X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fwrite.c;h=e7107ab1d392bb0a740d304737677117e7f6a8a2;hb=bbedc3e1934278c2ff2746b25b688477cffc9d83;hp=6b67498009edbfdb7f344568d28cababc75ede28;hpb=1449ca7e974c27bec75cb6784873be3c0e6eb78a;p=wimlib diff --git a/src/write.c b/src/write.c index 6b674980..e7107ab1 100644 --- a/src/write.c +++ b/src/write.c @@ -3,966 +3,3167 @@ * * 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 "wimlib_internal.h" -#include "io.h" -#include "lookup_table.h" -#include "dentry.h" -#include "sha1.h" -#include "lzx.h" -#include "xml.h" -#include "xpress.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 "wimlib/list.h" so that "wimlib/list.h" can + * overwrite the LIST_HEAD macro. */ +# include +#endif + +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/header.h" +#include "wimlib/integrity.h" +#include "wimlib/lookup_table.h" +#include "wimlib/metadata.h" +#include "wimlib/resource.h" +#include "wimlib/write.h" +#include "wimlib/xml.h" + +#ifdef __WIN32__ +# include "wimlib/win32.h" /* win32_get_number_of_processors() */ +#endif + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +# include +#endif + +#include +#include +#include +#include #include +#ifdef HAVE_ALLOCA_H +# include +#endif -/* Used for buffering FILE IO */ -#define BUFFER_SIZE 4096 +#ifndef __WIN32__ +# include /* for `struct iovec' */ +#endif -/* - * Copies bytes between two file streams. - * - * Copies @len bytes from @in to @out, at the current position in @out, and at - * an offset of @in_offset in @in. - */ -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; - } +/* 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) +{ + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) + return false; + if (lte->resource_location != RESOURCE_IN_WIM) + return false; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + return false; + if (lte->rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + return false; + if (!(lte->rspec->flags & WIM_RESHDR_FLAG_COMPRESSED)) + return false; + if (lte->rspec->wim->compression_type != out_ctype) + return false; + if (lte->rspec->wim->chunk_size != out_chunk_size) + return false; + return true; +} - if (fwrite(buf, 1, n, out) != n) { - ERROR("Error copying data between files: %m\n"); - return WIMLIB_ERR_WRITE; + +/* Return true if the specified resource must be recompressed when the specified + * output parameters are used. */ +static bool +must_compress_stream(const struct wim_lookup_table_entry *lte, + int write_resource_flags, int out_ctype, u32 out_chunk_size) +{ + return (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE + && ((write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) + || !can_raw_copy(lte, write_resource_flags, + out_ctype, out_chunk_size))); +} + +static unsigned +compress_chunk(const void * uncompressed_data, + unsigned uncompressed_len, + void *compressed_data, + int out_ctype, + struct wimlib_lzx_context *comp_ctx) +{ + switch (out_ctype) { + case WIMLIB_COMPRESSION_TYPE_XPRESS: + return wimlib_xpress_compress(uncompressed_data, + uncompressed_len, + compressed_data); + case WIMLIB_COMPRESSION_TYPE_LZX: + return wimlib_lzx_compress2(uncompressed_data, + uncompressed_len, + compressed_data, + comp_ctx); + case WIMLIB_COMPRESSION_TYPE_LZMS: + /* TODO */ + WARNING("LZMS compression not yet implemented!"); + return 0; + + default: + wimlib_assert(0); + return 0; + } +} + +/* Chunk table that's located at the beginning of each compressed resource in + * the WIM. (This is not the on-disk format; the on-disk format just has an + * array of offsets.) */ +struct chunk_table { + u64 original_resource_size; + u64 num_chunks; + u64 table_disk_size; + unsigned bytes_per_chunk_entry; + void *cur_offset_p; + union { + u32 cur_offset_u32; + u64 cur_offset_u64; + }; + /* Beginning of chunk offsets, in either 32-bit or 64-bit little endian + * integers, including the first offset of 0, which will not be written. + * */ + u8 offsets[] _aligned_attribute(8); +}; + +/* Allocate and initializes a chunk table, then reserve space for it in the + * output file unless writing a pipable resource. */ +static int +begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, + struct filedes *out_fd, + u32 out_chunk_size, + struct chunk_table **chunk_tab_ret, + int resource_flags) +{ + u64 size; + u64 num_chunks; + unsigned bytes_per_chunk_entry; + size_t alloc_size; + struct chunk_table *chunk_tab; + int ret; + + size = lte->size; + num_chunks = DIV_ROUND_UP(size, out_chunk_size); + bytes_per_chunk_entry = (size > (1ULL << 32)) ? 8 : 4; + alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); + chunk_tab = CALLOC(1, alloc_size); + + if (!chunk_tab) { + ERROR("Failed to allocate chunk table for %"PRIu64" byte " + "resource", size); + return WIMLIB_ERR_NOMEM; + } + chunk_tab->num_chunks = num_chunks; + chunk_tab->original_resource_size = size; + chunk_tab->bytes_per_chunk_entry = bytes_per_chunk_entry; + chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * + (num_chunks - 1); + chunk_tab->cur_offset_p = chunk_tab->offsets; + + /* We don't know the correct offsets yet; so just write zeroes to + * reserve space for the table, so we can go back to it later after + * we've written the compressed chunks following it. + * + * Special case: if writing a pipable WIM, compressed resources are in a + * modified format (see comment above write_pipable_wim()) and do not + * have a chunk table at the beginning, so don't reserve any space for + * one. */ + if (!(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { + ret = full_write(out_fd, chunk_tab->offsets, + chunk_tab->table_disk_size); + if (ret) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + FREE(chunk_tab); + return ret; } - len -= n; } + *chunk_tab_ret = chunk_tab; return 0; } +/* Add the offset for the next chunk to the chunk table being constructed for a + * compressed stream. */ +static void +chunk_tab_record_chunk(struct chunk_table *chunk_tab, unsigned out_chunk_size) +{ + if (chunk_tab->bytes_per_chunk_entry == 4) { + *(le32*)chunk_tab->cur_offset_p = cpu_to_le32(chunk_tab->cur_offset_u32); + chunk_tab->cur_offset_p = (le32*)chunk_tab->cur_offset_p + 1; + chunk_tab->cur_offset_u32 += out_chunk_size; + } else { + *(le64*)chunk_tab->cur_offset_p = cpu_to_le64(chunk_tab->cur_offset_u64); + chunk_tab->cur_offset_p = (le64*)chunk_tab->cur_offset_p + 1; + chunk_tab->cur_offset_u64 += out_chunk_size; + } +} -/* - * Uncompresses a WIM file resource and writes it uncompressed to a file stream. - * - * @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. +/* Finishes a WIM chunk table and writes it to the output file at the correct + * offset. */ +static int +finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, + struct filedes *out_fd, + off_t res_start_offset, + int write_resource_flags) +{ + int ret; + + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + ret = full_write(out_fd, + chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + chunk_tab->table_disk_size); + } else { + ret = full_pwrite(out_fd, + chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + chunk_tab->table_disk_size, + res_start_offset); + } + if (ret) + ERROR_WITH_ERRNO("Write error"); + return ret; +} + +/* Write the header for a stream in a pipable WIM. */ -static int uncompress_resource(FILE *in, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out) +static int +write_pwm_stream_header(const struct wim_lookup_table_entry *lte, + struct filedes *out_fd, + int additional_reshdr_flags) { + struct pwm_stream_hdr stream_hdr; + u32 reshdr_flags; 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; - if (fwrite(buf, 1, uncompressed_chunk_size, out) != - uncompressed_chunk_size) { - ERROR("Failed to write file resource: %m\n"); - return WIMLIB_ERR_WRITE; + stream_hdr.magic = PWM_STREAM_MAGIC; + stream_hdr.uncompressed_size = cpu_to_le64(lte->size); + if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) { + zero_out_hash(stream_hdr.hash); + } else { + wimlib_assert(!lte->unhashed); + copy_hash(stream_hdr.hash, lte->hash); + } + + reshdr_flags = lte->flags & ~(WIM_RESHDR_FLAG_COMPRESSED | WIM_RESHDR_FLAG_PACKED_STREAMS); + reshdr_flags |= additional_reshdr_flags; + stream_hdr.flags = cpu_to_le32(reshdr_flags); + ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr)); + if (ret) + ERROR_WITH_ERRNO("Error writing stream header"); + return ret; +} + +static int +seek_and_truncate(struct filedes *out_fd, off_t offset) +{ + if (filedes_seek(out_fd, offset) == -1 || + ftruncate(out_fd->fd, offset)) + { + ERROR_WITH_ERRNO("Failed to truncate output WIM file"); + return WIMLIB_ERR_WRITE; + } + return 0; +} + +static int +finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) +{ + u8 md[SHA1_HASH_SIZE]; + + sha1_final(md, sha_ctx); + if (lte->unhashed) { + copy_hash(lte->hash, md); + } else if (!hashes_equal(md, lte->hash)) { + ERROR("WIM resource has incorrect hash!"); + if (lte_filename_valid(lte)) { + ERROR("We were reading it from \"%"TS"\"; maybe " + "it changed while we were reading it.", + lte->file_on_disk); } + return WIMLIB_ERR_INVALID_RESOURCE_HASH; } return 0; } -/* - * 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. +struct write_resource_ctx { + int out_ctype; + u32 out_chunk_size; + struct wimlib_lzx_context *comp_ctx; + struct chunk_table *chunk_tab; + struct filedes *out_fd; + SHA_CTX sha_ctx; + bool doing_sha; + int resource_flags; +}; + +static int +write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) +{ + struct write_resource_ctx *ctx = _ctx; + const void *out_chunk; + unsigned out_chunk_size; + int ret; + void *compressed_chunk = NULL; + unsigned compressed_size; + bool compressed_chunk_malloced = false; + + if (ctx->doing_sha) + sha1_update(&ctx->sha_ctx, chunk, chunk_size); + + out_chunk = chunk; + out_chunk_size = chunk_size; + if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + + /* Compress the chunk. */ + if (chunk_size <= STACK_MAX) { + compressed_chunk = alloca(chunk_size); + } else { + compressed_chunk = MALLOC(chunk_size); + if (compressed_chunk == NULL) + return WIMLIB_ERR_NOMEM; + compressed_chunk_malloced = true; + } + + compressed_size = compress_chunk(chunk, chunk_size, + compressed_chunk, + ctx->out_ctype, + ctx->comp_ctx); + /* Use compressed data if compression to less than input size + * was successful. */ + if (compressed_size) { + out_chunk = compressed_chunk; + out_chunk_size = compressed_size; + } + } + + if (ctx->chunk_tab) { + /* Update chunk table accounting. */ + chunk_tab_record_chunk(ctx->chunk_tab, out_chunk_size); + + /* If writing compressed chunks to a pipable WIM, before the + * chunk data write a chunk header that provides the compressed + * chunk size. */ + if (ctx->resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + struct pwm_chunk_hdr chunk_hdr = { + .compressed_size = cpu_to_le32(out_chunk_size), + }; + ret = full_write(ctx->out_fd, &chunk_hdr, + sizeof(chunk_hdr)); + if (ret) + goto error; + } + } + + /* Write the chunk data. */ + ret = full_write(ctx->out_fd, out_chunk, out_chunk_size); + if (ret) + goto error; + +out_free_memory: + if (compressed_chunk_malloced) + FREE(compressed_chunk); + return ret; + +error: + ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); + goto out_free_memory; +} + +/* + * write_wim_resource()- + * + * Write a resource to an output WIM. + * + * @lte: + * Lookup table entry for the resource, which could be in another WIM, in + * an external file, or in another location. + * + * @out_fd: + * File descriptor opened to the output WIM. + * + * @out_ctype: + * One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate which + * compression algorithm to use. + * + * @out_chunk_size: + * Compressed chunk size to use. + * + * @out_reshdr: + * On success, this is filled in with the offset, flags, compressed size, + * and uncompressed size of the resource in the output WIM. + * + * @resource_flags: + * * WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed even + * if it could otherwise be copied directly from the input; + * * WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE if writing a resource for a pipable WIM + * (and the output file descriptor may be a pipe). + * + * @comp_ctx: + * Location of LZX compression context pointer, which will be allocated or + * updated if needed. (Initialize to NULL.) + * + * Additional notes: The SHA1 message digest of the uncompressed data is + * calculated (except when doing a raw copy --- see below). If the @unhashed + * flag is set on the lookup table entry, this message digest is simply copied + * to it; otherwise, the message digest is compared with the existing one, and + * this function will fail if they do not match. */ -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 +write_wim_resource(struct wim_lookup_table_entry *lte, + struct filedes *out_fd, int out_ctype, + u32 out_chunk_size, + struct wim_reshdr *out_reshdr, + int resource_flags, + struct wimlib_lzx_context **comp_ctx) { + struct write_resource_ctx write_ctx; + off_t res_start_offset; + u64 read_size; int ret; - int (*compress)(const void *, uint, void *, uint *); - if (output_ctype == WIM_COMPRESSION_TYPE_LZX) - compress = lzx_compress; + + /* Mask out any irrelevant flags, since this function also uses this + * variable to store WIMLIB_READ_RESOURCE flags. */ + resource_flags &= WIMLIB_WRITE_RESOURCE_MASK; + + /* Get current position in output WIM. */ + res_start_offset = out_fd->offset; + + /* If we are not forcing the data to be recompressed, and the input + * resource is located in a WIM with a compression mode compatible with + * the output, we can simply copy the compressed data without + * recompressing it. This also means we must skip calculating the SHA1, + * as we never will see the uncompressed data. */ + if (can_raw_copy(lte, resource_flags, out_ctype, out_chunk_size)) { + /* Normally, for raw copies we can request a RAW_FULL read, but + * if we're reading from a pipable resource and writing a + * non-pipable resource or vice versa, then a RAW_CHUNKS read + * needs to be requested so that the written resource can be + * appropriately formatted. However, in neither case is any + * actual decompression needed. */ + if (lte->rspec->is_pipable == !!(resource_flags & + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + { + resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL; + read_size = lte->rspec->size_in_wim; + } else { + resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS; + read_size = lte->size; + } + write_ctx.doing_sha = false; + } else { + write_ctx.doing_sha = true; + sha1_init(&write_ctx.sha_ctx); + read_size = lte->size; + } + + /* Set the output compression mode and initialize chunk table if needed. + */ + write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + write_ctx.out_chunk_size = out_chunk_size; + write_ctx.chunk_tab = NULL; + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + wimlib_assert(out_chunk_size > 0); + if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) { + /* Compression needed. */ + write_ctx.out_ctype = out_ctype; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = wimlib_lzx_alloc_context(out_chunk_size, + NULL, comp_ctx); + if (ret) + goto out; + } + write_ctx.comp_ctx = *comp_ctx; + } + if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)) { + /* Chunk table needed. */ + ret = begin_wim_resource_chunk_tab(lte, out_fd, + out_chunk_size, + &write_ctx.chunk_tab, + resource_flags); + if (ret) + goto out; + } + } + + /* If writing a pipable resource, write the stream header and update + * @res_start_offset to be the end of the stream header. */ + if (resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + int reshdr_flags = 0; + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) + reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; + ret = write_pwm_stream_header(lte, out_fd, reshdr_flags); + if (ret) + goto out_free_chunk_tab; + res_start_offset = out_fd->offset; + } + + /* Write the entire resource by reading the entire resource and feeding + * the data through write_resource_cb(). */ + write_ctx.out_fd = out_fd; + write_ctx.resource_flags = resource_flags; +try_write_again: + ret = read_stream_prefix(lte, read_size, write_resource_cb, + &write_ctx, resource_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; + } + + /* Write chunk table if needed. */ + if (write_ctx.chunk_tab) { + ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, + out_fd, + res_start_offset, + resource_flags); + if (ret) + goto out_free_chunk_tab; + } + + /* Fill in out_reshdr with information about the newly written + * resource. */ + out_reshdr->size_in_wim = out_fd->offset - res_start_offset; + out_reshdr->flags = lte->flags & ~WIM_RESHDR_FLAG_PACKED_STREAMS; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + out_reshdr->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; else - compress = xpress_compress; + out_reshdr->flags |= WIM_RESHDR_FLAG_COMPRESSED; + out_reshdr->offset_in_wim = res_start_offset; + out_reshdr->uncompressed_size = lte->size; + + /* Check for resources compressed to greater than their original size + * and write them uncompressed instead. (But never do this if writing + * to a pipe, and don't bother if we did a raw copy.) */ + if (out_reshdr->size_in_wim > out_reshdr->uncompressed_size && + !(resource_flags & (WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE | + WIMLIB_READ_RESOURCE_FLAG_RAW))) + { + DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " + "writing uncompressed instead", + out_reshdr->uncompressed_size, out_reshdr->size_in_wim); + ret = seek_and_truncate(out_fd, res_start_offset); + if (ret) + goto out_free_chunk_tab; + out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + FREE(write_ctx.chunk_tab); + write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + write_ctx.chunk_tab = NULL; + write_ctx.doing_sha = false; + goto try_write_again; + } + if (resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW) { + DEBUG("Copied raw compressed data " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_reshdr->uncompressed_size, out_reshdr->size_in_wim, + out_reshdr->offset_in_wim, out_reshdr->flags); + } else if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + DEBUG("Wrote compressed resource " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_reshdr->uncompressed_size, out_reshdr->size_in_wim, + out_reshdr->offset_in_wim, out_reshdr->flags); + } else { + DEBUG("Wrote uncompressed resource " + "(%"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_reshdr->uncompressed_size, + out_reshdr->offset_in_wim, out_reshdr->flags); + } + ret = 0; +out_free_chunk_tab: + FREE(write_ctx.chunk_tab); +out: + return ret; +} + +/* Like write_wim_resource(), but the resource is specified by a buffer of + * uncompressed data rather a lookup table entry. Also writes the SHA1 message + * digest of the buffer to @hash_ret if it is non-NULL. */ +int +write_wim_resource_from_buffer(const void *buf, size_t buf_size, + int reshdr_flags, struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wim_reshdr *out_reshdr, + u8 *hash_ret, int write_resource_flags, + struct wimlib_lzx_context **comp_ctx) +{ + int ret; + struct wim_lookup_table_entry *lte; + + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ - u8 uncompressed_buf[WIM_CHUNK_SIZE]; - u8 compressed_buf[WIM_CHUNK_SIZE - 1]; + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; - /* Determine how many compressed chunks the file needs to be divided - * into. */ - u64 num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = (void*)buf; + lte->size = buf_size; + lte->flags = reshdr_flags; - u64 num_chunk_entries = num_chunks - 1; + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + sha1_buffer(buf, buf_size, lte->hash); + lte->unhashed = 0; + } else { + lte->unhashed = 1; + } - /* 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; + ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, + out_reshdr, write_resource_flags, comp_ctx); + if (ret) + goto out_free_lte; + if (hash_ret) + copy_hash(hash_ret, lte->hash); + ret = 0; +out_free_lte: + lte->resource_location = RESOURCE_NONEXISTENT; + free_lookup_table_entry(lte); + return ret; +} - /* Array in which to construct the chunk offset table. */ - u64 chunk_offsets[num_chunk_entries]; - /* Offset of the start of the chunk table in the output file. */ - off_t chunk_tab_offset = ftello(out); +#ifdef ENABLE_MULTITHREADED_COMPRESSION - /* Total size of the chunk table (as written to the file) */ - u64 chunk_tab_size = chunk_entry_size * num_chunk_entries; +/* 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; +}; - /* 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"); - return WIMLIB_ERR_WRITE; +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; +} - /* 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++) { +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); +} - u64 uncompressed_offset = i * WIM_CHUNK_SIZE; - u64 uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); +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); - const u8 *uncompressed_p; - if (uncompressed_resource != NULL) { - uncompressed_p = uncompressed_resource + - uncompressed_offset; + q->back = (q->back + 1) % q->size; + q->array[q->back] = obj; + q->filled_slots++; + pthread_cond_broadcast(&q->msg_avail_cond); + pthread_mutex_unlock(&q->lock); +} + +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); + + obj = q->array[q->front]; + q->array[q->front] = NULL; + q->front = (q->front + 1) % q->size; + q->filled_slots--; + + 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; + int out_ctype; + struct wimlib_lzx_context *comp_ctx; +}; + +#define MAX_CHUNKS_PER_MSG 2 + +struct message { + struct wim_lookup_table_entry *lte; + u32 out_chunk_size; + u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; + u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; + unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; + struct iovec out_chunks[MAX_CHUNKS_PER_MSG]; + unsigned num_chunks; + struct list_head list; + bool complete; + u64 begin_chunk; +}; + +static void +compress_chunks(struct message *msg, int out_ctype, + struct wimlib_lzx_context *comp_ctx) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) { + unsigned len; + + len = compress_chunk(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i], + out_ctype, + comp_ctx); + + void *out_chunk; + unsigned out_len; + if (len) { + /* To be written compressed */ + out_chunk = msg->compressed_chunks[i]; + out_len = len; + } else { + /* To be written uncompressed */ + out_chunk = msg->uncompressed_chunks[i]; + out_len = msg->uncompressed_chunk_sizes[i]; + } + msg->out_chunks[i].iov_base = out_chunk; + msg->out_chunks[i].iov_len = out_len; + } +} + +/* 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; + struct message *msg; + + DEBUG("Compressor thread ready"); + while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { + compress_chunks(msg, params->out_ctype, params->comp_ctx); + shared_queue_put(compressed_res_queue, msg); + } + DEBUG("Compressor thread terminating"); + return NULL; +} +#endif /* ENABLE_MULTITHREADED_COMPRESSION */ + +struct write_streams_progress_data { + wimlib_progress_func_t progress_func; + union wimlib_progress_info progress; + uint64_t next_progress; + WIMStruct *prev_wim_part; +}; + +static void +do_write_streams_progress(struct write_streams_progress_data *progress_data, + struct wim_lookup_table_entry *lte, + bool stream_discarded) +{ + union wimlib_progress_info *progress = &progress_data->progress; + bool new_wim_part; + + if (stream_discarded) { + progress->write_streams.total_bytes -= lte->size; + 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 += lte->size; + } + new_wim_part = false; + if (lte->resource_location == RESOURCE_IN_WIM && + lte->rspec->wim != progress_data->prev_wim_part) + { + if (progress_data->prev_wim_part) { + new_wim_part = true; + progress->write_streams.completed_parts++; + } + progress_data->prev_wim_part = lte->rspec->wim; + } + progress->write_streams.completed_streams++; + if (progress_data->progress_func + && (progress->write_streams.completed_bytes >= progress_data->next_progress + || new_wim_part)) + { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress_data->next_progress == progress->write_streams.total_bytes) { + progress_data->next_progress = ~(uint64_t)0; } 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) + progress_data->next_progress = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); + } + } +} + +struct serial_write_stream_ctx { + struct filedes *out_fd; + int out_ctype; + u32 out_chunk_size; + struct wimlib_lzx_context **comp_ctx; + int write_resource_flags; +}; + +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, + ctx->out_chunk_size, + <e->out_reshdr, + ctx->write_resource_flags, + ctx->comp_ctx); +} + + +/* 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. */ +static int +do_write_stream_list(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + int (*write_stream_cb)(struct wim_lookup_table_entry *, void *), + void *write_stream_ctx, + struct write_streams_progress_data *progress_data) +{ + int ret = 0; + struct wim_lookup_table_entry *lte; + bool stream_discarded; + + /* For each stream in @stream_list ... */ + while (!list_empty(stream_list)) { + stream_discarded = false; + lte = container_of(stream_list->next, + struct wim_lookup_table_entry, + write_streams_list); + list_del(<e->write_streams_list); + if (lte->unhashed && !lte->unique_size) { + /* Unhashed stream that shares a size with some other + * stream in the WIM we are writing. The stream must be + * checksummed to know if we need to write it or not. */ + struct wim_lookup_table_entry *tmp; + u32 orig_out_refcnt = lte->out_refcnt; + + ret = hash_unhashed_stream(lte, lookup_table, &tmp); + if (ret) + break; + if (tmp != lte) { + /* We found a duplicate stream. 'lte' was + * freed, so replace it with the duplicate. */ + lte = tmp; + + /* 'out_refcnt' was transferred to the + * duplicate, and we can detect if the duplicate + * stream was already referenced for writing by + * checking if its 'out_refcnt' is higher than + * that of the original stream. In such cases, + * the current stream can be discarded. We can + * also discard the current stream if it was + * previously marked as filtered (e.g. already + * present in the WIM being written). */ + if (lte->out_refcnt > orig_out_refcnt || + lte->filtered) { + DEBUG("Discarding duplicate stream of " + "length %"PRIu64, + lte->size); + lte->no_progress = 0; + stream_discarded = true; + goto skip_to_progress; + } + } + } + + /* Here, @lte is either a hashed stream or an unhashed stream + * with a unique size. In either case we know that the stream + * has to be written. In either case the SHA1 message digest + * will be calculated over the stream while writing it; however, + * in the former case this is done merely to check the data, + * while in the latter case this is done because we do not have + * the SHA1 message digest yet. */ + wimlib_assert(lte->out_refcnt != 0); + lte->deferred = 0; + lte->no_progress = 0; + ret = (*write_stream_cb)(lte, write_stream_ctx); + if (ret) + break; + /* In parallel mode, some streams are deferred for later, + * serialized processing; ignore them here. */ + if (lte->deferred) + continue; + if (lte->unhashed) { + list_del(<e->unhashed_list); + lookup_table_insert(lookup_table, lte); + lte->unhashed = 0; + } + skip_to_progress: + if (!lte->no_progress) { + do_write_streams_progress(progress_data, + lte, stream_discarded); + } + } + return ret; +} + +static int +do_write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wimlib_lzx_context **comp_ctx, + int write_resource_flags, + struct write_streams_progress_data *progress_data) +{ + struct serial_write_stream_ctx ctx = { + .out_fd = out_fd, + .out_ctype = out_ctype, + .out_chunk_size = out_chunk_size, + .write_resource_flags = write_resource_flags, + .comp_ctx = comp_ctx, + }; + return do_write_stream_list(stream_list, + lookup_table, + serial_write_stream, + &ctx, + progress_data); +} + +static inline int +write_flags_to_resource_flags(int write_flags) +{ + int resource_flags = 0; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS; + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; + return resource_flags; +} + +static int +write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wimlib_lzx_context **comp_ctx, + int write_resource_flags, + struct write_streams_progress_data *progress_data) +{ + union wimlib_progress_info *progress = &progress_data->progress; + DEBUG("Writing stream list of size %"PRIu64" (serial version)", + progress->write_streams.total_streams); + progress->write_streams.num_threads = 1; + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } + return do_write_stream_list_serial(stream_list, + lookup_table, + out_fd, + out_ctype, + out_chunk_size, + comp_ctx, + write_resource_flags, + progress_data); +} + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +static int +write_wim_chunks(struct message *msg, struct filedes *out_fd, + struct chunk_table *chunk_tab, + int write_resource_flags) +{ + struct iovec *vecs; + struct pwm_chunk_hdr *chunk_hdrs; + unsigned nvecs; + int ret; + + for (unsigned i = 0; i < msg->num_chunks; i++) + chunk_tab_record_chunk(chunk_tab, msg->out_chunks[i].iov_len); + + if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { + nvecs = msg->num_chunks; + vecs = msg->out_chunks; + } else { + /* Special case: If writing a compressed resource to a pipable + * WIM, prefix each compressed chunk with a header that gives + * its compressed size. */ + nvecs = msg->num_chunks * 2; + vecs = alloca(nvecs * sizeof(vecs[0])); + chunk_hdrs = alloca(msg->num_chunks * sizeof(chunk_hdrs[0])); + + for (unsigned i = 0; i < msg->num_chunks; i++) { + chunk_hdrs[i].compressed_size = cpu_to_le32(msg->out_chunks[i].iov_len); + vecs[i * 2].iov_base = &chunk_hdrs[i]; + vecs[i * 2].iov_len = sizeof(chunk_hdrs[i]); + vecs[i * 2 + 1].iov_base = msg->out_chunks[i].iov_base; + vecs[i * 2 + 1].iov_len = msg->out_chunks[i].iov_len; + } + } + ret = full_writev(out_fd, vecs, nvecs); + if (ret) + ERROR_WITH_ERRNO("Write error"); + return ret; +} + +struct main_writer_thread_ctx { + struct list_head *stream_list; + struct wim_lookup_table *lookup_table; + struct filedes *out_fd; + off_t res_start_offset; + int out_ctype; + u32 out_chunk_size; + struct wimlib_lzx_context **comp_ctx; + int write_resource_flags; + struct shared_queue *res_to_compress_queue; + struct shared_queue *compressed_res_queue; + size_t num_messages; + struct write_streams_progress_data *progress_data; + + struct list_head available_msgs; + struct list_head outstanding_streams; + struct list_head serial_streams; + size_t num_outstanding_messages; + + SHA_CTX next_sha_ctx; + u64 next_chunk; + u64 next_num_chunks; + struct wim_lookup_table_entry *next_lte; + + struct message *msgs; + struct message *next_msg; + struct chunk_table *cur_chunk_tab; +}; + +static int +init_message(struct message *msg, u32 out_chunk_size) +{ + msg->out_chunk_size = out_chunk_size; + for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { + msg->compressed_chunks[i] = MALLOC(out_chunk_size); + msg->uncompressed_chunks[i] = MALLOC(out_chunk_size); + if (msg->compressed_chunks[i] == NULL || + msg->uncompressed_chunks[i] == NULL) + return WIMLIB_ERR_NOMEM; + } + return 0; +} + +static void +destroy_message(struct message *msg) +{ + for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { + FREE(msg->compressed_chunks[i]); + FREE(msg->uncompressed_chunks[i]); + } +} + +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, u32 out_chunk_size) +{ + struct message *msgs; + + msgs = CALLOC(num_messages, sizeof(struct message)); + if (msgs == NULL) + return NULL; + for (size_t i = 0; i < num_messages; i++) { + if (init_message(&msgs[i], out_chunk_size)) { + free_messages(msgs, num_messages); + return NULL; + } + } + 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); +} + +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, ctx->out_chunk_size); + if (ctx->msgs == NULL) + 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; + + /* Message currently being prepared. */ + ctx->next_msg = NULL; + + /* Resources that don't need any chunks compressed are added to this + * list and written directly by the main thread. */ + INIT_LIST_HEAD(&ctx->serial_streams); + + /* Pointer to chunk table for stream currently being written. */ + ctx->cur_chunk_tab = 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; + + 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) { + /* First set of chunks. */ + + /* Write pipable WIM stream header if needed. */ + if (ctx->write_resource_flags & + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) + { + ret = write_pwm_stream_header(cur_lte, ctx->out_fd, + WIM_RESHDR_FLAG_COMPRESSED); + if (ret) + return ret; + } + + /* Save current offset. */ + ctx->res_start_offset = ctx->out_fd->offset; + + /* Begin building the chunk table, and leave space for + * it if needed. */ + ret = begin_wim_resource_chunk_tab(cur_lte, + ctx->out_fd, + ctx->out_chunk_size, + &ctx->cur_chunk_tab, + ctx->write_resource_flags); + if (ret) return ret; - uncompressed_p = uncompressed_buf; } - if (i != 0) - chunk_offsets[i - 1] = cur_chunk_offset; + /* Write the compressed chunks from the message. */ + ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab, + ctx->write_resource_flags); + if (ret) + return ret; - uint compressed_len; + /* Was this the last chunk of the stream? If so, finish the + * stream by writing the chunk table. */ + if (list_empty(&cur_lte->msg_list) && + msg->begin_chunk + msg->num_chunks == ctx->cur_chunk_tab->num_chunks) + { + u64 res_csize; + + ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, + ctx->out_fd, + ctx->res_start_offset, + ctx->write_resource_flags); + if (ret) + return ret; - ret = compress(uncompressed_p, uncompressed_chunk_size, - compressed_buf, &compressed_len); + list_del(&cur_lte->being_compressed_list); + + res_csize = ctx->out_fd->offset - ctx->res_start_offset; + + FREE(ctx->cur_chunk_tab); + ctx->cur_chunk_tab = NULL; + + /* Check for resources compressed to greater than or + * equal to their original size and write them + * uncompressed instead. (But never do this if writing + * to a pipe.) */ + if (res_csize >= cur_lte->size && + !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + { + DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " + "writing uncompressed instead", + cur_lte->size, res_csize); + ret = seek_and_truncate(ctx->out_fd, ctx->res_start_offset); + if (ret) + return ret; + ret = write_wim_resource(cur_lte, + ctx->out_fd, + WIMLIB_COMPRESSION_TYPE_NONE, + 0, + &cur_lte->out_reshdr, + ctx->write_resource_flags, + ctx->comp_ctx); + if (ret) + return ret; + } else { + cur_lte->out_reshdr.size_in_wim = + res_csize; + + cur_lte->out_reshdr.uncompressed_size = + cur_lte->size; + + cur_lte->out_reshdr.offset_in_wim = + ctx->res_start_offset; + + cur_lte->out_reshdr.flags = + cur_lte->flags | + WIM_RESHDR_FLAG_COMPRESSED; + + DEBUG("Wrote compressed resource " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + cur_lte->out_reshdr.uncompressed_size, + cur_lte->out_reshdr.size_in_wim, + cur_lte->out_reshdr.offset_in_wim, + cur_lte->out_reshdr.flags); + } - /* 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; + do_write_streams_progress(ctx->progress_data, + cur_lte, false); + + /* Since we just finished writing a stream, write any + * streams that have been added to the serial_streams + * list for direct writing by the main thread (e.g. + * resources that don't need to be compressed because + * the desired compression type is the same as the + * previous compression type). */ + if (!list_empty(&ctx->serial_streams)) { + ret = do_write_stream_list_serial(&ctx->serial_streams, + ctx->lookup_table, + ctx->out_fd, + ctx->out_ctype, + ctx->out_chunk_size, + ctx->comp_ctx, + ctx->write_resource_flags, + ctx->progress_data); + if (ret) + return ret; + } + + /* 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. */ +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; + u64 next_chunk_in_msg; + + /* Update SHA1 message digest for the stream currently being read by the + * main thread. */ + sha1_update(&ctx->next_sha_ctx, chunk, chunk_size); + + /* We send chunks of data to the compressor chunks in batches which we + * refer to as "messages". @next_msg is the message that is currently + * being prepared to send off. If it is NULL, that indicates that we + * need to start a new message. */ + next_msg = ctx->next_msg; + if (!next_msg) { + /* We need to start a new message. First check to see if there + * is a message available in the list of available messages. If + * so, we can just take one. If not, all the messages (there is + * a fixed number of them, proportional to the number of + * threads) have been sent off to the compressor threads, so we + * receive messages from the compressor threads containing + * compressed chunks of data. + * + * We may need to receive multiple messages before one is + * actually available to use because messages received that are + * *not* for the very next set of chunks to compress must be + * buffered until it's time to write those chunks. */ + while (list_empty(&ctx->available_msgs)) { + ret = receive_compressed_chunks(ctx); + if (ret) + return ret; + } + + 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; + } + return 0; +} + +static int +main_writer_thread_finish(void *_ctx) +{ + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + while (ctx->num_outstanding_messages != 0) { + ret = receive_compressed_chunks(ctx); + 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->out_chunk_size, + ctx->comp_ctx, + ctx->write_resource_flags, + ctx->progress_data); +} + +static int +submit_stream_for_compression(struct wim_lookup_table_entry *lte, + struct main_writer_thread_ctx *ctx) +{ + int ret; + + /* Read the entire stream @lte, feeding its data chunks to the + * compressor threads. Also SHA1-sum the stream; this is required in + * the case that @lte is unhashed, and a nice additional verification + * when @lte is already hashed. */ + sha1_init(&ctx->next_sha_ctx); + ctx->next_chunk = 0; + ctx->next_num_chunks = DIV_ROUND_UP(lte->size, ctx->out_chunk_size); + ctx->next_lte = lte; + INIT_LIST_HEAD(<e->msg_list); + list_add_tail(<e->being_compressed_list, &ctx->outstanding_streams); + ret = read_stream_prefix(lte, lte->size, main_writer_thread_cb, ctx, 0); + if (ret) + return ret; + wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); + return finalize_and_check_sha1(&ctx->next_sha_ctx, lte); +} + +static int +main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) +{ + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + + if (lte->size < 1000 || + !must_compress_stream(lte, ctx->write_resource_flags, + ctx->out_ctype, ctx->out_chunk_size)) + { + /* Stream is too small or isn't being compressed. Process it by + * the main thread when we have a chance. We can't necessarily + * process it right here, as the main thread could be in the + * middle of writing a different stream. */ + 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; +} + +static long +get_default_num_threads(void) +{ +#ifdef __WIN32__ + return win32_get_number_of_processors(); +#else + return sysconf(_SC_NPROCESSORS_ONLN); +#endif +} + +/* 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 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, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wimlib_lzx_context **comp_ctx, + int write_resource_flags, + struct write_streams_progress_data *progress_data, + unsigned num_threads) +{ + int ret; + struct shared_queue res_to_compress_queue; + struct shared_queue compressed_res_queue; + pthread_t *compressor_threads = NULL; + union wimlib_progress_info *progress = &progress_data->progress; + unsigned num_started_threads; + bool can_retry = true; + + if (num_threads == 0) { + long nthreads = get_default_num_threads(); + if (nthreads < 1 || nthreads > UINT_MAX) { + WARNING("Could not determine number of processors! Assuming 1"); + goto out_serial_quiet; + } else if (nthreads == 1) { + goto out_serial_quiet; } else { - buf_to_write = uncompressed_p; - len_to_write = uncompressed_chunk_size; + num_threads = nthreads; } + } - 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; + DEBUG("Writing stream list of size %"PRIu64" " + "(parallel version, num_threads=%u)", + progress->write_streams.total_streams, 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 = CALLOC(num_threads, sizeof(params[0])); + if (params == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } + + for (unsigned i = 0; i < num_threads; i++) { + params[i].res_to_compress_queue = &res_to_compress_queue; + params[i].compressed_res_queue = &compressed_res_queue; + params[i].out_ctype = out_ctype; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = wimlib_lzx_alloc_context(out_chunk_size, + NULL, ¶ms[i].comp_ctx); + if (ret) + goto out_free_params; + } + } + + compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + if (compressor_threads == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out_free_params; + } + + for (unsigned i = 0; i < num_threads; i++) { + DEBUG("pthread_create thread %u of %u", i + 1, num_threads); + ret = pthread_create(&compressor_threads[i], NULL, + compressor_thread_proc, ¶ms[i]); + if (ret) { + errno = ret; + ret = -1; + ERROR_WITH_ERRNO("Failed to create compressor " + "thread %u of %u", + i + 1, num_threads); + num_started_threads = i; + goto out_join; + } + } + num_started_threads = num_threads; + + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } + + struct main_writer_thread_ctx ctx; + + memset(&ctx, 0, sizeof(ctx)); + + ctx.stream_list = stream_list; + ctx.lookup_table = lookup_table; + ctx.out_fd = out_fd; + ctx.out_ctype = out_ctype; + ctx.out_chunk_size = out_chunk_size; + ctx.comp_ctx = comp_ctx; + ctx.res_to_compress_queue = &res_to_compress_queue; + ctx.compressed_res_queue = &compressed_res_queue; + ctx.num_messages = queue_size; + ctx.write_resource_flags = write_resource_flags; + ctx.progress_data = progress_data; + ret = main_writer_thread_init_ctx(&ctx); + if (ret) + goto out_join; + + can_retry = false; + ret = do_write_stream_list(stream_list, lookup_table, + main_thread_process_next_stream, + &ctx, progress_data); + if (ret) + goto out_destroy_ctx; + + /* The main thread has finished reading all streams that are going to be + * compressed in parallel, and it now needs to wait for all remaining + * chunks to be compressed so that the remaining streams can actually be + * written to the output file. Furthermore, any remaining streams that + * had processing deferred to the main thread need to be handled. These + * tasks are done by the main_writer_thread_finish() function. */ + ret = main_writer_thread_finish(&ctx); +out_destroy_ctx: + main_writer_thread_destroy_ctx(&ctx); +out_join: + for (unsigned i = 0; i < num_started_threads; i++) + shared_queue_put(&res_to_compress_queue, NULL); + + for (unsigned i = 0; i < num_started_threads; i++) { + if (pthread_join(compressor_threads[i], NULL)) { + WARNING_WITH_ERRNO("Failed to join compressor " + "thread %u of %u", + i + 1, num_threads); + } + } + FREE(compressor_threads); +out_free_params: + for (unsigned i = 0; i < num_threads; i++) + wimlib_lzx_free_context(params[i].comp_ctx); + FREE(params); +out_destroy_compressed_res_queue: + shared_queue_destroy(&compressed_res_queue); +out_destroy_res_to_compress_queue: + shared_queue_destroy(&res_to_compress_queue); + if (!can_retry || (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, + out_chunk_size, + comp_ctx, + write_resource_flags, + progress_data); + +} +#endif + +/* Write a list of streams to a WIM (@out_fd) using the compression type + * @out_ctype, chunk size @out_chunk_size, and up to @num_threads compressor + * threads. */ +static int +write_stream_list(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + struct filedes *out_fd, int out_ctype, + u32 out_chunk_size, + struct wimlib_lzx_context **comp_ctx, + int write_flags, + unsigned num_threads, wimlib_progress_func_t progress_func) +{ + int ret; + int write_resource_flags; + u64 total_bytes; + u64 total_compression_bytes; + unsigned total_parts; + WIMStruct *prev_wim_part; + size_t num_streams; + struct wim_lookup_table_entry *lte; + struct write_streams_progress_data progress_data; + + if (list_empty(stream_list)) { + DEBUG("No streams to write."); + return 0; + } + + write_resource_flags = write_flags_to_resource_flags(write_flags); + + DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", + out_fd->offset, write_resource_flags); + + /* Sort the stream list into a good order for reading. */ + ret = sort_stream_list_by_sequential_order(stream_list, + offsetof(struct wim_lookup_table_entry, + write_streams_list)); + if (ret) + return ret; + + /* Calculate the total size of the streams to be written. Note: this + * will be the uncompressed size, as we may not know the compressed size + * yet, and also this will assume that every unhashed stream will be + * written (which will not necessarily be the case). */ + total_bytes = 0; + total_compression_bytes = 0; + num_streams = 0; + total_parts = 0; + prev_wim_part = NULL; + list_for_each_entry(lte, stream_list, write_streams_list) { + num_streams++; + total_bytes += lte->size; + if (must_compress_stream(lte, write_resource_flags, + out_ctype, out_chunk_size)) + total_compression_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++; + } } - 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; + memset(&progress_data, 0, sizeof(progress_data)); + progress_data.progress_func = progress_func; + + progress_data.progress.write_streams.total_bytes = total_bytes; + progress_data.progress.write_streams.total_streams = num_streams; + progress_data.progress.write_streams.completed_bytes = 0; + progress_data.progress.write_streams.completed_streams = 0; + progress_data.progress.write_streams.num_threads = num_threads; + progress_data.progress.write_streams.compression_type = out_ctype; + progress_data.progress.write_streams.total_parts = total_parts; + progress_data.progress.write_streams.completed_parts = 0; + + progress_data.next_progress = 0; + progress_data.prev_wim_part = NULL; + +#ifdef ENABLE_MULTITHREADED_COMPRESSION + if (total_compression_bytes >= 2000000 && num_threads != 1) + ret = write_stream_list_parallel(stream_list, + lookup_table, + out_fd, + out_ctype, + out_chunk_size, + comp_ctx, + write_resource_flags, + &progress_data, + num_threads); + else +#endif + ret = write_stream_list_serial(stream_list, + lookup_table, + out_fd, + out_ctype, + out_chunk_size, + comp_ctx, + write_resource_flags, + &progress_data); + if (ret == 0) + DEBUG("Successfully wrote stream list."); + else + DEBUG("Failed to write stream list (ret=%d).", ret); + return ret; +} + +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) +{ + FREE(tab->array); +} + +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(lte->size) % tab->capacity; + lte->unique_size = 1; + hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) { + if (same_size_lte->size == lte->size) { + lte->unique_size = 0; + same_size_lte->unique_size = 0; + break; + } + } + + hlist_add_head(<e->hash_list_2, &tab->array[pos]); + tab->num_entries++; + return 0; +} + +struct find_streams_ctx { + WIMStruct *wim; + int write_flags; + struct list_head stream_list; + struct stream_size_table stream_size_tab; +}; + +static void +lte_reference_for_logical_write(struct wim_lookup_table_entry *lte, + struct find_streams_ctx *ctx, + unsigned nref) +{ + if (lte->out_refcnt == 0) { + stream_size_table_insert(lte, &ctx->stream_size_tab); + list_add_tail(<e->write_streams_list, &ctx->stream_list); + } + lte->out_refcnt += nref; +} + +static int +do_lte_full_reference_for_logical_write(struct wim_lookup_table_entry *lte, + void *_ctx) +{ + struct find_streams_ctx *ctx = _ctx; + lte->out_refcnt = 0; + lte_reference_for_logical_write(lte, ctx, + (lte->refcnt ? lte->refcnt : 1)); + return 0; +} + +static int +inode_find_streams_to_write(struct wim_inode *inode, + struct wim_lookup_table *table, + struct find_streams_ctx *ctx) +{ + struct wim_lookup_table_entry *lte; + unsigned i; + + for (i = 0; i <= inode->i_num_ads; i++) { + lte = inode_stream_lte(inode, i, table); + if (lte) + lte_reference_for_logical_write(lte, ctx, inode->i_nlink); + else if (!is_zero_hash(inode_stream_hash(inode, i))) + return WIMLIB_ERR_RESOURCE_NOT_FOUND; + } + return 0; +} + +static int +image_find_streams_to_write(WIMStruct *wim) +{ + struct find_streams_ctx *ctx; + struct wim_image_metadata *imd; + struct wim_inode *inode; + struct wim_lookup_table_entry *lte; + int ret; + + ctx = wim->private; + imd = wim_get_current_image_metadata(wim); - /* Now that all entries of the chunk table are determined, rewind the - * stream to where the chunk table was, and write it back out. */ + image_for_each_unhashed_stream(lte, imd) + lte->out_refcnt = 0; - if (fseeko(out, chunk_tab_offset, SEEK_SET) != 0) { - ERROR("Failed to seek to beginning of chunk table: %m\n"); - return WIMLIB_ERR_READ; + /* Go through this image's inodes to find any streams that have not been + * found yet. */ + image_for_each_inode(inode, imd) { + ret = inode_find_streams_to_write(inode, wim->lookup_table, ctx); + if (ret) + return ret; } + return 0; +} - if (chunk_entry_size == 8) { - array_to_le64(chunk_offsets, num_chunk_entries); +/* + * Build a list of streams (via `struct wim_lookup_table_entry's) included in + * the "logical write" of the WIM, meaning all streams that are referenced at + * least once by dentries in the the image(s) being written. 'out_refcnt' on + * each stream being included in the logical write is set to the number of + * references from dentries in the image(s). Furthermore, 'unique_size' on each + * stream being included in the logical write is set to indicate whether that + * stream has a unique size relative to the streams being included in the + * logical write. Still furthermore, 'part_number' on each stream being + * included in the logical write is set to the part number given in the + * in-memory header of @p wim. + * + * This is considered a "logical write" because it does not take into account + * filtering out streams already present in the WIM (in the case of an in place + * overwrite) or present in other WIMs (in case of creating delta WIM). + */ +static int +prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, + struct find_streams_ctx *ctx) +{ + int ret; - if (fwrite(chunk_offsets, 1, chunk_tab_size, out) != - chunk_tab_size) { - ERROR("Failed to write chunk table: %m\n"); - return WIMLIB_ERR_WRITE; + 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; + + for_lookup_table_entry(wim->lookup_table, + do_lte_full_reference_for_logical_write, ctx); + for (i = 0; i < wim->hdr.image_count; i++) { + imd = wim->image_metadata[i]; + image_for_each_unhashed_stream(lte, imd) + do_lte_full_reference_for_logical_write(lte, ctx); } } 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"); - return WIMLIB_ERR_WRITE; - } + /* Slow case: Walk through the images being written and + * determine the streams referenced. */ + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + wim->private = ctx; + ret = for_image(wim, image, image_find_streams_to_write); + if (ret) + return ret; } - if (fseeko(out, 0, SEEK_END) != 0) { - ERROR("Failed to seek to end of output file: %m\n"); - return WIMLIB_ERR_WRITE; + return 0; +} + +static int +process_filtered_stream(struct wim_lookup_table_entry *lte, void *_ctx) +{ + struct find_streams_ctx *ctx = _ctx; + u16 filtered = 0; + + /* Calculate and set lte->filtered. */ + if (lte->resource_location == RESOURCE_IN_WIM) { + if (lte->rspec->wim == ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE)) + filtered |= FILTERED_SAME_WIM; + if (lte->rspec->wim != ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)) + filtered |= FILTERED_EXTERNAL_WIM; } + lte->filtered = filtered; + + /* Filtered streams get inserted into the stream size table too, unless + * they already were. This is because streams that are checksummed + * on-the-fly during the write should not be written if they are + * duplicates of filtered stream. */ + if (lte->filtered && lte->out_refcnt == 0) + stream_size_table_insert(lte, &ctx->stream_size_tab); + return 0; +} +static int +mark_stream_not_filtered(struct wim_lookup_table_entry *lte, void *_ignore) +{ + lte->filtered = 0; return 0; } -int write_resource_from_memory(const u8 resource[], int out_ctype, - u64 resource_original_size, FILE *out, - u64 *resource_size_ret) +/* Given the list of streams to include in a logical write of a WIM, handle + * filtering out streams already present in the WIM or already present in + * external WIMs, depending on the write flags provided. */ +static void +handle_stream_filtering(struct find_streams_ctx *ctx) { - 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; - } - *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. + struct wim_lookup_table_entry *lte, *tmp; + + if (!(ctx->write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE | + WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS))) + { + for_lookup_table_entry(ctx->wim->lookup_table, + mark_stream_not_filtered, ctx); + return; + } + + for_lookup_table_entry(ctx->wim->lookup_table, + process_filtered_stream, ctx); + + /* Streams in logical write list that were filtered can be removed. */ + list_for_each_entry_safe(lte, tmp, &ctx->stream_list, + write_streams_list) + if (lte->filtered) + list_del(<e->write_streams_list); +} + +/* Prepares list of streams to write for the specified WIM image(s). This wraps + * around prepare_logical_stream_list() to handle filtering out streams already + * present in the WIM or already present in external WIMs, depending on the + * write flags provided. + * + * Note: some additional data is stored in each `struct wim_lookup_table_entry': + * + * - 'out_refcnt' is set to the number of references found for the logical write. + * This will be nonzero on all streams in the list returned by this function, + * but will also be nonzero on streams not in the list that were included in + * the logical write list, but filtered out from the returned list. + * - 'filtered' is set to nonzero if the stream was filtered. Filtered streams + * are not included in the list of streams returned by this function. + * - 'unique_size' is set if the stream has a unique size among all streams in + * the logical write plus any filtered streams in the entire WIM that could + * potentially turn out to have the same checksum as a yet-to-be-checksummed + * stream being written. */ -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) +static int +prepare_stream_list(WIMStruct *wim, int image, int write_flags, + struct list_head *stream_list) { int ret; + bool streams_ok; + struct find_streams_ctx ctx; - /* Handle zero-length files */ - if (original_size == 0) { - memset(output_res_entry, 0, sizeof(*output_res_entry)); - return 0; - } + INIT_LIST_HEAD(&ctx.stream_list); + ret = init_stream_size_table(&ctx.stream_size_tab, + wim->lookup_table->capacity); + if (ret) + return ret; + ctx.write_flags = write_flags; + ctx.wim = wim; - /* 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; - } + streams_ok = ((write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) != 0); - if (output_ctype == input_ctype) { - /* The same compression types; simply copy the resource. */ + ret = prepare_logical_stream_list(wim, image, streams_ok, &ctx); + if (ret) + goto out_destroy_table; - ret = copy_between_files(in, offset, out, size); - if (ret != 0) + handle_stream_filtering(&ctx); + list_transfer(&ctx.stream_list, stream_list); + ret = 0; +out_destroy_table: + destroy_stream_size_table(&ctx.stream_size_tab); + return ret; +} + +static int +write_wim_streams(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func, + struct list_head *stream_list_override) +{ + int ret; + struct list_head _stream_list; + struct list_head *stream_list; + struct wim_lookup_table_entry *lte; + + if (stream_list_override == NULL) { + /* Normal case: prepare stream list from image(s) being written. + */ + stream_list = &_stream_list; + ret = prepare_stream_list(wim, image, write_flags, stream_list); + if (ret) 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) - return ret; - output_res_entry->size = new_size; - } - + /* 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; + list_for_each_entry(lte, stream_list, write_streams_list) + lte->out_refcnt = (lte->refcnt ? lte->refcnt : 1); } - 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; - return 0; + return write_stream_list(stream_list, + wim->lookup_table, + &wim->out_fd, + wim->out_compression_type, + wim->out_chunk_size, + &wim->lzx_context, + write_flags, + num_threads, + progress_func); } -/* - * Writes a file resource to the output file. - * - * @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. - */ -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 +write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, + wimlib_progress_func_t progress_func) +{ int ret; + int start_image; + int end_image; + int write_resource_flags; - w = wim_p; - out = w->out_fp; - - /* Directories don't need file resources. */ - if (dentry_is_directory(dentry)) + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) { + DEBUG("Not writing any metadata resources."); return 0; + } - /* Get the lookup entry for the file resource. */ - lte = wim_lookup_resource(w, dentry); - if (!lte) - return 0; - - /* 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; - - out_wim_ctype = wimlib_get_compression_type(w); - output_res_entry = <e->output_resource_entry; - - /* 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) { - - /* Read from disk (uncompressed) */ - - len = lte->resource_entry.original_size; + write_resource_flags = write_flags_to_resource_flags(write_flags); - 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; - } + DEBUG("Writing metadata resources (offset=%"PRIu64")", + wim->out_fd.offset); - if (w->verbose) - puts(lte->file_on_disk); + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); - ret = transfer_file_resource(in, len, len, 0, - WIM_COMPRESSION_TYPE_NONE, out, - out_wim_ctype, output_res_entry); - fclose(in); + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; } else { + start_image = image; + end_image = image; + } - /* Read from input WIM (possibly compressed) */ - - /* 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; + 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); + wim_res_spec_to_hdr(imd->metadata_lte->rspec, + &imd->metadata_lte->out_reshdr); + ret = 0; } else { - /* Same WIM file. */ - in = w->fp; - in_wim_ctype = out_wim_ctype; - } - 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); + 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, + &imd->metadata_lte->out_reshdr, + write_resource_flags, + &wim->lzx_context); + } + if (ret) + return ret; } - return ret; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + return 0; } -/* Reopens the FILE* for a WIM read-write. */ -static int reopen_rw(WIMStruct *w) +static int +open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) { - FILE *fp; + int raw_fd; + DEBUG("Opening \"%"TS"\" for writing.", path); - 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); + 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; } - w->fp = fp; + 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 a WIM file to the original file that it was read from, overwriting it. +/* + * finish_write(): + * + * Finish writing a WIM file: write the lookup table, xml data, and integrity + * table, then overwrite the WIM header. By default, closes the WIM file + * descriptor (@wim->out_fd) 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_FSYNC: + * fsync() the output file before closing it. + * + * (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_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_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. */ -WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int flags) +static int +finish_write(WIMStruct *wim, int image, int write_flags, + wimlib_progress_func_t progress_func, + struct list_head *stream_list_override) { - const char *wimfile_name; - size_t wim_name_len; int ret; - - wimfile_name = w->filename; + off_t hdr_offset; + int write_resource_flags; + off_t old_lookup_table_end; + off_t new_lookup_table_end; + u64 xml_totalbytes; - DEBUG("Replacing WIM file `%s'\n", wimfile_name); + DEBUG("image=%d, write_flags=%08x", image, write_flags); - if (!wimfile_name) - return WIMLIB_ERR_NO_FILENAME; + write_resource_flags = write_flags_to_resource_flags(write_flags); - /* 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'; + /* 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 (wim->hdr.boot_idx == 0) { + zero_reshdr(&wim->hdr.boot_metadata_reshdr); + } else { + copy_reshdr(&wim->hdr.boot_metadata_reshdr, + &wim->image_metadata[wim->hdr.boot_idx- 1 + ]->metadata_lte->out_reshdr); + } - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, flags); - if (ret != 0) { - ERROR("Failed to write the WIM file `%s'!\n", tmpfile); - return ret; + /* 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_wim_lookup_table(wim, image, write_flags, + &wim->hdr.lookup_table_reshdr, + stream_list_override); + if (ret) + return ret; } - DEBUG("Closing original WIM file.\n"); - /* Close the original WIM file that was opened for reading. */ - if (w->fp) { - if (fclose(w->fp) != 0) { - WARNING("Failed to close the file `%s'\n", - wimfile_name); + /* 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) + 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, &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) + return ret; } - w->fp = NULL; + + if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) + old_lookup_table_end = 0; + + 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); + if (ret) + return ret; + } else { + /* No integrity table. */ + zero_reshdr(&wim->hdr.integrity_table_reshdr); } - DEBUG("Renaming `%s' to `%s'\n", tmpfile, wimfile_name); + /* 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) + 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) { + DEBUG("Syncing WIM file."); + if (fsync(wim->out_fd.fd)) { + ERROR_WITH_ERRNO("Error syncing data to WIM file"); + return WIMLIB_ERR_WRITE; + } + } - /* 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; + if (close_wim_writable(wim, write_flags)) { + ERROR_WITH_ERRNO("Failed to close the output WIM file"); + return WIMLIB_ERR_WRITE; } return 0; } +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +int +lock_wim(WIMStruct *wim, int fd) +{ + int ret = 0; + 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!", wim->filename); + ret = WIMLIB_ERR_ALREADY_LOCKED; + } else { + WARNING_WITH_ERRNO("Failed to lock `%"TS"'", + wim->filename); + ret = 0; + } + } else { + wim->wim_locked = 1; + } + } + return ret; +} +#endif -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *w, int flags) +/* + * 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 +write_pipable_wim(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, wimlib_progress_func_t progress_func, + struct list_head *stream_list_override) { int ret; - FILE *fp; - u8 *integrity_table = NULL; - off_t xml_end; - off_t xml_size; - size_t bytes_written; - - DEBUG("Overwriting XML and header of `%s', flags = %d\n", - w->filename, flags); - if (!w->filename) - return WIMLIB_ERR_NO_FILENAME; + 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; - ret = reopen_rw(w); - if (ret != 0) + /* Write extra copy of the XML data. */ + ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, + &xml_reshdr, + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE); + if (ret) return ret; - fp = w->fp; + /* Write metadata resources for the image(s) being included in the + * output WIM. */ + ret = write_wim_metadata_resources(wim, image, write_flags, + progress_func); + if (ret) + return ret; - /* 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; + /* Write streams needed for the image(s) being included in the output + * WIM, or streams needed for the split WIM part. */ + return write_wim_streams(wim, image, write_flags, num_threads, + progress_func, stream_list_override); - 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"); - } + /* The lookup table, XML data, and header at end are handled by + * finish_write(). */ +} - 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; +/* 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, + wimlib_progress_func_t progress_func, + 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 lt_stream_list_override; + + 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_REBUILD) + DEBUG("\tREBUILD"); + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + DEBUG("\tRECOMPRESS"); + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) + DEBUG("\tFSYNC"); + if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE) + DEBUG("\tFSYNC"); + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) + DEBUG("\tIGNORE_READONLY_FLAG"); + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + DEBUG("\tPIPABLE"); + if (write_flags & WIMLIB_WRITE_FLAG_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", (progress_func ? "yes" : "no")); + DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect")); + DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ? + "specified" : "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 > wim->hdr.image_count)) + return WIMLIB_ERR_INVALID_IMAGE; + + /* 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 and pipable flags. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE))) + if (wim_is_pipable(wim)) + 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)) + write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + + /* Set appropriate magic number. */ + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + wim->hdr.magic = PWM_MAGIC; + else + wim->hdr.magic = WIM_MAGIC; + + /* 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); + + /* Set SPANNED header flag if writing part of a split WIM. */ + if (total_parts != 1) + wim->hdr.flags |= WIM_HDR_FLAG_SPANNED; + + /* 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); } - 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; + /* Set chunk size if different. */ + wim->hdr.chunk_size = wim->out_chunk_size; + + /* Use GUID if specified; otherwise generate a new one. */ + if (guid) + memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN); + else if (!wim->guid_set_explicitly) + 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; } - 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"); - ret = WIMLIB_ERR_WRITE; - goto err; + + /* 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; } - FREE(integrity_table); - } 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; } + } 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; + /* 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_restore_hdr; + + if (stream_list_override) { + struct wim_lookup_table_entry *lte; + INIT_LIST_HEAD(<_stream_list_override); + list_for_each_entry(lte, stream_list_override, + write_streams_list) + { + list_add_tail(<e->lookup_table_list, + <_stream_list_override); } - memset(&w->hdr.integrity, 0, sizeof(struct resource_entry)); } - 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; + /* Write metadata resources and streams. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) { + /* Default case: create a normal (non-pipable) WIM. */ + ret = write_wim_streams(wim, image, write_flags, num_threads, + progress_func, stream_list_override); + if (ret) + goto out_restore_hdr; + + ret = write_wim_metadata_resources(wim, image, write_flags, + progress_func); + if (ret) + goto out_restore_hdr; + } else { + /* Non-default case: create pipable WIM. */ + ret = write_pipable_wim(wim, image, write_flags, num_threads, + progress_func, stream_list_override); + if (ret) + goto out_restore_hdr; + write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END; } - ret = write_header(&w->hdr, fp); - if (ret != 0) - return ret;; + if (stream_list_override) + stream_list_override = <_stream_list_override; - DEBUG("Closing file.\n"); - if (fclose(fp) != 0) { - ERROR("Failed to close `%s': %m\n", w->filename); - return WIMLIB_ERR_WRITE; - } - w->fp = NULL; - DEBUG("Done.\n"); - return 0; -err: - FREE(integrity_table); + /* Write lookup table, XML data, and (optional) integrity table. */ + ret = finish_write(wim, image, write_flags, progress_func, + stream_list_override); +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 the metadata resource for the current image. */ -int write_metadata_resource(WIMStruct *w) +/* 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, + wimlib_progress_func_t progress_func) { - 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; + return write_wim_part(wim, path_or_fd, image, write_flags, + num_threads, progress_func, 1, 1, NULL, NULL); +} - 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; - } - 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 +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write(WIMStruct *wim, const tchar *path, + int image, int write_flags, unsigned num_threads, + wimlib_progress_func_t progress_func) +{ + if (!path) + return WIMLIB_ERR_INVALID_PARAM; - 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; + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - /* 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; + return write_standalone_wim(wim, path, image, write_flags, + num_threads, progress_func); } -/* Write the file resources for the current image. */ -static int write_file_resources(WIMStruct *w) +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write_to_fd(WIMStruct *wim, int fd, + int image, int write_flags, unsigned num_threads, + wimlib_progress_func_t progress_func) { + if (fd < 0) + return WIMLIB_ERR_INVALID_PARAM; + + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR; - DEBUG("Writing file resources for image %u\n", w->current_image); - return for_dentry_in_tree(wim_root_dentry(w), write_file_resource, w); + return write_standalone_wim(wim, &fd, image, write_flags, + num_threads, progress_func); } -/* Write lookup table, xml data, lookup table, and rewrite header - * - * write_lt is zero iff the lookup table is not to be written; i.e. it is - * handled elsewhere. */ -int finish_write(WIMStruct *w, int image, int flags, int write_lt) +static bool +any_images_modified(WIMStruct *wim) { - 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; - FILE *out = w->out_fp; - - if (write_lt) { - lookup_table_offset = ftello(out); - if (lookup_table_offset == -1) - return WIMLIB_ERR_WRITE; + for (int i = 0; i < wim->hdr.image_count; i++) + if (wim->image_metadata[i]->modified) + return true; + return false; +} - DEBUG("Writing lookup table.\n"); - /* Write the lookup table. */ - ret = write_lookup_table(w->lookup_table, out); - if (ret != 0) - return ret; - } +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; - DEBUG("Writing XML data.\n"); + 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; +} - xml_data_offset = ftello(out); - if (xml_data_offset == -1) - return WIMLIB_ERR_WRITE; +/* 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; - /* @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)); - if (write_lt) { - 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 = hdr.lookup_table_res_entry.size; - hdr.lookup_table_res_entry.flags = WIM_RESHDR_FLAG_METADATA; - - ret = write_xml_data(w->wim_info, image, out); - if (ret != 0) + wim->private = &end_offset; + ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim); + if (ret) return ret; - 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) + for (i = 0; i < wim->hdr.image_count; i++) { + ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim); + if (ret) 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; + return 0; +} - DEBUG("Updating WIM header.\n"); +/* + * 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 *wim, 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; + struct wim_header hdr_save; + + DEBUG("Overwriting `%"TS"' in-place", wim->filename); + + /* 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 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 = wim->hdr.xml_data_reshdr.offset_in_wim; + old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim; + old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + + wim->hdr.lookup_table_reshdr.size_in_wim; + if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 && + wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) { + WARNING("Didn't expect the integrity table to be before the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } + if (old_lookup_table_end > old_xml_begin) { + WARNING("Didn't expect the lookup table to be after the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } - /* - * 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)); + /* 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 (!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 + * 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 (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) { + /* Old WIM has an integrity table; begin writing new streams + * after it. */ + old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim + + wim->hdr.integrity_table_reshdr.size_in_wim; } else { - memcpy(&hdr.boot_metadata_res_entry, - &w->image_metadata[hdr.boot_idx - 1].lookup_table_entry-> - output_resource_entry, - sizeof(struct resource_entry)); + /* No existing integrity table; begin writing new streams after + * the old XML data. */ + old_wim_end = old_xml_end; } - /* 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; - } + ret = check_resource_offsets(wim, old_wim_end); + if (ret) + return ret; + ret = prepare_stream_list(wim, WIMLIB_ALL_IMAGES, write_flags, + &stream_list); + if (ret) + return ret; - if (fseeko(out, 0, SEEK_SET) != 0) - return WIMLIB_ERR_WRITE; + ret = open_wim_writable(wim, wim->filename, O_RDWR); + if (ret) + return ret; - return write_header(&hdr, out); -} + ret = lock_wim(wim, wim->out_fd.fd); + if (ret) + goto out_close_wim; -/* Open file stream and write dummy header for WIM. */ -int begin_write(WIMStruct *w, const char *path, int flags) -{ - const char *mode; - DEBUG("Opening `%s' for new WIM\n", path); + /* Save original header so it can be restored in case of error */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); - /* 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"; + /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */ + wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd); + if (ret) { + ERROR_WITH_ERRNO("Error updating WIM header flags"); + goto out_restore_memory_hdr; + } - w->out_fp = fopen(path, mode); - if (!w->out_fp) { - ERROR("Failed to open the file `%s' for writing!\n", path); - return WIMLIB_ERR_OPEN; + if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + ret = WIMLIB_ERR_WRITE; + goto out_restore_physical_hdr; } - /* Write dummy header. It will be overwritten later. */ - return write_header(&w->hdr, w->out_fp); + ret = write_stream_list(&stream_list, + wim->lookup_table, + &wim->out_fd, + wim->compression_type, + wim->chunk_size, + &wim->lzx_context, + write_flags, + num_threads, + progress_func); + if (ret) + goto out_truncate; + + ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, + write_flags, progress_func); + if (ret) + goto out_truncate; + + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; + ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, + progress_func, NULL); + if (ret) + goto out_truncate; + + goto out_unlock_wim; + +out_truncate: + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", + wim->filename, old_wim_end); + /* Return value of ftruncate() is ignored because this is + * already an error path. */ + (void)ftruncate(wim->out_fd.fd, old_wim_end); + } +out_restore_physical_hdr: + (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd); +out_restore_memory_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); +out_close_wim: + (void)close_wim_writable(wim, write_flags); +out_unlock_wim: + wim->wim_locked = 0; + return ret; } -/* 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 *wim, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { + size_t wim_name_len; int ret; - if (image != WIM_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) - return WIMLIB_ERR_INVALID_IMAGE; + DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename); - if (image == WIM_ALL_IMAGES) - DEBUG("Writing all images to `%s'\n", path); - else - DEBUG("Writing image %d to `%s'\n", image, path); + /* Write the WIM to a temporary file in the same directory as the + * original WIM. */ + wim_name_len = tstrlen(wim->filename); + tchar tmpfile[wim_name_len + 10]; + 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 = begin_write(w, path, flags); - if (ret != 0) - goto done; + ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, + write_flags | WIMLIB_WRITE_FLAG_FSYNC, + num_threads, progress_func); + if (ret) { + tunlink(tmpfile); + return ret; + } - for_lookup_table_entry(w->lookup_table, zero_out_refcnts, NULL); + close_wim(wim); + + /* Rename the new WIM file to the original WIM file. Note: on Windows + * this actually calls win32_rename_replacement(), not _wrename(), so + * that removing the existing destination file can be handled. */ + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename); + ret = trename(tmpfile, wim->filename); + if (ret) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", + tmpfile, wim->filename); + #ifdef __WIN32__ + if (ret < 0) + #endif + { + tunlink(tmpfile); + } + return WIMLIB_ERR_RENAME; + } - 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 = wim->filename; + progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } + return 0; +} + +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_overwrite(WIMStruct *wim, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) +{ + int ret; + u32 orig_hdr_flags; - ret = for_image(w, image, write_metadata_resource); + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - if (ret != 0) { - ERROR("Failed to write image metadata!\n"); - goto done; - } + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) + return WIMLIB_ERR_INVALID_PARAM; - ret = finish_write(w, image, flags, 1); + if (!wim->filename) + return WIMLIB_ERR_NO_FILENAME; -done: - DEBUG("Closing output file.\n"); - if (w->out_fp != NULL) { - if (fclose(w->out_fp) != 0) { - ERROR("Failed to close the file `%s': %m\n", path); - ret = WIMLIB_ERR_WRITE; - } - w->out_fp = NULL; + 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 ((!wim->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + && !(write_flags & (WIMLIB_WRITE_FLAG_REBUILD | + WIMLIB_WRITE_FLAG_PIPABLE)) + && !(wim_is_pipable(wim)) + && wim->compression_type == wim->out_compression_type + && wim->chunk_size == wim->out_chunk_size) + { + ret = overwrite_wim_inplace(wim, write_flags, num_threads, + progress_func); + if (ret != WIMLIB_ERR_RESOURCE_ORDER) + return ret; + WARNING("Falling back to re-building entire WIM"); } - return ret; + return overwrite_wim_via_tmpfile(wim, write_flags, num_threads, + progress_func); }