X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=8554d1ab573fc23af05533c2f9f1cde750e0914e;hp=80022182896cfd36caa9f051bf6715d8c00fdd03;hb=49a63aa13cdeb4c1348697ccd92207a1a65ec7b0;hpb=039a0b038851fc9aecd595d182c8eee02093247d diff --git a/src/write.c b/src/write.c index 80022182..8554d1ab 100644 --- a/src/write.c +++ b/src/write.c @@ -68,6 +68,53 @@ # 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. */ +static bool +can_raw_copy(const struct wim_lookup_table_entry *lte, + int write_resource_flags, int out_ctype, u32 out_chunk_size) +{ + return (out_ctype == wim_resource_compression_type(lte) + && out_chunk_size == wim_resource_chunk_size(lte) + && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE); +} + + +/* 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); + default: + wimlib_assert(0); + return 0; + } +} + /* 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.) */ @@ -92,6 +139,7 @@ struct chunk_table { 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) { @@ -103,7 +151,7 @@ begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, int ret; size = wim_resource_size(lte); - num_chunks = wim_resource_chunks(lte); + 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); @@ -158,41 +206,6 @@ chunk_tab_record_chunk(struct chunk_table *chunk_tab, unsigned out_chunk_size) } } -/* - * compress_func_t- Pointer to a function to compresses a chunk - * of a WIM resource. This may be either - * wimlib_xpress_compress() (xpress-compress.c) or - * wimlib_lzx_compress() (lzx-compress.c). - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk, in bytes. - * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes. - * - * Returns the size of the compressed data written to @out in bytes, or 0 if the - * data could not be compressed to (@chunk_size - 1) bytes or fewer. - * - * As a special requirement, the compression code is optimized for the WIM - * format and therefore requires (@chunk_size <= 32768). - * - * As another special requirement, the compression code will read up to 8 bytes - * off the end of the @chunk array for performance reasons. The values of these - * bytes will not affect the output of the compression, but the calling code - * must make sure that the buffer holding the uncompressed chunk is actually at - * least (@chunk_size + 8) bytes, or at least that these extra bytes are in - * mapped memory that will not cause a memory access violation if accessed. - */ -typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, - void *out); - -static compress_func_t -get_compress_func(int out_ctype) -{ - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return wimlib_lzx_compress; - else - return wimlib_xpress_compress; -} - /* Finishes a WIM chunk table and writes it to the output file at the correct * offset. */ static int @@ -284,7 +297,9 @@ finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) } struct write_resource_ctx { - compress_func_t compress; + int out_ctype; + u32 out_chunk_size; + struct wimlib_lzx_context *comp_ctx; struct chunk_table *chunk_tab; struct filedes *out_fd; SHA_CTX sha_ctx; @@ -299,21 +314,32 @@ write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) const void *out_chunk; unsigned out_chunk_size; int ret; + void *compressed_chunk = NULL; + unsigned compressed_size; + bool compressed_chunk_malloced = false; + size_t stack_max = 32768; if (ctx->doing_sha) sha1_update(&ctx->sha_ctx, chunk, chunk_size); out_chunk = chunk; out_chunk_size = chunk_size; - if (ctx->compress) { - void *compressed_chunk; - unsigned compressed_size; + if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { /* Compress the chunk. */ - compressed_chunk = alloca(chunk_size); - compressed_size = (*ctx->compress)(chunk, chunk_size, - compressed_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; + } + 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) { @@ -344,11 +370,15 @@ write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) ret = full_write(ctx->out_fd, out_chunk, out_chunk_size); if (ret) goto error; - return 0; + +out_free_memory: + if (compressed_chunk_malloced) + FREE(compressed_chunk); + return ret; error: ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return ret; + goto out_free_memory; } /* @@ -371,12 +401,16 @@ error: * On success, this is filled in with the offset, flags, compressed size, * and uncompressed size of the resource in the output WIM. * - * @write_resource_flags: + * @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 @@ -386,11 +420,14 @@ error: int write_wim_resource(struct wim_lookup_table_entry *lte, struct filedes *out_fd, int out_ctype, + u32 out_chunk_size, struct resource_entry *out_res_entry, - int resource_flags) + int resource_flags, + struct wimlib_lzx_context **comp_ctx) { struct write_resource_ctx write_ctx; off_t res_start_offset; + u32 in_chunk_size; u64 read_size; int ret; @@ -406,11 +443,7 @@ write_wim_resource(struct wim_lookup_table_entry *lte, * desired other than no compression, 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 (lte->resource_location == RESOURCE_IN_WIM && - out_ctype == wim_resource_compression_type(lte) && - out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && - !(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS)) - { + if (can_raw_copy(lte, resource_flags, out_ctype, out_chunk_size)) { /* Normally 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 @@ -422,8 +455,8 @@ write_wim_resource(struct wim_lookup_table_entry *lte, resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL; else resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS; - write_ctx.doing_sha = false; read_size = lte->resource_entry.size; + write_ctx.doing_sha = false; } else { write_ctx.doing_sha = true; sha1_init(&write_ctx.sha_ctx); @@ -434,13 +467,23 @@ write_wim_resource(struct wim_lookup_table_entry *lte, * table and set the function to use for chunk compression. Exceptions: * no compression function is needed if doing a raw copy; also, no chunk * table is needed if doing a *full* (not per-chunk) raw copy. */ - write_ctx.compress = NULL; + 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) { - if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) - write_ctx.compress = get_compress_func(out_ctype); + wimlib_assert(out_chunk_size > 0); + if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) { + write_ctx.out_ctype = out_ctype; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = wimlib_lzx_alloc_context(NULL, comp_ctx); + if (ret) + goto out; + } + write_ctx.comp_ctx = *comp_ctx; + } if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)) { ret = begin_wim_resource_chunk_tab(lte, out_fd, + out_chunk_size, &write_ctx.chunk_tab, resource_flags); if (ret) @@ -465,8 +508,13 @@ write_wim_resource(struct wim_lookup_table_entry *lte, write_ctx.out_fd = out_fd; write_ctx.resource_flags = resource_flags; try_write_again: + if (write_ctx.out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + in_chunk_size = wim_resource_chunk_size(lte); + else + in_chunk_size = out_chunk_size; ret = read_resource_prefix(lte, read_size, - write_resource_cb, &write_ctx, resource_flags); + write_resource_cb, + in_chunk_size, &write_ctx, resource_flags); if (ret) goto out_free_chunk_tab; @@ -514,7 +562,7 @@ try_write_again: goto out_free_chunk_tab; out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; FREE(write_ctx.chunk_tab); - write_ctx.compress = NULL; + write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; write_ctx.chunk_tab = NULL; write_ctx.doing_sha = false; goto try_write_again; @@ -549,8 +597,10 @@ 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 resource_entry *out_res_entry, - u8 *hash_ret, int write_resource_flags) + u8 *hash_ret, int write_resource_flags, + struct wimlib_lzx_context **comp_ctx) { /* Set up a temporary lookup table entry to provide to * write_wim_resource(). */ @@ -561,6 +611,7 @@ write_wim_resource_from_buffer(const void *buf, size_t buf_size, lte.attached_buffer = (void*)buf; lte.resource_entry.original_size = buf_size; lte.resource_entry.flags = reshdr_flags; + lte.compression_type = WIMLIB_COMPRESSION_TYPE_NONE; if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { sha1_buffer(buf, buf_size, lte.hash); @@ -569,8 +620,8 @@ write_wim_resource_from_buffer(const void *buf, size_t buf_size, lte.unhashed = 1; } - ret = write_wim_resource(<e, out_fd, out_ctype, out_res_entry, - write_resource_flags); + ret = write_wim_resource(<e, out_fd, out_ctype, out_chunk_size, + out_res_entry, write_resource_flags, comp_ctx); if (ret) return ret; if (hash_ret) @@ -671,13 +722,15 @@ shared_queue_get(struct shared_queue *q) struct compressor_thread_params { struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; - compress_func_t compress; + int out_ctype; + struct wimlib_lzx_context *comp_ctx; }; #define MAX_CHUNKS_PER_MSG 2 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]; @@ -689,12 +742,18 @@ struct message { }; static void -compress_chunks(struct message *msg, compress_func_t compress) +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 = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_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) { @@ -722,12 +781,11 @@ compressor_thread_proc(void *arg) 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; - compress_func_t compress = params->compress; struct message *msg; DEBUG("Compressor thread ready"); while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { - compress_chunks(msg, compress); + compress_chunks(msg, params->out_ctype, params->comp_ctx); shared_queue_put(compressed_res_queue, msg); } DEBUG("Compressor thread terminating"); @@ -791,6 +849,8 @@ do_write_streams_progress(struct write_streams_progress_data *progress_data, 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; }; @@ -799,8 +859,11 @@ 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, <e->output_resource_entry, - ctx->write_resource_flags); + ctx->out_ctype, + ctx->out_chunk_size, + <e->output_resource_entry, + ctx->write_resource_flags, + ctx->comp_ctx); } @@ -898,13 +961,17 @@ 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, @@ -930,6 +997,8 @@ 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) { @@ -945,6 +1014,8 @@ write_stream_list_serial(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + out_chunk_size, + comp_ctx, write_resource_flags, progress_data); } @@ -994,6 +1065,8 @@ struct main_writer_thread_ctx { 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; @@ -1016,11 +1089,12 @@ struct main_writer_thread_ctx { }; static int -init_message(struct message *msg) +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(WIM_CHUNK_SIZE); - msg->uncompressed_chunks[i] = MALLOC(WIM_CHUNK_SIZE); + 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; @@ -1048,7 +1122,7 @@ free_messages(struct message *msgs, size_t num_messages) } static struct message * -allocate_messages(size_t num_messages) +allocate_messages(size_t num_messages, u32 out_chunk_size) { struct message *msgs; @@ -1056,7 +1130,7 @@ allocate_messages(size_t num_messages) if (!msgs) return NULL; for (size_t i = 0; i < num_messages; i++) { - if (init_message(&msgs[i])) { + if (init_message(&msgs[i], out_chunk_size)) { free_messages(msgs, num_messages); return NULL; } @@ -1078,7 +1152,7 @@ main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) { /* Pre-allocate all the buffers that will be needed to do the chunk * compression. */ - ctx->msgs = allocate_messages(ctx->num_messages); + ctx->msgs = allocate_messages(ctx->num_messages, ctx->out_chunk_size); if (!ctx->msgs) return WIMLIB_ERR_NOMEM; @@ -1164,6 +1238,7 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) * 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) @@ -1214,8 +1289,10 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) ret = write_wim_resource(cur_lte, ctx->out_fd, WIMLIB_COMPRESSION_TYPE_NONE, + 0, &cur_lte->output_resource_entry, - ctx->write_resource_flags); + ctx->write_resource_flags, + ctx->comp_ctx); if (ret) return ret; } else { @@ -1254,6 +1331,8 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) 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) @@ -1352,6 +1431,8 @@ main_writer_thread_finish(void *_ctx) ctx->lookup_table, ctx->out_fd, ctx->out_ctype, + ctx->out_chunk_size, + ctx->comp_ctx, ctx->write_resource_flags, ctx->progress_data); } @@ -1368,12 +1449,14 @@ submit_stream_for_compression(struct wim_lookup_table_entry *lte, * when @lte is already hashed. */ sha1_init(&ctx->next_sha_ctx); ctx->next_chunk = 0; - ctx->next_num_chunks = wim_resource_chunks(lte); + ctx->next_num_chunks = DIV_ROUND_UP(wim_resource_size(lte), + 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_resource_prefix(lte, wim_resource_size(lte), - main_writer_thread_cb, ctx, 0); + main_writer_thread_cb, + ctx->out_chunk_size, ctx, 0); if (ret) return ret; wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); @@ -1387,10 +1470,8 @@ main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) int ret; if (wim_resource_size(lte) < 1000 || - ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || - (lte->resource_location == RESOURCE_IN_WIM && - !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) && - lte->wim->compression_type == ctx->out_ctype)) + !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 @@ -1422,15 +1503,15 @@ get_default_num_threads(void) * create the number of threads requested. * * High level description of the algorithm for writing compressed streams in - * parallel: We perform compression on chunks of size WIM_CHUNK_SIZE bytes - * 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. + * 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 @@ -1447,6 +1528,8 @@ 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) @@ -1488,21 +1571,35 @@ write_stream_list_parallel(struct list_head *stream_list, if (ret) goto out_destroy_res_to_compress_queue; - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); + struct compressor_thread_params *params; + + params = CALLOC(num_threads, sizeof(params[0])); + if (params == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } + + 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(NULL, ¶ms[i].comp_ctx); + if (ret) + goto out_free_params; + } + } compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); if (!compressor_threads) { ret = WIMLIB_ERR_NOMEM; - goto out_destroy_compressed_res_queue; + 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); + compressor_thread_proc, ¶ms[i]); if (ret != 0) { ret = -1; ERROR_WITH_ERRNO("Failed to create compressor " @@ -1523,6 +1620,8 @@ write_stream_list_parallel(struct list_head *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; @@ -1558,6 +1657,10 @@ out_join: } } 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: @@ -1571,6 +1674,8 @@ out_serial_quiet: lookup_table, out_fd, out_ctype, + out_chunk_size, + comp_ctx, write_resource_flags, progress_data); @@ -1584,7 +1689,10 @@ out_serial_quiet: static int write_stream_list(struct list_head *stream_list, struct wim_lookup_table *lookup_table, - struct filedes *out_fd, int out_ctype, int write_flags, + 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) { struct wim_lookup_table_entry *lte; @@ -1618,12 +1726,9 @@ write_stream_list(struct list_head *stream_list, list_for_each_entry(lte, stream_list, write_streams_list) { num_streams++; total_bytes += wim_resource_size(lte); - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE - && (wim_resource_compression_type(lte) != out_ctype || - (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS))) - { + if (must_compress_stream(lte, write_resource_flags, + out_ctype, out_chunk_size)) total_compression_bytes += wim_resource_size(lte); - } if (lte->resource_location == RESOURCE_IN_WIM) { if (prev_wim_part != lte->wim) { prev_wim_part = lte->wim; @@ -1653,6 +1758,8 @@ write_stream_list(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + out_chunk_size, + comp_ctx, write_resource_flags, &progress_data, num_threads); @@ -1662,6 +1769,8 @@ write_stream_list(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + out_chunk_size, + comp_ctx, write_resource_flags, &progress_data); if (ret == 0) @@ -1983,7 +2092,9 @@ write_wim_streams(WIMStruct *wim, int image, int write_flags, return write_stream_list(stream_list, wim->lookup_table, &wim->out_fd, - wim->compression_type, + wim->out_compression_type, + wim->out_chunk_size, + &wim->lzx_context, write_flags, num_threads, progress_func); @@ -2042,9 +2153,11 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, "metadata resource.", i); ret = write_wim_resource(imd->metadata_lte, &wim->out_fd, - wim->compression_type, + wim->out_compression_type, + wim->out_chunk_size, &imd->metadata_lte->output_resource_entry, - write_resource_flags); + write_resource_flags, + &wim->lzx_context); } if (ret) return ret; @@ -2074,10 +2187,12 @@ close_wim_writable(WIMStruct *wim, int write_flags) { int ret = 0; - if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) + if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) { + DEBUG("Closing WIM file."); if (filedes_valid(&wim->out_fd)) if (filedes_close(&wim->out_fd)) ret = WIMLIB_ERR_WRITE; + } filedes_invalidate(&wim->out_fd); return ret; } @@ -2087,7 +2202,7 @@ close_wim_writable(WIMStruct *wim, int write_flags) * * Finish writing a WIM file: write the lookup table, xml data, and integrity * table, then overwrite the WIM header. By default, closes the WIM file - * descriptor (@wim->out_fd) in both success and error cases. + * descriptor (@wim->out_fd) if successful. * * write_flags is a bitwise OR of the following: * @@ -2140,6 +2255,8 @@ finish_write(WIMStruct *wim, int image, int write_flags, off_t new_lookup_table_end; u64 xml_totalbytes; + DEBUG("image=%d, write_flags=%08x", image, write_flags); + write_resource_flags = write_flags_to_resource_flags(write_flags); /* In the WIM header, there is room for the resource entry for a @@ -2163,7 +2280,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, &wim->hdr.lookup_table_res_entry, stream_list_override); if (ret) - goto out_close_wim; + return ret; } /* Write XML data. */ @@ -2174,7 +2291,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, &wim->hdr.xml_res_entry, write_resource_flags); if (ret) - goto out_close_wim; + return ret; /* Write integrity table (optional). */ if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { @@ -2186,7 +2303,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, ret = write_wim_header_at_offset(&checkpoint_hdr, &wim->out_fd, 0); if (ret) - goto out_close_wim; + return ret; } if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) @@ -2200,7 +2317,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, old_lookup_table_end, progress_func); if (ret) - goto out_close_wim; + return ret; } else { /* No integrity table. */ zero_resource_entry(&wim->hdr.integrity); @@ -2214,9 +2331,10 @@ finish_write(WIMStruct *wim, int image, int write_flags, hdr_offset = 0; if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END) hdr_offset = wim->out_fd.offset; + DEBUG("Writing new header @ %"PRIu64".", hdr_offset); ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset); if (ret) - goto out_close_wim; + return ret; /* Possibly sync file data to disk before closing. On POSIX systems, it * is necessary to do this before using rename() to overwrite an @@ -2225,22 +2343,19 @@ finish_write(WIMStruct *wim, int image, int write_flags, * operation has been written to disk, but the new file data has not. */ if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + DEBUG("Syncing WIM file."); if (fsync(wim->out_fd.fd)) { ERROR_WITH_ERRNO("Error syncing data to WIM file"); - ret = WIMLIB_ERR_WRITE; - goto out_close_wim; + return WIMLIB_ERR_WRITE; } } - ret = 0; -out_close_wim: if (close_wim_writable(wim, write_flags)) { - if (ret == 0) { - ERROR_WITH_ERRNO("Failed to close the output WIM file"); - ret = WIMLIB_ERR_WRITE; - } + ERROR_WITH_ERRNO("Failed to close the output WIM file"); + return WIMLIB_ERR_WRITE; } - return ret; + + return 0; } #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) @@ -2522,6 +2637,13 @@ write_wim_part(WIMStruct *wim, wim->hdr.part_number = part_number; wim->hdr.total_parts = total_parts; + /* Set compression type if different. */ + if (wim->compression_type != wim->out_compression_type) + wim->hdr.flags = get_wim_hdr_cflags(wim->out_compression_type); + + /* Set chunk size if different. */ + wim->hdr.chunk_size = wim->out_chunk_size; + /* Use GUID if specified; otherwise generate a new one. */ if (guid) memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN); @@ -2624,7 +2746,8 @@ write_wim_part(WIMStruct *wim, stream_list_override); out_restore_hdr: memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); - close_wim_writable(wim, write_flags); + (void)close_wim_writable(wim, write_flags); + DEBUG("ret=%d", ret); return ret; } @@ -2778,7 +2901,7 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, 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; DEBUG("Overwriting `%"TS"' in-place", wim->filename); @@ -2849,31 +2972,32 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, return ret; ret = lock_wim(wim, wim->out_fd.fd); - if (ret) { - close_wim_writable(wim, write_flags); - return ret; - } + if (ret) + goto out_close_wim; + + /* Save original header so it can be restored in case of error */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */ - ret = write_wim_header_flags(wim->hdr.flags | WIM_HDR_FLAG_WRITE_IN_PROGRESS, - &wim->out_fd); + wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd); if (ret) { ERROR_WITH_ERRNO("Error updating WIM header flags"); - close_wim_writable(wim, write_flags); - goto out_unlock_wim; + goto out_restore_memory_hdr; } if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - close_wim_writable(wim, write_flags); ret = WIMLIB_ERR_WRITE; - goto out_unlock_wim; + goto out_restore_physical_hdr; } ret = write_stream_list(&stream_list, wim->lookup_table, &wim->out_fd, wim->compression_type, + wim->chunk_size, + &wim->lzx_context, write_flags, num_threads, progress_func); @@ -2888,15 +3012,25 @@ 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); + if (ret) + goto out_truncate; + + goto out_unlock_wim; + out_truncate: - close_wim_writable(wim, write_flags); - if (ret && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", wim->filename, old_wim_end); - /* Return value of truncate() is ignored because this is already - * an error path. */ - (void)ttruncate(wim->filename, old_wim_end); + /* Return value of ftruncate() is ignored because this is + * already an error path. */ + (void)ftruncate(wim->out_fd.fd, old_wim_end); } +out_restore_physical_hdr: + (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd); +out_restore_memory_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); +out_close_wim: + (void)close_wim_writable(wim, write_flags); out_unlock_wim: wim->wim_locked = 0; return ret; @@ -2923,18 +3057,28 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, write_flags | WIMLIB_WRITE_FLAG_FSYNC, num_threads, progress_func); - if (ret) - goto out_unlink; + if (ret) { + tunlink(tmpfile); + return ret; + } close_wim(wim); + /* Rename the new WIM file to the original WIM file. Note: on Windows + * this actually calls win32_rename_replacement(), not _wrename(), so + * that removing the existing destination file can be handled. */ DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename); - /* Rename the new file to the old file .*/ - if (trename(tmpfile, wim->filename) != 0) { + ret = trename(tmpfile, wim->filename); + if (ret) { ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", tmpfile, wim->filename); - ret = WIMLIB_ERR_RENAME; - goto out_unlink; + #ifdef __WIN32__ + if (ret < 0) + #endif + { + tunlink(tmpfile); + } + return WIMLIB_ERR_RENAME; } if (progress_func) { @@ -2944,11 +3088,6 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } return 0; - -out_unlink: - /* Remove temporary file. */ - tunlink(tmpfile); - return ret; } /* API function documented in wimlib.h */ @@ -2979,7 +3118,9 @@ wimlib_overwrite(WIMStruct *wim, int write_flags, 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_is_pipable(wim)) + && wim->compression_type == wim->out_compression_type + && wim->chunk_size == wim->out_chunk_size) { ret = overwrite_wim_inplace(wim, write_flags, num_threads, progress_func);