list_head_offset); @@ -379,13 +374,13 @@ lte_bind_wim_resource_spec(struct wim_lookup_table_entry *lte, { lte->resource_location = RESOURCE_IN_WIM; lte->rspec = rspec; - list_add_tail(<e->wim_resource_list, &rspec->stream_list); + list_add_tail(<e->rspec_node, &rspec->stream_list); } static inline void lte_unbind_wim_resource_spec(struct wim_lookup_table_entry *lte) { - list_del(<e->wim_resource_list); + list_del(<e->rspec_node); lte->rspec = NULL; lte->resource_location = RESOURCE_NONEXISTENT; } diff --git a/include/wimlib/ntfs_3g.h b/include/wimlib/ntfs_3g.h index b8ff3723..00b9fbdd 100644 --- a/include/wimlib/ntfs_3g.h +++ b/include/wimlib/ntfs_3g.h @@ -14,8 +14,7 @@ extern int read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, consume_data_callback_t cb, - void *cb_ctx, - int _ignored_flags); + void *cb_ctx); extern int diff --git a/include/wimlib/resource.h b/include/wimlib/resource.h index 9b176879..3d58bd24 100644 --- a/include/wimlib/resource.h +++ b/include/wimlib/resource.h @@ -48,6 +48,9 @@ struct wim_resource_spec { * resource will be in a slightly different format if it is compressed. * This is a wimlib extension. */ u32 is_pipable : 1; + + /* Temporary flag. */ + u32 raw_copy_ok : 1; }; /* On-disk version of a WIM resource header. */ @@ -99,6 +102,10 @@ struct wim_reshdr { * should be WIM_VERSION_PACKED_STREAMS. */ #define WIM_RESHDR_FLAG_PACKED_STREAMS 0x10 +/* Magic number in the 'uncompressed_size' field of the resource header that + * identifies the main entry for a pack. */ +#define WIM_PACK_MAGIC_NUMBER 0x100000000ULL + /* Returns true if the specified WIM resource is compressed, using either the * original chunk table layout or the alternate layout for resources that may * contain multiple packed streams. */ @@ -137,17 +144,31 @@ void put_wim_reshdr(const struct wim_reshdr *reshdr, struct wim_reshdr_disk *disk_reshdr); -/* wimlib internal flags used when reading or writing resources. */ +/* Alternate chunk table format for resources with + * WIM_RESHDR_FLAG_PACKED_STREAMS set. */ +struct alt_chunk_table_header_disk { + /* Uncompressed size of the resource in bytes. */ + le64 res_usize; + + /* Number of bytes each compressed chunk decompresses into, except + * possibly the last which decompresses into the remainder. */ + le32 chunk_size; + + /* Compression format used for compressed chunks: + * 0 = None + * 1 = LZX + * 2 = XPRESS + * 3 = LZMS */ + le32 compression_format; + + /* This header is directly followed by a table of compressed sizes of + * the chunks. */ +} _packed_attribute; + +/* wimlib internal flags used when writing resources. */ #define WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001 #define WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE 0x00000002 -#define WIMLIB_WRITE_RESOURCE_MASK 0x0000ffff - -#define WIMLIB_READ_RESOURCE_FLAG_RAW_FULL 0x80000000 -#define WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS 0x40000000 -#define WIMLIB_READ_RESOURCE_FLAG_RAW (WIMLIB_READ_RESOURCE_FLAG_RAW_FULL | \ - WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) -#define WIMLIB_READ_RESOURCE_MASK 0xffff0000 - +#define WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004 /* Functions to read streams */ @@ -172,7 +193,7 @@ skip_wim_stream(struct wim_lookup_table_entry *lte); extern int read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, consume_data_callback_t cb, - void *cb_ctx, int flags); + void *cb_ctx); typedef int (*read_stream_list_begin_stream_t)(struct wim_lookup_table_entry *lte, bool is_partial_res, @@ -181,6 +202,11 @@ typedef int (*read_stream_list_end_stream_t)(struct wim_lookup_table_entry *lte, int status, void *ctx); +#define VERIFY_STREAM_HASHES 0x1 +#define COMPUTE_MISSING_STREAM_HASHES 0x2 +#define STREAM_LIST_ALREADY_SORTED 0x4 +#define BEGIN_STREAM_STATUS_SKIP_STREAM -1 + /* Callback functions and contexts for read_stream_list(). */ struct read_stream_list_callbacks { @@ -206,8 +232,8 @@ struct read_stream_list_callbacks { extern int read_stream_list(struct list_head *stream_list, size_t list_head_offset, - u32 cb_chunk_size, - const struct read_stream_list_callbacks *cbs); + const struct read_stream_list_callbacks *cbs, + int flags); /* Functions to extract streams. */ diff --git a/include/wimlib/util.h b/include/wimlib/util.h index cd0d599f..8e9d695e 100644 --- a/include/wimlib/util.h +++ b/include/wimlib/util.h @@ -138,6 +138,17 @@ hash_u64(u64 n) return n * 0x9e37fffffffc0001ULL; } +static inline int +cmp_u64(u64 n1, u64 n2) +{ + if (n1 < n2) + return -1; + else if (n1 > n2) + return 1; + else + return 0; +} + /* is_any_path_separator() - characters treated as path separators in WIM path * specifications and capture configuration files (the former will be translated * to WIM_PATH_SEPARATOR; the latter will be translated to diff --git a/include/wimlib/write.h b/include/wimlib/write.h index 1240c89f..9857fc03 100644 --- a/include/wimlib/write.h +++ b/include/wimlib/write.h @@ -46,7 +46,8 @@ write_wim_resource_from_buffer(const void *buf, size_t buf_size, int out_ctype, u32 out_chunk_size, struct wim_reshdr *out_reshdr, - u8 *hash_ret, int write_resource_flags, + u8 *hash, + int write_resource_flags, struct wimlib_lzx_context **comp_ctx); #endif /* _WIMLIB_WRITE_H */ diff --git a/programs/imagex.c b/programs/imagex.c index c3187f74..b983c7f3 100644 --- a/programs/imagex.c +++ b/programs/imagex.c @@ -142,7 +142,9 @@ enum { IMAGEX_NORPFIX_OPTION, IMAGEX_NOCHECK_OPTION, IMAGEX_NO_ACLS_OPTION, + IMAGEX_NO_PACK_STREAMS_OPTION, IMAGEX_NOT_PIPABLE_OPTION, + IMAGEX_PACK_STREAMS_OPTION, IMAGEX_PATH_OPTION, IMAGEX_PIPABLE_OPTION, IMAGEX_REBUILD_OPTION, @@ -192,6 +194,7 @@ static const struct option capture_or_append_options[] = { {T("compress"), required_argument, NULL, IMAGEX_COMPRESS_OPTION}, {T("compress-slow"), no_argument, NULL, IMAGEX_COMPRESS_SLOW_OPTION}, {T("chunk-size"), required_argument, NULL, IMAGEX_CHUNK_SIZE_OPTION}, + {T("pack-streams"), no_argument, NULL, IMAGEX_PACK_STREAMS_OPTION}, {T("config"), required_argument, NULL, IMAGEX_CONFIG_OPTION}, {T("dereference"), no_argument, NULL, IMAGEX_DEREFERENCE_OPTION}, {T("flags"), required_argument, NULL, IMAGEX_FLAGS_OPTION}, @@ -1024,22 +1027,6 @@ stdin_get_text_contents(size_t *num_tchars_ret) #define TO_PERCENT(numerator, denominator) \ (((denominator) == 0) ? 0 : ((numerator) * 100 / (denominator))) -/* Given an enumerated value for WIM compression type, return a descriptive - * string. */ -static const tchar * -get_data_type(int ctype) -{ - switch (ctype) { - case WIMLIB_COMPRESSION_TYPE_NONE: - return T("uncompressed"); - case WIMLIB_COMPRESSION_TYPE_LZX: - return T("LZX-compressed"); - case WIMLIB_COMPRESSION_TYPE_XPRESS: - return T("XPRESS-compressed"); - } - return NULL; -} - #define GIBIBYTE_MIN_NBYTES 10000000000ULL #define MEBIBYTE_MIN_NBYTES 10000000ULL #define KIBIBYTE_MIN_NBYTES 10000ULL @@ -1079,11 +1066,9 @@ imagex_progress_func(enum wimlib_progress_msg msg, info->write_streams.total_bytes); if (info->write_streams.completed_streams == 0) { - const tchar *data_type; - - data_type = get_data_type(info->write_streams.compression_type); - imagex_printf(T("Writing %"TS" data using %u thread%"TS"\n"), - data_type, info->write_streams.num_threads, + imagex_printf(T("Writing %"TS"-compressed data using %u thread%"TS"\n"), + wimlib_get_compression_type_string(info->write_streams.compression_type), + info->write_streams.num_threads, (info->write_streams.num_threads == 1) ? T("") : T("s")); } if (info->write_streams.total_parts <= 1) { @@ -1753,6 +1738,9 @@ imagex_capture_or_append(int argc, tchar **argv, int cmd) if (chunk_size == UINT32_MAX) goto out_err; break; + case IMAGEX_PACK_STREAMS_OPTION: + write_flags |= WIMLIB_WRITE_FLAG_PACK_STREAMS; + break; case IMAGEX_FLAGS_OPTION: flags_element = optarg; break; @@ -2773,6 +2761,7 @@ print_wim_information(const tchar *wimfile, const struct wimlib_wim_info *info) tprintf(T("GUID: 0x")); print_byte_field(info->guid, sizeof(info->guid)); tputchar(T('\n')); + tprintf(T("Version: %u\n"), info->wim_version); tprintf(T("Image Count: %d\n"), info->image_count); tprintf(T("Compression: %"TS"\n"), wimlib_get_compression_type_string(info->compression_type)); @@ -2796,7 +2785,7 @@ print_resource(const struct wimlib_resource_entry *resource, { tprintf(T("Uncompressed size = %"PRIu64" bytes\n"), resource->uncompressed_size); - if (resource->is_partial) { + if (resource->is_packed_streams) { tprintf(T("Raw uncompressed size = %"PRIu64" bytes\n"), resource->raw_resource_uncompressed_size); @@ -2833,6 +2822,8 @@ print_resource(const struct wimlib_resource_entry *resource, tprintf(T("WIM_RESHDR_FLAG_FREE ")); if (resource->is_spanned) tprintf(T("WIM_RESHDR_FLAG_SPANNED ")); + if (resource->is_packed_streams) + tprintf(T("WIM_RESHDR_FLAG_PACKED_STREAMS ")); tputchar(T('\n')); tputchar(T('\n')); return 0; diff --git a/src/extract.c b/src/extract.c index 3438ca73..a7e62697 100644 --- a/src/extract.c +++ b/src/extract.c @@ -1308,7 +1308,8 @@ need_tmpfile_to_extract(struct wim_lookup_table_entry *lte, static int begin_extract_stream_to_tmpfile(struct wim_lookup_table_entry *lte, - bool is_partial_res, void *_ctx) + bool is_partial_res, + void *_ctx) { struct apply_ctx *ctx = _ctx; int ret; @@ -1320,10 +1321,7 @@ begin_extract_stream_to_tmpfile(struct wim_lookup_table_entry *lte, if (ret) return ret; - /* Negative return value here means the function was successful, - * but the consume_chunk and end_chunk callbacks need not be - * called. */ - return -1; + return BEGIN_STREAM_STATUS_SKIP_STREAM; } DEBUG("Temporary file needed for stream (size=%"PRIu64")", lte->size); @@ -1390,7 +1388,7 @@ extract_stream_list(struct apply_ctx *ctx) }; return read_stream_list(&ctx->stream_list, offsetof(struct wim_lookup_table_entry, extraction_list), - 0, &cbs); + &cbs, VERIFY_STREAM_HASHES); } else { /* Extract the streams in unsorted order. */ struct wim_lookup_table_entry *lte; diff --git a/src/lookup_table.c b/src/lookup_table.c index 44ed7966..a899bf95 100644 --- a/src/lookup_table.c +++ b/src/lookup_table.c @@ -91,13 +91,13 @@ clone_lookup_table_entry(const struct wim_lookup_table_entry *old) struct wim_lookup_table_entry *new; new = memdup(old, sizeof(struct wim_lookup_table_entry)); - if (!new) + if (new == NULL) return NULL; new->extracted_file = NULL; switch (new->resource_location) { case RESOURCE_IN_WIM: - list_add(&new->wim_resource_list, &new->rspec->stream_list); + list_add(&new->rspec_node, &new->rspec->stream_list); break; case RESOURCE_IN_FILE_ON_DISK: @@ -110,12 +110,12 @@ clone_lookup_table_entry(const struct wim_lookup_table_entry *old) (void*)&old->staging_file_name); #endif new->file_on_disk = TSTRDUP(old->file_on_disk); - if (!new->file_on_disk) + if (new->file_on_disk == NULL) goto out_free; break; case RESOURCE_IN_ATTACHED_BUFFER: new->attached_buffer = memdup(old->attached_buffer, old->size); - if (!new->attached_buffer) + if (new->attached_buffer == NULL) goto out_free; break; #ifdef WITH_NTFS_3G @@ -123,18 +123,18 @@ clone_lookup_table_entry(const struct wim_lookup_table_entry *old) if (old->ntfs_loc) { struct ntfs_location *loc; loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location)); - if (!loc) + if (loc == NULL) goto out_free; loc->path = NULL; loc->stream_name = NULL; new->ntfs_loc = loc; loc->path = STRDUP(old->ntfs_loc->path); - if (!loc->path) + if (loc->path == NULL) goto out_free; - if (loc->stream_name_nchars) { + if (loc->stream_name_nchars != 0) { loc->stream_name = memdup(old->ntfs_loc->stream_name, loc->stream_name_nchars * 2); - if (!loc->stream_name) + if (loc->stream_name == NULL) goto out_free; } } @@ -155,7 +155,7 @@ free_lookup_table_entry(struct wim_lookup_table_entry *lte) if (lte) { switch (lte->resource_location) { case RESOURCE_IN_WIM: - list_del(<e->wim_resource_list); + list_del(<e->rspec_node); if (list_empty(<e->rspec->stream_list)) FREE(lte->rspec); break; @@ -298,7 +298,7 @@ for_lookup_table_entry(struct wim_lookup_table *table, } /* qsort() callback that sorts streams (represented by `struct - * wim_lookup_table_entry's) into an order optimized for reading and writing. + * wim_lookup_table_entry's) into an order optimized for reading. * * Sorting is done primarily by resource location, then secondarily by a * per-resource location order. For example, resources in WIM files are sorted @@ -310,7 +310,6 @@ cmp_streams_by_sequential_order(const void *p1, const void *p2) const struct wim_lookup_table_entry *lte1, *lte2; int v; WIMStruct *wim1, *wim2; - u64 offset1, offset2; lte1 = *(const struct wim_lookup_table_entry**)p1; lte2 = *(const struct wim_lookup_table_entry**)p2; @@ -338,15 +337,9 @@ cmp_streams_by_sequential_order(const void *p1, const void *p2) if (v) return v; - /* Compare by offset. */ - offset1 = lte1->rspec->offset_in_wim + lte1->offset_in_res; - offset2 = lte2->rspec->offset_in_wim + lte2->offset_in_res; + return cmp_u64(lte1->rspec->offset_in_wim + lte1->offset_in_res, + lte2->rspec->offset_in_wim + lte2->offset_in_res); - if (offset1 < offset2) - return -1; - if (offset1 > offset2) - return 1; - return 0; case RESOURCE_IN_FILE_ON_DISK: #ifdef WITH_FUSE case RESOURCE_IN_STAGING_FILE: @@ -370,8 +363,9 @@ cmp_streams_by_sequential_order(const void *p1, const void *p2) } int -sort_stream_list_by_sequential_order(struct list_head *stream_list, - size_t list_head_offset) +sort_stream_list(struct list_head *stream_list, + size_t list_head_offset, + int (*compar)(const void *, const void*)) { struct list_head *cur; struct wim_lookup_table_entry **array; @@ -382,10 +376,14 @@ sort_stream_list_by_sequential_order(struct list_head *stream_list, list_for_each(cur, stream_list) num_streams++; + if (num_streams <= 1) + return 0; + array_size = num_streams * sizeof(array[0]); array = MALLOC(array_size); if (array == NULL) return WIMLIB_ERR_NOMEM; + cur = stream_list->next; for (i = 0; i < num_streams; i++) { array[i] = (struct wim_lookup_table_entry*)((u8*)cur - @@ -393,8 +391,7 @@ sort_stream_list_by_sequential_order(struct list_head *stream_list, cur = cur->next; } - qsort(array, num_streams, sizeof(array[0]), - cmp_streams_by_sequential_order); + qsort(array, num_streams, sizeof(array[0]), compar); INIT_LIST_HEAD(stream_list); for (i = 0; i < num_streams; i++) { @@ -406,6 +403,15 @@ sort_stream_list_by_sequential_order(struct list_head *stream_list, return 0; } +/* Sort the specified list of streams in an order optimized for reading. */ +int +sort_stream_list_by_sequential_order(struct list_head *stream_list, + size_t list_head_offset) +{ + return sort_stream_list(stream_list, list_head_offset, + cmp_streams_by_sequential_order); +} + static int add_lte_to_array(struct wim_lookup_table_entry *lte, @@ -482,13 +488,13 @@ validate_resource(const struct wim_resource_spec *rspec) /* Verify that each stream in the resource has a valid offset and size, * and that no streams overlap. */ cur_offset = 0; - list_for_each_entry(lte, &rspec->stream_list, wim_resource_list) { + list_for_each_entry(lte, &rspec->stream_list, rspec_node) { if (lte->offset_in_res + lte->size < lte->size || lte->offset_in_res + lte->size > rspec->uncompressed_size || lte->offset_in_res < cur_offset) goto invalid; - cur_offset = lte->offset_in_res; + cur_offset = lte->offset_in_res + lte->size; } return 0; @@ -523,6 +529,7 @@ read_wim_lookup_table(WIMStruct *wim) struct wim_lookup_table_entry *cur_entry, *duplicate_entry; struct wim_resource_spec *cur_rspec; void *buf; + bool back_to_back_pack; DEBUG("Reading lookup table."); @@ -586,11 +593,35 @@ read_wim_lookup_table(WIMStruct *wim) continue; } + if (!(reshdr.flags & (WIM_RESHDR_FLAG_PACKED_STREAMS | + WIM_RESHDR_FLAG_COMPRESSED))) { + if (reshdr.uncompressed_size != reshdr.size_in_wim) { + ERROR("Invalid resource entry!"); + ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY; + goto out_free_cur_entry; + } + } + + back_to_back_pack = false; if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) || - cur_rspec == NULL) + cur_rspec == NULL || + (back_to_back_pack = + ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER && + cur_rspec != NULL && + cur_rspec->size_in_wim != 0))) { /* Starting new run of streams that share the same WIM * resource. */ + struct wim_lookup_table_entry *prev_entry = NULL; + + if (back_to_back_pack) { + prev_entry = list_entry(cur_rspec->stream_list.prev, + struct wim_lookup_table_entry, + rspec_node); + lte_unbind_wim_resource_spec(prev_entry); + cur_rspec->uncompressed_size -= prev_entry->size; + } if (cur_rspec != NULL) { ret = validate_resource(cur_rspec); if (ret) @@ -614,32 +645,43 @@ read_wim_lookup_table(WIMStruct *wim) if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { cur_rspec->size_in_wim = 0; cur_rspec->uncompressed_size = 0; - cur_rspec->flags = 0; + cur_rspec->flags = WIM_RESHDR_FLAG_PACKED_STREAMS; } + + if (prev_entry) { + lte_bind_wim_resource_spec(prev_entry, cur_rspec); + cur_rspec->uncompressed_size = prev_entry->size; + } + } + + if ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) + { + /* Found the specification for the packed resource. + * Transfer the values to the `struct + * wim_resource_spec', and discard the current stream + * since this lookup table entry did not, in fact, + * correspond to a "stream". */ + + cur_rspec->offset_in_wim = reshdr.offset_in_wim; + cur_rspec->size_in_wim = reshdr.size_in_wim; + cur_rspec->flags = reshdr.flags; + DEBUG("Full pack is %"PRIu64" compressed bytes " + "at file offset %"PRIu64" (flags 0x%02x)", + cur_rspec->size_in_wim, + cur_rspec->offset_in_wim, + cur_rspec->flags); + free_lookup_table_entry(cur_entry); + continue; } if (is_zero_hash(cur_entry->hash)) { - if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { - /* Found the specification for the packed resource. - * Transfer the values to the `struct - * wim_resource_spec', and discard the current stream - * since this lookup table entry did not, in fact, - * correspond to a "stream". */ - cur_rspec->offset_in_wim = reshdr.offset_in_wim; - cur_rspec->size_in_wim = reshdr.size_in_wim; - cur_rspec->flags = reshdr.flags; - DEBUG("Full run is %"PRIu64" compressed bytes " - "at file offset %"PRIu64" (flags 0x%02x)", - cur_rspec->size_in_wim, - cur_rspec->offset_in_wim, - cur_rspec->flags); - } free_lookup_table_entry(cur_entry); continue; } if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { - /* Continuing the run with another stream. */ + /* Continuing the pack with another stream. */ DEBUG("Continuing packed run with stream: " "%"PRIu64" uncompressed bytes @ resource offset %"PRIu64")", reshdr.size_in_wim, reshdr.offset_in_wim); @@ -666,14 +708,16 @@ read_wim_lookup_table(WIMStruct *wim) if (cur_entry->flags & WIM_RESHDR_FLAG_METADATA) { /* Lookup table entry for a metadata resource */ + + /* Metadata entries with no references must be ignored; + * see for example the WinPE WIMs from the WAIK v2.1. + * */ + if (cur_entry->refcnt == 0) { + free_lookup_table_entry(cur_entry); + continue; + } + if (cur_entry->refcnt != 1) { - /* Metadata entries with no references must be - * ignored. See for example the WinPE WIMs from - * WAIK v2.1. */ - if (cur_entry->refcnt == 0) { - free_lookup_table_entry(cur_entry); - continue; - } if (wimlib_print_errors) { ERROR("Found metadata resource with refcnt != 1:"); print_lookup_table_entry(cur_entry, stderr); @@ -738,7 +782,7 @@ read_wim_lookup_table(WIMStruct *wim) if (cur_rspec != NULL) { ret = validate_resource(cur_rspec); if (ret) - goto out_free_cur_entry; + goto out_free_lookup_table; } if (wim->hdr.part_number == 1 && wim->current_image != wim->hdr.image_count) { @@ -767,19 +811,18 @@ out: return ret; } - static void -write_wim_lookup_table_entry(const struct wim_lookup_table_entry *lte, - struct wim_lookup_table_entry_disk *disk_entry, - u16 part_number) +put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry, + const struct wim_reshdr *out_reshdr, + u16 part_number, u32 refcnt, const u8 *hash) { - put_wim_reshdr(<e->out_reshdr, &disk_entry->reshdr); + put_wim_reshdr(out_reshdr, &disk_entry->reshdr); disk_entry->part_number = cpu_to_le16(part_number); - disk_entry->refcnt = cpu_to_le32(lte->out_refcnt); - copy_hash(disk_entry->hash, lte->hash); + disk_entry->refcnt = cpu_to_le32(refcnt); + copy_hash(disk_entry->hash, hash); } -static int +int write_wim_lookup_table_from_stream_list(struct list_head *stream_list, struct filedes *out_fd, u16 part_number, @@ -792,23 +835,68 @@ write_wim_lookup_table_from_stream_list(struct list_head *stream_list, struct wim_lookup_table_entry_disk *table_buf; struct wim_lookup_table_entry_disk *table_buf_ptr; int ret; + u64 prev_res_offset_in_wim = ~0ULL; table_size = 0; - list_for_each_entry(lte, stream_list, lookup_table_list) + list_for_each_entry(lte, stream_list, lookup_table_list) { table_size += sizeof(struct wim_lookup_table_entry_disk); + if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS && + lte->out_res_offset_in_wim != prev_res_offset_in_wim) + { + table_size += sizeof(struct wim_lookup_table_entry_disk); + prev_res_offset_in_wim = lte->out_res_offset_in_wim; + } + } + DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")", table_size, out_fd->offset); table_buf = MALLOC(table_size); - if (!table_buf) { + if (table_buf == NULL) { ERROR("Failed to allocate %zu bytes for temporary lookup table", table_size); return WIMLIB_ERR_NOMEM; } table_buf_ptr = table_buf; - list_for_each_entry(lte, stream_list, lookup_table_list) - write_wim_lookup_table_entry(lte, table_buf_ptr++, part_number); + + prev_res_offset_in_wim = ~0ULL; + list_for_each_entry(lte, stream_list, lookup_table_list) { + + put_wim_lookup_table_entry(table_buf_ptr++, + <e->out_reshdr, + part_number, + lte->out_refcnt, + lte->hash); + if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS && + lte->out_res_offset_in_wim != prev_res_offset_in_wim) + { + /* Put the main resource entry for the pack. */ + + struct wim_reshdr reshdr; + + reshdr.offset_in_wim = lte->out_res_offset_in_wim; + reshdr.size_in_wim = lte->out_res_size_in_wim; + reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER; + reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS; + + DEBUG("Putting main entry for pack: " + "size_in_wim=%"PRIu64", " + "offset_in_wim=%"PRIu64", " + "uncompressed_size=%"PRIu64, + reshdr.size_in_wim, + reshdr.offset_in_wim, + reshdr.uncompressed_size); + + put_wim_lookup_table_entry(table_buf_ptr++, + &reshdr, + part_number, + 1, zero_hash); + prev_res_offset_in_wim = lte->out_res_offset_in_wim; + } + + } + wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size); /* Write the lookup table uncompressed. Although wimlib can handle a * compressed lookup table, MS software cannot. */ @@ -827,94 +915,6 @@ write_wim_lookup_table_from_stream_list(struct list_head *stream_list, return ret; } -static int -append_lookup_table_entry(struct wim_lookup_table_entry *lte, void *_list) -{ - /* Lookup table entries with 'out_refcnt' == 0 correspond to streams not - * written and not present in the resulting WIM file, and should not be - * included in the lookup table. - * - * Lookup table entries marked as filtered (EXTERNAL_WIM) with - * 'out_refcnt != 0' were referenced as part of the logical write but - * correspond to streams that were not in fact written, and should not - * be included in the lookup table. - * - * Lookup table entries marked as filtered (SAME_WIM) with 'out_refcnt - * != 0' were referenced as part of the logical write but correspond to - * streams that were not in fact written, but nevertheless were already - * present in the WIM being overwritten in-place. These entries must be - * included in the lookup table, and the resource information to write - * needs to be copied from the resource information read originally. - */ - if (lte->out_refcnt != 0 && !(lte->filtered & FILTERED_EXTERNAL_WIM)) { - if (lte->filtered & FILTERED_SAME_WIM) - wim_res_spec_to_hdr(lte->rspec, <e->out_reshdr); - list_add_tail(<e->lookup_table_list, (struct list_head*)_list); - } - return 0; -} - -int -write_wim_lookup_table(WIMStruct *wim, int image, int write_flags, - struct wim_reshdr *out_reshdr, - struct list_head *stream_list_override) -{ - int write_resource_flags; - struct list_head _stream_list; - struct list_head *stream_list; - - if (stream_list_override) { - stream_list = stream_list_override; - } else { - stream_list = &_stream_list; - INIT_LIST_HEAD(stream_list); - } - - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) { - int start_image; - int end_image; - - if (image == WIMLIB_ALL_IMAGES) { - start_image = 1; - end_image = wim->hdr.image_count; - } else { - start_image = image; - end_image = image; - } - - /* Push metadata resource lookup table entries onto the front of - * the list in reverse order, so that they're written in order. - */ - for (int i = end_image; i >= start_image; i--) { - struct wim_lookup_table_entry *metadata_lte; - - metadata_lte = wim->image_metadata[i - 1]->metadata_lte; - metadata_lte->out_refcnt = 1; - metadata_lte->out_reshdr.flags |= WIM_RESHDR_FLAG_METADATA; - list_add(&metadata_lte->lookup_table_list, stream_list); - } - } - - /* Append additional lookup table entries that need to be written, with - * some special handling for streams that have been marked as filtered. - */ - if (!stream_list_override) { - for_lookup_table_entry(wim->lookup_table, - append_lookup_table_entry, stream_list); - } - - write_resource_flags = 0; - if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) - write_resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; - return write_wim_lookup_table_from_stream_list(stream_list, - &wim->out_fd, - wim->hdr.part_number, - out_reshdr, - write_resource_flags, - &wim->lzx_context); -} - - int lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore) { @@ -950,7 +950,7 @@ print_lookup_table_entry(const struct wim_lookup_table_entry *lte, FILE *out) tprintf(T("Uncompressed size = %"PRIu64" bytes\n"), lte->size); - if (lte_is_partial(lte)) { + if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { tprintf(T("Offset = %"PRIu64" bytes\n"), lte->offset_in_res); @@ -1029,7 +1029,7 @@ lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte, wentry->uncompressed_size = lte->size; if (lte->resource_location == RESOURCE_IN_WIM) { wentry->part_number = lte->rspec->wim->hdr.part_number; - if (lte_is_partial(lte)) { + if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { wentry->compressed_size = 0; wentry->offset = lte->offset_in_res; } else { @@ -1046,7 +1046,7 @@ lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte, wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0; wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0; wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0; - wentry->is_partial = lte_is_partial(lte); + wentry->is_packed_streams = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0; } struct iterate_lte_context { @@ -1438,24 +1438,23 @@ hash_unhashed_stream(struct wim_lookup_table_entry *lte, list_del(<e->unhashed_list); if (duplicate_lte) { /* We have a duplicate stream. Transfer the reference counts - * from this stream to the duplicate, update the reference to + * from this stream to the duplicate and update the reference to * this stream (in an inode or ads_entry) to point to the - * duplicate, then free this stream. */ + * duplicate. The caller is responsible for freeing @lte if + * needed. */ wimlib_assert(!(duplicate_lte->unhashed)); + wimlib_assert(duplicate_lte->size == lte->size); duplicate_lte->refcnt += lte->refcnt; - duplicate_lte->out_refcnt += lte->out_refcnt; + lte->refcnt = 0; *back_ptr = duplicate_lte; - free_lookup_table_entry(lte); lte = duplicate_lte; } else { - /* No duplicate stream, so we need to insert - * this stream into the lookup table and treat - * it as a hashed stream. */ + /* No duplicate stream, so we need to insert this stream into + * the lookup table and treat it as a hashed stream. */ lookup_table_insert(lookup_table, lte); lte->unhashed = 0; } - if (lte_ret) - *lte_ret = lte; + *lte_ret = lte; return 0; } @@ -1468,7 +1467,7 @@ lte_clone_if_new(struct wim_lookup_table_entry *lte, void *_lookup_table) return 0; /* Resource already present. */ lte = clone_lookup_table_entry(lte); - if (!lte) + if (lte == NULL) return WIMLIB_ERR_NOMEM; lte->out_refcnt = 1; lookup_table_insert(lookup_table, lte); diff --git a/src/ntfs-3g_capture.c b/src/ntfs-3g_capture.c index 7a07cd3e..17fa9f06 100644 --- a/src/ntfs-3g_capture.c +++ b/src/ntfs-3g_capture.c @@ -75,11 +75,8 @@ open_ntfs_attr(ntfs_inode *ni, struct ntfs_location *loc) } int -read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *cb_ctx, - int _ignored_flags) +read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, + consume_data_callback_t cb, void *cb_ctx) { struct ntfs_location *loc = lte->ntfs_loc; ntfs_volume *vol = loc->ntfs_vol; diff --git a/src/resource.c b/src/resource.c index b8b8b921..373279c1 100644 --- a/src/resource.c +++ b/src/resource.c @@ -118,27 +118,6 @@ struct data_range { u64 size; }; -/* Alternate chunk table format for resources with - * WIM_RESHDR_FLAG_PACKED_STREAMS set. */ -struct alt_chunk_table_header_disk { - /* Uncompressed size of the resource in bytes. */ - le64 res_usize; - - /* Number of bytes each compressed chunk decompresses into, except - * possibly the last which decompresses into the remainder. */ - le32 chunk_size; - - /* Compression format used for compressed chunks: - * 0 = None - * 1 = LZX - * 2 = XPRESS - * 3 = LZMS */ - le32 compression_format; - - /* This header is directly followed by a table of compressed sizes of - * the chunks. */ -} _packed_attribute; - /* * read_compressed_wim_resource() - * @@ -153,16 +132,11 @@ struct alt_chunk_table_header_disk { * Number of ranges in @ranges; must be at least 1. * @cb * Callback function to feed the data being read. Each call provides the - * next chunk of the requested data. Each chunk will be of nonzero size - * and will not cross range boundaries, but otherwise is of unspecified - * size. + * next chunk of the requested data, uncompressed. Each chunk will be of + * nonzero size and will not cross range boundaries, but otherwise will be + * of unspecified size. * @cb_ctx * Parameter to pass to @cb_ctx. - * @raw_chunks_mode - * If %true, this function will provide the raw compressed chunks of the - * resource rather than the uncompressed data. In this mode, only a single - * data range can be requested, and it must cover the entire uncompressed - * resource. * * Possible return values: * @@ -179,8 +153,7 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, const struct data_range * const ranges, const size_t num_ranges, const consume_data_callback_t cb, - void * const cb_ctx, - const bool raw_chunks_mode) + void * const cb_ctx) { int ret; int errno_save; @@ -208,12 +181,6 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, for (size_t i = 0; i < num_ranges - 1; i++) wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset); - if (raw_chunks_mode) { - wimlib_assert(num_ranges == 1); - wimlib_assert(ranges[0].offset == 0); - wimlib_assert(ranges[0].size == rspec->uncompressed_size); - } - /* Get the offsets of the first and last bytes of the read. */ const u64 first_offset = ranges[0].offset; const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1; @@ -428,19 +395,17 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, ubuf_malloced = true; } - /* Unless the raw compressed data was requested, allocate a temporary - * buffer for reading compressed chunks, each of which can be at most - * @chunk_size - 1 bytes. This excludes compressed chunks that are a - * full @chunk_size bytes, which are actually stored uncompressed. */ - if (!raw_chunks_mode) { - if (chunk_size - 1 <= STACK_MAX) { - cbuf = alloca(chunk_size - 1); - } else { - cbuf = MALLOC(chunk_size - 1); - if (cbuf == NULL) - goto oom; - cbuf_malloced = true; - } + /* Allocate a temporary buffer for reading compressed chunks, each of + * which can be at most @chunk_size - 1 bytes. This excludes compressed + * chunks that are a full @chunk_size bytes, which are actually stored + * uncompressed. */ + if (chunk_size - 1 <= STACK_MAX) { + cbuf = alloca(chunk_size - 1); + } else { + cbuf = MALLOC(chunk_size - 1); + if (cbuf == NULL) + goto oom; + cbuf_malloced = true; } /* Set current data range. */ @@ -513,7 +478,7 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, * function. */ u8 *read_buf; - if (chunk_csize == chunk_usize || raw_chunks_mode) + if (chunk_csize == chunk_usize) read_buf = ubuf; else read_buf = cbuf; @@ -527,7 +492,7 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, if (read_buf == cbuf) { DEBUG("Decompressing chunk %"PRIu64" " - "(csize=%"PRIu64" usize=%"PRIu64"", + "(csize=%"PRIu32" usize=%"PRIu32")", i, chunk_csize, chunk_usize); ret = decompress(cbuf, chunk_csize, @@ -557,10 +522,7 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, end = min(cur_range_end, chunk_end_offset) - chunk_start_offset; size = end - start; - if (raw_chunks_mode) - ret = (*cb)(&ubuf[0], chunk_csize, cb_ctx); - else - ret = (*cb)(&ubuf[start], size, cb_ctx); + ret = (*cb)(&ubuf[start], size, cb_ctx); if (ret) goto out_free_memory; @@ -658,24 +620,21 @@ bufferer_cb(const void *chunk, size_t size, void *_ctx) * read_partial_wim_resource()- * * Read a range of data from an uncompressed or compressed resource in a WIM - * file. Data is fed chunk-by-chunk into the callback function @cb, passing it - * the argument @cb_ctx. The chunks are of unspecified size unless the - * RAW_CHUNKS mode is requested. - * - * By default, this function provides the uncompressed data of the resource, and - * @offset and @size and interpreted relative to the uncompressed contents of - * the resource. This behavior can be modified by either of the following - * flags: - * - * WIMLIB_READ_RESOURCE_FLAG_RAW_FULL: - * Read @size bytes at @offset of the raw contents of the compressed - * resource. In the case of pipable resources, this excludes the stream - * header. Exclusive with WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS. + * file. * - * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS: - * Read the raw compressed chunks of the compressed resource. @size must - * be the full uncompressed size, @offset must be 0, and @cb_chunk_size - * must be the resource chunk size. + * @rspec + * Specification of the WIM resource to read from. + * @offset + * Offset within the uncompressed resource at which to start reading. + * @size + * Number of bytes to read. + * @cb + * Callback function to feed the data being read. Each call provides the + * next chunk of the requested data, uncompressed. Each chunk will be of + * nonzero size and will not cross range boundaries, but otherwise will be + * of unspecified size. + * @cb_ctx + * Parameter to pass to @cb_ctx. * * Return values: * WIMLIB_ERR_SUCCESS (0) @@ -688,47 +647,35 @@ bufferer_cb(const void *chunk, size_t size, void *_ctx) */ static int read_partial_wim_resource(const struct wim_resource_spec *rspec, - u64 offset, u64 size, consume_data_callback_t cb, - void *cb_ctx, int flags) + u64 offset, u64 size, + consume_data_callback_t cb, void *cb_ctx) { /* Sanity checks. */ - if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) { - wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)); - wimlib_assert(offset + size >= offset); - wimlib_assert(offset + size <= rspec->size_in_wim); - } else if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { - wimlib_assert(offset == 0); - wimlib_assert(size == rspec->uncompressed_size); - } else { - wimlib_assert(offset + size >= offset); - wimlib_assert(offset + size <= rspec->uncompressed_size); - } + wimlib_assert(offset + size >= offset); + wimlib_assert(offset + size <= rspec->uncompressed_size); DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource " - "%"PRIu64" => %"PRIu64" @ %"PRIu64" (flags 0x%08x)", + "%"PRIu64" => %"PRIu64" @ %"PRIu64, size, offset, rspec->uncompressed_size, - rspec->size_in_wim, rspec->offset_in_wim, flags); + rspec->size_in_wim, rspec->offset_in_wim); /* Trivial case. */ if (size == 0) return 0; - if ((flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) || - !resource_is_compressed(rspec)) - { - return read_raw_file_data(&rspec->wim->in_fd, - size, - cb, - cb_ctx, - rspec->offset_in_wim + offset); - } else { - bool raw_chunks = (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS); + if (resource_is_compressed(rspec)) { struct data_range range = { .offset = offset, .size = size, }; return read_compressed_wim_resource(rspec, &range, 1, - cb, cb_ctx, raw_chunks); + cb, cb_ctx); + } else { + return read_raw_file_data(&rspec->wim->in_fd, + size, + cb, + cb_ctx, + rspec->offset_in_wim + offset); } } @@ -746,8 +693,7 @@ read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte, lte->offset_in_res + offset, size, bufferer_cb, - &buf, - 0); + &buf); } /* A consume_data_callback_t implementation that simply ignores the data @@ -764,25 +710,21 @@ int skip_wim_stream(struct wim_lookup_table_entry *lte) { wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); - wimlib_assert(!lte_is_partial(lte)); + wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); + DEBUG("Skipping stream (size=%"PRIu64")", lte->size); return read_partial_wim_resource(lte->rspec, 0, lte->rspec->uncompressed_size, skip_chunk_cb, - NULL, - WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS); + NULL); } static int read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, void *cb_ctx, int flags) + consume_data_callback_t cb, void *cb_ctx) { - return read_partial_wim_resource(lte->rspec, - lte->offset_in_res, - size, - cb, - cb_ctx, - flags); + return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size, + cb, cb_ctx); } #ifndef __WIN32__ @@ -796,8 +738,7 @@ read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, * encrypted), so Windows uses its own code for its equivalent case. */ static int read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, void *cb_ctx, - int _ignored_flags) + consume_data_callback_t cb, void *cb_ctx) { int ret; int raw_fd; @@ -823,8 +764,7 @@ read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, * fact, already located in an in-memory buffer. */ static int read_buffer_prefix(const struct wim_lookup_table_entry *lte, - u64 size, consume_data_callback_t cb, - void *cb_ctx, int _ignored_flags) + u64 size, consume_data_callback_t cb, void *cb_ctx) { wimlib_assert(size <= lte->size); return (*cb)(lte->attached_buffer, size, cb_ctx); @@ -833,7 +773,7 @@ read_buffer_prefix(const struct wim_lookup_table_entry *lte, typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte, u64 size, consume_data_callback_t cb, - void *cb_ctx, int flags); + void *cb_ctx); /* * read_stream_prefix()- @@ -845,9 +785,6 @@ typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry * This function feeds the data to a callback function @cb in chunks of * unspecified size. * - * If the stream is located in a WIM file, @flags can be set as documented in - * read_partial_wim_resource(). Otherwise @flags are ignored. - * * Returns 0 on success; nonzero on error. A nonzero value will be returned if * the stream data cannot be successfully read (for a number of different * reasons, depending on the stream location), or if @cb returned nonzero in @@ -855,7 +792,7 @@ typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry */ int read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, void *cb_ctx, int flags) + consume_data_callback_t cb, void *cb_ctx) { static const read_stream_prefix_handler_t handlers[] = { [RESOURCE_IN_WIM] = read_wim_stream_prefix, @@ -877,7 +814,7 @@ read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, }; wimlib_assert(lte->resource_location < ARRAY_LEN(handlers) && handlers[lte->resource_location] != NULL); - return handlers[lte->resource_location](lte, size, cb, cb_ctx, flags); + return handlers[lte->resource_location](lte, size, cb, cb_ctx); } /* Read the full uncompressed data of the specified stream into the specified @@ -886,7 +823,7 @@ int read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *_buf) { u8 *buf = _buf; - return read_stream_prefix(lte, lte->size, bufferer_cb, &buf, 0); + return read_stream_prefix(lte, lte->size, bufferer_cb, &buf); } /* Retrieve the full uncompressed data of the specified stream. A buffer large @@ -930,7 +867,6 @@ wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret) if (lte == NULL) return WIMLIB_ERR_NOMEM; - lte->unhashed = 1; lte_bind_wim_resource_spec(lte, rspec); lte->flags = rspec->flags; lte->size = rspec->uncompressed_size; @@ -1000,6 +936,13 @@ streamifier_cb(const void *chunk, size_t size, void *_ctx) if (ctx->cur_stream_offset == ctx->cur_stream->size) { /* Finished reading all the data for a stream. */ + + struct list_head *cur, *next; + + cur = (struct list_head *) + ((u8*)ctx->cur_stream + ctx->list_head_offset); + next = cur->next; + DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size); ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0, ctx->cbs.end_stream_ctx); @@ -1008,12 +951,6 @@ streamifier_cb(const void *chunk, size_t size, void *_ctx) if (ctx->cur_stream != ctx->final_stream) { /* Advance to next stream. */ - struct list_head *cur, *next; - - cur = (struct list_head *) - ((u8*)ctx->cur_stream + ctx->list_head_offset); - next = cur->next; - ctx->cur_stream = (struct wim_lookup_table_entry *) ((u8*)next - ctx->list_head_offset); @@ -1028,6 +965,7 @@ streamifier_cb(const void *chunk, size_t size, void *_ctx) struct hasher_context { SHA_CTX sha_ctx; + int flags; struct read_stream_list_callbacks cbs; }; @@ -1083,26 +1021,30 @@ hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx) sha1_final(hash, &ctx->sha_ctx); if (lte->unhashed) { - /* No SHA1 message digest was previously present for the stream. - * Set it to the one just calculated. */ - DEBUG("Set SHA1 message digest for stream (size=%"PRIu64").", lte->size); - copy_hash(lte->hash, hash); + if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) { + /* No SHA1 message digest was previously present for the + * stream. Set it to the one just calculated. */ + DEBUG("Set SHA1 message digest for stream " + "(size=%"PRIu64").", lte->size); + copy_hash(lte->hash, hash); + } } else { - /* The stream already had a SHA1 message digest present. Verify - * that it is the same as the calculated value. */ - if (!hashes_equal(hash, lte->hash)) { - if (wimlib_print_errors) { - ERROR("Invalid SHA1 message digest " - "on the following WIM stream:"); - print_lookup_table_entry(lte, stderr); - if (lte->resource_location == RESOURCE_IN_WIM) - ERROR("The WIM file appears to be corrupt!"); + if (ctx->flags & VERIFY_STREAM_HASHES) { + /* The stream already had a SHA1 message digest present. Verify + * that it is the same as the calculated value. */ + if (!hashes_equal(hash, lte->hash)) { + if (wimlib_print_errors) { + ERROR("Invalid SHA1 message digest " + "on the following WIM stream:"); + print_lookup_table_entry(lte, stderr); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + errno = EINVAL; + goto out_next_cb; } - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - errno = EINVAL; - goto out_next_cb; + DEBUG("SHA1 message digest okay for " + "stream (size=%"PRIu64").", lte->size); } - DEBUG("SHA1 message digest okay for stream (size=%"PRIu64").", lte->size); } ret = 0; out_next_cb: @@ -1112,89 +1054,43 @@ out_next_cb: return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx); } -/* Read the full data of the specified stream, passing the data into the - * specified callbacks (all of which are optional) and either checking or - * computing the SHA1 message digest of the stream. */ static int -read_full_stream_with_sha1(struct wim_lookup_table_entry *lte, - const struct read_stream_list_callbacks *cbs) +read_full_stream_with_cbs(struct wim_lookup_table_entry *lte, + const struct read_stream_list_callbacks *cbs) { int ret; - struct hasher_context hasher_ctx = { - .cbs = *cbs, - }; - - ret = hasher_begin_stream(lte, false, &hasher_ctx); + ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx); if (ret) return ret; - ret = read_stream_prefix(lte, lte->size, hasher_consume_chunk, - &hasher_ctx, 0); + ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk, + cbs->consume_chunk_ctx); - return hasher_end_stream(lte, ret, &hasher_ctx); + return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx); } -struct rechunkifier_context { - u8 *buffer; - u32 buffer_filled; - u32 cb_chunk_size; - - const struct data_range *ranges; - size_t num_ranges; - size_t cur_range; - u64 range_bytes_remaining; - - consume_data_callback_t cb; - void *cb_ctx; -}; - -/* Wrapper callback for adjusting the data chunk size. */ +/* Read the full data of the specified stream, passing the data into the + * specified callbacks (all of which are optional) and either checking or + * computing the SHA1 message digest of the stream. */ static int -rechunkifier_cb(const void *chunk, size_t size, void *_ctx) +read_full_stream_with_sha1(struct wim_lookup_table_entry *lte, + const struct read_stream_list_callbacks *cbs) { - struct rechunkifier_context *ctx = _ctx; - const u8 *chunkptr = chunk; - size_t bytes_to_copy; - int ret; - - wimlib_assert(ctx->cur_range != ctx->num_ranges); - - while (size) { - - /* Append more data to the buffer. */ - bytes_to_copy = size; - - if (bytes_to_copy > ctx->cb_chunk_size - ctx->buffer_filled) - bytes_to_copy = ctx->cb_chunk_size - ctx->buffer_filled; - - if (bytes_to_copy > ctx->range_bytes_remaining - ctx->buffer_filled) - bytes_to_copy = ctx->range_bytes_remaining - ctx->buffer_filled; - - memcpy(&ctx->buffer[ctx->buffer_filled], chunkptr, bytes_to_copy); - - ctx->buffer_filled += bytes_to_copy; - chunkptr += bytes_to_copy; - size -= bytes_to_copy; - ctx->range_bytes_remaining -= bytes_to_copy; - - if (ctx->buffer_filled == ctx->cb_chunk_size || - ctx->range_bytes_remaining == 0) - { - /* Maximum chunk size reached, or current range ended. - * Call the next consume_data_callback_t and empty the - * buffer */ - ret = (*ctx->cb)(ctx->buffer, ctx->buffer_filled, ctx->cb_ctx); - if (ret) - return ret; - ctx->buffer_filled = 0; + struct hasher_context hasher_ctx = { + .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES, + .cbs = *cbs, + }; + struct read_stream_list_callbacks hasher_cbs = { + .begin_stream = hasher_begin_stream, + .begin_stream_ctx = &hasher_ctx, + .consume_chunk = hasher_consume_chunk, + .consume_chunk_ctx = &hasher_ctx, + .end_stream = hasher_end_stream, + .end_stream_ctx = &hasher_ctx, - if (ctx->range_bytes_remaining == 0 && - ++ctx->cur_range != ctx->num_ranges) - ctx->range_bytes_remaining = ctx->ranges[ctx->cur_range].size; - } - } - return 0; + }; + return read_full_stream_with_cbs(lte, &hasher_cbs); } /* @@ -1210,11 +1106,10 @@ rechunkifier_cb(const void *chunk, size_t size, void *_ctx) * @list_head_offset * Offset of the `struct list_head' within each `struct * wim_lookup_table_entry' that makes up the @stream_list. - * @cb_chunk_size - * Size of chunks to provide to @consume_chunk. For a given stream, all - * the chunks will be this size, except possibly the last which will be the - * remainder. If @ - * @cbs TODO + * @cbs + * Callback functions to accept the stream data. + * @flags + * * * Returns 0 on success; a nonzero error code on failure. Failure can occur due * to an error reading the data or due to an error status being returned by any @@ -1223,16 +1118,39 @@ rechunkifier_cb(const void *chunk, size_t size, void *_ctx) int read_stream_list(struct list_head *stream_list, size_t list_head_offset, - u32 cb_chunk_size, - const struct read_stream_list_callbacks *cbs) + const struct read_stream_list_callbacks *cbs, + int flags) { int ret; struct list_head *cur, *next; struct wim_lookup_table_entry *lte; + struct hasher_context *hasher_ctx; + struct read_stream_list_callbacks *sink_cbs; - ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset); - if (ret) - return ret; + if (!(flags & STREAM_LIST_ALREADY_SORTED)) { + ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset); + if (ret) + return ret; + } + + if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) { + hasher_ctx = alloca(sizeof(*hasher_ctx)); + *hasher_ctx = (struct hasher_context) { + .flags = flags, + .cbs = *cbs, + }; + sink_cbs = alloca(sizeof(*sink_cbs)); + *sink_cbs = (struct read_stream_list_callbacks) { + .begin_stream = hasher_begin_stream, + .begin_stream_ctx = hasher_ctx, + .consume_chunk = hasher_consume_chunk, + .consume_chunk_ctx = hasher_ctx, + .end_stream = hasher_end_stream, + .end_stream_ctx = hasher_ctx, + }; + } else { + sink_cbs = (struct read_stream_list_callbacks*)cbs; + } for (cur = stream_list->next, next = cur->next; cur != stream_list; @@ -1240,7 +1158,9 @@ read_stream_list(struct list_head *stream_list, { lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset); - if (lte_is_partial(lte)) { + if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS && + lte->size != lte->rspec->uncompressed_size) + { struct wim_lookup_table_entry *lte_next, *lte_last; struct list_head *next2; @@ -1296,73 +1216,8 @@ read_stream_list(struct list_head *stream_list, } } - /* Set up a chain of callbacks. - * - * The first level is the - * streamifier_cb, - * which takes in chunks of data and divides - * them into the constituent streams. - * - * The second level are the SHA1 message digest - * callbacks, which checksum each stream. - * - * rechunkifier_cb handles dividing the read - * data into chunks of maximum size - * @cb_chunk_size. If @cb_chunk_size is 0, then - * this callback is not needed. - * - * Finally, the last level of callbacks are - * @cbs, passed as arguments to this function. - */ - - struct rechunkifier_context *rechunkifier_ctx = NULL; - consume_data_callback_t last_cb; - void *last_cb_ctx; - - if (cb_chunk_size != 0) { - rechunkifier_ctx = alloca(sizeof(*rechunkifier_ctx)); - *rechunkifier_ctx = (struct rechunkifier_context) { - .buffer = MALLOC(cb_chunk_size), - .buffer_filled = 0, - .cb_chunk_size = cb_chunk_size, - .ranges = ranges, - .num_ranges = stream_count, - .cur_range = 0, - .range_bytes_remaining = ranges[0].size, - .cb = cbs->consume_chunk, - .cb_ctx = cbs->consume_chunk_ctx, - }; - - if (rechunkifier_ctx->buffer == NULL) - return WIMLIB_ERR_NOMEM; - last_cb = rechunkifier_cb; - last_cb_ctx = rechunkifier_ctx; - } else { - rechunkifier_ctx = NULL; - last_cb = cbs->consume_chunk; - last_cb_ctx = cbs->consume_chunk_ctx; - } - - struct hasher_context hasher_ctx = { - .cbs = { - .begin_stream = cbs->begin_stream, - .begin_stream_ctx = cbs->begin_stream_ctx, - .consume_chunk = last_cb, - .consume_chunk_ctx = last_cb_ctx, - .end_stream = cbs->end_stream, - .end_stream_ctx = cbs->end_stream_ctx, - }, - }; - struct streamifier_context streamifier_ctx = { - .cbs = { - .begin_stream = hasher_begin_stream, - .begin_stream_ctx = &hasher_ctx, - .consume_chunk = hasher_consume_chunk, - .consume_chunk_ctx = &hasher_ctx, - .end_stream = hasher_end_stream, - .end_stream_ctx = &hasher_ctx, - }, + .cbs = *sink_cbs, .cur_stream = lte, .cur_stream_offset = 0, .final_stream = lte_last, @@ -1373,10 +1228,7 @@ read_stream_list(struct list_head *stream_list, ranges, stream_count, streamifier_cb, - &streamifier_ctx, - false); - if (rechunkifier_ctx != NULL) - FREE(rechunkifier_ctx->buffer); + &streamifier_ctx); if (ret) { if (streamifier_ctx.cur_stream_offset != 0) { @@ -1391,8 +1243,8 @@ read_stream_list(struct list_head *stream_list, } } - ret = read_full_stream_with_sha1(lte, cbs); - if (ret > 0) + ret = read_full_stream_with_cbs(lte, sink_cbs); + if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM) return ret; } return 0; @@ -1419,7 +1271,7 @@ extract_stream(struct wim_lookup_table_entry *lte, u64 size, } else { /* Don't do SHA1. */ return read_stream_prefix(lte, size, extract_chunk, - extract_chunk_arg, 0); + extract_chunk_arg); } } @@ -1499,7 +1351,6 @@ get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr, ((u64)disk_reshdr->size_in_wim[6] << 48)); reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size); reshdr->flags = disk_reshdr->flags; - return 0; } /* Translates a WIM resource header from an in-memory format into the on-disk diff --git a/src/split.c b/src/split.c index 167105eb..01bb75fc 100644 --- a/src/split.c +++ b/src/split.c @@ -148,8 +148,13 @@ add_stream_to_swm(struct wim_lookup_table_entry *lte, void *_swm_info) struct swm_info *swm_info = _swm_info; u64 stream_size; - /* TODO */ - stream_size = lte->rspec->size_in_wim; + /* We want the compressed size of the stream, but use the uncompressed + * size if the compressed size is not available or not relevant. */ + if (lte->resource_location == RESOURCE_IN_WIM && + (!lte_is_partial(lte) || lte->rspec->size_in_wim < lte->size)) + stream_size = lte->rspec->size_in_wim; + else + stream_size = lte->size; /* - Start first part if no parts have been started so far; * - Start next part if adding this stream exceeds maximum part size, diff --git a/src/wim.c b/src/wim.c index 11bb2e50..3d173f9a 100644 --- a/src/wim.c +++ b/src/wim.c @@ -140,8 +140,7 @@ wim_chunk_size_valid(u32 chunk_size, int ctype) return order >= 15 && order <= 26; case WIMLIB_COMPRESSION_TYPE_LZMS: - /* TODO */ - return 131072; + return order >= 15 && order <= 26; } return false; } @@ -836,9 +835,12 @@ wim_checksum_unhashed_streams(WIMStruct *wim) struct wim_lookup_table_entry *lte, *tmp; struct wim_image_metadata *imd = wim->image_metadata[i]; image_for_each_unhashed_stream_safe(lte, tmp, imd) { - ret = hash_unhashed_stream(lte, wim->lookup_table, NULL); + struct wim_lookup_table_entry *new_lte; + ret = hash_unhashed_stream(lte, wim->lookup_table, &new_lte); if (ret) return ret; + if (new_lte != lte) + free_lookup_table_entry(lte); } } return 0; diff --git a/src/write.c b/src/write.c index e7107ab1..95365744 100644 --- a/src/write.c +++ b/src/write.c @@ -34,6 +34,7 @@ # include #endif +#include "wimlib/compress_chunks.h" #include "wimlib/endianness.h" #include "wimlib/error.h" #include "wimlib/file_io.h" @@ -45,17 +46,8 @@ #include "wimlib/write.h" #include "wimlib/xml.h" -#ifdef __WIN32__ -# include "wimlib/win32.h" /* win32_get_number_of_processors() */ -#endif - -#ifdef ENABLE_MULTITHREADED_COMPRESSION -# include -#endif - #include #include -#include #include #include @@ -63,10 +55,88 @@ # include #endif +static inline int +write_flags_to_resource_flags(int write_flags) +{ + int write_resource_flags = 0; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS; + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + write_resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) + write_resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS; + return write_resource_flags; +} + +struct filter_context { + int write_flags; + WIMStruct *wim; +}; + +/* Determine specified stream should be filtered out from the write. + * + * Return values: + * + * < 0 : The stream should be hard-filtered; that is, not included in the + * output WIM at all. + * 0 : The stream should not be filtered out. + * > 0 : The stream should be soft-filtered; that is, it already exists in the + * WIM file and may not need to be written again. + */ +static int +stream_filtered(const struct wim_lookup_table_entry *lte, + const struct filter_context *ctx) +{ + int write_flags = ctx->write_flags; + WIMStruct *wim = ctx->wim; + + if (ctx == NULL) + return 0; + + if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE && + lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->wim == wim) + return 1; + + if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS && + lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->wim != wim) + return -1; + + return 0; +} + +static bool +stream_hard_filtered(const struct wim_lookup_table_entry *lte, + struct filter_context *ctx) +{ + return stream_filtered(lte, ctx) < 0; +} + +static inline int +may_soft_filter_streams(const struct filter_context *ctx) +{ + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE; +} + +static inline int +may_hard_filter_streams(const struct filter_context *ctx) +{ + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS; +} + +static inline int +may_filter_streams(const struct filter_context *ctx) +{ + return (may_soft_filter_streams(ctx) || + may_hard_filter_streams(ctx)); +} -#ifndef __WIN32__ -# include /* for `struct iovec' */ -#endif /* Return true if the specified resource is compressed and the compressed data * can be reused with the specified output parameters. */ @@ -74,184 +144,95 @@ static bool can_raw_copy(const struct wim_lookup_table_entry *lte, int write_resource_flags, int out_ctype, u32 out_chunk_size) { + const struct wim_resource_spec *rspec; + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) return false; - if (lte->resource_location != RESOURCE_IN_WIM) - return false; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) return false; - if (lte->rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) - return false; - if (!(lte->rspec->flags & WIM_RESHDR_FLAG_COMPRESSED)) - return false; - if (lte->rspec->wim->compression_type != out_ctype) - return false; - if (lte->rspec->wim->chunk_size != out_chunk_size) + + if (lte->resource_location != RESOURCE_IN_WIM) return false; - return true; -} + rspec = lte->rspec; -/* Return true if the specified resource must be recompressed when the specified - * output parameters are used. */ -static bool -must_compress_stream(const struct wim_lookup_table_entry *lte, - int write_resource_flags, int out_ctype, u32 out_chunk_size) -{ - return (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE - && ((write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) - || !can_raw_copy(lte, write_resource_flags, - out_ctype, out_chunk_size))); -} - -static unsigned -compress_chunk(const void * uncompressed_data, - unsigned uncompressed_len, - void *compressed_data, - int out_ctype, - struct wimlib_lzx_context *comp_ctx) -{ - switch (out_ctype) { - case WIMLIB_COMPRESSION_TYPE_XPRESS: - return wimlib_xpress_compress(uncompressed_data, - uncompressed_len, - compressed_data); - case WIMLIB_COMPRESSION_TYPE_LZX: - return wimlib_lzx_compress2(uncompressed_data, - uncompressed_len, - compressed_data, - comp_ctx); - case WIMLIB_COMPRESSION_TYPE_LZMS: - /* TODO */ - WARNING("LZMS compression not yet implemented!"); - return 0; + if (rspec->is_pipable != !!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + return false; - default: - wimlib_assert(0); - return 0; + + if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) { + /* Normal compressed resource: Must use same compression type + * and chunk size. */ + return (rspec->wim->compression_type == out_ctype && + rspec->wim->chunk_size == out_chunk_size); } -} -/* Chunk table that's located at the beginning of each compressed resource in - * the WIM. (This is not the on-disk format; the on-disk format just has an - * array of offsets.) */ -struct chunk_table { - u64 original_resource_size; - u64 num_chunks; - u64 table_disk_size; - unsigned bytes_per_chunk_entry; - void *cur_offset_p; - union { - u32 cur_offset_u32; - u64 cur_offset_u64; - }; - /* Beginning of chunk offsets, in either 32-bit or 64-bit little endian - * integers, including the first offset of 0, which will not be written. - * */ - u8 offsets[] _aligned_attribute(8); -}; + if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS)) + { + /* Packed resource: Such resources may contain multiple streams, + * and in general only a subset of them need to be written. As + * a heuristic, re-use the raw data if at least half the + * uncompressed size is being written. */ -/* Allocate and initializes a chunk table, then reserve space for it in the - * output file unless writing a pipable resource. */ -static int -begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, - struct filedes *out_fd, - u32 out_chunk_size, - struct chunk_table **chunk_tab_ret, - int resource_flags) -{ - u64 size; - u64 num_chunks; - unsigned bytes_per_chunk_entry; - size_t alloc_size; - struct chunk_table *chunk_tab; - int ret; + /* Note: packed resources contain a header that specifies the + * compression type and chunk size; therefore we don't need to + * check if they are compatible with @out_ctype and + * @out_chunk_size. */ - size = lte->size; - num_chunks = DIV_ROUND_UP(size, out_chunk_size); - bytes_per_chunk_entry = (size > (1ULL << 32)) ? 8 : 4; - alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); - chunk_tab = CALLOC(1, alloc_size); + struct wim_lookup_table_entry *res_stream; + u64 write_size = 0; - if (!chunk_tab) { - ERROR("Failed to allocate chunk table for %"PRIu64" byte " - "resource", size); - return WIMLIB_ERR_NOMEM; - } - chunk_tab->num_chunks = num_chunks; - chunk_tab->original_resource_size = size; - chunk_tab->bytes_per_chunk_entry = bytes_per_chunk_entry; - chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * - (num_chunks - 1); - chunk_tab->cur_offset_p = chunk_tab->offsets; - - /* We don't know the correct offsets yet; so just write zeroes to - * reserve space for the table, so we can go back to it later after - * we've written the compressed chunks following it. - * - * Special case: if writing a pipable WIM, compressed resources are in a - * modified format (see comment above write_pipable_wim()) and do not - * have a chunk table at the beginning, so don't reserve any space for - * one. */ - if (!(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { - ret = full_write(out_fd, chunk_tab->offsets, - chunk_tab->table_disk_size); - if (ret) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - FREE(chunk_tab); - return ret; - } + list_for_each_entry(res_stream, &rspec->stream_list, rspec_node) + if (res_stream->will_be_in_output_wim) + write_size += res_stream->size; + + return (write_size > rspec->uncompressed_size / 2); } - *chunk_tab_ret = chunk_tab; - return 0; + return false; } -/* Add the offset for the next chunk to the chunk table being constructed for a - * compressed stream. */ -static void -chunk_tab_record_chunk(struct chunk_table *chunk_tab, unsigned out_chunk_size) +static u8 +filter_resource_flags(u8 flags) { - if (chunk_tab->bytes_per_chunk_entry == 4) { - *(le32*)chunk_tab->cur_offset_p = cpu_to_le32(chunk_tab->cur_offset_u32); - chunk_tab->cur_offset_p = (le32*)chunk_tab->cur_offset_p + 1; - chunk_tab->cur_offset_u32 += out_chunk_size; - } else { - *(le64*)chunk_tab->cur_offset_p = cpu_to_le64(chunk_tab->cur_offset_u64); - chunk_tab->cur_offset_p = (le64*)chunk_tab->cur_offset_p + 1; - chunk_tab->cur_offset_u64 += out_chunk_size; - } + return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS | + WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SPANNED | + WIM_RESHDR_FLAG_FREE)); } -/* Finishes a WIM chunk table and writes it to the output file at the correct - * offset. */ -static int -finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - struct filedes *out_fd, - off_t res_start_offset, - int write_resource_flags) +static void +stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte) { - int ret; + const struct wim_resource_spec *rspec; - if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { - ret = full_write(out_fd, - chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - chunk_tab->table_disk_size); + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + rspec = lte->rspec; + + if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { + + wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS); + + lte->out_reshdr.offset_in_wim = lte->offset_in_res; + lte->out_reshdr.uncompressed_size = 0; + lte->out_reshdr.size_in_wim = lte->size; + + lte->out_res_offset_in_wim = rspec->offset_in_wim; + lte->out_res_size_in_wim = rspec->size_in_wim; + lte->out_res_uncompressed_size = rspec->uncompressed_size; } else { - ret = full_pwrite(out_fd, - chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - chunk_tab->table_disk_size, - res_start_offset); + wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); + + lte->out_reshdr.offset_in_wim = rspec->offset_in_wim; + lte->out_reshdr.uncompressed_size = rspec->uncompressed_size; + lte->out_reshdr.size_in_wim = rspec->size_in_wim; } - if (ret) - ERROR_WITH_ERRNO("Write error"); - return ret; + lte->out_reshdr.flags = lte->flags; } -/* Write the header for a stream in a pipable WIM. - */ + +/* Write the header for a stream in a pipable WIM. */ static int write_pwm_stream_header(const struct wim_lookup_table_entry *lte, struct filedes *out_fd, @@ -270,15 +251,16 @@ write_pwm_stream_header(const struct wim_lookup_table_entry *lte, copy_hash(stream_hdr.hash, lte->hash); } - reshdr_flags = lte->flags & ~(WIM_RESHDR_FLAG_COMPRESSED | WIM_RESHDR_FLAG_PACKED_STREAMS); + reshdr_flags = filter_resource_flags(lte->flags); reshdr_flags |= additional_reshdr_flags; stream_hdr.flags = cpu_to_le32(reshdr_flags); ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr)); if (ret) - ERROR_WITH_ERRNO("Error writing stream header"); + ERROR_WITH_ERRNO("Write error"); return ret; } +#if 0 static int seek_and_truncate(struct filedes *out_fd, off_t offset) { @@ -290,1530 +272,1119 @@ seek_and_truncate(struct filedes *out_fd, off_t offset) } return 0; } +#endif -static int -finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) -{ - u8 md[SHA1_HASH_SIZE]; - - sha1_final(md, sha_ctx); - if (lte->unhashed) { - copy_hash(lte->hash, md); - } else if (!hashes_equal(md, lte->hash)) { - ERROR("WIM resource has incorrect hash!"); - if (lte_filename_valid(lte)) { - ERROR("We were reading it from \"%"TS"\"; maybe " - "it changed while we were reading it.", - lte->file_on_disk); +struct write_streams_progress_data { + wimlib_progress_func_t progress_func; + union wimlib_progress_info progress; + uint64_t next_progress; + WIMStruct *prev_wim_part; +}; + +static void +do_write_streams_progress(struct write_streams_progress_data *progress_data, + struct wim_lookup_table_entry *lte, + bool stream_discarded) +{ + union wimlib_progress_info *progress = &progress_data->progress; + bool new_wim_part; + + if (stream_discarded) { + progress->write_streams.total_bytes -= lte->size; + if (progress_data->next_progress != ~(uint64_t)0 && + progress_data->next_progress > progress->write_streams.total_bytes) + { + progress_data->next_progress = progress->write_streams.total_bytes; + } + } else { + progress->write_streams.completed_bytes += lte->size; + } + new_wim_part = false; + if (lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->wim != progress_data->prev_wim_part) + { + if (progress_data->prev_wim_part) { + new_wim_part = true; + progress->write_streams.completed_parts++; + } + progress_data->prev_wim_part = lte->rspec->wim; + } + progress->write_streams.completed_streams++; + if (progress_data->progress_func + && (progress->write_streams.completed_bytes >= progress_data->next_progress + || new_wim_part)) + { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress_data->next_progress == progress->write_streams.total_bytes) { + progress_data->next_progress = ~(uint64_t)0; + } else { + progress_data->next_progress = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); } - return WIMLIB_ERR_INVALID_RESOURCE_HASH; } - return 0; } -struct write_resource_ctx { +struct write_streams_ctx { + /* File descriptor the streams are being written to. */ + struct filedes *out_fd; + + /* Lookup table for the WIMStruct on whose behalf the streams are being + * written. */ + struct wim_lookup_table *lookup_table; + + /* Compression format to use. */ int out_ctype; + + /* Maximum uncompressed chunk size in compressed resources to use. */ u32 out_chunk_size; - struct wimlib_lzx_context *comp_ctx; - struct chunk_table *chunk_tab; - struct filedes *out_fd; - SHA_CTX sha_ctx; - bool doing_sha; - int resource_flags; -}; -static int -write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) -{ - struct write_resource_ctx *ctx = _ctx; - const void *out_chunk; - unsigned out_chunk_size; - int ret; - void *compressed_chunk = NULL; - unsigned compressed_size; - bool compressed_chunk_malloced = false; + /* Flags that affect how the streams will be written. */ + int write_resource_flags; - if (ctx->doing_sha) - sha1_update(&ctx->sha_ctx, chunk, chunk_size); + /* Data used for issuing WRITE_STREAMS progress. */ + struct write_streams_progress_data progress_data; - out_chunk = chunk; - out_chunk_size = chunk_size; - if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + struct filter_context *filter_ctx; - /* Compress the chunk. */ - if (chunk_size <= STACK_MAX) { - compressed_chunk = alloca(chunk_size); - } else { - compressed_chunk = MALLOC(chunk_size); - if (compressed_chunk == NULL) - return WIMLIB_ERR_NOMEM; - compressed_chunk_malloced = true; - } + /* Upper bound on the total number of bytes that need to be compressed. + * */ + u64 num_bytes_to_compress; - compressed_size = compress_chunk(chunk, chunk_size, - compressed_chunk, - ctx->out_ctype, - ctx->comp_ctx); - /* Use compressed data if compression to less than input size - * was successful. */ - if (compressed_size) { - out_chunk = compressed_chunk; - out_chunk_size = compressed_size; - } - } + /* Pointer to the chunk_compressor implementation being used for + * compressing chunks of data, or NULL if chunks are being written + * uncompressed. */ + struct chunk_compressor *compressor; - if (ctx->chunk_tab) { - /* Update chunk table accounting. */ - chunk_tab_record_chunk(ctx->chunk_tab, out_chunk_size); + /* Buffer for dividing the read data into chunks of size + * @out_chunk_size. */ + u8 *chunk_buf; - /* If writing compressed chunks to a pipable WIM, before the - * chunk data write a chunk header that provides the compressed - * chunk size. */ - if (ctx->resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { - struct pwm_chunk_hdr chunk_hdr = { - .compressed_size = cpu_to_le32(out_chunk_size), - }; - ret = full_write(ctx->out_fd, &chunk_hdr, - sizeof(chunk_hdr)); - if (ret) - goto error; - } - } + /* Number of bytes in @chunk_buf that are currently filled. */ + size_t chunk_buf_filled; - /* Write the chunk data. */ - ret = full_write(ctx->out_fd, out_chunk, out_chunk_size); - if (ret) - goto error; + /* List of streams that currently have chunks being compressed. */ + struct list_head pending_streams; -out_free_memory: - if (compressed_chunk_malloced) - FREE(compressed_chunk); - return ret; + /* Set to true if the stream currently being read was a duplicate, and + * therefore the corresponding stream entry needs to be freed once the + * read finishes. (In this case we add the duplicate entry to + * pending_streams rather than the entry being read.) */ + bool stream_was_duplicate; -error: - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - goto out_free_memory; + /* Current uncompressed offset in the resource being read. */ + u64 cur_read_res_offset; + + /* Uncompressed size of the resource currently being read. */ + u64 cur_read_res_size; + + /* Current uncompressed offset in the resource being written. */ + u64 cur_write_res_offset; + + /* Uncompressed size of resource currently being written. */ + u64 cur_write_res_size; + + /* Array that is filled in with compressed chunk sizes as a resource is + * being written. */ + u64 *chunk_csizes; + + /* Index of next entry in @chunk_csizes to fill in. */ + size_t chunk_index; + + /* Number of entries in @chunk_csizes currently allocated. */ + size_t num_alloc_chunks; + + /* Offset in the output file of the start of the chunks of the resource + * currently being written. */ + u64 chunks_start_offset; +}; + +static u64 +get_chunk_entry_size(u64 res_size) +{ + if (res_size <= UINT32_MAX) + return 4; + else + return 8; } -/* - * write_wim_resource()- - * - * Write a resource to an output WIM. - * - * @lte: - * Lookup table entry for the resource, which could be in another WIM, in - * an external file, or in another location. - * - * @out_fd: - * File descriptor opened to the output WIM. - * - * @out_ctype: - * One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate which - * compression algorithm to use. - * - * @out_chunk_size: - * Compressed chunk size to use. - * - * @out_reshdr: - * On success, this is filled in with the offset, flags, compressed size, - * and uncompressed size of the resource in the output WIM. - * - * @resource_flags: - * * WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed even - * if it could otherwise be copied directly from the input; - * * WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE if writing a resource for a pipable WIM - * (and the output file descriptor may be a pipe). - * - * @comp_ctx: - * Location of LZX compression context pointer, which will be allocated or - * updated if needed. (Initialize to NULL.) - * - * Additional notes: The SHA1 message digest of the uncompressed data is - * calculated (except when doing a raw copy --- see below). If the @unhashed - * flag is set on the lookup table entry, this message digest is simply copied - * to it; otherwise, the message digest is compared with the existing one, and - * this function will fail if they do not match. - */ +/* Reserve space for the chunk table and prepare to accumulate the chunk table + * in memory. */ static int -write_wim_resource(struct wim_lookup_table_entry *lte, - struct filedes *out_fd, int out_ctype, - u32 out_chunk_size, - struct wim_reshdr *out_reshdr, - int resource_flags, - struct wimlib_lzx_context **comp_ctx) +begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) { - struct write_resource_ctx write_ctx; - off_t res_start_offset; - u64 read_size; + u64 expected_num_chunks; + u64 expected_num_chunk_entries; + size_t reserve_size; int ret; - /* Mask out any irrelevant flags, since this function also uses this - * variable to store WIMLIB_READ_RESOURCE flags. */ - resource_flags &= WIMLIB_WRITE_RESOURCE_MASK; - - /* Get current position in output WIM. */ - res_start_offset = out_fd->offset; - - /* If we are not forcing the data to be recompressed, and the input - * resource is located in a WIM with a compression mode compatible with - * the output, we can simply copy the compressed data without - * recompressing it. This also means we must skip calculating the SHA1, - * as we never will see the uncompressed data. */ - if (can_raw_copy(lte, resource_flags, out_ctype, out_chunk_size)) { - /* Normally, for raw copies we can request a RAW_FULL read, but - * if we're reading from a pipable resource and writing a - * non-pipable resource or vice versa, then a RAW_CHUNKS read - * needs to be requested so that the written resource can be - * appropriately formatted. However, in neither case is any - * actual decompression needed. */ - if (lte->rspec->is_pipable == !!(resource_flags & - WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) - { - resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL; - read_size = lte->rspec->size_in_wim; - } else { - resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS; - read_size = lte->size; + /* Calculate the number of chunks and chunk entries that should be + * needed for the resource. These normally will be the final values, + * but in PACKED_STREAMS mode some of the streams we're planning to + * write into the resource may be duplicates, and therefore discarded, + * potentially decreasing the number of chunk entries needed. */ + expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size); + expected_num_chunk_entries = expected_num_chunks; + if (!(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS)) + expected_num_chunk_entries--; + + /* Make sure the chunk_csizes array is long enough to store the + * compressed size of each chunk. */ + if (expected_num_chunks > ctx->num_alloc_chunks) { + u64 new_length = expected_num_chunks + 50; + + if ((size_t)new_length != new_length) { + ERROR("Resource size too large (%"PRIu64" bytes!", + res_expected_size); + return WIMLIB_ERR_NOMEM; } - write_ctx.doing_sha = false; - } else { - write_ctx.doing_sha = true; - sha1_init(&write_ctx.sha_ctx); - read_size = lte->size; - } - /* Set the output compression mode and initialize chunk table if needed. - */ - write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - write_ctx.out_chunk_size = out_chunk_size; - write_ctx.chunk_tab = NULL; - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - wimlib_assert(out_chunk_size > 0); - if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) { - /* Compression needed. */ - write_ctx.out_ctype = out_ctype; - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { - ret = wimlib_lzx_alloc_context(out_chunk_size, - NULL, comp_ctx); - if (ret) - goto out; - } - write_ctx.comp_ctx = *comp_ctx; - } - if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)) { - /* Chunk table needed. */ - ret = begin_wim_resource_chunk_tab(lte, out_fd, - out_chunk_size, - &write_ctx.chunk_tab, - resource_flags); - if (ret) - goto out; + FREE(ctx->chunk_csizes); + ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0])); + if (ctx->chunk_csizes == NULL) { + ctx->num_alloc_chunks = 0; + return WIMLIB_ERR_NOMEM; } - } - - /* If writing a pipable resource, write the stream header and update - * @res_start_offset to be the end of the stream header. */ - if (resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { - int reshdr_flags = 0; - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) - reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; - ret = write_pwm_stream_header(lte, out_fd, reshdr_flags); + ctx->num_alloc_chunks = new_length; + } + + ctx->chunk_index = 0; + + if (!(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { + /* Reserve space for the chunk table in the output file. In the + * case of packed resources this reserves the upper bound for + * the needed space, not necessarily the exact space which will + * prove to be needed. At this point, we just use @chunk_csizes + * for a buffer of 0's because the actual compressed chunk sizes + * are unknown. */ + reserve_size = expected_num_chunk_entries * get_chunk_entry_size(res_expected_size); + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) + reserve_size += sizeof(struct alt_chunk_table_header_disk); + memset(ctx->chunk_csizes, 0, reserve_size); + ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size); if (ret) - goto out_free_chunk_tab; - res_start_offset = out_fd->offset; + return ret; } + return 0; +} - /* Write the entire resource by reading the entire resource and feeding - * the data through write_resource_cb(). */ - write_ctx.out_fd = out_fd; - write_ctx.resource_flags = resource_flags; -try_write_again: - ret = read_stream_prefix(lte, read_size, write_resource_cb, - &write_ctx, resource_flags); - if (ret) - goto out_free_chunk_tab; +static int +begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size) +{ + int ret; - /* Verify SHA1 message digest of the resource, or set the hash for the - * first time. */ - if (write_ctx.doing_sha) { - ret = finalize_and_check_sha1(&write_ctx.sha_ctx, lte); - if (ret) - goto out_free_chunk_tab; - } + wimlib_assert(res_expected_size != 0); - /* Write chunk table if needed. */ - if (write_ctx.chunk_tab) { - ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, - out_fd, - res_start_offset, - resource_flags); + if (ctx->compressor != NULL) { + ret = begin_chunk_table(ctx, res_expected_size); if (ret) - goto out_free_chunk_tab; + return ret; } - /* Fill in out_reshdr with information about the newly written - * resource. */ - out_reshdr->size_in_wim = out_fd->offset - res_start_offset; - out_reshdr->flags = lte->flags & ~WIM_RESHDR_FLAG_PACKED_STREAMS; - if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) - out_reshdr->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; - else - out_reshdr->flags |= WIM_RESHDR_FLAG_COMPRESSED; - out_reshdr->offset_in_wim = res_start_offset; - out_reshdr->uncompressed_size = lte->size; - - /* Check for resources compressed to greater than their original size - * and write them uncompressed instead. (But never do this if writing - * to a pipe, and don't bother if we did a raw copy.) */ - if (out_reshdr->size_in_wim > out_reshdr->uncompressed_size && - !(resource_flags & (WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE | - WIMLIB_READ_RESOURCE_FLAG_RAW))) - { - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - out_reshdr->uncompressed_size, out_reshdr->size_in_wim); - ret = seek_and_truncate(out_fd, res_start_offset); - if (ret) - goto out_free_chunk_tab; - out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - FREE(write_ctx.chunk_tab); - write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - write_ctx.chunk_tab = NULL; - write_ctx.doing_sha = false; - goto try_write_again; - } - if (resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW) { - DEBUG("Copied raw compressed data " - "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - out_reshdr->uncompressed_size, out_reshdr->size_in_wim, - out_reshdr->offset_in_wim, out_reshdr->flags); - } else if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - DEBUG("Wrote compressed resource " - "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - out_reshdr->uncompressed_size, out_reshdr->size_in_wim, - out_reshdr->offset_in_wim, out_reshdr->flags); - } else { - DEBUG("Wrote uncompressed resource " - "(%"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - out_reshdr->uncompressed_size, - out_reshdr->offset_in_wim, out_reshdr->flags); - } - ret = 0; -out_free_chunk_tab: - FREE(write_ctx.chunk_tab); -out: - return ret; + /* Output file descriptor is now positioned at the offset at which to + * write the first chunk of the resource. */ + ctx->chunks_start_offset = ctx->out_fd->offset; + ctx->cur_write_res_offset = 0; + ctx->cur_write_res_size = res_expected_size; + return 0; } -/* Like write_wim_resource(), but the resource is specified by a buffer of - * uncompressed data rather a lookup table entry. Also writes the SHA1 message - * digest of the buffer to @hash_ret if it is non-NULL. */ -int -write_wim_resource_from_buffer(const void *buf, size_t buf_size, - int reshdr_flags, struct filedes *out_fd, - int out_ctype, - u32 out_chunk_size, - struct wim_reshdr *out_reshdr, - u8 *hash_ret, int write_resource_flags, - struct wimlib_lzx_context **comp_ctx) +static int +end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, + u64 *res_start_offset_ret, u64 *res_store_size_ret) { + size_t actual_num_chunks; + size_t actual_num_chunk_entries; + size_t chunk_entry_size; int ret; - struct wim_lookup_table_entry *lte; - /* Set up a temporary lookup table entry to provide to - * write_wim_resource(). */ + actual_num_chunks = ctx->chunk_index; + actual_num_chunk_entries = actual_num_chunks; + if (!(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS)) + actual_num_chunk_entries--; - lte = new_lookup_table_entry(); - if (lte == NULL) - return WIMLIB_ERR_NOMEM; + chunk_entry_size = get_chunk_entry_size(res_actual_size); - lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte->attached_buffer = (void*)buf; - lte->size = buf_size; - lte->flags = reshdr_flags; + typedef le64 __attribute__((may_alias)) aliased_le64_t; + typedef le32 __attribute__((may_alias)) aliased_le32_t; - if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { - sha1_buffer(buf, buf_size, lte->hash); - lte->unhashed = 0; + if (chunk_entry_size == 4) { + aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + for (size_t i = 0; i < actual_num_chunk_entries; i++) + entries[i] = cpu_to_le32(ctx->chunk_csizes[i]); + } else { + u32 offset = ctx->chunk_csizes[0]; + for (size_t i = 0; i < actual_num_chunk_entries; i++) { + u32 next_size = ctx->chunk_csizes[i + 1]; + entries[i] = cpu_to_le32(offset); + offset += next_size; + } + } } else { - lte->unhashed = 1; + aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + for (size_t i = 0; i < actual_num_chunk_entries; i++) + entries[i] = cpu_to_le64(ctx->chunk_csizes[i]); + } else { + u64 offset = ctx->chunk_csizes[0]; + for (size_t i = 0; i < actual_num_chunk_entries; i++) { + u64 next_size = ctx->chunk_csizes[i + 1]; + entries[i] = cpu_to_le64(offset); + offset += next_size; + } + } } - ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, - out_reshdr, write_resource_flags, comp_ctx); - if (ret) - goto out_free_lte; - if (hash_ret) - copy_hash(hash_ret, lte->hash); - ret = 0; -out_free_lte: - lte->resource_location = RESOURCE_NONEXISTENT; - free_lookup_table_entry(lte); - return ret; -} + size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size; + u64 res_start_offset; + u64 res_end_offset = ctx->out_fd->offset; + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size); + if (ret) + goto error; + res_end_offset = ctx->out_fd->offset; + res_start_offset = ctx->chunks_start_offset; + } else { + res_end_offset = ctx->out_fd->offset; -#ifdef ENABLE_MULTITHREADED_COMPRESSION + u64 chunk_table_offset; -/* Blocking shared queue (solves the producer-consumer problem) */ -struct shared_queue { - unsigned size; - unsigned front; - unsigned back; - unsigned filled_slots; - void **array; - pthread_mutex_t lock; - pthread_cond_t msg_avail_cond; - pthread_cond_t space_avail_cond; -}; + chunk_table_offset = ctx->chunks_start_offset - chunk_table_size; -static int -shared_queue_init(struct shared_queue *q, unsigned size) -{ - wimlib_assert(size != 0); - q->array = CALLOC(sizeof(q->array[0]), size); - if (!q->array) - goto err; - q->filled_slots = 0; - q->front = 0; - q->back = size - 1; - q->size = size; - if (pthread_mutex_init(&q->lock, NULL)) { - ERROR_WITH_ERRNO("Failed to initialize mutex"); - goto err; - } - if (pthread_cond_init(&q->msg_avail_cond, NULL)) { - ERROR_WITH_ERRNO("Failed to initialize condition variable"); - goto err_destroy_lock; - } - if (pthread_cond_init(&q->space_avail_cond, NULL)) { - ERROR_WITH_ERRNO("Failed to initialize condition variable"); - goto err_destroy_msg_avail_cond; + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + struct alt_chunk_table_header_disk hdr; + + hdr.res_usize = cpu_to_le64(res_actual_size); + hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size); + hdr.compression_format = cpu_to_le32(ctx->out_ctype); + + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3); + + ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr), + chunk_table_offset - sizeof(hdr)); + if (ret) + goto error; + res_start_offset = chunk_table_offset - sizeof(hdr); + } else { + res_start_offset = chunk_table_offset; + } + + ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes, + chunk_table_size, chunk_table_offset); + if (ret) + goto error; } + + *res_start_offset_ret = res_start_offset; + *res_store_size_ret = res_end_offset - res_start_offset; + return 0; -err_destroy_msg_avail_cond: - pthread_cond_destroy(&q->msg_avail_cond); -err_destroy_lock: - pthread_mutex_destroy(&q->lock); -err: - return WIMLIB_ERR_NOMEM; -} -static void -shared_queue_destroy(struct shared_queue *q) -{ - FREE(q->array); - pthread_mutex_destroy(&q->lock); - pthread_cond_destroy(&q->msg_avail_cond); - pthread_cond_destroy(&q->space_avail_cond); +error: + ERROR_WITH_ERRNO("Write error"); + return ret; } -static void -shared_queue_put(struct shared_queue *q, void *obj) +/* Finish writing a WIM resource by writing or updating the chunk table (if not + * writing the data uncompressed) and loading its metadata into @out_reshdr. */ +static int +end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr) { - pthread_mutex_lock(&q->lock); - while (q->filled_slots == q->size) - pthread_cond_wait(&q->space_avail_cond, &q->lock); + int ret; + u64 res_size_in_wim; + u64 res_uncompressed_size; + u64 res_offset_in_wim; - q->back = (q->back + 1) % q->size; - q->array[q->back] = obj; - q->filled_slots++; + wimlib_assert(ctx->cur_write_res_size == ctx->cur_write_res_offset); + res_uncompressed_size = ctx->cur_write_res_size; - pthread_cond_broadcast(&q->msg_avail_cond); - pthread_mutex_unlock(&q->lock); + if (ctx->compressor) { + ret = end_chunk_table(ctx, res_uncompressed_size, + &res_offset_in_wim, &res_size_in_wim); + if (ret) + return ret; + } else { + res_offset_in_wim = ctx->chunks_start_offset; + res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim; + } + out_reshdr->uncompressed_size = res_uncompressed_size; + out_reshdr->size_in_wim = res_size_in_wim; + out_reshdr->offset_in_wim = res_offset_in_wim; + DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"", + res_uncompressed_size, res_size_in_wim, res_offset_in_wim); + return 0; } -static void * -shared_queue_get(struct shared_queue *q) +/* Begin processing a stream for writing. */ +static int +write_stream_begin_read(struct wim_lookup_table_entry *lte, + bool is_partial_res, void *_ctx) { - void *obj; + struct write_streams_ctx *ctx = _ctx; + int ret; - pthread_mutex_lock(&q->lock); - while (q->filled_slots == 0) - pthread_cond_wait(&q->msg_avail_cond, &q->lock); + wimlib_assert(lte->size > 0); - obj = q->array[q->front]; - q->array[q->front] = NULL; - q->front = (q->front + 1) % q->size; - q->filled_slots--; + ctx->cur_read_res_offset = 0; + ctx->cur_read_res_size = lte->size; - pthread_cond_broadcast(&q->space_avail_cond); - pthread_mutex_unlock(&q->lock); - return obj; -} + /* As an optimization, we allow some streams to be "unhashed", meaning + * their SHA1 message digests are unknown. This is the case with + * streams that are added by scanning a directry tree with + * wimlib_add_image(), for example. Since WIM uses single-instance + * streams, we don't know whether such each such stream really need to + * written until it is actually checksummed, unless it has a unique + * size. In such cases we read and checksum the stream in this + * function, thereby advancing ahead of read_stream_list(), which will + * still provide the data again to write_stream_process_chunk(). This + * is okay because an unhashed stream cannot be in a WIM resource, which + * might be costly to decompress. */ + ctx->stream_was_duplicate = false; + if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) { -struct compressor_thread_params { - struct shared_queue *res_to_compress_queue; - struct shared_queue *compressed_res_queue; - int out_ctype; - struct wimlib_lzx_context *comp_ctx; -}; + wimlib_assert(!is_partial_res); + wimlib_assert(ctx->lookup_table != NULL); -#define MAX_CHUNKS_PER_MSG 2 + struct wim_lookup_table_entry *lte_new; -struct message { - struct wim_lookup_table_entry *lte; - u32 out_chunk_size; - u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; - u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; - unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; - struct iovec out_chunks[MAX_CHUNKS_PER_MSG]; - unsigned num_chunks; - struct list_head list; - bool complete; - u64 begin_chunk; -}; + ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new); + if (ret) + return ret; + if (lte_new != lte) { + /* Duplicate stream detected. */ -static void -compress_chunks(struct message *msg, int out_ctype, - struct wimlib_lzx_context *comp_ctx) -{ - for (unsigned i = 0; i < msg->num_chunks; i++) { - unsigned len; - - len = compress_chunk(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i], - out_ctype, - comp_ctx); - - void *out_chunk; - unsigned out_len; - if (len) { - /* To be written compressed */ - out_chunk = msg->compressed_chunks[i]; - out_len = len; - } else { - /* To be written uncompressed */ - out_chunk = msg->uncompressed_chunks[i]; - out_len = msg->uncompressed_chunk_sizes[i]; + if (lte_new->will_be_in_output_wim || + stream_filtered(lte_new, ctx->filter_ctx)) + { + /* The duplicate stream is already being + * included in the output WIM, or it would be + * filtered out if it had been. Skip writing + * this stream (and reading it again) entirely, + * passing its output reference count to the + * duplicate stream in the former case. */ + DEBUG("Discarding duplicate stream of " + "length %"PRIu64, lte->size); + do_write_streams_progress(&ctx->progress_data, + lte, true); + list_del(<e->write_streams_list); + list_del(<e->lookup_table_list); + if (lte_new->will_be_in_output_wim) + lte_new->out_refcnt += lte->out_refcnt; + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) + ctx->cur_write_res_size -= lte->size; + free_lookup_table_entry(lte); + return BEGIN_STREAM_STATUS_SKIP_STREAM; + } else { + /* The duplicate stream can validly be written, + * but was not marked as such. Discard the + * current stream entry and use the duplicate, + * but actually freeing the current entry must + * wait until read_stream_list() has finished + * reading its data. */ + DEBUG("Stream duplicate, but not already " + "selected for writing."); + list_replace(<e->write_streams_list, + <e_new->write_streams_list); + list_replace(<e->lookup_table_list, + <e_new->lookup_table_list); + lte_new->out_refcnt = lte->out_refcnt; + lte_new->will_be_in_output_wim = 1; + ctx->stream_was_duplicate = true; + lte = lte_new; + } } - msg->out_chunks[i].iov_base = out_chunk; - msg->out_chunks[i].iov_len = out_len; } + list_move_tail(<e->write_streams_list, &ctx->pending_streams); + return 0; } -/* Compressor thread routine. This is a lot simpler than the main thread - * routine: just repeatedly get a group of chunks from the - * res_to_compress_queue, compress them, and put them in the - * compressed_res_queue. A NULL pointer indicates that the thread should stop. - * */ -static void * -compressor_thread_proc(void *arg) +/* Write the next chunk of (typically compressed) data to the output WIM, + * handling the writing of the chunk table. */ +static int +write_chunk(struct write_streams_ctx *ctx, const void *cchunk, + size_t csize, size_t usize) { - struct compressor_thread_params *params = arg; - struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; - struct shared_queue *compressed_res_queue = params->compressed_res_queue; - struct message *msg; + int ret; - DEBUG("Compressor thread ready"); - while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { - compress_chunks(msg, params->out_ctype, params->comp_ctx); - shared_queue_put(compressed_res_queue, msg); - } - DEBUG("Compressor thread terminating"); - return NULL; -} -#endif /* ENABLE_MULTITHREADED_COMPRESSION */ + if (ctx->cur_write_res_offset == 0 && + !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS)) + { + /* Starting to write a new stream in non-packed mode. */ -struct write_streams_progress_data { - wimlib_progress_func_t progress_func; - union wimlib_progress_info progress; - uint64_t next_progress; - WIMStruct *prev_wim_part; -}; + struct wim_lookup_table_entry *lte; -static void -do_write_streams_progress(struct write_streams_progress_data *progress_data, - struct wim_lookup_table_entry *lte, - bool stream_discarded) -{ - union wimlib_progress_info *progress = &progress_data->progress; - bool new_wim_part; + lte = list_entry(ctx->pending_streams.next, + struct wim_lookup_table_entry, write_streams_list); - if (stream_discarded) { - progress->write_streams.total_bytes -= lte->size; - if (progress_data->next_progress != ~(uint64_t)0 && - progress_data->next_progress > progress->write_streams.total_bytes) - { - progress_data->next_progress = progress->write_streams.total_bytes; - } - } else { - progress->write_streams.completed_bytes += lte->size; - } - new_wim_part = false; - if (lte->resource_location == RESOURCE_IN_WIM && - lte->rspec->wim != progress_data->prev_wim_part) - { - if (progress_data->prev_wim_part) { - new_wim_part = true; - progress->write_streams.completed_parts++; + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + int additional_reshdr_flags = 0; + if (ctx->compressor != NULL) + additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; + + DEBUG("Writing pipable WIM stream header " + "(offset=%"PRIu64")", ctx->out_fd->offset); + + ret = write_pwm_stream_header(lte, ctx->out_fd, + additional_reshdr_flags); + if (ret) + return ret; } - progress_data->prev_wim_part = lte->rspec->wim; + + ret = begin_write_resource(ctx, lte->size); + if (ret) + return ret; } - progress->write_streams.completed_streams++; - if (progress_data->progress_func - && (progress->write_streams.completed_bytes >= progress_data->next_progress - || new_wim_part)) - { - progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - if (progress_data->next_progress == progress->write_streams.total_bytes) { - progress_data->next_progress = ~(uint64_t)0; - } else { - progress_data->next_progress = - min(progress->write_streams.total_bytes, - progress->write_streams.completed_bytes + - progress->write_streams.total_bytes / 100); + + if (ctx->compressor != NULL) { + /* Record the compresed chunk size. */ + wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks); + ctx->chunk_csizes[ctx->chunk_index++] = csize; + + /* If writing a pipable WIM, before the chunk data write a chunk + * header that provides the compressed chunk size. */ + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + struct pwm_chunk_hdr chunk_hdr = { + .compressed_size = cpu_to_le32(csize), + }; + ret = full_write(ctx->out_fd, &chunk_hdr, + sizeof(chunk_hdr)); + if (ret) + goto error; } } -} -struct serial_write_stream_ctx { - struct filedes *out_fd; - int out_ctype; - u32 out_chunk_size; - struct wimlib_lzx_context **comp_ctx; - int write_resource_flags; -}; + /* Write the chunk data. */ + ret = full_write(ctx->out_fd, cchunk, csize); + if (ret) + goto error; -static int -serial_write_stream(struct wim_lookup_table_entry *lte, void *_ctx) -{ - struct serial_write_stream_ctx *ctx = _ctx; - return write_wim_resource(lte, ctx->out_fd, - ctx->out_ctype, - ctx->out_chunk_size, - <e->out_reshdr, - ctx->write_resource_flags, - ctx->comp_ctx); -} + ctx->cur_write_res_offset += usize; + if (ctx->cur_write_res_offset == ctx->cur_write_res_size && + !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS)) + { + struct wim_lookup_table_entry *lte; -/* Write a list of streams, taking into account that some streams may be - * duplicates that are checksummed and discarded on the fly, and also delegating - * the actual writing of a stream to a function @write_stream_cb, which is - * passed the context @write_stream_ctx. */ -static int -do_write_stream_list(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - int (*write_stream_cb)(struct wim_lookup_table_entry *, void *), - void *write_stream_ctx, - struct write_streams_progress_data *progress_data) -{ - int ret = 0; - struct wim_lookup_table_entry *lte; - bool stream_discarded; - - /* For each stream in @stream_list ... */ - while (!list_empty(stream_list)) { - stream_discarded = false; - lte = container_of(stream_list->next, - struct wim_lookup_table_entry, - write_streams_list); - list_del(<e->write_streams_list); - if (lte->unhashed && !lte->unique_size) { - /* Unhashed stream that shares a size with some other - * stream in the WIM we are writing. The stream must be - * checksummed to know if we need to write it or not. */ - struct wim_lookup_table_entry *tmp; - u32 orig_out_refcnt = lte->out_refcnt; - - ret = hash_unhashed_stream(lte, lookup_table, &tmp); - if (ret) - break; - if (tmp != lte) { - /* We found a duplicate stream. 'lte' was - * freed, so replace it with the duplicate. */ - lte = tmp; - - /* 'out_refcnt' was transferred to the - * duplicate, and we can detect if the duplicate - * stream was already referenced for writing by - * checking if its 'out_refcnt' is higher than - * that of the original stream. In such cases, - * the current stream can be discarded. We can - * also discard the current stream if it was - * previously marked as filtered (e.g. already - * present in the WIM being written). */ - if (lte->out_refcnt > orig_out_refcnt || - lte->filtered) { - DEBUG("Discarding duplicate stream of " - "length %"PRIu64, - lte->size); - lte->no_progress = 0; - stream_discarded = true; - goto skip_to_progress; - } - } - } + lte = list_entry(ctx->pending_streams.next, + struct wim_lookup_table_entry, write_streams_list); + wimlib_assert(ctx->cur_write_res_offset == lte->size); + + /* Finished writing a stream in non-packed mode. */ - /* Here, @lte is either a hashed stream or an unhashed stream - * with a unique size. In either case we know that the stream - * has to be written. In either case the SHA1 message digest - * will be calculated over the stream while writing it; however, - * in the former case this is done merely to check the data, - * while in the latter case this is done because we do not have - * the SHA1 message digest yet. */ - wimlib_assert(lte->out_refcnt != 0); - lte->deferred = 0; - lte->no_progress = 0; - ret = (*write_stream_cb)(lte, write_stream_ctx); + do_write_streams_progress(&ctx->progress_data, lte, false); + + ret = end_write_resource(ctx, <e->out_reshdr); if (ret) - break; - /* In parallel mode, some streams are deferred for later, - * serialized processing; ignore them here. */ - if (lte->deferred) - continue; - if (lte->unhashed) { - list_del(<e->unhashed_list); - lookup_table_insert(lookup_table, lte); - lte->unhashed = 0; - } - skip_to_progress: - if (!lte->no_progress) { - do_write_streams_progress(progress_data, - lte, stream_discarded); - } + return ret; + + wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size); + + lte->out_reshdr.flags = filter_resource_flags(lte->flags); + if (ctx->compressor != NULL) + lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED; + + list_del(<e->write_streams_list); + ctx->cur_write_res_offset = 0; } + + return 0; + +error: + ERROR_WITH_ERRNO("Write error"); return ret; } static int -do_write_stream_list_serial(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, - int out_ctype, - u32 out_chunk_size, - struct wimlib_lzx_context **comp_ctx, - int write_resource_flags, - struct write_streams_progress_data *progress_data) -{ - struct serial_write_stream_ctx ctx = { - .out_fd = out_fd, - .out_ctype = out_ctype, - .out_chunk_size = out_chunk_size, - .write_resource_flags = write_resource_flags, - .comp_ctx = comp_ctx, - }; - return do_write_stream_list(stream_list, - lookup_table, - serial_write_stream, - &ctx, - progress_data); -} - -static inline int -write_flags_to_resource_flags(int write_flags) +submit_chunk_for_compression(struct write_streams_ctx *ctx, + const void *chunk, size_t size) { - int resource_flags = 0; + /* While we are unable to submit the chunk for compression (due to too + * many chunks already outstanding), retrieve and write the next + * compressed chunk. */ + while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) { + const void *cchunk; + unsigned csize; + unsigned usize; + bool bret; + int ret; - if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS; - if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) - resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; - return resource_flags; -} + bret = ctx->compressor->get_chunk(ctx->compressor, + &cchunk, &csize, &usize); -static int -write_stream_list_serial(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, - int out_ctype, - u32 out_chunk_size, - struct wimlib_lzx_context **comp_ctx, - int write_resource_flags, - struct write_streams_progress_data *progress_data) -{ - union wimlib_progress_info *progress = &progress_data->progress; - DEBUG("Writing stream list of size %"PRIu64" (serial version)", - progress->write_streams.total_streams); - progress->write_streams.num_threads = 1; - if (progress_data->progress_func) { - progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); + wimlib_assert(bret); + + ret = write_chunk(ctx, cchunk, csize, usize); + if (ret) + return ret; } - return do_write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - out_chunk_size, - comp_ctx, - write_resource_flags, - progress_data); + return 0; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION +/* Process the next chunk of data to be written to a WIM resource. */ static int -write_wim_chunks(struct message *msg, struct filedes *out_fd, - struct chunk_table *chunk_tab, - int write_resource_flags) +write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) { - struct iovec *vecs; - struct pwm_chunk_hdr *chunk_hdrs; - unsigned nvecs; + struct write_streams_ctx *ctx = _ctx; int ret; + const u8 *chunkptr, *chunkend; - for (unsigned i = 0; i < msg->num_chunks; i++) - chunk_tab_record_chunk(chunk_tab, msg->out_chunks[i].iov_len); + wimlib_assert(size != 0); - if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { - nvecs = msg->num_chunks; - vecs = msg->out_chunks; - } else { - /* Special case: If writing a compressed resource to a pipable - * WIM, prefix each compressed chunk with a header that gives - * its compressed size. */ - nvecs = msg->num_chunks * 2; - vecs = alloca(nvecs * sizeof(vecs[0])); - chunk_hdrs = alloca(msg->num_chunks * sizeof(chunk_hdrs[0])); - - for (unsigned i = 0; i < msg->num_chunks; i++) { - chunk_hdrs[i].compressed_size = cpu_to_le32(msg->out_chunks[i].iov_len); - vecs[i * 2].iov_base = &chunk_hdrs[i]; - vecs[i * 2].iov_len = sizeof(chunk_hdrs[i]); - vecs[i * 2 + 1].iov_base = msg->out_chunks[i].iov_base; - vecs[i * 2 + 1].iov_len = msg->out_chunks[i].iov_len; + if (ctx->compressor == NULL) { + /* Write chunk uncompressed. */ + ret = write_chunk(ctx, chunk, size, size); + if (ret) + return ret; + ctx->cur_read_res_offset += size; + return 0; + } + + /* Submit the chunk for compression, but take into account that the + * @size the chunk was provided in may not correspond to the + * @out_chunk_size being used for compression. */ + chunkptr = chunk; + chunkend = chunkptr + size; + do { + const u8 *resized_chunk; + size_t needed_chunk_size; + + if (ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + needed_chunk_size = ctx->out_chunk_size; + } else { + u64 res_bytes_remaining; + + res_bytes_remaining = ctx->cur_read_res_size - + ctx->cur_read_res_offset; + needed_chunk_size = min(ctx->out_chunk_size, + ctx->chunk_buf_filled + + res_bytes_remaining); } - } - ret = full_writev(out_fd, vecs, nvecs); - if (ret) - ERROR_WITH_ERRNO("Write error"); - return ret; -} -struct main_writer_thread_ctx { - struct list_head *stream_list; - struct wim_lookup_table *lookup_table; - struct filedes *out_fd; - off_t res_start_offset; - int out_ctype; - u32 out_chunk_size; - struct wimlib_lzx_context **comp_ctx; - int write_resource_flags; - struct shared_queue *res_to_compress_queue; - struct shared_queue *compressed_res_queue; - size_t num_messages; - struct write_streams_progress_data *progress_data; - - struct list_head available_msgs; - struct list_head outstanding_streams; - struct list_head serial_streams; - size_t num_outstanding_messages; - - SHA_CTX next_sha_ctx; - u64 next_chunk; - u64 next_num_chunks; - struct wim_lookup_table_entry *next_lte; - - struct message *msgs; - struct message *next_msg; - struct chunk_table *cur_chunk_tab; -}; + if (ctx->chunk_buf_filled == 0 && + chunkend - chunkptr >= needed_chunk_size) + { + /* No intermediate buffering needed. */ + resized_chunk = chunkptr; + chunkptr += needed_chunk_size; + ctx->cur_read_res_offset += needed_chunk_size; + } else { + /* Intermediate buffering needed. */ + size_t bytes_consumed; -static int -init_message(struct message *msg, u32 out_chunk_size) -{ - msg->out_chunk_size = out_chunk_size; - for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { - msg->compressed_chunks[i] = MALLOC(out_chunk_size); - msg->uncompressed_chunks[i] = MALLOC(out_chunk_size); - if (msg->compressed_chunks[i] == NULL || - msg->uncompressed_chunks[i] == NULL) - return WIMLIB_ERR_NOMEM; - } - return 0; -} + bytes_consumed = min(chunkend - chunkptr, + needed_chunk_size - ctx->chunk_buf_filled); -static void -destroy_message(struct message *msg) -{ - for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { - FREE(msg->compressed_chunks[i]); - FREE(msg->uncompressed_chunks[i]); - } -} + memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled], + chunkptr, bytes_consumed); -static void -free_messages(struct message *msgs, size_t num_messages) -{ - if (msgs) { - for (size_t i = 0; i < num_messages; i++) - destroy_message(&msgs[i]); - FREE(msgs); - } -} + resized_chunk = ctx->chunk_buf; -static struct message * -allocate_messages(size_t num_messages, u32 out_chunk_size) -{ - struct message *msgs; + chunkptr += bytes_consumed; + ctx->cur_read_res_offset += bytes_consumed; + ctx->chunk_buf_filled += bytes_consumed; + if (ctx->chunk_buf_filled == needed_chunk_size) { + resized_chunk = ctx->chunk_buf; + ctx->chunk_buf_filled = 0; + } else { + break; + } - msgs = CALLOC(num_messages, sizeof(struct message)); - if (msgs == NULL) - return NULL; - for (size_t i = 0; i < num_messages; i++) { - if (init_message(&msgs[i], out_chunk_size)) { - free_messages(msgs, num_messages); - return NULL; } + + ret = submit_chunk_for_compression(ctx, resized_chunk, + needed_chunk_size); + if (ret) + return ret; + + } while (chunkptr != chunkend); + return 0; +} + +/* Finish processing a stream for writing. It may not have been completely + * written yet, as the chunk_compressor implementation may still have chunks + * buffered or being compressed. */ +static int +write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx) +{ + struct write_streams_ctx *ctx = _ctx; + if (status == 0) + wimlib_assert(ctx->cur_read_res_offset == ctx->cur_read_res_size); + if (ctx->stream_was_duplicate) { + free_lookup_table_entry(lte); + } else if (lte->unhashed && ctx->lookup_table != NULL) { + list_del(<e->unhashed_list); + lookup_table_insert(ctx->lookup_table, lte); + lte->unhashed = 0; } - return msgs; + return status; } +/* Compute statistics about a list of streams that will be written. + * + * Assumes the streams are sorted such that all streams located in each distinct + * WIM (specified by WIMStruct) are together. */ static void -main_writer_thread_destroy_ctx(struct main_writer_thread_ctx *ctx) +compute_stream_list_stats(struct list_head *stream_list, + struct write_streams_ctx *ctx) { - while (ctx->num_outstanding_messages--) - shared_queue_get(ctx->compressed_res_queue); - free_messages(ctx->msgs, ctx->num_messages); - FREE(ctx->cur_chunk_tab); -} + struct wim_lookup_table_entry *lte; + u64 total_bytes = 0; + u64 num_streams = 0; + u64 total_parts = 0; + WIMStruct *prev_wim_part = NULL; -static int -main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) + list_for_each_entry(lte, stream_list, write_streams_list) { + num_streams++; + total_bytes += lte->size; + if (lte->resource_location == RESOURCE_IN_WIM) { + if (prev_wim_part != lte->rspec->wim) { + prev_wim_part = lte->rspec->wim; + total_parts++; + } + } + } + ctx->progress_data.progress.write_streams.total_bytes = total_bytes; + ctx->progress_data.progress.write_streams.total_streams = num_streams; + ctx->progress_data.progress.write_streams.completed_bytes = 0; + ctx->progress_data.progress.write_streams.completed_streams = 0; + ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype; + ctx->progress_data.progress.write_streams.total_parts = total_parts; + ctx->progress_data.progress.write_streams.completed_parts = 0; + ctx->progress_data.next_progress = 0; + ctx->progress_data.prev_wim_part = NULL; +} + +/* Find streams in @stream_list that can be copied to the output WIM in raw form + * rather than compressed. Delete these streams from @stream_list, and move one + * per resource to @raw_copy_resources. Return the total uncompressed size of + * the streams that need to be compressed. */ +static u64 +find_raw_copy_resources(struct list_head *stream_list, + int write_resource_flags, + int out_ctype, + u32 out_chunk_size, + struct list_head *raw_copy_resources) { - /* Pre-allocate all the buffers that will be needed to do the chunk - * compression. */ - ctx->msgs = allocate_messages(ctx->num_messages, ctx->out_chunk_size); - if (ctx->msgs == NULL) - return WIMLIB_ERR_NOMEM; - - /* Initially, all the messages are available to use. */ - INIT_LIST_HEAD(&ctx->available_msgs); - for (size_t i = 0; i < ctx->num_messages; i++) - list_add_tail(&ctx->msgs[i].list, &ctx->available_msgs); - - /* outstanding_streams is the list of streams that currently have had - * chunks sent off for compression. - * - * The first stream in outstanding_streams is the stream that is - * currently being written. - * - * The last stream in outstanding_streams is the stream that is - * currently being read and having chunks fed to the compressor threads. - * */ - INIT_LIST_HEAD(&ctx->outstanding_streams); - ctx->num_outstanding_messages = 0; + struct wim_lookup_table_entry *lte, *tmp; + u64 num_bytes_to_compress = 0; - /* Message currently being prepared. */ - ctx->next_msg = NULL; + INIT_LIST_HEAD(raw_copy_resources); - /* Resources that don't need any chunks compressed are added to this - * list and written directly by the main thread. */ - INIT_LIST_HEAD(&ctx->serial_streams); + /* Initialize temporary raw_copy_ok flag. */ + list_for_each_entry(lte, stream_list, write_streams_list) + if (lte->resource_location == RESOURCE_IN_WIM) + lte->rspec->raw_copy_ok = 0; - /* Pointer to chunk table for stream currently being written. */ - ctx->cur_chunk_tab = NULL; + list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) { + if (lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->raw_copy_ok) + { + list_del(<e->write_streams_list); + } else if (can_raw_copy(lte, write_resource_flags, + out_ctype, out_chunk_size)) + { + lte->rspec->raw_copy_ok = 1; + list_move_tail(<e->write_streams_list, + raw_copy_resources); + } else { + num_bytes_to_compress += lte->size; + } + } - return 0; + return num_bytes_to_compress; } +/* Copy a raw compressed resource located in another WIM file to the WIM file + * being written. */ static int -receive_compressed_chunks(struct main_writer_thread_ctx *ctx) +write_raw_copy_resource(struct wim_resource_spec *in_rspec, + struct filedes *out_fd) { - struct message *msg; - struct wim_lookup_table_entry *cur_lte; + u64 cur_read_offset; + u64 end_read_offset; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; int ret; + struct filedes *in_fd; + struct wim_lookup_table_entry *lte; + u64 out_offset_in_wim; - wimlib_assert(!list_empty(&ctx->outstanding_streams)); - wimlib_assert(ctx->num_outstanding_messages != 0); - - cur_lte = container_of(ctx->outstanding_streams.next, - struct wim_lookup_table_entry, - being_compressed_list); - - /* Get the next message from the queue and process it. - * The message will contain 1 or more data chunks that have been - * compressed. */ - msg = shared_queue_get(ctx->compressed_res_queue); - msg->complete = true; - --ctx->num_outstanding_messages; - - /* Is this the next chunk in the current resource? If it's not (i.e., - * an earlier chunk in a same or different resource hasn't been - * compressed yet), do nothing, and keep this message around until all - * earlier chunks are received. - * - * Otherwise, write all the chunks we can. */ - while (cur_lte != NULL && - !list_empty(&cur_lte->msg_list) - && (msg = container_of(cur_lte->msg_list.next, - struct message, - list))->complete) - { - list_move(&msg->list, &ctx->available_msgs); - if (msg->begin_chunk == 0) { - /* First set of chunks. */ + DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", " + "uncompressed_size=%"PRIu64")", + in_rspec->size_in_wim, in_rspec->uncompressed_size); - /* Write pipable WIM stream header if needed. */ - if (ctx->write_resource_flags & - WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) - { - ret = write_pwm_stream_header(cur_lte, ctx->out_fd, - WIM_RESHDR_FLAG_COMPRESSED); - if (ret) - return ret; - } + /* Copy the raw data. */ + cur_read_offset = in_rspec->offset_in_wim; + end_read_offset = cur_read_offset + in_rspec->size_in_wim; - /* Save current offset. */ - ctx->res_start_offset = ctx->out_fd->offset; + out_offset_in_wim = out_fd->offset; - /* Begin building the chunk table, and leave space for - * it if needed. */ - ret = begin_wim_resource_chunk_tab(cur_lte, - ctx->out_fd, - ctx->out_chunk_size, - &ctx->cur_chunk_tab, - ctx->write_resource_flags); - if (ret) - return ret; - } + if (in_rspec->is_pipable) { + if (cur_read_offset < sizeof(struct pwm_stream_hdr)) + return WIMLIB_ERR_INVALID_PIPABLE_WIM; + cur_read_offset -= sizeof(struct pwm_stream_hdr); + out_offset_in_wim += sizeof(struct pwm_stream_hdr); + } + in_fd = &in_rspec->wim->in_fd; + while (cur_read_offset != end_read_offset) { - /* Write the compressed chunks from the message. */ - ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab, - ctx->write_resource_flags); + bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset); + + ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset); if (ret) return ret; - /* Was this the last chunk of the stream? If so, finish the - * stream by writing the chunk table. */ - if (list_empty(&cur_lte->msg_list) && - msg->begin_chunk + msg->num_chunks == ctx->cur_chunk_tab->num_chunks) - { - u64 res_csize; - - ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, - ctx->out_fd, - ctx->res_start_offset, - ctx->write_resource_flags); - if (ret) - return ret; - - list_del(&cur_lte->being_compressed_list); - - res_csize = ctx->out_fd->offset - ctx->res_start_offset; - - FREE(ctx->cur_chunk_tab); - ctx->cur_chunk_tab = NULL; - - /* Check for resources compressed to greater than or - * equal to their original size and write them - * uncompressed instead. (But never do this if writing - * to a pipe.) */ - if (res_csize >= cur_lte->size && - !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) - { - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - cur_lte->size, res_csize); - ret = seek_and_truncate(ctx->out_fd, ctx->res_start_offset); - if (ret) - return ret; - ret = write_wim_resource(cur_lte, - ctx->out_fd, - WIMLIB_COMPRESSION_TYPE_NONE, - 0, - &cur_lte->out_reshdr, - ctx->write_resource_flags, - ctx->comp_ctx); - if (ret) - return ret; - } else { - cur_lte->out_reshdr.size_in_wim = - res_csize; - - cur_lte->out_reshdr.uncompressed_size = - cur_lte->size; - - cur_lte->out_reshdr.offset_in_wim = - ctx->res_start_offset; - - cur_lte->out_reshdr.flags = - cur_lte->flags | - WIM_RESHDR_FLAG_COMPRESSED; + ret = full_write(out_fd, buf, bytes_to_read); + if (ret) + return ret; - DEBUG("Wrote compressed resource " - "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - cur_lte->out_reshdr.uncompressed_size, - cur_lte->out_reshdr.size_in_wim, - cur_lte->out_reshdr.offset_in_wim, - cur_lte->out_reshdr.flags); - } + cur_read_offset += bytes_to_read; + } - do_write_streams_progress(ctx->progress_data, - cur_lte, false); - - /* Since we just finished writing a stream, write any - * streams that have been added to the serial_streams - * list for direct writing by the main thread (e.g. - * resources that don't need to be compressed because - * the desired compression type is the same as the - * previous compression type). */ - if (!list_empty(&ctx->serial_streams)) { - ret = do_write_stream_list_serial(&ctx->serial_streams, - ctx->lookup_table, - ctx->out_fd, - ctx->out_ctype, - ctx->out_chunk_size, - ctx->comp_ctx, - ctx->write_resource_flags, - ctx->progress_data); - if (ret) - return ret; - } + list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) { + if (lte->will_be_in_output_wim) { + stream_set_out_reshdr_for_reuse(lte); + if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + lte->out_res_offset_in_wim = out_offset_in_wim; + else + lte->out_reshdr.offset_in_wim = out_offset_in_wim; - /* Advance to the next stream to write. */ - if (list_empty(&ctx->outstanding_streams)) { - cur_lte = NULL; - } else { - cur_lte = container_of(ctx->outstanding_streams.next, - struct wim_lookup_table_entry, - being_compressed_list); - } } } return 0; } -/* Called when the main thread has read a new chunk of data. */ +/* Copy a list of raw compressed resources located other WIM file(s) to the WIM + * file being written. */ static int -main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) +write_raw_copy_resources(struct list_head *raw_copy_resources, + struct filedes *out_fd) { - struct main_writer_thread_ctx *ctx = _ctx; + struct wim_lookup_table_entry *lte; int ret; - struct message *next_msg; - u64 next_chunk_in_msg; - - /* Update SHA1 message digest for the stream currently being read by the - * main thread. */ - sha1_update(&ctx->next_sha_ctx, chunk, chunk_size); - - /* We send chunks of data to the compressor chunks in batches which we - * refer to as "messages". @next_msg is the message that is currently - * being prepared to send off. If it is NULL, that indicates that we - * need to start a new message. */ - next_msg = ctx->next_msg; - if (!next_msg) { - /* We need to start a new message. First check to see if there - * is a message available in the list of available messages. If - * so, we can just take one. If not, all the messages (there is - * a fixed number of them, proportional to the number of - * threads) have been sent off to the compressor threads, so we - * receive messages from the compressor threads containing - * compressed chunks of data. - * - * We may need to receive multiple messages before one is - * actually available to use because messages received that are - * *not* for the very next set of chunks to compress must be - * buffered until it's time to write those chunks. */ - while (list_empty(&ctx->available_msgs)) { - ret = receive_compressed_chunks(ctx); - if (ret) - return ret; - } - next_msg = container_of(ctx->available_msgs.next, - struct message, list); - list_del(&next_msg->list); - next_msg->complete = false; - next_msg->begin_chunk = ctx->next_chunk; - next_msg->num_chunks = min(MAX_CHUNKS_PER_MSG, - ctx->next_num_chunks - ctx->next_chunk); - ctx->next_msg = next_msg; - } - - /* Fill in the next chunk to compress */ - next_chunk_in_msg = ctx->next_chunk - next_msg->begin_chunk; - - next_msg->uncompressed_chunk_sizes[next_chunk_in_msg] = chunk_size; - memcpy(next_msg->uncompressed_chunks[next_chunk_in_msg], - chunk, chunk_size); - ctx->next_chunk++; - if (++next_chunk_in_msg == next_msg->num_chunks) { - /* Send off an array of chunks to compress */ - list_add_tail(&next_msg->list, &ctx->next_lte->msg_list); - shared_queue_put(ctx->res_to_compress_queue, next_msg); - ++ctx->num_outstanding_messages; - ctx->next_msg = NULL; + list_for_each_entry(lte, raw_copy_resources, write_streams_list) { + ret = write_raw_copy_resource(lte->rspec, out_fd); + if (ret) + return ret; } return 0; } +/* Wait for and write all chunks pending in the compressor. */ static int -main_writer_thread_finish(void *_ctx) +finish_remaining_chunks(struct write_streams_ctx *ctx) { - struct main_writer_thread_ctx *ctx = _ctx; + const void *cdata; + unsigned csize; + unsigned usize; int ret; - while (ctx->num_outstanding_messages != 0) { - ret = receive_compressed_chunks(ctx); + + if (ctx->compressor == NULL) + return 0; + + if (ctx->chunk_buf_filled != 0) { + ret = submit_chunk_for_compression(ctx, ctx->chunk_buf, + ctx->chunk_buf_filled); if (ret) return ret; } - wimlib_assert(list_empty(&ctx->outstanding_streams)); - return do_write_stream_list_serial(&ctx->serial_streams, - ctx->lookup_table, - ctx->out_fd, - ctx->out_ctype, - ctx->out_chunk_size, - ctx->comp_ctx, - ctx->write_resource_flags, - ctx->progress_data); -} - -static int -submit_stream_for_compression(struct wim_lookup_table_entry *lte, - struct main_writer_thread_ctx *ctx) -{ - int ret; - /* Read the entire stream @lte, feeding its data chunks to the - * compressor threads. Also SHA1-sum the stream; this is required in - * the case that @lte is unhashed, and a nice additional verification - * when @lte is already hashed. */ - sha1_init(&ctx->next_sha_ctx); - ctx->next_chunk = 0; - ctx->next_num_chunks = DIV_ROUND_UP(lte->size, ctx->out_chunk_size); - ctx->next_lte = lte; - INIT_LIST_HEAD(<e->msg_list); - list_add_tail(<e->being_compressed_list, &ctx->outstanding_streams); - ret = read_stream_prefix(lte, lte->size, main_writer_thread_cb, ctx, 0); - if (ret) - return ret; - wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); - return finalize_and_check_sha1(&ctx->next_sha_ctx, lte); + while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) { + ret = write_chunk(ctx, cdata, csize, usize); + if (ret) + return ret; + } + return 0; } -static int -main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) +static void +remove_zero_length_streams(struct list_head *stream_list) { - struct main_writer_thread_ctx *ctx = _ctx; - int ret; + struct wim_lookup_table_entry *lte, *tmp; - if (lte->size < 1000 || - !must_compress_stream(lte, ctx->write_resource_flags, - ctx->out_ctype, ctx->out_chunk_size)) - { - /* Stream is too small or isn't being compressed. Process it by - * the main thread when we have a chance. We can't necessarily - * process it right here, as the main thread could be in the - * middle of writing a different stream. */ - list_add_tail(<e->write_streams_list, &ctx->serial_streams); - lte->deferred = 1; - ret = 0; - } else { - ret = submit_stream_for_compression(lte, ctx); + list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) { + wimlib_assert(lte->will_be_in_output_wim); + if (lte->size == 0) { + list_del(<e->write_streams_list); + lte->out_reshdr.offset_in_wim = 0; + lte->out_reshdr.size_in_wim = 0; + lte->out_reshdr.uncompressed_size = 0; + lte->out_reshdr.flags = filter_resource_flags(lte->flags); + } } - lte->no_progress = 1; - return ret; -} - -static long -get_default_num_threads(void) -{ -#ifdef __WIN32__ - return win32_get_number_of_processors(); -#else - return sysconf(_SC_NPROCESSORS_ONLN); -#endif } -/* Equivalent to write_stream_list_serial(), except this takes a @num_threads - * parameter and will perform compression using that many threads. Falls - * back to write_stream_list_serial() on certain errors, such as a failure to - * create the number of threads requested. - * - * High level description of the algorithm for writing compressed streams in - * parallel: We perform compression on chunks rather than on full files. The - * currently executing thread becomes the main thread and is entirely in charge - * of reading the data to compress (which may be in any location understood by - * the resource code--- such as in an external file being captured, or in - * another WIM file from which an image is being exported) and actually writing - * the compressed data to the output file. Additional threads are "compressor - * threads" and all execute the compressor_thread_proc, where they repeatedly - * retrieve buffers of data from the main thread, compress them, and hand them - * back to the main thread. - * - * Certain streams, such as streams that do not need to be compressed (e.g. - * input compression type same as output compression type) or streams of very - * small size are placed in a list (main_writer_thread_ctx.serial_list) and - * handled entirely by the main thread at an appropriate time. - * - * At any given point in time, multiple streams may be having chunks compressed - * concurrently. The stream that the main thread is currently *reading* may be - * later in the list that the stream that the main thread is currently - * *writing*. */ static int -write_stream_list_parallel(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, - int out_ctype, - u32 out_chunk_size, - struct wimlib_lzx_context **comp_ctx, - int write_resource_flags, - struct write_streams_progress_data *progress_data, - unsigned num_threads) +write_stream_list(struct list_head *stream_list, + struct filedes *out_fd, + int write_resource_flags, + int out_ctype, + u32 out_chunk_size, + unsigned num_threads, + struct wim_lookup_table *lookup_table, + struct filter_context *filter_ctx, + struct wimlib_lzx_context **comp_ctx, + wimlib_progress_func_t progress_func) { int ret; - struct shared_queue res_to_compress_queue; - struct shared_queue compressed_res_queue; - pthread_t *compressor_threads = NULL; - union wimlib_progress_info *progress = &progress_data->progress; - unsigned num_started_threads; - bool can_retry = true; - - if (num_threads == 0) { - long nthreads = get_default_num_threads(); - if (nthreads < 1 || nthreads > UINT_MAX) { - WARNING("Could not determine number of processors! Assuming 1"); - goto out_serial_quiet; - } else if (nthreads == 1) { - goto out_serial_quiet; - } else { - num_threads = nthreads; - } + struct write_streams_ctx ctx; + struct list_head raw_copy_resources; + + remove_zero_length_streams(stream_list); + + if (list_empty(stream_list)) { + DEBUG("No streams to write."); + return 0; } - DEBUG("Writing stream list of size %"PRIu64" " - "(parallel version, num_threads=%u)", - progress->write_streams.total_streams, num_threads); + memset(&ctx, 0, sizeof(ctx)); + + /* Pre-sorting the streams is required for compute_stream_list_stats(). + * Afterwards, read_stream_list() need not sort them again. */ + ret = sort_stream_list_by_sequential_order(stream_list, + offsetof(struct wim_lookup_table_entry, + write_streams_list)); + if (ret) + return ret; - progress->write_streams.num_threads = num_threads; + ctx.out_fd = out_fd; + ctx.lookup_table = lookup_table; + ctx.out_ctype = out_ctype; + ctx.out_chunk_size = out_chunk_size; + ctx.write_resource_flags = write_resource_flags; + ctx.filter_ctx = filter_ctx; - static const size_t MESSAGES_PER_THREAD = 2; - size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); + if (out_chunk_size <= STACK_MAX) { + ctx.chunk_buf = alloca(out_chunk_size); + } else { + ctx.chunk_buf = MALLOC(out_chunk_size); + if (ctx.chunk_buf == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_context; + } + } + ctx.chunk_buf_filled = 0; - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); + compute_stream_list_stats(stream_list, &ctx); - ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret) - goto out_serial; + ctx.progress_data.progress_func = progress_func; - ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret) - goto out_destroy_res_to_compress_queue; + ctx.num_bytes_to_compress = find_raw_copy_resources(stream_list, + write_resource_flags, + out_ctype, + out_chunk_size, + &raw_copy_resources); - struct compressor_thread_params *params; + DEBUG("Writing stream list " + "(offset = %"PRIu64", write_resource_flags=0x%08x, " + "out_ctype=%d, out_chunk_size=%u, num_threads=%u, " + "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")", + out_fd->offset, write_resource_flags, + out_ctype, out_chunk_size, num_threads, + ctx.progress_data.progress.write_streams.total_bytes, + ctx.num_bytes_to_compress); - params = CALLOC(num_threads, sizeof(params[0])); - if (params == NULL) { - ret = WIMLIB_ERR_NOMEM; - goto out_destroy_compressed_res_queue; + if (ctx.num_bytes_to_compress == 0) { + DEBUG("No compression needed; skipping to raw copy!"); + goto out_write_raw_copy_resources; } - for (unsigned i = 0; i < num_threads; i++) { - params[i].res_to_compress_queue = &res_to_compress_queue; - params[i].compressed_res_queue = &compressed_res_queue; - params[i].out_ctype = out_ctype; - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { - ret = wimlib_lzx_alloc_context(out_chunk_size, - NULL, ¶ms[i].comp_ctx); - if (ret) - goto out_free_params; + /* Unless uncompressed output was required, allocate a chunk_compressor + * to do compression. There are serial and parallel implementations of + * the chunk_compressor interface. We default to parallel using the + * specified number of threads, unless the upper bound on the number + * bytes needing to be compressed is less 2000000 (heuristic value). */ + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + + if (ctx.num_bytes_to_compress >= 2000000) { + ret = new_parallel_chunk_compressor(out_ctype, + out_chunk_size, + num_threads, 0, + &ctx.compressor); + if (ret) { + DEBUG("Couldn't create parallel chunk compressor " + "(status %d)", ret); + } } - } - compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); - if (compressor_threads == NULL) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_params; - } - - for (unsigned i = 0; i < num_threads; i++) { - DEBUG("pthread_create thread %u of %u", i + 1, num_threads); - ret = pthread_create(&compressor_threads[i], NULL, - compressor_thread_proc, ¶ms[i]); - if (ret) { - errno = ret; - ret = -1; - ERROR_WITH_ERRNO("Failed to create compressor " - "thread %u of %u", - i + 1, num_threads); - num_started_threads = i; - goto out_join; + if (ctx.compressor == NULL) { + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = wimlib_lzx_alloc_context(out_chunk_size, + NULL, + comp_ctx); + if (ret) + goto out_destroy_context; + } + ret = new_serial_chunk_compressor(out_ctype, out_chunk_size, + *comp_ctx, &ctx.compressor); + if (ret) + goto out_destroy_context; } } - num_started_threads = num_threads; - if (progress_data->progress_func) { - progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); + if (ctx.compressor) + ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads; + else + ctx.progress_data.progress.write_streams.num_threads = 1; + + DEBUG("Actually using %u threads", + ctx.progress_data.progress.write_streams.num_threads); + + INIT_LIST_HEAD(&ctx.pending_streams); + + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress); + if (ret) + goto out_destroy_context; } - struct main_writer_thread_ctx ctx; + /* Read the list of streams needing to be compressed, using the + * specified callbacks to execute processing of the data. */ - memset(&ctx, 0, sizeof(ctx)); + struct read_stream_list_callbacks cbs = { + .begin_stream = write_stream_begin_read, + .begin_stream_ctx = &ctx, + .consume_chunk = write_stream_process_chunk, + .consume_chunk_ctx = &ctx, + .end_stream = write_stream_end_read, + .end_stream_ctx = &ctx, + }; + + ret = read_stream_list(stream_list, + offsetof(struct wim_lookup_table_entry, write_streams_list), + &cbs, + STREAM_LIST_ALREADY_SORTED | + VERIFY_STREAM_HASHES | + COMPUTE_MISSING_STREAM_HASHES); - ctx.stream_list = stream_list; - ctx.lookup_table = lookup_table; - ctx.out_fd = out_fd; - ctx.out_ctype = out_ctype; - ctx.out_chunk_size = out_chunk_size; - ctx.comp_ctx = comp_ctx; - ctx.res_to_compress_queue = &res_to_compress_queue; - ctx.compressed_res_queue = &compressed_res_queue; - ctx.num_messages = queue_size; - ctx.write_resource_flags = write_resource_flags; - ctx.progress_data = progress_data; - ret = main_writer_thread_init_ctx(&ctx); if (ret) - goto out_join; + goto out_destroy_context; - can_retry = false; - ret = do_write_stream_list(stream_list, lookup_table, - main_thread_process_next_stream, - &ctx, progress_data); + ret = finish_remaining_chunks(&ctx); if (ret) - goto out_destroy_ctx; - - /* The main thread has finished reading all streams that are going to be - * compressed in parallel, and it now needs to wait for all remaining - * chunks to be compressed so that the remaining streams can actually be - * written to the output file. Furthermore, any remaining streams that - * had processing deferred to the main thread need to be handled. These - * tasks are done by the main_writer_thread_finish() function. */ - ret = main_writer_thread_finish(&ctx); -out_destroy_ctx: - main_writer_thread_destroy_ctx(&ctx); -out_join: - for (unsigned i = 0; i < num_started_threads; i++) - shared_queue_put(&res_to_compress_queue, NULL); - - for (unsigned i = 0; i < num_started_threads; i++) { - if (pthread_join(compressor_threads[i], NULL)) { - WARNING_WITH_ERRNO("Failed to join compressor " - "thread %u of %u", - i + 1, num_threads); + goto out_destroy_context; + + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS) { + struct wim_reshdr reshdr; + struct wim_lookup_table_entry *lte; + u64 offset_in_res; + + ret = end_write_resource(&ctx, &reshdr); + if (ret) + goto out_destroy_context; + + DEBUG("Ending packed resource: %lu %lu %lu.", + reshdr.offset_in_wim, + reshdr.size_in_wim, + reshdr.uncompressed_size); + + offset_in_res = 0; + list_for_each_entry(lte, &ctx.pending_streams, write_streams_list) { + lte->out_reshdr.size_in_wim = lte->size; + lte->out_reshdr.flags = filter_resource_flags(lte->flags); + lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS; + lte->out_reshdr.uncompressed_size = 0; + lte->out_reshdr.offset_in_wim = offset_in_res; + lte->out_res_offset_in_wim = reshdr.offset_in_wim; + lte->out_res_size_in_wim = reshdr.size_in_wim; + lte->out_res_uncompressed_size = reshdr.uncompressed_size; + offset_in_res += lte->size; } + wimlib_assert(offset_in_res == reshdr.uncompressed_size); } - FREE(compressor_threads); -out_free_params: - for (unsigned i = 0; i < num_threads; i++) - wimlib_lzx_free_context(params[i].comp_ctx); - FREE(params); -out_destroy_compressed_res_queue: - shared_queue_destroy(&compressed_res_queue); -out_destroy_res_to_compress_queue: - shared_queue_destroy(&res_to_compress_queue); - if (!can_retry || (ret >= 0 && ret != WIMLIB_ERR_NOMEM)) - return ret; -out_serial: - WARNING("Falling back to single-threaded compression"); -out_serial_quiet: - return write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - out_chunk_size, - comp_ctx, - write_resource_flags, - progress_data); +out_write_raw_copy_resources: + /* Copy any compressed resources for which the raw data can be reused + * without decompression. */ + ret = write_raw_copy_resources(&raw_copy_resources, ctx.out_fd); + +out_destroy_context: + if (out_chunk_size > STACK_MAX) + FREE(ctx.chunk_buf); + FREE(ctx.chunk_csizes); + if (ctx.compressor) + ctx.compressor->destroy(ctx.compressor); + DEBUG("Done (ret=%d)", ret); + return ret; } -#endif -/* Write a list of streams to a WIM (@out_fd) using the compression type - * @out_ctype, chunk size @out_chunk_size, and up to @num_threads compressor - * threads. */ static int -write_stream_list(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, int out_ctype, - u32 out_chunk_size, - struct wimlib_lzx_context **comp_ctx, - int write_flags, - unsigned num_threads, wimlib_progress_func_t progress_func) +write_wim_resource(struct wim_lookup_table_entry *lte, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + int write_resource_flags, + struct wimlib_lzx_context **comp_ctx) +{ + LIST_HEAD(stream_list); + list_add(<e->write_streams_list, &stream_list); + lte->will_be_in_output_wim = 1; + return write_stream_list(&stream_list, + out_fd, + write_resource_flags & ~WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS, + out_ctype, + out_chunk_size, + 1, + NULL, + NULL, + comp_ctx, + NULL); +} + +int +write_wim_resource_from_buffer(const void *buf, size_t buf_size, + int reshdr_flags, struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wim_reshdr *out_reshdr, + u8 *hash, + int write_resource_flags, + struct wimlib_lzx_context **comp_ctx) { int ret; - int write_resource_flags; - u64 total_bytes; - u64 total_compression_bytes; - unsigned total_parts; - WIMStruct *prev_wim_part; - size_t num_streams; struct wim_lookup_table_entry *lte; - struct write_streams_progress_data progress_data; - if (list_empty(stream_list)) { - DEBUG("No streams to write."); - return 0; - } + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ - write_resource_flags = write_flags_to_resource_flags(write_flags); + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; - DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", - out_fd->offset, write_resource_flags); + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = (void*)buf; + lte->size = buf_size; + lte->flags = reshdr_flags; - /* Sort the stream list into a good order for reading. */ - ret = sort_stream_list_by_sequential_order(stream_list, - offsetof(struct wim_lookup_table_entry, - write_streams_list)); + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + sha1_buffer(buf, buf_size, lte->hash); + lte->unhashed = 0; + } else { + lte->unhashed = 1; + } + + ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, + write_resource_flags, comp_ctx); if (ret) - return ret; + goto out_free_lte; - /* Calculate the total size of the streams to be written. Note: this - * will be the uncompressed size, as we may not know the compressed size - * yet, and also this will assume that every unhashed stream will be - * written (which will not necessarily be the case). */ - total_bytes = 0; - total_compression_bytes = 0; - num_streams = 0; - total_parts = 0; - prev_wim_part = NULL; - list_for_each_entry(lte, stream_list, write_streams_list) { - num_streams++; - total_bytes += lte->size; - if (must_compress_stream(lte, write_resource_flags, - out_ctype, out_chunk_size)) - total_compression_bytes += lte->size; - if (lte->resource_location == RESOURCE_IN_WIM) { - if (prev_wim_part != lte->rspec->wim) { - prev_wim_part = lte->rspec->wim; - total_parts++; - } - } - } + copy_reshdr(out_reshdr, <e->out_reshdr); - memset(&progress_data, 0, sizeof(progress_data)); - progress_data.progress_func = progress_func; - - progress_data.progress.write_streams.total_bytes = total_bytes; - progress_data.progress.write_streams.total_streams = num_streams; - progress_data.progress.write_streams.completed_bytes = 0; - progress_data.progress.write_streams.completed_streams = 0; - progress_data.progress.write_streams.num_threads = num_threads; - progress_data.progress.write_streams.compression_type = out_ctype; - progress_data.progress.write_streams.total_parts = total_parts; - progress_data.progress.write_streams.completed_parts = 0; - - progress_data.next_progress = 0; - progress_data.prev_wim_part = NULL; - -#ifdef ENABLE_MULTITHREADED_COMPRESSION - if (total_compression_bytes >= 2000000 && num_threads != 1) - ret = write_stream_list_parallel(stream_list, - lookup_table, - out_fd, - out_ctype, - out_chunk_size, - comp_ctx, - write_resource_flags, - &progress_data, - num_threads); - else -#endif - ret = write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - out_chunk_size, - comp_ctx, - write_resource_flags, - &progress_data); - if (ret == 0) - DEBUG("Successfully wrote stream list."); - else - DEBUG("Failed to write stream list (ret=%d).", ret); + if (hash) + copy_hash(hash, lte->hash); + ret = 0; +out_free_lte: + lte->resource_location = RESOURCE_NONEXISTENT; + free_lookup_table_entry(lte); return ret; } @@ -1827,7 +1398,7 @@ static int init_stream_size_table(struct stream_size_table *tab, size_t capacity) { tab->array = CALLOC(capacity, sizeof(tab->array[0])); - if (!tab->array) + if (tab->array == NULL) return WIMLIB_ERR_NOMEM; tab->num_entries = 0; tab->capacity = capacity; @@ -1871,40 +1442,42 @@ struct find_streams_ctx { }; static void -lte_reference_for_logical_write(struct wim_lookup_table_entry *lte, - struct find_streams_ctx *ctx, - unsigned nref) +reference_stream_for_write(struct wim_lookup_table_entry *lte, + struct list_head *stream_list, u32 nref) { - if (lte->out_refcnt == 0) { - stream_size_table_insert(lte, &ctx->stream_size_tab); - list_add_tail(<e->write_streams_list, &ctx->stream_list); + if (!lte->will_be_in_output_wim) { + lte->out_refcnt = 0; + list_add_tail(<e->write_streams_list, stream_list); + lte->will_be_in_output_wim = 1; } lte->out_refcnt += nref; } static int -do_lte_full_reference_for_logical_write(struct wim_lookup_table_entry *lte, - void *_ctx) +fully_reference_stream_for_write(struct wim_lookup_table_entry *lte, + void *_stream_list) { - struct find_streams_ctx *ctx = _ctx; - lte->out_refcnt = 0; - lte_reference_for_logical_write(lte, ctx, - (lte->refcnt ? lte->refcnt : 1)); + struct list_head *stream_list = _stream_list; + lte->will_be_in_output_wim = 0; + reference_stream_for_write(lte, stream_list, lte->refcnt); return 0; } static int -inode_find_streams_to_write(struct wim_inode *inode, - struct wim_lookup_table *table, - struct find_streams_ctx *ctx) +inode_find_streams_to_reference(const struct wim_inode *inode, + const struct wim_lookup_table *table, + struct list_head *stream_list) { struct wim_lookup_table_entry *lte; unsigned i; + wimlib_assert(inode->i_nlink > 0); + for (i = 0; i <= inode->i_num_ads; i++) { lte = inode_stream_lte(inode, i, table); if (lte) - lte_reference_for_logical_write(lte, ctx, inode->i_nlink); + reference_stream_for_write(lte, stream_list, + inode->i_nlink); else if (!is_zero_hash(inode_stream_hash(inode, i))) return WIMLIB_ERR_RESOURCE_NOT_FOUND; } @@ -1912,52 +1485,47 @@ inode_find_streams_to_write(struct wim_inode *inode, } static int -image_find_streams_to_write(WIMStruct *wim) +do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore) +{ + lte->will_be_in_output_wim = 0; + return 0; +} + +static int +image_find_streams_to_reference(WIMStruct *wim) { - struct find_streams_ctx *ctx; struct wim_image_metadata *imd; struct wim_inode *inode; struct wim_lookup_table_entry *lte; + struct list_head *stream_list; int ret; - ctx = wim->private; imd = wim_get_current_image_metadata(wim); image_for_each_unhashed_stream(lte, imd) - lte->out_refcnt = 0; + lte->will_be_in_output_wim = 0; - /* Go through this image's inodes to find any streams that have not been - * found yet. */ + stream_list = wim->private; image_for_each_inode(inode, imd) { - ret = inode_find_streams_to_write(inode, wim->lookup_table, ctx); + ret = inode_find_streams_to_reference(inode, + wim->lookup_table, + stream_list); if (ret) return ret; } return 0; } -/* - * Build a list of streams (via `struct wim_lookup_table_entry's) included in - * the "logical write" of the WIM, meaning all streams that are referenced at - * least once by dentries in the the image(s) being written. 'out_refcnt' on - * each stream being included in the logical write is set to the number of - * references from dentries in the image(s). Furthermore, 'unique_size' on each - * stream being included in the logical write is set to indicate whether that - * stream has a unique size relative to the streams being included in the - * logical write. Still furthermore, 'part_number' on each stream being - * included in the logical write is set to the part number given in the - * in-memory header of @p wim. - * - * This is considered a "logical write" because it does not take into account - * filtering out streams already present in the WIM (in the case of an in place - * overwrite) or present in other WIMs (in case of creating delta WIM). - */ static int -prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, - struct find_streams_ctx *ctx) +prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim, + int image, + int streams_ok, + struct list_head *stream_list_ret) { int ret; + INIT_LIST_HEAD(stream_list_ret); + if (streams_ok && (image == WIMLIB_ALL_IMAGES || (image == 1 && wim->hdr.image_count == 1))) { @@ -1968,18 +1536,21 @@ prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, unsigned i; for_lookup_table_entry(wim->lookup_table, - do_lte_full_reference_for_logical_write, ctx); + fully_reference_stream_for_write, + stream_list_ret); + for (i = 0; i < wim->hdr.image_count; i++) { imd = wim->image_metadata[i]; image_for_each_unhashed_stream(lte, imd) - do_lte_full_reference_for_logical_write(lte, ctx); + fully_reference_stream_for_write(lte, stream_list_ret); } } else { /* Slow case: Walk through the images being written and * determine the streams referenced. */ - for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); - wim->private = ctx; - ret = for_image(wim, image, image_find_streams_to_write); + for_lookup_table_entry(wim->lookup_table, + do_stream_set_not_in_output_wim, NULL); + wim->private = stream_list_ret; + ret = for_image(wim, image, image_find_streams_to_reference); if (ret) return ret; } @@ -1987,129 +1558,203 @@ prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, return 0; } +struct insert_other_if_hard_filtered_ctx { + struct stream_size_table *tab; + struct filter_context *filter_ctx; +}; + static int -process_filtered_stream(struct wim_lookup_table_entry *lte, void *_ctx) -{ - struct find_streams_ctx *ctx = _ctx; - u16 filtered = 0; - - /* Calculate and set lte->filtered. */ - if (lte->resource_location == RESOURCE_IN_WIM) { - if (lte->rspec->wim == ctx->wim && - (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE)) - filtered |= FILTERED_SAME_WIM; - if (lte->rspec->wim != ctx->wim && - (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)) - filtered |= FILTERED_EXTERNAL_WIM; - } - lte->filtered = filtered; - - /* Filtered streams get inserted into the stream size table too, unless - * they already were. This is because streams that are checksummed - * on-the-fly during the write should not be written if they are - * duplicates of filtered stream. */ - if (lte->filtered && lte->out_refcnt == 0) - stream_size_table_insert(lte, &ctx->stream_size_tab); +insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx) +{ + struct insert_other_if_hard_filtered_ctx *ctx = _ctx; + + if (!lte->will_be_in_output_wim && + stream_hard_filtered(lte, ctx->filter_ctx)) + stream_size_table_insert(lte, ctx->tab); return 0; } static int -mark_stream_not_filtered(struct wim_lookup_table_entry *lte, void *_ignore) +determine_stream_size_uniquity(struct list_head *stream_list, + struct wim_lookup_table *lt, + struct filter_context *filter_ctx) { - lte->filtered = 0; + int ret; + struct stream_size_table tab; + struct wim_lookup_table_entry *lte; + + ret = init_stream_size_table(&tab, lt->capacity); + if (ret) + return ret; + + if (may_hard_filter_streams(filter_ctx)) { + struct insert_other_if_hard_filtered_ctx ctx = { + .tab = &tab, + .filter_ctx = filter_ctx, + }; + for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx); + } + + list_for_each_entry(lte, stream_list, write_streams_list) + stream_size_table_insert(lte, &tab); + + destroy_stream_size_table(&tab); return 0; } -/* Given the list of streams to include in a logical write of a WIM, handle - * filtering out streams already present in the WIM or already present in - * external WIMs, depending on the write flags provided. */ static void -handle_stream_filtering(struct find_streams_ctx *ctx) +filter_stream_list_for_write(struct list_head *stream_list, + struct filter_context *filter_ctx) { struct wim_lookup_table_entry *lte, *tmp; - if (!(ctx->write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE | - WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS))) + list_for_each_entry_safe(lte, tmp, + stream_list, write_streams_list) { - for_lookup_table_entry(ctx->wim->lookup_table, - mark_stream_not_filtered, ctx); - return; - } + int status = stream_filtered(lte, filter_ctx); - for_lookup_table_entry(ctx->wim->lookup_table, - process_filtered_stream, ctx); - - /* Streams in logical write list that were filtered can be removed. */ - list_for_each_entry_safe(lte, tmp, &ctx->stream_list, - write_streams_list) - if (lte->filtered) + if (status == 0) { + /* Not filtered. */ + continue; + } else { + if (status > 0) { + /* Soft filtered. */ + } else { + /* Hard filtered. */ + lte->will_be_in_output_wim = 0; + list_del(<e->lookup_table_list); + } list_del(<e->write_streams_list); + } + } } -/* Prepares list of streams to write for the specified WIM image(s). This wraps - * around prepare_logical_stream_list() to handle filtering out streams already - * present in the WIM or already present in external WIMs, depending on the - * write flags provided. +/* + * prepare_stream_list_for_write() - + * + * Prepare the list of streams to write for writing a WIM containing the + * specified image(s) with the specified write flags. + * + * @wim + * The WIMStruct on whose behalf the write is occurring. + * + * @image + * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES. + * + * @write_flags + * WIMLIB_WRITE_FLAG_* flags for the write operation: * - * Note: some additional data is stored in each `struct wim_lookup_table_entry': + * STREAMS_OK: For writes of all images, assume that all streams in the + * lookup table of @wim and the per-image lists of unhashed streams should + * be taken as-is, and image metadata should not be searched for + * references. This does not exclude filtering with OVERWRITE and + * SKIP_EXTERNAL_WIMS, below. * - * - 'out_refcnt' is set to the number of references found for the logical write. - * This will be nonzero on all streams in the list returned by this function, - * but will also be nonzero on streams not in the list that were included in - * the logical write list, but filtered out from the returned list. - * - 'filtered' is set to nonzero if the stream was filtered. Filtered streams - * are not included in the list of streams returned by this function. - * - 'unique_size' is set if the stream has a unique size among all streams in - * the logical write plus any filtered streams in the entire WIM that could - * potentially turn out to have the same checksum as a yet-to-be-checksummed - * stream being written. + * OVERWRITE: Streams already present in @wim shall not be returned in + * @stream_list_ret. + * + * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not + * @wim, shall be be returned in neither @stream_list_ret nor + * @lookup_table_list_ret. + * + * @stream_list_ret + * List of streams, linked by write_streams_list, that need to be written + * will be returned here. + * + * Note that this function assumes that unhashed streams will be written; + * it does not take into account that they may become duplicates when + * actually hashed. + * + * @lookup_table_list_ret + * List of streams, linked by lookup_table_list, that need to be included + * in the WIM's lookup table will be returned here. This will be a + * superset of the streams in @stream_list_ret. + * + * This list will be a proper superset of @stream_list_ret if and only if + * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of + * the streams that would otherwise need to be written were already located + * in the WIM file. + * + * All streams in this list will have @out_refcnt set to the number of + * references to the stream in the output WIM. If + * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt + * may be as low as 0. + * + * @filter_ctx_ret + * A context for queries of stream filter status with stream_filtered() is + * returned in this location. + * + * In addition, @will_be_in_output_wim will be set to 1 in all stream entries + * inserted into @lookup_table_list_ret and to 0 in all stream entries in the + * lookup table of @wim not inserted into @lookup_table_list_ret. + * + * Still furthermore, @unique_size will be set to 1 on all stream entries in + * @stream_list_ret that have unique size among all stream entries in + * @stream_list_ret and among all stream entries in the lookup table of @wim + * that are ineligible for being written due to filtering. + * + * Returns 0 on success; nonzero on read error, memory allocation error, or + * otherwise. */ static int -prepare_stream_list(WIMStruct *wim, int image, int write_flags, - struct list_head *stream_list) +prepare_stream_list_for_write(WIMStruct *wim, int image, + int write_flags, + struct list_head *stream_list_ret, + struct list_head *lookup_table_list_ret, + struct filter_context *filter_ctx_ret) { int ret; - bool streams_ok; - struct find_streams_ctx ctx; + struct wim_lookup_table_entry *lte; + + filter_ctx_ret->write_flags = write_flags; + filter_ctx_ret->wim = wim; - INIT_LIST_HEAD(&ctx.stream_list); - ret = init_stream_size_table(&ctx.stream_size_tab, - wim->lookup_table->capacity); + ret = prepare_unfiltered_list_of_streams_in_output_wim( + wim, + image, + write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK, + stream_list_ret); if (ret) return ret; - ctx.write_flags = write_flags; - ctx.wim = wim; - streams_ok = ((write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) != 0); + INIT_LIST_HEAD(lookup_table_list_ret); + list_for_each_entry(lte, stream_list_ret, write_streams_list) + list_add_tail(<e->lookup_table_list, lookup_table_list_ret); - ret = prepare_logical_stream_list(wim, image, streams_ok, &ctx); + ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table, + filter_ctx_ret); if (ret) - goto out_destroy_table; + return ret; - handle_stream_filtering(&ctx); - list_transfer(&ctx.stream_list, stream_list); - ret = 0; -out_destroy_table: - destroy_stream_size_table(&ctx.stream_size_tab); - return ret; + if (may_filter_streams(filter_ctx_ret)) + filter_stream_list_for_write(stream_list_ret, filter_ctx_ret); + + return 0; } static int write_wim_streams(WIMStruct *wim, int image, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func, - struct list_head *stream_list_override) + struct list_head *stream_list_override, + struct list_head *lookup_table_list_ret) { int ret; struct list_head _stream_list; struct list_head *stream_list; struct wim_lookup_table_entry *lte; + struct filter_context _filter_ctx; + struct filter_context *filter_ctx; if (stream_list_override == NULL) { /* Normal case: prepare stream list from image(s) being written. */ stream_list = &_stream_list; - ret = prepare_stream_list(wim, image, write_flags, stream_list); + filter_ctx = &_filter_ctx; + ret = prepare_stream_list_for_write(wim, image, write_flags, + stream_list, + lookup_table_list_ret, + filter_ctx); if (ret) return ret; } else { @@ -2117,18 +1762,25 @@ write_wim_streams(WIMStruct *wim, int image, int write_flags, * use stream list already explicitly provided. Use existing * reference counts. */ stream_list = stream_list_override; - list_for_each_entry(lte, stream_list, write_streams_list) - lte->out_refcnt = (lte->refcnt ? lte->refcnt : 1); + filter_ctx = NULL; + INIT_LIST_HEAD(lookup_table_list_ret); + list_for_each_entry(lte, stream_list, write_streams_list) { + lte->out_refcnt = lte->refcnt; + lte->will_be_in_output_wim = 1; + lte->unique_size = 0; + list_add_tail(<e->lookup_table_list, lookup_table_list_ret); + } } return write_stream_list(stream_list, - wim->lookup_table, &wim->out_fd, + write_flags_to_resource_flags(write_flags), wim->out_compression_type, wim->out_chunk_size, - &wim->lzx_context, - write_flags, num_threads, + wim->lookup_table, + filter_ctx, + &wim->lzx_context, progress_func); } @@ -2148,6 +1800,8 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, write_resource_flags = write_flags_to_resource_flags(write_flags); + write_resource_flags &= ~WIMLIB_WRITE_RESOURCE_FLAG_PACK_STREAMS; + DEBUG("Writing metadata resources (offset=%"PRIu64")", wim->out_fd.offset); @@ -2177,8 +1831,7 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { DEBUG("Image %u was not modified; re-using existing " "metadata resource.", i); - wim_res_spec_to_hdr(imd->metadata_lte->rspec, - &imd->metadata_lte->out_reshdr); + stream_set_out_reshdr_for_reuse(imd->metadata_lte); ret = 0; } else { DEBUG("Image %u was not modified; copying existing " @@ -2187,7 +1840,6 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, &wim->out_fd, wim->out_compression_type, wim->out_chunk_size, - &imd->metadata_lte->out_reshdr, write_resource_flags, &wim->lzx_context); } @@ -2229,6 +1881,90 @@ close_wim_writable(WIMStruct *wim, int write_flags) return ret; } +static int +cmp_streams_by_out_rspec(const void *p1, const void *p2) +{ + const struct wim_lookup_table_entry *lte1, *lte2; + + lte1 = *(const struct wim_lookup_table_entry**)p1; + lte2 = *(const struct wim_lookup_table_entry**)p2; + + if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { + if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { + if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim) + return cmp_u64(lte1->out_res_offset_in_wim, + lte2->out_res_offset_in_wim); + } else { + return 1; + } + } else { + if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + return -1; + } + return cmp_u64(lte1->out_reshdr.offset_in_wim, + lte2->out_reshdr.offset_in_wim); +} + +static int +write_wim_lookup_table(WIMStruct *wim, int image, int write_flags, + struct wim_reshdr *out_reshdr, + struct list_head *lookup_table_list) +{ + int ret; + + /* Set output resource metadata for streams already present in WIM. */ + if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { + struct wim_lookup_table_entry *lte; + list_for_each_entry(lte, lookup_table_list, lookup_table_list) + { + if (lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->wim == wim) + { + stream_set_out_reshdr_for_reuse(lte); + } + } + } + + ret = sort_stream_list(lookup_table_list, + offsetof(struct wim_lookup_table_entry, lookup_table_list), + cmp_streams_by_out_rspec); + if (ret) + return ret; + + /* Add entries for metadata resources. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) { + int start_image; + int end_image; + + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; + } else { + start_image = image; + end_image = image; + } + + /* Push metadata resource lookup table entries onto the front of + * the list in reverse order, so that they're written in order. + */ + for (int i = end_image; i >= start_image; i--) { + struct wim_lookup_table_entry *metadata_lte; + + metadata_lte = wim->image_metadata[i - 1]->metadata_lte; + wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA); + metadata_lte->out_refcnt = 1; + list_add(&metadata_lte->lookup_table_list, lookup_table_list); + } + } + + return write_wim_lookup_table_from_stream_list(lookup_table_list, + &wim->out_fd, + wim->hdr.part_number, + out_reshdr, + write_flags_to_resource_flags(write_flags), + &wim->lzx_context); +} + /* * finish_write(): * @@ -2278,7 +2014,7 @@ close_wim_writable(WIMStruct *wim, int write_flags) static int finish_write(WIMStruct *wim, int image, int write_flags, wimlib_progress_func_t progress_func, - struct list_head *stream_list_override) + struct list_head *lookup_table_list) { int ret; off_t hdr_offset; @@ -2300,8 +2036,8 @@ finish_write(WIMStruct *wim, int image, int write_flags, zero_reshdr(&wim->hdr.boot_metadata_reshdr); } else { copy_reshdr(&wim->hdr.boot_metadata_reshdr, - &wim->image_metadata[wim->hdr.boot_idx- 1 - ]->metadata_lte->out_reshdr); + &wim->image_metadata[ + wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr); } /* Write lookup table. (Save old position first.) */ @@ -2310,7 +2046,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { ret = write_wim_lookup_table(wim, image, write_flags, &wim->hdr.lookup_table_reshdr, - stream_list_override); + lookup_table_list); if (ret) return ret; } @@ -2497,7 +2233,8 @@ lock_wim(WIMStruct *wim, int fd) static int write_pipable_wim(WIMStruct *wim, int image, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func, - struct list_head *stream_list_override) + struct list_head *stream_list_override, + struct list_head *lookup_table_list_ret) { int ret; struct wim_reshdr xml_reshdr; @@ -2538,7 +2275,8 @@ write_pipable_wim(WIMStruct *wim, int image, int write_flags, /* Write streams needed for the image(s) being included in the output * WIM, or streams needed for the split WIM part. */ return write_wim_streams(wim, image, write_flags, num_threads, - progress_func, stream_list_override); + progress_func, stream_list_override, + lookup_table_list_ret); /* The lookup table, XML data, and header at end are handled by * finish_write(). */ @@ -2560,7 +2298,7 @@ write_wim_part(WIMStruct *wim, { int ret; struct wim_header hdr_save; - struct list_head lt_stream_list_override; + struct list_head lookup_table_list; if (total_parts == 1) DEBUG("Writing standalone WIM."); @@ -2575,26 +2313,46 @@ write_wim_part(WIMStruct *wim, else DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd); DEBUG("Write flags: 0x%08x", write_flags); + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) DEBUG("\tCHECK_INTEGRITY"); + if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) DEBUG("\tREBUILD"); + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) DEBUG("\tRECOMPRESS"); + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) DEBUG("\tFSYNC"); + if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE) DEBUG("\tFSYNC"); + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) DEBUG("\tIGNORE_READONLY_FLAG"); + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) DEBUG("\tPIPABLE"); + + if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE) + DEBUG("\tNOT_PIPABLE"); + + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) + DEBUG("\tPACK_STREAMS"); + + if (write_flags & WIMLIB_WRITE_FLAG_NO_PACK_STREAMS) + DEBUG("\tNO_PACK_STREAMS"); + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) DEBUG("\tFILE_DESCRIPTOR"); + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) DEBUG("\tNO_METADATA"); + if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES) DEBUG("\tUSE_EXISTING_TOTALBYTES"); + if (num_threads == 0) DEBUG("Number of threads: autodetect"); else @@ -2634,19 +2392,42 @@ write_wim_part(WIMStruct *wim, WIMLIB_WRITE_FLAG_NOT_PIPABLE)) return WIMLIB_ERR_INVALID_PARAM; + if ((write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS | + WIMLIB_WRITE_FLAG_NO_PACK_STREAMS)) + == (WIMLIB_WRITE_FLAG_PACK_STREAMS | + WIMLIB_WRITE_FLAG_NO_PACK_STREAMS)) + return WIMLIB_ERR_INVALID_PARAM; + /* Save previous header, then start initializing the new one. */ memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); - /* Set default integrity and pipable flags. */ + /* Set default integrity, pipable, and packed stream flags. */ if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | WIMLIB_WRITE_FLAG_NOT_PIPABLE))) - if (wim_is_pipable(wim)) + if (wim_is_pipable(wim)) { + DEBUG("WIM is pipable; default to PIPABLE."); write_flags |= WIMLIB_WRITE_FLAG_PIPABLE; + } if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) - if (wim_has_integrity_table(wim)) + if (wim_has_integrity_table(wim)) { + DEBUG("Integrity table present; default to CHECK_INTEGRITY."); write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + } + + if (!(write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS | + WIMLIB_WRITE_FLAG_NO_PACK_STREAMS))) + if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS) { + DEBUG("WIM version 3584; default to PACK_STREAMS."); + write_flags |= WIMLIB_WRITE_FLAG_PACK_STREAMS; + } + + if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_PACK_STREAMS)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_PACK_STREAMS)) + return WIMLIB_ERR_INVALID_PARAM; /* Set appropriate magic number. */ if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) @@ -2654,6 +2435,12 @@ write_wim_part(WIMStruct *wim, else wim->hdr.magic = WIM_MAGIC; + /* Set appropriate version number. */ + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) + wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS; + else + wim->hdr.wim_version = WIM_VERSION_DEFAULT; + /* Clear header flags that will be set automatically. */ wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY | WIM_HDR_FLAG_RESOURCE_ONLY | @@ -2740,22 +2527,12 @@ write_wim_part(WIMStruct *wim, if (ret) goto out_restore_hdr; - if (stream_list_override) { - struct wim_lookup_table_entry *lte; - INIT_LIST_HEAD(<_stream_list_override); - list_for_each_entry(lte, stream_list_override, - write_streams_list) - { - list_add_tail(<e->lookup_table_list, - <_stream_list_override); - } - } - /* Write metadata resources and streams. */ if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) { /* Default case: create a normal (non-pipable) WIM. */ ret = write_wim_streams(wim, image, write_flags, num_threads, - progress_func, stream_list_override); + progress_func, stream_list_override, + &lookup_table_list); if (ret) goto out_restore_hdr; @@ -2766,18 +2543,17 @@ write_wim_part(WIMStruct *wim, } else { /* Non-default case: create pipable WIM. */ ret = write_pipable_wim(wim, image, write_flags, num_threads, - progress_func, stream_list_override); + progress_func, stream_list_override, + &lookup_table_list); if (ret) goto out_restore_hdr; write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END; } - if (stream_list_override) - stream_list_override = <_stream_list_override; /* Write lookup table, XML data, and (optional) integrity table. */ ret = finish_write(wim, image, write_flags, progress_func, - stream_list_override); + &lookup_table_list); out_restore_hdr: memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); (void)close_wim_writable(wim, write_flags); @@ -2932,10 +2708,12 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, wimlib_progress_func_t progress_func) { int ret; - struct list_head stream_list; off_t old_wim_end; u64 old_lookup_table_end, old_xml_begin, old_xml_end; struct wim_header hdr_save; + struct list_head stream_list; + struct list_head lookup_table_list; + struct filter_context filter_ctx; DEBUG("Overwriting `%"TS"' in-place", wim->filename); @@ -2945,6 +2723,12 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, if (wim_has_integrity_table(wim)) write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + /* Set default packed flag. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_PACK_STREAMS | + WIMLIB_WRITE_FLAG_NO_PACK_STREAMS))) + if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS) + write_flags |= WIMLIB_WRITE_FLAG_PACK_STREAMS; + /* Set additional flags for overwrite. */ write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE | WIMLIB_WRITE_FLAG_STREAMS_OK; @@ -2998,8 +2782,9 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, if (ret) return ret; - ret = prepare_stream_list(wim, WIMLIB_ALL_IMAGES, write_flags, - &stream_list); + ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags, + &stream_list, &lookup_table_list, + &filter_ctx); if (ret) return ret; @@ -3029,13 +2814,14 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, } ret = write_stream_list(&stream_list, - wim->lookup_table, &wim->out_fd, + write_flags_to_resource_flags(write_flags), wim->compression_type, wim->chunk_size, - &wim->lzx_context, - write_flags, num_threads, + wim->lookup_table, + &filter_ctx, + &wim->lzx_context, progress_func); if (ret) goto out_truncate; @@ -3047,7 +2833,7 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, - progress_func, NULL); + progress_func, &lookup_table_list); if (ret) goto out_truncate; @@ -3126,6 +2912,28 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, return 0; } +static bool +can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) +{ + if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) + return false; + + if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + return false; + + if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + return false; + + if (wim->hdr.wim_version != WIM_VERSION_PACKED_STREAMS) { + if (wim->compression_type != wim->out_compression_type) + return false; + if (wim->chunk_size != wim->out_chunk_size) + return false; + } + + return true; +} + /* API function documented in wimlib.h */ WIMLIBAPI int wimlib_overwrite(WIMStruct *wim, int write_flags, @@ -3151,13 +2959,7 @@ wimlib_overwrite(WIMStruct *wim, int write_flags, if (ret) return ret; - if ((!wim->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) - && !(write_flags & (WIMLIB_WRITE_FLAG_REBUILD | - WIMLIB_WRITE_FLAG_PIPABLE)) - && !(wim_is_pipable(wim)) - && wim->compression_type == wim->out_compression_type - && wim->chunk_size == wim->out_chunk_size) - { + if (can_overwrite_wim_inplace(wim, write_flags)) { ret = overwrite_wim_inplace(wim, write_flags, num_threads, progress_func); if (ret != WIMLIB_ERR_RESOURCE_ORDER) diff --git a/src/xml.c b/src/xml.c index 4a8bda9a..920eeeb5 100644 --- a/src/xml.c +++ b/src/xml.c @@ -1524,7 +1524,7 @@ write_wim_xml_data(WIMStruct *wim, int image, u64 total_bytes, write_resource_flags, &wim->lzx_context); FREE(xml_data); - DEBUG("ret=%d"); + DEBUG("ret=%d", ret); return ret; } -- 2.43.0