X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=c63f0f706dae9f3ad63cc7a927f8721b77587449;hp=0639fd76a0e6385ebb0c2eac1ebca77d18c1c267;hb=5944d11442d5df847c3393951a8563412d16e640;hpb=2254a0fc3f1d7af1151ee83f3458f44339b5028b diff --git a/src/write.c b/src/write.c index 0639fd76..c63f0f70 100644 --- a/src/write.c +++ b/src/write.c @@ -68,17 +68,28 @@ # include /* for `struct iovec' */ #endif -static int -alloc_lzx_context(int write_resource_flags, struct wimlib_lzx_context **ctx_pp) +/* 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) { - struct wimlib_lzx_params params; - params.size_of_this = sizeof(params); - if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_COMPRESS_SLOW) - params.algorithm = WIMLIB_LZX_ALGORITHM_SLOW; - else - params.algorithm = WIMLIB_LZX_ALGORITHM_FAST; - params.use_defaults = 1; - return wimlib_lzx_alloc_context(¶ms, ctx_pp); + 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 @@ -128,6 +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) { @@ -139,7 +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); @@ -210,16 +222,14 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, chunk_tab->bytes_per_chunk_entry, chunk_tab->table_disk_size); } else { - ret = full_pwrite(out_fd, - chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - chunk_tab->table_disk_size, - res_start_offset); - } - if (ret) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); + ret = full_pwrite(out_fd, + chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + chunk_tab->table_disk_size, + res_start_offset); } + if (ret) + ERROR_WITH_ERRNO("Write error"); return ret; } @@ -286,6 +296,7 @@ finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) struct write_resource_ctx { int out_ctype; + u32 out_chunk_size; struct wimlib_lzx_context *comp_ctx; struct chunk_table *chunk_tab; struct filedes *out_fd; @@ -301,6 +312,10 @@ 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); @@ -308,11 +323,16 @@ write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) out_chunk = chunk; out_chunk_size = chunk_size; if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - void *compressed_chunk; - unsigned compressed_size; /* Compress the chunk. */ - compressed_chunk = alloca(chunk_size); + 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, @@ -348,11 +368,15 @@ 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; } /* @@ -371,6 +395,9 @@ error: * One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate which * compression algorithm to use. * + * @out_chunk_size: + * Compressed chunk size to use. + * * @out_res_entry: * On success, this is filled in with the offset, flags, compressed size, * and uncompressed size of the resource in the output WIM. @@ -378,26 +405,30 @@ error: * @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_COMPRESS_SLOW to compress the data as much - * as possible; * * 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 - * the function will fail if they do not match. + * this function will fail if they do not match. */ -int +static 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, 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; @@ -409,53 +440,55 @@ write_wim_resource(struct wim_lookup_table_entry *lte, 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 the same compression type as that - * 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)) - { - /* 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 - * that the written resource can be appropriately formatted. - * However, in neither case is any actual decompression needed. - */ + * 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->is_pipable == !!(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + { resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL; - else + read_size = lte->resource_entry.size; + } else { resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS; + read_size = lte->resource_entry.original_size; + } write_ctx.doing_sha = false; - read_size = lte->resource_entry.size; } else { write_ctx.doing_sha = true; sha1_init(&write_ctx.sha_ctx); read_size = lte->resource_entry.original_size; } - - /* If the output resource is to be compressed, initialize the chunk - * 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. */ + /* 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 = alloc_lzx_context(resource_flags, comp_ctx); + 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) @@ -476,12 +509,17 @@ write_wim_resource(struct wim_lookup_table_entry *lte, } /* Write the entire resource by reading the entire resource and feeding - * the data through the write_resource_cb function. */ + * the data through write_resource_cb(). */ 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; @@ -534,7 +572,7 @@ try_write_again: write_ctx.doing_sha = false; goto try_write_again; } - if (resource_flags & (WIMLIB_READ_RESOURCE_FLAG_RAW)) { + if (resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW) { DEBUG("Copied raw compressed data " "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", out_res_entry->original_size, out_res_entry->size, @@ -558,40 +596,50 @@ out: } /* Like write_wim_resource(), but the resource is specified by a buffer of - * uncompressed data rather a lookup table entry; also writes the SHA1 hash of - * the buffer to @hash_ret. */ + * 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 resource_entry *out_res_entry, 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(). */ - struct wim_lookup_table_entry lte; int ret; + struct wim_lookup_table_entry *lte; - lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte.attached_buffer = (void*)buf; - lte.resource_entry.original_size = buf_size; - lte.resource_entry.flags = reshdr_flags; + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ + + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; + + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = (void*)buf; + lte->resource_entry.original_size = buf_size; + lte->resource_entry.flags = reshdr_flags; if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { - sha1_buffer(buf, buf_size, lte.hash); - lte.unhashed = 0; + sha1_buffer(buf, buf_size, lte->hash); + lte->unhashed = 0; } else { - lte.unhashed = 1; + lte->unhashed = 1; } - ret = write_wim_resource(<e, out_fd, out_ctype, out_res_entry, - write_resource_flags, comp_ctx); + ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, + out_res_entry, write_resource_flags, comp_ctx); if (ret) - return ret; + goto out_free_lte; if (hash_ret) - copy_hash(hash_ret, lte.hash); - return 0; + copy_hash(hash_ret, lte->hash); + ret = 0; +out_free_lte: + lte->resource_location = RESOURCE_NONEXISTENT; + free_lookup_table_entry(lte); + return ret; } @@ -695,6 +743,7 @@ struct compressor_thread_params { 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]; @@ -813,6 +862,7 @@ 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; }; @@ -822,7 +872,9 @@ 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->out_ctype, + ctx->out_chunk_size, + <e->output_resource_entry, ctx->write_resource_flags, ctx->comp_ctx); } @@ -900,7 +952,7 @@ do_write_stream_list(struct list_head *stream_list, if (ret) break; /* In parallel mode, some streams are deferred for later, - * serialized processing; ignore them here. */ + * serialized processing; ignore them here. */ if (lte->deferred) continue; if (lte->unhashed) { @@ -922,6 +974,7 @@ 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) @@ -929,6 +982,7 @@ do_write_stream_list_serial(struct list_head *stream_list, 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, }; @@ -946,8 +1000,6 @@ write_flags_to_resource_flags(int write_flags) if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS; - if (write_flags & WIMLIB_WRITE_FLAG_COMPRESS_SLOW) - resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_COMPRESS_SLOW; if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; return resource_flags; @@ -958,6 +1010,7 @@ 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) @@ -974,6 +1027,7 @@ 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); @@ -1014,7 +1068,7 @@ write_wim_chunks(struct message *msg, struct filedes *out_fd, } ret = full_writev(out_fd, vecs, nvecs); if (ret) - ERROR_WITH_ERRNO("Failed to write WIM chunks"); + ERROR_WITH_ERRNO("Write error"); return ret; } @@ -1024,6 +1078,7 @@ 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; @@ -1047,11 +1102,12 @@ 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; @@ -1079,15 +1135,15 @@ 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; msgs = CALLOC(num_messages, sizeof(struct message)); - if (!msgs) + if (msgs == NULL) 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; } @@ -1109,8 +1165,8 @@ 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); - if (!ctx->msgs) + 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. */ @@ -1130,12 +1186,14 @@ main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) INIT_LIST_HEAD(&ctx->outstanding_streams); ctx->num_outstanding_messages = 0; + /* Message currently being prepared. */ ctx->next_msg = NULL; /* Resources that don't need any chunks compressed are added to this - * list and written directly by the main thread. */ + * list and written directly by the main thread. */ INIT_LIST_HEAD(&ctx->serial_streams); + /* Pointer to chunk table for stream currently being written. */ ctx->cur_chunk_tab = NULL; return 0; @@ -1157,17 +1215,17 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) /* Get the next message from the queue and process it. * The message will contain 1 or more data chunks that have been - * compressed. */ + * 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. + /* 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. */ + * 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, @@ -1195,21 +1253,21 @@ 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) return ret; - } - /* Write the compressed chunks from the message. */ + /* Write the compressed chunks from the message. */ ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab, ctx->write_resource_flags); if (ret) return ret; - /* Was this the last chunk of the stream? If so, finish - * it. */ + /* 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) { @@ -1245,6 +1303,7 @@ 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->comp_ctx); @@ -1280,12 +1339,13 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) * 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). */ + * 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); @@ -1293,7 +1353,7 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) return ret; } - /* Advance to the next stream to write. */ + /* Advance to the next stream to write. */ if (list_empty(&ctx->outstanding_streams)) { cur_lte = NULL; } else { @@ -1306,7 +1366,7 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) return 0; } -/* Called when the main thread has read a new chunk of data. */ +/* Called when the main thread has read a new chunk of data. */ static int main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) { @@ -1385,6 +1445,7 @@ 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); @@ -1399,15 +1460,17 @@ submit_stream_for_compression(struct wim_lookup_table_entry *lte, /* 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. */ + * 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); @@ -1421,15 +1484,13 @@ 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 * process it right here, as the main thread could be in the - * middle of writing a different stream. */ + * middle of writing a different stream. */ list_add_tail(<e->write_streams_list, &ctx->serial_streams); lte->deferred = 1; ret = 0; @@ -1456,15 +1517,15 @@ 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 @@ -1474,13 +1535,13 @@ get_default_num_threads(void) * 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*. - */ + * *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, @@ -1491,12 +1552,14 @@ write_stream_list_parallel(struct list_head *stream_list, 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; + goto out_serial_quiet; } else if (nthreads == 1) { goto out_serial_quiet; } else { @@ -1536,15 +1599,15 @@ write_stream_list_parallel(struct list_head *stream_list, params[i].compressed_res_queue = &compressed_res_queue; params[i].out_ctype = out_ctype; if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { - ret = alloc_lzx_context(write_resource_flags, - ¶ms[i].comp_ctx); + ret = wimlib_lzx_alloc_context(out_chunk_size, + NULL, ¶ms[i].comp_ctx); if (ret) goto out_free_params; } } compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); - if (!compressor_threads) { + if (compressor_threads == NULL) { ret = WIMLIB_ERR_NOMEM; goto out_free_params; } @@ -1553,15 +1616,17 @@ write_stream_list_parallel(struct list_head *stream_list, 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 != 0) { + if (ret) { + errno = ret; ret = -1; ERROR_WITH_ERRNO("Failed to create compressor " "thread %u of %u", i + 1, num_threads); - num_threads = i; + num_started_threads = i; goto out_join; } } + num_started_threads = num_threads; if (progress_data->progress_func) { progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, @@ -1569,10 +1634,14 @@ write_stream_list_parallel(struct list_head *stream_list, } struct main_writer_thread_ctx ctx; + + memset(&ctx, 0, sizeof(ctx)); + 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; @@ -1582,6 +1651,8 @@ write_stream_list_parallel(struct list_head *stream_list, ret = main_writer_thread_init_ctx(&ctx); if (ret) goto out_join; + + can_retry = false; ret = do_write_stream_list(stream_list, lookup_table, main_thread_process_next_stream, &ctx, progress_data); @@ -1593,15 +1664,15 @@ write_stream_list_parallel(struct list_head *stream_list, * 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. */ + * 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_threads; i++) + for (unsigned i = 0; i < num_started_threads; i++) shared_queue_put(&res_to_compress_queue, NULL); - for (unsigned i = 0; i < num_threads; i++) { + 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", @@ -1617,7 +1688,7 @@ 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 (ret >= 0 && ret != WIMLIB_ERR_NOMEM) + if (!can_retry || (ret >= 0 && ret != WIMLIB_ERR_NOMEM)) return ret; out_serial: WARNING("Falling back to single-threaded compression"); @@ -1626,6 +1697,7 @@ out_serial_quiet: lookup_table, out_fd, out_ctype, + out_chunk_size, comp_ctx, write_resource_flags, progress_data); @@ -1633,27 +1705,27 @@ out_serial_quiet: } #endif -/* - * Write a list of streams to a WIM (@out_fd) using the compression type - * @out_ctype and up to @num_threads compressor threads. - */ +/* 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) { - struct wim_lookup_table_entry *lte; - size_t num_streams = 0; - u64 total_bytes = 0; - u64 total_compression_bytes = 0; - struct write_streams_progress_data progress_data; int ret; int write_resource_flags; - unsigned total_parts = 0; - WIMStruct *prev_wim_part = NULL; + 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."); @@ -1665,23 +1737,28 @@ write_stream_list(struct list_head *stream_list, DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", out_fd->offset, write_resource_flags); - sort_stream_list_by_sequential_order(stream_list, - offsetof(struct wim_lookup_table_entry, - write_streams_list)); + /* 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 (ret) + return ret; /* 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). */ + * 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 += 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; @@ -1711,6 +1788,7 @@ write_stream_list(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + out_chunk_size, comp_ctx, write_resource_flags, &progress_data, @@ -1721,13 +1799,14 @@ 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) DEBUG("Successfully wrote stream list."); else - DEBUG("Failed to write stream list."); + DEBUG("Failed to write stream list (ret=%d).", ret); return ret; } @@ -2043,7 +2122,8 @@ 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, @@ -2103,7 +2183,8 @@ 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, &wim->lzx_context); @@ -2136,10 +2217,12 @@ 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; } @@ -2202,6 +2285,8 @@ 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 @@ -2276,6 +2361,7 @@ 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) return ret; @@ -2287,6 +2373,7 @@ 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"); return WIMLIB_ERR_WRITE; @@ -2580,6 +2667,15 @@ 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) { + ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr); + wimlib_assert(ret == 0); + } + + /* 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); @@ -2683,6 +2779,7 @@ write_wim_part(WIMStruct *wim, out_restore_hdr: memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); (void)close_wim_writable(wim, write_flags); + DEBUG("ret=%d", ret); return ret; } @@ -2931,6 +3028,7 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, wim->lookup_table, &wim->out_fd, wim->compression_type, + wim->chunk_size, &wim->lzx_context, write_flags, num_threads, @@ -3052,7 +3150,9 @@ 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);