X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=6ae5599337b9552f556134effd691838feaa3318;hp=45b4993f5d3baa43f9cedec76afa3a444ade0408;hb=1f2091f5bb860769e20105af2d9eff3a0872c57b;hpb=885632f08c75c1d7bb5d25436231c78f6ad7e0c0 diff --git a/src/write.c b/src/write.c index 45b4993f..6ae55993 100644 --- a/src/write.c +++ b/src/write.c @@ -3,952 +3,2139 @@ * * Support for writing WIM files; write a WIM file, overwrite a WIM file, write * compressed file resources, etc. + */ + +/* + * Copyright (C) 2012, 2013 Eric Biggers * - * Copyright (C) 2010 Carl Thijssen - * Copyright (C) 2012 Eric Biggers - * - * wimlib - Library for working with WIM files + * This file is part of wimlib, a library for working with WIM files. * - * This library is free software; you can redistribute it and/or modify it under - * the terms of the GNU Lesser General Public License as published by the Free - * Software Foundation; either version 2.1 of the License, or (at your option) any - * later version. + * wimlib is free software; you can redistribute it and/or modify it under the + * terms of the GNU General Public License as published by the Free + * Software Foundation; either version 3 of the License, or (at your option) + * any later version. * - * This library is distributed in the hope that it will be useful, but WITHOUT ANY - * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A - * PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details. + * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY + * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR + * A PARTICULAR PURPOSE. See the GNU General Public License for more + * details. * - * You should have received a copy of the GNU Lesser General Public License along - * with this library; if not, write to the Free Software Foundation, Inc., 59 - * Temple Place, Suite 330, Boston, MA 02111-1307 USA + * You should have received a copy of the GNU General Public License + * along with wimlib; if not, see http://www.gnu.org/licenses/. */ + +#include "config.h" + +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +/* On BSD, this should be included before "list.h" so that "list.h" can + * overwrite the LIST_HEAD macro. */ +# include +#endif + +#ifdef __WIN32__ +# include "win32.h" +#endif + +#include "list.h" #include "wimlib_internal.h" -#include "io.h" -#include "lookup_table.h" +#include "buffer_io.h" #include "dentry.h" -#include "sha1.h" -#include "lzx.h" +#include "lookup_table.h" #include "xml.h" -#include "xpress.h" + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +# include +#endif + #include +#include + +#ifdef WITH_NTFS_3G +# include +# include +# include +# include +#endif + +#ifdef HAVE_ALLOCA_H +# include +#else +# include +#endif + +#include + +#if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE) +# define INVALID_HANDLE_VALUE ((HANDLE)(-1)) +#endif + +/* 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]; +}; +/* + * Allocates and initializes a chunk table, and reserves space for it in the + * output file. + */ +static int +begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, + FILE *out_fp, + off_t file_offset, + struct chunk_table **chunk_tab_ret) +{ + 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 ret; + if (!chunk_tab) { + ERROR("Failed to allocate chunk table for %"PRIu64" byte " + "resource", size); + ret = WIMLIB_ERR_NOMEM; + goto out; + } + 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 (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) != + chunk_tab->table_disk_size) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + FREE(chunk_tab); + ret = WIMLIB_ERR_WRITE; + goto out; + } -/* Used for buffering FILE IO */ -#define BUFFER_SIZE 4096 + ret = 0; + *chunk_tab_ret = chunk_tab; +out: + return ret; +} /* - * Copies bytes between two file streams. + * 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. * - * Copies @len bytes from @in to @out, at the current position in @out, and at - * an offset of @in_offset in @in. + * 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. */ -static int copy_between_files(FILE *in, off_t in_offset, FILE *out, size_t len) -{ - u8 buf[BUFFER_SIZE]; - size_t n; - - if (fseeko(in, in_offset, SEEK_SET) != 0) { - ERROR("Failed to seek to byte %"PRIu64" of input file: %m\n", - in_offset); - return WIMLIB_ERR_READ; - } - /* To reduce memory usage and improve speed, read and write BUFFER_SIZE - * bytes at a time. */ - while (len != 0) { - n = min(len, BUFFER_SIZE); - if (fread(buf, 1, n, in) != n) { - if (feof(in)) { - ERROR("Unexpected EOF when copying data " - "between files\n"); - } else { - ERROR("Error copying data between files: %m\n"); - } - return WIMLIB_ERR_READ; - } +typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, + void *out); - if (fwrite(buf, 1, n, out) != n) { - ERROR("Error copying data between files: %m\n"); - return WIMLIB_ERR_WRITE; +compress_func_t +get_compress_func(int out_ctype) +{ + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) + return wimlib_lzx_compress; + else + return wimlib_xpress_compress; +} + +/* + * 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_fp: FILE * 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 *chunk, unsigned chunk_size, + FILE *out_fp, compress_func_t compress, + struct chunk_table *chunk_tab) +{ + const u8 *out_chunk; + unsigned out_chunk_size; + if (compress) { + u8 *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; } - len -= n; + *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; + } + if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { + ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); + return WIMLIB_ERR_WRITE; } return 0; } - -/* - * Uncompresses a WIM file resource and writes it uncompressed to a file stream. +/* + * Finishes a WIM chunk table and writes it to the output file at the correct + * offset. * - * @in: The file stream that contains the file resource. - * @size: The size of the resource in the input file. - * @original_size: The original (uncompressed) size of the resource. - * @offset: The offset of the start of the resource in @in. - * @input_ctype: The compression type of the resource in @in. - * @out: The file stream to write the file resource to. + * The final size of the full compressed resource is returned in the + * @compressed_size_p. */ -static int uncompress_resource(FILE *in, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out) +static int +finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, + FILE *out_fp, u64 *compressed_size_p) { - int ret; - u8 buf[WIM_CHUNK_SIZE]; - /* Determine how many compressed chunks the file is divided into. */ - u64 num_chunks; - u64 i; - u64 uncompressed_offset; - u64 uncompressed_chunk_size; - - num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - - for (i = 0; i < num_chunks; i++) { - - uncompressed_offset = i * WIM_CHUNK_SIZE; - uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); - - ret = read_resource(in, size, original_size, offset, input_ctype, - uncompressed_chunk_size, - uncompressed_offset, buf); - if (ret != 0) - return ret; + size_t bytes_written; + if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output " + "WIM file", chunk_tab->file_offset); + return WIMLIB_ERR_WRITE; + } - if (fwrite(buf, 1, uncompressed_chunk_size, out) != - uncompressed_chunk_size) { - ERROR("Failed to write file resource: %m\n"); - return WIMLIB_ERR_WRITE; - } + if (chunk_tab->bytes_per_chunk_entry == 8) { + array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); + } 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 = fwrite((u8*)chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + 1, chunk_tab->table_disk_size, out_fp); + if (bytes_written != chunk_tab->table_disk_size) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + return WIMLIB_ERR_WRITE; + } + if (fseeko(out_fp, 0, SEEK_END) != 0) { + ERROR_WITH_ERRNO("Failed to seek to end of output WIM file"); + return WIMLIB_ERR_WRITE; + } + *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; return 0; } -/* - * Transfers a file resource between two files, writing it compressed. The file - * resource in the input file may be either compressed or uncompressed. - * Alternatively, the input resource may be in-memory, but it must be - * uncompressed. - * - * @in: The file stream that contains the file resource. Ignored - * if uncompressed_resource != NULL. - * @uncompressed_resource: If this pointer is not NULL, it points to an - * array of @original_size bytes that are - * the uncompressed input resource. - * @size: The size of the resource in the input file. - * @original_size: The original (uncompressed) size of the resource. - * @offset: The offset of the start of the resource in @in. Ignored - * if uncompressed_resource != NULL. - * @input_ctype: The compression type of the resource in @in. Ignored if - * uncompressed_resource != NULL. - * @out: The file stream to write the file resource to. - * @output_type: The compression type to use when writing the resource to - * @out. - * @new_size_ret: A location into which the new compressed size of the file - * resource in returned. - */ -static int recompress_resource(FILE *in, const u8 uncompressed_resource[], - u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out, - int output_ctype, u64 *new_size_ret) +static int +finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) { - int ret; - int (*compress)(const void *, uint, void *, uint *); - if (output_ctype == WIM_COMPRESSION_TYPE_LZX) - compress = lzx_compress; - else - compress = xpress_compress; + u8 md[SHA1_HASH_SIZE]; + sha1_final(md, sha_ctx); + if (lte->unhashed) { + copy_hash(lte->hash, md); + } else if (!hashes_equal(md, lte->hash)) { + ERROR("WIM resource has incorrect hash!"); + if (lte_filename_valid(lte)) { + ERROR("We were reading it from \"%"TS"\"; maybe " + "it changed while we were reading it.", + lte->file_on_disk); + } + return WIMLIB_ERR_INVALID_RESOURCE_HASH; + } + return 0; +} - u8 uncompressed_buf[WIM_CHUNK_SIZE]; - u8 compressed_buf[WIM_CHUNK_SIZE - 1]; - /* Determine how many compressed chunks the file needs to be divided - * into. */ - u64 num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; +struct write_resource_ctx { + compress_func_t compress; + struct chunk_table *chunk_tab; + FILE *out_fp; + SHA_CTX sha_ctx; + bool doing_sha; +}; - u64 num_chunk_entries = num_chunks - 1; +static int +write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) +{ + struct write_resource_ctx *ctx = _ctx; - /* Size of the chunk entries--- 8 bytes for files over 4GB, otherwise 4 - * bytes */ - uint chunk_entry_size = (original_size >= (u64)1 << 32) ? 8 : 4; + if (ctx->doing_sha) + sha1_update(&ctx->sha_ctx, chunk, chunk_size); - /* Array in which to construct the chunk offset table. */ - u64 chunk_offsets[num_chunk_entries]; + if (ctx->compress) { + return write_wim_resource_chunk(chunk, chunk_size, + ctx->out_fp, ctx->compress, + ctx->chunk_tab); + } else { + if (fwrite(chunk, 1, chunk_size, ctx->out_fp) != chunk_size) { + ERROR_WITH_ERRNO("Error writing to output WIM"); + return WIMLIB_ERR_WRITE; + } else { + return 0; + } + } +} - /* Offset of the start of the chunk table in the output file. */ - off_t chunk_tab_offset = ftello(out); +/* + * 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 * 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, + FILE *out_fp, 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; + int ret; - /* Total size of the chunk table (as written to the file) */ - u64 chunk_tab_size = chunk_entry_size * num_chunk_entries; + flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS; - /* Reserve space for the chunk table. */ - if (fwrite(chunk_offsets, 1, chunk_tab_size, out) != chunk_tab_size) { - ERROR("Failed to write chunk offset table: %m\n"); + /* Get current position in output WIM */ + offset = ftello(out_fp); + if (offset == -1) { + ERROR_WITH_ERRNO("Can't get position in output WIM"); return WIMLIB_ERR_WRITE; } - /* Read each chunk of the file, compress it, write it to the output - * file, and update th chunk offset table. */ - u64 cur_chunk_offset = 0; - for (u64 i = 0; i < num_chunks; i++) { + /* 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; + } - u64 uncompressed_offset = i * WIM_CHUNK_SIZE; - u64 uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); + /* 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; + } else { + write_ctx.compress = get_compress_func(out_ctype); + ret = begin_wim_resource_chunk_tab(lte, out_fp, + offset, + &write_ctx.chunk_tab); + if (ret) + return ret; + } - const u8 *uncompressed_p; - if (uncompressed_resource != NULL) { - uncompressed_p = uncompressed_resource + - uncompressed_offset; + /* Write the entire resource by reading the entire resource and feeding + * the data through the write_resource_cb function. */ + write_ctx.out_fp = out_fp; +try_write_again: + ret = read_resource_prefix(lte, read_size, + write_resource_cb, &write_ctx, flags); + if (ret) + goto out_free_chunk_tab; + + /* Verify SHA1 message digest of the resource, or set the hash for the + * first time. */ + if (write_ctx.doing_sha) { + ret = finalize_and_check_sha1(&write_ctx.sha_ctx, lte); + if (ret) + goto out_free_chunk_tab; + } - } else { - /* Read chunk i of the file into uncompressed_buf. */ - ret = read_resource(in, size, original_size, offset, input_ctype, - uncompressed_chunk_size, - uncompressed_offset, - uncompressed_buf); - if (ret != 0) - return ret; - uncompressed_p = uncompressed_buf; + 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; + } 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_fp, + &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. */ + if (fseeko(out_fp, offset, SEEK_SET) || + fflush(out_fp) || + ftruncate(fileno(out_fp), + offset + wim_resource_size(lte))) + { + ERROR_WITH_ERRNO("Failed to flush and/or truncate " + "output WIM file"); + ret = WIMLIB_ERR_WRITE; + goto out_free_chunk_tab; + } + DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " + "writing uncompressed instead", + wim_resource_size(lte), new_size); + write_ctx.compress = NULL; + write_ctx.doing_sha = false; + out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + goto try_write_again; } + 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 + +/* 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; +}; + +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) + return WIMLIB_ERR_NOMEM; + q->filled_slots = 0; + q->front = 0; + q->back = size - 1; + q->size = size; + pthread_mutex_init(&q->lock, NULL); + pthread_cond_init(&q->msg_avail_cond, NULL); + pthread_cond_init(&q->space_avail_cond, NULL); + return 0; +} + +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); +} + +static void +shared_queue_put(struct shared_queue *q, void *obj) +{ + pthread_mutex_lock(&q->lock); + while (q->filled_slots == q->size) + pthread_cond_wait(&q->space_avail_cond, &q->lock); - if (i != 0) - chunk_offsets[i - 1] = cur_chunk_offset; + q->back = (q->back + 1) % q->size; + q->array[q->back] = obj; + q->filled_slots++; - uint compressed_len; + pthread_cond_broadcast(&q->msg_avail_cond); + pthread_mutex_unlock(&q->lock); +} + +static void * +shared_queue_get(struct shared_queue *q) +{ + void *obj; + + pthread_mutex_lock(&q->lock); + while (q->filled_slots == 0) + pthread_cond_wait(&q->msg_avail_cond, &q->lock); - ret = compress(uncompressed_p, uncompressed_chunk_size, - compressed_buf, &compressed_len); + obj = q->array[q->front]; + q->array[q->front] = NULL; + q->front = (q->front + 1) % q->size; + q->filled_slots--; - /* if compress() returned nonzero, the compressed chunk would - * have been at least as large as the uncompressed chunk. In - * this situation, the WIM format requires that the uncompressed - * chunk be written instead. */ - const u8 *buf_to_write; - uint len_to_write; - if (ret == 0) { - buf_to_write = compressed_buf; - len_to_write = compressed_len; + pthread_cond_broadcast(&q->space_avail_cond); + pthread_mutex_unlock(&q->lock); + return obj; +} + +struct compressor_thread_params { + struct shared_queue *res_to_compress_queue; + struct shared_queue *compressed_res_queue; + compress_func_t compress; +}; + +#define MAX_CHUNKS_PER_MSG 2 + +struct message { + struct wim_lookup_table_entry *lte; + u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; + u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG]; + u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; + unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; + unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; + unsigned num_chunks; + struct list_head list; + bool complete; + u64 begin_chunk; +}; + +static void +compress_chunks(struct message *msg, compress_func_t compress) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) { + DEBUG2("compress chunk %u of %u", i, msg->num_chunks); + unsigned len = compress(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i]); + if (len) { + /* To be written compressed */ + msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + msg->compressed_chunk_sizes[i] = len; } else { - buf_to_write = uncompressed_p; - len_to_write = uncompressed_chunk_size; - } + /* To be written uncompressed */ + msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; + msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; - if (fwrite(buf_to_write, 1, len_to_write, out) != len_to_write) { - ERROR("Failed to write compressed file resource: %m\n"); - return WIMLIB_ERR_WRITE; } - cur_chunk_offset += len_to_write; } +} - /* The chunk offset after the last chunk, plus the size of the chunk - * table, gives the total compressed size of the resource. */ - *new_size_ret = cur_chunk_offset + chunk_tab_size; - - /* Now that all entries of the chunk table are determined, rewind the - * stream to where the chunk table was, and write it back out. */ +/* 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) +{ + struct compressor_thread_params *params = arg; + struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; + struct shared_queue *compressed_res_queue = params->compressed_res_queue; + compress_func_t compress = params->compress; + struct message *msg; + + DEBUG("Compressor thread ready"); + while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { + compress_chunks(msg, compress); + shared_queue_put(compressed_res_queue, msg); + } + DEBUG("Compressor thread terminating"); + return NULL; +} +#endif /* ENABLE_MULTITHREADED_COMPRESSION */ - if (fseeko(out, chunk_tab_offset, SEEK_SET) != 0) { - ERROR("Failed to seek to beginning of chunk table: %m\n"); - return WIMLIB_ERR_READ; +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); + } } +} - if (chunk_entry_size == 8) { - array_to_le64(chunk_offsets, num_chunk_entries); +static int +do_write_stream_list(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + int write_resource_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) +{ + int ret = 0; + struct wim_lookup_table_entry *lte; + + /* 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); + 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)); + goto skip_to_progress; + } + } + } - if (fwrite(chunk_offsets, 1, chunk_tab_size, out) != - chunk_tab_size) { - ERROR("Failed to write chunk table: %m\n"); - return WIMLIB_ERR_WRITE; + /* 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); + ret = write_wim_resource(lte, + out_fp, + out_ctype, + <e->output_resource_entry, + write_resource_flags); + if (ret) + break; + if (lte->unhashed) { + list_del(<e->unhashed_list); + lookup_table_insert(lookup_table, lte); + lte->unhashed = 0; } - } else { - u32 chunk_entries_small[num_chunk_entries]; - for (u64 i = 0; i < num_chunk_entries; i++) - chunk_entries_small[i] = to_le32(chunk_offsets[i]); - if (fwrite(chunk_entries_small, 1, chunk_tab_size, out) != - chunk_tab_size) { - ERROR("Failed to write chunk table: %m\n"); + skip_to_progress: + do_write_streams_progress(progress, + progress_func, + wim_resource_size(lte)); + } + return ret; +} + +static int +write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + int write_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) +{ + int write_resource_flags = 0; + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; + + progress->write_streams.num_threads = 1; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + return do_write_stream_list(stream_list, + lookup_table, + out_fp, + out_ctype, + write_resource_flags, + progress_func, + progress); +} + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +static int +write_wim_chunks(struct message *msg, FILE *out_fp, + struct chunk_table *chunk_tab) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) { + unsigned chunk_csize = msg->compressed_chunk_sizes[i]; + + DEBUG2("Write wim chunk %u of %u (csize = %u)", + i, msg->num_chunks, chunk_csize); + + if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) + != chunk_csize) + { + ERROR_WITH_ERRNO("Failed to write WIM chunk"); return WIMLIB_ERR_WRITE; } - } - if (fseeko(out, 0, SEEK_END) != 0) { - ERROR("Failed to seek to end of output file: %m\n"); - return WIMLIB_ERR_WRITE; + *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; + chunk_tab->cur_offset += chunk_csize; } + return 0; +} +struct main_writer_thread_ctx { + struct list_head *stream_list; + struct wim_lookup_table *lookup_table; + FILE *out_fp; + int out_ctype; + struct shared_queue *res_to_compress_queue; + struct shared_queue *compressed_res_queue; + size_t num_messages; + int write_flags; + 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; + u64 next_chunk; + u64 next_num_chunks; + struct message *msgs; + struct message *next_msg; + size_t next_chunk_in_msg; + struct wim_lookup_table_entry *cur_lte; + struct chunk_table *cur_chunk_tab; + struct wim_lookup_table_entry *next_lte; + SHA_CTX sha_ctx; + u8 next_hash[20]; +}; + +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; + } return 0; } -int write_resource_from_memory(const u8 resource[], int out_ctype, - u64 resource_original_size, FILE *out, - u64 *resource_size_ret) +static void +destroy_message(struct message *msg) { - if (out_ctype == WIM_COMPRESSION_TYPE_NONE) { - if (fwrite(resource, 1, resource_original_size, out) != - resource_original_size) { - ERROR("Failed to write resource of length " - "%"PRIu64": %m\n", - resource_original_size); - return WIMLIB_ERR_WRITE; + for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { + FREE(msg->compressed_chunks[i]); + FREE(msg->uncompressed_chunks[i]); + } +} + +static void +free_messages(struct message *msgs, size_t num_messages) +{ + if (msgs) { + for (size_t i = 0; i < num_messages; i++) + destroy_message(&msgs[i]); + FREE(msgs); + } +} + +static struct message * +allocate_messages(size_t num_messages) +{ + struct message *msgs; + + 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; } - *resource_size_ret = resource_original_size; - return 0; - } else { - return recompress_resource(NULL, resource, resource_original_size, - resource_original_size, 0, 0, out, out_ctype, - resource_size_ret); - } -} - - -/* - * Transfers a file resource from a FILE* opened for reading to a FILE* opened - * for writing, possibly changing the compression type. - * - * @in: The FILE* that contains the file resource. - * @size: The (compressed) size of the file resource. - * @original_size: The uncompressed size of the file resource. - * @offset: The offset of the file resource in the input file. - * @input_ctype: The compression type of the file resource in the input - * file. - * @out: The FILE* for the output file. The file resource is - * written at the current position of @out. - * @output_ctype: The compression type to which the file resource will be - * converted. - * @output_res_entry: A pointer to a resource entry that, upon successful - * return of this function, will have the size, - * original size, offset, and flags fields filled - * in for the file resource written to the output - * file. - */ -static int transfer_file_resource(FILE *in, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out, - int output_ctype, - struct resource_entry *output_res_entry) + } + return msgs; +} + +static void +main_writer_thread_destroy_ctx(struct main_writer_thread_ctx *ctx) +{ + free_messages(ctx->msgs, ctx->num_messages); + FREE(ctx->cur_chunk_tab); +} + + +static int +main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) +{ + /* Pre-allocate all the buffers that will be needed to do the chunk + * compression. */ + ctx->msgs = allocate_messages(ctx->num_messages); + if (!ctx->msgs) + 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 (cur_lte). + * + * The last stream in outstanding_streams is the stream that is + * currently being read and chunks fed to the compressor threads. */ + INIT_LIST_HEAD(&ctx->outstanding_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); + + ctx->cur_lte = NULL; + return 0; +} + +static int +receive_compressed_chunks(struct main_writer_thread_ctx *ctx) { + struct message *msg; + struct wim_lookup_table_entry *cur_lte; int ret; - /* Handle zero-length files */ - if (original_size == 0) { - memset(output_res_entry, 0, sizeof(*output_res_entry)); - return 0; - } + wimlib_assert(!list_empty(&ctx->outstanding_streams)); + + /* 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; + cur_lte = ctx->cur_lte; + + /* 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) + { + 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 = ftello(ctx->out_fp); + if (cur_offset == -1) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + ret = begin_wim_resource_chunk_tab(cur_lte, + ctx->out_fp, + cur_offset, + &ctx->cur_chunk_tab); + if (ret) + goto out; + } - /* Get current offset in the output file. */ - output_res_entry->offset = ftello(out); - if (output_res_entry->offset == -1) { - ERROR("Failed to get output position: %m\n"); - return WIMLIB_ERR_WRITE; + /* Write the compressed chunks from the message. */ + ret = write_wim_chunks(msg, ctx->out_fp, ctx->cur_chunk_tab); + if (ret) + goto out; + + list_del(&msg->list); + + /* This message is available to use for different chunks + * now. */ + list_add(&msg->list, &ctx->available_msgs); + + /* 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) + { + DEBUG2("Finish wim chunk tab"); + u64 res_csize; + ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, + ctx->out_fp, + &res_csize); + if (ret) + goto out; + +#if 0 + if (res_csize >= wim_resource_size(cur_lte)) { + /* Oops! We compressed the resource to + * larger than the original size. Write + * the resource uncompressed instead. */ + ret = write_uncompressed_resource_and_truncate( + cur_lte, + ctx->out_fp, + ctx->cur_chunk_tab->file_offset, + &cur_lte->output_resource_entry); + if (ret) + goto out; + } else +#endif + { + cur_lte->output_resource_entry.size = + res_csize; + + cur_lte->output_resource_entry.original_size = + cur_lte->resource_entry.original_size; + + cur_lte->output_resource_entry.offset = + ctx->cur_chunk_tab->file_offset; + + cur_lte->output_resource_entry.flags = + cur_lte->resource_entry.flags | + WIM_RESHDR_FLAG_COMPRESSED; + } + do_write_streams_progress(ctx->progress, ctx->progress_func, + wim_resource_size(cur_lte)); + FREE(ctx->cur_chunk_tab); + ctx->cur_chunk_tab = NULL; + + struct list_head *next = cur_lte->write_streams_list.next; + list_del(&cur_lte->write_streams_list); + + if (next == &ctx->outstanding_streams) + cur_lte = NULL; + else + cur_lte = container_of(cur_lte->write_streams_list.next, + struct wim_lookup_table_entry, + write_streams_list); + + /* 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). */ + ret = do_write_stream_list(&ctx->serial_streams, + ctx->lookup_table, + ctx->out_fp, + ctx->out_ctype, + ctx->progress_func, + ctx->progress, + 0); + if (ret) + goto out; + } } +out: + ctx->cur_lte = cur_lte; + return ret; +} - if (output_ctype == input_ctype) { - /* The same compression types; simply copy the resource. */ +static int +main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) +{ + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + struct message *next_msg; - ret = copy_between_files(in, offset, out, size); - if (ret != 0) - return ret; - output_res_entry->size = size; - } else { - /* Different compression types. */ - - if (output_ctype == WIM_COMPRESSION_TYPE_NONE) { - /* Uncompress a compressed file resource */ - ret = uncompress_resource(in, size, - original_size, offset, - input_ctype, out); - if (ret != 0) - return ret; - output_res_entry->size = original_size; - } else { - u64 new_size; - /* Compress an uncompressed file resource, or compress a - * compressed file resource using a different - * compression type (the latter is currently unsupported - * since only LZX compression is supported. */ - ret = recompress_resource(in, NULL, size, original_size, - offset, input_ctype, out, - output_ctype, &new_size); - if (ret != 0) + next_msg = ctx->next_msg; + + sha1_update(&ctx->sha_ctx, chunk, chunk_size); + + if (!next_msg) { + if (list_empty(&ctx->available_msgs)) { + ret = receive_compressed_chunks(ctx); + if (ret) return ret; - output_res_entry->size = new_size; } + wimlib_assert(!list_empty(&ctx->available_msgs)); + + 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_chunk_in_msg = 0; } - output_res_entry->original_size = original_size; - if (output_ctype == WIM_COMPRESSION_TYPE_NONE) - output_res_entry->flags = 0; - else - output_res_entry->flags = WIM_RESHDR_FLAG_COMPRESSED; + wimlib_assert(next_msg != NULL); + wimlib_assert(ctx->next_chunk_in_msg < next_msg->num_chunks); + + next_msg->uncompressed_chunk_sizes[ctx->next_chunk_in_msg] = chunk_size; + memcpy(next_msg->uncompressed_chunks[ctx->next_chunk_in_msg], + chunk, chunk_size); + + if (++ctx->next_chunk_in_msg == next_msg->num_chunks) { + shared_queue_put(ctx->res_to_compress_queue, + next_msg); + ctx->next_msg = NULL; + } return 0; } -/* - * Writes a file resource to the output file. +static int +submit_stream_for_compression(struct wim_lookup_table_entry *lte, + struct main_writer_thread_ctx *ctx) +{ + int ret; + + sha1_init(&ctx->sha_ctx); + ctx->next_num_chunks = wim_resource_chunks(lte); + ret = read_resource_prefix(lte, wim_resource_size(lte), + main_writer_thread_cb, ctx, 0); + if (ret) + return ret; + ret = finalize_and_check_sha1(&ctx->sha_ctx, lte); + if (ret) + return ret; +} + +/* + * This function is executed by the main thread when the resources are being + * compressed in parallel. The main thread is in change of all reading of the + * uncompressed data and writing of the compressed data. The compressor threads + * *only* do compression from/to in-memory buffers. * - * @dentry: The dentry for the file resource. - * @wim_p: A pointer to the WIMStruct. The fields of interest to this - * function are the input and output file streams and the lookup table. - * @return zero on success, nonzero on failure. + * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG + * chunks of compressed data to compress, represented in a `struct message'. + * Each message is passed from the main thread to a worker thread through the + * res_to_compress_queue, and it is passed back through the + * compressed_res_queue. */ -static int write_file_resource(struct dentry *dentry, void *wim_p) -{ - WIMStruct *w; - FILE *out; - FILE *in; - struct lookup_table_entry *lte; - int in_wim_ctype; - int out_wim_ctype; - int input_res_ctype; - struct resource_entry *input_res_entry; - struct resource_entry *output_res_entry; - u64 len; +static int +main_writer_thread_proc(struct main_writer_thread_ctx *ctx) +{ int ret; + struct list_head *stream_list; + struct wim_lookup_table_entry *lte; + + ret = main_writer_thread_init_ctx(ctx); + if (ret) + goto out_destroy_ctx; + + stream_list = ctx->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) { + struct wim_lookup_table_entry *tmp; + u32 orig_refcnt = lte->out_refcnt; + + ret = hash_unhashed_stream(lte, ctx->lookup_table, &tmp); + if (ret) + goto out_destroy_ctx; + if (tmp != lte) { + lte = tmp; + if (orig_refcnt != tmp->out_refcnt) { + DEBUG("Discarding duplicate stream of length %"PRIu64, + wim_resource_size(lte)); + goto skip_to_progress; + } + } + } - w = wim_p; - out = w->out_fp; + if (wim_resource_size(lte) < 1000 || + ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || + (lte->resource_location == RESOURCE_IN_WIM && + wimlib_get_compression_type(lte->wim) == ctx->out_ctype)) + { + list_add(<e->write_streams_list, + &ctx->serial_streams); + } else { + ret = submit_stream_for_compression(lte, ctx); + if (ret) + goto out_destroy_ctx; + if (lte->unhashed) { + list_del(<e->unhashed_list); + lookup_table_insert(ctx->lookup_table, lte); + lte->unhashed = 0; + } + } + skip_to_progress: + do_write_streams_progress(ctx->progress, + ctx->progress_func, + wim_resource_size(lte)); + } - /* Directories don't need file resources. */ - if (dentry_is_directory(dentry)) - return 0; + while (!list_empty(&ctx->outstanding_streams)) { + ret = receive_compressed_chunks(ctx); + if (ret) + goto out_destroy_ctx; + } + ret = 0; +out_destroy_ctx: + main_writer_thread_destroy_ctx(ctx); + return ret; +} - /* Get the lookup entry for the file resource. */ - lte = wim_lookup_resource(w, dentry); - if (!lte) - return 0; +static long +get_default_num_threads() +{ +#ifdef __WIN32__ + return win32_get_number_of_processors(); +#else + return sysconf(_SC_NPROCESSORS_ONLN); +#endif +} - /* No need to write file resources twice. (This indicates file - * resources that are part of a hard link set.) */ - if (++lte->out_refcnt != 1) - return 0; +static int +write_stream_list_parallel(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) +{ + 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 { + num_threads = nthreads; + } + } + + progress->write_streams.num_threads = num_threads; + + static const double MESSAGES_PER_THREAD = 2.0; + 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 != 0) + goto out_serial; - out_wim_ctype = wimlib_get_compression_type(w); - output_res_entry = <e->output_resource_entry; + ret = shared_queue_init(&compressed_res_queue, queue_size); + if (ret != 0) + goto out_destroy_res_to_compress_queue; - /* Figure out if we can read the resource from the WIM file, or - * if we have to read it from the filesystem outside. */ - if (lte->file_on_disk) { + 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); - /* Read from disk (uncompressed) */ + compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + if (!compressor_threads) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } - len = lte->resource_entry.original_size; + for (unsigned i = 0; i < num_threads; i++) { + DEBUG("pthread_create thread %u", i); + 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", i); + num_threads = i; + goto out_join; + } + } - in = fopen(lte->file_on_disk, "rb"); - if (!in) { - ERROR("Failed to open the file `%s': %m\n", - lte->file_on_disk); - return WIMLIB_ERR_OPEN; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + + struct main_writer_thread_ctx ctx; + memset(&ctx, 0, sizeof(ctx)); + ctx.stream_list = stream_list; + ctx.lookup_table = lookup_table; + ctx.out_fp = out_fp; + 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_flags = write_flags; + ctx.progress_func = progress_func; + ctx.progress = progress; + ret = main_writer_thread_proc(&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", i); } + } + 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"); + return write_stream_list_serial(stream_list, + lookup_table, + out_fp, + out_ctype, + write_flags, + progress_func, + progress); - if (w->verbose) - puts(lte->file_on_disk); +} +#endif - ret = transfer_file_resource(in, len, len, 0, - WIM_COMPRESSION_TYPE_NONE, out, - out_wim_ctype, output_res_entry); - fclose(in); - } else { +/* + * Write a list of streams to a WIM (@out_fp) 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, + FILE *out_fp, int out_ctype, int write_flags, + unsigned num_threads, wimlib_progress_func_t progress_func) +{ + struct wim_lookup_table_entry *lte; + size_t num_streams = 0; + u64 total_bytes = 0; + u64 total_compression_bytes = 0; + union wimlib_progress_info progress; + int ret; - /* Read from input WIM (possibly compressed) */ + if (list_empty(stream_list)) + return 0; - /* It may be a different WIM file, in the case of - * exporting images from one WIM file to another */ - if (lte->other_wim_fp) { - /* Different WIM file. */ - in = lte->other_wim_fp; - in_wim_ctype = lte->other_wim_ctype; - } else { - /* Same WIM file. */ - in = w->fp; - in_wim_ctype = out_wim_ctype; + /* 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_flags & WIMLIB_WRITE_FLAG_RECOMPRESS))) + { + total_compression_bytes += wim_resource_size(lte); } - input_res_entry = <e->resource_entry; - input_res_ctype = resource_compression_type( - in_wim_ctype, - input_res_entry->flags); - - ret = transfer_file_resource(in, - input_res_entry->size, - input_res_entry->original_size, - input_res_entry->offset, - input_res_ctype, - out, - out_wim_ctype, - output_res_entry); } + 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_fp, + out_ctype, + write_flags, + num_threads, + progress_func, + &progress); + else +#endif + ret = write_stream_list_serial(stream_list, + lookup_table, + out_fp, + out_ctype, + write_flags, + progress_func, + &progress); return ret; } -/* Reopens the FILE* for a WIM read-write. */ -static int reopen_rw(WIMStruct *w) +struct stream_size_table { + struct hlist_head *array; + size_t num_entries; + size_t capacity; +}; + +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) + return WIMLIB_ERR_NOMEM; + tab->num_entries = 0; + tab->capacity = capacity; + return 0; +} + +static void +destroy_stream_size_table(struct stream_size_table *tab) { - FILE *fp; + FREE(tab->array); +} - if (fclose(w->fp) != 0) - ERROR("Failed to close the file `%s': %m\n", w->filename); - fp = fopen(w->filename, "r+b"); - if (!fp) { - ERROR("Failed to open `%s' for reading and writing: " - "%m\n", w->filename); - return WIMLIB_ERR_OPEN; +static int +stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) +{ + struct stream_size_table *tab = _tab; + size_t pos; + struct wim_lookup_table_entry *same_size_lte; + struct hlist_node *tmp; + + pos = hash_u64(wim_resource_size(lte)) % 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)) { + lte->unique_size = 0; + same_size_lte->unique_size = 0; + break; + } } - w->fp = fp; + + hlist_add_head(<e->hash_list_2, &tab->array[pos]); + tab->num_entries++; return 0; } +struct lte_overwrite_prepare_args { + WIMStruct *wim; + off_t end_offset; + 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 int +lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *_args) +{ + struct lte_overwrite_prepare_args *args = _args; -/* - * Writes a WIM file to the original file that it was read from, overwriting it. + 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); + 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) +{ + struct lte_overwrite_prepare_args *args = _args; + + 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); + } + 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. */ -WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int flags) +static int +prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset, + struct list_head *stream_list) { - const char *wimfile_name; - size_t wim_name_len; int ret; - - wimfile_name = w->filename; + 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; - DEBUG("Replacing WIM file `%s'\n", wimfile_name); + INIT_LIST_HEAD(&args.stream_list); + for (i = 0; i < wim->hdr.image_count; i++) { + struct wim_image_metadata *imd; + struct wim_lookup_table_entry *lte; - if (!wimfile_name) - return WIMLIB_ERR_NO_FILENAME; + 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 (i = 0; i < wim->hdr.image_count; i++) { + ret = lte_overwrite_prepare_2(wim->image_metadata[i]->metadata_lte, + &args); + if (ret) + goto out_destroy_stream_size_table; + } + 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; +} - /* Write the WIM to a temporary file. */ - /* XXX should the temporary file be somewhere else? */ - wim_name_len = strlen(wimfile_name); - char tmpfile[wim_name_len + 10]; - memcpy(tmpfile, wimfile_name, wim_name_len); - randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, flags); - if (ret != 0) { - ERROR("Failed to write the WIM file `%s'!\n", tmpfile); - return ret; - } +struct find_streams_ctx { + struct list_head stream_list; + struct stream_size_table stream_size_tab; +}; - DEBUG("Closing original WIM file.\n"); - /* Close the original WIM file that was opened for reading. */ - if (w->fp) { - if (fclose(w->fp) != 0) { - DEBUG("WARNING: Failed to close the file `%s'\n", - wimfile_name); +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) +{ + 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); + } + lte->out_refcnt += inode->i_nlink; } - w->fp = NULL; } +} - DEBUG("Renaming `%s' to `%s'\n", tmpfile, wimfile_name); - - /* Rename the new file to the old file .*/ - if (rename(tmpfile, wimfile_name) != 0) { - ERROR("Failed to rename `%s' to `%s': %m\n", tmpfile, - wimfile_name); - /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - ERROR("Failed to remove `%s': %m\n", tmpfile); - return WIMLIB_ERR_RENAME; +static int +image_find_streams_to_write(WIMStruct *w) +{ + struct find_streams_ctx *ctx; + struct wim_image_metadata *imd; + struct wim_inode *inode; + struct wim_lookup_table_entry *lte; + + ctx = w->private; + imd = wim_get_current_image_metadata(w); + + image_for_each_unhashed_stream(lte, imd) + lte->out_refcnt = 0; + + /* 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; } - -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *w, int flags) +/* 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) { int ret; - FILE *fp; - u8 *integrity_table = NULL; - off_t xml_end; - off_t xml_size; - size_t bytes_written; + struct find_streams_ctx ctx; - DEBUG("Overwriting XML and header of `%s', flags = %d\n", - w->filename, flags); - if (!w->filename) - return WIMLIB_ERR_NO_FILENAME; + 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); + 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); + return ret; +} - ret = reopen_rw(w); - if (ret != 0) +/* 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) +{ + int ret; + struct list_head stream_list; + + ret = prepare_stream_list(wim, image, &stream_list); + if (ret) return ret; + return write_stream_list(&stream_list, + wim->lookup_table, + wim->out_fp, + wimlib_get_compression_type(wim), + write_flags, + num_threads, + progress_func); +} - fp = w->fp; - - /* The old integrity table is still OK, as the SHA1 message digests in - * the integrity table include neither the header nor the XML data. - * Save it for later if it exists and an integrity table was required. - * */ - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY && - w->hdr.integrity.offset != 0) { - DEBUG("Reading existing integrity table.\n"); - integrity_table = MALLOC(w->hdr.integrity.size); - if (!integrity_table) - return WIMLIB_ERR_NOMEM; +/* + * Finish writing a WIM file: write the lookup table, xml data, and integrity + * table (optional), then overwrite the WIM header. + * + * 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. + * + * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: + * Don't write the lookup table. + * + * (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_FSYNC: + * fsync() the output file before closing it. + * + */ +int +finish_write(WIMStruct *w, int image, int write_flags, + wimlib_progress_func_t progress_func) +{ + int ret; + struct wim_header hdr; + FILE *out = w->out_fp; + + /* @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)); - ret = read_uncompressed_resource(fp, w->hdr.integrity.offset, - w->hdr.integrity.original_size, - integrity_table); - if (ret != 0) - goto err; - DEBUG("Done reading existing integrity table.\n"); + /* 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; } - DEBUG("Overwriting XML data.\n"); - /* Overwrite the XML data. */ - if (fseeko(fp, w->hdr.xml_res_entry.offset, SEEK_SET) != 0) { - ERROR("Failed to seek to byte %"PRIu64" for XML data: " - "%m\n", w->hdr.xml_res_entry.offset); - ret = WIMLIB_ERR_WRITE; - goto err; + /* 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); + } else { + copy_resource_entry(&hdr.boot_metadata_res_entry, + &w->image_metadata[ hdr.boot_idx- 1 + ]->metadata_lte->output_resource_entry); } - ret = write_xml_data(w->wim_info, WIM_ALL_IMAGES, fp); - if (ret != 0) - goto err; - DEBUG("Updating XML resource entry.\n"); - /* Update the XML resource entry in the WIM header. */ - xml_end = ftello(fp); - if (xml_end == -1) { - ret = WIMLIB_ERR_WRITE; - goto err; - } - xml_size = xml_end - w->hdr.xml_res_entry.offset; - w->hdr.xml_res_entry.size = xml_size; - w->hdr.xml_res_entry.original_size = xml_size; - - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - DEBUG("Writing integrity table.\n"); - w->hdr.integrity.offset = xml_end; - if (integrity_table) { - /* The existing integrity table was saved. */ - bytes_written = fwrite(integrity_table, 1, - w->hdr.integrity.size, fp); - if (bytes_written != w->hdr.integrity.size) { - ERROR("Failed to write integrity table: %m\n"); + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); + if (ret) + goto out_close_wim; + } + + ret = write_xml_data(w->wim_info, image, out, + (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? + wim_info_get_total_bytes(w->wim_info) : 0, + &hdr.xml_res_entry); + if (ret) + goto out_close_wim; + + 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); + if (fseeko(out, 0, SEEK_SET)) { + ERROR_WITH_ERRNO("Failed to seek to beginning " + "of WIM being written"); ret = WIMLIB_ERR_WRITE; - goto err; + goto out_close_wim; } - FREE(integrity_table); + ret = write_header(&checkpoint_hdr, out); + if (ret) + goto out_close_wim; + + if (fflush(out) != 0) { + ERROR_WITH_ERRNO("Can't write data to WIM"); + ret = WIMLIB_ERR_WRITE; + goto out_close_wim; + } + + if (fseeko(out, 0, SEEK_END) != 0) { + ERROR_WITH_ERRNO("Failed to seek to end " + "of WIM being written"); + ret = WIMLIB_ERR_WRITE; + goto out_close_wim; + } + } + + 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 { - /* There was no existing integrity table, so a new one - * must be calculated. */ - ret = write_integrity_table(fp, WIM_HEADER_DISK_SIZE, - w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size, - flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); - if (ret != 0) - goto err; - - off_t integrity_size = ftello(fp) - xml_end; - w->hdr.integrity.size = integrity_size; - w->hdr.integrity.original_size = integrity_size; - w->hdr.integrity.flags = 0; + 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(out, + &hdr.integrity, + new_lookup_table_end, + old_lookup_table_end, + progress_func); + if (ret) + goto out_close_wim; } else { - DEBUG("Truncating file to end of XML data.\n"); - /* No integrity table to write. The file should be truncated - * because it's possible that the old file was longer (due to it - * including an integrity table, or due to its XML data being - * longer) */ - if (fflush(fp) != 0) { - ERROR("Failed to flush stream for file `%s': %m\n", - w->filename); - return WIMLIB_ERR_WRITE; - } - if (ftruncate(fileno(fp), xml_end) != 0) { - ERROR("Failed to truncate `%s' to %"PRIu64" " - "bytes: %m\n", - w->filename, xml_end); - return WIMLIB_ERR_WRITE; - } - memset(&w->hdr.integrity, 0, sizeof(struct resource_entry)); + zero_resource_entry(&hdr.integrity); } - DEBUG("Overwriting header.\n"); - /* Overwrite the header. */ - if (fseeko(fp, 0, SEEK_SET) != 0) { - ERROR("Failed to seek to beginning of `%s': %m\n", - w->filename); - return WIMLIB_ERR_WRITE; + if (fseeko(out, 0, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to beginning of WIM " + "being written"); + ret = WIMLIB_ERR_WRITE; + goto out_close_wim; } - ret = write_header(&w->hdr, fp); - if (ret != 0) - return ret;; + ret = write_header(&hdr, out); + if (ret) + goto out_close_wim; - DEBUG("Closing file.\n"); - if (fclose(fp) != 0) { - ERROR("Failed to close `%s': %m\n", w->filename); - return WIMLIB_ERR_WRITE; + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + if (fflush(out) != 0 + || fsync(fileno(out)) != 0) + { + ERROR_WITH_ERRNO("Error flushing data to WIM file"); + ret = WIMLIB_ERR_WRITE; + } } - w->fp = NULL; - DEBUG("Done.\n"); - return 0; -err: - FREE(integrity_table); +out_close_wim: + if (fclose(out) != 0) { + ERROR_WITH_ERRNO("Failed to close the output WIM file"); + if (ret == 0) + ret = WIMLIB_ERR_WRITE; + } + w->out_fp = NULL; return ret; } -/* Write the metadata resource for the current image. */ -static int write_metadata_resource(WIMStruct *w) +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +int +lock_wim(WIMStruct *w, FILE *fp) { - FILE *out; - u8 *buf; - u8 *p; - int ret; - off_t subdir_offset; - struct dentry *root; - struct lookup_table_entry *lte; - struct resource_entry *res_entry; - off_t metadata_offset; - u64 metadata_original_size; - u64 metadata_compressed_size; - int metadata_ctype; - u8 hash[WIM_HASH_SIZE]; - - DEBUG("Writing metadata resource for image %u\n", w->current_image); - - out = w->out_fp; - root = wim_root_dentry(w); - metadata_ctype = wimlib_get_compression_type(w); - metadata_offset = ftello(out); - if (metadata_offset == -1) - return WIMLIB_ERR_WRITE; - - subdir_offset = 8 + root->length + 8; - calculate_subdir_offsets(root, &subdir_offset); - metadata_original_size = subdir_offset; - buf = MALLOC(metadata_original_size); - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for " - "metadata resource\n", metadata_original_size); - return WIMLIB_ERR_NOMEM; + int ret = 0; + if (fp && !w->wim_locked) { + ret = flock(fileno(fp), 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); + ret = WIMLIB_ERR_ALREADY_LOCKED; + } else { + WARNING_WITH_ERRNO("Failed to lock `%"TS"'", + w->filename); + ret = 0; + } + } else { + w->wim_locked = 1; + } } - p = buf; - #if 0 - /* Write the security data. */ - p = write_security_data(wim_security_data(w), p); - #else - p = put_u32(p, 8); /* Total length of security data. */ - p = put_u32(p, 0); /* Number of security data entries. */ - #endif - - DEBUG("Writing dentry tree.\n"); - p = write_dentry_tree(root, p); - - /* Like file resources, the lookup table entry for a metadata resource - * uses for the hash code a SHA1 message digest of its uncompressed - * contents. */ - sha1_buffer(buf, metadata_original_size, hash); - - ret = write_resource_from_memory(buf, - metadata_ctype, - metadata_original_size, - out, - &metadata_compressed_size); - FREE(buf); - if (ret != 0) - return ret; + return ret; +} +#endif - /* Update the lookup table entry, including the hash and output resource - * entry fields, for this image's metadata resource. */ - lte = wim_metadata_lookup_table_entry(w); - res_entry = <e->output_resource_entry; - lte->out_refcnt++; - if (memcmp(hash, lte->hash, WIM_HASH_SIZE) != 0) { - lookup_table_unlink(w->lookup_table, lte); - memcpy(lte->hash, hash, WIM_HASH_SIZE); - lookup_table_insert(w->lookup_table, lte); - } - res_entry->original_size = metadata_original_size; - res_entry->offset = metadata_offset; - res_entry->size = metadata_compressed_size; - res_entry->flags = WIM_RESHDR_FLAG_METADATA; - if (metadata_ctype != WIM_COMPRESSION_TYPE_NONE) - res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; - return 0; +static int +open_wim_writable(WIMStruct *w, const tchar *path, + bool trunc, bool also_readable) +{ + const tchar *mode; + if (trunc) + if (also_readable) + mode = T("w+b"); + else + mode = T("wb"); + else + mode = T("r+b"); + + wimlib_assert(w->out_fp == NULL); + w->out_fp = tfopen(path, mode); + if (w->out_fp) { + return 0; + } else { + ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path); + return WIMLIB_ERR_OPEN; + } } -/* Write the file resources for the current image. */ -static int write_file_resources(WIMStruct *w) + +void +close_wim_writable(WIMStruct *w) { + if (w->out_fp) { + if (fclose(w->out_fp) != 0) { + WARNING_WITH_ERRNO("Failed to close output WIM"); + } + w->out_fp = NULL; + } +} - DEBUG("Writing file resources for image %u\n", w->current_image); - return for_dentry_in_tree(wim_root_dentry(w), write_file_resource, w); +/* Open file stream and write dummy header for WIM. */ +int +begin_write(WIMStruct *w, const tchar *path, int write_flags) +{ + int ret; + ret = open_wim_writable(w, path, true, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); + if (ret) + return ret; + /* Write dummy header. It will be overwritten later. */ + return write_header(&w->hdr, w->out_fp); } -/* Write lookup table, xml data, lookup table, and rewrite header */ -static int finish_write(WIMStruct *w, int image, FILE *out, int flags) +/* 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) { - off_t lookup_table_offset; - off_t xml_data_offset; - off_t lookup_table_size; - off_t integrity_offset; - off_t xml_data_size; - off_t end_offset; - off_t integrity_size; int ret; - int i; - struct wim_header hdr; - lookup_table_offset = ftello(out); - if (lookup_table_offset == -1) - return WIMLIB_ERR_WRITE; + if (!path) + return WIMLIB_ERR_INVALID_PARAM; - DEBUG("Writing lookup table.\n"); - /* Write the lookup table. */ - ret = write_lookup_table(w->lookup_table, out); - if (ret != 0) - return ret; + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - DEBUG("Writing XML data.\n"); + if (image != WIMLIB_ALL_IMAGES && + (image < 1 || image > w->hdr.image_count)) + return WIMLIB_ERR_INVALID_IMAGE; - xml_data_offset = ftello(out); - if (xml_data_offset == -1) - return WIMLIB_ERR_WRITE; + if (w->hdr.total_parts != 1) { + ERROR("Cannot call wimlib_write() on part of a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; + } - /* @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)); - lookup_table_size = xml_data_offset - lookup_table_offset; - hdr.lookup_table_res_entry.offset = lookup_table_offset; - hdr.lookup_table_res_entry.size = lookup_table_size; - hdr.lookup_table_res_entry.original_size = lookup_table_size; - hdr.lookup_table_res_entry.flags = WIM_RESHDR_FLAG_METADATA; + ret = begin_write(w, path, write_flags); + if (ret) + goto out_close_wim; - ret = write_xml_data(w->wim_info, image, out); - if (ret != 0) - return ret; + ret = write_wim_streams(w, image, write_flags, num_threads, + progress_func); + if (ret) + goto out_close_wim; - integrity_offset = ftello(out); - if (integrity_offset == -1) - return WIMLIB_ERR_WRITE; - xml_data_size = integrity_offset - xml_data_offset; - - hdr.xml_res_entry.offset = xml_data_offset; - hdr.xml_res_entry.size = xml_data_size; - hdr.xml_res_entry.original_size = xml_data_size; - hdr.xml_res_entry.flags = 0; - - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - ret = write_integrity_table(out, WIM_HEADER_DISK_SIZE, - xml_data_offset, - flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); - if (ret != 0) - return ret; - end_offset = ftello(out); - if (end_offset == -1) - return WIMLIB_ERR_WRITE; - integrity_size = end_offset - integrity_offset; - hdr.integrity.offset = integrity_offset; - hdr.integrity.size = integrity_size; - hdr.integrity.original_size = integrity_size; - } else { - hdr.integrity.offset = 0; - hdr.integrity.size = 0; - hdr.integrity.original_size = 0; - } - hdr.integrity.flags = 0; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); - DEBUG("Updating WIM header.\n"); + ret = for_image(w, image, write_metadata_resource); + if (ret) + goto out_close_wim; + + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + + 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); + return ret; +} +static bool +any_images_modified(WIMStruct *w) +{ + for (int i = 0; i < w->hdr.image_count; i++) + if (w->image_metadata[i]->modified) + return true; + return false; +} - /* - * 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 || !w->image_metadata - || (image != WIM_ALL_IMAGES && image != hdr.boot_idx)) { - memset(&hdr.boot_metadata_res_entry, 0, - sizeof(struct resource_entry)); - } else { - memcpy(&hdr.boot_metadata_res_entry, - &w->image_metadata[hdr.boot_idx - 1].lookup_table_entry-> - output_resource_entry, - sizeof(struct resource_entry)); +/* + * Overwrite a WIM, possibly appending streams to it. + * + * A WIM looks like (or is supposed to look like) the following: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * If we are not adding any streams or metadata resources, the lookup table is + * unchanged--- so we only need to overwrite the XML data, integrity table, and + * header. This operation is potentially unsafe if the program is abruptly + * terminated while the XML data or integrity table are being overwritten, but + * before the new header has been written. To partially alleviate this problem, + * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to + * finish_write() to cause a temporary WIM header to be written after the XML + * data has been written. This may prevent the WIM from becoming corrupted if + * the program is terminated while the integrity table is being calculated (but + * no guarantees, due to write re-ordering...). + * + * If we are adding new streams or images (metadata resources), the lookup table + * needs to be changed, and those streams need to be written. In this case, we + * try to perform a safe update of the WIM file by writing the streams *after* + * the end of the previous WIM, then writing the new lookup table, XML data, and + * (optionally) integrity table following the new streams. This will produce a + * layout like the following: + * + * Header (212 bytes) + * (OLD) Streams and metadata resources (variable size) + * (OLD) Lookup table (variable size) + * (OLD) XML data (variable size) + * (OLD) Integrity table (optional) (variable size) + * (NEW) Streams and metadata resources (variable size) + * (NEW) Lookup table (variable size) + * (NEW) XML data (variable size) + * (NEW) Integrity table (optional) (variable size) + * + * At all points, the WIM is valid as nothing points to the new data yet. Then, + * the header is overwritten to point to the new lookup table, XML data, and + * integrity table, to produce the following layout: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Nothing (variable size) + * More Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * 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 + * 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) +{ + int ret; + struct list_head stream_list; + off_t old_wim_end; + u64 old_lookup_table_end, old_xml_begin, old_xml_end; + + DEBUG("Overwriting `%"TS"' in-place", w->filename); + + /* 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; } - /* Set image count and boot index correctly for single image writes */ - if (image != WIM_ALL_IMAGES) { - hdr.image_count = 1; - if (hdr.boot_idx == image) - hdr.boot_idx = 1; - else - hdr.boot_idx = 0; + 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; } + /* 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 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 + * immediately after the lookup table. Note that this may + * overwrite an existing integrity table. */ + DEBUG("Skipping writing lookup table " + "(no images modified or deleted)"); + 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) { + /* Old WIM has an integrity table; begin writing new streams + * after it. */ + old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size; + } 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); + if (ret) + return ret; + + ret = open_wim_writable(w, w->filename, false, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); + if (ret) + return ret; + + ret = lock_wim(w, w->out_fp); + if (ret) { + close_wim_writable(w); + return ret; + } - if (fseeko(out, 0, SEEK_SET) != 0) + if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + close_wim_writable(w); + w->wim_locked = 0; return WIMLIB_ERR_WRITE; + } - return write_header(&hdr, out); + DEBUG("Writing newly added streams (offset = %"PRIu64")", + old_wim_end); + ret = write_stream_list(&stream_list, + w->lookup_table, + w->out_fp, + wimlib_get_compression_type(w), + write_flags, + num_threads, + progress_func); + 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); +out_truncate: + close_wim_writable(w); + if (ret != 0 && !(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; + return ret; } -/* Writes the WIM to a file. */ -WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, int image, int flags) +static int +overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { + size_t wim_name_len; int ret; - const char *mode; - FILE *out; - if (image != WIM_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) - return WIMLIB_ERR_INVALID_IMAGE; - - if (image == WIM_ALL_IMAGES) - DEBUG("Writing all images to `%s'\n", path); - else - DEBUG("Writing image %d to `%s'\n", image, path); - - /* checking the integrity requires going back over the file to read it. - * XXX - * (It also would be possible to keep a running sha1sum as the file - * as written-- this would be faster, but a bit more complicated) */ - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - mode = "w+b"; - else - mode = "wb"; + DEBUG("Overwriting `%"TS"' via a temporary file", w->filename); - out = fopen(path, mode); - if (!out) { - ERROR("Failed to open the file `%s' for writing!\n", - path); - return WIMLIB_ERR_OPEN; + /* Write the WIM to a temporary file in the same directory as the + * original WIM. */ + wim_name_len = tstrlen(w->filename); + tchar tmpfile[wim_name_len + 10]; + tmemcpy(tmpfile, w->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); + if (ret) { + ERROR("Failed to write the WIM file `%"TS"'", tmpfile); + goto out_unlink; } - w->out_fp = out; + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); - /* Write dummy header. It will be overwritten later. */ - ret = write_header(&w->hdr, out); - if (ret != 0) - goto done; +#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; + } +#endif - for_lookup_table_entry(w->lookup_table, zero_out_refcnts, NULL); + /* Rename the new file to the old file .*/ + if (trename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", + tmpfile, w->filename); + ret = WIMLIB_ERR_RENAME; + goto out_unlink; + } - ret = for_image(w, image, write_file_resources); - if (ret != 0) { - ERROR("Failed to write file resources!\n"); - goto done; + if (progress_func) { + union wimlib_progress_info progress; + progress.rename.from = tmpfile; + progress.rename.to = w->filename; + progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } - ret = for_image(w, image, write_metadata_resource); + /* Close the original WIM file that was opened for reading. */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } - if (ret != 0) { - ERROR("Failed to write image metadata!\n"); - goto done; + /* Re-open the WIM read-only. */ + w->fp = tfopen(w->filename, T("rb")); + if (w->fp == NULL) { + ret = WIMLIB_ERR_REOPEN; + WARNING_WITH_ERRNO("Failed to re-open `%"TS"' read-only", + w->filename); + FREE(w->filename); + w->filename = NULL; } + goto out; +out_unlink: + /* Remove temporary file. */ + if (tunlink(tmpfile) != 0) + WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); +out: + return ret; +} - ret = finish_write(w, image, out, flags); +/* + * Writes a WIM file to the original file that it was read from, overwriting it. + */ +WIMLIBAPI int +wimlib_overwrite(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) +{ + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; -done: - DEBUG("Closing output file.\n"); - w->out_fp = NULL; - if (fclose(out) != 0) { - ERROR("Failed to close the file `%s': %m\n", path); - ret = WIMLIB_ERR_WRITE; + if (!w->filename) + return WIMLIB_ERR_NO_FILENAME; + + if (w->hdr.total_parts != 1) { + ERROR("Cannot modify a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; } - 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 + return ret; + } + return overwrite_wim_via_tmpfile(w, write_flags, num_threads, + progress_func); }