X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=27463d90d059879426f9df0f17121df8e08f4fea;hp=9965f5c06bf2d21f853bb6d1bfedda0cbf3d98a3;hb=d82950b8ae73d50bffebb88c2b5c10b49f4082f4;hpb=f9695b9f40035f1a20968293255761a8301eaba0 diff --git a/src/write.c b/src/write.c index 9965f5c0..27463d90 100644 --- a/src/write.c +++ b/src/write.c @@ -24,1430 +24,1647 @@ * along with wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "config.h" +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) -/* On BSD, this should be included before "list.h" so that "list.h" can +/* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can * overwrite the LIST_HEAD macro. */ # include #endif +#include "wimlib/chunk_compressor.h" +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/header.h" +#include "wimlib/inode.h" +#include "wimlib/integrity.h" +#include "wimlib/lookup_table.h" +#include "wimlib/metadata.h" +#include "wimlib/progress.h" +#include "wimlib/resource.h" #ifdef __WIN32__ -# include "win32.h" -#endif - -#include "list.h" -#include "wimlib_internal.h" -#include "buffer_io.h" -#include "dentry.h" -#include "lookup_table.h" -#include "xml.h" - -#ifdef ENABLE_MULTITHREADED_COMPRESSION -# include +# include "wimlib/win32.h" /* win32_rename_replacement() */ #endif +#include "wimlib/write.h" +#include "wimlib/xml.h" -#include #include - -#ifdef WITH_NTFS_3G -# include -# include -# include -# include -#endif +#include +#include +#include #ifdef HAVE_ALLOCA_H # include -#else -# include #endif -#include - -#include /* writev() */ - -/* 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 { - off_t file_offset; - u64 num_chunks; - u64 original_resource_size; - u64 bytes_per_chunk_entry; - u64 table_disk_size; - u64 cur_offset; - u64 *cur_offset_p; - u64 offsets[0]; +/* 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 + +static inline int +write_flags_to_resource_flags(int write_flags) +{ + int write_resource_flags = 0; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= 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; +} + +struct filter_context { + int write_flags; + WIMStruct *wim; }; -/* - * Allocates and initializes a chunk table, and reserves space for it in the - * output file. +/* 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, - int out_fd, - off_t file_offset, - struct chunk_table **chunk_tab_ret) +stream_filtered(const struct wim_lookup_table_entry *lte, + const struct filter_context *ctx) { - u64 size = wim_resource_size(lte); - u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); - struct chunk_table *chunk_tab = CALLOC(1, alloc_size); + int write_flags = ctx->write_flags; + WIMStruct *wim = ctx->wim; - DEBUG("Begin chunk table for stream with size %"PRIu64, 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->file_offset = file_offset; - chunk_tab->num_chunks = num_chunks; - chunk_tab->original_resource_size = size; - chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4; - chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * - (num_chunks - 1); - chunk_tab->cur_offset = 0; - chunk_tab->cur_offset_p = chunk_tab->offsets; - - if (full_write(out_fd, chunk_tab, - chunk_tab->table_disk_size) != chunk_tab->table_disk_size) - { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - FREE(chunk_tab); - return WIMLIB_ERR_WRITE; - } - *chunk_tab_ret = chunk_tab; return 0; } -/* - * compress_func_t- Pointer to a function to compresses a chunk - * of a WIM resource. This may be either - * wimlib_xpress_compress() (xpress-compress.c) or - * wimlib_lzx_compress() (lzx-compress.c). - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk, in bytes. - * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes. - * - * Returns the size of the compressed data written to @out in bytes, or 0 if the - * data could not be compressed to (@chunk_size - 1) bytes or fewer. - * - * As a special requirement, the compression code is optimized for the WIM - * format and therefore requires (@chunk_size <= 32768). - * - * As another special requirement, the compression code will read up to 8 bytes - * off the end of the @chunk array for performance reasons. The values of these - * bytes will not affect the output of the compression, but the calling code - * must make sure that the buffer holding the uncompressed chunk is actually at - * least (@chunk_size + 8) bytes, or at least that these extra bytes are in - * mapped memory that will not cause a memory access violation if accessed. - */ -typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, - void *out); +static bool +stream_hard_filtered(const struct wim_lookup_table_entry *lte, + struct filter_context *ctx) +{ + return stream_filtered(lte, ctx) < 0; +} -static compress_func_t -get_compress_func(int out_ctype) +static inline int +may_soft_filter_streams(const struct filter_context *ctx) { - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return wimlib_lzx_compress; - else - return wimlib_xpress_compress; + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE; } -/* - * Writes a chunk of a WIM resource to an output file. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) - * @out_fd: FILE descriptor to write the chunk to. - * @compress: Compression function to use (NULL if writing uncompressed - * data). - * @chunk_tab: Pointer to chunk table being created. It is updated with the - * offset of the chunk we write. - * - * Returns 0 on success; nonzero on failure. - */ -static int -write_wim_resource_chunk(const void * restrict chunk, - unsigned chunk_size, - filedes_t out_fd, - compress_func_t compress, - struct chunk_table * restrict chunk_tab) -{ - const void *out_chunk; - unsigned out_chunk_size; - if (compress) { - void *compressed_chunk = alloca(chunk_size); - - out_chunk_size = (*compress)(chunk, chunk_size, compressed_chunk); - if (out_chunk_size) { - /* Write compressed */ - out_chunk = compressed_chunk; - } else { - /* Write uncompressed */ - out_chunk = chunk; - out_chunk_size = chunk_size; - } - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += out_chunk_size; - } else { - /* Write uncompressed */ - out_chunk = chunk; - out_chunk_size = chunk_size; +static inline int +may_hard_filter_streams(const struct filter_context *ctx) +{ + if (ctx == NULL) + return 0; + return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS; +} + +static inline int +may_filter_streams(const struct filter_context *ctx) +{ + return (may_soft_filter_streams(ctx) || + may_hard_filter_streams(ctx)); +} + + +/* 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) +{ + const struct wim_resource_spec *rspec; + + if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS) + return false; + + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + return false; + + if (lte->resource_location != RESOURCE_IN_WIM) + return false; + + rspec = lte->rspec; + + if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) + return false; + + if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) { + /* Normal compressed resource: Must use same compression type + * and chunk size. */ + return (rspec->compression_type == out_ctype && + rspec->chunk_size == out_chunk_size); } - if (full_write(out_fd, out_chunk, out_chunk_size) != out_chunk_size) { - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return WIMLIB_ERR_WRITE; + + if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) + { + /* Packed resource: Such resources may contain multiple streams, + * and in general only a subset of them need to be written. As + * a heuristic, re-use the raw data if more than two-thirds 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. */ + + struct wim_lookup_table_entry *res_stream; + u64 write_size = 0; + + list_for_each_entry(res_stream, &rspec->stream_list, rspec_node) + if (res_stream->will_be_in_output_wim) + write_size += res_stream->size; + + return (write_size > rspec->uncompressed_size * 2 / 3); } - return 0; + + return false; } -/* - * Finishes a WIM chunk table and writes it to the output file at the correct - * offset. - * - * The final size of the full compressed resource is returned in the - * @compressed_size_p. - */ -static int -finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - filedes_t out_fd, u64 *compressed_size_p) +static u8 +filter_resource_flags(u8 flags) +{ + return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS | + WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SPANNED | + WIM_RESHDR_FLAG_FREE)); +} + +static void +stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte) { - size_t bytes_written; + const struct wim_resource_spec *rspec; + + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + rspec = lte->rspec; - if (chunk_tab->bytes_per_chunk_entry == 8) { - array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); + if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { + + wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS); + + lte->out_reshdr.offset_in_wim = lte->offset_in_res; + lte->out_reshdr.uncompressed_size = 0; + lte->out_reshdr.size_in_wim = lte->size; + + lte->out_res_offset_in_wim = rspec->offset_in_wim; + lte->out_res_size_in_wim = rspec->size_in_wim; + lte->out_res_uncompressed_size = rspec->uncompressed_size; } else { - for (u64 i = 0; i < chunk_tab->num_chunks; i++) - ((u32*)chunk_tab->offsets)[i] = - cpu_to_le32(chunk_tab->offsets[i]); - } - bytes_written = full_pwrite(out_fd, - (u8*)chunk_tab->offsets + chunk_tab->bytes_per_chunk_entry, - chunk_tab->table_disk_size, - chunk_tab->file_offset); - if (bytes_written != chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - return WIMLIB_ERR_WRITE; + wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); + + lte->out_reshdr.offset_in_wim = rspec->offset_in_wim; + lte->out_reshdr.uncompressed_size = rspec->uncompressed_size; + lte->out_reshdr.size_in_wim = rspec->size_in_wim; } - *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; - return 0; + lte->out_reshdr.flags = lte->flags; } + +/* Write the header for a stream in a pipable WIM. */ static int -seek_and_truncate(filedes_t out_fd, off_t offset) +write_pwm_stream_header(const struct wim_lookup_table_entry *lte, + struct filedes *out_fd, + int additional_reshdr_flags) { - if (lseek(out_fd, offset, SEEK_SET) == -1 || - ftruncate(out_fd, offset)) - { - ERROR_WITH_ERRNO("Failed to truncate output WIM file"); - return WIMLIB_ERR_WRITE; + struct pwm_stream_hdr stream_hdr; + u32 reshdr_flags; + int ret; + + 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 { - return 0; + wimlib_assert(!lte->unhashed); + copy_hash(stream_hdr.hash, lte->hash); } + + 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; } +struct write_streams_progress_data { + wimlib_progress_func_t progfunc; + void *progctx; + union wimlib_progress_info progress; + uint64_t next_progress; +}; + static int -finalize_and_check_sha1(SHA_CTX * restrict sha_ctx, - struct wim_lookup_table_entry * restrict 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); +do_write_streams_progress(struct write_streams_progress_data *progress_data, + struct wim_lookup_table_entry *cur_stream, + u64 complete_size, + u32 complete_count, + bool discarded) +{ + union wimlib_progress_info *progress = &progress_data->progress; + int ret; + + 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 += complete_size; + progress->write_streams.completed_streams += complete_count; + } + + if (progress->write_streams.completed_bytes >= progress_data->next_progress) + { + ret = call_progress(progress_data->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress, + progress_data->progctx); + if (ret) + return ret; + + if (progress_data->next_progress == progress->write_streams.total_bytes) { + progress_data->next_progress = ~(uint64_t)0; + } else { + progress_data->next_progress = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); } - return WIMLIB_ERR_INVALID_RESOURCE_HASH; } return 0; } +struct write_streams_ctx { + /* File descriptor the streams are being written to. */ + struct filedes *out_fd; -struct write_resource_ctx { - compress_func_t compress; - struct chunk_table *chunk_tab; - filedes_t out_fd; - SHA_CTX sha_ctx; - bool doing_sha; + /* Lookup table for the WIMStruct on whose behalf the streams are being + * written. */ + struct wim_lookup_table *lookup_table; + + /* Compression format to use. */ + int out_ctype; + + /* Maximum uncompressed chunk size in compressed resources to use. */ + u32 out_chunk_size; + + /* Flags that affect how the streams will be written. */ + int write_resource_flags; + + /* 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; + + /* 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 -write_resource_cb(const void *restrict chunk, size_t chunk_size, - void *restrict _ctx) +begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) { - struct write_resource_ctx *ctx = _ctx; + u64 expected_num_chunks; + u64 expected_num_chunk_entries; + size_t reserve_size; + int ret; - if (ctx->doing_sha) - sha1_update(&ctx->sha_ctx, chunk, chunk_size); - return write_wim_resource_chunk(chunk, chunk_size, - ctx->out_fd, ctx->compress, - ctx->chunk_tab); + /* 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; + } + + 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 0; } -/* - * 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_fp: 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. - * - * @flags: WIMLIB_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed - * even if it could otherwise be copied directly from the input. - * - * 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, - filedes_t out_fd, int out_ctype, - struct resource_entry *out_res_entry, - int flags) -{ - struct write_resource_ctx write_ctx; - u64 read_size; - u64 new_size; - off_t offset; +static int +begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size) +{ int ret; - flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS; + wimlib_assert(res_expected_size != 0); - /* Get current position in output WIM */ - offset = filedes_offset(out_fd); - if (offset == -1) { - ERROR_WITH_ERRNO("Can't get position in output WIM"); - return WIMLIB_ERR_WRITE; + if (ctx->compressor != NULL) { + ret = begin_chunk_table(ctx, res_expected_size); + if (ret) + return ret; } - /* 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 (!(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && - lte->resource_location == RESOURCE_IN_WIM && - out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && - wimlib_get_compression_type(lte->wim) == out_ctype) - { - flags |= WIMLIB_RESOURCE_FLAG_RAW; - 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; - } + /* 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; +} + +static int +end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, + u64 *res_start_offset_ret, u64 *res_store_size_ret) +{ + size_t actual_num_chunks; + size_t actual_num_chunk_entries; + size_t chunk_entry_size; + int ret; + + 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--; + + 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; - /* Initialize the chunk table and set the compression function if - * compressing the resource. */ - if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || - (flags & WIMLIB_RESOURCE_FLAG_RAW)) { - write_ctx.compress = NULL; - write_ctx.chunk_tab = NULL; + 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 { - write_ctx.compress = get_compress_func(out_ctype); - ret = begin_wim_resource_chunk_tab(lte, out_fd, - offset, - &write_ctx.chunk_tab); - if (ret) - return ret; + 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; + } + } } - /* 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; -try_write_again: - ret = read_resource_prefix(lte, read_size, - write_resource_cb, &write_ctx, flags); - if (ret) - goto out_free_chunk_tab; + size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size; + u64 res_start_offset; + u64 res_end_offset; - /* 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 (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size); if (ret) - goto out_free_chunk_tab; - } - - out_res_entry->flags = lte->resource_entry.flags; - out_res_entry->original_size = wim_resource_size(lte); - out_res_entry->offset = offset; - if (flags & WIMLIB_RESOURCE_FLAG_RAW) { - /* Doing a raw write: The new compressed size is the same as - * the compressed size in the other WIM. */ - new_size = lte->resource_entry.size; - } else if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) { - /* Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size - * is the original size. */ - new_size = lte->resource_entry.original_size; - out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; + goto error; + res_end_offset = ctx->out_fd->offset; + res_start_offset = ctx->chunks_start_offset; } else { - /* Using a different compression type: Call - * finish_wim_resource_chunk_tab() and it will provide the new - * compressed size. */ - ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fd, - &new_size); - if (ret) - goto out_free_chunk_tab; - if (new_size >= wim_resource_size(lte)) { - /* Oops! We compressed the resource to larger than the original - * size. Write the resource uncompressed instead. */ - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - wim_resource_size(lte), new_size); - ret = seek_and_truncate(out_fd, offset); + 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_XPRESS != 1); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 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 out_free_chunk_tab; - write_ctx.compress = NULL; - write_ctx.doing_sha = false; - out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - goto try_write_again; + goto error; + res_start_offset = chunk_table_offset - sizeof(hdr); + } else { + res_start_offset = chunk_table_offset; } - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; - } - out_res_entry->size = new_size; - ret = 0; -out_free_chunk_tab: - FREE(write_ctx.chunk_tab); - return ret; -} -#ifdef ENABLE_MULTITHREADED_COMPRESSION + ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes, + chunk_table_size, chunk_table_offset); + if (ret) + goto error; + } -/* 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; -}; + *res_start_offset_ret = res_start_offset; + *res_store_size_ret = res_end_offset - res_start_offset; -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 0; -err_destroy_msg_avail_cond: - pthread_cond_destroy(&q->msg_avail_cond); -err_destroy_lock: - pthread_mutex_destroy(&q->lock); -err: - return WIMLIB_ERR_NOMEM; -} -static void -shared_queue_destroy(struct shared_queue *q) -{ - FREE(q->array); - pthread_mutex_destroy(&q->lock); - pthread_cond_destroy(&q->msg_avail_cond); - pthread_cond_destroy(&q->space_avail_cond); +error: + ERROR_WITH_ERRNO("Write error"); + return ret; } -static void -shared_queue_put(struct shared_queue *q, void *obj) +/* Finish writing a WIM resource by writing or updating the chunk table (if not + * writing the data uncompressed) and loading its metadata into @out_reshdr. */ +static int +end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr) { - pthread_mutex_lock(&q->lock); - while (q->filled_slots == q->size) - pthread_cond_wait(&q->space_avail_cond, &q->lock); + int ret; + u64 res_size_in_wim; + u64 res_uncompressed_size; + u64 res_offset_in_wim; - q->back = (q->back + 1) % q->size; - q->array[q->back] = obj; - q->filled_slots++; + wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size || + (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)); + res_uncompressed_size = ctx->cur_write_res_size; - pthread_cond_broadcast(&q->msg_avail_cond); - pthread_mutex_unlock(&q->lock); + if (ctx->compressor) { + ret = end_chunk_table(ctx, res_uncompressed_size, + &res_offset_in_wim, &res_size_in_wim); + if (ret) + return ret; + } else { + res_offset_in_wim = ctx->chunks_start_offset; + res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim; + } + out_reshdr->uncompressed_size = res_uncompressed_size; + out_reshdr->size_in_wim = res_size_in_wim; + out_reshdr->offset_in_wim = res_offset_in_wim; + DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"", + res_uncompressed_size, res_size_in_wim, res_offset_in_wim); + return 0; } -static void * -shared_queue_get(struct shared_queue *q) +/* Begin processing a stream for writing. */ +static int +write_stream_begin_read(struct wim_lookup_table_entry *lte, + u32 flags, void *_ctx) { - void *obj; + struct write_streams_ctx *ctx = _ctx; + int ret; - pthread_mutex_lock(&q->lock); - while (q->filled_slots == 0) - pthread_cond_wait(&q->msg_avail_cond, &q->lock); + wimlib_assert(lte->size > 0); - obj = q->array[q->front]; - q->array[q->front] = NULL; - q->front = (q->front + 1) % q->size; - q->filled_slots--; + ctx->cur_read_stream_offset = 0; + ctx->cur_read_stream_size = lte->size; - pthread_cond_broadcast(&q->space_avail_cond); - pthread_mutex_unlock(&q->lock); - return obj; -} + /* As an optimization, we allow some streams to be "unhashed", meaning + * their SHA1 message digests are unknown. This is the case with + * streams that are added by scanning a directry tree with + * wimlib_add_image(), for example. Since WIM uses single-instance + * streams, we don't know whether such each such stream really need to + * written until it is actually checksummed, unless it has a unique + * size. In such cases we read and checksum the stream in this + * function, thereby advancing ahead of read_stream_list(), which will + * still provide the data again to write_stream_process_chunk(). This + * is okay because an unhashed stream cannot be in a WIM resource, which + * might be costly to decompress. */ + ctx->stream_was_duplicate = false; + if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) { -struct compressor_thread_params { - struct shared_queue *res_to_compress_queue; - struct shared_queue *compressed_res_queue; - compress_func_t compress; -}; + wimlib_assert(!(flags & BEGIN_STREAM_FLAG_PARTIAL_RESOURCE)); -#define MAX_CHUNKS_PER_MSG 2 + struct wim_lookup_table_entry *lte_new; -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]; - size_t total_out_bytes; - unsigned num_chunks; - struct list_head list; - bool complete; - u64 begin_chunk; -}; - -static void -compress_chunks(struct message *msg, compress_func_t compress) -{ - msg->total_out_bytes = 0; - for (unsigned i = 0; i < msg->num_chunks; i++) { - unsigned len = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i]); - 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]; + 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); + ret = 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); + if (ret) + return ret; + return BEGIN_STREAM_STATUS_SKIP_STREAM; + } else { + /* The duplicate stream can validly be written, + * but was not marked as such. Discard the + * current stream entry and use the duplicate, + * but actually freeing the current entry must + * wait until read_stream_list() has finished + * reading its data. */ + DEBUG("Stream duplicate, but not already " + "selected for writing."); + list_replace(<e->write_streams_list, + <e_new->write_streams_list); + list_replace(<e->lookup_table_list, + <e_new->lookup_table_list); + lte_new->out_refcnt = lte->out_refcnt; + lte_new->will_be_in_output_wim = 1; + ctx->stream_was_duplicate = true; + lte = lte_new; + } } - msg->out_chunks[i].iov_base = out_chunk; - msg->out_chunks[i].iov_len = out_len; - msg->total_out_bytes += out_len; } + list_move_tail(<e->write_streams_list, &ctx->pending_streams); + return 0; } -/* Compressor thread routine. This is a lot simpler than the main thread - * routine: just repeatedly get a group of chunks from the - * res_to_compress_queue, compress them, and put them in the - * compressed_res_queue. A NULL pointer indicates that the thread should stop. - * */ -static void * -compressor_thread_proc(void *arg) +/* 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 +write_stream_uncompressed(struct wim_lookup_table_entry *lte, + struct filedes *out_fd) { - struct compressor_thread_params *params = arg; - struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; - struct shared_queue *compressed_res_queue = params->compressed_res_queue; - compress_func_t compress = params->compress; - struct message *msg; + int ret; + u64 begin_offset = lte->out_reshdr.offset_in_wim; + u64 end_offset = out_fd->offset; - DEBUG("Compressor thread ready"); - while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { - compress_chunks(msg, compress); - shared_queue_put(compressed_res_queue, msg); - } - DEBUG("Compressor thread terminating"); - return NULL; -} -#endif /* ENABLE_MULTITHREADED_COMPRESSION */ + if (filedes_seek(out_fd, begin_offset) == -1) + return 0; -static void -do_write_streams_progress(union wimlib_progress_info *progress, - wimlib_progress_func_t progress_func, - uint64_t size_added) -{ - progress->write_streams.completed_bytes += size_added; - progress->write_streams.completed_streams++; - if (progress_func && - progress->write_streams.completed_bytes >= progress->write_streams._private) - { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - if (progress->write_streams._private == progress->write_streams.total_bytes) { - progress->write_streams._private = ~0; - } else { - progress->write_streams._private = - min(progress->write_streams.total_bytes, - progress->write_streams.completed_bytes + - progress->write_streams.total_bytes / 100); + 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; } -} -struct serial_write_stream_ctx { - filedes_t out_fd; - int out_ctype; - int write_resource_flags; -}; + wimlib_assert(out_fd->offset - begin_offset == lte->size); -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); + 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 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. */ +/* Write the next chunk of (typically compressed) data to the output WIM, + * handling the writing of the chunk table. */ 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, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) +write_chunk(struct write_streams_ctx *ctx, const void *cchunk, + size_t csize, size_t usize) { - int ret = 0; + int ret; + 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)) + { + /* Starting to write a new stream in non-packed mode. */ + + 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; + + DEBUG("Writing pipable WIM stream header " + "(offset=%"PRIu64")", ctx->out_fd->offset); - /* For each stream in @stream_list ... */ - while (!list_empty(stream_list)) { - 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_refcnt = lte->out_refcnt; - - ret = hash_unhashed_stream(lte, lookup_table, &tmp); + ret = write_pwm_stream_header(lte, ctx->out_fd, + additional_reshdr_flags); if (ret) - break; - if (tmp != lte) { - lte = tmp; - /* We found a duplicate stream. */ - if (orig_refcnt != tmp->out_refcnt) { - /* We have already written, or are going - * to write, the duplicate stream. So - * just skip to the next stream. */ - DEBUG("Discarding duplicate stream of length %"PRIu64, - wim_resource_size(lte)); - lte->no_progress = 0; - goto skip_to_progress; - } - } + return 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); + ret = begin_write_resource(ctx, lte->size); 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; + return ret; + } + + if (ctx->compressor != NULL) { + /* Record the compresed chunk size. */ + wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks); + ctx->chunk_csizes[ctx->chunk_index++] = csize; + + /* If writing a pipable WIM, before the chunk data write a chunk + * header that provides the compressed chunk size. */ + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + struct pwm_chunk_hdr chunk_hdr = { + .compressed_size = cpu_to_le32(csize), + }; + ret = full_write(ctx->out_fd, &chunk_hdr, + sizeof(chunk_hdr)); + if (ret) + goto error; + } + } + + /* 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++; } - skip_to_progress: - if (!lte->no_progress) { - do_write_streams_progress(progress, - progress_func, - wim_resource_size(lte)); + } 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(<e->write_streams_list); + + wimlib_assert(ctx->cur_write_stream_offset == + ctx->cur_write_res_size); + + ret = end_write_resource(ctx, <e->out_reshdr); + if (ret) + return ret; + + lte->out_reshdr.flags = filter_resource_flags(lte->flags); + if (ctx->compressor != NULL) + lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED; + + 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)) + { + /* 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; + } + wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size); + + ctx->cur_write_stream_offset = 0; } } + + return do_write_streams_progress(&ctx->progress_data, lte, + completed_size, completed_stream_count, + false); + +error: + ERROR_WITH_ERRNO("Write error"); return ret; } static int -do_write_stream_list_serial(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - filedes_t out_fd, - int out_ctype, - int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) -{ - struct serial_write_stream_ctx ctx = { - .out_fd = out_fd, - .out_ctype = out_ctype, - .write_resource_flags = write_resource_flags, - }; - return do_write_stream_list(stream_list, - lookup_table, - serial_write_stream, - &ctx, - progress_func, - progress); -} - -static inline int -write_flags_to_resource_flags(int write_flags) +submit_chunk_for_compression(struct write_streams_ctx *ctx, + const void *chunk, size_t size) { - int resource_flags = 0; + /* While we are unable to submit the chunk for compression (due to too + * many chunks already outstanding), retrieve and write the next + * compressed chunk. */ + while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) { + const void *cchunk; + u32 csize; + u32 usize; + bool bret; + int ret; - if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; - return resource_flags; + bret = ctx->compressor->get_chunk(ctx->compressor, + &cchunk, &csize, &usize); + + 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_list_serial(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - filedes_t out_fd, - int out_ctype, - int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) -{ - DEBUG("Writing stream list (serial version)"); - progress->write_streams.num_threads = 1; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); - return do_write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - write_resource_flags, - progress_func, - progress); -} - -#ifdef ENABLE_MULTITHREADED_COMPRESSION -static int -write_wim_chunks(struct message *msg, filedes_t out_fd, - struct chunk_table *chunk_tab) +write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) { - ssize_t bytes_remaining = msg->total_out_bytes; - struct iovec *vecs = msg->out_chunks; - unsigned nvecs = msg->num_chunks; + struct write_streams_ctx *ctx = _ctx; int ret; + const u8 *chunkptr, *chunkend; - wimlib_assert(nvecs != 0); - wimlib_assert(msg->total_out_bytes != 0); + wimlib_assert(size != 0); - for (unsigned i = 0; i < msg->num_chunks; i++) { - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += vecs[i].iov_len; - } - for (;;) { - ssize_t bytes_written; + 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; - bytes_written = writev(out_fd, vecs, nvecs); - if (bytes_written <= 0) { - if (bytes_written < 0 && errno == EINTR) - continue; - else if (bytes_written == 0) - errno = EIO; - ERROR_WITH_ERRNO("Failed to write WIM chunks"); - ret = WIMLIB_ERR_WRITE; - break; - } - bytes_remaining -= bytes_written; - if (bytes_remaining <= 0) { - ret = 0; - break; + 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); } - while (bytes_written >= 0) { - wimlib_assert(nvecs != 0); - if (bytes_written >= vecs[0].iov_len) { - vecs++; - nvecs--; - bytes_written -= vecs[0].iov_len; + + 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 { - vecs[0].iov_base += bytes_written; - vecs[0].iov_len -= bytes_written; - bytes_written = 0; + break; } + } - } - return ret; -} -struct main_writer_thread_ctx { - struct list_head *stream_list; - struct wim_lookup_table *lookup_table; - filedes_t out_fd; - int out_ctype; - int write_resource_flags; - struct shared_queue *res_to_compress_queue; - struct shared_queue *compressed_res_queue; - size_t num_messages; - wimlib_progress_func_t progress_func; - union wimlib_progress_info *progress; - - 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; -}; + ret = submit_chunk_for_compression(ctx, resized_chunk, + needed_chunk_size); + if (ret) + return ret; -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; - } + } while (chunkptr != chunkend); return 0; } -static void -destroy_message(struct message *msg) +/* Finish processing a stream for writing. It may not have been completely + * written yet, as the chunk_compressor implementation may still have chunks + * buffered or being compressed. */ +static int +write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx) { - for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { - FREE(msg->compressed_chunks[i]); - FREE(msg->uncompressed_chunks[i]); - } + 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 -free_messages(struct message *msgs, size_t num_messages) +compute_stream_list_stats(struct list_head *stream_list, + struct write_streams_ctx *ctx) { - if (msgs) { - for (size_t i = 0; i < num_messages; i++) - destroy_message(&msgs[i]); - FREE(msgs); - } -} - -static struct message * -allocate_messages(size_t num_messages) -{ - struct message *msgs; + struct wim_lookup_table_entry *lte; + u64 total_bytes = 0; + u64 num_streams = 0; + u64 total_parts = 0; + WIMStruct *prev_wim_part = NULL; - 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; + 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++; + } } } - 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); + 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; } -static int -main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) +/* 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 them + * to @raw_copy_streams. Return the total uncompressed size of the streams that + * need to be compressed. */ +static u64 +find_raw_copy_streams(struct list_head *stream_list, + int write_resource_flags, + int out_ctype, + u32 out_chunk_size, + struct list_head *raw_copy_streams) { - /* 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; - - /* Initially, all the messages are available to use. */ - INIT_LIST_HEAD(&ctx->available_msgs); - for (size_t i = 0; i < ctx->num_messages; i++) - list_add_tail(&ctx->msgs[i].list, &ctx->available_msgs); - - /* outstanding_streams is the list of streams that currently have had - * chunks sent off for compression. - * - * The first stream in outstanding_streams is the stream that is - * currently being written. - * - * The last stream in outstanding_streams is the stream that is - * currently being read and having chunks fed to the compressor threads. - * */ - INIT_LIST_HEAD(&ctx->outstanding_streams); - ctx->num_outstanding_messages = 0; + struct wim_lookup_table_entry *lte, *tmp; + u64 num_bytes_to_compress = 0; - ctx->next_msg = NULL; + INIT_LIST_HEAD(raw_copy_streams); - /* Resources that don't need any chunks compressed are added to this - * list and written directly by the main thread. */ - INIT_LIST_HEAD(&ctx->serial_streams); + /* Initialize temporary raw_copy_ok flag. */ + list_for_each_entry(lte, stream_list, write_streams_list) + if (lte->resource_location == RESOURCE_IN_WIM) + lte->rspec->raw_copy_ok = 0; - ctx->cur_chunk_tab = NULL; + list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) { + if (lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->raw_copy_ok) + { + list_move_tail(<e->write_streams_list, + raw_copy_streams); + } 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_streams); + } else { + num_bytes_to_compress += lte->size; + } + } - return 0; + return num_bytes_to_compress; } +/* Copy a raw compressed resource located in another WIM file to the WIM file + * being written. */ static int -receive_compressed_chunks(struct main_writer_thread_ctx *ctx) +write_raw_copy_resource(struct wim_resource_spec *in_rspec, + struct filedes *out_fd) { - struct message *msg; - struct wim_lookup_table_entry *cur_lte; + u64 cur_read_offset; + u64 end_read_offset; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; int ret; + struct filedes *in_fd; + struct wim_lookup_table_entry *lte; + u64 out_offset_in_wim; - wimlib_assert(!list_empty(&ctx->outstanding_streams)); - wimlib_assert(ctx->num_outstanding_messages != 0); - - cur_lte = container_of(ctx->outstanding_streams.next, - struct wim_lookup_table_entry, - being_compressed_list); - - /* Get the next message from the queue and process it. - * The message will contain 1 or more data chunks that have been - * compressed. */ - msg = shared_queue_get(ctx->compressed_res_queue); - msg->complete = true; - --ctx->num_outstanding_messages; - - /* Is this the next chunk in the current resource? If it's not - * (i.e., an earlier chunk in a same or different resource - * hasn't been compressed yet), do nothing, and keep this - * message around until all earlier chunks are received. - * - * Otherwise, write all the chunks we can. */ - while (cur_lte != NULL && - !list_empty(&cur_lte->msg_list) - && (msg = container_of(cur_lte->msg_list.next, - struct message, - list))->complete) - { - list_move(&msg->list, &ctx->available_msgs); - if (msg->begin_chunk == 0) { - /* This is the first set of chunks. Leave space - * for the chunk table in the output file. */ - off_t cur_offset = filedes_offset(ctx->out_fd); - if (cur_offset == -1) - return WIMLIB_ERR_WRITE; - ret = begin_wim_resource_chunk_tab(cur_lte, - ctx->out_fd, - cur_offset, - &ctx->cur_chunk_tab); - if (ret) - return ret; - } - - /* Write the compressed chunks from the message. */ - ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab); - 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; - off_t offset; + DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", " + "uncompressed_size=%"PRIu64")", + in_rspec->size_in_wim, in_rspec->uncompressed_size); - ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, - ctx->out_fd, - &res_csize); - if (ret) - return ret; + /* Copy the raw data. */ + cur_read_offset = in_rspec->offset_in_wim; + end_read_offset = cur_read_offset + in_rspec->size_in_wim; - list_del(&cur_lte->being_compressed_list); + out_offset_in_wim = out_fd->offset; - /* Grab the offset of this stream in the output file - * from the chunk table before we free it. */ - offset = ctx->cur_chunk_tab->file_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 { - FREE(ctx->cur_chunk_tab); - ctx->cur_chunk_tab = NULL; + bytes_to_read = min(sizeof(buf), end_read_offset - cur_read_offset); - if (res_csize >= wim_resource_size(cur_lte)) { - /* Oops! We compressed the resource to - * larger than the original size. Write - * the resource uncompressed instead. */ - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - wim_resource_size(cur_lte), res_csize); - ret = seek_and_truncate(ctx->out_fd, 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); - if (ret) - return ret; - } else { - cur_lte->output_resource_entry.size = - res_csize; + ret = full_pread(in_fd, buf, bytes_to_read, cur_read_offset); + if (ret) + return ret; - cur_lte->output_resource_entry.original_size = - cur_lte->resource_entry.original_size; + ret = full_write(out_fd, buf, bytes_to_read); + if (ret) + return ret; - cur_lte->output_resource_entry.offset = - offset; + cur_read_offset += bytes_to_read; - cur_lte->output_resource_entry.flags = - cur_lte->resource_entry.flags | - WIM_RESHDR_FLAG_COMPRESSED; - } + } while (cur_read_offset != end_read_offset); - do_write_streams_progress(ctx->progress, - ctx->progress_func, - wim_resource_size(cur_lte)); - - /* 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->write_resource_flags, - ctx->progress_func, - ctx->progress); - if (ret) - return ret; - } + list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) { + if (lte->will_be_in_output_wim) { + stream_set_out_reshdr_for_reuse(lte); + if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + lte->out_res_offset_in_wim = out_offset_in_wim; + else + lte->out_reshdr.offset_in_wim = out_offset_in_wim; - /* Advance to the next stream to write. */ - if (list_empty(&ctx->outstanding_streams)) { - cur_lte = NULL; - } else { - cur_lte = container_of(ctx->outstanding_streams.next, - struct wim_lookup_table_entry, - being_compressed_list); - } } } return 0; } -/* Called when the main thread has read a new chunk of data. */ +/* Copy a list of raw compressed resources located in other WIM file(s) to the + * WIM file being written. */ static int -main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) +write_raw_copy_resources(struct list_head *raw_copy_streams, + struct filedes *out_fd, + struct write_streams_progress_data *progress_data) { - struct main_writer_thread_ctx *ctx = _ctx; + struct wim_lookup_table_entry *lte; int ret; - struct message *next_msg; - u64 next_chunk_in_msg; - - /* Update SHA1 message digest for the stream currently being read by the - * main thread. */ - sha1_update(&ctx->next_sha_ctx, chunk, chunk_size); - - /* We send chunks of data to the compressor chunks in batches which we - * refer to as "messages". @next_msg is the message that is currently - * being prepared to send off. If it is NULL, that indicates that we - * need to start a new message. */ - next_msg = ctx->next_msg; - if (!next_msg) { - /* We need to start a new message. First check to see if there - * is a message available in the list of available messages. If - * so, we can just take one. If not, all the messages (there is - * a fixed number of them, proportional to the number of - * threads) have been sent off to the compressor threads, so we - * receive messages from the compressor threads containing - * compressed chunks of data. - * - * We may need to receive multiple messages before one is - * actually available to use because messages received that are - * *not* for the very next set of chunks to compress must be - * buffered until it's time to write those chunks. */ - while (list_empty(&ctx->available_msgs)) { - ret = receive_compressed_chunks(ctx); + + list_for_each_entry(lte, raw_copy_streams, write_streams_list) + lte->rspec->raw_copy_ok = 1; + + list_for_each_entry(lte, raw_copy_streams, write_streams_list) { + if (lte->rspec->raw_copy_ok) { + /* Write each packed resource only one time, no matter + * how many streams reference it. */ + ret = write_raw_copy_resource(lte->rspec, out_fd); if (ret) return ret; + lte->rspec->raw_copy_ok = 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; + ret = do_write_streams_progress(progress_data, lte, lte->size, + 1, false); + if (ret) + return ret; } return 0; } +/* Wait for and write all chunks pending in the compressor. */ static int -main_writer_thread_finish(void *_ctx) +finish_remaining_chunks(struct write_streams_ctx *ctx) { - struct main_writer_thread_ctx *ctx = _ctx; + const void *cdata; + u32 csize; + u32 usize; int ret; - while (ctx->num_outstanding_messages != 0) { - ret = receive_compressed_chunks(ctx); + + if (ctx->compressor == NULL) + return 0; + + if (ctx->chunk_buf_filled != 0) { + ret = submit_chunk_for_compression(ctx, ctx->chunk_buf, + ctx->chunk_buf_filled); + if (ret) + return ret; + } + + while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) { + ret = write_chunk(ctx, cdata, csize, usize); if (ret) return ret; } - wimlib_assert(list_empty(&ctx->outstanding_streams)); - return do_write_stream_list_serial(&ctx->serial_streams, - ctx->lookup_table, - ctx->out_fd, - ctx->out_ctype, - ctx->write_resource_flags, - ctx->progress_func, - ctx->progress); + return 0; } +static void +remove_zero_length_streams(struct list_head *stream_list) +{ + 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); + } + } +} + +/* + * 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. + * + * 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 -submit_stream_for_compression(struct wim_lookup_table_entry *lte, - struct main_writer_thread_ctx *ctx) +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 progfunc, + void *progctx) { int ret; + struct write_streams_ctx ctx; + struct list_head raw_copy_streams; - /* 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 == 0) { - wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); - ret = finalize_and_check_sha1(&ctx->next_sha_ctx, lte); + 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; + + 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 { + 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.progfunc = progfunc; + ctx.progress_data.progctx = progctx; + + ctx.num_bytes_to_compress = find_raw_copy_streams(stream_list, + write_resource_flags, + out_ctype, + out_chunk_size, + &raw_copy_streams); + + 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 than a heuristic value. */ + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + + #ifdef ENABLE_MULTITHREADED_COMPRESSION + if (ctx.num_bytes_to_compress > max(2000000, out_chunk_size)) { + 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); + + ret = call_progress(ctx.progress_data.progfunc, + WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + &ctx.progress_data.progress, + ctx.progress_data.progctx); + if (ret) + goto out_destroy_context; + + 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; + } + + /* Read the list of streams needing to be compressed, using the + * specified callbacks to execute processing of the data. */ + + struct read_stream_list_callbacks cbs = { + .begin_stream = write_stream_begin_read, + .begin_stream_ctx = &ctx, + .consume_chunk = write_stream_process_chunk, + .consume_chunk_ctx = &ctx, + .end_stream = write_stream_end_read, + .end_stream_ctx = &ctx, + }; + + ret = read_stream_list(stream_list, + offsetof(struct wim_lookup_table_entry, write_streams_list), + &cbs, + STREAM_LIST_ALREADY_SORTED | + VERIFY_STREAM_HASHES | + COMPUTE_MISSING_STREAM_HASHES); + + if (ret) + goto out_destroy_context; + + ret = finish_remaining_chunks(&ctx); + if (ret) + goto out_destroy_context; + + if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + struct wim_reshdr reshdr; + struct wim_lookup_table_entry *lte; + u64 offset_in_res; + + ret = end_write_resource(&ctx, &reshdr); + if (ret) + goto out_destroy_context; + + DEBUG("Ending packed resource: %lu %lu %lu.", + reshdr.offset_in_wim, + reshdr.size_in_wim, + reshdr.uncompressed_size); + + offset_in_res = 0; + list_for_each_entry(lte, &ctx.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; + } + 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_streams, 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; } static int -main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) +is_stream_packed(struct wim_lookup_table_entry *lte, void *_ignore) { - struct main_writer_thread_ctx *ctx = _ctx; - int ret; - - if (wim_resource_size(lte) < 1000 || - ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || - (lte->resource_location == RESOURCE_IN_WIM && - !(ctx->write_resource_flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && - wimlib_get_compression_type(lte->wim) == 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); - } - lte->no_progress = 1; - return ret; + return lte_is_partial(lte); } -static long -get_default_num_threads() +static bool +wim_has_packed_streams(WIMStruct *wim) { -#ifdef __WIN32__ - return win32_get_number_of_processors(); -#else - return sysconf(_SC_NPROCESSORS_ONLN); -#endif + return for_lookup_table_entry(wim->lookup_table, is_stream_packed, NULL); } -/* 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*. - */ static int -write_stream_list_parallel(struct list_head *stream_list, - struct wim_lookup_table *lookup_table, - filedes_t out_fd, - int out_ctype, - int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress, - unsigned num_threads) +wim_write_stream_list(WIMStruct *wim, + struct list_head *stream_list, + int write_flags, + unsigned num_threads, + struct filter_context *filter_ctx) { - int ret; - struct shared_queue res_to_compress_queue; - struct shared_queue compressed_res_queue; - pthread_t *compressor_threads = NULL; - - 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; - } else { - num_threads = nthreads; - } - } - - DEBUG("Writing stream list (parallel version, num_threads=%u)", - num_threads); - - progress->write_streams.num_threads = num_threads; - - static const size_t MESSAGES_PER_THREAD = 2; - size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); - - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); - - ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret) - goto out_serial; - - ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret) - goto out_destroy_res_to_compress_queue; - - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); + int out_ctype; + u32 out_chunk_size; + int write_resource_flags; - compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); - if (!compressor_threads) { - ret = WIMLIB_ERR_NOMEM; - goto out_destroy_compressed_res_queue; - } + write_resource_flags = write_flags_to_resource_flags(write_flags); - 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); - 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; - } + /* wimlib v1.7.0: pack streams by default if the WIM version has been + * set to WIM_VERSION_PACKED_STREAMS and at least one stream in the + * WIM's lookup table is located in a packed resource (may be the same + * WIM, or a different one in the case of export). */ + if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS && + wim_has_packed_streams(wim)) + { + write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS; } - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); - - 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.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_func = progress_func; - ctx.progress = progress; - 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_func, progress); - 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_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, - write_resource_flags, - progress_func, - progress); + return write_stream_list(stream_list, + &wim->out_fd, + write_resource_flags, + out_ctype, + out_chunk_size, + num_threads, + wim->lookup_table, + filter_ctx, + wim->progfunc, + wim->progctx); } -#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, - filedes_t out_fd, int out_ctype, 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, + 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; - union wimlib_progress_info progress; int ret; - int write_resource_flags; + struct wim_lookup_table_entry *lte; - if (list_empty(stream_list)) - return 0; + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ - write_resource_flags = write_flags_to_resource_flags(write_flags); + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; - /* 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_RESOURCE_FLAG_RECOMPRESS))) - { - total_compression_bytes += wim_resource_size(lte); - } + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = (void*)buf; + lte->size = buf_size; + lte->flags = reshdr_flags; + + if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + sha1_buffer(buf, buf_size, lte->hash); + lte->unhashed = 0; + } else { + lte->unhashed = 1; } - progress.write_streams.total_bytes = total_bytes; - progress.write_streams.total_streams = num_streams; - progress.write_streams.completed_bytes = 0; - progress.write_streams.completed_streams = 0; - progress.write_streams.num_threads = num_threads; - progress.write_streams.compression_type = out_ctype; - progress.write_streams._private = 0; - -#ifdef ENABLE_MULTITHREADED_COMPRESSION - if (total_compression_bytes >= 1000000 && num_threads != 1) - ret = write_stream_list_parallel(stream_list, - lookup_table, - out_fd, - out_ctype, - write_resource_flags, - progress_func, - &progress, - num_threads); - else -#endif - ret = write_stream_list_serial(stream_list, - lookup_table, - out_fd, - out_ctype, - write_resource_flags, - progress_func, - &progress); + + 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; } @@ -1461,7 +1678,7 @@ 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; @@ -1482,10 +1699,10 @@ 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; @@ -1497,481 +1714,1206 @@ stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) return 0; } - -struct lte_overwrite_prepare_args { +struct find_streams_ctx { WIMStruct *wim; - off_t end_offset; + int write_flags; struct list_head stream_list; struct stream_size_table stream_size_tab; }; -/* First phase of preparing streams for an in-place overwrite. This is called - * on all streams, both hashed and unhashed, except the metadata resources. */ +static void +reference_stream_for_write(struct wim_lookup_table_entry *lte, + struct list_head *stream_list, u32 nref) +{ + 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 +fully_reference_stream_for_write(struct wim_lookup_table_entry *lte, + void *_stream_list) +{ + 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 -lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *_args) +inode_find_streams_to_reference(const struct wim_inode *inode, + const struct wim_lookup_table *table, + struct list_head *stream_list) { - struct lte_overwrite_prepare_args *args = _args; + struct wim_lookup_table_entry *lte; + unsigned i; + + wimlib_assert(inode->i_nlink > 0); - wimlib_assert(!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)); - if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) - list_add_tail(<e->write_streams_list, &args->stream_list); - lte->out_refcnt = lte->refcnt; - stream_size_table_insert(lte, &args->stream_size_tab); + for (i = 0; i <= inode->i_num_ads; i++) { + lte = inode_stream_lte(inode, i, table); + if (lte) + 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; + } return 0; } -/* Second phase of preparing streams for an in-place overwrite. This is called - * on existing metadata resources and hashed streams, but not unhashed streams. - * - * NOTE: lte->output_resource_entry is in union with lte->hash_list_2, so - * lte_overwrite_prepare_2() must be called after lte_overwrite_prepare(), as - * the latter uses lte->hash_list_2, while the former expects to set - * lte->output_resource_entry. */ static int -lte_overwrite_prepare_2(struct wim_lookup_table_entry *lte, void *_args) +do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore) { - struct lte_overwrite_prepare_args *args = _args; + lte->will_be_in_output_wim = 0; + return 0; +} - if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == args->wim) { - /* We can't do an in place overwrite on the WIM if there are - * streams after the XML data. */ - if (lte->resource_entry.offset + - lte->resource_entry.size > args->end_offset) - { - #ifdef ENABLE_ERROR_MESSAGES - ERROR("The following resource is after the XML data:"); - print_lookup_table_entry(lte, stderr); - #endif - return WIMLIB_ERR_RESOURCE_ORDER; - } - copy_resource_entry(<e->output_resource_entry, - <e->resource_entry); +static int +image_find_streams_to_reference(WIMStruct *wim) +{ + struct wim_image_metadata *imd; + struct wim_inode *inode; + struct wim_lookup_table_entry *lte; + struct list_head *stream_list; + int ret; + + imd = wim_get_current_image_metadata(wim); + + image_for_each_unhashed_stream(lte, imd) + lte->will_be_in_output_wim = 0; + + stream_list = wim->private; + image_for_each_inode(inode, imd) { + ret = inode_find_streams_to_reference(inode, + wim->lookup_table, + stream_list); + if (ret) + return ret; } return 0; } -/* Given a WIM that we are going to overwrite in place with zero or more - * additional streams added, construct a list the list of new unique streams - * ('struct wim_lookup_table_entry's) that must be written, plus any unhashed - * streams that need to be added but may be identical to other hashed or - * unhashed streams. These unhashed streams are checksummed while the streams - * are being written. To aid this process, the member @unique_size is set to 1 - * on streams that have a unique size and therefore must be written. - * - * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to - * indicate the number of times the stream is referenced in only the streams - * that are being written; this may still be adjusted later when unhashed - * streams are being resolved. - */ static int -prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset, - struct list_head *stream_list) +prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim, + int image, + int streams_ok, + struct list_head *stream_list_ret) { int ret; - struct lte_overwrite_prepare_args args; - unsigned i; - args.wim = wim; - args.end_offset = end_offset; - ret = init_stream_size_table(&args.stream_size_tab, - wim->lookup_table->capacity); - if (ret) - return ret; + INIT_LIST_HEAD(stream_list_ret); - INIT_LIST_HEAD(&args.stream_list); - for (i = 0; i < wim->hdr.image_count; i++) { - struct wim_image_metadata *imd; + if (streams_ok && (image == WIMLIB_ALL_IMAGES || + (image == 1 && wim->hdr.image_count == 1))) + { + /* Fast case: Assume that all streams are being written and + * that the reference counts are correct. */ struct wim_lookup_table_entry *lte; + struct wim_image_metadata *imd; + unsigned i; - imd = wim->image_metadata[i]; - image_for_each_unhashed_stream(lte, imd) - lte_overwrite_prepare(lte, &args); - } - for_lookup_table_entry(wim->lookup_table, lte_overwrite_prepare, &args); - list_transfer(&args.stream_list, stream_list); + for_lookup_table_entry(wim->lookup_table, + fully_reference_stream_for_write, + stream_list_ret); - for (i = 0; i < wim->hdr.image_count; i++) { - ret = lte_overwrite_prepare_2(wim->image_metadata[i]->metadata_lte, - &args); + for (i = 0; i < wim->hdr.image_count; i++) { + imd = wim->image_metadata[i]; + image_for_each_unhashed_stream(lte, imd) + 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, + 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) - goto out_destroy_stream_size_table; + return ret; } - ret = for_lookup_table_entry(wim->lookup_table, - lte_overwrite_prepare_2, &args); -out_destroy_stream_size_table: - destroy_stream_size_table(&args.stream_size_tab); - return ret; -} + return 0; +} -struct find_streams_ctx { - struct list_head stream_list; - struct stream_size_table stream_size_tab; +struct insert_other_if_hard_filtered_ctx { + struct stream_size_table *tab; + struct filter_context *filter_ctx; }; -static void -inode_find_streams_to_write(struct wim_inode *inode, - struct wim_lookup_table *table, - struct list_head *stream_list, - struct stream_size_table *tab) +static int +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 +determine_stream_size_uniquity(struct list_head *stream_list, + struct wim_lookup_table *lt, + struct filter_context *filter_ctx) { + int ret; + struct stream_size_table tab; struct wim_lookup_table_entry *lte; - for (unsigned i = 0; i <= inode->i_num_ads; i++) { - lte = inode_stream_lte(inode, i, table); - if (lte) { - if (lte->out_refcnt == 0) { - if (lte->unhashed) - stream_size_table_insert(lte, tab); - list_add_tail(<e->write_streams_list, stream_list); + + ret = init_stream_size_table(&tab, 9001); + 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; +} + +static void +filter_stream_list_for_write(struct list_head *stream_list, + struct filter_context *filter_ctx) +{ + struct wim_lookup_table_entry *lte, *tmp; + + list_for_each_entry_safe(lte, tmp, + stream_list, write_streams_list) + { + int status = stream_filtered(lte, filter_ctx); + + 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); } - lte->out_refcnt += inode->i_nlink; + list_del(<e->write_streams_list); } } } +/* + * 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 returned in neither @stream_list_ret nor + * @lookup_table_list_ret. + * + * @stream_list_ret + * List of streams, linked by write_streams_list, that need to be written + * will be returned here. + * + * Note that this function assumes that unhashed streams will be written; + * it does not take into account that they may become duplicates when + * actually hashed. + * + * @lookup_table_list_ret + * List of streams, linked by lookup_table_list, that need to be included + * in the WIM's lookup table will be returned here. This will be a + * superset of the streams in @stream_list_ret. + * + * This list will be a proper superset of @stream_list_ret if and only if + * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of + * the streams that would otherwise need to be written were already located + * in the WIM file. + * + * All streams in this list will have @out_refcnt set to the number of + * references to the stream in the output WIM. If + * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt + * may be as low as 0. + * + * @filter_ctx_ret + * A context for queries of stream filter status with stream_filtered() is + * returned in this location. + * + * In addition, @will_be_in_output_wim will be set to 1 in all stream entries + * inserted into @lookup_table_list_ret and to 0 in all stream entries in the + * lookup table of @wim not inserted into @lookup_table_list_ret. + * + * Still furthermore, @unique_size will be set to 1 on all stream entries in + * @stream_list_ret that have unique size among all stream entries in + * @stream_list_ret and among all stream entries in the lookup table of @wim + * that are ineligible for being written due to filtering. + * + * Returns 0 on success; nonzero on read error, memory allocation error, or + * otherwise. + */ static int -image_find_streams_to_write(WIMStruct *w) +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) { - struct find_streams_ctx *ctx; - struct wim_image_metadata *imd; - struct wim_inode *inode; + int ret; struct wim_lookup_table_entry *lte; - ctx = w->private; - imd = wim_get_current_image_metadata(w); + filter_ctx_ret->write_flags = write_flags; + filter_ctx_ret->wim = wim; - image_for_each_unhashed_stream(lte, imd) - lte->out_refcnt = 0; + 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; + + 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 = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table, + filter_ctx_ret); + if (ret) + return ret; + + if (may_filter_streams(filter_ctx_ret)) + filter_stream_list_for_write(stream_list_ret, filter_ctx_ret); - /* Go through this image's inodes to find any streams that have not been - * found yet. */ - image_for_each_inode(inode, imd) { - inode_find_streams_to_write(inode, w->lookup_table, - &ctx->stream_list, - &ctx->stream_size_tab); - } return 0; } -/* Given a WIM that from which one or all of the images is being written, build - * the list of unique streams ('struct wim_lookup_table_entry's) that must be - * written, plus any unhashed streams that need to be written but may be - * identical to other hashed or unhashed streams being written. These unhashed - * streams are checksummed while the streams are being written. To aid this - * process, the member @unique_size is set to 1 on streams that have a unique - * size and therefore must be written. - * - * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to - * indicate the number of times the stream is referenced in only the streams - * that are being written; this may still be adjusted later when unhashed - * streams are being resolved. - */ static int -prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) +write_wim_streams(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + 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; + 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 { + /* Currently only as a result of wimlib_split() being called: + * 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->will_be_in_output_wim = 1; + lte->unique_size = 0; + list_add_tail(<e->lookup_table_list, lookup_table_list_ret); + } + } + + return wim_write_stream_list(wim, + stream_list, + write_flags, + num_threads, + filter_ctx); +} + +static int +write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags) { int ret; - struct find_streams_ctx ctx; + int start_image; + int end_image; + int write_resource_flags; + + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) { + DEBUG("Not writing any metadata resources."); + return 0; + } + + write_resource_flags = write_flags_to_resource_flags(write_flags); - for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); - ret = init_stream_size_table(&ctx.stream_size_tab, - wim->lookup_table->capacity); + write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS; + + DEBUG("Writing metadata resources (offset=%"PRIu64")", + wim->out_fd.offset); + + ret = call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, + NULL, wim->progctx); if (ret) return ret; - for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, - &ctx.stream_size_tab); - INIT_LIST_HEAD(&ctx.stream_list); - wim->private = &ctx; - ret = for_image(wim, image, image_find_streams_to_write); - destroy_stream_size_table(&ctx.stream_size_tab); - if (ret == 0) - list_transfer(&ctx.stream_list, stream_list); + + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; + } else { + start_image = image; + end_image = image; + } + + for (int i = start_image; i <= end_image; i++) { + struct wim_image_metadata *imd; + + imd = wim->image_metadata[i - 1]; + /* Build a new metadata resource only if image was modified from + * the original (or was newly added). Otherwise just copy the + * existing one. */ + if (imd->modified) { + DEBUG("Image %u was modified; building and writing new " + "metadata resource", i); + ret = write_metadata_resource(wim, i, + write_resource_flags); + } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { + DEBUG("Image %u was not modified; re-using existing " + "metadata resource.", i); + stream_set_out_reshdr_for_reuse(imd->metadata_lte); + ret = 0; + } else { + DEBUG("Image %u was not modified; copying existing " + "metadata resource.", i); + ret = write_wim_resource(imd->metadata_lte, + &wim->out_fd, + wim->out_compression_type, + wim->out_chunk_size, + write_resource_flags); + } + if (ret) + return ret; + } + + return call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, + NULL, wim->progctx); +} + +static int +open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) +{ + int raw_fd; + DEBUG("Opening \"%"TS"\" for writing.", path); + + raw_fd = topen(path, open_flags | O_BINARY, 0644); + if (raw_fd < 0) { + ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path); + return WIMLIB_ERR_OPEN; + } + filedes_init(&wim->out_fd, raw_fd); + return 0; +} + +static int +close_wim_writable(WIMStruct *wim, int write_flags) +{ + int ret = 0; + + if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) { + DEBUG("Closing WIM file."); + if (filedes_valid(&wim->out_fd)) + if (filedes_close(&wim->out_fd)) + ret = WIMLIB_ERR_WRITE; + } + filedes_invalidate(&wim->out_fd); return ret; } -/* Writes the streams for the specified @image in @wim to @wim->out_fp. - */ static int -write_wim_streams(WIMStruct *wim, int image, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +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; - struct list_head stream_list; - ret = prepare_stream_list(wim, image, &stream_list); + /* 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; - return write_stream_list(&stream_list, - wim->lookup_table, - wim->out_fd, - wimlib_get_compression_type(wim), - write_flags, - num_threads, - progress_func); + + /* 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(): + * * Finish writing a WIM file: write the lookup table, xml data, and integrity - * table (optional), then overwrite the WIM header. + * table, then overwrite the WIM header. By default, closes the WIM file + * descriptor (@wim->out_fd) if successful. * * write_flags is a bitwise OR of the following: * - * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: - * Include an integrity table. - * - * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS: - * Show progress information when (if) writing the integrity table. + * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: + * Include an integrity table. * - * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: - * Don't write the lookup table. + * (public) WIMLIB_WRITE_FLAG_FSYNC: + * fsync() the output file before closing it. * - * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: - * When (if) writing the integrity table, re-use entries from the - * existing integrity table, if possible. + * (public) WIMLIB_WRITE_FLAG_PIPABLE: + * Writing a pipable WIM, possibly to a pipe; include pipable WIM + * stream headers before the lookup table and XML data, and also + * write the WIM header at the end instead of seeking to the + * beginning. Can't be combined with + * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY. * - * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: - * After writing the XML data but before writing the integrity - * table, write a temporary WIM header and flush the stream so that - * the WIM is less likely to become corrupted upon abrupt program - * termination. + * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: + * Don't write the lookup table. * - * (private) WIMLIB_WRITE_FLAG_FSYNC: - * fsync() the output file before closing it. + * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: + * When (if) writing the integrity table, re-use entries from the + * existing integrity table, if possible. * + * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: + * After writing the XML data but before writing the integrity + * table, write a temporary WIM header and flush the stream so that + * the WIM is less likely to become corrupted upon abrupt program + * termination. + * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END: + * Instead of overwriting the WIM header at the beginning of the + * file, simply append it to the end of the file. (Used when + * writing to pipe.) + * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR: + * Do not close the file descriptor @wim->out_fd on either success + * on failure. + * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES: + * Use the existing stored in the in-memory XML + * information, rather than setting it to the offset of the XML + * data being written. */ -int -finish_write(WIMStruct *w, int image, int write_flags, - wimlib_progress_func_t progress_func) +static int +finish_write(WIMStruct *wim, int image, int write_flags, + struct list_head *lookup_table_list) { int ret; - struct wim_header hdr; + off_t hdr_offset; + int write_resource_flags; + off_t old_lookup_table_end; + off_t new_lookup_table_end; + u64 xml_totalbytes; - /* @hdr will be the header for the new WIM. First copy all the data - * from the header in the WIMStruct; then set all the fields that may - * have changed, including the resource entries, boot index, and image - * count. */ - memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); + DEBUG("image=%d, write_flags=%08x", image, write_flags); - /* Set image count and boot index correctly for single image writes */ - if (image != WIMLIB_ALL_IMAGES) { - hdr.image_count = 1; - if (hdr.boot_idx == image) - hdr.boot_idx = 1; - else - hdr.boot_idx = 0; - } + write_resource_flags = write_flags_to_resource_flags(write_flags); /* In the WIM header, there is room for the resource entry for a * metadata resource labeled as the "boot metadata". This entry should * be zeroed out if there is no bootable image (boot_idx 0). Otherwise, * it should be a copy of the resource entry for the image that is * marked as bootable. This is not well documented... */ - if (hdr.boot_idx == 0) { - zero_resource_entry(&hdr.boot_metadata_res_entry); + if (wim->hdr.boot_idx == 0) { + zero_reshdr(&wim->hdr.boot_metadata_reshdr); } else { - copy_resource_entry(&hdr.boot_metadata_res_entry, - &w->image_metadata[ 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_reshdr.offset_in_wim + + wim->hdr.lookup_table_reshdr.size_in_wim; if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); + ret = write_wim_lookup_table(wim, image, write_flags, + &wim->hdr.lookup_table_reshdr, + lookup_table_list); if (ret) - goto out_close_wim; + return ret; } - ret = write_xml_data(w->wim_info, image, w->out_fd, - (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? - wim_info_get_total_bytes(w->wim_info) : 0, - &hdr.xml_res_entry); + /* Write XML data. */ + xml_totalbytes = wim->out_fd.offset; + 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_data_reshdr, + write_resource_flags); if (ret) - goto out_close_wim; + return ret; + /* Write integrity table (optional). */ if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { struct wim_header checkpoint_hdr; - memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); - zero_resource_entry(&checkpoint_hdr.integrity); - ret = write_header(&checkpoint_hdr, w->out_fd); + memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header)); + 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); if (ret) - goto out_close_wim; + return ret; } - off_t old_lookup_table_end; - off_t new_lookup_table_end; - if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { - old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; - } else { + if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) old_lookup_table_end = 0; - } - new_lookup_table_end = hdr.lookup_table_res_entry.offset + - hdr.lookup_table_res_entry.size; - ret = write_integrity_table(w->out_fd, - &hdr.integrity, + 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, - old_lookup_table_end, - progress_func); + old_lookup_table_end); if (ret) - goto out_close_wim; + return ret; } else { - zero_resource_entry(&hdr.integrity); - } - - ret = write_header(&hdr, w->out_fd); + /* No integrity table. */ + zero_reshdr(&wim->hdr.integrity_table_reshdr); + } + + /* Now that all information in the WIM header has been determined, the + * preliminary header written earlier can be overwritten, the header of + * the existing WIM file can be overwritten, or the final header can be + * written to the end of the pipable WIM. */ + wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; + hdr_offset = 0; + if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END) + hdr_offset = wim->out_fd.offset; + DEBUG("Writing new header @ %"PRIu64".", hdr_offset); + ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset); if (ret) - goto out_close_wim; + return ret; + /* Possibly sync file data to disk before closing. On POSIX systems, it + * is necessary to do this before using rename() to overwrite an + * existing file with a new file. Otherwise, data loss would occur if + * the system is abruptly terminated when the metadata for the rename + * operation has been written to disk, but the new file data has not. + */ if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { - if (fsync(w->out_fd)) { + DEBUG("Syncing WIM file."); + if (fsync(wim->out_fd.fd)) { ERROR_WITH_ERRNO("Error syncing data to WIM file"); - ret = WIMLIB_ERR_WRITE; + return WIMLIB_ERR_WRITE; } } -out_close_wim: - if (close(w->out_fd)) { + + if (close_wim_writable(wim, write_flags)) { ERROR_WITH_ERRNO("Failed to close the output WIM file"); - if (ret == 0) - ret = WIMLIB_ERR_WRITE; + return WIMLIB_ERR_WRITE; } - w->out_fd = INVALID_FILEDES; - return ret; + + return 0; } #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) int -lock_wim(WIMStruct *w, filedes_t fd) +lock_wim(WIMStruct *wim, int fd) { int ret = 0; - if (fd != INVALID_FILEDES && !w->wim_locked) { + if (fd != -1 && !wim->wim_locked) { ret = flock(fd, LOCK_EX | LOCK_NB); if (ret != 0) { if (errno == EWOULDBLOCK) { ERROR("`%"TS"' is already being modified or has been " "mounted read-write\n" - " by another process!", w->filename); + " by another process!", wim->filename); ret = WIMLIB_ERR_ALREADY_LOCKED; } else { WARNING_WITH_ERRNO("Failed to lock `%"TS"'", - w->filename); + wim->filename); ret = 0; } } else { - w->wim_locked = 1; + wim->wim_locked = 1; } } return ret; } #endif +/* + * write_pipable_wim(): + * + * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM + * capable of being applied from a pipe). + * + * Pipable WIMs are a wimlib-specific modification of the WIM format such that + * images can be applied from them sequentially when the file data is sent over + * a pipe. In addition, a pipable WIM can be written sequentially to a pipe. + * The modifications made to the WIM format for pipable WIMs are: + * + * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead + * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also + * stops other software from trying to read the file as a normal WIM. + * + * - The header at the beginning of the file does not contain all the normal + * information; in particular it will have all 0's for the lookup table and + * XML data resource entries. This is because this information cannot be + * determined until the lookup table and XML data have been written. + * Consequently, wimlib will write the full header at the very end of the + * file. The header at the end, however, is only used when reading the WIM + * from a seekable file (not a pipe). + * + * - An extra copy of the XML data is placed directly after the header. This + * allows image names and sizes to be determined at an appropriate time when + * reading the WIM from a pipe. This copy of the XML data is ignored if the + * WIM is read from a seekable file (not a pipe). + * + * - The format of resources, or streams, has been modified to allow them to be + * used before the "lookup table" has been read. Each stream is prefixed with + * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct + * wim_lookup_table_entry_disk' that only contains the SHA1 message digest, + * uncompressed stream size, and flags that indicate whether the stream is + * compressed. The data of uncompressed streams then follows literally, while + * the data of compressed streams follows in a modified format. Compressed + * streams do not begin with a chunk table, since the chunk table cannot be + * written until all chunks have been compressed. Instead, each compressed + * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size. + * Furthermore, the chunk table is written at the end of the resource instead + * of the start. Note: chunk offsets are given in the chunk table as if the + * `struct pwm_chunk_hdr's were not present; also, the chunk table is only + * used if the WIM is being read from a seekable file (not a pipe). + * + * - Metadata resources always come before other file resources (streams). + * (This does not by itself constitute an incompatibility with normal WIMs, + * since this is valid in normal WIMs.) + * + * - At least up to the end of the file resources, all components must be packed + * as tightly as possible; there cannot be any "holes" in the WIM. (This does + * not by itself consititute an incompatibility with normal WIMs, since this + * is valid in normal WIMs.) + * + * Note: the lookup table, XML data, and header at the end are not used when + * applying from a pipe. They exist to support functionality such as image + * application and export when the WIM is *not* read from a pipe. + * + * Layout of pipable WIM: + * + * ---------+----------+--------------------+----------------+--------------+-----------+--------+ + * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header | + * ---------+----------+--------------------+----------------+--------------+-----------+--------+ + * + * Layout of normal WIM: + * + * +--------+-----------------------------+-------------------------+ + * | Header | File and metadata resources | Lookup table | XML data | + * +--------+-----------------------------+-------------------------+ + * + * An optional integrity table can follow the final XML data in both normal and + * pipable WIMs. However, due to implementation details, wimlib currently can + * only include an integrity table in a pipable WIM when writing it to a + * seekable file (not a pipe). + * + * Do note that since pipable WIMs are not supported by Microsoft's software, + * wimlib does not create them unless explicitly requested (with + * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic + * characters to identify the file. + */ static int -open_wim_writable(WIMStruct *w, const tchar *path, int open_flags) -{ - wimlib_assert(w->out_fd == INVALID_FILEDES); - w->out_fd = open(path, open_flags, 0644); - if (w->out_fd == INVALID_FILEDES) { - ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path); - return WIMLIB_ERR_OPEN; - } - return 0; -} - - -void -close_wim_writable(WIMStruct *w) -{ - if (w->out_fd != INVALID_FILEDES) { - if (close(w->out_fd)) - WARNING_WITH_ERRNO("Failed to close output WIM"); - w->out_fd = INVALID_FILEDES; - } -} - -/* Open file stream and write dummy header for WIM. */ -int -begin_write(WIMStruct *w, const tchar *path, int write_flags) +write_pipable_wim(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + struct list_head *stream_list_override, + struct list_head *lookup_table_list_ret) { int ret; - int open_flags = O_TRUNC | O_CREAT; - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - open_flags |= O_RDWR; - else - open_flags |= O_WRONLY; - ret = open_wim_writable(w, path, open_flags); + struct wim_reshdr xml_reshdr; + + WARNING("Creating a pipable WIM, which will " + "be incompatible\n" + " with Microsoft's software (wimgapi/imagex/Dism)."); + + /* At this point, the header at the beginning of the file has already + * been written. */ + + /* For efficiency, when wimlib adds an image to the WIM with + * wimlib_add_image(), the SHA1 message digests of files is not + * calculated; instead, they are calculated while the files are being + * written. However, this does not work when writing a pipable WIM, + * since when writing a stream to a pipable WIM, its SHA1 message digest + * needs to be known before the stream data is written. Therefore, + * before getting much farther, we need to pre-calculate the SHA1 + * message digests of all streams that will be written. */ + ret = wim_checksum_unhashed_streams(wim); if (ret) return ret; - /* Write dummy header. It will be overwritten later. */ - ret = write_header(&w->hdr, w->out_fd); + + /* Write extra copy of the XML data. */ + ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, + &xml_reshdr, + WRITE_RESOURCE_FLAG_PIPABLE); if (ret) return ret; - if (lseek(w->out_fd, 0, SEEK_END) == -1) { - ERROR_WITH_ERRNO("Failed to seek to end of WIM"); - return WIMLIB_ERR_WRITE; - } - return 0; + + /* Write metadata resources for the image(s) being included in the + * output WIM. */ + ret = write_wim_metadata_resources(wim, image, write_flags); + if (ret) + return ret; + + /* 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, + stream_list_override, lookup_table_list_ret); + + /* The lookup table, XML data, and header at end are handled by + * finish_write(). */ } -/* Writes a stand-alone WIM to a file. */ -WIMLIBAPI int -wimlib_write(WIMStruct *w, const tchar *path, - int image, int write_flags, unsigned num_threads, - wimlib_progress_func_t progress_func) +/* Write a standalone WIM or split WIM (SWM) part to a new file or to a file + * descriptor. */ +int +write_wim_part(WIMStruct *wim, + const void *path_or_fd, + int image, + int write_flags, + unsigned num_threads, + unsigned part_number, + unsigned total_parts, + struct list_head *stream_list_override, + const u8 *guid) { int ret; + struct wim_header hdr_save; + struct list_head lookup_table_list; - if (!path) - return WIMLIB_ERR_INVALID_PARAM; + if (total_parts == 1) + DEBUG("Writing standalone WIM."); + else + DEBUG("Writing split WIM part %u/%u", part_number, total_parts); + if (image == WIMLIB_ALL_IMAGES) + DEBUG("Including all images."); + else + DEBUG("Including image %d only.", image); + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) + DEBUG("File descriptor: %d", *(const int*)path_or_fd); + 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_NO_CHECK_INTEGRITY) + DEBUG("\tNO_CHECK_INTEGRITY"); + + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + DEBUG("\tPIPABLE"); - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE) + DEBUG("\tNOT_PIPABLE"); + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + DEBUG("\tRECOMPRESS"); + + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) + DEBUG("\tFSYNC"); + + if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) + DEBUG("\tREBUILD"); + + if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE) + DEBUG("\tSOFT_DELETE"); + + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) + DEBUG("\tIGNORE_READONLY_FLAG"); + + if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS) + DEBUG("\tSKIP_EXTERNAL_WIMS"); + + if (write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) + DEBUG("\tSTREAMS_OK"); + + if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID) + DEBUG("\tRETAIN_GUID"); + + 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 + DEBUG("Number of threads: %u", num_threads); + DEBUG("Progress function: %s", (wim->progfunc ? "yes" : "no")); + DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect")); + DEBUG("GUID: %s", (write_flags & + WIMLIB_WRITE_FLAG_RETAIN_GUID) ? "retain" + : guid ? "explicit" : "generate new"); + + /* Internally, this is always called with a valid part number and total + * parts. */ + wimlib_assert(total_parts >= 1); + wimlib_assert(part_number >= 1 && part_number <= total_parts); + + /* A valid image (or all images) must be specified. */ if (image != WIMLIB_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) + (image < 1 || image > wim->hdr.image_count)) return WIMLIB_ERR_INVALID_IMAGE; - if (w->hdr.total_parts != 1) { - ERROR("Cannot call wimlib_write() on part of a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; + /* If we need to write metadata resources, make sure the ::WIMStruct has + * the needed information attached (e.g. is not a resource-only WIM, + * such as a non-first part of a split WIM). */ + if (!wim_has_metadata(wim) && + !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) + return WIMLIB_ERR_METADATA_NOT_FOUND; + + /* Check for contradictory flags. */ + if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + return WIMLIB_ERR_INVALID_PARAM; + + if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + return WIMLIB_ERR_INVALID_PARAM; + + /* Save previous header, then start initializing the new one. */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); + + /* Set default integrity, pipable, and packed stream flags. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE))) + 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)) { + 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; } - ret = begin_write(w, path, write_flags); - if (ret) - goto out_close_wim; + /* Set appropriate magic number. */ + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + wim->hdr.magic = PWM_MAGIC; + else + wim->hdr.magic = WIM_MAGIC; - ret = write_wim_streams(w, image, write_flags, num_threads, - progress_func); - if (ret) - goto out_close_wim; + /* 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 | + WIM_HDR_FLAG_SPANNED | + WIM_HDR_FLAG_WRITE_IN_PROGRESS); - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); + /* Set SPANNED header flag if writing part of a split WIM. */ + if (total_parts != 1) + wim->hdr.flags |= WIM_HDR_FLAG_SPANNED; - ret = for_image(w, image, write_metadata_resource); + /* Set part number and total parts of split WIM. This will be 1 and 1 + * if the WIM is standalone. */ + wim->hdr.part_number = part_number; + wim->hdr.total_parts = total_parts; + + /* Set compression type if different. */ + if (wim->compression_type != wim->out_compression_type) { + ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr); + wimlib_assert(ret == 0); + } + + /* Set chunk size if different. */ + wim->hdr.chunk_size = wim->out_chunk_size; + + /* Set GUID. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)) { + if (guid) + memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN); + else + randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN); + } + + /* Clear references to resources that have not been written yet. */ + 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) { + wim->hdr.image_count = 1; + if (wim->hdr.boot_idx == image) + wim->hdr.boot_idx = 1; + else + wim->hdr.boot_idx = 0; + } + + /* Split WIMs can't be bootable. */ + if (total_parts != 1) + wim->hdr.boot_idx = 0; + + /* Initialize output file descriptor. */ + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) { + /* File descriptor was explicitly provided. Return error if + * file descriptor is not seekable, unless writing a pipable WIM + * was requested. */ + wim->out_fd.fd = *(const int*)path_or_fd; + wim->out_fd.offset = 0; + if (!filedes_is_seekable(&wim->out_fd)) { + ret = WIMLIB_ERR_INVALID_PARAM; + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + goto out_restore_hdr; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + ERROR("Can't include integrity check when " + "writing pipable WIM to pipe!"); + goto out_restore_hdr; + } + } + + } else { + /* Filename of WIM to write was provided; open file descriptor + * to it. */ + ret = open_wim_writable(wim, (const tchar*)path_or_fd, + O_TRUNC | O_CREAT | O_RDWR); + if (ret) + goto out_restore_hdr; + } + + /* Write initial header. This is merely a "dummy" header since it + * doesn't have all the information yet, so it will be overwritten later + * (unless writing a pipable WIM). */ + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header(&wim->hdr, &wim->out_fd); + wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; if (ret) - goto out_close_wim; + goto out_restore_hdr; + + /* 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, + stream_list_override, + &lookup_table_list); + if (ret) + goto out_restore_hdr; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + ret = write_wim_metadata_resources(wim, image, write_flags); + if (ret) + goto out_restore_hdr; + } else { + /* Non-default case: create pipable WIM. */ + ret = write_pipable_wim(wim, image, write_flags, num_threads, + stream_list_override, + &lookup_table_list); + if (ret) + goto out_restore_hdr; + write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END; + } - ret = finish_write(w, image, write_flags, progress_func); - /* finish_write() closed the WIM for us */ - goto out; -out_close_wim: - close_wim_writable(w); -out: - DEBUG("wimlib_write(path=%"TS") = %d", path, ret); + + /* Write lookup table, XML data, and (optional) integrity table. */ + ret = finish_write(wim, image, write_flags, &lookup_table_list); +out_restore_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); + (void)close_wim_writable(wim, write_flags); + DEBUG("ret=%d", ret); return ret; } +/* Write a standalone WIM to a file or file descriptor. */ +static int +write_standalone_wim(WIMStruct *wim, const void *path_or_fd, + int image, int write_flags, unsigned num_threads) +{ + return write_wim_part(wim, path_or_fd, image, write_flags, + num_threads, 1, 1, NULL, NULL); +} + +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write(WIMStruct *wim, const tchar *path, + int image, int write_flags, unsigned num_threads) +{ + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; + + if (path == NULL || path[0] == T('\0')) + return WIMLIB_ERR_INVALID_PARAM; + + return write_standalone_wim(wim, path, image, write_flags, num_threads); +} + +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write_to_fd(WIMStruct *wim, int fd, + int image, int write_flags, unsigned num_threads) +{ + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; + + if (fd < 0) + return WIMLIB_ERR_INVALID_PARAM; + + write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR; + + return write_standalone_wim(wim, &fd, image, write_flags, num_threads); +} + static bool -any_images_modified(WIMStruct *w) +any_images_modified(WIMStruct *wim) { - for (int i = 0; i < w->hdr.image_count; i++) - if (w->image_metadata[i]->modified) + for (int i = 0; i < wim->hdr.image_count; i++) + if (wim->image_metadata[i]->modified) return true; return false; } +static int +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->rspec->wim == wim && + lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset) + return WIMLIB_ERR_RESOURCE_ORDER; + return 0; +} + +/* Make sure no file or metadata resources are located after the XML data (or + * integrity table if present)--- otherwise we can't safely overwrite the WIM in + * place and we return WIMLIB_ERR_RESOURCE_ORDER. */ +static int +check_resource_offsets(WIMStruct *wim, off_t end_offset) +{ + int ret; + unsigned i; + + wim->private = &end_offset; + ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim); + if (ret) + return ret; + + for (i = 0; i < wim->hdr.image_count; i++) { + ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim); + if (ret) + return ret; + } + return 0; +} + /* * Overwrite a WIM, possibly appending streams to it. * @@ -2024,47 +2966,67 @@ any_images_modified(WIMStruct *w) * Integrity table (optional) (variable size) * * This method allows an image to be appended to a large WIM very quickly, and - * is is crash-safe except in the case of write re-ordering, but the + * is crash-safe except in the case of write re-ordering, but the * disadvantage is that a small hole is left in the WIM where the old lookup * table, xml data, and integrity table were. (These usually only take up a * small amount of space compared to the streams, however.) */ static int -overwrite_wim_inplace(WIMStruct *w, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads) { int ret; - struct list_head stream_list; off_t old_wim_end; u64 old_lookup_table_end, old_xml_begin, old_xml_end; - int open_flags; + 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); - DEBUG("Overwriting `%"TS"' in-place", w->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; /* Make sure that the integrity table (if present) is after the XML * 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 = w->hdr.xml_res_entry.offset; - old_xml_end = old_xml_begin + w->hdr.xml_res_entry.size; - old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; - if (w->hdr.integrity.offset != 0 && w->hdr.integrity.offset < old_xml_end) { - ERROR("Didn't expect the integrity table to be before the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; + 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"); + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out_restore_memory_hdr; } if (old_lookup_table_end > old_xml_begin) { - ERROR("Didn't expect the lookup table to be after the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; + WARNING("Didn't expect the lookup table to be after the XML data"); + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out_restore_memory_hdr; } /* Set @old_wim_end, which indicates the point beyond which we don't * allow any file and metadata resources to appear without returning * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise * overwrite these resources). */ - if (!w->deletion_occurred && !any_images_modified(w)) { + if (!wim->deletion_occurred && !any_images_modified(wim)) { /* If no images have been modified and no images have been * deleted, a new lookup table does not need to be written. We * shall write the new XML data and optional integrity table @@ -2075,169 +3037,199 @@ overwrite_wim_inplace(WIMStruct *w, 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 (w->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 = w->hdr.integrity.offset + w->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. */ old_wim_end = old_xml_end; } - ret = prepare_streams_for_overwrite(w, old_wim_end, &stream_list); + ret = check_resource_offsets(wim, old_wim_end); if (ret) - return ret; + goto out_restore_memory_hdr; - open_flags = 0; - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - open_flags |= O_RDWR; - else - open_flags |= O_WRONLY; - ret = open_wim_writable(w, w->filename, open_flags); + 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) + goto out_restore_memory_hdr; + + ret = lock_wim(wim, wim->out_fd.fd); + if (ret) + goto out_close_wim; - ret = lock_wim(w, w->out_fd); + /* 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) { - close_wim_writable(w); - return ret; + ERROR_WITH_ERRNO("Error updating WIM header flags"); + goto out_unlock_wim; } - if (lseek(w->out_fd, old_wim_end, SEEK_SET) == -1) { + if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - close_wim_writable(w); - w->wim_locked = 0; - return WIMLIB_ERR_WRITE; + ret = WIMLIB_ERR_WRITE; + goto out_restore_physical_hdr; } - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); - ret = write_stream_list(&stream_list, - w->lookup_table, - w->out_fd, - wimlib_get_compression_type(w), - write_flags, - num_threads, - progress_func); + ret = wim_write_stream_list(wim, + &stream_list, + write_flags, + num_threads, + &filter_ctx); + if (ret) + goto out_truncate; + + ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags); if (ret) goto out_truncate; - for (int i = 0; i < w->hdr.image_count; i++) { - if (w->image_metadata[i]->modified) { - select_wim_image(w, i + 1); - ret = write_metadata_resource(w); - if (ret) - goto out_truncate; - } - } write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; - ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, - progress_func); + ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, + &lookup_table_list); + if (ret) + goto out_truncate; + + wim->wim_locked = 0; + return 0; + out_truncate: - close_wim_writable(w); - if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", - w->filename, old_wim_end); - /* Return value of truncate() is ignored because this is already - * an error path. */ - (void)ttruncate(w->filename, old_wim_end); - } - w->wim_locked = 0; + wim->filename, old_wim_end); + /* Return value of ftruncate() is ignored because this is + * already an error path. */ + (void)ftruncate(wim->out_fd.fd, old_wim_end); + } +out_restore_physical_hdr: + (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd); +out_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; } static int -overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads) { size_t wim_name_len; int ret; - DEBUG("Overwriting `%"TS"' via a temporary file", w->filename); + DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename); /* Write the WIM to a temporary file in the same directory as the * original WIM. */ - wim_name_len = tstrlen(w->filename); + wim_name_len = tstrlen(wim->filename); tchar tmpfile[wim_name_len + 10]; - tmemcpy(tmpfile, w->filename, wim_name_len); + tmemcpy(tmpfile, wim->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); tmpfile[wim_name_len + 9] = T('\0'); - ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, - write_flags | WIMLIB_WRITE_FLAG_FSYNC, - num_threads, progress_func); + ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, + write_flags | + WIMLIB_WRITE_FLAG_FSYNC | + WIMLIB_WRITE_FLAG_RETAIN_GUID, + num_threads); if (ret) { - ERROR("Failed to write the WIM file `%"TS"'", tmpfile); - goto out_unlink; + tunlink(tmpfile); + return ret; } - DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); - -#ifdef __WIN32__ - /* Windows won't let you delete open files unless FILE_SHARE_DELETE was - * specified to CreateFile(). The WIM was opened with fopen(), which - * didn't provided this flag to CreateFile, so the handle must be closed - * before executing the rename(). */ - if (w->fp != NULL) { - fclose(w->fp); - w->fp = NULL; + if (filedes_valid(&wim->in_fd)) { + filedes_close(&wim->in_fd); + filedes_invalidate(&wim->in_fd); } -#endif - /* Rename the new file to the old file .*/ - if (trename(tmpfile, w->filename) != 0) { + /* Rename the new WIM file to the original WIM file. Note: on Windows + * this actually calls win32_rename_replacement(), not _wrename(), so + * that removing the existing destination file can be handled. */ + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename); + ret = trename(tmpfile, wim->filename); + if (ret) { ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", - tmpfile, w->filename); - ret = WIMLIB_ERR_RENAME; - goto out_unlink; - } - - if (progress_func) { - union wimlib_progress_info progress; - progress.rename.from = tmpfile; - progress.rename.to = w->filename; - progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); - } - goto out; -out_unlink: - /* Remove temporary file. */ - if (tunlink(tmpfile) != 0) - WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); -out: - return ret; + tmpfile, wim->filename); + #ifdef __WIN32__ + if (ret < 0) + #endif + { + tunlink(tmpfile); + } + return WIMLIB_ERR_RENAME; + } + + union wimlib_progress_info progress; + progress.rename.from = tmpfile; + progress.rename.to = wim->filename; + return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME, + &progress, wim->progctx); } -/* - * Writes a WIM file to the original file that it was read from, overwriting it. - */ +/* 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; + + /* 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 *w, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads) { - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + int ret; + u32 orig_hdr_flags; - if (!w->filename) + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; + + if (!wim->filename) return WIMLIB_ERR_NO_FILENAME; - if (w->hdr.total_parts != 1) { - ERROR("Cannot modify a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; - } + orig_hdr_flags = wim->hdr.flags; + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) + wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY; + ret = can_modify_wim(wim); + wim->hdr.flags = orig_hdr_flags; + if (ret) + return ret; - if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) - && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) - { - int ret; - ret = overwrite_wim_inplace(w, write_flags, num_threads, - progress_func); - if (ret == WIMLIB_ERR_RESOURCE_ORDER) - WARNING("Falling back to re-building entire WIM"); - else + if (can_overwrite_wim_inplace(wim, write_flags)) { + ret = overwrite_wim_inplace(wim, write_flags, num_threads); + if (ret != WIMLIB_ERR_RESOURCE_ORDER) return ret; + WARNING("Falling back to re-building entire WIM"); } - return overwrite_wim_via_tmpfile(w, write_flags, num_threads, - progress_func); + return overwrite_wim_via_tmpfile(wim, write_flags, num_threads); }