X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=86730ff7fec7a1158b6ed8bed305ec784b9375aa;hp=061f0e7b5443c8b1977e19ad079e0d1199de505b;hb=11b249a3d62d88ca4c5123570f21e754715628e5;hpb=c79590300def03992af2062d97afebade0fc4fdf diff --git a/src/write.c b/src/write.c index 061f0e7b..86730ff7 100644 --- a/src/write.c +++ b/src/write.c @@ -6,7 +6,6 @@ */ /* - * Copyright (C) 2010 Carl Thijssen * Copyright (C) 2012 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. @@ -35,7 +34,7 @@ #include "list.h" #include "wimlib_internal.h" -#include "io.h" +#include "buffer_io.h" #include "dentry.h" #include "lookup_table.h" #include "xml.h" @@ -62,24 +61,15 @@ #include #endif - -static int do_fflush(FILE *fp) +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 " @@ -108,7 +98,7 @@ 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) @@ -218,7 +208,7 @@ 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 @@ -260,7 +250,7 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, /* 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 +static int prepare_resource_for_read(struct wim_lookup_table_entry *lte #ifdef WITH_NTFS_3G , ntfs_inode **ni_ret @@ -309,7 +299,7 @@ static int prepare_resource_for_read(struct lookup_table_entry *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 +static void end_wim_resource_read(struct wim_lookup_table_entry *lte #ifdef WITH_NTFS_3G , ntfs_inode *ni #endif @@ -332,6 +322,27 @@ static void end_wim_resource_read(struct lookup_table_entry *lte #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) +{ + 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; + } + ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE, + out_res_entry, 0); + if (ret != 0) + return ret; + + return fflush_and_ftruncate(out_fp, + file_offset + wim_resource_size(lte)); +} + /* * Writes a WIM resource to a FILE * opened for writing. The resource may be * written uncompressed or compressed depending on the @out_ctype parameter. @@ -353,7 +364,7 @@ static void end_wim_resource_read(struct lookup_table_entry *lte * * Returns 0 on success; nonzero on failure. */ -int write_wim_resource(struct lookup_table_entry *lte, +int write_wim_resource(struct wim_lookup_table_entry *lte, FILE *out_fp, int out_ctype, struct resource_entry *out_res_entry, int flags) @@ -507,18 +518,10 @@ int write_wim_resource(struct lookup_table_entry *lte, { /* 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); + ret = write_uncompressed_resource_and_truncate(lte, + out_fp, + file_offset, + out_res_entry); if (ret != 0) goto out_fclose; } else { @@ -545,6 +548,8 @@ out: } #ifdef ENABLE_MULTITHREADED_COMPRESSION + +/* Blocking shared queue (solves the producer-consumer problem) */ struct shared_queue { unsigned size; unsigned front; @@ -558,6 +563,7 @@ struct shared_queue { 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; @@ -620,7 +626,7 @@ 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]; @@ -649,6 +655,11 @@ static void compress_chunks(struct message *msg, compress_func_t compress) } } +/* 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; @@ -675,7 +686,7 @@ static int do_write_stream_list(struct list_head *my_resources, int write_resource_flags) { int ret; - struct lookup_table_entry *lte, *tmp; + struct wim_lookup_table_entry *lte, *tmp; list_for_each_entry_safe(lte, tmp, my_resources, staging_list) { ret = write_wim_resource(lte, @@ -758,19 +769,25 @@ static int main_writer_thread_proc(struct list_head *stream_list, int out_ctype, struct shared_queue *res_to_compress_queue, struct shared_queue *compressed_res_queue, - size_t queue_size, + 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 @@ -790,55 +807,42 @@ 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 lz.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; } @@ -853,7 +857,95 @@ 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 `%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) { + // 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 @@ -880,13 +972,11 @@ 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); @@ -908,81 +998,11 @@ 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; } @@ -990,26 +1010,20 @@ 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) { @@ -1041,8 +1055,8 @@ 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) { @@ -1054,6 +1068,32 @@ static int main_writer_thread_proc(struct list_head *stream_list, if (ret != 0) goto out; + 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.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; + } + progress->write_streams.completed_bytes += wim_resource_size(cur_lte); progress->write_streams.completed_streams++; @@ -1063,34 +1103,18 @@ static int main_writer_thread_proc(struct list_head *stream_list, progress); } - 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 { + 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 @@ -1112,38 +1136,50 @@ 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; } @@ -1193,6 +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); @@ -1230,6 +1270,7 @@ out_join: } } 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); @@ -1256,24 +1297,21 @@ static int write_stream_list(struct list_head *stream_list, FILE *out_fp, 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) { 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; @@ -1282,54 +1320,96 @@ static int write_stream_list(struct list_head *stream_list, FILE *out_fp, progress.write_streams.num_threads = num_threads; progress.write_streams.compression_type = out_ctype; - if (num_streams == 0) { - ret = 0; - goto out; - } - #ifdef ENABLE_MULTITHREADED_COMPRESSION - if (compression_needed && total_bytes >= 1000000 && num_threads != 1) { - ret = write_stream_list_parallel(stream_list, - out_fp, - out_ctype, - write_flags, - num_threads, - progress_func, - &progress); - } + if (total_compression_bytes >= 1000000 && num_threads != 1) + return write_stream_list_parallel(stream_list, + out_fp, + out_ctype, + write_flags, + num_threads, + progress_func, + &progress); else #endif + return write_stream_list_serial(stream_list, + out_fp, + out_ctype, + write_flags, + progress_func, + &progress); +} + +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) { - ret = write_stream_list_serial(stream_list, - out_fp, - out_ctype, - write_flags, - progress_func, - &progress); + ERROR("The following resource is after the XML data:"); + print_lookup_table_entry(lte); + return WIMLIB_ERR_RESOURCE_ORDER; } -out: - return ret; + + 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)) { + wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT); + if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) + list_add(<e->staging_list, args->stream_list); + } + return 0; } +static int wim_find_new_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, + }; + + return for_lookup_table_entry(wim->lookup_table, + lte_overwrite_prepare, &args); +} -static int dentry_find_streams_to_write(struct dentry *dentry, - void *wim) +static int inode_find_streams_to_write(struct wim_inode *inode, + struct wim_lookup_table *table, + struct list_head *stream_list) { - 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) + list_add_tail(<e->staging_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_inode *inode; + struct hlist_node *cur; + struct hlist_head *inode_list; + + inode_list = &wim_get_current_image_metadata(w)->inode_list; + hlist_for_each_entry(inode, cur, inode_list, i_hlist) { + inode_find_streams_to_write(inode, w->lookup_table, + (struct list_head*)w->private); + } + return 0; } static int write_wim_streams(WIMStruct *w, int image, int write_flags, @@ -1340,7 +1420,7 @@ static int write_wim_streams(WIMStruct *w, int image, int write_flags, 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); + for_image(w, image, 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); @@ -1407,6 +1487,8 @@ 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; } @@ -1421,6 +1503,8 @@ 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; } @@ -1455,16 +1539,6 @@ 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) { @@ -1475,7 +1549,19 @@ 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; } @@ -1503,22 +1589,24 @@ out: } #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) -int lock_wim(FILE *fp, const char *path) +int lock_wim(WIMStruct *w, FILE *fp) { int ret = 0; - if (fp) { + if (fp && !w->wim_locked) { ret = flock(fileno(fp), LOCK_EX | LOCK_NB); if (ret != 0) { if (errno == EWOULDBLOCK) { ERROR("`%s' is already being modified or has been " "mounted read-write\n" - " by another process!", path); + " by another process!", w->filename); ret = WIMLIB_ERR_ALREADY_LOCKED; } else { WARNING("Failed to lock `%s': %s", - path, strerror(errno)); + w->filename, strerror(errno)); ret = 0; } + } else { + w->wim_locked = 1; } } return ret; @@ -1578,7 +1666,7 @@ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, { int ret; - if (!w || !path) + if (!path) return WIMLIB_ERR_INVALID_PARAM; write_flags &= WIMLIB_WRITE_MASK_PUBLIC; @@ -1617,37 +1705,12 @@ out: 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; } /* @@ -1705,16 +1768,16 @@ 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) + wimlib_progress_func_t progress_func) { int ret; struct list_head stream_list; off_t old_wim_end; + bool found_modified_image; DEBUG("Overwriting `%s' in-place", w->filename); @@ -1733,27 +1796,13 @@ 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); - 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. */ old_wim_end = w->hdr.lookup_table_res_entry.offset + @@ -1761,17 +1810,17 @@ static int overwrite_wim_inplace(WIMStruct *w, int write_flags, 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_find_new_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->out_fp, w->filename); + ret = lock_wim(w, w->out_fp); if (ret != 0) { fclose(w->out_fp); w->out_fp = NULL; @@ -1780,8 +1829,10 @@ static int overwrite_wim_inplace(WIMStruct *w, int write_flags, if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - ret = WIMLIB_ERR_WRITE; - goto out_ftruncate; + fclose(w->out_fp); + w->out_fp = NULL; + w->wim_locked = 0; + return WIMLIB_ERR_WRITE; } if (!list_empty(&stream_list)) { @@ -1797,22 +1848,30 @@ 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; + found_modified_image = false; + for (int i = 0; i < w->hdr.image_count; i++) { + if (!found_modified_image) + found_modified_image = w->image_metadata[i].modified; + if (found_modified_image) { + 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, WIMLIB_ALL_IMAGES, write_flags, progress_func); out_ftruncate: close_wim_writable(w); - if (ret != 0) { + if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { WARNING("Truncating `%s' 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)truncate(w->filename, old_wim_end); } + w->wim_locked = 0; return ret; } @@ -1888,9 +1947,6 @@ 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) @@ -1904,18 +1960,13 @@ 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);