X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=ccad7f07caa5c3f4b7af309995ab5974d2742aa0;hp=a08bc8503a6c5a57f908ab7f535787205fe3fded;hb=062465170ddb4520ca6fbc64a0f6e27e786de101;hpb=9b9a502863318d6208da2818b1d522346e5eee9e diff --git a/src/write.c b/src/write.c index a08bc850..ccad7f07 100644 --- a/src/write.c +++ b/src/write.c @@ -6,8 +6,7 @@ */ /* - * Copyright (C) 2010 Carl Thijssen - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -25,52 +24,61 @@ * along with wimlib; if not, see http://www.gnu.org/licenses/. */ +#include "config.h" + +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +/* On BSD, this should be included before "list.h" so that "list.h" can + * overwrite the LIST_HEAD macro. */ +# include +#endif + +#ifdef __WIN32__ +# include "win32.h" +#endif + +#include "list.h" #include "wimlib_internal.h" -#include "io.h" +#include "buffer_io.h" #include "dentry.h" #include "lookup_table.h" #include "xml.h" -#include "lzx.h" -#include "xpress.h" -#include #ifdef ENABLE_MULTITHREADED_COMPRESSION -#include -#include -#include +# include #endif +#include +#include + #ifdef WITH_NTFS_3G -#include -#include -#include -#include +# include +# include +# include +# include #endif - #ifdef HAVE_ALLOCA_H -#include +# include #else -#include +# include #endif -static int do_fflush(FILE *fp) +#include + +#if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE) +# define INVALID_HANDLE_VALUE ((HANDLE)(-1)) +#endif + +static int +fflush_and_ftruncate(FILE *fp, off_t size) { - int ret = fflush(fp); + int ret; + + 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; - - 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 " @@ -99,7 +107,7 @@ struct chunk_table { * output file. */ static int -begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, +begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, FILE *out_fp, off_t file_offset, struct chunk_table **chunk_tab_ret) @@ -140,28 +148,38 @@ out: } /* - * Pointer to function to compresses a chunk of a WIM resource. + * compress_func_t- Pointer to a function to compresses a chunk + * of a WIM resource. This may be either + * wimlib_xpress_compress() (xpress-compress.c) or + * wimlib_lzx_compress() (lzx-compress.c). * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk in bytes. - * @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. + * @chunk: Uncompressed data of the chunk. + * @chunk_size: Size of the uncompressed chunk, in bytes. + * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes. + * + * Returns the size of the compressed data written to @out in bytes, or 0 if the + * data could not be compressed to (@chunk_size - 1) bytes or fewer. * - * 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. + * As a special requirement, the compression code is optimized for the WIM + * format and therefore requires (@chunk_size <= 32768). + * + * As another special requirement, the compression code will read up to 8 bytes + * off the end of the @chunk array for performance reasons. The values of these + * bytes will not affect the output of the compression, but the calling code + * must make sure that the buffer holding the uncompressed chunk is actually at + * least (@chunk_size + 8) bytes, or at least that these extra bytes are in + * mapped memory that will not cause a memory access violation if accessed. */ -typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); +typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, + void *out); -compress_func_t get_compress_func(int out_ctype) +compress_func_t +get_compress_func(int out_ctype) { if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return lzx_compress; + return wimlib_lzx_compress; else - return xpress_compress; + return wimlib_xpress_compress; } /* @@ -169,35 +187,37 @@ compress_func_t get_compress_func(int out_ctype) * * @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) + * @out_fp: FILE * to write the chunk to. + * @compress: Compression function to use (NULL if writing uncompressed + * data). * @chunk_tab: Pointer to chunk table being created. It is updated with the * offset of the chunk we write. * * Returns 0 on success; nonzero on failure. */ -static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, - FILE *out_fp, compress_func_t compress, - struct chunk_table *chunk_tab) +static int +write_wim_resource_chunk(const void *chunk, unsigned chunk_size, + FILE *out_fp, compress_func_t compress, + struct chunk_table *chunk_tab) { const u8 *out_chunk; unsigned out_chunk_size; - if (chunk_tab) { + if (compress) { u8 *compressed_chunk = alloca(chunk_size); - int ret; - ret = compress(chunk, chunk_size, compressed_chunk, - &out_chunk_size); - if (ret == 0) { + out_chunk_size = compress(chunk, chunk_size, compressed_chunk); + if (out_chunk_size) { + /* Write compressed */ out_chunk = compressed_chunk; } else { + /* Write uncompressed */ out_chunk = chunk; out_chunk_size = chunk_size; } *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; chunk_tab->cur_offset += out_chunk_size; } else { + /* Write uncompressed */ out_chunk = chunk; out_chunk_size = chunk_size; } @@ -209,7 +229,7 @@ static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, } /* - * Finishes a WIM chunk tale and writes it to the output file at the correct + * Finishes a WIM chunk table and writes it to the output file at the correct * offset. * * The final size of the full compressed resource is returned in the @@ -249,348 +269,242 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, return 0; } -/* 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 - - #ifdef WITH_NTFS_3G - , ntfs_inode **ni_ret - #endif - ) +static int +write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte, + FILE *out_fp, + off_t file_offset, + struct resource_entry *out_res_entry) { - 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; + int ret; + if (fseeko(out_fp, file_offset, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of " + "output WIM file", file_offset); + return WIMLIB_ERR_WRITE; } -#endif - return 0; + ret = write_wim_resource(lte, out_fp, + WIMLIB_COMPRESSION_TYPE_NONE, + out_res_entry, + 0); + if (ret) + return ret; + + return fflush_and_ftruncate(out_fp, + file_offset + wim_resource_size(lte)); } -/* 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 - ) +struct write_resource_ctx { + compress_func_t compress; + struct chunk_table *chunk_tab; + FILE *out_fp; + SHA_CTX sha_ctx; + bool doing_sha; +}; + +static int +write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) { - 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; + struct write_resource_ctx *ctx = _ctx; + + if (ctx->doing_sha) + sha1_update(&ctx->sha_ctx, chunk, chunk_size); + + if (ctx->compress) { + return write_wim_resource_chunk(chunk, chunk_size, + ctx->out_fp, ctx->compress, + ctx->chunk_tab); + } else { + if (fwrite(chunk, 1, chunk_size, ctx->out_fp) != chunk_size) { + ERROR_WITH_ERRNO("Error writing to output WIM"); + return WIMLIB_ERR_WRITE; + } else { + return 0; } - if (ni) - ntfs_inode_close(ni); } -#endif } -/* - * 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) +int +write_wim_resource(struct wim_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; + struct write_resource_ctx write_ctx; + u64 new_size; + off_t offset; int ret; - struct chunk_table *chunk_tab = NULL; - bool raw; - off_t file_offset; - compress_func_t compress = NULL; -#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); + if (wim_resource_size(lte) == 0) { + /* Empty resource; nothing needs to be done, so just return + * success. */ + return 0; + } - /* Current offset in output file */ - file_offset = ftello(out_fp); - if (file_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset in output " - "stream"); + offset = ftello(out_fp); + if (offset == -1) { + ERROR_WITH_ERRNO("Can't get position in output WIM"); return WIMLIB_ERR_WRITE; } - /* 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 != WIMLIB_COMPRESSION_TYPE_NONE - && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); + /* Can we simply copy the compressed data without recompressing it? */ - if (raw) { + if (!(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && + lte->resource_location == RESOURCE_IN_WIM && + wimlib_get_compression_type(lte->wim) == out_ctype) + { flags |= WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = old_compressed_size; + write_ctx.doing_sha = false; } else { - flags &= ~WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = original_size; + write_ctx.doing_sha = true; + sha1_init(&write_ctx.sha_ctx); } - /* Empty resource; nothing needs to be done, so just return success. */ - if (bytes_remaining == 0) - return 0; - - /* Buffer for reading chunks for the resource */ - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - - /* If we are writing a compressed resource and not doing a raw copy, we - * need to initialize the chunk table */ - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) { - ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, - &chunk_tab); - if (ret != 0) - goto out; - } - - /* 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 WIMLIB_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; + /* Initialize the chunk table and set the compression function if + * compressing the resource. */ + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || + (flags & WIMLIB_RESOURCE_FLAG_RAW)) { + write_ctx.compress = NULL; + write_ctx.chunk_tab = NULL; } else { - if (out_ctype == WIMLIB_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) - goto out_fclose; - } + write_ctx.compress = get_compress_func(out_ctype); + ret = begin_wim_resource_chunk_tab(lte, out_fp, + offset, + &write_ctx.chunk_tab); + if (ret) + return ret; } - /* 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) { + /* Write the data */ + write_ctx.out_fp = out_fp; + ret = read_resource_prefix(lte, wim_resource_size(lte), + write_resource_cb, &write_ctx, 0); + + /* Verify SHA1 message digest of the resource, or set the hash for the + * first time. */ + if (write_ctx.doing_sha) { u8 md[SHA1_HASH_SIZE]; - sha1_final(md, &ctx); - if (is_zero_hash(lte->hash)) { + sha1_final(md, &write_ctx.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->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; maybe it changed " - "while we were reading it.", + if (lte_filename_valid(lte)) { + ERROR("We were reading it from \"%"TS"\"; maybe " + "it changed while we were reading it.", lte->file_on_disk); } ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out_fclose; + goto out_free_chunk_tab; } } - if (!raw && new_compressed_size >= original_size && - out_ctype != WIMLIB_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, WIMLIB_COMPRESSION_TYPE_NONE, - out_res_entry, flags); - if (ret != 0) - goto out_fclose; - - ret = fflush_and_ftruncate(out_fp, file_offset + out_res_entry->size); - if (ret != 0) - goto out_fclose; + out_res_entry->flags = lte->resource_entry.flags; + out_res_entry->original_size = wim_resource_size(lte); + out_res_entry->offset = offset; + if (flags & WIMLIB_RESOURCE_FLAG_RAW) { + /* Doing a raw write: The new compressed size is the same as + * the compressed size in the other WIM. */ + new_size = lte->resource_entry.size; + } else if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) { + /* Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size + * is the original size. */ + new_size = lte->resource_entry.original_size; + out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; } else { - 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 != WIMLIB_COMPRESSION_TYPE_NONE) - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; + /* Using a different compression type: Call + * finish_wim_resource_chunk_tab() and it will provide the new + * compressed size. */ + ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fp, + &new_size); + if (ret) + goto out_free_chunk_tab; + if (new_size >= wim_resource_size(lte)) { + /* Oops! We compressed the resource to larger than the original + * size. Write the resource uncompressed instead. */ + ret = write_uncompressed_resource_and_truncate(lte, + out_fp, + offset, + out_res_entry); + goto out_free_chunk_tab; } + out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } + out_res_entry->size = new_size; 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); +out_free_chunk_tab: + FREE(write_ctx.chunk_tab); return ret; } #ifdef ENABLE_MULTITHREADED_COMPRESSION + +/* Blocking shared queue (solves the producer-consumer problem) */ struct shared_queue { - sem_t filled_slots; - sem_t empty_slots; - pthread_mutex_t lock; + unsigned size; unsigned front; unsigned back; + unsigned filled_slots; void **array; - unsigned size; + pthread_mutex_t lock; + pthread_cond_t msg_avail_cond; + pthread_cond_t space_avail_cond; }; -static int shared_queue_init(struct shared_queue *q, unsigned size) +static int +shared_queue_init(struct shared_queue *q, unsigned size) { + wimlib_assert(size != 0); q->array = CALLOC(sizeof(q->array[0]), size); if (!q->array) return WIMLIB_ERR_NOMEM; - - sem_init(&q->filled_slots, 0, 0); - sem_init(&q->empty_slots, 0, size); - pthread_mutex_init(&q->lock, NULL); + q->filled_slots = 0; q->front = 0; q->back = size - 1; q->size = size; + pthread_mutex_init(&q->lock, NULL); + pthread_cond_init(&q->msg_avail_cond, NULL); + pthread_cond_init(&q->space_avail_cond, NULL); return 0; } -static void shared_queue_destroy(struct shared_queue *q) +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); + pthread_mutex_destroy(&q->lock); + pthread_cond_destroy(&q->msg_avail_cond); + pthread_cond_destroy(&q->space_avail_cond); } -static void shared_queue_put(struct shared_queue *q, void *obj) +static void +shared_queue_put(struct shared_queue *q, void *obj) { - sem_wait(&q->empty_slots); pthread_mutex_lock(&q->lock); + while (q->filled_slots == q->size) + pthread_cond_wait(&q->space_avail_cond, &q->lock); q->back = (q->back + 1) % q->size; q->array[q->back] = obj; + q->filled_slots++; - sem_post(&q->filled_slots); + pthread_cond_broadcast(&q->msg_avail_cond); pthread_mutex_unlock(&q->lock); } -static void *shared_queue_get(struct shared_queue *q) +static void * +shared_queue_get(struct shared_queue *q) { - sem_wait(&q->filled_slots); + void *obj; + pthread_mutex_lock(&q->lock); + while (q->filled_slots == 0) + pthread_cond_wait(&q->msg_avail_cond, &q->lock); - void *obj = q->array[q->front]; + obj = q->array[q->front]; q->array[q->front] = NULL; q->front = (q->front + 1) % q->size; + q->filled_slots--; - sem_post(&q->empty_slots); + pthread_cond_broadcast(&q->space_avail_cond); pthread_mutex_unlock(&q->lock); return obj; } @@ -604,7 +518,7 @@ struct compressor_thread_params { #define MAX_CHUNKS_PER_MSG 2 struct message { - struct lookup_table_entry *lte; + struct wim_lookup_table_entry *lte; u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG]; u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; @@ -616,24 +530,34 @@ struct message { u64 begin_chunk; }; -static void compress_chunks(struct message *msg, compress_func_t compress) +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) { + unsigned len = compress(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i]); + if (len) { + /* To be written compressed */ msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + msg->compressed_chunk_sizes[i] = len; } else { + /* To be written uncompressed */ msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; + } } } -static void *compressor_thread_proc(void *arg) +/* 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; @@ -649,44 +573,137 @@ static void *compressor_thread_proc(void *arg) DEBUG("Compressor thread terminating"); return NULL; } -#endif +#endif /* ENABLE_MULTITHREADED_COMPRESSION */ + +static void +do_write_streams_progress(union wimlib_progress_info *progress, + wimlib_progress_func_t progress_func, + uint64_t size_added) +{ + progress->write_streams.completed_bytes += size_added; + progress->write_streams.completed_streams++; + if (progress_func && + progress->write_streams.completed_bytes >= progress->write_streams._private) + { + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress->write_streams._private == progress->write_streams.total_bytes) { + progress->write_streams._private = ~0; + } else { + progress->write_streams._private = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); + } + } +} -static int do_write_stream_list(struct list_head *my_resources, - FILE *out_fp, - int out_ctype, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress, - int write_resource_flags) +static int +sha1_chunk(const void *buf, size_t len, void *ctx) +{ + sha1_update(ctx, buf, len); + return 0; +} + +static int +sha1_resource(struct wim_lookup_table_entry *lte) { int ret; - struct lookup_table_entry *lte, *tmp; + SHA_CTX sha_ctx; + + sha1_init(&sha_ctx); + ret = read_resource_prefix(lte, wim_resource_size(lte), + sha1_chunk, &sha_ctx, 0); + if (ret == 0) + sha1_final(lte->hash, &sha_ctx); + return ret; +} + +enum { + STREAMS_MERGED = 0, + STREAMS_NOT_MERGED = 1, +}; + +static int +do_write_stream_list(struct list_head *my_resources, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress, + int write_resource_flags) +{ + int ret; + struct wim_lookup_table_entry *lte; + + while (!list_empty(my_resources)) { + lte = container_of(my_resources->next, + struct wim_lookup_table_entry, + write_streams_list); + list_del(<e->write_streams_list); + if (lte->unhashed && !lte->unique_size) { + struct wim_lookup_table_entry *duplicate_lte; + struct wim_lookup_table_entry **my_ptr; + + my_ptr = lte->my_ptr; + ret = sha1_resource(lte); + if (ret) + return ret; + duplicate_lte = __lookup_resource(lookup_table, lte->hash); + if (duplicate_lte) { + bool new_stream = (duplicate_lte->out_refcnt == 0); + duplicate_lte->refcnt += lte->refcnt; + duplicate_lte->out_refcnt += lte->refcnt; + *my_ptr = duplicate_lte; + free_lookup_table_entry(lte); + + if (new_stream) { + lte = duplicate_lte; + DEBUG("Stream of length %"PRIu64" is duplicate " + "with one already in WIM", + wim_resource_size(duplicate_lte)); + } else { + DEBUG("Discarding duplicate stream of length %"PRIu64, + wim_resource_size(duplicate_lte)); + goto skip_to_progress; + } + + } else { + lookup_table_insert(lookup_table, lte); + lte->out_refcnt = lte->refcnt; + lte->unhashed = 0; + } + } + + wimlib_assert(lte->out_refcnt != 0); - list_for_each_entry_safe(lte, tmp, my_resources, staging_list) { ret = write_wim_resource(lte, out_fp, out_ctype, <e->output_resource_entry, write_resource_flags); - if (ret != 0) + if (ret) return ret; - list_del(<e->staging_list); - progress->write_streams.completed_bytes += - wim_resource_size(lte); - progress->write_streams.completed_streams++; - if (progress_func) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); + if (lte->unhashed) { + lookup_table_insert(lookup_table, lte); + lte->unhashed = 0; } + skip_to_progress: + do_write_streams_progress(progress, + progress_func, + wim_resource_size(lte)); } return 0; } -static int write_stream_list_serial(struct list_head *stream_list, - FILE *out_fp, - int out_ctype, - int write_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) +static int +write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + int write_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { int write_resource_flags; @@ -697,14 +714,17 @@ static int write_stream_list_serial(struct list_head *stream_list, progress->write_streams.num_threads = 1; if (progress_func) progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); - return do_write_stream_list(stream_list, out_fp, + return do_write_stream_list(stream_list, + lookup_table, + out_fp, out_ctype, progress_func, progress, write_resource_flags); } #ifdef ENABLE_MULTITHREADED_COMPRESSION -static int write_wim_chunks(struct message *msg, FILE *out_fp, - struct chunk_table *chunk_tab) +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]; @@ -737,24 +757,31 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, * res_to_compress_queue, and it is passed back through the * compressed_res_queue. */ -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, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) +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 num_messages, + int write_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { int ret; - - struct message msgs[queue_size]; - ZERO_ARRAY(msgs); + struct chunk_table *cur_chunk_tab = NULL; + struct message *msgs = CALLOC(num_messages, sizeof(struct message)); + struct wim_lookup_table_entry *next_lte = NULL; // Initially, all the messages are available to use. LIST_HEAD(available_msgs); - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) + + if (!msgs) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } + + for (size_t i = 0; i < num_messages; i++) list_add(&msgs[i].list, &available_msgs); // outstanding_resources is the list of resources that currently have @@ -774,55 +801,43 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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); + u64 next_num_chunks = 0; // 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 wim_lookup_table_entry *cur_lte = NULL; struct message *msg; #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); + num_messages * 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 i = 0; i < num_messages; 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); + + // The extra 8 bytes is because longest_match() in + // lz77.c may read a little bit off the end of the + // uncompressed data. It doesn't need to be + // initialized--- we really just need to avoid accessing + // an unmapped page. + msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8); 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; } @@ -837,7 +852,95 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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) { + while (!list_empty(&available_msgs)) { + if (next_chunk == next_num_chunks) { + // If next_chunk == next_num_chunks, there are + // no more chunks to write in the current + // stream. So, check the SHA1 message digest of + // the stream that was just finished (unless + // next_lte == NULL, which is the case the very + // first time this loop is entered, and also + // near the very end of the compression when + // there are no more streams.) Then, advance to + // the next stream (if there is one). + if (next_lte != NULL) { + #ifdef WITH_NTFS_3G + end_wim_resource_read(next_lte, ni); + ni = NULL; + #else + end_wim_resource_read(next_lte); + #endif + 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 `%"TS"'; " + "maybe it changed while we were " + "reading it.", + next_lte->file_on_disk); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + goto out; + } + } + + // 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) { + // No more resources to send for + // compression + next_lte = NULL; + break; + } + next_lte = container_of(next_resource, + struct wim_lookup_table_entry, + staging_list); + next_resource = next_resource->next; + if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + && wim_resource_compression_type(next_lte) == 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; + if (cur_lte == NULL) { + // Set cur_lte for the + // first time + cur_lte = next_lte; + } + break; + } + } + } + + if (next_lte == NULL) { + // No more resources to send for compression + break; + } // Get a message from the available messages // list @@ -864,13 +967,11 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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; + if (next_chunk == next_num_chunks - 1) { + size = MODULO_NONZERO(wim_resource_size(next_lte), + WIM_CHUNK_SIZE); } - DEBUG2("Read resource (size=%u, offset=%zu)", size, next_chunk * WIM_CHUNK_SIZE); @@ -892,81 +993,11 @@ static int main_writer_thread_proc(struct list_head *stream_list, 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; - } - - // 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; - } - } } // 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; } @@ -974,26 +1005,20 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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; - DEBUG2("Received msg (begin_chunk=%"PRIu64")", msg->begin_chunk); - - list_for_each_entry(msg, &cur_lte->msg_list, list) { - DEBUG2("complete=%d", msg->complete); - } - // 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) + while (cur_lte != NULL && + !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) { @@ -1025,8 +1050,8 @@ static int main_writer_thread_proc(struct list_head *stream_list, // now. list_add(&msg->list, &available_msgs); - // Was this the last chunk of the stream? If so, - // finish it. + // 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) { @@ -1038,27 +1063,34 @@ static int main_writer_thread_proc(struct list_head *stream_list, if (ret != 0) goto out; - progress->write_streams.completed_bytes += - wim_resource_size(cur_lte); - progress->write_streams.completed_streams++; - - if (progress_func) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - } + if (res_csize >= wim_resource_size(cur_lte)) { + /* Oops! We compressed the resource to + * larger than the original size. Write + * the resource uncompressed instead. */ + ret = write_uncompressed_resource_and_truncate( + cur_lte, + out_fp, + cur_chunk_tab->file_offset, + &cur_lte->output_resource_entry); + if (ret != 0) + goto out; + } else { + cur_lte->output_resource_entry.size = + res_csize; - 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.original_size = - cur_lte->resource_entry.original_size; + cur_lte->output_resource_entry.offset = + cur_chunk_tab->file_offset; - 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; + } - cur_lte->output_resource_entry.flags = - cur_lte->resource_entry.flags | - WIM_RESHDR_FLAG_COMPRESSED; + do_write_streams_progress(progress, progress_func, + wim_resource_size(cur_lte)); FREE(cur_chunk_tab); cur_chunk_tab = NULL; @@ -1066,15 +1098,12 @@ static int main_writer_thread_proc(struct list_head *stream_list, 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 { + if (next == &outstanding_resources) + cur_lte = NULL; + else cur_lte = container_of(cur_lte->staging_list.next, - struct lookup_table_entry, + struct wim_lookup_table_entry, staging_list); - } // Since we just finished writing a stream, // write any streams that have been added to the @@ -1096,49 +1125,72 @@ static int main_writer_thread_proc(struct list_head *stream_list, } out: + if (ret == WIMLIB_ERR_NOMEM) { + ERROR("Could not allocate enough memory for " + "multi-threaded compression"); + } + + if (next_lte) { #ifdef WITH_NTFS_3G - end_wim_resource_read(cur_lte, ni); + end_wim_resource_read(next_lte, ni); #else - end_wim_resource_read(cur_lte); + end_wim_resource_read(next_lte); #endif + } + if (ret == 0) { ret = do_write_stream_list(&my_resources, out_fp, out_ctype, progress_func, progress, 0); } else { - size_t num_available_msgs = 0; - struct list_head *cur; + if (msgs) { + size_t num_available_msgs = 0; + struct list_head *cur; - list_for_each(cur, &available_msgs) { - num_available_msgs++; - } + list_for_each(cur, &available_msgs) { + num_available_msgs++; + } - while (num_available_msgs < ARRAY_LEN(msgs)) { - shared_queue_get(compressed_res_queue); - num_available_msgs++; + while (num_available_msgs < num_messages) { + shared_queue_get(compressed_res_queue); + num_available_msgs++; + } } } - 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]); + if (msgs) { + for (size_t i = 0; i < num_messages; i++) { + for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { + FREE(msgs[i].compressed_chunks[j]); + FREE(msgs[i].uncompressed_chunks[j]); + } } + FREE(msgs); } - if (cur_chunk_tab != NULL) - FREE(cur_chunk_tab); + FREE(cur_chunk_tab); return ret; } +static long +get_default_num_threads() +{ +#ifdef __WIN32__ + return win32_get_number_of_processors(); +#else + return sysconf(_SC_NPROCESSORS_ONLN); +#endif +} -static int write_stream_list_parallel(struct list_head *stream_list, - FILE *out_fp, - int out_ctype, - int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) +static int +write_stream_list_parallel(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, + int out_ctype, + int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { int ret; struct shared_queue res_to_compress_queue; @@ -1146,8 +1198,8 @@ static int write_stream_list_parallel(struct list_head *stream_list, pthread_t *compressor_threads = NULL; if (num_threads == 0) { - long nthreads = sysconf(_SC_NPROCESSORS_ONLN); - if (nthreads < 1) { + long nthreads = get_default_num_threads(); + if (nthreads < 1 || nthreads > UINT_MAX) { WARNING("Could not determine number of processors! Assuming 1"); goto out_serial; } else { @@ -1177,6 +1229,10 @@ static int write_stream_list_parallel(struct list_head *stream_list, params.compress = get_compress_func(out_ctype); compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + if (!compressor_threads) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } for (unsigned i = 0; i < num_threads; i++) { DEBUG("pthread_create thread %u", i); @@ -1209,11 +1265,12 @@ out_join: 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)); + WARNING_WITH_ERRNO("Failed to join compressor " + "thread %u", i); } } FREE(compressor_threads); +out_destroy_compressed_res_queue: shared_queue_destroy(&compressed_res_queue); out_destroy_res_to_compress_queue: shared_queue_destroy(&res_to_compress_queue); @@ -1222,6 +1279,7 @@ out_destroy_res_to_compress_queue: out_serial: WARNING("Falling back to single-threaded compression"); return write_stream_list_serial(stream_list, + lookup_table, out_fp, out_ctype, write_flags, @@ -1235,29 +1293,28 @@ out_serial: * 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, - wimlib_progress_func_t progress_func) +static int +write_stream_list(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + FILE *out_fp, int out_ctype, int write_flags, + unsigned num_threads, wimlib_progress_func_t progress_func) { - struct lookup_table_entry *lte; + struct wim_lookup_table_entry *lte; size_t num_streams = 0; u64 total_bytes = 0; - bool compression_needed = false; + u64 total_compression_bytes = 0; union wimlib_progress_info progress; int ret; - list_for_each_entry(lte, stream_list, staging_list) { + list_for_each_entry(lte, stream_list, write_streams_list) { num_streams++; total_bytes += wim_resource_size(lte); - if (!compression_needed - && - (out_ctype != WIMLIB_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; + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE + && (wim_resource_compression_type(lte) != out_ctype || + (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS))) + { + total_compression_bytes += wim_resource_size(lte); + } } progress.write_streams.total_bytes = total_bytes; progress.write_streams.total_streams = num_streams; @@ -1265,69 +1322,237 @@ static int write_stream_list(struct list_head *stream_list, FILE *out_fp, progress.write_streams.completed_streams = 0; progress.write_streams.num_threads = num_threads; progress.write_streams.compression_type = out_ctype; - - if (num_streams == 0) { - ret = 0; - goto out; - } + progress.write_streams._private = 0; #ifdef ENABLE_MULTITHREADED_COMPRESSION - if (compression_needed && total_bytes >= 1000000 && num_threads != 1) { + if (total_compression_bytes >= 1000000 && num_threads != 1) ret = write_stream_list_parallel(stream_list, + lookup_table, out_fp, out_ctype, write_flags, num_threads, progress_func, &progress); - } else #endif - { ret = write_stream_list_serial(stream_list, + lookup_table, out_fp, out_ctype, write_flags, progress_func, &progress); - } -out: return ret; } +struct lte_overwrite_prepare_args { + WIMStruct *wim; + off_t end_offset; + struct list_head *stream_list; +}; + +static int +lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg) +{ + struct lte_overwrite_prepare_args *args = arg; + + if (lte->resource_location == RESOURCE_IN_WIM && + lte->wim == args->wim && + lte->resource_entry.offset + lte->resource_entry.size > args->end_offset) + { + #ifdef ENABLE_ERROR_MESSAGES + ERROR("The following resource is after the XML data:"); + print_lookup_table_entry(lte, stderr); + #endif + return WIMLIB_ERR_RESOURCE_ORDER; + } + + lte->out_refcnt = lte->refcnt; + memcpy(<e->output_resource_entry, <e->resource_entry, + sizeof(struct resource_entry)); + if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) + if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) + list_add(<e->staging_list, args->stream_list); + return 0; +} + +static int +wim_prepare_streams(WIMStruct *wim, off_t end_offset, + struct list_head *stream_list) +{ + struct lte_overwrite_prepare_args args = { + .wim = wim, + .end_offset = end_offset, + .stream_list = stream_list, + }; + int ret; + + for (int i = 0; i < wim->hdr.image_count; i++) { + ret = lte_overwrite_prepare(wim->image_metadata[i]->metadata_lte, + &args); + if (ret) + return ret; + } + return for_lookup_table_entry(wim->lookup_table, + lte_overwrite_prepare, &args); +} + +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 *hashed_lte; + struct hlist_node *tmp; + + pos = hash_u64(wim_resource_size(lte)) % tab->capacity; + lte->unique_size = 1; + hlist_for_each_entry(hashed_lte, tmp, &tab->array[pos], hash_list_2) { + if (wim_resource_size(hashed_lte) == wim_resource_size(lte)) { + lte->unique_size = 0; + hashed_lte->unique_size = 0; + break; + } + } + + hlist_add_head(<e->hash_list_2, &tab->array[pos]); + tab->num_entries++; + return 0; +} + -static int dentry_find_streams_to_write(struct dentry *dentry, - void *wim) +struct find_streams_ctx { + struct list_head stream_list; + struct stream_size_table stream_size_tab; +}; + +static int +inode_find_streams_to_write(struct wim_inode *inode, + struct wim_lookup_table *table, + struct list_head *stream_list, + struct stream_size_table *tab) { - 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); + struct wim_lookup_table_entry *lte; + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + lte = inode_stream_lte(inode, i, table); + if (lte) { + if (lte->out_refcnt == 0) { + if (lte->unhashed) + stream_size_table_insert(lte, tab); + list_add_tail(<e->write_streams_list, stream_list); + } + lte->out_refcnt += inode->i_nlink; + } } return 0; } -static int find_streams_to_write(WIMStruct *w) +static int +image_find_streams_to_write(WIMStruct *w) { - return for_dentry_in_tree(wim_root_dentry(w), - dentry_find_streams_to_write, w); + struct wim_image_metadata *imd; + struct find_streams_ctx *ctx; + struct wim_inode *inode; + struct wim_lookup_table_entry *lte; + + ctx = w->private; + imd = wim_get_current_image_metadata(w); + + image_for_each_unhashed_stream(lte, imd) { + lte->out_refcnt = 0; + wimlib_assert(lte->unhashed); + wimlib_assert(lte->my_ptr != NULL); + } + + /* Go through this image's inodes to find any streams that have not been + * found yet. */ + image_for_each_inode(inode, imd) { + inode_find_streams_to_write(inode, w->lookup_table, + &ctx->stream_list, + &ctx->stream_size_tab); + } + return 0; } -static int write_wim_streams(WIMStruct *w, int image, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +/* Given a WIM that from which one or all of the images is being written, build + * the list of unique streams ('struct wim_lookup_table_entry's) that must be + * written, plus any unhashed streams that need to be written but may be + * identical to other hashed or unhashed streams being written. These unhashed + * streams are checksummed while the streams are being written. To aid this + * process, the member @unique_size is set to 1 on streams that have a unique + * size and therefore must be written. + * + * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to + * indicate the number of times the stream is referenced in only the streams + * that are being written; this may still be adjusted later when unhashed + * streams are being resolved. + */ +static int +prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) { + int ret; + struct find_streams_ctx ctx; - 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, progress_func); + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + ret = init_stream_size_table(&ctx.stream_size_tab, 9001); + if (ret) + return ret; + for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, + &ctx.stream_size_tab); + INIT_LIST_HEAD(&ctx.stream_list); + wim->private = &ctx; + for_image(wim, image, image_find_streams_to_write); + destroy_stream_size_table(&ctx.stream_size_tab); + + INIT_LIST_HEAD(stream_list); + list_splice(&ctx.stream_list, stream_list); + return 0; +} + +/* Writes the streams for the specified @image in @wim to @wim->out_fp. + */ +static int +write_wim_streams(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) +{ + int ret; + struct list_head stream_list; + + ret = prepare_stream_list(wim, image, &stream_list); + if (ret) + return ret; + return write_stream_list(&stream_list, + wim->lookup_table, + wim->out_fp, + wimlib_get_compression_type(wim), + write_flags, + num_threads, + progress_func); } /* @@ -1359,8 +1584,9 @@ static int write_wim_streams(WIMStruct *w, int image, int write_flags, * fsync() the output file before closing it. * */ -int finish_write(WIMStruct *w, int image, int write_flags, - wimlib_progress_func_t progress_func) +int +finish_write(WIMStruct *w, int image, int write_flags, + wimlib_progress_func_t progress_func) { int ret; struct wim_header hdr; @@ -1373,7 +1599,7 @@ int finish_write(WIMStruct *w, int image, int write_flags, memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); + ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); if (ret != 0) goto out; } @@ -1391,6 +1617,8 @@ int finish_write(WIMStruct *w, int image, int write_flags, memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); if (fseeko(out, 0, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to beginning " + "of WIM being written"); ret = WIMLIB_ERR_WRITE; goto out; } @@ -1405,6 +1633,8 @@ int finish_write(WIMStruct *w, int image, int write_flags, } if (fseeko(out, 0, SEEK_END) != 0) { + ERROR_WITH_ERRNO("Failed to seek to end " + "of WIM being written"); ret = WIMLIB_ERR_WRITE; goto out; } @@ -1439,16 +1669,6 @@ int finish_write(WIMStruct *w, int image, int write_flags, * it should be a copy of the resource entry for the image that is * marked as bootable. This is not well documented... */ - if (hdr.boot_idx == 0 || !w->image_metadata - || (image != WIMLIB_ALL_IMAGES && image != hdr.boot_idx)) { - memset(&hdr.boot_metadata_res_entry, 0, - sizeof(struct resource_entry)); - } else { - memcpy(&hdr.boot_metadata_res_entry, - &w->image_metadata[ - hdr.boot_idx - 1].metadata_lte->output_resource_entry, - sizeof(struct resource_entry)); - } /* Set image count and boot index correctly for single image writes */ if (image != WIMLIB_ALL_IMAGES) { @@ -1459,13 +1679,25 @@ int finish_write(WIMStruct *w, int image, int write_flags, hdr.boot_idx = 0; } + if (hdr.boot_idx == 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]->metadata_lte->output_resource_entry, + sizeof(struct resource_entry)); + } + if (fseeko(out, 0, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to beginning of WIM " + "being written"); ret = WIMLIB_ERR_WRITE; goto out; } ret = write_header(&hdr, out); - if (ret != 0) + if (ret) goto out; if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { @@ -1486,27 +1718,74 @@ out: return ret; } -static void close_wim_writable(WIMStruct *w) +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +int +lock_wim(WIMStruct *w, FILE *fp) +{ + int ret = 0; + if (fp && !w->wim_locked) { + ret = flock(fileno(fp), LOCK_EX | LOCK_NB); + if (ret != 0) { + if (errno == EWOULDBLOCK) { + ERROR("`%"TS"' is already being modified or has been " + "mounted read-write\n" + " by another process!", w->filename); + ret = WIMLIB_ERR_ALREADY_LOCKED; + } else { + WARNING_WITH_ERRNO("Failed to lock `%"TS"'", + w->filename); + ret = 0; + } + } else { + w->wim_locked = 1; + } + } + return ret; +} +#endif + +static int +open_wim_writable(WIMStruct *w, const tchar *path, + bool trunc, bool also_readable) +{ + const tchar *mode; + if (trunc) + if (also_readable) + mode = T("w+b"); + else + mode = T("wb"); + else + mode = T("r+b"); + + wimlib_assert(w->out_fp == NULL); + w->out_fp = tfopen(path, mode); + if (w->out_fp) { + return 0; + } else { + ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path); + return WIMLIB_ERR_OPEN; + } +} + + +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)); + WARNING_WITH_ERRNO("Failed to close output WIM"); } 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 +begin_write(WIMStruct *w, const tchar *path, int write_flags) { int ret; - bool need_readable = false; - bool trunc = true; - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - need_readable = true; - - ret = open_wim_writable(w, path, trunc, need_readable); + ret = open_wim_writable(w, path, true, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); if (ret != 0) return ret; /* Write dummy header. It will be overwritten later. */ @@ -1514,13 +1793,14 @@ int begin_write(WIMStruct *w, const char *path, int write_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, - wimlib_progress_func_t progress_func) +WIMLIBAPI int +wimlib_write(WIMStruct *w, const tchar *path, + int image, int write_flags, unsigned num_threads, + wimlib_progress_func_t progress_func) { int ret; - if (!w || !path) + if (!path) return WIMLIB_ERR_INVALID_PARAM; write_flags &= WIMLIB_WRITE_MASK_PUBLIC; @@ -1535,19 +1815,19 @@ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, } ret = begin_write(w, path, write_flags); - if (ret != 0) + if (ret) goto out; ret = write_wim_streams(w, image, write_flags, num_threads, progress_func); - if (ret != 0) + if (ret) goto out; if (progress_func) progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); ret = for_image(w, image, write_metadata_resource); - if (ret != 0) + if (ret) goto out; if (progress_func) @@ -1556,40 +1836,17 @@ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, ret = finish_write(w, image, write_flags, progress_func); out: close_wim_writable(w); + DEBUG("wimlib_write(path=%"TS") = %d", path, ret); return ret; } -static int lte_overwrite_prepare(struct lookup_table_entry *lte, - void *ignore) +static bool +any_images_modified(WIMStruct *w) { - 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 - lte->out_refcnt = lte->refcnt; - return 0; + for (int i = 0; i < w->hdr.image_count; i++) + if (w->image_metadata[i]->modified) + return true; + return false; } /* @@ -1647,18 +1904,18 @@ static int find_new_streams(struct lookup_table_entry *lte, void *arg) * 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. + * small amount of space compared to the streams, however.) */ -static int overwrite_wim_inplace(WIMStruct *w, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func, - int modified_image_idx) +static int +overwrite_wim_inplace(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { int ret; struct list_head stream_list; off_t old_wim_end; - DEBUG("Overwriting `%s' in-place", w->filename); + DEBUG("Overwriting `%"TS"' in-place", w->filename); /* Make sure that the integrity table (if present) is after the XML * data, and that there are no stream resources, metadata resources, or @@ -1675,58 +1932,51 @@ static int overwrite_wim_inplace(WIMStruct *w, int write_flags, return WIMLIB_ERR_RESOURCE_ORDER; } - 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); - } 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) - return ret; - - if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) { + if (!w->deletion_occurred && !any_images_modified(w)) { /* If no images have been modified and no images have been * deleted, a new lookup table does not need to be written. */ - 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; } - INIT_LIST_HEAD(&stream_list); - for_lookup_table_entry(w->lookup_table, find_new_streams, - &stream_list); + ret = wim_prepare_streams(w, old_wim_end, &stream_list); + if (ret != 0) + return ret; ret = open_wim_writable(w, w->filename, false, (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); if (ret != 0) return ret; + ret = lock_wim(w, w->out_fp); + if (ret != 0) { + fclose(w->out_fp); + w->out_fp = NULL; + return ret; + } + if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); + fclose(w->out_fp); + w->out_fp = NULL; + w->wim_locked = 0; 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, + ret = write_stream_list(&stream_list, + w->lookup_table, + w->out_fp, wimlib_get_compression_type(w), write_flags, num_threads, progress_func); @@ -1736,61 +1986,72 @@ static int overwrite_wim_inplace(WIMStruct *w, int write_flags, 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; + for (int i = 0; i < w->hdr.image_count; i++) { + if (w->image_metadata[i]->modified) { + select_wim_image(w, i + 1); + ret = write_metadata_resource(w); + if (ret) + goto out_ftruncate; + } } write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, progress_func); out_ftruncate: close_wim_writable(w); - if (ret != 0) { - WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", w->filename, old_wim_end); - truncate(w->filename, old_wim_end); + /* Return value of truncate() is ignored because this is already + * an error path. */ + (void)ttruncate(w->filename, old_wim_end); } + w->wim_locked = 0; return ret; } -static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +static int +overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { size_t wim_name_len; int ret; - DEBUG("Overwriting `%s' via a temporary file", w->filename); + DEBUG("Overwriting `%"TS"' 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); + wim_name_len = tstrlen(w->filename); + tchar tmpfile[wim_name_len + 10]; + tmemcpy(tmpfile, w->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; + tmpfile[wim_name_len + 9] = T('\0'); ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, write_flags | WIMLIB_WRITE_FLAG_FSYNC, num_threads, progress_func); if (ret != 0) { - ERROR("Failed to write the WIM file `%s'", tmpfile); + ERROR("Failed to write the WIM file `%"TS"'", tmpfile); goto err; } - /* Close the original WIM file that was opened for reading. */ + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); + +#ifdef __WIN32__ + /* Windows won't let you delete open files unless FILE_SHARE_DELETE was + * specified to CreateFile(). The WIM was opened with fopen(), which + * didn't provided this flag to CreateFile, so the handle must be closed + * before executing the rename(). */ if (w->fp != NULL) { fclose(w->fp); w->fp = NULL; } - - DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); +#endif /* Rename the new file to the old file .*/ - if (rename(tmpfile, w->filename) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + if (trename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", tmpfile, w->filename); ret = WIMLIB_ERR_RENAME; goto err; @@ -1803,31 +2064,37 @@ static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } + /* Close the original WIM file that was opened for reading. */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } + /* Re-open the WIM read-only. */ - w->fp = fopen(w->filename, "rb"); + w->fp = tfopen(w->filename, T("rb")); if (w->fp == NULL) { ret = WIMLIB_ERR_REOPEN; - WARNING("Failed to re-open `%s' read-only: %s", - w->filename, strerror(errno)); + WARNING_WITH_ERRNO("Failed to re-open `%"TS"' read-only", + w->filename); + FREE(w->filename); + w->filename = NULL; } return ret; err: /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - WARNING("Failed to remove `%s': %s", tmpfile, strerror(errno)); + if (tunlink(tmpfile) != 0) + WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); 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, - wimlib_progress_func_t progress_func) +WIMLIBAPI int +wimlib_overwrite(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { - if (!w) - return WIMLIB_ERR_INVALID_PARAM; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; if (!w->filename) @@ -1841,18 +2108,13 @@ WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags, 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, - progress_func, - modified_image_idx); - } + int ret; + ret = overwrite_wim_inplace(w, write_flags, num_threads, + progress_func); + if (ret == WIMLIB_ERR_RESOURCE_ORDER) + WARNING("Falling back to re-building entire WIM"); + else + return ret; } return overwrite_wim_via_tmpfile(w, write_flags, num_threads, progress_func);