X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=3fd1d045f0e5d7fbd582f33c5351cbd658521b78;hp=a127a99b486881f7cfa77c50d01e9929476a5292;hb=1c1db3fbc159f6657d533b702ad0f2a4c26142e3;hpb=239adf84c95212272cef9a1d281bd5b8723ee86d diff --git a/src/write.c b/src/write.c index a127a99b..3fd1d045 100644 --- a/src/write.c +++ b/src/write.c @@ -34,6 +34,7 @@ # include #endif +#include "wimlib/chunk_compressor.h" #include "wimlib/endianness.h" #include "wimlib/error.h" #include "wimlib/file_io.h" @@ -42,20 +43,14 @@ #include "wimlib/lookup_table.h" #include "wimlib/metadata.h" #include "wimlib/resource.h" -#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 +# include "wimlib/win32.h" /* win32_rename_replacement() */ #endif +#include "wimlib/write.h" +#include "wimlib/xml.h" #include #include -#include #include #include @@ -63,688 +58,220 @@ # include #endif +/* wimlib internal flags used when writing resources. */ +#define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001 +#define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002 +#define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004 -#ifndef __WIN32__ -# include /* for `struct iovec' */ -#endif +static inline int +write_flags_to_resource_flags(int write_flags) +{ + int write_resource_flags = 0; -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; - } + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS; + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE; + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) + write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS; + return write_resource_flags; } -/* 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); +struct filter_context { + int write_flags; + WIMStruct *wim; }; -/* Allocate and initializes a chunk table, then reserve space for it in the - * output file unless writing a pipable resource. */ +/* 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 -begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, - struct filedes *out_fd, - 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; +stream_filtered(const struct wim_lookup_table_entry *lte, + const struct filter_context *ctx) +{ + int write_flags = ctx->write_flags; + WIMStruct *wim = ctx->wim; - size = wim_resource_size(lte); - num_chunks = wim_resource_chunks(lte); - 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); + 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; - 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; - } - } - *chunk_tab_ret = chunk_tab; return 0; } -/* 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 bool +stream_hard_filtered(const struct wim_lookup_table_entry *lte, + struct filter_context *ctx) { - 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 stream_filtered(lte, ctx) < 0; } -/* 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 inline int +may_soft_filter_streams(const struct filter_context *ctx) { - int ret; - - 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); - } 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"); - } - return ret; + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE; } -/* 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, - int additional_reshdr_flags) +static inline int +may_hard_filter_streams(const struct filter_context *ctx) { - struct pwm_stream_hdr stream_hdr; - u32 reshdr_flags; - int ret; - - stream_hdr.magic = PWM_STREAM_MAGIC; - stream_hdr.uncompressed_size = cpu_to_le64(lte->resource_entry.original_size); - if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) { - zero_out_hash(stream_hdr.hash); - } else { - wimlib_assert(!lte->unhashed); - copy_hash(stream_hdr.hash, lte->hash); - } - - reshdr_flags = lte->resource_entry.flags & ~WIM_RESHDR_FLAG_COMPRESSED; - 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"); - return ret; + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS; } -static int -seek_and_truncate(struct filedes *out_fd, off_t offset) +static inline int +may_filter_streams(const struct filter_context *ctx) { - if (filedes_seek(out_fd, offset) == -1 || - ftruncate(out_fd->fd, offset)) - { - ERROR_WITH_ERRNO("Failed to truncate output WIM file"); - return WIMLIB_ERR_WRITE; - } - return 0; -} - -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); - } - return WIMLIB_ERR_INVALID_RESOURCE_HASH; - } - return 0; + return (may_soft_filter_streams(ctx) || + may_hard_filter_streams(ctx)); } -struct write_resource_ctx { - int out_ctype; - 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) +/* 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 write_resource_ctx *ctx = _ctx; - const void *out_chunk; - unsigned out_chunk_size; - int ret; - - if (ctx->doing_sha) - sha1_update(&ctx->sha_ctx, chunk, chunk_size); - - 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); - - 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; - } - } - - if (ctx->chunk_tab) { - /* Update chunk table accounting. */ - chunk_tab_record_chunk(ctx->chunk_tab, out_chunk_size); - - /* 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; - } - } - - /* Write the chunk data. */ - ret = full_write(ctx->out_fd, out_chunk, out_chunk_size); - if (ret) - goto error; - return 0; - -error: - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return ret; -} - -/* - * 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_res_entry: - * 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 - * the function will fail if they do not match. - */ -int -write_wim_resource(struct wim_lookup_table_entry *lte, - struct filedes *out_fd, int out_ctype, - 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; - u64 read_size; - int ret; + const struct wim_resource_spec *rspec; - /* 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 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. - */ - if (lte->is_pipable == !!(resource_flags & - WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) - 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; - } else { - write_ctx.doing_sha = true; - sha1_init(&write_ctx.sha_ctx); - read_size = lte->resource_entry.original_size; - } + if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS) + return false; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + return false; - /* 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. */ - write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - write_ctx.chunk_tab = NULL; - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - 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, - &write_ctx.chunk_tab, - resource_flags); - if (ret) - goto out; - } - } + if (lte->resource_location != RESOURCE_IN_WIM) + return false; - /* 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); - if (ret) - goto out_free_chunk_tab; - res_start_offset = out_fd->offset; - } + rspec = lte->rspec; - /* Write the entire resource by reading the entire resource and feeding - * the data through the write_resource_cb function. */ - write_ctx.out_fd = out_fd; - write_ctx.resource_flags = resource_flags; -try_write_again: - ret = read_resource_prefix(lte, read_size, - write_resource_cb, &write_ctx, resource_flags); - if (ret) - goto out_free_chunk_tab; + if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) + return false; - /* 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; - } - /* 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 (ret) - goto out_free_chunk_tab; + 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); } - /* Fill in out_res_entry with information about the newly written - * resource. */ - out_res_entry->size = out_fd->offset - res_start_offset; - out_res_entry->flags = lte->resource_entry.flags; - if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) - out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; - else - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; - out_res_entry->offset = res_start_offset; - out_res_entry->original_size = wim_resource_size(lte); - - /* 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_res_entry->size > out_res_entry->original_size && - !(resource_flags & (WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE | - WIMLIB_READ_RESOURCE_FLAG_RAW))) + /* XXX: For compatibility, we can't allow multiple packed resources per + * WIM. */ +#if 0 + if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) { - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - out_res_entry->original_size, out_res_entry->size); - 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_res_entry->original_size, out_res_entry->size, - out_res_entry->offset, out_res_entry->flags); - } else if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - DEBUG("Wrote compressed resource " - "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - out_res_entry->original_size, out_res_entry->size, - out_res_entry->offset, out_res_entry->flags); - } else { - DEBUG("Wrote uncompressed resource " - "(%"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - out_res_entry->original_size, - out_res_entry->offset, out_res_entry->flags); - } - ret = 0; -out_free_chunk_tab: - FREE(write_ctx.chunk_tab); -out: - return ret; -} - -/* Like write_wim_resource(), but the resource is specified by a buffer of - * uncompressed data rather a lookup table entry; also writes the SHA1 hash of - * the buffer to @hash_ret. */ -int -write_wim_resource_from_buffer(const void *buf, size_t buf_size, - int reshdr_flags, struct filedes *out_fd, - int out_ctype, - 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; - - 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; - } else { - lte.unhashed = 1; - } - - ret = write_wim_resource(<e, out_fd, out_ctype, out_res_entry, - write_resource_flags, comp_ctx); - if (ret) - return ret; - if (hash_ret) - copy_hash(hash_ret, lte.hash); - return 0; -} + /* 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. */ + /* 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. */ -#ifdef ENABLE_MULTITHREADED_COMPRESSION + struct wim_lookup_table_entry *res_stream; + u64 write_size = 0; -/* 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; -}; + list_for_each_entry(res_stream, &rspec->stream_list, rspec_node) + if (res_stream->will_be_in_output_wim) + write_size += res_stream->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; + return (write_size > rspec->uncompressed_size / 2); } - 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; -} +#endif -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); + return false; } -static void -shared_queue_put(struct shared_queue *q, void *obj) +static u8 +filter_resource_flags(u8 flags) { - pthread_mutex_lock(&q->lock); - while (q->filled_slots == q->size) - pthread_cond_wait(&q->space_avail_cond, &q->lock); - - q->back = (q->back + 1) % q->size; - q->array[q->back] = obj; - q->filled_slots++; - - pthread_cond_broadcast(&q->msg_avail_cond); - pthread_mutex_unlock(&q->lock); + return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS | + WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SPANNED | + WIM_RESHDR_FLAG_FREE)); } -static void * -shared_queue_get(struct shared_queue *q) +static void +stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte) { - void *obj; - - pthread_mutex_lock(&q->lock); - while (q->filled_slots == 0) - pthread_cond_wait(&q->msg_avail_cond, &q->lock); + const struct wim_resource_spec *rspec; - obj = q->array[q->front]; - q->array[q->front] = NULL; - q->front = (q->front + 1) % q->size; - q->filled_slots--; + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + rspec = lte->rspec; - pthread_cond_broadcast(&q->space_avail_cond); - pthread_mutex_unlock(&q->lock); - return obj; -} + if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { -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(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS); -#define MAX_CHUNKS_PER_MSG 2 + lte->out_reshdr.offset_in_wim = lte->offset_in_res; + lte->out_reshdr.uncompressed_size = 0; + lte->out_reshdr.size_in_wim = lte->size; -struct message { - struct wim_lookup_table_entry *lte; - 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; -}; + 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 { + wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); -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]; - } - msg->out_chunks[i].iov_base = out_chunk; - msg->out_chunks[i].iov_len = out_len; + 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; } + lte->out_reshdr.flags = lte->flags; } -/* 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 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, + int additional_reshdr_flags) { - 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; + struct pwm_stream_hdr stream_hdr; + u32 reshdr_flags; + 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); + stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC); + stream_hdr.uncompressed_size = cpu_to_le64(lte->size); + if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) { + zero_out_hash(stream_hdr.hash); + } else { + wimlib_assert(!lte->unhashed); + copy_hash(stream_hdr.hash, lte->hash); } - DEBUG("Compressor thread terminating"); - return NULL; + + 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("Write error"); + return ret; } -#endif /* ENABLE_MULTITHREADED_COMPRESSION */ struct write_streams_progress_data { wimlib_progress_func_t progress_func; @@ -755,33 +282,38 @@ struct write_streams_progress_data { static void do_write_streams_progress(struct write_streams_progress_data *progress_data, - struct wim_lookup_table_entry *lte, - bool stream_discarded) + struct wim_lookup_table_entry *cur_stream, + u64 complete_size, + u32 complete_count, + bool discarded) { union wimlib_progress_info *progress = &progress_data->progress; bool new_wim_part; - if (stream_discarded) { - progress->write_streams.total_bytes -= wim_resource_size(lte); + if (discarded) { + progress->write_streams.total_bytes -= complete_size; + progress->write_streams.total_streams -= complete_count; 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 += wim_resource_size(lte); + progress->write_streams.completed_bytes += complete_size; + progress->write_streams.completed_streams += complete_count; } + new_wim_part = false; - if (lte->resource_location == RESOURCE_IN_WIM && - lte->wim != progress_data->prev_wim_part) + if (cur_stream->resource_location == RESOURCE_IN_WIM && + cur_stream->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->wim; + progress_data->prev_wim_part = cur_stream->rspec->wim; } - progress->write_streams.completed_streams++; + if (progress_data->progress_func && (progress->write_streams.completed_bytes >= progress_data->next_progress || new_wim_part)) @@ -799,921 +331,1319 @@ do_write_streams_progress(struct write_streams_progress_data *progress_data, } } -struct serial_write_stream_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; - struct wimlib_lzx_context **comp_ctx; + + /* Maximum uncompressed chunk size in compressed resources to use. */ + u32 out_chunk_size; + + /* Flags that affect how the streams will be written. */ int write_resource_flags; -}; -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, <e->output_resource_entry, - ctx->write_resource_flags, - ctx->comp_ctx); -} + /* Data used for issuing WRITE_STREAMS progress. */ + struct write_streams_progress_data progress_data; + + struct filter_context *filter_ctx; + + /* Upper bound on the total number of bytes that need to be compressed. + * */ + u64 num_bytes_to_compress; + + /* 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; + + /* Buffer for dividing the read data into chunks of size + * @out_chunk_size. */ + u8 *chunk_buf; + + /* Number of bytes in @chunk_buf that are currently filled. */ + size_t chunk_buf_filled; + + /* List of streams that currently have chunks being compressed. */ + struct list_head pending_streams; + + /* List of streams in the resource pack. Streams are moved here after + * @pending_streams only when writing a packed resource. */ + struct list_head pack_streams; + + /* 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; + + /* Current uncompressed offset in the stream being read. */ + u64 cur_read_stream_offset; + /* Uncompressed size of the stream currently being read. */ + u64 cur_read_stream_size; -/* 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. */ + /* Current uncompressed offset in the stream being written. */ + u64 cur_write_stream_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; +}; + +/* Reserve space for the chunk table and prepare to accumulate the chunk table + * in memory. */ 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) +begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) { - 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, - wim_resource_size(lte)); - lte->no_progress = 0; - stream_discarded = true; - goto skip_to_progress; - } - } - } + u64 expected_num_chunks; + u64 expected_num_chunk_entries; + size_t reserve_size; + int ret; - /* 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); - 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; + /* 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 & 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; } - skip_to_progress: - if (!lte->no_progress) { - do_write_streams_progress(progress_data, - lte, stream_discarded); + + 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; } + ctx->num_alloc_chunks = new_length; + } + + ctx->chunk_index = 0; + + if (!(ctx->write_resource_flags & 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, + 0 != (ctx->write_resource_flags & + WIM_RESHDR_FLAG_PACKED_STREAMS)); + if (ctx->write_resource_flags & 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) + return ret; } - return ret; + return 0; } 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, - 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, - .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) +begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size) { - int resource_flags = 0; + 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; -} + wimlib_assert(res_expected_size != 0); -static int -write_stream_list_serial(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, - int out_ctype, - 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); + if (ctx->compressor != NULL) { + ret = begin_chunk_table(ctx, res_expected_size); + if (ret) + return ret; } - return do_write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - comp_ctx, - write_resource_flags, - progress_data); + + /* 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_stream_offset = 0; + ctx->cur_write_res_size = res_expected_size; + return 0; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION static int -write_wim_chunks(struct message *msg, struct filedes *out_fd, - struct chunk_table *chunk_tab, - int write_resource_flags) +end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, + u64 *res_start_offset_ret, u64 *res_store_size_ret) { - struct iovec *vecs; - struct pwm_chunk_hdr *chunk_hdrs; - unsigned nvecs; + size_t actual_num_chunks; + size_t actual_num_chunk_entries; + size_t chunk_entry_size; int ret; - for (unsigned i = 0; i < msg->num_chunks; i++) - chunk_tab_record_chunk(chunk_tab, msg->out_chunks[i].iov_len); + actual_num_chunks = ctx->chunk_index; + actual_num_chunk_entries = actual_num_chunks; + if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) + actual_num_chunk_entries--; - if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { - nvecs = msg->num_chunks; - vecs = msg->out_chunks; + chunk_entry_size = get_chunk_entry_size(res_actual_size, + 0 != (ctx->write_resource_flags & + WIM_RESHDR_FLAG_PACKED_STREAMS)); + + typedef le64 __attribute__((may_alias)) aliased_le64_t; + typedef le32 __attribute__((may_alias)) aliased_le32_t; + + if (chunk_entry_size == 4) { + aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & 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 { - /* 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; + aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & 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 = full_writev(out_fd, vecs, nvecs); - if (ret) - ERROR_WITH_ERRNO("Failed to write WIM chunks"); - 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; - 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; -}; + size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size; + u64 res_start_offset; + u64 res_end_offset; -static int -init_message(struct message *msg) -{ - 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); - if (msg->compressed_chunks[i] == NULL || - msg->uncompressed_chunks[i] == NULL) - return WIMLIB_ERR_NOMEM; + if (ctx->write_resource_flags & 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; + + u64 chunk_table_offset; + + chunk_table_offset = ctx->chunks_start_offset - chunk_table_size; + + if (ctx->write_resource_flags & 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; -} -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]); - } +error: + ERROR_WITH_ERRNO("Write error"); + return ret; } -static void -free_messages(struct message *msgs, size_t num_messages) +/* 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) { - if (msgs) { - for (size_t i = 0; i < num_messages; i++) - destroy_message(&msgs[i]); - FREE(msgs); - } + int ret; + u64 res_size_in_wim; + u64 res_uncompressed_size; + u64 res_offset_in_wim; + + wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size || + (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)); + res_uncompressed_size = ctx->cur_write_res_size; + + 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 struct message * -allocate_messages(size_t num_messages) +/* Begin processing a stream for writing. */ +static int +write_stream_begin_read(struct wim_lookup_table_entry *lte, + bool is_partial_res, void *_ctx) { - struct message *msgs; + struct write_streams_ctx *ctx = _ctx; + int ret; + + wimlib_assert(lte->size > 0); + + ctx->cur_read_stream_offset = 0; + ctx->cur_read_stream_size = lte->size; + + /* 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) { + + wimlib_assert(!is_partial_res); - msgs = CALLOC(num_messages, sizeof(struct message)); - if (!msgs) - return NULL; - for (size_t i = 0; i < num_messages; i++) { - if (init_message(&msgs[i])) { - free_messages(msgs, num_messages); - return NULL; + struct wim_lookup_table_entry *lte_new; + + ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new); + if (ret) + return ret; + if (lte_new != lte) { + /* Duplicate stream detected. */ + + 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, lte->size, + 1, 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 & 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; + } } } - return msgs; -} - -static void -main_writer_thread_destroy_ctx(struct main_writer_thread_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); + list_move_tail(<e->write_streams_list, &ctx->pending_streams); + return 0; } +/* Rewrite a stream that was just written compressed as uncompressed instead. + * This function is optional, but if a stream did not compress to less than its + * original size, it might as well be written uncompressed. */ static int -main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) +write_stream_uncompressed(struct wim_lookup_table_entry *lte, + struct filedes *out_fd) { - /* Pre-allocate all the buffers that will be needed to do the chunk - * compression. */ - ctx->msgs = allocate_messages(ctx->num_messages); - if (!ctx->msgs) - return WIMLIB_ERR_NOMEM; + int ret; + u64 begin_offset = lte->out_reshdr.offset_in_wim; + u64 end_offset = out_fd->offset; - /* 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; + if (filedes_seek(out_fd, begin_offset) == -1) + return 0; - ctx->next_msg = NULL; + ret = extract_full_stream_to_fd(lte, out_fd); + if (ret) { + /* Error reading the uncompressed data. */ + if (out_fd->offset == begin_offset && + filedes_seek(out_fd, end_offset) != -1) + { + /* Nothing was actually written yet, and we successfully + * seeked to the end of the compressed resource, so + * don't issue a hard error; just keep the compressed + * resource instead. */ + WARNING("Recovered compressed stream of " + "size %"PRIu64", continuing on.", + lte->size); + return 0; + } + return ret; + } - /* 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); + wimlib_assert(out_fd->offset - begin_offset == lte->size); - ctx->cur_chunk_tab = NULL; + if (out_fd->offset < end_offset && + 0 != ftruncate(out_fd->fd, out_fd->offset)) + { + ERROR_WITH_ERRNO("Can't truncate output file to " + "offset %"PRIu64, out_fd->offset); + return WIMLIB_ERR_WRITE; + } + lte->out_reshdr.size_in_wim = lte->size; + lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_PACKED_STREAMS); return 0; } +/* Write the next chunk of (typically compressed) data to the output WIM, + * handling the writing of the chunk table. */ static int -receive_compressed_chunks(struct main_writer_thread_ctx *ctx) +write_chunk(struct write_streams_ctx *ctx, const void *cchunk, + size_t csize, size_t usize) { - struct message *msg; - struct wim_lookup_table_entry *cur_lte; int ret; - 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) + struct wim_lookup_table_entry *lte; + u32 completed_stream_count; + u32 completed_size; + + lte = list_entry(ctx->pending_streams.next, + struct wim_lookup_table_entry, write_streams_list); + + if (ctx->cur_write_stream_offset == 0 && + !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) { - list_move(&msg->list, &ctx->available_msgs); - if (msg->begin_chunk == 0) { - /* First set of chunks. */ + /* Starting to write a new stream in non-packed mode. */ - /* 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; - } + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + int additional_reshdr_flags = 0; + if (ctx->compressor != NULL) + additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; - /* Save current offset. */ - ctx->res_start_offset = ctx->out_fd->offset; + DEBUG("Writing pipable WIM stream header " + "(offset=%"PRIu64")", ctx->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->cur_chunk_tab, - ctx->write_resource_flags); + ret = write_pwm_stream_header(lte, ctx->out_fd, + additional_reshdr_flags); if (ret) return ret; - } - /* Write the compressed chunks from the message. */ - ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab, - ctx->write_resource_flags); + ret = begin_write_resource(ctx, lte->size); if (ret) return ret; + } - /* Was this the last chunk of the stream? If so, finish - * it. */ - if (list_empty(&cur_lte->msg_list) && - msg->begin_chunk + msg->num_chunks == ctx->cur_chunk_tab->num_chunks) - { - u64 res_csize; + 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; - ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, - ctx->out_fd, - ctx->res_start_offset, - ctx->write_resource_flags); + /* If writing a pipable WIM, before the chunk data write a chunk + * header that provides the compressed chunk size. */ + if (ctx->write_resource_flags & 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) - return ret; + goto error; + } + } + + /* Write the chunk data. */ + ret = full_write(ctx->out_fd, cchunk, csize); + if (ret) + goto error; + + ctx->cur_write_stream_offset += usize; + + completed_size = usize; + completed_stream_count = 0; + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + /* Wrote chunk in packed mode. It may have finished multiple + * streams. */ + while (ctx->cur_write_stream_offset > lte->size) { + struct wim_lookup_table_entry *next; + + ctx->cur_write_stream_offset -= lte->size; + + wimlib_assert(!list_is_singular(&ctx->pending_streams) && + !list_empty(&ctx->pending_streams)); + next = list_entry(lte->write_streams_list.next, + struct wim_lookup_table_entry, + write_streams_list); + list_move_tail(<e->write_streams_list, + &ctx->pack_streams); + lte = next; + completed_stream_count++; + } + if (ctx->cur_write_stream_offset == lte->size) { + ctx->cur_write_stream_offset = 0; + list_move_tail(<e->write_streams_list, + &ctx->pack_streams); + completed_stream_count++; + } + } else { + /* Wrote chunk in non-packed mode. It may have finished a + * stream. */ + if (ctx->cur_write_stream_offset == lte->size) { + + completed_stream_count++; - list_del(&cur_lte->being_compressed_list); + list_del(<e->write_streams_list); - res_csize = ctx->out_fd->offset - ctx->res_start_offset; + wimlib_assert(ctx->cur_write_stream_offset == + ctx->cur_write_res_size); + + ret = end_write_resource(ctx, <e->out_reshdr); + if (ret) + return ret; - FREE(ctx->cur_chunk_tab); - ctx->cur_chunk_tab = NULL; + lte->out_reshdr.flags = filter_resource_flags(lte->flags); + if (ctx->compressor != NULL) + lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED; - /* 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 >= wim_resource_size(cur_lte) && - !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + if (ctx->compressor != NULL && + lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size && + !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) && + !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) { - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - wim_resource_size(cur_lte), 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, - &cur_lte->output_resource_entry, - ctx->write_resource_flags, - ctx->comp_ctx); + /* Stream did not compress to less than its original + * size. If we're not writing a pipable WIM (which + * could mean the output file descriptor is + * non-seekable), and the stream isn't located in a + * resource pack (which would make reading it again + * costly), truncate the file to the start of the stream + * and write it uncompressed instead. */ + DEBUG("Stream of size %"PRIu64" did not compress to " + "less than original size; writing uncompressed.", + lte->size); + ret = write_stream_uncompressed(lte, ctx->out_fd); if (ret) return ret; - } else { - cur_lte->output_resource_entry.size = - res_csize; + } + wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size); - cur_lte->output_resource_entry.original_size = - cur_lte->resource_entry.original_size; + ctx->cur_write_stream_offset = 0; + } + } - cur_lte->output_resource_entry.offset = - ctx->res_start_offset; + do_write_streams_progress(&ctx->progress_data, lte, + completed_size, completed_stream_count, + false); - cur_lte->output_resource_entry.flags = - cur_lte->resource_entry.flags | - WIM_RESHDR_FLAG_COMPRESSED; + return 0; - DEBUG("Wrote compressed resource " - "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", - cur_lte->output_resource_entry.original_size, - cur_lte->output_resource_entry.size, - cur_lte->output_resource_entry.offset, - cur_lte->output_resource_entry.flags); - } +error: + ERROR_WITH_ERRNO("Write error"); + return ret; +} - 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->comp_ctx, - ctx->write_resource_flags, - ctx->progress_data); - if (ret) - return ret; - } +static int +submit_chunk_for_compression(struct write_streams_ctx *ctx, + const void *chunk, size_t size) +{ + /* 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; + + bret = ctx->compressor->get_chunk(ctx->compressor, + &cchunk, &csize, &usize); - /* Advance to the next stream to write. */ - if (list_empty(&ctx->outstanding_streams)) { - cur_lte = NULL; + wimlib_assert(bret); + + ret = write_chunk(ctx, cchunk, csize, usize); + if (ret) + return ret; + } + return 0; +} + +/* Process the next chunk of data to be written to a WIM resource. */ +static int +write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) +{ + struct write_streams_ctx *ctx = _ctx; + int ret; + const u8 *chunkptr, *chunkend; + + wimlib_assert(size != 0); + + if (ctx->compressor == NULL) { + /* Write chunk uncompressed. */ + ret = write_chunk(ctx, chunk, size, size); + if (ret) + return ret; + ctx->cur_read_stream_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 & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + needed_chunk_size = ctx->out_chunk_size; + } else { + u64 res_bytes_remaining; + + res_bytes_remaining = ctx->cur_read_stream_size - + ctx->cur_read_stream_offset; + needed_chunk_size = min(ctx->out_chunk_size, + ctx->chunk_buf_filled + + res_bytes_remaining); + } + + 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_stream_offset += needed_chunk_size; + } else { + /* Intermediate buffering needed. */ + size_t bytes_consumed; + + bytes_consumed = min(chunkend - chunkptr, + needed_chunk_size - ctx->chunk_buf_filled); + + memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled], + chunkptr, bytes_consumed); + + chunkptr += bytes_consumed; + ctx->cur_read_stream_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 { - cur_lte = container_of(ctx->outstanding_streams.next, - struct wim_lookup_table_entry, - being_compressed_list); + break; } + } - } + + ret = submit_chunk_for_compression(ctx, resized_chunk, + needed_chunk_size); + if (ret) + return ret; + + } while (chunkptr != chunkend); return 0; } -/* Called when the main thread has read a new chunk of data. */ +/* 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 -main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) +write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx) { - struct main_writer_thread_ctx *ctx = _ctx; - 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; + struct write_streams_ctx *ctx = _ctx; + if (status == 0) + wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_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 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 +compute_stream_list_stats(struct list_head *stream_list, + struct write_streams_ctx *ctx) +{ + struct wim_lookup_table_entry *lte; + u64 total_bytes = 0; + u64 num_streams = 0; + u64 total_parts = 0; + WIMStruct *prev_wim_part = NULL; + + 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) +{ + struct wim_lookup_table_entry *lte, *tmp; + u64 num_bytes_to_compress = 0; - 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; + INIT_LIST_HEAD(raw_copy_resources); + + /* 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; + + 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 -main_writer_thread_finish(void *_ctx) +write_raw_copy_resource(struct wim_resource_spec *in_rspec, + struct filedes *out_fd) { - struct main_writer_thread_ctx *ctx = _ctx; + u64 cur_read_offset; + u64 end_read_offset; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; int ret; - while (ctx->num_outstanding_messages != 0) { - ret = receive_compressed_chunks(ctx); + struct filedes *in_fd; + struct wim_lookup_table_entry *lte; + u64 out_offset_in_wim; + + DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", " + "uncompressed_size=%"PRIu64")", + in_rspec->size_in_wim, in_rspec->uncompressed_size); + + /* Copy the raw data. */ + cur_read_offset = in_rspec->offset_in_wim; + end_read_offset = cur_read_offset + in_rspec->size_in_wim; + + out_offset_in_wim = out_fd->offset; + + 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; + wimlib_assert(cur_read_offset != end_read_offset); + do { + + 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; + + ret = full_write(out_fd, buf, bytes_to_read); if (ret) return ret; + + cur_read_offset += bytes_to_read; + + } while (cur_read_offset != end_read_offset); + + 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; + + } } - 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->comp_ctx, - ctx->write_resource_flags, - ctx->progress_data); + return 0; } +/* Copy a list of raw compressed resources located other WIM file(s) to the WIM + * file being written. */ static int -submit_stream_for_compression(struct wim_lookup_table_entry *lte, - struct main_writer_thread_ctx *ctx) +write_raw_copy_resources(struct list_head *raw_copy_resources, + struct filedes *out_fd, + struct write_streams_progress_data *progress_data) { + struct wim_lookup_table_entry *lte; 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 = wim_resource_chunks(lte); - 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); - if (ret) - return ret; - wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); - return finalize_and_check_sha1(&ctx->next_sha_ctx, lte); + list_for_each_entry(lte, raw_copy_resources, write_streams_list) { + ret = write_raw_copy_resource(lte->rspec, out_fd); + if (ret) + return ret; + do_write_streams_progress(progress_data, lte, lte->size, + 1, false); + } + return 0; } +/* Wait for and write all chunks pending in the compressor. */ static int -main_thread_process_next_stream(struct wim_lookup_table_entry *lte, 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; - 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)) - { - /* 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); + 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; } - lte->no_progress = 1; - return ret; + + while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) { + ret = write_chunk(ctx, cdata, csize, usize); + if (ret) + return ret; + } + return 0; } -static long -get_default_num_threads(void) +static void +remove_zero_length_streams(struct list_head *stream_list) { -#ifdef __WIN32__ - return win32_get_number_of_processors(); -#else - return sysconf(_SC_NPROCESSORS_ONLN); -#endif + struct wim_lookup_table_entry *lte, *tmp; + + 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); + } + } } -/* 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 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. - * - * 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*. +/* + * Write a list of streams to the output WIM file. + * + * @stream_list + * The list of streams to write, specified by a list of `struct + * wim_lookup_table_entry's linked by the 'write_streams_list' member. + * + * @out_fd + * The file descriptor, opened for writing, to which to write the streams. + * + * @write_resource_flags + * Flags to modify how the streams are written: + * + * WRITE_RESOURCE_FLAG_RECOMPRESS: + * Force compression of all resources, even if they could otherwise + * be re-used by copying the raw data, due to being located in a WIM + * file with compatible compression parameters. + * + * WRITE_RESOURCE_FLAG_PIPABLE: + * Write the resources in the wimlib-specific pipable format, and + * furthermore do so in such a way that no seeking backwards in + * @out_fd will be performed (so it may be a pipe). + * + * WRITE_RESOURCE_FLAG_PACK_STREAMS: + * Pack all the streams into a single resource rather than writing + * them in separate resources. This flag is only valid if the WIM + * version number has been, or will be, set to + * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with + * WRITE_RESOURCE_FLAG_PIPABLE. + * + * @out_ctype + * Compression format to use to write the output streams, specified as one + * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting + * WIMLIB_COMPRESSION_TYPE_INVALID but including + * WIMLIB_COMPRESSION_TYPE_NONE. + * + * @out_chunk_size + * Chunk size to use to write the streams. It must be a valid chunk size + * for the specified compression format @out_ctype, unless @out_ctype is + * WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter is ignored. + * + * @num_threads + * Number of threads to use to compress data. If 0, a default number of + * threads will be chosen. The number of threads still may be decreased + * from the specified value if insufficient memory is detected. + * + * @lookup_table + * If on-the-fly deduplication of unhashed streams is desired, this + * parameter must be pointer to the lookup table for the WIMStruct on whose + * behalf the streams are being written. Otherwise, this parameter can be + * NULL. + * + * @filter_ctx + * If on-the-fly deduplication of unhashed streams is desired, this + * parameter can be a pointer to a context for stream filtering used to + * detect whether the duplicate stream has been hard-filtered or not. If + * no streams are hard-filtered or no streams are unhashed, this parameter + * can be NULL. + * + * @progress_func + * If non-NULL, a progress function that will be called periodically with + * WIMLIB_PROGRESS_MSG_WRITE_STREAMS messages. Note that on-the-fly + * deduplication of unhashed streams may result in the total bytes provided + * in the progress data to decrease from one message to the next. + * + * This function will write the streams in @stream_list to resources in + * consecutive positions in the output WIM file, or to a single packed resource + * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags. + * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for + * each stream written will be updated to specify its location, size, and flags + * in the output WIM. In the packed resource case, + * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each + * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim + * of each @out_reshdr will be set to the offset and size, respectively, in the + * output WIM of the packed resource containing the corresponding stream. + * + * Each of the streams to write may be in any location supported by the + * resource-handling code (specifically, read_stream_list()), such as the + * contents of external file that has been logically added to the output WIM, or + * a stream in another WIM file that has been imported, or even a stream in the + * "same" WIM file of which a modified copy is being written. In the case that + * a stream is already in a WIM file and uses compatible compression parameters, + * by default this function will re-use the raw data instead of decompressing + * it, then recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS + * specified in @write_resource_flags, this is not done. + * + * As a further requirement, this function requires that the + * @will_be_in_output_wim member be set to 1 on all streams in @stream_list as + * well as any other streams not in @stream_list that will be in the output WIM + * file, but set to 0 on any other streams in the output WIM's lookup table or + * sharing a packed resource with a stream in @stream_list. Still furthermore, + * if on-the-fly deduplication of streams is possible, then all streams in + * @stream_list must also be linked by @lookup_table_list along with any other + * streams that have @will_be_in_output_wim set. + * + * This function handles on-the-fly deduplication of streams for which SHA1 + * message digests have not yet been calculated. Such streams may or may not + * need to be written. If @lookup_table is non-NULL, then each stream in + * @stream_list that has @unhashed set but not @unique_size set is checksummed + * immediately before it would otherwise be read for writing in order to + * determine if it is identical to another stream already being written or one + * that would be filtered out of the output WIM using stream_filtered() with the + * context @filter_ctx. Each such duplicate stream will be removed from + * @stream_list, its reference count transfered to the pre-existing duplicate + * stream, its memory freed, and will not be written. Alternatively, if a + * stream in @stream_list is a duplicate with any stream in @lookup_table that + * has not been marked for writing or would not be hard-filtered, it is freed + * and the pre-existing duplicate is written instead, taking ownership of the + * reference count and slot in the @lookup_table_list. + * + * Returns 0 if every stream was either written successfully or did not need to + * be written; otherwise returns a non-zero error code. */ static int -write_stream_list_parallel(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - struct filedes *out_fd, - int out_ctype, - 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, + 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; + struct write_streams_ctx ctx; + struct list_head raw_copy_resources; + + wimlib_assert((write_resource_flags & + (WRITE_RESOURCE_FLAG_PACK_STREAMS | + WRITE_RESOURCE_FLAG_PIPABLE)) != + (WRITE_RESOURCE_FLAG_PACK_STREAMS | + WRITE_RESOURCE_FLAG_PIPABLE)); + + remove_zero_length_streams(stream_list); + + if (list_empty(stream_list)) { + DEBUG("No streams to write."); + return 0; + } + + 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; - 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; - } else if (nthreads == 1) { - goto out_serial_quiet; + 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; + + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + wimlib_assert(out_chunk_size != 0); + if (out_chunk_size <= STACK_MAX) { + ctx.chunk_buf = alloca(out_chunk_size); } else { - num_threads = nthreads; + 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; + + compute_stream_list_stats(stream_list, &ctx); + + ctx.progress_data.progress_func = progress_func; + + ctx.num_bytes_to_compress = find_raw_copy_resources(stream_list, + write_resource_flags, + out_ctype, + out_chunk_size, + &raw_copy_resources); + + 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); + + if (ctx.num_bytes_to_compress == 0) { + DEBUG("No compression needed; skipping to raw copy!"); + goto out_write_raw_copy_resources; + } + + /* 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) { + + #ifdef ENABLE_MULTITHREADED_COMPRESSION + 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); + } + } + #endif + + if (ctx.compressor == NULL) { + ret = new_serial_chunk_compressor(out_ctype, out_chunk_size, + &ctx.compressor); + if (ret) + goto out_destroy_context; + } + } + + 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); + INIT_LIST_HEAD(&ctx.pack_streams); + + if (ctx.progress_data.progress_func) { + (*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + &ctx.progress_data.progress); + } - DEBUG("Writing stream list of size %"PRIu64" " - "(parallel version, num_threads=%u)", - progress->write_streams.total_streams, num_threads); + if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress); + if (ret) + goto out_destroy_context; + } - progress->write_streams.num_threads = num_threads; + /* Read the list of streams needing to be compressed, using the + * specified callbacks to execute processing of the data. */ - static const size_t MESSAGES_PER_THREAD = 2; - size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); + 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, + }; - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); + 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); - ret = shared_queue_init(&res_to_compress_queue, queue_size); if (ret) - goto out_serial; + goto out_destroy_context; - ret = shared_queue_init(&compressed_res_queue, queue_size); + ret = finish_remaining_chunks(&ctx); if (ret) - goto out_destroy_res_to_compress_queue; + goto out_destroy_context; - 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; - } + if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + struct wim_reshdr reshdr; + struct wim_lookup_table_entry *lte; + u64 offset_in_res; - 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; + 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.pack_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; } - } - - compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); - if (!compressor_threads) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_params; - } + wimlib_assert(offset_in_res == reshdr.uncompressed_size); + } + +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, + &ctx.progress_data); + +out_destroy_context: + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && 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; +} - 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 != 0) { - ret = -1; - ERROR_WITH_ERRNO("Failed to create compressor " - "thread %u of %u", - i + 1, num_threads); - num_threads = i; - goto out_join; - } - } +static int +wim_write_stream_list(WIMStruct *wim, + struct list_head *stream_list, + int write_flags, + unsigned num_threads, + struct filter_context *filter_ctx, + wimlib_progress_func_t progress_func) +{ + int out_ctype; + u32 out_chunk_size; + int write_resource_flags; - if (progress_data->progress_func) { - progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - } + write_resource_flags = write_flags_to_resource_flags(write_flags); - struct main_writer_thread_ctx ctx; - ctx.stream_list = stream_list; - ctx.lookup_table = lookup_table; - ctx.out_fd = out_fd; - ctx.out_ctype = out_ctype; - 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; - ret = do_write_stream_list(stream_list, lookup_table, - main_thread_process_next_stream, - &ctx, progress_data); - 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_threads; i++) - shared_queue_put(&res_to_compress_queue, NULL); - - for (unsigned i = 0; i < num_threads; i++) { - if (pthread_join(compressor_threads[i], NULL)) { - WARNING_WITH_ERRNO("Failed to join compressor " - "thread %u of %u", - i + 1, num_threads); - } + if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + out_chunk_size = wim->out_pack_chunk_size; + out_ctype = wim->out_pack_compression_type; + } else { + out_chunk_size = wim->out_chunk_size; + out_ctype = wim->out_compression_type; } - 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 (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, - comp_ctx, - write_resource_flags, - progress_data); + return write_stream_list(stream_list, + &wim->out_fd, + write_resource_flags, + out_ctype, + out_chunk_size, + num_threads, + wim->lookup_table, + filter_ctx, + progress_func); } -#endif -/* - * Write a list of streams to a WIM (@out_fd) using the compression type - * @out_ctype 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, - 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) +{ + 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 & ~WRITE_RESOURCE_FLAG_PACK_STREAMS, + out_ctype, + out_chunk_size, + 1, + NULL, + NULL, + 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 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; - - if (list_empty(stream_list)) { - DEBUG("No streams to write."); - return 0; - } + struct wim_lookup_table_entry *lte; - write_resource_flags = write_flags_to_resource_flags(write_flags); + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ - DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", - out_fd->offset, write_resource_flags); + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; - sort_stream_list_by_sequential_order(stream_list, - offsetof(struct wim_lookup_table_entry, - write_streams_list)); + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = (void*)buf; + lte->size = buf_size; + lte->flags = reshdr_flags; - /* 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). */ - 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))) - { - 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; - total_parts++; - } - } + if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + sha1_buffer(buf, buf_size, lte->hash); + lte->unhashed = 0; + } else { + lte->unhashed = 1; } - 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, - comp_ctx, - write_resource_flags, - &progress_data, - num_threads); - else -#endif - ret = write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - comp_ctx, - write_resource_flags, - &progress_data); - if (ret == 0) - DEBUG("Successfully wrote stream list."); - else - DEBUG("Failed to write stream list."); + ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, + write_resource_flags); + if (ret) + goto out_free_lte; + + copy_reshdr(out_reshdr, <e->out_reshdr); + + if (hash) + copy_hash(hash, lte->hash); + ret = 0; +out_free_lte: + lte->resource_location = RESOURCE_NONEXISTENT; + free_lookup_table_entry(lte); return ret; } @@ -1727,7 +1657,7 @@ 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; @@ -1748,10 +1678,10 @@ stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) struct wim_lookup_table_entry *same_size_lte; struct hlist_node *tmp; - pos = hash_u64(wim_resource_size(lte)) % tab->capacity; + pos = hash_u64(lte->size) % tab->capacity; lte->unique_size = 1; hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) { - if (wim_resource_size(same_size_lte) == wim_resource_size(lte)) { + if (same_size_lte->size == lte->size) { lte->unique_size = 0; same_size_lte->unique_size = 0; break; @@ -1771,40 +1701,42 @@ 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; } @@ -1812,52 +1744,46 @@ 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; - struct wim_lookup_table_entry *lte; + + INIT_LIST_HEAD(stream_list_ret); if (streams_ok && (image == WIMLIB_ALL_IMAGES || (image == 1 && wim->hdr.image_count == 1))) @@ -1869,150 +1795,225 @@ 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; } - list_for_each_entry(lte, &ctx->stream_list, write_streams_list) - lte->part_number = wim->hdr.part_number; 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->wim == ctx->wim && - (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE)) - filtered |= FILTERED_SAME_WIM; - if (lte->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; - } - - for_lookup_table_entry(ctx->wim->lookup_table, - process_filtered_stream, ctx); + int status = stream_filtered(lte, filter_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: + * + * 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. + * + * 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. * - * Note: some additional data is stored in each `struct wim_lookup_table_entry': + * 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. * - * - '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. + * @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; - INIT_LIST_HEAD(&ctx.stream_list); - ret = init_stream_size_table(&ctx.stream_size_tab, - wim->lookup_table->capacity); + filter_ctx_ret->write_flags = write_flags; + filter_ctx_ret->wim = wim; + + 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 { @@ -2020,20 +2021,22 @@ 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; + 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->refcnt : 1); - lte->part_number = wim->hdr.part_number; + 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, - wim->out_compression_type, - &wim->lzx_context, - write_flags, - num_threads, - progress_func); + return wim_write_stream_list(wim, + stream_list, + write_flags, + num_threads, + filter_ctx, + progress_func); } static int @@ -2052,6 +2055,8 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, write_resource_flags = write_flags_to_resource_flags(write_flags); + write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS; + DEBUG("Writing metadata resources (offset=%"PRIu64")", wim->out_fd.offset); @@ -2081,8 +2086,7 @@ 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); - copy_resource_entry(&imd->metadata_lte->output_resource_entry, - &imd->metadata_lte->resource_entry); + stream_set_out_reshdr_for_reuse(imd->metadata_lte); ret = 0; } else { DEBUG("Image %u was not modified; copying existing " @@ -2090,9 +2094,8 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, ret = write_wim_resource(imd->metadata_lte, &wim->out_fd, wim->out_compression_type, - &imd->metadata_lte->output_resource_entry, - write_resource_flags, - &wim->lzx_context); + wim->out_chunk_size, + write_resource_flags); } if (ret) return ret; @@ -2132,6 +2135,89 @@ 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)); +} + /* * finish_write(): * @@ -2181,7 +2267,7 @@ 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; @@ -2200,20 +2286,20 @@ finish_write(WIMStruct *wim, int image, int write_flags, * it should be a copy of the resource entry for the image that is * marked as bootable. This is not well documented... */ if (wim->hdr.boot_idx == 0) { - zero_resource_entry(&wim->hdr.boot_metadata_res_entry); + zero_reshdr(&wim->hdr.boot_metadata_reshdr); } else { - copy_resource_entry(&wim->hdr.boot_metadata_res_entry, - &wim->image_metadata[wim->hdr.boot_idx- 1 - ]->metadata_lte->output_resource_entry); + copy_reshdr(&wim->hdr.boot_metadata_reshdr, + &wim->image_metadata[ + wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr); } /* Write lookup table. (Save old position first.) */ - old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + - wim->hdr.lookup_table_res_entry.size; + old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + + wim->hdr.lookup_table_reshdr.size_in_wim; if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { ret = write_wim_lookup_table(wim, image, write_flags, - &wim->hdr.lookup_table_res_entry, - stream_list_override); + &wim->hdr.lookup_table_reshdr, + lookup_table_list); if (ret) return ret; } @@ -2223,7 +2309,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES) xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING; ret = write_wim_xml_data(wim, image, xml_totalbytes, - &wim->hdr.xml_res_entry, + &wim->hdr.xml_data_reshdr, write_resource_flags); if (ret) return ret; @@ -2233,7 +2319,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { struct wim_header checkpoint_hdr; memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header)); - zero_resource_entry(&checkpoint_hdr.integrity); + zero_reshdr(&checkpoint_hdr.integrity_table_reshdr); checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; ret = write_wim_header_at_offset(&checkpoint_hdr, &wim->out_fd, 0); @@ -2244,8 +2330,8 @@ finish_write(WIMStruct *wim, int image, int write_flags, if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) old_lookup_table_end = 0; - new_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + - wim->hdr.lookup_table_res_entry.size; + new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + + wim->hdr.lookup_table_reshdr.size_in_wim; ret = write_integrity_table(wim, new_lookup_table_end, @@ -2255,7 +2341,7 @@ finish_write(WIMStruct *wim, int image, int write_flags, return ret; } else { /* No integrity table. */ - zero_resource_entry(&wim->hdr.integrity); + zero_reshdr(&wim->hdr.integrity_table_reshdr); } /* Now that all information in the WIM header has been determined, the @@ -2400,10 +2486,11 @@ 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 resource_entry xml_res_entry; + struct wim_reshdr xml_reshdr; WARNING("Creating a pipable WIM, which will " "be incompatible\n" @@ -2426,8 +2513,8 @@ write_pipable_wim(WIMStruct *wim, int image, int write_flags, /* Write extra copy of the XML data. */ ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, - &xml_res_entry, - WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE); + &xml_reshdr, + WRITE_RESOURCE_FLAG_PIPABLE); if (ret) return ret; @@ -2441,28 +2528,13 @@ 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(). */ } -/* API function documented in wimlib.h */ -WIMLIBAPI int -wimlib_set_output_compression_type(WIMStruct *wim, int ctype) -{ - switch (ctype) { - case WIMLIB_COMPRESSION_TYPE_INVALID: - break; - case WIMLIB_COMPRESSION_TYPE_NONE: - case WIMLIB_COMPRESSION_TYPE_LZX: - case WIMLIB_COMPRESSION_TYPE_XPRESS: - wim->out_compression_type = ctype; - return 0; - } - return WIMLIB_ERR_INVALID_PARAM; -} - /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file * descriptor. */ int @@ -2479,7 +2551,7 @@ 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."); @@ -2494,26 +2566,43 @@ 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_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 @@ -2556,16 +2645,29 @@ write_wim_part(WIMStruct *wim, /* 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_PIPABLE | + WIMLIB_WRITE_FLAG_PACK_STREAMS)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_PACK_STREAMS)) + { + ERROR("Cannot specify both PIPABLE and PACK_STREAMS!"); + return WIMLIB_ERR_INVALID_PARAM; + } /* Set appropriate magic number. */ if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) @@ -2573,6 +2675,13 @@ write_wim_part(WIMStruct *wim, else wim->hdr.magic = WIM_MAGIC; + /* Set appropriate version number. */ + if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) || + wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS) + 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 | @@ -2589,8 +2698,13 @@ write_wim_part(WIMStruct *wim, 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); + 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) @@ -2599,10 +2713,10 @@ write_wim_part(WIMStruct *wim, randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN); /* Clear references to resources that have not been written yet. */ - zero_resource_entry(&wim->hdr.lookup_table_res_entry); - zero_resource_entry(&wim->hdr.xml_res_entry); - zero_resource_entry(&wim->hdr.boot_metadata_res_entry); - zero_resource_entry(&wim->hdr.integrity); + zero_reshdr(&wim->hdr.lookup_table_reshdr); + zero_reshdr(&wim->hdr.xml_data_reshdr); + zero_reshdr(&wim->hdr.boot_metadata_reshdr); + zero_reshdr(&wim->hdr.integrity_table_reshdr); /* Set image count and boot index correctly for single image writes. */ if (image != WIMLIB_ALL_IMAGES) { @@ -2654,22 +2768,12 @@ 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; @@ -2680,18 +2784,17 @@ 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); @@ -2755,8 +2858,8 @@ check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim) const WIMStruct *wim = _wim; off_t end_offset = *(const off_t*)wim->private; - if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == wim && - lte->resource_entry.offset + lte->resource_entry.size > end_offset) + if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim && + lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset) return WIMLIB_ERR_RESOURCE_ORDER; return 0; } @@ -2846,19 +2949,28 @@ 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); + /* Save original header so it can be restored in case of error */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); + /* Set default integrity flag. */ if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) if (wim_has_integrity_table(wim)) write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + /* Set WIM version if adding packed streams. */ + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) + wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS; + /* Set additional flags for overwrite. */ write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE | WIMLIB_WRITE_FLAG_STREAMS_OK; @@ -2867,18 +2979,21 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, * data, and that there are no stream resources, metadata resources, or * lookup tables after the XML data. Otherwise, these data would be * overwritten. */ - old_xml_begin = wim->hdr.xml_res_entry.offset; - old_xml_end = old_xml_begin + wim->hdr.xml_res_entry.size; - old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + - wim->hdr.lookup_table_res_entry.size; - if (wim->hdr.integrity.offset != 0 && wim->hdr.integrity.offset < old_xml_end) { + old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim; + old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim; + old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + + wim->hdr.lookup_table_reshdr.size_in_wim; + if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 && + wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) { WARNING("Didn't expect the integrity table to be before the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out_restore_memory_hdr; } if (old_lookup_table_end > old_xml_begin) { WARNING("Didn't expect the lookup table to be after the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out_restore_memory_hdr; } /* Set @old_wim_end, which indicates the point beyond which we don't @@ -2896,10 +3011,11 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, old_wim_end = old_lookup_table_end; write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; - } else if (wim->hdr.integrity.offset) { + } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) { /* Old WIM has an integrity table; begin writing new streams * after it. */ - old_wim_end = wim->hdr.integrity.offset + wim->hdr.integrity.size; + old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim + + wim->hdr.integrity_table_reshdr.size_in_wim; } else { /* No existing integrity table; begin writing new streams after * the old XML data. */ @@ -2908,30 +3024,28 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, ret = check_resource_offsets(wim, old_wim_end); if (ret) - return ret; + goto out_restore_memory_hdr; - 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; + goto out_restore_memory_hdr; ret = open_wim_writable(wim, wim->filename, O_RDWR); if (ret) - return ret; + goto out_restore_memory_hdr; ret = lock_wim(wim, wim->out_fd.fd); 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. */ 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"); - goto out_restore_memory_hdr; + goto out_unlock_wim; } if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { @@ -2940,14 +3054,12 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, goto out_restore_physical_hdr; } - ret = write_stream_list(&stream_list, - wim->lookup_table, - &wim->out_fd, - wim->compression_type, - &wim->lzx_context, - write_flags, - num_threads, - progress_func); + ret = wim_write_stream_list(wim, + &stream_list, + write_flags, + num_threads, + &filter_ctx, + progress_func); if (ret) goto out_truncate; @@ -2958,11 +3070,12 @@ 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; - goto out_unlock_wim; + wim->wim_locked = 0; + return 0; out_truncate: if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { @@ -2974,12 +3087,12 @@ out_truncate: } 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; +out_close_wim: + (void)close_wim_writable(wim, write_flags); +out_restore_memory_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); return ret; } @@ -3037,6 +3150,43 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, return 0; } +/* Determine if the specified WIM file may be updated by appending in-place + * rather than writing and replacing it with an entirely new file. */ +static bool +can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) +{ + /* REBUILD flag forces full rebuild. */ + if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) + return false; + + /* Deletions cause full rebuild by default. */ + if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + return false; + + /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be + * turned into a pipable WIM in-place. */ + if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + return false; + + /* wimlib allows multiple packs in a single WIM, but they don't seem to + * be compatible with WIMGAPI, so force all streams to be repacked if + * the WIM already may have contained a pack and PACK_STREAMS was + * requested. */ + if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS && + wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS) + return false; + + /* The default compression type and compression chunk size selected for + * the output WIM must be the same as those currently used for the WIM. + */ + 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, @@ -3062,12 +3212,7 @@ 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) - { + 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)