X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fwrite.c;h=42b9fcfff84e38f682d05b5406dcfa0b58d7118a;hb=47898041237677686c46ccac6dd88485a63ddf93;hp=96ada4929be1316e353091aa547e64f32b14960a;hpb=61db93f82eca3fe9f7676355c709c58cc425a6ad;p=wimlib diff --git a/src/write.c b/src/write.c index 96ada492..42b9fcff 100644 --- a/src/write.c +++ b/src/write.c @@ -53,17 +53,16 @@ # include #endif -#include -#include #include +#include +#include +#include +#include #ifdef HAVE_ALLOCA_H # include -#else -# include #endif -#include #ifndef __WIN32__ # include /* for `struct iovec' */ @@ -683,7 +682,6 @@ struct message { u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; struct iovec out_chunks[MAX_CHUNKS_PER_MSG]; - size_t total_out_bytes; unsigned num_chunks; struct list_head list; bool complete; @@ -693,7 +691,6 @@ struct message { static void compress_chunks(struct message *msg, compress_func_t compress) { - msg->total_out_bytes = 0; for (unsigned i = 0; i < msg->num_chunks; i++) { unsigned len = compress(msg->uncompressed_chunks[i], msg->uncompressed_chunk_sizes[i], @@ -711,7 +708,6 @@ compress_chunks(struct message *msg, compress_func_t compress) } msg->out_chunks[i].iov_base = out_chunk; msg->out_chunks[i].iov_len = out_len; - msg->total_out_bytes += out_len; } } @@ -739,32 +735,52 @@ compressor_thread_proc(void *arg) } #endif /* ENABLE_MULTITHREADED_COMPRESSION */ +struct write_streams_progress_data { + wimlib_progress_func_t progress_func; + union wimlib_progress_info progress; + uint64_t next_progress; + WIMStruct *prev_wim_part; +}; + static void -do_write_streams_progress(union wimlib_progress_info *progress, - wimlib_progress_func_t progress_func, - uint64_t size_added, +do_write_streams_progress(struct write_streams_progress_data *progress_data, + struct wim_lookup_table_entry *lte, bool stream_discarded) { + union wimlib_progress_info *progress = &progress_data->progress; + bool new_wim_part; + if (stream_discarded) { - progress->write_streams.total_bytes -= size_added; - if (progress->write_streams._private != ~(uint64_t)0 && - progress->write_streams._private > progress->write_streams.total_bytes) + progress->write_streams.total_bytes -= wim_resource_size(lte); + if (progress_data->next_progress != ~(uint64_t)0 && + progress_data->next_progress > progress->write_streams.total_bytes) { - progress->write_streams._private = progress->write_streams.total_bytes; + progress_data->next_progress = progress->write_streams.total_bytes; } } else { - progress->write_streams.completed_bytes += size_added; + progress->write_streams.completed_bytes += wim_resource_size(lte); + } + new_wim_part = false; + if (lte->resource_location == RESOURCE_IN_WIM && + lte->wim != progress_data->prev_wim_part) + { + if (progress_data->prev_wim_part) { + new_wim_part = true; + progress->write_streams.completed_parts++; + } + progress_data->prev_wim_part = lte->wim; } progress->write_streams.completed_streams++; - if (progress_func && - progress->write_streams.completed_bytes >= progress->write_streams._private) + if (progress_data->progress_func + && (progress->write_streams.completed_bytes >= progress_data->next_progress + || new_wim_part)) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - if (progress->write_streams._private == progress->write_streams.total_bytes) { - progress->write_streams._private = ~(uint64_t)0; + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress_data->next_progress == progress->write_streams.total_bytes) { + progress_data->next_progress = ~(uint64_t)0; } else { - progress->write_streams._private = + progress_data->next_progress = min(progress->write_streams.total_bytes, progress->write_streams.completed_bytes + progress->write_streams.total_bytes / 100); @@ -787,6 +803,7 @@ serial_write_stream(struct wim_lookup_table_entry *lte, void *_ctx) ctx->write_resource_flags); } + /* Write a list of streams, taking into account that some streams may be * duplicates that are checksummed and discarded on the fly, and also delegating * the actual writing of a stream to a function @write_stream_cb, which is @@ -796,8 +813,7 @@ do_write_stream_list(struct list_head *stream_list, struct wim_lookup_table *lookup_table, int (*write_stream_cb)(struct wim_lookup_table_entry *, void *), void *write_stream_ctx, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { int ret = 0; struct wim_lookup_table_entry *lte; @@ -815,19 +831,29 @@ do_write_stream_list(struct list_head *stream_list, * stream in the WIM we are writing. The stream must be * checksummed to know if we need to write it or not. */ struct wim_lookup_table_entry *tmp; - u32 orig_refcnt = lte->out_refcnt; + u32 orig_out_refcnt = lte->out_refcnt; ret = hash_unhashed_stream(lte, lookup_table, &tmp); if (ret) break; if (tmp != lte) { + /* We found a duplicate stream. 'lte' was + * freed, so replace it with the duplicate. */ lte = tmp; - /* We found a duplicate stream. */ - if (orig_refcnt != tmp->out_refcnt) { - /* We have already written, or are going - * to write, the duplicate stream. So - * just skip to the next stream. */ - DEBUG("Discarding duplicate stream of length %"PRIu64, + + /* 'out_refcnt' was transferred to the + * duplicate, and we can detect if the duplicate + * stream was already referenced for writing by + * checking if its 'out_refcnt' is higher than + * that of the original stream. In such cases, + * the current stream can be discarded. We can + * also discard the current stream if it was + * previously marked as filtered (e.g. already + * present in the WIM being written). */ + if (lte->out_refcnt > orig_out_refcnt || + lte->filtered) { + DEBUG("Discarding duplicate stream of " + "length %"PRIu64, wim_resource_size(lte)); lte->no_progress = 0; stream_discarded = true; @@ -860,10 +886,8 @@ do_write_stream_list(struct list_head *stream_list, } skip_to_progress: if (!lte->no_progress) { - do_write_streams_progress(progress, - progress_func, - wim_resource_size(lte), - stream_discarded); + do_write_streams_progress(progress_data, + lte, stream_discarded); } } return ret; @@ -875,8 +899,7 @@ do_write_stream_list_serial(struct list_head *stream_list, struct filedes *out_fd, int out_ctype, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { struct serial_write_stream_ctx ctx = { .out_fd = out_fd, @@ -887,8 +910,7 @@ do_write_stream_list_serial(struct list_head *stream_list, lookup_table, serial_write_stream, &ctx, - progress_func, - progress); + progress_data); } static inline int @@ -909,21 +931,22 @@ write_stream_list_serial(struct list_head *stream_list, struct filedes *out_fd, int out_ctype, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { + union wimlib_progress_info *progress = &progress_data->progress; DEBUG("Writing stream list of size %"PRIu64" (serial version)", progress->write_streams.total_streams); progress->write_streams.num_threads = 1; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } return do_write_stream_list_serial(stream_list, lookup_table, out_fd, out_ctype, write_resource_flags, - progress_func, - progress); + progress_data); } #ifdef ENABLE_MULTITHREADED_COMPRESSION @@ -935,7 +958,6 @@ write_wim_chunks(struct message *msg, struct filedes *out_fd, struct iovec *vecs; struct pwm_chunk_hdr *chunk_hdrs; unsigned nvecs; - size_t nbytes; int ret; for (unsigned i = 0; i < msg->num_chunks; i++) @@ -944,7 +966,6 @@ write_wim_chunks(struct message *msg, struct filedes *out_fd, if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { nvecs = msg->num_chunks; vecs = msg->out_chunks; - nbytes = msg->total_out_bytes; } else { /* Special case: If writing a compressed resource to a pipable * WIM, prefix each compressed chunk with a header that gives @@ -960,7 +981,6 @@ write_wim_chunks(struct message *msg, struct filedes *out_fd, vecs[i * 2 + 1].iov_base = msg->out_chunks[i].iov_base; vecs[i * 2 + 1].iov_len = msg->out_chunks[i].iov_len; } - nbytes = msg->total_out_bytes + msg->num_chunks * sizeof(chunk_hdrs[0]); } ret = full_writev(out_fd, vecs, nvecs); if (ret) @@ -978,8 +998,7 @@ struct main_writer_thread_ctx { struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; size_t num_messages; - wimlib_progress_func_t progress_func; - union wimlib_progress_info *progress; + struct write_streams_progress_data *progress_data; struct list_head available_msgs; struct list_head outstanding_streams; @@ -1221,10 +1240,8 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) cur_lte->output_resource_entry.flags); } - do_write_streams_progress(ctx->progress, - ctx->progress_func, - wim_resource_size(cur_lte), - false); + do_write_streams_progress(ctx->progress_data, + cur_lte, false); /* Since we just finished writing a stream, write any * streams that have been added to the serial_streams @@ -1238,8 +1255,7 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) ctx->out_fd, ctx->out_ctype, ctx->write_resource_flags, - ctx->progress_func, - ctx->progress); + ctx->progress_data); if (ret) return ret; } @@ -1337,8 +1353,7 @@ main_writer_thread_finish(void *_ctx) ctx->out_fd, ctx->out_ctype, ctx->write_resource_flags, - ctx->progress_func, - ctx->progress); + ctx->progress_data); } static int @@ -1433,14 +1448,14 @@ write_stream_list_parallel(struct list_head *stream_list, struct filedes *out_fd, int out_ctype, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress, + struct write_streams_progress_data *progress_data, unsigned num_threads) { int ret; struct shared_queue res_to_compress_queue; struct shared_queue compressed_res_queue; pthread_t *compressor_threads = NULL; + union wimlib_progress_info *progress = &progress_data->progress; if (num_threads == 0) { long nthreads = get_default_num_threads(); @@ -1498,8 +1513,10 @@ write_stream_list_parallel(struct list_head *stream_list, } } - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } struct main_writer_thread_ctx ctx; ctx.stream_list = stream_list; @@ -1510,14 +1527,13 @@ write_stream_list_parallel(struct list_head *stream_list, ctx.compressed_res_queue = &compressed_res_queue; ctx.num_messages = queue_size; ctx.write_resource_flags = write_resource_flags; - ctx.progress_func = progress_func; - ctx.progress = progress; + ctx.progress_data = progress_data; ret = main_writer_thread_init_ctx(&ctx); if (ret) goto out_join; ret = do_write_stream_list(stream_list, lookup_table, main_thread_process_next_stream, - &ctx, progress_func, progress); + &ctx, progress_data); if (ret) goto out_destroy_ctx; @@ -1556,8 +1572,7 @@ out_serial_quiet: out_fd, out_ctype, write_resource_flags, - progress_func, - progress); + progress_data); } #endif @@ -1576,16 +1591,25 @@ write_stream_list(struct list_head *stream_list, size_t num_streams = 0; u64 total_bytes = 0; u64 total_compression_bytes = 0; - union wimlib_progress_info progress; + struct write_streams_progress_data progress_data; int ret; int write_resource_flags; + unsigned total_parts = 0; + WIMStruct *prev_wim_part = NULL; - if (list_empty(stream_list)) + if (list_empty(stream_list)) { + DEBUG("No streams to write."); return 0; + } write_resource_flags = write_flags_to_resource_flags(write_flags); - DEBUG("write_resource_flags=0x%08x", write_resource_flags); + DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", + out_fd->offset, write_resource_flags); + + sort_stream_list_by_sequential_order(stream_list, + offsetof(struct wim_lookup_table_entry, + write_streams_list)); /* Calculate the total size of the streams to be written. Note: this * will be the uncompressed size, as we may not know the compressed size @@ -1600,14 +1624,28 @@ write_stream_list(struct list_head *stream_list, { total_compression_bytes += wim_resource_size(lte); } + if (lte->resource_location == RESOURCE_IN_WIM) { + if (prev_wim_part != lte->wim) { + prev_wim_part = lte->wim; + total_parts++; + } + } } - progress.write_streams.total_bytes = total_bytes; - progress.write_streams.total_streams = num_streams; - progress.write_streams.completed_bytes = 0; - progress.write_streams.completed_streams = 0; - progress.write_streams.num_threads = num_threads; - progress.write_streams.compression_type = out_ctype; - progress.write_streams._private = 0; + + memset(&progress_data, 0, sizeof(progress_data)); + progress_data.progress_func = progress_func; + + progress_data.progress.write_streams.total_bytes = total_bytes; + progress_data.progress.write_streams.total_streams = num_streams; + progress_data.progress.write_streams.completed_bytes = 0; + progress_data.progress.write_streams.completed_streams = 0; + progress_data.progress.write_streams.num_threads = num_threads; + progress_data.progress.write_streams.compression_type = out_ctype; + progress_data.progress.write_streams.total_parts = total_parts; + progress_data.progress.write_streams.completed_parts = 0; + + progress_data.next_progress = 0; + progress_data.prev_wim_part = NULL; #ifdef ENABLE_MULTITHREADED_COMPRESSION if (total_compression_bytes >= 2000000 && num_threads != 1) @@ -1616,8 +1654,7 @@ write_stream_list(struct list_head *stream_list, out_fd, out_ctype, write_resource_flags, - progress_func, - &progress, + &progress_data, num_threads); else #endif @@ -1626,8 +1663,7 @@ write_stream_list(struct list_head *stream_list, out_fd, out_ctype, write_resource_flags, - progress_func, - &progress); + &progress_data); if (ret == 0) DEBUG("Successfully wrote stream list."); else @@ -1681,136 +1717,52 @@ stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) return 0; } - -struct lte_overwrite_prepare_args { +struct find_streams_ctx { WIMStruct *wim; - off_t end_offset; + int write_flags; struct list_head stream_list; struct stream_size_table stream_size_tab; }; -/* First phase of preparing streams for an in-place overwrite. This is called - * on all streams, both hashed and unhashed, except the metadata resources. */ -static int -lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *_args) +static void +lte_reference_for_logical_write(struct wim_lookup_table_entry *lte, + struct find_streams_ctx *ctx, + unsigned nref) { - struct lte_overwrite_prepare_args *args = _args; - - wimlib_assert(!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)); - if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) - list_add_tail(<e->write_streams_list, &args->stream_list); - lte->out_refcnt = lte->refcnt; - stream_size_table_insert(lte, &args->stream_size_tab); - return 0; + if (lte->out_refcnt == 0) { + stream_size_table_insert(lte, &ctx->stream_size_tab); + list_add_tail(<e->write_streams_list, &ctx->stream_list); + } + lte->out_refcnt += nref; } -/* Second phase of preparing streams for an in-place overwrite. This is called - * on existing metadata resources and hashed streams, but not unhashed streams. - * - * NOTE: lte->output_resource_entry is in union with lte->hash_list_2, so - * lte_overwrite_prepare_2() must be called after lte_overwrite_prepare(), as - * the latter uses lte->hash_list_2, while the former expects to set - * lte->output_resource_entry. */ static int -lte_overwrite_prepare_2(struct wim_lookup_table_entry *lte, void *_args) +do_lte_full_reference_for_logical_write(struct wim_lookup_table_entry *lte, + void *_ctx) { - struct lte_overwrite_prepare_args *args = _args; - - if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == args->wim) { - /* We can't do an in place overwrite on the WIM if there are - * streams after the XML data. */ - if (lte->resource_entry.offset + - lte->resource_entry.size > args->end_offset) - { - if (wimlib_print_errors) { - ERROR("The following resource is after the XML data:"); - print_lookup_table_entry(lte, stderr); - } - return WIMLIB_ERR_RESOURCE_ORDER; - } - copy_resource_entry(<e->output_resource_entry, - <e->resource_entry); - } + struct find_streams_ctx *ctx = _ctx; + lte->out_refcnt = 0; + lte_reference_for_logical_write(lte, ctx, + (lte->refcnt ? lte->refcnt : 1)); return 0; } -/* Given a WIM that we are going to overwrite in place with zero or more - * additional streams added, construct a list the list of new unique streams - * ('struct wim_lookup_table_entry's) that must be written, plus any unhashed - * streams that need to be added but may be identical to other hashed or - * unhashed streams. These unhashed streams are checksummed while the streams - * are being written. To aid this process, the member @unique_size is set to 1 - * on streams that have a unique size and therefore must be written. - * - * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to - * indicate the number of times the stream is referenced in only the streams - * that are being written; this may still be adjusted later when unhashed - * streams are being resolved. - */ static int -prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset, - struct list_head *stream_list) -{ - int ret; - struct lte_overwrite_prepare_args args; - unsigned i; - - args.wim = wim; - args.end_offset = end_offset; - ret = init_stream_size_table(&args.stream_size_tab, - wim->lookup_table->capacity); - if (ret) - return ret; - - INIT_LIST_HEAD(&args.stream_list); - for (i = 0; i < wim->hdr.image_count; i++) { - struct wim_image_metadata *imd; - struct wim_lookup_table_entry *lte; - - imd = wim->image_metadata[i]; - image_for_each_unhashed_stream(lte, imd) - lte_overwrite_prepare(lte, &args); - } - for_lookup_table_entry(wim->lookup_table, lte_overwrite_prepare, &args); - list_transfer(&args.stream_list, stream_list); - - for (i = 0; i < wim->hdr.image_count; i++) { - ret = lte_overwrite_prepare_2(wim->image_metadata[i]->metadata_lte, - &args); - if (ret) - goto out_destroy_stream_size_table; - } - ret = for_lookup_table_entry(wim->lookup_table, - lte_overwrite_prepare_2, &args); -out_destroy_stream_size_table: - destroy_stream_size_table(&args.stream_size_tab); - return ret; -} - - -struct find_streams_ctx { - struct list_head stream_list; - struct stream_size_table stream_size_tab; -}; - -static void inode_find_streams_to_write(struct wim_inode *inode, struct wim_lookup_table *table, - struct list_head *stream_list, - struct stream_size_table *tab) + struct find_streams_ctx *ctx) { struct wim_lookup_table_entry *lte; - for (unsigned i = 0; i <= inode->i_num_ads; i++) { + unsigned i; + + for (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; - } + if (lte) + lte_reference_for_logical_write(lte, ctx, inode->i_nlink); + else if (!is_zero_hash(inode_stream_hash(inode, i))) + return WIMLIB_ERR_RESOURCE_NOT_FOUND; } + return 0; } static int @@ -1820,6 +1772,7 @@ image_find_streams_to_write(WIMStruct *wim) struct wim_image_metadata *imd; struct wim_inode *inode; struct wim_lookup_table_entry *lte; + int ret; ctx = wim->private; imd = wim_get_current_image_metadata(wim); @@ -1830,55 +1783,174 @@ image_find_streams_to_write(WIMStruct *wim) /* 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, wim->lookup_table, - &ctx->stream_list, - &ctx->stream_size_tab); + ret = inode_find_streams_to_write(inode, wim->lookup_table, ctx); + if (ret) + return ret; } return 0; } -/* 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. +/* + * Build a list of streams (via `struct wim_lookup_table_entry's) included in + * the "logical write" of the WIM, meaning all streams that are referenced at + * least once by dentries in the the image(s) being written. 'out_refcnt' on + * each stream being included in the logical write is set to the number of + * references from dentries in the image(s). Furthermore, 'unique_size' on each + * stream being included in the logical write is set to indicate whether that + * stream has a unique size relative to the streams being included in the + * logical write. Still furthermore, 'part_number' on each stream being + * included in the logical write is set to the part number given in the + * in-memory header of @p wim. * - * 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. + * This is considered a "logical write" because it does not take into account + * filtering out streams already present in the WIM (in the case of an in place + * overwrite) or present in other WIMs (in case of creating delta WIM). */ static int -prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) +prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, + struct find_streams_ctx *ctx) { int ret; - struct find_streams_ctx ctx; + struct wim_lookup_table_entry *lte; + + if (streams_ok && (image == WIMLIB_ALL_IMAGES || + (image == 1 && wim->hdr.image_count == 1))) + { + /* Fast case: Assume that all streams are being written and + * that the reference counts are correct. */ + struct wim_lookup_table_entry *lte; + struct wim_image_metadata *imd; + unsigned i; + + for_lookup_table_entry(wim->lookup_table, + do_lte_full_reference_for_logical_write, ctx); + for (i = 0; i < wim->hdr.image_count; i++) { + imd = wim->image_metadata[i]; + image_for_each_unhashed_stream(lte, imd) + do_lte_full_reference_for_logical_write(lte, ctx); + } + } else { + /* Slow case: Walk through the images being written and + * determine the streams referenced. */ + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + wim->private = ctx; + ret = for_image(wim, image, image_find_streams_to_write); + if (ret) + return ret; + } + + list_for_each_entry(lte, &ctx->stream_list, write_streams_list) + lte->part_number = wim->hdr.part_number; + return 0; +} + +static int +process_filtered_stream(struct wim_lookup_table_entry *lte, void *_ctx) +{ + struct find_streams_ctx *ctx = _ctx; + u16 filtered = 0; + + /* Calculate and set lte->filtered. */ + if (lte->resource_location == RESOURCE_IN_WIM) { + if (lte->wim == ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE)) + filtered |= FILTERED_SAME_WIM; + if (lte->wim != ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)) + filtered |= FILTERED_EXTERNAL_WIM; + } + lte->filtered = filtered; + + /* Filtered streams get inserted into the stream size table too, unless + * they already were. This is because streams that are checksummed + * on-the-fly during the write should not be written if they are + * duplicates of filtered stream. */ + if (lte->filtered && lte->out_refcnt == 0) + stream_size_table_insert(lte, &ctx->stream_size_tab); + return 0; +} + +static int +mark_stream_not_filtered(struct wim_lookup_table_entry *lte, void *_ignore) +{ + lte->filtered = 0; + return 0; +} + +/* Given the list of streams to include in a logical write of a WIM, handle + * filtering out streams already present in the WIM or already present in + * external WIMs, depending on the write flags provided. */ +static void +handle_stream_filtering(struct find_streams_ctx *ctx) +{ + struct wim_lookup_table_entry *lte, *tmp; + + if (!(ctx->write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE | + WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS))) + { + for_lookup_table_entry(ctx->wim->lookup_table, + mark_stream_not_filtered, ctx); + return; + } - DEBUG("Preparing list of streams to write for image %d.", image); + for_lookup_table_entry(ctx->wim->lookup_table, + process_filtered_stream, ctx); - for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + /* Streams in logical write list that were filtered can be removed. */ + list_for_each_entry_safe(lte, tmp, &ctx->stream_list, + write_streams_list) + if (lte->filtered) + list_del(<e->write_streams_list); +} + +/* Prepares list of streams to write for the specified WIM image(s). This wraps + * around prepare_logical_stream_list() to handle filtering out streams already + * present in the WIM or already present in external WIMs, depending on the + * write flags provided. + * + * Note: some additional data is stored in each `struct wim_lookup_table_entry': + * + * - 'out_refcnt' is set to the number of references found for the logical write. + * This will be nonzero on all streams in the list returned by this function, + * but will also be nonzero on streams not in the list that were included in + * the logical write list, but filtered out from the returned list. + * - 'filtered' is set to nonzero if the stream was filtered. Filtered streams + * are not included in the list of streams returned by this function. + * - 'unique_size' is set if the stream has a unique size among all streams in + * the logical write plus any filtered streams in the entire WIM that could + * potentially turn out to have the same checksum as a yet-to-be-checksummed + * stream being written. + */ +static int +prepare_stream_list(WIMStruct *wim, int image, int write_flags, + struct list_head *stream_list) +{ + int ret; + bool streams_ok; + struct find_streams_ctx ctx; + + INIT_LIST_HEAD(&ctx.stream_list); ret = init_stream_size_table(&ctx.stream_size_tab, wim->lookup_table->capacity); if (ret) return ret; - for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, - &ctx.stream_size_tab); - INIT_LIST_HEAD(&ctx.stream_list); - wim->private = &ctx; - ret = for_image(wim, image, image_find_streams_to_write); - destroy_stream_size_table(&ctx.stream_size_tab); + ctx.write_flags = write_flags; + ctx.wim = wim; + + streams_ok = ((write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) != 0); + + ret = prepare_logical_stream_list(wim, image, streams_ok, &ctx); if (ret) - return ret; + goto out_destroy_table; + + handle_stream_filtering(&ctx); list_transfer(&ctx.stream_list, stream_list); - return 0; + ret = 0; +out_destroy_table: + destroy_stream_size_table(&ctx.stream_size_tab); + return ret; } -/* Writes the streams for the specified @image in @wim to @wim->out_fd. - * Alternatively, if @stream_list_override is specified, it is taken to be the - * list of streams to write (connected with 'write_streams_list') and @image is - * ignored. */ static int write_wim_streams(WIMStruct *wim, int image, int write_flags, unsigned num_threads, @@ -1890,22 +1962,24 @@ write_wim_streams(WIMStruct *wim, int image, int write_flags, struct list_head *stream_list; struct wim_lookup_table_entry *lte; - if (stream_list_override) { - stream_list = stream_list_override; - list_for_each_entry(lte, stream_list, write_streams_list) { - if (lte->refcnt) - lte->out_refcnt = lte->refcnt; - else - lte->out_refcnt = 1; - } - } else { + if (stream_list_override == NULL) { + /* Normal case: prepare stream list from image(s) being written. + */ stream_list = &_stream_list; - ret = prepare_stream_list(wim, image, stream_list); + ret = prepare_stream_list(wim, image, write_flags, stream_list); if (ret) return ret; + } else { + /* Currently only as a result of wimlib_split() being called: + * use stream list already explicitly provided. Use existing + * reference counts. */ + stream_list = stream_list_override; + list_for_each_entry(lte, stream_list, write_streams_list) { + lte->out_refcnt = (lte->refcnt ? lte->refcnt : 1); + lte->part_number = wim->hdr.part_number; + } } - list_for_each_entry(lte, stream_list, write_streams_list) - lte->part_number = wim->hdr.part_number; + return write_stream_list(stream_list, wim->lookup_table, &wim->out_fd, @@ -1924,8 +1998,10 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, int end_image; int write_resource_flags; - if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) { + DEBUG("Not writing any metadata resources."); return 0; + } write_resource_flags = write_flags_to_resource_flags(write_flags); @@ -1947,10 +2023,23 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, struct wim_image_metadata *imd; imd = wim->image_metadata[i - 1]; + /* Build a new metadata resource only if image was modified from + * the original (or was newly added). Otherwise just copy the + * existing one. */ if (imd->modified) { + DEBUG("Image %u was modified; building and writing new " + "metadata resource", i); ret = write_metadata_resource(wim, i, write_resource_flags); + } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { + DEBUG("Image %u was not modified; re-using existing " + "metadata resource.", i); + copy_resource_entry(&imd->metadata_lte->output_resource_entry, + &imd->metadata_lte->resource_entry); + ret = 0; } else { + DEBUG("Image %u was not modified; copying existing " + "metadata resource.", i); ret = write_wim_resource(imd->metadata_lte, &wim->out_fd, wim->compression_type, @@ -1965,6 +2054,34 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, return 0; } +static int +open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) +{ + int raw_fd; + DEBUG("Opening \"%"TS"\" for writing.", path); + + raw_fd = topen(path, open_flags | O_BINARY, 0644); + if (raw_fd < 0) { + ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path); + return WIMLIB_ERR_OPEN; + } + filedes_init(&wim->out_fd, raw_fd); + return 0; +} + +static int +close_wim_writable(WIMStruct *wim, int write_flags) +{ + int ret = 0; + + if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) + if (filedes_valid(&wim->out_fd)) + if (filedes_close(&wim->out_fd)) + ret = WIMLIB_ERR_WRITE; + filedes_invalidate(&wim->out_fd); + return ret; +} + /* * Finish writing a WIM file: write the lookup table, xml data, and integrity * table, then overwrite the WIM header. Always closes the WIM file descriptor @@ -2100,12 +2217,12 @@ finish_write(WIMStruct *wim, int image, int write_flags, ret = 0; out_close_wim: - if (filedes_close(&wim->out_fd)) { - ERROR_WITH_ERRNO("Failed to close the output WIM file"); - if (ret == 0) + if (close_wim_writable(wim, write_flags)) { + if (ret == 0) { + ERROR_WITH_ERRNO("Failed to close the output WIM file"); ret = WIMLIB_ERR_WRITE; + } } - filedes_invalidate(&wim->out_fd); return ret; } @@ -2135,32 +2252,6 @@ lock_wim(WIMStruct *wim, int fd) } #endif -static int -open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) -{ - int raw_fd; - DEBUG("Opening \"%"TS"\" for writing.", path); - - raw_fd = topen(path, open_flags | O_BINARY, 0644); - if (raw_fd < 0) { - ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path); - return WIMLIB_ERR_OPEN; - } - filedes_init(&wim->out_fd, raw_fd); - return 0; -} - - -static void -close_wim_writable(WIMStruct *wim) -{ - if (filedes_valid(&wim->out_fd)) { - if (filedes_close(&wim->out_fd)) - WARNING_WITH_ERRNO("Failed to close output WIM"); - filedes_invalidate(&wim->out_fd); - } -} - /* * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM * capable of being applied from a pipe). Such a WIM looks like: @@ -2336,7 +2427,8 @@ write_wim_part(WIMStruct *wim, DEBUG("Number of threads: %u", num_threads); DEBUG("Progress function: %s", (progress_func ? "yes" : "no")); DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect")); - DEBUG("GUID: %s", (guid ? "specified" : "generate new")); + DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ? + "specified" : "generate new")); /* Internally, this is always called with a valid part number and total * parts. */ @@ -2348,9 +2440,12 @@ write_wim_part(WIMStruct *wim, (image < 1 || image > wim->hdr.image_count)) return WIMLIB_ERR_INVALID_IMAGE; - /* @wim must specify a standalone WIM. */ - if (wim->hdr.total_parts != 1) - return WIMLIB_ERR_SPLIT_UNSUPPORTED; + /* If we need to write metadata resources, make sure the ::WIMStruct has + * the needed information attached (e.g. is not a resource-only WIM, + * such as a non-first part of a split WIM). */ + if (!wim_has_metadata(wim) && + !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) + return WIMLIB_ERR_METADATA_NOT_FOUND; /* Check for contradictory flags. */ if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | @@ -2403,7 +2498,7 @@ write_wim_part(WIMStruct *wim, /* Use GUID if specified; otherwise generate a new one. */ if (guid) memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN); - else + else if (!wim->guid_set_explicitly) randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN); /* Clear references to resources that have not been written yet. */ @@ -2502,7 +2597,7 @@ write_wim_part(WIMStruct *wim, stream_list_override); out_restore_hdr: memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); - close_wim_writable(wim); + close_wim_writable(wim, write_flags); return ret; } @@ -2556,6 +2651,40 @@ any_images_modified(WIMStruct *wim) return false; } +static int +check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim) +{ + const WIMStruct *wim = _wim; + off_t end_offset = *(const off_t*)wim->private; + + if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == wim && + lte->resource_entry.offset + lte->resource_entry.size > end_offset) + return WIMLIB_ERR_RESOURCE_ORDER; + return 0; +} + +/* Make sure no file or metadata resources are located after the XML data (or + * integrity table if present)--- otherwise we can't safely overwrite the WIM in + * place and we return WIMLIB_ERR_RESOURCE_ORDER. */ +static int +check_resource_offsets(WIMStruct *wim, off_t end_offset) +{ + int ret; + unsigned i; + + wim->private = &end_offset; + ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim); + if (ret) + return ret; + + for (i = 0; i < wim->hdr.image_count; i++) { + ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim); + if (ret) + return ret; + } + return 0; +} + /* * Overwrite a WIM, possibly appending streams to it. * @@ -2632,6 +2761,10 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, if (wim_has_integrity_table(wim)) write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + /* Set additional flags for overwrite. */ + write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE | + WIMLIB_WRITE_FLAG_STREAMS_OK; + /* Make sure that the integrity table (if present) is after the XML * data, and that there are no stream resources, metadata resources, or * lookup tables after the XML data. Otherwise, these data would be @@ -2641,12 +2774,12 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + wim->hdr.lookup_table_res_entry.size; if (wim->hdr.integrity.offset != 0 && wim->hdr.integrity.offset < old_xml_end) { - ERROR("Didn't expect the integrity table to be before the XML data"); + WARNING("Didn't expect the integrity table to be before the XML data"); return WIMLIB_ERR_RESOURCE_ORDER; } if (old_lookup_table_end > old_xml_begin) { - ERROR("Didn't expect the lookup table to be after the XML data"); + WARNING("Didn't expect the lookup table to be after the XML data"); return WIMLIB_ERR_RESOURCE_ORDER; } @@ -2675,7 +2808,12 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, old_wim_end = old_xml_end; } - ret = prepare_streams_for_overwrite(wim, old_wim_end, &stream_list); + ret = check_resource_offsets(wim, old_wim_end); + if (ret) + return ret; + + ret = prepare_stream_list(wim, WIMLIB_ALL_IMAGES, write_flags, + &stream_list); if (ret) return ret; @@ -2685,7 +2823,7 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, ret = lock_wim(wim, wim->out_fd.fd); if (ret) { - close_wim_writable(wim); + close_wim_writable(wim, write_flags); return ret; } @@ -2694,19 +2832,17 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, &wim->out_fd); if (ret) { ERROR_WITH_ERRNO("Error updating WIM header flags"); - close_wim_writable(wim); + close_wim_writable(wim, write_flags); goto out_unlock_wim; } if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - close_wim_writable(wim); + close_wim_writable(wim, write_flags); ret = WIMLIB_ERR_WRITE; goto out_unlock_wim; } - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); ret = write_stream_list(&stream_list, wim->lookup_table, &wim->out_fd, @@ -2717,19 +2853,17 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, if (ret) goto out_truncate; - for (unsigned i = 1; i <= wim->hdr.image_count; i++) { - if (wim->image_metadata[i - 1]->modified) { - ret = write_metadata_resource(wim, i, 0); - if (ret) - goto out_truncate; - } - } + ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, + write_flags, progress_func); + if (ret) + goto out_truncate; + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, progress_func, NULL); out_truncate: - close_wim_writable(wim); - if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + close_wim_writable(wim, write_flags); + if (ret && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", wim->filename, old_wim_end); /* Return value of truncate() is ignored because this is already