X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=61a29962f6916521bc1ca255ccb7b39bc2162cb0;hp=45b4993f5d3baa43f9cedec76afa3a444ade0408;hb=5cdd60306facd14cc9dcc24471386451294ca73c;hpb=885632f08c75c1d7bb5d25436231c78f6ad7e0c0 diff --git a/src/write.c b/src/write.c index 45b4993f..61a29962 100644 --- a/src/write.c +++ b/src/write.c @@ -3,857 +3,1488 @@ * * Support for writing WIM files; write a WIM file, overwrite a WIM file, write * compressed file resources, etc. - * + */ + +/* * 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 "lookup_table.h" #include "xml.h" +#include "lzx.h" #include "xpress.h" #include +#ifdef ENABLE_MULTITHREADED_COMPRESSION +#include +#include +#include +#endif +#ifdef WITH_NTFS_3G +#include +#include +#include +#include +#endif -/* Used for buffering FILE IO */ -#define BUFFER_SIZE 4096 -/* - * 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. - */ -static int copy_between_files(FILE *in, off_t in_offset, FILE *out, size_t len) -{ - u8 buf[BUFFER_SIZE]; - size_t n; - - if (fseeko(in, in_offset, SEEK_SET) != 0) { - ERROR("Failed to seek to byte %"PRIu64" of input file: %m\n", - in_offset); - return WIMLIB_ERR_READ; - } - /* To reduce memory usage and improve speed, read and write BUFFER_SIZE - * bytes at a time. */ - while (len != 0) { - n = min(len, BUFFER_SIZE); - if (fread(buf, 1, n, in) != n) { - if (feof(in)) { - ERROR("Unexpected EOF when copying data " - "between files\n"); - } else { - ERROR("Error copying data between files: %m\n"); - } - return WIMLIB_ERR_READ; - } +#ifdef HAVE_ALLOCA_H +#include +#else +#include +#endif - if (fwrite(buf, 1, n, out) != n) { - ERROR("Error copying data between files: %m\n"); - return WIMLIB_ERR_WRITE; - } - len -= n; +static int do_fflush(FILE *fp) +{ + int ret = fflush(fp); + if (ret != 0) { + ERROR_WITH_ERRNO("Failed to flush data to output WIM file"); + return WIMLIB_ERR_WRITE; } return 0; } +static int fflush_and_ftruncate(FILE *fp, off_t size) +{ + int ret; -/* - * 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. + ret = do_fflush(fp); + if (ret != 0) + return ret; + ret = ftruncate(fileno(fp), size); + if (ret != 0) { + ERROR_WITH_ERRNO("Failed to truncate output WIM file to " + "%"PRIu64" bytes", size); + return WIMLIB_ERR_WRITE; + } + 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 { + off_t file_offset; + u64 num_chunks; + u64 original_resource_size; + u64 bytes_per_chunk_entry; + u64 table_disk_size; + u64 cur_offset; + u64 *cur_offset_p; + u64 offsets[0]; +}; + +/* + * Allocates and initializes a chunk table, and reserves space for it in the + * output file. */ -static int uncompress_resource(FILE *in, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out) +static int +begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, + FILE *out_fp, + off_t file_offset, + struct chunk_table **chunk_tab_ret) { + u64 size = wim_resource_size(lte); + u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; + size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); + struct chunk_table *chunk_tab = CALLOC(1, alloc_size); int ret; - 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++) { + if (!chunk_tab) { + ERROR("Failed to allocate chunk table for %"PRIu64" byte " + "resource", size); + ret = WIMLIB_ERR_NOMEM; + goto out; + } + chunk_tab->file_offset = file_offset; + chunk_tab->num_chunks = num_chunks; + chunk_tab->original_resource_size = size; + chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4; + chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * + (num_chunks - 1); + chunk_tab->cur_offset = 0; + chunk_tab->cur_offset_p = chunk_tab->offsets; + + if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) != + chunk_tab->table_disk_size) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + ret = WIMLIB_ERR_WRITE; + goto out; + } - uncompressed_offset = i * WIM_CHUNK_SIZE; - uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); + ret = 0; +out: + *chunk_tab_ret = chunk_tab; + return ret; +} - ret = read_resource(in, size, original_size, offset, input_ctype, - uncompressed_chunk_size, - uncompressed_offset, buf); - if (ret != 0) - return ret; +/* + * Pointer to function to compresses a chunk of a WIM resource. + * + * @chunk: Uncompressed data of the chunk. + * @chunk_size: Size of the uncompressed chunk in bytes. + * @compressed_chunk: Pointer to output buffer of size at least + * (@chunk_size - 1) bytes. + * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size + * of the compressed chunk will be + * returned. + * + * Returns zero if compressed succeeded, and nonzero if the chunk could not be + * compressed to any smaller than @chunk_size. This function cannot fail for + * any other reasons. + */ +typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); - if (fwrite(buf, 1, uncompressed_chunk_size, out) != - uncompressed_chunk_size) { - ERROR("Failed to write file resource: %m\n"); - return WIMLIB_ERR_WRITE; +compress_func_t get_compress_func(int out_ctype) +{ + if (out_ctype == WIM_COMPRESSION_TYPE_LZX) + return lzx_compress; + else + return xpress_compress; +} + +/* + * Writes a chunk of a WIM resource to an output file. + * + * @chunk: Uncompressed data of the chunk. + * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) + * @out_fp: FILE * to write tho chunk to. + * @out_ctype: Compression type to use when writing the chunk (ignored if no + * chunk table provided) + * @chunk_tab: Pointer to chunk table being created. It is updated with the + * offset of the chunk we write. + * + * Returns 0 on success; nonzero on failure. + */ +static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, + FILE *out_fp, compress_func_t compress, + struct chunk_table *chunk_tab) +{ + const u8 *out_chunk; + unsigned out_chunk_size; + if (chunk_tab) { + u8 *compressed_chunk = alloca(chunk_size); + int ret; + + ret = compress(chunk, chunk_size, compressed_chunk, + &out_chunk_size); + if (ret == 0) { + out_chunk = compressed_chunk; + } else { + out_chunk = chunk; + out_chunk_size = chunk_size; } + *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; + chunk_tab->cur_offset += out_chunk_size; + } else { + out_chunk = chunk; + out_chunk_size = chunk_size; + } + if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { + ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); + return WIMLIB_ERR_WRITE; } return 0; } -/* - * 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. +/* + * Finishes a WIM chunk tale and writes it to the output file at the correct + * offset. * - * @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. + * The final size of the full compressed resource is returned in the + * @compressed_size_p. */ -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 +finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, + FILE *out_fp, u64 *compressed_size_p) { - int ret; - int (*compress)(const void *, uint, void *, uint *); - if (output_ctype == WIM_COMPRESSION_TYPE_LZX) - compress = lzx_compress; - else - compress = xpress_compress; - - u8 uncompressed_buf[WIM_CHUNK_SIZE]; - u8 compressed_buf[WIM_CHUNK_SIZE - 1]; + size_t bytes_written; + if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output " + "WIM file", chunk_tab->file_offset); + return WIMLIB_ERR_WRITE; + } - /* Determine how many compressed chunks the file needs to be divided - * into. */ - u64 num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; + if (chunk_tab->bytes_per_chunk_entry == 8) { + array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); + } else { + for (u64 i = 0; i < chunk_tab->num_chunks; i++) + ((u32*)chunk_tab->offsets)[i] = + cpu_to_le32(chunk_tab->offsets[i]); + } + bytes_written = fwrite((u8*)chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + 1, chunk_tab->table_disk_size, out_fp); + if (bytes_written != chunk_tab->table_disk_size) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + return WIMLIB_ERR_WRITE; + } + if (fseeko(out_fp, 0, SEEK_END) != 0) { + ERROR_WITH_ERRNO("Failed to seek to end of output WIM file"); + return WIMLIB_ERR_WRITE; + } + *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; + return 0; +} - u64 num_chunk_entries = num_chunks - 1; +/* Prepare for multiple reads to a resource by caching a FILE * or NTFS + * attribute pointer in the lookup table entry. */ +static int prepare_resource_for_read(struct lookup_table_entry *lte - /* 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; + #ifdef WITH_NTFS_3G + , ntfs_inode **ni_ret + #endif + ) +{ + if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK + && !lte->file_on_disk_fp) + { + wimlib_assert(lte->file_on_disk); + lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb"); + if (!lte->file_on_disk_fp) { + ERROR_WITH_ERRNO("Failed to open the file `%s' for " + "reading", lte->file_on_disk); + return WIMLIB_ERR_OPEN; + } + } +#ifdef WITH_NTFS_3G + else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME + && !lte->attr) + { + struct ntfs_location *loc = lte->ntfs_loc; + ntfs_inode *ni; + wimlib_assert(loc); + ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8); + if (!ni) { + ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS " + "volume", loc->path_utf8); + return WIMLIB_ERR_NTFS_3G; + } + lte->attr = ntfs_attr_open(ni, + loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, + (ntfschar*)loc->stream_name_utf16, + loc->stream_name_utf16_num_chars); + if (!lte->attr) { + ERROR_WITH_ERRNO("Failed to open attribute of `%s' in " + "NTFS volume", loc->path_utf8); + ntfs_inode_close(ni); + return WIMLIB_ERR_NTFS_3G; + } + *ni_ret = ni; + } +#endif + return 0; +} - /* Array in which to construct the chunk offset table. */ - u64 chunk_offsets[num_chunk_entries]; +/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS + * attribute. */ +static void end_wim_resource_read(struct lookup_table_entry *lte + #ifdef WITH_NTFS_3G + , ntfs_inode *ni + #endif + ) +{ + if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK + && lte->file_on_disk_fp) { + fclose(lte->file_on_disk_fp); + lte->file_on_disk_fp = NULL; + } +#ifdef WITH_NTFS_3G + else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) { + if (lte->attr) { + ntfs_attr_close(lte->attr); + lte->attr = NULL; + } + if (ni) + ntfs_inode_close(ni); + } +#endif +} - /* Offset of the start of the chunk table in the output file. */ - off_t chunk_tab_offset = ftello(out); +/* + * Writes a WIM resource to a FILE * opened for writing. The resource may be + * written uncompressed or compressed depending on the @out_ctype parameter. + * + * If by chance the resource compresses to more than the original size (this may + * happen with random data or files than are pre-compressed), the resource is + * instead written uncompressed (and this is reflected in the @out_res_entry by + * removing the WIM_RESHDR_FLAG_COMPRESSED flag). + * + * @lte: The lookup table entry for the WIM resource. + * @out_fp: The FILE * to write the resource to. + * @out_ctype: The compression type of the resource to write. Note: if this is + * the same as the compression type of the WIM resource we + * need to read, we simply copy the data (i.e. we do not + * uncompress it, then compress it again). + * @out_res_entry: If non-NULL, a resource entry that is filled in with the + * offset, original size, compressed size, and compression flag + * of the output resource. + * + * Returns 0 on success; nonzero on failure. + */ +int write_wim_resource(struct lookup_table_entry *lte, + FILE *out_fp, int out_ctype, + struct resource_entry *out_res_entry, + int flags) +{ + u64 bytes_remaining; + u64 original_size; + u64 old_compressed_size; + u64 new_compressed_size; + u64 offset; + int ret; + struct chunk_table *chunk_tab = NULL; + bool raw; + off_t file_offset; + compress_func_t compress; +#ifdef WITH_NTFS_3G + ntfs_inode *ni = NULL; +#endif + + wimlib_assert(lte); + + /* Original size of the resource */ + original_size = wim_resource_size(lte); + + /* Compressed size of the resource (as it exists now) */ + old_compressed_size = wim_resource_compressed_size(lte); + + /* Current offset in output file */ + file_offset = ftello(out_fp); + if (file_offset == -1) { + ERROR_WITH_ERRNO("Failed to get offset in output " + "stream"); + return WIMLIB_ERR_WRITE; + } - /* Total size of the chunk table (as written to the file) */ - u64 chunk_tab_size = chunk_entry_size * num_chunk_entries; + /* Are the compression types the same? If so, do a raw copy (copy + * without decompressing and recompressing the data). */ + raw = (wim_resource_compression_type(lte) == out_ctype + && out_ctype != WIM_COMPRESSION_TYPE_NONE + && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); - /* Reserve space for the chunk table. */ - if (fwrite(chunk_offsets, 1, chunk_tab_size, out) != chunk_tab_size) { - ERROR("Failed to write chunk offset table: %m\n"); - return WIMLIB_ERR_WRITE; + if (raw) { + flags |= WIMLIB_RESOURCE_FLAG_RAW; + bytes_remaining = old_compressed_size; + } else { + flags &= ~WIMLIB_RESOURCE_FLAG_RAW; + bytes_remaining = original_size; } - /* 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++) { + /* Empty resource; nothing needs to be done, so just return success. */ + if (bytes_remaining == 0) + return 0; - u64 uncompressed_offset = i * WIM_CHUNK_SIZE; - u64 uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); + /* Buffer for reading chunks for the resource */ + u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - const u8 *uncompressed_p; - if (uncompressed_resource != NULL) { - uncompressed_p = uncompressed_resource + - uncompressed_offset; + /* If we are writing a compressed resource and not doing a raw copy, we + * need to initialize the chunk table */ + if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) { + ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, + &chunk_tab); + if (ret != 0) + goto out; + } - } 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 the WIM resource is in an external file, open a FILE * to it so we + * don't have to open a temporary one in read_wim_resource() for each + * chunk. */ +#ifdef WITH_NTFS_3G + ret = prepare_resource_for_read(lte, &ni); +#else + ret = prepare_resource_for_read(lte); +#endif + if (ret != 0) + goto out; + + /* If we aren't doing a raw copy, we will compute the SHA1 message + * digest of the resource as we read it, and verify it's the same as the + * hash given in the lookup table entry once we've finished reading the + * resource. */ + SHA_CTX ctx; + if (!raw) { + sha1_init(&ctx); + compress = get_compress_func(out_ctype); + } + offset = 0; + + /* While there are still bytes remaining in the WIM resource, read a + * chunk of the resource, update SHA1, then write that chunk using the + * desired compression type. */ + do { + u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); + ret = read_wim_resource(lte, buf, to_read, offset, flags); + if (ret != 0) + goto out_fclose; + if (!raw) + sha1_update(&ctx, buf, to_read); + ret = write_wim_resource_chunk(buf, to_read, out_fp, + compress, chunk_tab); + if (ret != 0) + goto out_fclose; + bytes_remaining -= to_read; + offset += to_read; + } while (bytes_remaining); + + /* Raw copy: The new compressed size is the same as the old compressed + * size + * + * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the + * original size + * + * Using a different compression type: Call + * finish_wim_resource_chunk_tab() and it will provide the new + * compressed size. + */ + if (raw) { + new_compressed_size = old_compressed_size; + } else { + if (out_ctype == WIM_COMPRESSION_TYPE_NONE) + new_compressed_size = original_size; + else { + ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp, + &new_compressed_size); if (ret != 0) - return ret; - uncompressed_p = uncompressed_buf; + goto out_fclose; } + } - if (i != 0) - chunk_offsets[i - 1] = cur_chunk_offset; - - uint compressed_len; - - ret = compress(uncompressed_p, uncompressed_chunk_size, - compressed_buf, &compressed_len); + /* Verify SHA1 message digest of the resource, unless we are doing a raw + * write (in which case we never even saw the uncompressed data). Or, + * if the hash we had before is all 0's, just re-set it to be the new + * hash. */ + if (!raw) { + u8 md[SHA1_HASH_SIZE]; + sha1_final(md, &ctx); + if (is_zero_hash(lte->hash)) { + copy_hash(lte->hash, md); + } else if (!hashes_equal(md, lte->hash)) { + ERROR("WIM resource has incorrect hash!"); + if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { + ERROR("We were reading it from `%s'; maybe it changed " + "while we were reading it.", + lte->file_on_disk); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + goto out_fclose; + } + } - /* 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; - } else { - buf_to_write = uncompressed_p; - len_to_write = uncompressed_chunk_size; + if (!raw && new_compressed_size >= original_size && + out_ctype != WIM_COMPRESSION_TYPE_NONE) + { + /* Oops! We compressed the resource to larger than the original + * size. Write the resource uncompressed instead. */ + if (fseeko(out_fp, file_offset, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" " + "of output WIM file", file_offset); + ret = WIMLIB_ERR_WRITE; + goto out_fclose; } + ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE, + out_res_entry, flags); + if (ret != 0) + goto out_fclose; - 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; + ret = fflush_and_ftruncate(out_fp, file_offset + out_res_entry->size); + if (ret != 0) + goto out_fclose; + } else { + if (out_res_entry) { + out_res_entry->size = new_compressed_size; + out_res_entry->original_size = original_size; + out_res_entry->offset = file_offset; + out_res_entry->flags = lte->resource_entry.flags + & ~WIM_RESHDR_FLAG_COMPRESSED; + if (out_ctype != WIM_COMPRESSION_TYPE_NONE) + out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } - cur_chunk_offset += len_to_write; } + ret = 0; +out_fclose: +#ifdef WITH_NTFS_3G + end_wim_resource_read(lte, ni); +#else + end_wim_resource_read(lte); +#endif +out: + FREE(chunk_tab); + return ret; +} - /* 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; +#ifdef ENABLE_MULTITHREADED_COMPRESSION +struct shared_queue { + sem_t filled_slots; + sem_t empty_slots; + pthread_mutex_t lock; + unsigned front; + unsigned back; + void **array; + unsigned size; +}; + +static int shared_queue_init(struct shared_queue *q, unsigned size) +{ + q->array = CALLOC(sizeof(q->array[0]), size); + if (!q->array) + return WIMLIB_ERR_NOMEM; - /* Now that all entries of the chunk table are determined, rewind the - * stream to where the chunk table was, and write it back out. */ + sem_init(&q->filled_slots, 0, 0); + sem_init(&q->empty_slots, 0, size); + pthread_mutex_init(&q->lock, NULL); + q->front = 0; + q->back = size - 1; + q->size = size; + return 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; - } +static void shared_queue_destroy(struct shared_queue *q) +{ + sem_destroy(&q->filled_slots); + sem_destroy(&q->empty_slots); + pthread_mutex_destroy(&q->lock); + FREE(q->array); +} - if (chunk_entry_size == 8) { - array_to_le64(chunk_offsets, num_chunk_entries); +static void shared_queue_put(struct shared_queue *q, void *obj) +{ + sem_wait(&q->empty_slots); + pthread_mutex_lock(&q->lock); - if (fwrite(chunk_offsets, 1, chunk_tab_size, out) != - chunk_tab_size) { - ERROR("Failed to write chunk table: %m\n"); - return WIMLIB_ERR_WRITE; - } - } 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; + q->back = (q->back + 1) % q->size; + q->array[q->back] = obj; + + sem_post(&q->filled_slots); + pthread_mutex_unlock(&q->lock); +} + +static void *shared_queue_get(struct shared_queue *q) +{ + sem_wait(&q->filled_slots); + pthread_mutex_lock(&q->lock); + + void *obj = q->array[q->front]; + q->array[q->front] = NULL; + q->front = (q->front + 1) % q->size; + + sem_post(&q->empty_slots); + pthread_mutex_unlock(&q->lock); + return obj; +} + +struct compressor_thread_params { + struct shared_queue *res_to_compress_queue; + struct shared_queue *compressed_res_queue; + compress_func_t compress; +}; + +#define MAX_CHUNKS_PER_MSG 2 + +struct message { + struct lookup_table_entry *lte; + u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; + u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG]; + u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; + unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; + unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; + unsigned num_chunks; + struct list_head list; + bool complete; + u64 begin_chunk; +}; + +static void compress_chunks(struct message *msg, compress_func_t compress) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) { + DEBUG2("compress chunk %u of %u", i, msg->num_chunks); + int ret = compress(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i], + &msg->compressed_chunk_sizes[i]); + if (ret == 0) { + msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + } else { + msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; + msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; } } +} - if (fseeko(out, 0, SEEK_END) != 0) { - ERROR("Failed to seek to end of output file: %m\n"); - return WIMLIB_ERR_WRITE; +static void *compressor_thread_proc(void *arg) +{ + struct compressor_thread_params *params = arg; + struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; + struct shared_queue *compressed_res_queue = params->compressed_res_queue; + compress_func_t compress = params->compress; + struct message *msg; + + DEBUG("Compressor thread ready"); + while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { + compress_chunks(msg, compress); + shared_queue_put(compressed_res_queue, msg); } - - return 0; + DEBUG("Compressor thread terminating"); } +#endif -int write_resource_from_memory(const u8 resource[], int out_ctype, - u64 resource_original_size, FILE *out, - u64 *resource_size_ret) +void show_stream_op_progress(u64 *cur_size, u64 *next_size, + u64 total_size, u64 one_percent, + unsigned *cur_percent, + const struct lookup_table_entry *cur_lte, + const char *op) { - 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); + if (*cur_size >= *next_size) { + printf("\r%"PRIu64" MiB of %"PRIu64" MiB " + "(uncompressed) %s (%u%% done)", + *cur_size >> 20, + total_size >> 20, op, *cur_percent); + fflush(stdout); + *next_size += one_percent; + (*cur_percent)++; } + *cur_size += wim_resource_size(cur_lte); } +void finish_stream_op_progress(u64 total_size, const char *op) +{ + printf("\r%"PRIu64" MiB of %"PRIu64" MiB " + "(uncompressed) %s (100%% done)\n", + total_size >> 20, total_size >> 20, op); + fflush(stdout); +} -/* - * Transfers a file resource from a FILE* opened for reading to a FILE* opened - * for writing, possibly changing the compression type. - * - * @in: The FILE* that contains the file resource. - * @size: The (compressed) size of the file resource. - * @original_size: The uncompressed size of the file resource. - * @offset: The offset of the file resource in the input file. - * @input_ctype: The compression type of the file resource in the input - * file. - * @out: The FILE* for the output file. The file resource is - * written at the current position of @out. - * @output_ctype: The compression type to which the file resource will be - * converted. - * @output_res_entry: A pointer to a resource entry that, upon successful - * return of this function, will have the size, - * original size, offset, and flags fields filled - * in for the file resource written to the output - * file. - */ -static int transfer_file_resource(FILE *in, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out, - int output_ctype, - struct resource_entry *output_res_entry) +static int write_stream_list_serial(struct list_head *stream_list, + FILE *out_fp, int out_ctype, + int write_flags, u64 total_size) { + struct lookup_table_entry *lte; int ret; - /* Handle zero-length files */ - if (original_size == 0) { - memset(output_res_entry, 0, sizeof(*output_res_entry)); - return 0; - } - - /* 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; - } + u64 one_percent = total_size / 100; + u64 cur_size = 0; + u64 next_size = 0; + unsigned cur_percent = 0; + int write_resource_flags = 0; - if (output_ctype == input_ctype) { - /* The same compression types; simply copy the resource. */ + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; - ret = copy_between_files(in, offset, out, size); + list_for_each_entry(lte, stream_list, staging_list) { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_op_progress(&cur_size, &next_size, + total_size, one_percent, + &cur_percent, lte, "written"); + } + ret = write_wim_resource(lte, out_fp, out_ctype, + <e->output_resource_entry, + write_resource_flags); if (ret != 0) return ret; - output_res_entry->size = size; - } else { - /* Different compression types. */ + } + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + finish_stream_op_progress(total_size, "written"); + return 0; +} - 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; +#ifdef ENABLE_MULTITHREADED_COMPRESSION +static int write_wim_chunks(struct message *msg, FILE *out_fp, + struct chunk_table *chunk_tab) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) { + unsigned chunk_csize = msg->compressed_chunk_sizes[i]; + + DEBUG2("Write wim chunk %u of %u (csize = %u)", + i, msg->num_chunks, chunk_csize); + + if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) + != chunk_csize) + { + ERROR_WITH_ERRNO("Failed to write WIM chunk"); + return WIMLIB_ERR_WRITE; } + *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; + chunk_tab->cur_offset += chunk_csize; } - - 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; } -/* - * Writes a file resource to the output file. +/* + * This function is executed by the main thread when the resources are being + * compressed in parallel. The main thread is in change of all reading of the + * uncompressed data and writing of the compressed data. The compressor threads + * *only* do compression from/to in-memory buffers. * - * @dentry: The dentry for the file resource. - * @wim_p: A pointer to the WIMStruct. The fields of interest to this - * function are the input and output file streams and the lookup table. - * @return zero on success, nonzero on failure. + * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG + * chunks of compressed data to compress, represented in a `struct message'. + * Each message is passed from the main thread to a worker thread through the + * res_to_compress_queue, and it is passed back through the + * compressed_res_queue. */ -static int write_file_resource(struct dentry *dentry, void *wim_p) +static int main_writer_thread_proc(struct list_head *stream_list, + FILE *out_fp, + int out_ctype, + struct shared_queue *res_to_compress_queue, + struct shared_queue *compressed_res_queue, + size_t queue_size, + int write_flags, + u64 total_size) { - 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; int ret; - w = wim_p; - out = w->out_fp; - - /* Directories don't need file resources. */ - if (dentry_is_directory(dentry)) - return 0; + struct message msgs[queue_size]; + ZERO_ARRAY(msgs); + + // Initially, all the messages are available to use. + LIST_HEAD(available_msgs); + for (size_t i = 0; i < ARRAY_LEN(msgs); i++) + list_add(&msgs[i].list, &available_msgs); + + // outstanding_resources is the list of resources that currently have + // had chunks sent off for compression. + // + // The first stream in outstanding_resources is the stream that is + // currently being written (cur_lte). + // + // The last stream in outstanding_resources is the stream that is + // currently being read and chunks fed to the compressor threads + // (next_lte). + // + // Depending on the number of threads and the sizes of the resource, + // the outstanding streams list may contain streams between cur_lte and + // next_lte that have all their chunks compressed or being compressed, + // but haven't been written yet. + // + LIST_HEAD(outstanding_resources); + struct list_head *next_resource = stream_list->next; + struct lookup_table_entry *next_lte = container_of(next_resource, + struct lookup_table_entry, + staging_list); + next_resource = next_resource->next; + u64 next_chunk = 0; + u64 next_num_chunks = wim_resource_chunks(next_lte); + INIT_LIST_HEAD(&next_lte->msg_list); + list_add_tail(&next_lte->staging_list, &outstanding_resources); + + // As in write_wim_resource(), each resource we read is checksummed. + SHA_CTX next_sha_ctx; + sha1_init(&next_sha_ctx); + u8 next_hash[SHA1_HASH_SIZE]; + + // Resources that don't need any chunks compressed are added to this + // list and written directly by the main thread. + LIST_HEAD(my_resources); + + struct lookup_table_entry *cur_lte = next_lte; + struct chunk_table *cur_chunk_tab = NULL; + struct lookup_table_entry *lte; + struct message *msg; + + u64 one_percent = total_size / 100; + u64 cur_size = 0; + u64 next_size = 0; + unsigned cur_percent = 0; + +#ifdef WITH_NTFS_3G + ntfs_inode *ni = NULL; +#endif + +#ifdef WITH_NTFS_3G + ret = prepare_resource_for_read(next_lte, &ni); +#else + ret = prepare_resource_for_read(next_lte); +#endif + if (ret != 0) + goto out; + + DEBUG("Initializing buffers for uncompressed " + "and compressed data (%zu bytes needed)", + queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2); + + // Pre-allocate all the buffers that will be needed to do the chunk + // compression. + for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { + for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { + msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); + msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); + if (msgs[i].compressed_chunks[j] == NULL || + msgs[i].uncompressed_chunks[j] == NULL) + { + ERROR("Could not allocate enough memory for " + "multi-threaded compression"); + ret = WIMLIB_ERR_NOMEM; + goto out; + } + } + } - /* Get the lookup entry for the file resource. */ - lte = wim_lookup_resource(w, dentry); - if (!lte) - return 0; + // This loop is executed until all resources have been written, except + // possibly a few that have been added to the @my_resources list for + // writing later. + while (1) { + // Send chunks to the compressor threads until either (a) there + // are no more messages available since they were all sent off, + // or (b) there are no more resources that need to be + // compressed. + while (!list_empty(&available_msgs) && next_lte != NULL) { + + // Get a message from the available messages + // list + msg = container_of(available_msgs.next, + struct message, + list); + + // ... and delete it from the available messages + // list + list_del(&msg->list); + + // Initialize the message with the chunks to + // compress. + msg->num_chunks = min(next_num_chunks - next_chunk, + MAX_CHUNKS_PER_MSG); + msg->lte = next_lte; + msg->complete = false; + msg->begin_chunk = next_chunk; + + unsigned size = WIM_CHUNK_SIZE; + for (unsigned i = 0; i < msg->num_chunks; i++) { + + // Read chunk @next_chunk of the stream into the + // message so that a compressor thread can + // compress it. + + if (next_chunk == next_num_chunks - 1 && + wim_resource_size(next_lte) % WIM_CHUNK_SIZE != 0) + { + size = wim_resource_size(next_lte) % WIM_CHUNK_SIZE; + } + + + DEBUG2("Read resource (size=%u, offset=%zu)", + size, next_chunk * WIM_CHUNK_SIZE); + + msg->uncompressed_chunk_sizes[i] = size; + + ret = read_wim_resource(next_lte, + msg->uncompressed_chunks[i], + size, + next_chunk * WIM_CHUNK_SIZE, + 0); + if (ret != 0) + goto out; + sha1_update(&next_sha_ctx, + msg->uncompressed_chunks[i], size); + next_chunk++; + } - /* 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; + // Send the compression request + list_add_tail(&msg->list, &next_lte->msg_list); + shared_queue_put(res_to_compress_queue, msg); + DEBUG2("Compression request sent"); + + if (next_chunk != next_num_chunks) + // More chunks to send for this resource + continue; + + // Done sending compression requests for a resource! + // Check the SHA1 message digest. + DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)", next_num_chunks); + sha1_final(next_hash, &next_sha_ctx); + if (!hashes_equal(next_lte->hash, next_hash)) { + ERROR("WIM resource has incorrect hash!"); + if (next_lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { + ERROR("We were reading it from `%s'; maybe it changed " + "while we were reading it.", + next_lte->file_on_disk); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + goto out; + } - out_wim_ctype = wimlib_get_compression_type(w); - output_res_entry = <e->output_resource_entry; + // Advance to the next resource. + // + // If the next resource needs no compression, just write + // it with this thread (not now though--- we could be in + // the middle of writing another resource.) Keep doing + // this until we either get to the end of the resources + // list, or we get to a resource that needs compression. + + while (1) { + if (next_resource == stream_list) { + next_lte = NULL; + break; + } + #ifdef WITH_NTFS_3G + end_wim_resource_read(next_lte, ni); + ni = NULL; + #else + end_wim_resource_read(next_lte); + #endif + + next_lte = container_of(next_resource, + struct lookup_table_entry, + staging_list); + next_resource = next_resource->next; + if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + && next_lte->resource_location == RESOURCE_IN_WIM + && wimlib_get_compression_type(next_lte->wim) == out_ctype) + || wim_resource_size(next_lte) == 0) + { + list_add_tail(&next_lte->staging_list, + &my_resources); + } else { + list_add_tail(&next_lte->staging_list, + &outstanding_resources); + next_chunk = 0; + next_num_chunks = wim_resource_chunks(next_lte); + sha1_init(&next_sha_ctx); + INIT_LIST_HEAD(&next_lte->msg_list); + #ifdef WITH_NTFS_3G + ret = prepare_resource_for_read(next_lte, &ni); + #else + ret = prepare_resource_for_read(next_lte); + #endif + if (ret != 0) + goto out; + DEBUG2("Updated next_lte"); + break; + } + } + } - /* 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) { + // If there are no outstanding resources, there are no more + // resources that need to be written. + if (list_empty(&outstanding_resources)) { + DEBUG("No outstanding resources! Done"); + ret = 0; + goto out; + } - /* Read from disk (uncompressed) */ + // Get the next message from the queue and process it. + // The message will contain 1 or more data chunks that have been + // compressed. + DEBUG2("Waiting for message"); + msg = shared_queue_get(compressed_res_queue); + msg->complete = true; - len = lte->resource_entry.original_size; + DEBUG2("Received msg (begin_chunk=%"PRIu64")", msg->begin_chunk); - 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; + list_for_each_entry(msg, &cur_lte->msg_list, list) { + DEBUG2("complete=%d", msg->complete); } - if (w->verbose) - puts(lte->file_on_disk); + // 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 (!list_empty(&cur_lte->msg_list) + && (msg = container_of(cur_lte->msg_list.next, + struct message, + list))->complete) + { + DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk); + if (msg->begin_chunk == 0) { + DEBUG2("Begin chunk tab"); + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_op_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + cur_lte, + "written"); + } + + // This is the first set of chunks. Leave space + // for the chunk table in the output file. + off_t cur_offset = ftello(out_fp); + if (cur_offset == -1) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + ret = begin_wim_resource_chunk_tab(cur_lte, + out_fp, + cur_offset, + &cur_chunk_tab); + if (ret != 0) + goto out; + } - ret = transfer_file_resource(in, len, len, 0, - WIM_COMPRESSION_TYPE_NONE, out, - out_wim_ctype, output_res_entry); - fclose(in); + // Write the compressed chunks from the message. + ret = write_wim_chunks(msg, out_fp, cur_chunk_tab); + if (ret != 0) + goto out; + + list_del(&msg->list); + + // This message is available to use for different chunks + // now. + list_add(&msg->list, &available_msgs); + + // Was this the last chunk of the stream? If so, + // finish it. + if (list_empty(&cur_lte->msg_list) && + msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks) + { + DEBUG2("Finish wim chunk tab"); + u64 res_csize; + ret = finish_wim_resource_chunk_tab(cur_chunk_tab, + out_fp, + &res_csize); + if (ret != 0) + goto out; + + + cur_lte->output_resource_entry.size = + res_csize; + + cur_lte->output_resource_entry.original_size = + cur_lte->resource_entry.original_size; + + cur_lte->output_resource_entry.offset = + cur_chunk_tab->file_offset; + + cur_lte->output_resource_entry.flags = + cur_lte->resource_entry.flags | + WIM_RESHDR_FLAG_COMPRESSED; + + FREE(cur_chunk_tab); + cur_chunk_tab = NULL; + + struct list_head *next = cur_lte->staging_list.next; + list_del(&cur_lte->staging_list); + + if (next == &outstanding_resources) { + DEBUG("No more outstanding resources"); + ret = 0; + goto out; + } else { + cur_lte = container_of(cur_lte->staging_list.next, + struct lookup_table_entry, + staging_list); + } + + // Since we just finished writing a stream, + // write any streams that have been added to the + // my_resources 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). + struct lookup_table_entry *tmp; + list_for_each_entry_safe(lte, + tmp, + &my_resources, + staging_list) + { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_op_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + lte, + "written"); + } + + ret = write_wim_resource(lte, + out_fp, + out_ctype, + <e->output_resource_entry, + 0); + list_del(<e->staging_list); + if (ret != 0) + goto out; + } + } + } + } + +out: +#ifdef WITH_NTFS_3G + end_wim_resource_read(cur_lte, ni); +#else + end_wim_resource_read(cur_lte); +#endif + if (ret == 0) { + list_for_each_entry(lte, &my_resources, staging_list) { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + show_stream_op_progress(&cur_size, + &next_size, + total_size, + one_percent, + &cur_percent, + lte, + "written"); + } + ret = write_wim_resource(lte, out_fp, + out_ctype, + <e->output_resource_entry, + 0); + if (ret != 0) + break; + } + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + finish_stream_op_progress(total_size, "written"); } else { + size_t num_available_msgs = 0; + struct list_head *cur; - /* Read from input WIM (possibly compressed) */ + list_for_each(cur, &available_msgs) { + num_available_msgs++; + } - /* It may be a different WIM file, in the case of - * exporting images from one WIM file to another */ - if (lte->other_wim_fp) { - /* Different WIM file. */ - in = lte->other_wim_fp; - in_wim_ctype = lte->other_wim_ctype; - } else { - /* Same WIM file. */ - in = w->fp; - in_wim_ctype = out_wim_ctype; + while (num_available_msgs < ARRAY_LEN(msgs)) { + shared_queue_get(compressed_res_queue); + num_available_msgs++; } - 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); } - return ret; -} -/* Reopens the FILE* for a WIM read-write. */ -static int reopen_rw(WIMStruct *w) -{ - FILE *fp; + DEBUG("Freeing messages"); - if (fclose(w->fp) != 0) - ERROR("Failed to close the file `%s': %m\n", w->filename); - fp = fopen(w->filename, "r+b"); - if (!fp) { - ERROR("Failed to open `%s' for reading and writing: " - "%m\n", w->filename); - return WIMLIB_ERR_OPEN; + for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { + for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { + FREE(msgs[i].compressed_chunks[j]); + FREE(msgs[i].uncompressed_chunks[j]); + } } - w->fp = fp; - return 0; + + if (cur_chunk_tab != NULL) + FREE(cur_chunk_tab); + return ret; } +static const char *get_data_type(int ctype) +{ + switch (ctype) { + case WIM_COMPRESSION_TYPE_NONE: + return "uncompressed"; + case WIM_COMPRESSION_TYPE_LZX: + return "LZX-compressed"; + case WIM_COMPRESSION_TYPE_XPRESS: + return "XPRESS-compressed"; + } +} -/* - * Writes a WIM file to the original file that it was read from, overwriting it. - */ -WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int flags) +static int write_stream_list_parallel(struct list_head *stream_list, + FILE *out_fp, int out_ctype, + int write_flags, u64 total_size, + unsigned num_threads) { - const char *wimfile_name; - size_t wim_name_len; int ret; - - wimfile_name = w->filename; + struct shared_queue res_to_compress_queue; + struct shared_queue compressed_res_queue; + pthread_t *compressor_threads = NULL; + + if (num_threads == 0) { + long nthreads = sysconf(_SC_NPROCESSORS_ONLN); + if (nthreads < 1) { + WARNING("Could not determine number of processors! Assuming 1"); + goto out_serial; + } else { + num_threads = nthreads; + } + } - DEBUG("Replacing WIM file `%s'\n", wimfile_name); + wimlib_assert(stream_list->next != stream_list); - if (!wimfile_name) - return WIMLIB_ERR_NO_FILENAME; + static const double MESSAGES_PER_THREAD = 2.0; + size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); - /* 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'; + DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, flags); - if (ret != 0) { - ERROR("Failed to write the WIM file `%s'!\n", tmpfile); - return ret; - } + ret = shared_queue_init(&res_to_compress_queue, queue_size); + if (ret != 0) + goto out_serial; - DEBUG("Closing original WIM file.\n"); - /* Close the original WIM file that was opened for reading. */ - if (w->fp) { - if (fclose(w->fp) != 0) { - DEBUG("WARNING: Failed to close the file `%s'\n", - wimfile_name); + ret = shared_queue_init(&compressed_res_queue, queue_size); + if (ret != 0) + goto out_destroy_res_to_compress_queue; + + struct compressor_thread_params params; + params.res_to_compress_queue = &res_to_compress_queue; + params.compressed_res_queue = &compressed_res_queue; + params.compress = get_compress_func(out_ctype); + + compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + + for (unsigned i = 0; i < num_threads; i++) { + DEBUG("pthread_create thread %u", i); + ret = pthread_create(&compressor_threads[i], NULL, + compressor_thread_proc, ¶ms); + if (ret != 0) { + ret = -1; + ERROR_WITH_ERRNO("Failed to create compressor " + "thread %u", i); + num_threads = i; + goto out_join; } - w->fp = NULL; } - DEBUG("Renaming `%s' to `%s'\n", tmpfile, wimfile_name); - - /* Rename the new file to the old file .*/ - if (rename(tmpfile, wimfile_name) != 0) { - ERROR("Failed to rename `%s' to `%s': %m\n", tmpfile, - wimfile_name); - /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - ERROR("Failed to remove `%s': %m\n", tmpfile); - return WIMLIB_ERR_RENAME; + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + printf("Writing %s data using %u threads...\n", + get_data_type(out_ctype), num_threads); } - return 0; + ret = main_writer_thread_proc(stream_list, + out_fp, + out_ctype, + &res_to_compress_queue, + &compressed_res_queue, + queue_size, + write_flags, + total_size); + +out_join: + for (unsigned i = 0; i < num_threads; i++) + shared_queue_put(&res_to_compress_queue, NULL); + + for (unsigned i = 0; i < num_threads; i++) { + if (pthread_join(compressor_threads[i], NULL)) { + WARNING("Failed to join compressor thread %u: %s", + i, strerror(errno)); + } + } + FREE(compressor_threads); + shared_queue_destroy(&compressed_res_queue); +out_destroy_res_to_compress_queue: + shared_queue_destroy(&res_to_compress_queue); + if (ret >= 0 && ret != WIMLIB_ERR_NOMEM) + return ret; +out_serial: + WARNING("Falling back to single-threaded compression"); + return write_stream_list_serial(stream_list, out_fp, + out_ctype, write_flags, total_size); } +#endif - -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *w, int flags) +/* + * Write a list of streams to a WIM (@out_fp) using the compression type + * @out_ctype and up to @num_threads compressor threads. + */ +static int write_stream_list(struct list_head *stream_list, FILE *out_fp, + int out_ctype, int write_flags, + unsigned num_threads) { - 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; - - ret = reopen_rw(w); - if (ret != 0) - return ret; - - fp = w->fp; - - /* The old integrity table is still OK, as the SHA1 message digests in - * the integrity table include neither the header nor the XML data. - * Save it for later if it exists and an integrity table was required. - * */ - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY && - w->hdr.integrity.offset != 0) { - DEBUG("Reading existing integrity table.\n"); - integrity_table = MALLOC(w->hdr.integrity.size); - if (!integrity_table) - return WIMLIB_ERR_NOMEM; - - 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"); + struct lookup_table_entry *lte; + size_t num_streams = 0; + u64 total_size = 0; + bool compression_needed = false; + + list_for_each_entry(lte, stream_list, staging_list) { + num_streams++; + total_size += wim_resource_size(lte); + if (!compression_needed + && + (out_ctype != WIM_COMPRESSION_TYPE_NONE + && (lte->resource_location != RESOURCE_IN_WIM + || wimlib_get_compression_type(lte->wim) != out_ctype + || (write_flags & WIMLIB_WRITE_FLAG_REBUILD))) + && wim_resource_size(lte) != 0) + compression_needed = true; } - 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; + if (num_streams == 0) { + if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) + printf("No streams to write\n"); + return 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; + if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { + printf("Preparing to write %zu streams " + "(%"PRIu64" total bytes uncompressed)\n", + num_streams, total_size); + printf("Using compression type %s\n", + wimlib_get_compression_type_string(out_ctype)); } - 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; - } - 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; - } - memset(&w->hdr.integrity, 0, sizeof(struct resource_entry)); +#ifdef ENABLE_MULTITHREADED_COMPRESSION + if (compression_needed && total_size >= 1000000 && num_threads != 1) { + return write_stream_list_parallel(stream_list, out_fp, + out_ctype, write_flags, + total_size, num_threads); } + else +#endif + { + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { + const char *reason = ""; + if (!compression_needed) + reason = " (no compression needed)"; + printf("Writing %s data using 1 thread%s\n", + get_data_type(out_ctype), reason); + } - 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; + return write_stream_list_serial(stream_list, out_fp, + out_ctype, write_flags, + total_size); } +} - ret = write_header(&w->hdr, fp); - if (ret != 0) - return ret;; - DEBUG("Closing file.\n"); - if (fclose(fp) != 0) { - ERROR("Failed to close `%s': %m\n", w->filename); - return WIMLIB_ERR_WRITE; +static int dentry_find_streams_to_write(struct dentry *dentry, + void *wim) +{ + WIMStruct *w = wim; + struct list_head *stream_list = w->private; + struct lookup_table_entry *lte; + for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) { + lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table); + if (lte && ++lte->out_refcnt == 1) + list_add_tail(<e->staging_list, stream_list); } - w->fp = NULL; - DEBUG("Done.\n"); return 0; -err: - FREE(integrity_table); - return ret; } -/* Write the metadata resource for the current image. */ -static int write_metadata_resource(WIMStruct *w) +static int find_streams_to_write(WIMStruct *w) { - FILE *out; - u8 *buf; - u8 *p; - int ret; - off_t subdir_offset; - struct dentry *root; - struct lookup_table_entry *lte; - struct resource_entry *res_entry; - off_t metadata_offset; - u64 metadata_original_size; - u64 metadata_compressed_size; - int metadata_ctype; - u8 hash[WIM_HASH_SIZE]; - - DEBUG("Writing metadata resource for image %u\n", w->current_image); - - out = w->out_fp; - root = wim_root_dentry(w); - metadata_ctype = wimlib_get_compression_type(w); - metadata_offset = ftello(out); - if (metadata_offset == -1) - return WIMLIB_ERR_WRITE; - - subdir_offset = 8 + root->length + 8; - calculate_subdir_offsets(root, &subdir_offset); - metadata_original_size = subdir_offset; - buf = MALLOC(metadata_original_size); - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for " - "metadata resource\n", metadata_original_size); - return WIMLIB_ERR_NOMEM; - } - p = buf; - #if 0 - /* Write the security data. */ - p = write_security_data(wim_security_data(w), p); - #else - p = put_u32(p, 8); /* Total length of security data. */ - p = put_u32(p, 0); /* Number of security data entries. */ - #endif - - DEBUG("Writing dentry tree.\n"); - p = write_dentry_tree(root, p); - - /* Like file resources, the lookup table entry for a metadata resource - * uses for the hash code a SHA1 message digest of its uncompressed - * contents. */ - sha1_buffer(buf, metadata_original_size, hash); - - ret = write_resource_from_memory(buf, - metadata_ctype, - metadata_original_size, - out, - &metadata_compressed_size); - FREE(buf); - if (ret != 0) - return ret; - - /* 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 for_dentry_in_tree(wim_root_dentry(w), + dentry_find_streams_to_write, w); } -/* Write the file resources for the current image. */ -static int write_file_resources(WIMStruct *w) +static int write_wim_streams(WIMStruct *w, int image, int write_flags, + unsigned num_threads) { - DEBUG("Writing file resources for image %u\n", w->current_image); - return for_dentry_in_tree(wim_root_dentry(w), write_file_resource, w); + for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); + LIST_HEAD(stream_list); + w->private = &stream_list; + for_image(w, image, find_streams_to_write); + return write_stream_list(&stream_list, w->out_fp, + wimlib_get_compression_type(w), write_flags, + num_threads); } -/* Write lookup table, xml data, lookup table, and rewrite header */ -static int finish_write(WIMStruct *w, int image, FILE *out, int flags) +/* + * Finish writing a WIM file: write the lookup table, xml data, and integrity + * table (optional), then overwrite the WIM header. + * + * write_flags is a bitwise OR of the following: + * + * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: + * Include an integrity table. + * + * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS: + * Show progress information when (if) writing the integrity table. + * + * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: + * Don't write the lookup table. + * + * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: + * When (if) writing the integrity table, re-use entries from the + * existing integrity table, if possible. + * + * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: + * After writing the XML data but before writing the integrity + * table, write a temporary WIM header and flush the stream so that + * the WIM is less likely to become corrupted upon abrupt program + * termination. + * + * (private) WIMLIB_WRITE_FLAG_FSYNC: + * fsync() the output file before closing it. + * + */ +int finish_write(WIMStruct *w, int image, int write_flags) { - off_t lookup_table_offset; - off_t xml_data_offset; - off_t lookup_table_size; - off_t integrity_offset; - off_t xml_data_size; - off_t end_offset; - off_t integrity_size; int ret; - int i; struct wim_header hdr; - - lookup_table_offset = ftello(out); - if (lookup_table_offset == -1) - return WIMLIB_ERR_WRITE; - - DEBUG("Writing lookup table.\n"); - /* Write the lookup table. */ - ret = write_lookup_table(w->lookup_table, out); - if (ret != 0) - return ret; - - DEBUG("Writing XML data.\n"); - - xml_data_offset = ftello(out); - if (xml_data_offset == -1) - return WIMLIB_ERR_WRITE; + FILE *out = w->out_fp; /* @hdr will be the header for the new WIM. First copy all the data * from the header in the WIMStruct; then set all the fields that may * have changed, including the resource entries, boot index, and image * count. */ memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); - lookup_table_size = xml_data_offset - lookup_table_offset; - hdr.lookup_table_res_entry.offset = lookup_table_offset; - hdr.lookup_table_res_entry.size = lookup_table_size; - hdr.lookup_table_res_entry.original_size = lookup_table_size; - hdr.lookup_table_res_entry.flags = WIM_RESHDR_FLAG_METADATA; - ret = write_xml_data(w->wim_info, image, out); + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); + if (ret != 0) + goto out; + } + + ret = write_xml_data(w->wim_info, image, out, + (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? + wim_info_get_total_bytes(w->wim_info) : 0, + &hdr.xml_res_entry); if (ret != 0) - return ret; + goto out; + + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { + struct wim_header checkpoint_hdr; + memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); + memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); + if (fseeko(out, 0, SEEK_SET) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + ret = write_header(&checkpoint_hdr, out); + if (ret != 0) + goto out; - integrity_offset = ftello(out); - if (integrity_offset == -1) - return WIMLIB_ERR_WRITE; - xml_data_size = integrity_offset - xml_data_offset; + if (fflush(out) != 0) { + ERROR_WITH_ERRNO("Can't write data to WIM"); + ret = WIMLIB_ERR_WRITE; + goto out; + } - 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 (fseeko(out, 0, SEEK_END) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } + } - 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); + off_t old_lookup_table_end; + off_t new_lookup_table_end; + bool show_progress; + if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { + old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + } else { + old_lookup_table_end = 0; + } + new_lookup_table_end = hdr.lookup_table_res_entry.offset + + hdr.lookup_table_res_entry.size; + show_progress = ((write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) != 0); + + ret = write_integrity_table(out, + &hdr.integrity, + new_lookup_table_end, + old_lookup_table_end, + show_progress); if (ret != 0) - return ret; - end_offset = ftello(out); - if (end_offset == -1) - return WIMLIB_ERR_WRITE; - integrity_size = end_offset - integrity_offset; - hdr.integrity.offset = integrity_offset; - hdr.integrity.size = integrity_size; - hdr.integrity.original_size = integrity_size; + goto out; } else { - hdr.integrity.offset = 0; - hdr.integrity.size = 0; - hdr.integrity.original_size = 0; + memset(&hdr.integrity, 0, sizeof(struct resource_entry)); } - hdr.integrity.flags = 0; - - DEBUG("Updating WIM header.\n"); - - /* + /* * 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, @@ -862,13 +1493,13 @@ static int finish_write(WIMStruct *w, int image, FILE *out, int flags) */ if (hdr.boot_idx == 0 || !w->image_metadata || (image != WIM_ALL_IMAGES && image != hdr.boot_idx)) { - memset(&hdr.boot_metadata_res_entry, 0, + memset(&hdr.boot_metadata_res_entry, 0, sizeof(struct resource_entry)); } else { - memcpy(&hdr.boot_metadata_res_entry, - &w->image_metadata[hdr.boot_idx - 1].lookup_table_entry-> - output_resource_entry, - sizeof(struct resource_entry)); + memcpy(&hdr.boot_metadata_res_entry, + &w->image_metadata[ + hdr.boot_idx - 1].metadata_lte->output_resource_entry, + sizeof(struct resource_entry)); } /* Set image count and boot index correctly for single image writes */ @@ -880,75 +1511,398 @@ static int finish_write(WIMStruct *w, int image, FILE *out, int flags) hdr.boot_idx = 0; } + if (fseeko(out, 0, SEEK_SET) != 0) { + ret = WIMLIB_ERR_WRITE; + goto out; + } - if (fseeko(out, 0, SEEK_SET) != 0) - return WIMLIB_ERR_WRITE; + ret = write_header(&hdr, out); + if (ret != 0) + goto out; + + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + if (fflush(out) != 0 + || fsync(fileno(out)) != 0) + { + ERROR_WITH_ERRNO("Error flushing data to WIM file"); + ret = WIMLIB_ERR_WRITE; + } + } +out: + if (fclose(out) != 0) { + ERROR_WITH_ERRNO("Failed to close the WIM file"); + if (ret == 0) + ret = WIMLIB_ERR_WRITE; + } + w->out_fp = NULL; + return ret; +} + +static void close_wim_writable(WIMStruct *w) +{ + if (w->out_fp) { + if (fclose(w->out_fp) != 0) { + WARNING("Failed to close output WIM: %s", + strerror(errno)); + } + w->out_fp = NULL; + } +} + +/* Open file stream and write dummy header for WIM. */ +int begin_write(WIMStruct *w, const char *path, int write_flags) +{ + int ret; + bool need_readable = false; + bool trunc = true; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) + need_readable = true; - return write_header(&hdr, out); + ret = open_wim_writable(w, path, trunc, need_readable); + if (ret != 0) + return ret; + /* Write dummy header. It will be overwritten later. */ + return write_header(&w->hdr, w->out_fp); } -/* Writes the WIM to a file. */ -WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, int image, int flags) +/* Writes a stand-alone WIM to a file. */ +WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, + int image, int write_flags, unsigned num_threads) { int ret; - const char *mode; - FILE *out; - if (image != WIM_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) + if (!w || !path) + return WIMLIB_ERR_INVALID_PARAM; + + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + + if (image != WIM_ALL_IMAGES && + (image < 1 || image > w->hdr.image_count)) return WIMLIB_ERR_INVALID_IMAGE; + if (w->hdr.total_parts != 1) { + ERROR("Cannot call wimlib_write() on part of a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; + } + if (image == WIM_ALL_IMAGES) - DEBUG("Writing all images to `%s'\n", path); + DEBUG("Writing all images to `%s'.", path); else - DEBUG("Writing image %d to `%s'\n", image, path); - - /* checking the integrity requires going back over the file to read it. - * XXX - * (It also would be possible to keep a running sha1sum as the file - * as written-- this would be faster, but a bit more complicated) */ - if (flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - mode = "w+b"; + DEBUG("Writing image %d to `%s'.", image, path); + + ret = begin_write(w, path, write_flags); + if (ret != 0) + goto out; + + ret = write_wim_streams(w, image, write_flags, num_threads); + if (ret != 0) + goto out; + + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + printf("Writing image metadata...\n"); + + ret = for_image(w, image, write_metadata_resource); + if (ret != 0) + goto out; + + ret = finish_write(w, image, write_flags); + if (ret == 0 && (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)) + printf("Successfully wrote `%s'\n", path); +out: + close_wim_writable(w); + return ret; +} + +static int lte_overwrite_prepare(struct lookup_table_entry *lte, + void *ignore) +{ + memcpy(<e->output_resource_entry, <e->resource_entry, + sizeof(struct resource_entry)); + lte->out_refcnt = 0; + return 0; +} + +static int check_resource_offset(struct lookup_table_entry *lte, void *arg) +{ + off_t end_offset = *(u64*)arg; + + wimlib_assert(lte->out_refcnt <= lte->refcnt); + if (lte->out_refcnt < lte->refcnt) { + if (lte->resource_entry.offset + lte->resource_entry.size > end_offset) { + ERROR("The following resource is after the XML data:"); + print_lookup_table_entry(lte); + return WIMLIB_ERR_RESOURCE_ORDER; + } + } + return 0; +} + +static int find_new_streams(struct lookup_table_entry *lte, void *arg) +{ + if (lte->out_refcnt == lte->refcnt) + list_add(<e->staging_list, (struct list_head*)arg); else - mode = "wb"; + lte->out_refcnt = lte->refcnt; + return 0; +} - out = fopen(path, mode); - if (!out) { - ERROR("Failed to open the file `%s' for writing!\n", - path); - return WIMLIB_ERR_OPEN; +/* + * Overwrite a WIM, possibly appending streams to it. + * + * A WIM looks like (or is supposed to look like) the following: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * If we are not adding any streams or metadata resources, the lookup table is + * unchanged--- so we only need to overwrite the XML data, integrity table, and + * header. This operation is potentially unsafe if the program is abruptly + * terminated while the XML data or integrity table are being overwritten, but + * before the new header has been written. To partially alleviate this problem, + * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to + * finish_write() to cause a temporary WIM header to be written after the XML + * data has been written. This may prevent the WIM from becoming corrupted if + * the program is terminated while the integrity table is being calculated (but + * no guarantees, due to write re-ordering...). + * + * If we are adding new streams or images (metadata resources), the lookup table + * needs to be changed, and those streams need to be written. In this case, we + * try to perform a safe update of the WIM file by writing the streams *after* + * the end of the previous WIM, then writing the new lookup table, XML data, and + * (optionally) integrity table following the new streams. This will produce a + * layout like the following: + * + * Header (212 bytes) + * (OLD) Streams and metadata resources (variable size) + * (OLD) Lookup table (variable size) + * (OLD) XML data (variable size) + * (OLD) Integrity table (optional) (variable size) + * (NEW) Streams and metadata resources (variable size) + * (NEW) Lookup table (variable size) + * (NEW) XML data (variable size) + * (NEW) Integrity table (optional) (variable size) + * + * At all points, the WIM is valid as nothing points to the new data yet. Then, + * the header is overwritten to point to the new lookup table, XML data, and + * integrity table, to produce the following layout: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Nothing (variable size) + * More Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * This method allows an image to be appended to a large WIM very quickly, and + * is is crash-safe except in the case of write re-ordering, but the + * disadvantage is that a small hole is left in the WIM where the old lookup + * table, xml data, and integrity table were. (These usually only take up a + * small amount of space compared to the streams, however. + */ +static int overwrite_wim_inplace(WIMStruct *w, int write_flags, + unsigned num_threads, + int modified_image_idx) +{ + int ret; + struct list_head stream_list; + off_t old_wim_end; + + DEBUG("Overwriting `%s' in-place", w->filename); + + /* Make sure that the integrity table (if present) is after the XML + * data, and that there are no stream resources, metadata resources, or + * lookup tables after the XML data. Otherwise, these data would be + * overwritten. */ + if (w->hdr.integrity.offset != 0 && + w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the integrity table to be before the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } + + if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the lookup table to be after the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; } - w->out_fp = out; + DEBUG("Identifying newly added streams"); + for_lookup_table_entry(w->lookup_table, lte_overwrite_prepare, NULL); + INIT_LIST_HEAD(&stream_list); + for (int i = modified_image_idx; i < w->hdr.image_count; i++) { + DEBUG("Identifiying streams in image %d", i + 1); + wimlib_assert(w->image_metadata[i].modified); + wimlib_assert(!w->image_metadata[i].has_been_mounted_rw); + wimlib_assert(w->image_metadata[i].root_dentry != NULL); + wimlib_assert(w->image_metadata[i].metadata_lte != NULL); + w->private = &stream_list; + for_dentry_in_tree(w->image_metadata[i].root_dentry, + dentry_find_streams_to_write, w); + } - /* Write dummy header. It will be overwritten later. */ - ret = write_header(&w->hdr, out); + if (w->hdr.integrity.offset) + old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size; + else + old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size; + + ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, + &old_wim_end); if (ret != 0) - goto done; + return ret; - for_lookup_table_entry(w->lookup_table, zero_out_refcnts, NULL); + if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) { + /* If no images have been modified and no images have been + * deleted, a new lookup table does not need to be written. */ + wimlib_assert(list_empty(&stream_list)); + old_wim_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | + WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; + } - ret = for_image(w, image, write_file_resources); + INIT_LIST_HEAD(&stream_list); + for_lookup_table_entry(w->lookup_table, find_new_streams, + &stream_list); + + ret = open_wim_writable(w, w->filename, false, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); + if (ret != 0) + return ret; + + if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + return WIMLIB_ERR_WRITE; + } + + if (!list_empty(&stream_list)) { + DEBUG("Writing newly added streams (offset = %"PRIu64")", + old_wim_end); + ret = write_stream_list(&stream_list, w->out_fp, + wimlib_get_compression_type(w), + write_flags, num_threads); + if (ret != 0) + goto out_ftruncate; + } else { + DEBUG("No new streams were added"); + } + + for (int i = modified_image_idx; i < w->hdr.image_count; i++) { + select_wim_image(w, i + 1); + ret = write_metadata_resource(w); + if (ret != 0) + goto out_ftruncate; + } + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; + ret = finish_write(w, WIM_ALL_IMAGES, write_flags); +out_ftruncate: + close_wim_writable(w); if (ret != 0) { - ERROR("Failed to write file resources!\n"); - goto done; + WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + w->filename, old_wim_end); + truncate(w->filename, old_wim_end); } + return ret; +} - ret = for_image(w, image, write_metadata_resource); +static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, + unsigned num_threads) +{ + size_t wim_name_len; + int ret; + + DEBUG("Overwrining `%s' via a temporary file", w->filename); + + /* Write the WIM to a temporary file in the same directory as the + * original WIM. */ + wim_name_len = strlen(w->filename); + char tmpfile[wim_name_len + 10]; + memcpy(tmpfile, w->filename, wim_name_len); + randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); + tmpfile[wim_name_len + 9] = '\0'; + ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, + write_flags | WIMLIB_WRITE_FLAG_FSYNC, + num_threads); if (ret != 0) { - ERROR("Failed to write image metadata!\n"); - goto done; + ERROR("Failed to write the WIM file `%s'", tmpfile); + goto err; } - ret = finish_write(w, image, out, flags); + /* Close the original WIM file that was opened for reading. */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } -done: - DEBUG("Closing output file.\n"); - w->out_fp = NULL; - if (fclose(out) != 0) { - ERROR("Failed to close the file `%s': %m\n", path); - ret = WIMLIB_ERR_WRITE; + DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); + + /* Rename the new file to the old file .*/ + if (rename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + tmpfile, w->filename); + ret = WIMLIB_ERR_RENAME; + goto err; + } + + if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) + printf("Successfully renamed `%s' to `%s'\n", tmpfile, w->filename); + + /* Re-open the WIM read-only. */ + w->fp = fopen(w->filename, "rb"); + if (w->fp == NULL) { + ret = WIMLIB_ERR_REOPEN; + WARNING("Failed to re-open `%s' read-only: %s", + w->filename, strerror(errno)); } return ret; +err: + /* Remove temporary file. */ + if (unlink(tmpfile) != 0) + WARNING("Failed to remove `%s': %s", tmpfile, strerror(errno)); + return ret; +} + +/* + * Writes a WIM file to the original file that it was read from, overwriting it. + */ +WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags, + unsigned num_threads) +{ + if (!w) + return WIMLIB_ERR_INVALID_PARAM; + + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + + if (!w->filename) + return WIMLIB_ERR_NO_FILENAME; + + if (w->hdr.total_parts != 1) { + ERROR("Cannot modify a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; + } + + if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) + { + int i, modified_image_idx; + for (i = 0; i < w->hdr.image_count && !w->image_metadata[i].modified; i++) + ; + modified_image_idx = i; + for (; i < w->hdr.image_count && w->image_metadata[i].modified && + !w->image_metadata[i].has_been_mounted_rw; i++) + ; + if (i == w->hdr.image_count) { + return overwrite_wim_inplace(w, write_flags, num_threads, + modified_image_idx); + } + } + return overwrite_wim_via_tmpfile(w, write_flags, num_threads); +} + +/* Deprecated */ +WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *wim, int write_flags) +{ + return wimlib_overwrite(wim, write_flags, 1); }