X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=a3f4213fef7513ab2bd744e2adda57c319530772;hp=1a5b50ec6180eba3c28b087d019f7d18a2af96e5;hb=707700a1f8f41a8e27bc87d4debffe4fb1d24ec7;hpb=3fa7b7a033ce803accfd6758029114323f6a6865 diff --git a/src/write.c b/src/write.c index 1a5b50ec..930c61b1 100644 --- a/src/write.c +++ b/src/write.c @@ -6,1949 +6,3353 @@ */ /* - * Copyright (C) 2010 Carl Thijssen - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers * - * This file is part of wimlib, a library for working with WIM files. + * This file is free software; you can redistribute it and/or modify it under + * the terms of the GNU Lesser General Public License as published by the Free + * Software Foundation; either version 3 of the License, or (at your option) any + * later version. * - * wimlib is free software; you can redistribute it and/or modify it under the - * terms of the GNU General Public License as published by the Free - * Software Foundation; either version 3 of the License, or (at your option) - * any later version. - * - * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY - * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR - * A PARTICULAR PURPOSE. See the GNU General Public License for more + * This file is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS + * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more * details. * - * You should have received a copy of the GNU General Public License - * along with wimlib; if not, see http://www.gnu.org/licenses/. + * You should have received a copy of the GNU Lesser General Public License + * along with this file; 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 +#ifdef HAVE_CONFIG_H +# include "config.h" #endif -#include "list.h" -#include "wimlib_internal.h" -#include "io.h" -#include "dentry.h" -#include "lookup_table.h" -#include "xml.h" -#include "lzx.h" -#include "xpress.h" - -#ifdef ENABLE_MULTITHREADED_COMPRESSION -#include +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +/* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can + * override the LIST_HEAD macro. */ +# include #endif -#include #include +#include +#include +#include -#ifdef WITH_NTFS_3G -#include -#include -#include -#include -#endif +#include "wimlib/alloca.h" +#include "wimlib/assert.h" +#include "wimlib/blob_table.h" +#include "wimlib/chunk_compressor.h" +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/header.h" +#include "wimlib/inode.h" +#include "wimlib/integrity.h" +#include "wimlib/metadata.h" +#include "wimlib/paths.h" +#include "wimlib/progress.h" +#include "wimlib/resource.h" +#include "wimlib/solid.h" +#include "wimlib/win32.h" /* win32_rename_replacement() */ +#include "wimlib/write.h" +#include "wimlib/xml.h" + + +/* wimlib internal flags used when writing resources. */ +#define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001 +#define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002 +#define WRITE_RESOURCE_FLAG_SOLID 0x00000004 +#define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008 +#define WRITE_RESOURCE_FLAG_SOLID_SORT 0x00000010 -#ifdef HAVE_ALLOCA_H -#include -#else -#include -#endif +static int +write_flags_to_resource_flags(int write_flags) +{ + int write_resource_flags = 0; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS; + + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE; + + if (write_flags & WIMLIB_WRITE_FLAG_SOLID) + write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID; + + if (write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES) + write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE; + + if ((write_flags & (WIMLIB_WRITE_FLAG_SOLID | + WIMLIB_WRITE_FLAG_NO_SOLID_SORT)) == + WIMLIB_WRITE_FLAG_SOLID) + write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID_SORT; + + return write_resource_flags; +} + +struct filter_context { + int write_flags; + WIMStruct *wim; +}; + +/* + * Determine whether the specified blob should be filtered out from the write. + * + * Return values: + * + * < 0 : The blob should be hard-filtered; that is, not included in the output + * WIM file at all. + * 0 : The blob should not be filtered out. + * > 0 : The blob should be soft-filtered; that is, it already exists in the + * WIM file and may not need to be written again. + */ +static int +blob_filtered(const struct blob_descriptor *blob, + const struct filter_context *ctx) +{ + int write_flags; + WIMStruct *wim; + + if (ctx == NULL) + return 0; + + write_flags = ctx->write_flags; + wim = ctx->wim; + + if (write_flags & WIMLIB_WRITE_FLAG_APPEND && + blob->blob_location == BLOB_IN_WIM && + blob->rdesc->wim == wim) + return 1; + + if (write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS && + blob->blob_location == BLOB_IN_WIM && + blob->rdesc->wim != wim) + return -1; + + return 0; +} + +static bool +blob_hard_filtered(const struct blob_descriptor *blob, + struct filter_context *ctx) +{ + return blob_filtered(blob, ctx) < 0; +} + +static inline bool +may_soft_filter_blobs(const struct filter_context *ctx) +{ + return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_APPEND); +} + +static inline bool +may_hard_filter_blobs(const struct filter_context *ctx) +{ + return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS); +} + +static inline bool +may_filter_blobs(const struct filter_context *ctx) +{ + return (may_soft_filter_blobs(ctx) || may_hard_filter_blobs(ctx)); +} -static int fflush_and_ftruncate(FILE *fp, off_t size) +/* Return true if the specified blob is located in a WIM resource which can be + * reused in the output WIM file, without being recompressed. */ +static bool +can_raw_copy(const struct blob_descriptor *blob, int write_resource_flags, + int out_ctype, u32 out_chunk_size) { + const struct wim_resource_descriptor *rdesc; + + /* Recompress everything if requested. */ + if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS) + return false; + + /* A blob not located in a WIM resource cannot be reused. */ + if (blob->blob_location != BLOB_IN_WIM) + return false; + + rdesc = blob->rdesc; + + /* In the case of an in-place compaction, always reuse resources located + * in the WIM being compacted. */ + if (rdesc->wim->being_compacted) + return true; + + /* Otherwise, only reuse compressed resources. */ + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || + !(rdesc->flags & (WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SOLID))) + return false; + + /* When writing a pipable WIM, we can only reuse pipable resources; and + * when writing a non-pipable WIM, we can only reuse non-pipable + * resources. */ + if (rdesc->is_pipable != + !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) + return false; + + /* When writing a solid WIM, we can only reuse solid resources; and when + * writing a non-solid WIM, we can only reuse non-solid resources. */ + if (!!(rdesc->flags & WIM_RESHDR_FLAG_SOLID) != + !!(write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) + return false; + + /* Note: it is theoretically possible to copy chunks of compressed data + * between non-solid, solid, and pipable resources. However, we don't + * currently implement this optimization because it would be complex and + * would usually go unused. */ + + if (rdesc->flags & WIM_RESHDR_FLAG_COMPRESSED) { + /* To re-use a non-solid resource, it must use the desired + * compression type and chunk size. */ + return (rdesc->compression_type == out_ctype && + rdesc->chunk_size == out_chunk_size); + } else { + /* Solid resource: Such resources may contain multiple blobs, + * and in general only a subset of them need to be written. As + * a heuristic, re-use the raw data if more than two-thirds the + * uncompressed size is being written. */ + + /* Note: solid resources contain a header that specifies the + * compression type and chunk size; therefore we don't need to + * check if they are compatible with @out_ctype and + * @out_chunk_size. */ + + /* Did we already decide to reuse the resource? */ + if (rdesc->raw_copy_ok) + return true; + + struct blob_descriptor *res_blob; + u64 write_size = 0; + + list_for_each_entry(res_blob, &rdesc->blob_list, rdesc_node) + if (res_blob->will_be_in_output_wim) + write_size += res_blob->size; + + return (write_size > rdesc->uncompressed_size * 2 / 3); + } +} + +static u32 +reshdr_flags_for_blob(const struct blob_descriptor *blob) +{ + u32 reshdr_flags = 0; + if (blob->is_metadata) + reshdr_flags |= WIM_RESHDR_FLAG_METADATA; + return reshdr_flags; +} + +static void +blob_set_out_reshdr_for_reuse(struct blob_descriptor *blob) +{ + const struct wim_resource_descriptor *rdesc; + + wimlib_assert(blob->blob_location == BLOB_IN_WIM); + rdesc = blob->rdesc; + + if (rdesc->flags & WIM_RESHDR_FLAG_SOLID) { + blob->out_reshdr.offset_in_wim = blob->offset_in_res; + blob->out_reshdr.uncompressed_size = 0; + blob->out_reshdr.size_in_wim = blob->size; + + blob->out_res_offset_in_wim = rdesc->offset_in_wim; + blob->out_res_size_in_wim = rdesc->size_in_wim; + blob->out_res_uncompressed_size = rdesc->uncompressed_size; + } else { + blob->out_reshdr.offset_in_wim = rdesc->offset_in_wim; + blob->out_reshdr.uncompressed_size = rdesc->uncompressed_size; + blob->out_reshdr.size_in_wim = rdesc->size_in_wim; + } + blob->out_reshdr.flags = rdesc->flags; +} + + +/* Write the header for a blob in a pipable WIM. */ +static int +write_pwm_blob_header(const struct blob_descriptor *blob, + struct filedes *out_fd, bool compressed) +{ + struct pwm_blob_hdr blob_hdr; + u32 reshdr_flags; int ret; - ret = fflush(fp); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to flush data to output WIM file"); - return WIMLIB_ERR_WRITE; + wimlib_assert(!blob->unhashed); + + blob_hdr.magic = cpu_to_le64(PWM_BLOB_MAGIC); + blob_hdr.uncompressed_size = cpu_to_le64(blob->size); + copy_hash(blob_hdr.hash, blob->hash); + reshdr_flags = reshdr_flags_for_blob(blob); + if (compressed) + reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; + blob_hdr.flags = cpu_to_le32(reshdr_flags); + ret = full_write(out_fd, &blob_hdr, sizeof(blob_hdr)); + if (ret) + ERROR_WITH_ERRNO("Write error"); + return ret; +} + +struct write_blobs_progress_data { + wimlib_progress_func_t progfunc; + void *progctx; + union wimlib_progress_info progress; + u64 next_progress; +}; + +static int +do_write_blobs_progress(struct write_blobs_progress_data *progress_data, + u64 complete_size, u32 complete_count, bool discarded) +{ + union wimlib_progress_info *progress = &progress_data->progress; + int ret; + + if (discarded) { + progress->write_streams.total_bytes -= complete_size; + progress->write_streams.total_streams -= complete_count; + if (progress_data->next_progress != ~(u64)0 && + progress_data->next_progress > progress->write_streams.total_bytes) + { + progress_data->next_progress = progress->write_streams.total_bytes; + } + } else { + progress->write_streams.completed_bytes += complete_size; + progress->write_streams.completed_streams += complete_count; } - ret = ftruncate(fileno(fp), size); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to truncate output WIM file to " - "%"PRIu64" bytes", size); - return WIMLIB_ERR_WRITE; + + if (progress->write_streams.completed_bytes >= progress_data->next_progress) { + + ret = call_progress(progress_data->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress, + progress_data->progctx); + if (ret) + return ret; + + set_next_progress(progress->write_streams.completed_bytes, + progress->write_streams.total_bytes, + &progress_data->next_progress); } return 0; } -/* Chunk table that's located at the beginning of each compressed resource in - * the WIM. (This is not the on-disk format; the on-disk format just has an - * array of offsets.) */ -struct chunk_table { - off_t file_offset; - u64 num_chunks; - u64 original_resource_size; - u64 bytes_per_chunk_entry; - u64 table_disk_size; - u64 cur_offset; - u64 *cur_offset_p; - u64 offsets[0]; +struct write_blobs_ctx { + /* File descriptor to which the blobs are being written. */ + struct filedes *out_fd; + + /* Blob table for the WIMStruct on whose behalf the blobs are being + * written. */ + struct blob_table *blob_table; + + /* Compression format to use. */ + int out_ctype; + + /* Maximum uncompressed chunk size in compressed resources to use. */ + u32 out_chunk_size; + + /* Flags that affect how the blobs will be written. */ + int write_resource_flags; + + /* Data used for issuing WRITE_STREAMS progress. */ + struct write_blobs_progress_data progress_data; + + struct filter_context *filter_ctx; + + /* Pointer to the chunk_compressor implementation being used for + * compressing chunks of data, or NULL if chunks are being written + * uncompressed. */ + struct chunk_compressor *compressor; + + /* A buffer of size @out_chunk_size that has been loaned out from the + * chunk compressor and is currently being filled with the uncompressed + * data of the next chunk. */ + u8 *cur_chunk_buf; + + /* Number of bytes in @cur_chunk_buf that are currently filled. */ + size_t cur_chunk_buf_filled; + + /* List of blobs that currently have chunks being compressed. */ + struct list_head blobs_being_compressed; + + /* List of blobs in the solid resource. Blobs are moved here after + * @blobs_being_compressed only when writing a solid resource. */ + struct list_head blobs_in_solid_resource; + + /* Current uncompressed offset in the blob being read. */ + u64 cur_read_blob_offset; + + /* Uncompressed size of the blob currently being read. */ + u64 cur_read_blob_size; + + /* Current uncompressed offset in the blob being written. */ + u64 cur_write_blob_offset; + + /* Uncompressed size of resource currently being written. */ + u64 cur_write_res_size; + + /* Array that is filled in with compressed chunk sizes as a resource is + * being written. */ + u64 *chunk_csizes; + + /* Index of next entry in @chunk_csizes to fill in. */ + size_t chunk_index; + + /* Number of entries in @chunk_csizes currently allocated. */ + size_t num_alloc_chunks; + + /* Offset in the output file of the start of the chunks of the resource + * currently being written. */ + u64 chunks_start_offset; }; -/* - * Allocates and initializes a chunk table, and reserves space for it in the - * output file. - */ +/* Reserve space for the chunk table and prepare to accumulate the chunk table + * in memory. */ static int -begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, - FILE *out_fp, - off_t file_offset, - struct chunk_table **chunk_tab_ret) -{ - u64 size = wim_resource_size(lte); - u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); - struct chunk_table *chunk_tab = CALLOC(1, alloc_size); +begin_chunk_table(struct write_blobs_ctx *ctx, u64 res_expected_size) +{ + u64 expected_num_chunks; + u64 expected_num_chunk_entries; + size_t reserve_size; int ret; - if (!chunk_tab) { - ERROR("Failed to allocate chunk table for %"PRIu64" byte " - "resource", size); - ret = WIMLIB_ERR_NOMEM; - goto out; + /* Calculate the number of chunks and chunk entries that should be + * needed for the resource. These normally will be the final values, + * but in SOLID mode some of the blobs we're planning to write into the + * resource may be duplicates, and therefore discarded, potentially + * decreasing the number of chunk entries needed. */ + expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size); + expected_num_chunk_entries = expected_num_chunks; + if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) + expected_num_chunk_entries--; + + /* Make sure the chunk_csizes array is long enough to store the + * compressed size of each chunk. */ + if (expected_num_chunks > ctx->num_alloc_chunks) { + u64 new_length = expected_num_chunks + 50; + + if ((size_t)new_length != new_length) { + ERROR("Resource size too large (%"PRIu64" bytes!", + res_expected_size); + return WIMLIB_ERR_NOMEM; + } + + FREE(ctx->chunk_csizes); + ctx->chunk_csizes = MALLOC(new_length * sizeof(ctx->chunk_csizes[0])); + if (ctx->chunk_csizes == NULL) { + ctx->num_alloc_chunks = 0; + return WIMLIB_ERR_NOMEM; + } + ctx->num_alloc_chunks = new_length; + } + + ctx->chunk_index = 0; + + if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) { + /* Reserve space for the chunk table in the output file. In the + * case of solid resources this reserves the upper bound for the + * needed space, not necessarily the exact space which will + * prove to be needed. At this point, we just use @chunk_csizes + * for a buffer of 0's because the actual compressed chunk sizes + * are unknown. */ + reserve_size = expected_num_chunk_entries * + get_chunk_entry_size(res_expected_size, + 0 != (ctx->write_resource_flags & + WRITE_RESOURCE_FLAG_SOLID)); + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) + reserve_size += sizeof(struct alt_chunk_table_header_disk); + memset(ctx->chunk_csizes, 0, reserve_size); + ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size); + if (ret) + return ret; } - chunk_tab->file_offset = file_offset; - chunk_tab->num_chunks = num_chunks; - chunk_tab->original_resource_size = size; - chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4; - chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * - (num_chunks - 1); - chunk_tab->cur_offset = 0; - chunk_tab->cur_offset_p = chunk_tab->offsets; - - if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) != - chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - ret = WIMLIB_ERR_WRITE; - goto out; + return 0; +} + +static int +begin_write_resource(struct write_blobs_ctx *ctx, u64 res_expected_size) +{ + int ret; + + wimlib_assert(res_expected_size != 0); + + if (ctx->compressor != NULL) { + ret = begin_chunk_table(ctx, res_expected_size); + if (ret) + return ret; } - ret = 0; -out: - *chunk_tab_ret = chunk_tab; + /* Output file descriptor is now positioned at the offset at which to + * write the first chunk of the resource. */ + ctx->chunks_start_offset = ctx->out_fd->offset; + ctx->cur_write_blob_offset = 0; + ctx->cur_write_res_size = res_expected_size; + return 0; +} + +static int +end_chunk_table(struct write_blobs_ctx *ctx, u64 res_actual_size, + u64 *res_start_offset_ret, u64 *res_store_size_ret) +{ + size_t actual_num_chunks; + size_t actual_num_chunk_entries; + size_t chunk_entry_size; + int ret; + + actual_num_chunks = ctx->chunk_index; + actual_num_chunk_entries = actual_num_chunks; + if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) + actual_num_chunk_entries--; + + chunk_entry_size = get_chunk_entry_size(res_actual_size, + 0 != (ctx->write_resource_flags & + WRITE_RESOURCE_FLAG_SOLID)); + + typedef le64 _may_alias_attribute aliased_le64_t; + typedef le32 _may_alias_attribute aliased_le32_t; + + if (chunk_entry_size == 4) { + aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + for (size_t i = 0; i < actual_num_chunk_entries; i++) + entries[i] = cpu_to_le32(ctx->chunk_csizes[i]); + } else { + u32 offset = ctx->chunk_csizes[0]; + for (size_t i = 0; i < actual_num_chunk_entries; i++) { + u32 next_size = ctx->chunk_csizes[i + 1]; + entries[i] = cpu_to_le32(offset); + offset += next_size; + } + } + } else { + aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes; + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + for (size_t i = 0; i < actual_num_chunk_entries; i++) + entries[i] = cpu_to_le64(ctx->chunk_csizes[i]); + } else { + u64 offset = ctx->chunk_csizes[0]; + for (size_t i = 0; i < actual_num_chunk_entries; i++) { + u64 next_size = ctx->chunk_csizes[i + 1]; + entries[i] = cpu_to_le64(offset); + offset += next_size; + } + } + } + + size_t chunk_table_size = actual_num_chunk_entries * chunk_entry_size; + u64 res_start_offset; + u64 res_end_offset; + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size); + if (ret) + goto write_error; + res_end_offset = ctx->out_fd->offset; + res_start_offset = ctx->chunks_start_offset; + } else { + res_end_offset = ctx->out_fd->offset; + + u64 chunk_table_offset; + + chunk_table_offset = ctx->chunks_start_offset - chunk_table_size; + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + struct alt_chunk_table_header_disk hdr; + + hdr.res_usize = cpu_to_le64(res_actual_size); + hdr.chunk_size = cpu_to_le32(ctx->out_chunk_size); + hdr.compression_format = cpu_to_le32(ctx->out_ctype); + + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2); + STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3); + + ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr), + chunk_table_offset - sizeof(hdr)); + if (ret) + goto write_error; + res_start_offset = chunk_table_offset - sizeof(hdr); + } else { + res_start_offset = chunk_table_offset; + } + + ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes, + chunk_table_size, chunk_table_offset); + if (ret) + goto write_error; + } + + *res_start_offset_ret = res_start_offset; + *res_store_size_ret = res_end_offset - res_start_offset; + + return 0; + +write_error: + ERROR_WITH_ERRNO("Write error"); return ret; } -/* - * Pointer to function to compresses a chunk of a WIM resource. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk in bytes. - * @compressed_chunk: Pointer to output buffer of size at least - * (@chunk_size - 1) bytes. - * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size - * of the compressed chunk will be - * returned. - * - * Returns zero if compressed succeeded, and nonzero if the chunk could not be - * compressed to any smaller than @chunk_size. This function cannot fail for - * any other reasons. - */ -typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); +/* Finish writing a WIM resource by writing or updating the chunk table (if not + * writing the data uncompressed) and loading its metadata into @out_reshdr. */ +static int +end_write_resource(struct write_blobs_ctx *ctx, struct wim_reshdr *out_reshdr) +{ + int ret; + u64 res_size_in_wim; + u64 res_uncompressed_size; + u64 res_offset_in_wim; + + wimlib_assert(ctx->cur_write_blob_offset == ctx->cur_write_res_size || + (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)); + res_uncompressed_size = ctx->cur_write_res_size; + + if (ctx->compressor) { + ret = end_chunk_table(ctx, res_uncompressed_size, + &res_offset_in_wim, &res_size_in_wim); + if (ret) + return ret; + } else { + res_offset_in_wim = ctx->chunks_start_offset; + res_size_in_wim = ctx->out_fd->offset - res_offset_in_wim; + } + out_reshdr->uncompressed_size = res_uncompressed_size; + out_reshdr->size_in_wim = res_size_in_wim; + out_reshdr->offset_in_wim = res_offset_in_wim; + return 0; +} -compress_func_t get_compress_func(int out_ctype) +/* Call when no more data from the file at @path is needed. */ +static int +done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx) { - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return lzx_compress; - else - return xpress_compress; + union wimlib_progress_info info; + + info.done_with_file.path_to_file = path; + + return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE, + &info, progctx); } -/* - * Writes a chunk of a WIM resource to an output file. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) - * @out_fp: FILE * to write tho chunk to. - * @out_ctype: Compression type to use when writing the chunk (ignored if no - * chunk table provided) - * @chunk_tab: Pointer to chunk table being created. It is updated with the - * offset of the chunk we write. - * - * Returns 0 on success; nonzero on failure. - */ -static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, - FILE *out_fp, compress_func_t compress, - struct chunk_table *chunk_tab) -{ - const u8 *out_chunk; - unsigned out_chunk_size; - if (chunk_tab) { - u8 *compressed_chunk = alloca(chunk_size); - int ret; +static int +do_done_with_blob(struct blob_descriptor *blob, + wimlib_progress_func_t progfunc, void *progctx) +{ + int ret; + struct wim_inode *inode; + tchar *cookie1; + tchar *cookie2; - ret = compress(chunk, chunk_size, compressed_chunk, - &out_chunk_size); - if (ret == 0) { - out_chunk = compressed_chunk; - } else { - out_chunk = chunk; - out_chunk_size = chunk_size; + if (!blob->may_send_done_with_file) + return 0; + + inode = blob->file_inode; + + wimlib_assert(inode != NULL); + wimlib_assert(inode->i_num_remaining_streams > 0); + if (--inode->i_num_remaining_streams > 0) + return 0; + + cookie1 = progress_get_streamless_path(blob->file_on_disk); + cookie2 = progress_get_win32_path(blob->file_on_disk); + + ret = done_with_file(blob->file_on_disk, progfunc, progctx); + + progress_put_win32_path(cookie2); + progress_put_streamless_path(cookie1); + + return ret; +} + +/* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */ +static inline int +done_with_blob(struct blob_descriptor *blob, struct write_blobs_ctx *ctx) +{ + if (likely(!(ctx->write_resource_flags & + WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))) + return 0; + return do_done_with_blob(blob, ctx->progress_data.progfunc, + ctx->progress_data.progctx); +} + +/* Begin processing a blob for writing. */ +static int +write_blob_begin_read(struct blob_descriptor *blob, void *_ctx) +{ + struct write_blobs_ctx *ctx = _ctx; + int ret; + + wimlib_assert(blob->size > 0); + + ctx->cur_read_blob_offset = 0; + ctx->cur_read_blob_size = blob->size; + + /* As an optimization, we allow some blobs to be "unhashed", meaning + * their SHA-1 message digests are unknown. This is the case with blobs + * that are added by scanning a directory tree with wimlib_add_image(), + * for example. Since WIM uses single-instance blobs, we don't know + * whether such each such blob really need to written until it is + * actually checksummed, unless it has a unique size. In such cases we + * read and checksum the blob in this function, thereby advancing ahead + * of read_blob_list(), which will still provide the data again to + * write_blob_process_chunk(). This is okay because an unhashed blob + * cannot be in a WIM resource, which might be costly to decompress. */ + if (ctx->blob_table != NULL && blob->unhashed && !blob->unique_size) { + + struct blob_descriptor *new_blob; + + ret = hash_unhashed_blob(blob, ctx->blob_table, &new_blob); + if (ret) + return ret; + if (new_blob != blob) { + /* Duplicate blob detected. */ + + if (new_blob->will_be_in_output_wim || + blob_filtered(new_blob, ctx->filter_ctx)) + { + /* The duplicate blob is already being included + * in the output WIM, or it would be filtered + * out if it had been. Skip writing this blob + * (and reading it again) entirely, passing its + * output reference count to the duplicate blob + * in the former case. */ + ret = do_write_blobs_progress(&ctx->progress_data, + blob->size, 1, true); + list_del(&blob->write_blobs_list); + list_del(&blob->blob_table_list); + if (new_blob->will_be_in_output_wim) + new_blob->out_refcnt += blob->out_refcnt; + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) + ctx->cur_write_res_size -= blob->size; + if (!ret) + ret = done_with_blob(blob, ctx); + free_blob_descriptor(blob); + if (ret) + return ret; + return BEGIN_BLOB_STATUS_SKIP_BLOB; + } else { + /* The duplicate blob can validly be written, + * but was not marked as such. Discard the + * current blob descriptor and use the + * duplicate, but actually freeing the current + * blob descriptor must wait until + * read_blob_list() has finished reading its + * data. */ + list_replace(&blob->write_blobs_list, + &new_blob->write_blobs_list); + list_replace(&blob->blob_table_list, + &new_blob->blob_table_list); + blob->will_be_in_output_wim = 0; + new_blob->out_refcnt = blob->out_refcnt; + new_blob->will_be_in_output_wim = 1; + new_blob->may_send_done_with_file = 0; + blob = new_blob; + } + } + } + list_move_tail(&blob->write_blobs_list, &ctx->blobs_being_compressed); + return 0; +} + +/* Rewrite a blob that was just written compressed (as a non-solid WIM resource) + * as uncompressed instead. */ +static int +write_blob_uncompressed(struct blob_descriptor *blob, struct filedes *out_fd) +{ + int ret; + u64 begin_offset = blob->out_reshdr.offset_in_wim; + u64 end_offset = out_fd->offset; + + if (filedes_seek(out_fd, begin_offset) == -1) + return 0; + + ret = extract_blob_to_fd(blob, out_fd); + if (ret) { + /* Error reading the uncompressed data. */ + if (out_fd->offset == begin_offset && + filedes_seek(out_fd, end_offset) != -1) + { + /* Nothing was actually written yet, and we successfully + * seeked to the end of the compressed resource, so + * don't issue a hard error; just keep the compressed + * resource instead. */ + WARNING("Recovered compressed resource of " + "size %"PRIu64", continuing on.", blob->size); + return 0; + } + return ret; + } + + wimlib_assert(out_fd->offset - begin_offset == blob->size); + + /* We could ftruncate() the file to 'out_fd->offset' here, but there + * isn't much point. Usually we will only be truncating by a few bytes + * and will just overwrite the data immediately. */ + + blob->out_reshdr.size_in_wim = blob->size; + blob->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SOLID); + return 0; +} + +/* Returns true if the specified blob, which was written as a non-solid + * resource, should be truncated from the WIM file and re-written uncompressed. + * blob->out_reshdr must be filled in from the initial write of the blob. */ +static bool +should_rewrite_blob_uncompressed(const struct write_blobs_ctx *ctx, + const struct blob_descriptor *blob) +{ + /* If the compressed data is smaller than the uncompressed data, prefer + * the compressed data. */ + if (blob->out_reshdr.size_in_wim < blob->out_reshdr.uncompressed_size) + return false; + + /* If we're not actually writing compressed data, then there's no need + * for re-writing. */ + if (!ctx->compressor) + return false; + + /* If writing a pipable WIM, everything we write to the output is final + * (it might actually be a pipe!). */ + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) + return false; + + /* If the blob that would need to be re-read is located in a solid + * resource in another WIM file, then re-reading it would be costly. So + * don't do it. + * + * Exception: if the compressed size happens to be *exactly* the same as + * the uncompressed size, then the blob *must* be written uncompressed + * in order to remain compatible with the Windows Overlay Filesystem + * Filter Driver (WOF). + * + * TODO: we are currently assuming that the optimization for + * single-chunk resources in maybe_rewrite_blob_uncompressed() prevents + * this case from being triggered too often. To fully prevent excessive + * decompressions in degenerate cases, we really should obtain the + * uncompressed data by decompressing the compressed data we wrote to + * the output file. + */ + if (blob->blob_location == BLOB_IN_WIM && + blob->size != blob->rdesc->uncompressed_size && + blob->size != blob->out_reshdr.size_in_wim) + return false; + + return true; +} + +static int +maybe_rewrite_blob_uncompressed(struct write_blobs_ctx *ctx, + struct blob_descriptor *blob) +{ + if (!should_rewrite_blob_uncompressed(ctx, blob)) + return 0; + + /* Regular (non-solid) WIM resources with exactly one chunk and + * compressed size equal to uncompressed size are exactly the same as + * the corresponding compressed data --- since there must be 0 entries + * in the chunk table and the only chunk must be stored uncompressed. + * In this case, there's no need to rewrite anything. */ + if (ctx->chunk_index == 1 && + blob->out_reshdr.size_in_wim == blob->out_reshdr.uncompressed_size) + { + blob->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED; + return 0; + } + + return write_blob_uncompressed(blob, ctx->out_fd); +} + +/* Write the next chunk of (typically compressed) data to the output WIM, + * handling the writing of the chunk table. */ +static int +write_chunk(struct write_blobs_ctx *ctx, const void *cchunk, + size_t csize, size_t usize) +{ + int ret; + struct blob_descriptor *blob; + u32 completed_blob_count; + u32 completed_size; + + blob = list_entry(ctx->blobs_being_compressed.next, + struct blob_descriptor, write_blobs_list); + + if (ctx->cur_write_blob_offset == 0 && + !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) + { + /* Starting to write a new blob in non-solid mode. */ + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + ret = write_pwm_blob_header(blob, ctx->out_fd, + ctx->compressor != NULL); + if (ret) + return ret; + } + + ret = begin_write_resource(ctx, blob->size); + if (ret) + return ret; + } + + if (ctx->compressor != NULL) { + /* Record the compresed chunk size. */ + wimlib_assert(ctx->chunk_index < ctx->num_alloc_chunks); + ctx->chunk_csizes[ctx->chunk_index++] = csize; + + /* If writing a pipable WIM, before the chunk data write a chunk + * header that provides the compressed chunk size. */ + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { + struct pwm_chunk_hdr chunk_hdr = { + .compressed_size = cpu_to_le32(csize), + }; + ret = full_write(ctx->out_fd, &chunk_hdr, + sizeof(chunk_hdr)); + if (ret) + goto write_error; + } + } + + /* Write the chunk data. */ + ret = full_write(ctx->out_fd, cchunk, csize); + if (ret) + goto write_error; + + ctx->cur_write_blob_offset += usize; + + completed_size = usize; + completed_blob_count = 0; + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + /* Wrote chunk in solid mode. It may have finished multiple + * blobs. */ + struct blob_descriptor *next_blob; + + while (blob && ctx->cur_write_blob_offset >= blob->size) { + + ctx->cur_write_blob_offset -= blob->size; + + if (ctx->cur_write_blob_offset) + next_blob = list_entry(blob->write_blobs_list.next, + struct blob_descriptor, + write_blobs_list); + else + next_blob = NULL; + + ret = done_with_blob(blob, ctx); + if (ret) + return ret; + list_move_tail(&blob->write_blobs_list, &ctx->blobs_in_solid_resource); + completed_blob_count++; + + blob = next_blob; } - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += out_chunk_size; } else { - out_chunk = chunk; - out_chunk_size = chunk_size; + /* Wrote chunk in non-solid mode. It may have finished a + * blob. */ + if (ctx->cur_write_blob_offset == blob->size) { + + wimlib_assert(ctx->cur_write_blob_offset == + ctx->cur_write_res_size); + + ret = end_write_resource(ctx, &blob->out_reshdr); + if (ret) + return ret; + + blob->out_reshdr.flags = reshdr_flags_for_blob(blob); + if (ctx->compressor != NULL) + blob->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED; + + ret = maybe_rewrite_blob_uncompressed(ctx, blob); + if (ret) + return ret; + + wimlib_assert(blob->out_reshdr.uncompressed_size == blob->size); + + ctx->cur_write_blob_offset = 0; + + ret = done_with_blob(blob, ctx); + if (ret) + return ret; + list_del(&blob->write_blobs_list); + completed_blob_count++; + } } - if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return WIMLIB_ERR_WRITE; + + return do_write_blobs_progress(&ctx->progress_data, completed_size, + completed_blob_count, false); + +write_error: + ERROR_WITH_ERRNO("Write error"); + return ret; +} + +static int +prepare_chunk_buffer(struct write_blobs_ctx *ctx) +{ + /* While we are unable to get a new chunk buffer due to too many chunks + * already outstanding, retrieve and write the next compressed chunk. */ + while (!(ctx->cur_chunk_buf = + ctx->compressor->get_chunk_buffer(ctx->compressor))) + { + const void *cchunk; + u32 csize; + u32 usize; + bool bret; + int ret; + + bret = ctx->compressor->get_compression_result(ctx->compressor, + &cchunk, + &csize, + &usize); + wimlib_assert(bret); + + ret = write_chunk(ctx, cchunk, csize, usize); + if (ret) + return ret; } return 0; } +/* Process the next chunk of data to be written to a WIM resource. */ +static int +write_blob_process_chunk(const void *chunk, size_t size, void *_ctx) +{ + struct write_blobs_ctx *ctx = _ctx; + int ret; + const u8 *chunkptr, *chunkend; + + wimlib_assert(size != 0); + + if (ctx->compressor == NULL) { + /* Write chunk uncompressed. */ + ret = write_chunk(ctx, chunk, size, size); + if (ret) + return ret; + ctx->cur_read_blob_offset += size; + return 0; + } + + /* Submit the chunk for compression, but take into account that the + * @size the chunk was provided in may not correspond to the + * @out_chunk_size being used for compression. */ + chunkptr = chunk; + chunkend = chunkptr + size; + do { + size_t needed_chunk_size; + size_t bytes_consumed; + + if (!ctx->cur_chunk_buf) { + ret = prepare_chunk_buffer(ctx); + if (ret) + return ret; + } + + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + needed_chunk_size = ctx->out_chunk_size; + } else { + needed_chunk_size = min(ctx->out_chunk_size, + ctx->cur_chunk_buf_filled + + (ctx->cur_read_blob_size - + ctx->cur_read_blob_offset)); + } + + bytes_consumed = min(chunkend - chunkptr, + needed_chunk_size - ctx->cur_chunk_buf_filled); + + memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled], + chunkptr, bytes_consumed); + + chunkptr += bytes_consumed; + ctx->cur_read_blob_offset += bytes_consumed; + ctx->cur_chunk_buf_filled += bytes_consumed; + + if (ctx->cur_chunk_buf_filled == needed_chunk_size) { + ctx->compressor->signal_chunk_filled(ctx->compressor, + ctx->cur_chunk_buf_filled); + ctx->cur_chunk_buf = NULL; + ctx->cur_chunk_buf_filled = 0; + } + } while (chunkptr != chunkend); + return 0; +} + +/* Finish processing a blob for writing. It may not have been completely + * written yet, as the chunk_compressor implementation may still have chunks + * buffered or being compressed. */ +static int +write_blob_end_read(struct blob_descriptor *blob, int status, void *_ctx) +{ + struct write_blobs_ctx *ctx = _ctx; + + wimlib_assert(ctx->cur_read_blob_offset == ctx->cur_read_blob_size || status); + + if (!blob->will_be_in_output_wim) { + /* The blob was a duplicate. Now that its data has finished + * being read, it is being discarded in favor of the duplicate + * entry. It therefore is no longer needed, and we can fire the + * DONE_WITH_FILE callback because the file will not be read + * again. + * + * Note: we can't yet fire DONE_WITH_FILE for non-duplicate + * blobs, since it needs to be possible to re-read the file if + * it does not compress to less than its original size. */ + if (!status) + status = done_with_blob(blob, ctx); + free_blob_descriptor(blob); + } else if (!status && blob->unhashed && ctx->blob_table != NULL) { + /* The blob was not a duplicate and was previously unhashed. + * Since we passed COMPUTE_MISSING_BLOB_HASHES to + * read_blob_list(), blob->hash is now computed and valid. So + * turn this blob into a "hashed" blob. */ + list_del(&blob->unhashed_list); + blob_table_insert(ctx->blob_table, blob); + blob->unhashed = 0; + } + return status; +} + /* - * Finishes a WIM chunk table and writes it to the output file at the correct - * offset. + * Compute statistics about a list of blobs that will be written. + * + * Assumes the blobs are sorted such that all blobs located in each distinct WIM + * (specified by WIMStruct) are together. * - * The final size of the full compressed resource is returned in the - * @compressed_size_p. + * For compactions, also verify that there are no overlapping resources. This + * really should be checked earlier, but for now it's easiest to check here. */ static int -finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - FILE *out_fp, u64 *compressed_size_p) +compute_blob_list_stats(struct list_head *blob_list, + struct write_blobs_ctx *ctx) { - size_t bytes_written; - if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output " - "WIM file", chunk_tab->file_offset); - return WIMLIB_ERR_WRITE; + struct blob_descriptor *blob; + u64 total_bytes = 0; + u64 num_blobs = 0; + u64 total_parts = 0; + WIMStruct *prev_wim_part = NULL; + const struct wim_resource_descriptor *prev_rdesc = NULL; + + list_for_each_entry(blob, blob_list, write_blobs_list) { + num_blobs++; + total_bytes += blob->size; + if (blob->blob_location == BLOB_IN_WIM) { + const struct wim_resource_descriptor *rdesc = blob->rdesc; + WIMStruct *wim = rdesc->wim; + + if (prev_wim_part != wim) { + prev_wim_part = wim; + total_parts++; + } + if (unlikely(wim->being_compacted) && rdesc != prev_rdesc) { + if (prev_rdesc != NULL && + rdesc->offset_in_wim < + prev_rdesc->offset_in_wim + + prev_rdesc->size_in_wim) + { + WARNING("WIM file contains overlapping " + "resources! Compaction is not " + "possible."); + return WIMLIB_ERR_RESOURCE_ORDER; + } + prev_rdesc = rdesc; + } + } } - - if (chunk_tab->bytes_per_chunk_entry == 8) { - array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); - } else { - for (u64 i = 0; i < chunk_tab->num_chunks; i++) - ((u32*)chunk_tab->offsets)[i] = - cpu_to_le32(chunk_tab->offsets[i]); - } - bytes_written = fwrite((u8*)chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - 1, chunk_tab->table_disk_size, out_fp); - if (bytes_written != chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - return WIMLIB_ERR_WRITE; - } - if (fseeko(out_fp, 0, SEEK_END) != 0) { - ERROR_WITH_ERRNO("Failed to seek to end of output WIM file"); - return WIMLIB_ERR_WRITE; - } - *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; + ctx->progress_data.progress.write_streams.total_bytes = total_bytes; + ctx->progress_data.progress.write_streams.total_streams = num_blobs; + ctx->progress_data.progress.write_streams.completed_bytes = 0; + ctx->progress_data.progress.write_streams.completed_streams = 0; + ctx->progress_data.progress.write_streams.compression_type = ctx->out_ctype; + ctx->progress_data.progress.write_streams.total_parts = total_parts; + ctx->progress_data.progress.write_streams.completed_parts = 0; + ctx->progress_data.next_progress = 0; 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 - ) +/* Find blobs in @blob_list that can be copied to the output WIM in raw form + * rather than compressed. Delete these blobs from @blob_list and move them to + * @raw_copy_blobs. Return the total uncompressed size of the blobs that need + * to be compressed. */ +static u64 +find_raw_copy_blobs(struct list_head *blob_list, int write_resource_flags, + int out_ctype, u32 out_chunk_size, + struct list_head *raw_copy_blobs) { - 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; + struct blob_descriptor *blob, *tmp; + u64 num_nonraw_bytes = 0; + + INIT_LIST_HEAD(raw_copy_blobs); + + /* Initialize temporary raw_copy_ok flag. */ + list_for_each_entry(blob, blob_list, write_blobs_list) + if (blob->blob_location == BLOB_IN_WIM) + blob->rdesc->raw_copy_ok = 0; + + list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) { + if (can_raw_copy(blob, write_resource_flags, + out_ctype, out_chunk_size)) + { + blob->rdesc->raw_copy_ok = 1; + list_move_tail(&blob->write_blobs_list, raw_copy_blobs); + } else { + num_nonraw_bytes += blob->size; } } -#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; + + return num_nonraw_bytes; +} + +/* Copy a raw compressed resource located in another WIM file to the WIM file + * being written. */ +static int +write_raw_copy_resource(struct wim_resource_descriptor *in_rdesc, + struct filedes *out_fd) +{ + u64 cur_read_offset; + u64 end_read_offset; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; + int ret; + struct filedes *in_fd; + struct blob_descriptor *blob; + u64 out_offset_in_wim; + + /* Copy the raw data. */ + cur_read_offset = in_rdesc->offset_in_wim; + end_read_offset = cur_read_offset + in_rdesc->size_in_wim; + + out_offset_in_wim = out_fd->offset; + + if (in_rdesc->is_pipable) { + if (cur_read_offset < sizeof(struct pwm_blob_hdr)) + return WIMLIB_ERR_INVALID_PIPABLE_WIM; + cur_read_offset -= sizeof(struct pwm_blob_hdr); + out_offset_in_wim += sizeof(struct pwm_blob_hdr); + } + in_fd = &in_rdesc->wim->in_fd; + wimlib_assert(cur_read_offset != end_read_offset); + + if (likely(!in_rdesc->wim->being_compacted) || + in_rdesc->offset_in_wim > out_fd->offset) { + do { + bytes_to_read = min(sizeof(buf), + end_read_offset - cur_read_offset); + + ret = full_pread(in_fd, buf, bytes_to_read, + cur_read_offset); + if (ret) + return ret; + + ret = full_write(out_fd, buf, bytes_to_read); + if (ret) + return ret; + + cur_read_offset += bytes_to_read; + + } while (cur_read_offset != end_read_offset); + } else { + /* Optimization: the WIM file is being compacted and the + * resource being written is already in the desired location. + * Skip over the data instead of re-writing it. */ + + /* Due the earlier check for overlapping resources, it should + * never be the case that we already overwrote the resource. */ + wimlib_assert(!(in_rdesc->offset_in_wim < out_fd->offset)); + + if (-1 == filedes_seek(out_fd, out_fd->offset + in_rdesc->size_in_wim)) + return WIMLIB_ERR_WRITE; + } + + list_for_each_entry(blob, &in_rdesc->blob_list, rdesc_node) { + if (blob->will_be_in_output_wim) { + blob_set_out_reshdr_for_reuse(blob); + if (in_rdesc->flags & WIM_RESHDR_FLAG_SOLID) + blob->out_res_offset_in_wim = out_offset_in_wim; + else + blob->out_reshdr.offset_in_wim = out_offset_in_wim; + } - 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; + } + return 0; +} + +/* Copy a list of raw compressed resources located in other WIM file(s) to the + * WIM file being written. */ +static int +write_raw_copy_resources(struct list_head *raw_copy_blobs, + struct filedes *out_fd, + struct write_blobs_progress_data *progress_data) +{ + struct blob_descriptor *blob; + int ret; + + list_for_each_entry(blob, raw_copy_blobs, write_blobs_list) + blob->rdesc->raw_copy_ok = 1; + + list_for_each_entry(blob, raw_copy_blobs, write_blobs_list) { + if (blob->rdesc->raw_copy_ok) { + /* Write each solid resource only one time. */ + ret = write_raw_copy_resource(blob->rdesc, out_fd); + if (ret) + return ret; + blob->rdesc->raw_copy_ok = 0; } - *ni_ret = ni; + ret = do_write_blobs_progress(progress_data, blob->size, + 1, false); + if (ret) + return ret; + } + return 0; +} + +/* Wait for and write all chunks pending in the compressor. */ +static int +finish_remaining_chunks(struct write_blobs_ctx *ctx) +{ + const void *cdata; + u32 csize; + u32 usize; + int ret; + + if (ctx->compressor == NULL) + return 0; + + if (ctx->cur_chunk_buf_filled != 0) { + ctx->compressor->signal_chunk_filled(ctx->compressor, + ctx->cur_chunk_buf_filled); + } + + while (ctx->compressor->get_compression_result(ctx->compressor, &cdata, + &csize, &usize)) + { + ret = write_chunk(ctx, cdata, csize, usize); + if (ret) + return ret; } -#endif return 0; } -/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS - * attribute. */ -static void end_wim_resource_read(struct lookup_table_entry *lte - #ifdef WITH_NTFS_3G - , ntfs_inode *ni - #endif - ) -{ - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && lte->file_on_disk_fp) { - fclose(lte->file_on_disk_fp); - lte->file_on_disk_fp = NULL; - } -#ifdef WITH_NTFS_3G - else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) { - if (lte->attr) { - ntfs_attr_close(lte->attr); - lte->attr = NULL; - } - if (ni) - ntfs_inode_close(ni); +static void +validate_blob_list(struct list_head *blob_list) +{ + struct blob_descriptor *blob; + + list_for_each_entry(blob, blob_list, write_blobs_list) { + wimlib_assert(blob->will_be_in_output_wim); + wimlib_assert(blob->size != 0); } -#endif } -static int -write_uncompressed_resource_and_truncate(struct lookup_table_entry *lte, - FILE *out_fp, - off_t file_offset, - struct resource_entry *out_res_entry) +static void +init_done_with_file_info(struct list_head *blob_list) { - 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; + struct blob_descriptor *blob; + + list_for_each_entry(blob, blob_list, write_blobs_list) { + if (blob_is_in_file(blob)) { + blob->file_inode->i_num_remaining_streams = 0; + blob->may_send_done_with_file = 1; + } else { + blob->may_send_done_with_file = 0; + } + } - return fflush_and_ftruncate(out_fp, - file_offset + wim_resource_size(lte)); + list_for_each_entry(blob, blob_list, write_blobs_list) + if (blob->may_send_done_with_file) + blob->file_inode->i_num_remaining_streams++; } /* - * 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. + * Write a list of blobs to the output WIM file. + * + * @blob_list + * The list of blobs to write, specified by a list of 'struct blob_descriptor' linked + * by the 'write_blobs_list' member. + * + * @out_fd + * The file descriptor, opened for writing, to which to write the blobs. + * + * @write_resource_flags + * Flags to modify how the blobs are written: + * + * WRITE_RESOURCE_FLAG_RECOMPRESS: + * Force compression of all resources, even if they could otherwise + * be re-used by copying the raw data, due to being located in a WIM + * file with compatible compression parameters. + * + * WRITE_RESOURCE_FLAG_PIPABLE: + * Write the resources in the wimlib-specific pipable format, and + * furthermore do so in such a way that no seeking backwards in + * @out_fd will be performed (so it may be a pipe). + * + * WRITE_RESOURCE_FLAG_SOLID: + * Combine all the blobs into a single resource rather than writing + * them in separate resources. This flag is only valid if the WIM + * version number has been, or will be, set to WIM_VERSION_SOLID. + * This flag may not be combined with WRITE_RESOURCE_FLAG_PIPABLE. + * + * @out_ctype + * Compression format to use in the output resources, specified as one of + * the WIMLIB_COMPRESSION_TYPE_* constants. WIMLIB_COMPRESSION_TYPE_NONE + * is allowed. + * + * @out_chunk_size + * Compression chunk size to use in the output resources. It must be a + * valid chunk size for the specified compression format @out_ctype, unless + * @out_ctype is WIMLIB_COMPRESSION_TYPE_NONE, in which case this parameter + * is ignored. + * + * @num_threads + * Number of threads to use to compress data. If 0, a default number of + * threads will be chosen. The number of threads still may be decreased + * from the specified value if insufficient memory is detected. + * + * @blob_table + * If on-the-fly deduplication of unhashed blobs is desired, this parameter + * must be pointer to the blob table for the WIMStruct on whose behalf the + * blobs are being written. Otherwise, this parameter can be NULL. + * + * @filter_ctx + * If on-the-fly deduplication of unhashed blobs is desired, this parameter + * can be a pointer to a context for blob filtering used to detect whether + * the duplicate blob has been hard-filtered or not. If no blobs are + * hard-filtered or no blobs are unhashed, this parameter can be NULL. + * + * This function will write the blobs in @blob_list to resources in + * consecutive positions in the output WIM file, or to a single solid resource + * if WRITE_RESOURCE_FLAG_SOLID was specified in @write_resource_flags. In both + * cases, the @out_reshdr of the `struct blob_descriptor' for each blob written will be + * updated to specify its location, size, and flags in the output WIM. In the + * solid resource case, WIM_RESHDR_FLAG_SOLID will be set in the @flags field of + * each @out_reshdr, and furthermore @out_res_offset_in_wim and + * @out_res_size_in_wim of each @out_reshdr will be set to the offset and size, + * respectively, in the output WIM of the solid resource containing the + * corresponding blob. + * + * Each of the blobs to write may be in any location supported by the + * resource-handling code (specifically, read_blob_list()), such as the contents + * of external file that has been logically added to the output WIM, or a blob + * in another WIM file that has been imported, or even a blob in the "same" WIM + * file of which a modified copy is being written. In the case that a blob is + * already in a WIM file and uses compatible compression parameters, by default + * this function will re-use the raw data instead of decompressing it, then + * recompressing it; however, with WRITE_RESOURCE_FLAG_RECOMPRESS + * specified in @write_resource_flags, this is not done. + * + * As a further requirement, this function requires that the + * @will_be_in_output_wim member be set to 1 on all blobs in @blob_list as well + * as any other blobs not in @blob_list that will be in the output WIM file, but + * set to 0 on any other blobs in the output WIM's blob table or sharing a solid + * resource with a blob in @blob_list. Still furthermore, if on-the-fly + * deduplication of blobs is possible, then all blobs in @blob_list must also be + * linked by @blob_table_list along with any other blobs that have + * @will_be_in_output_wim set. + * + * This function handles on-the-fly deduplication of blobs for which SHA-1 + * message digests have not yet been calculated. Such blobs may or may not need + * to be written. If @blob_table is non-NULL, then each blob in @blob_list that + * has @unhashed set but not @unique_size set is checksummed immediately before + * it would otherwise be read for writing in order to determine if it is + * identical to another blob already being written or one that would be filtered + * out of the output WIM using blob_filtered() with the context @filter_ctx. + * Each such duplicate blob will be removed from @blob_list, its reference count + * transferred to the pre-existing duplicate blob, its memory freed, and will + * not be written. Alternatively, if a blob in @blob_list is a duplicate with + * any blob in @blob_table that has not been marked for writing or would not be + * hard-filtered, it is freed and the pre-existing duplicate is written instead, + * taking ownership of the reference count and slot in the @blob_table_list. + * + * Returns 0 if every blob was either written successfully or did not need to be + * written; otherwise returns a non-zero error code. */ -int write_wim_resource(struct lookup_table_entry *lte, - FILE *out_fp, int out_ctype, - struct resource_entry *out_res_entry, - int flags) -{ - u64 bytes_remaining; - u64 original_size; - u64 old_compressed_size; - u64 new_compressed_size; - u64 offset; +static int +write_blob_list(struct list_head *blob_list, + struct filedes *out_fd, + int write_resource_flags, + int out_ctype, + u32 out_chunk_size, + unsigned num_threads, + struct blob_table *blob_table, + struct filter_context *filter_ctx, + wimlib_progress_func_t progfunc, + void *progctx) +{ 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); + struct write_blobs_ctx ctx; + struct list_head raw_copy_blobs; + u64 num_nonraw_bytes; - /* Original size of the resource */ - original_size = wim_resource_size(lte); + wimlib_assert((write_resource_flags & + (WRITE_RESOURCE_FLAG_SOLID | + WRITE_RESOURCE_FLAG_PIPABLE)) != + (WRITE_RESOURCE_FLAG_SOLID | + WRITE_RESOURCE_FLAG_PIPABLE)); - /* Compressed size of the resource (as it exists now) */ - old_compressed_size = wim_resource_compressed_size(lte); - - /* Current offset in output file */ - file_offset = ftello(out_fp); - if (file_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset in output " - "stream"); - return WIMLIB_ERR_WRITE; - } + validate_blob_list(blob_list); - /* 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)); + if (list_empty(blob_list)) + return 0; - if (raw) { - flags |= WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = old_compressed_size; - } else { - flags &= ~WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = original_size; - } + /* If needed, set auxiliary information so that we can detect when the + * library has finished using each external file. */ + if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)) + init_done_with_file_info(blob_list); - /* Empty resource; nothing needs to be done, so just return success. */ - if (bytes_remaining == 0) - return 0; + memset(&ctx, 0, sizeof(ctx)); - /* Buffer for reading chunks for the resource */ - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; + ctx.out_fd = out_fd; + ctx.blob_table = blob_table; + ctx.out_ctype = out_ctype; + ctx.out_chunk_size = out_chunk_size; + ctx.write_resource_flags = write_resource_flags; + ctx.filter_ctx = filter_ctx; - /* 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; - } + /* + * We normally sort the blobs to write by a "sequential" order that is + * optimized for reading. But when using solid compression, we instead + * sort the blobs by file extension and file name (when applicable; and + * we don't do this for blobs from solid resources) so that similar + * files are grouped together, which improves the compression ratio. + * This is somewhat of a hack since a blob does not necessarily + * correspond one-to-one with a filename, nor is there any guarantee + * that two files with similar names or extensions are actually similar + * in content. A potential TODO is to sort the blobs based on some + * measure of similarity of their actual contents. + */ - /* 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; + ret = sort_blob_list_by_sequential_order(blob_list, + offsetof(struct blob_descriptor, + write_blobs_list)); + if (ret) + return ret; - /* 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; + ret = compute_blob_list_stats(blob_list, &ctx); + if (ret) + return ret; - /* 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; - } 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; + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID_SORT) { + ret = sort_blob_list_for_solid_compression(blob_list); + if (unlikely(ret)) + WARNING("Failed to sort blobs for solid compression. Continuing anyways."); + } + + ctx.progress_data.progfunc = progfunc; + ctx.progress_data.progctx = progctx; + + num_nonraw_bytes = find_raw_copy_blobs(blob_list, write_resource_flags, + out_ctype, out_chunk_size, + &raw_copy_blobs); + + /* Unless no data needs to be compressed, allocate a chunk_compressor to + * do compression. There are serial and parallel implementations of the + * chunk_compressor interface. We default to parallel using the + * specified number of threads, unless the upper bound on the number + * bytes needing to be compressed is less than a heuristic value. */ + if (num_nonraw_bytes != 0 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + #ifdef ENABLE_MULTITHREADED_COMPRESSION + if (num_nonraw_bytes > max(2000000, out_chunk_size)) { + ret = new_parallel_chunk_compressor(out_ctype, + out_chunk_size, + num_threads, 0, + &ctx.compressor); + if (ret > 0) { + WARNING("Couldn't create parallel chunk compressor: %"TS".\n" + " Falling back to single-threaded compression.", + wimlib_get_error_string(ret)); + } } - } + #endif - /* Verify SHA1 message digest of the resource, unless we are doing a raw - * write (in which case we never even saw the uncompressed data). Or, - * if the hash we had before is all 0's, just re-set it to be the new - * hash. */ - if (!raw) { - u8 md[SHA1_HASH_SIZE]; - sha1_final(md, &ctx); - if (is_zero_hash(lte->hash)) { - copy_hash(lte->hash, md); - } else if (!hashes_equal(md, lte->hash)) { - ERROR("WIM resource has incorrect hash!"); - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; maybe it changed " - "while we were reading it.", - lte->file_on_disk); - } - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out_fclose; + if (ctx.compressor == NULL) { + ret = new_serial_chunk_compressor(out_ctype, out_chunk_size, + &ctx.compressor); + if (ret) + goto out_destroy_context; } } - 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. */ - ret = write_uncompressed_resource_and_truncate(lte, - out_fp, - file_offset, - out_res_entry); - if (ret != 0) - goto out_fclose; - } else { - if (out_res_entry) { - out_res_entry->size = new_compressed_size; - out_res_entry->original_size = original_size; - out_res_entry->offset = file_offset; - out_res_entry->flags = lte->resource_entry.flags - & ~WIM_RESHDR_FLAG_COMPRESSED; - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; + if (ctx.compressor) + ctx.progress_data.progress.write_streams.num_threads = ctx.compressor->num_threads; + else + ctx.progress_data.progress.write_streams.num_threads = 1; + + ret = call_progress(ctx.progress_data.progfunc, + WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + &ctx.progress_data.progress, + ctx.progress_data.progctx); + if (ret) + goto out_destroy_context; + + /* Copy any compressed resources for which the raw data can be reused + * without decompression. */ + ret = write_raw_copy_resources(&raw_copy_blobs, ctx.out_fd, + &ctx.progress_data); + + if (ret || num_nonraw_bytes == 0) + goto out_destroy_context; + + INIT_LIST_HEAD(&ctx.blobs_being_compressed); + + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + + INIT_LIST_HEAD(&ctx.blobs_in_solid_resource); + + ret = begin_write_resource(&ctx, num_nonraw_bytes); + if (ret) + goto out_destroy_context; + } + + /* Read the list of blobs needing to be compressed, using the specified + * callbacks to execute processing of the data. */ + + struct read_blob_callbacks cbs = { + .begin_blob = write_blob_begin_read, + .consume_chunk = write_blob_process_chunk, + .end_blob = write_blob_end_read, + .ctx = &ctx, + }; + + ret = read_blob_list(blob_list, + offsetof(struct blob_descriptor, write_blobs_list), + &cbs, + BLOB_LIST_ALREADY_SORTED | + VERIFY_BLOB_HASHES | + COMPUTE_MISSING_BLOB_HASHES); + + if (ret) + goto out_destroy_context; + + ret = finish_remaining_chunks(&ctx); + if (ret) + goto out_destroy_context; + + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + struct wim_reshdr reshdr; + struct blob_descriptor *blob; + u64 offset_in_res; + + ret = end_write_resource(&ctx, &reshdr); + if (ret) + goto out_destroy_context; + + offset_in_res = 0; + list_for_each_entry(blob, &ctx.blobs_in_solid_resource, write_blobs_list) { + blob->out_reshdr.size_in_wim = blob->size; + blob->out_reshdr.flags = reshdr_flags_for_blob(blob) | + WIM_RESHDR_FLAG_SOLID; + blob->out_reshdr.uncompressed_size = 0; + blob->out_reshdr.offset_in_wim = offset_in_res; + blob->out_res_offset_in_wim = reshdr.offset_in_wim; + blob->out_res_size_in_wim = reshdr.size_in_wim; + blob->out_res_uncompressed_size = reshdr.uncompressed_size; + offset_in_res += blob->size; } + wimlib_assert(offset_in_res == reshdr.uncompressed_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_destroy_context: + FREE(ctx.chunk_csizes); + if (ctx.compressor) + ctx.compressor->destroy(ctx.compressor); return ret; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION -struct shared_queue { - unsigned size; - unsigned front; - unsigned back; - unsigned filled_slots; - void **array; - pthread_mutex_t lock; - pthread_cond_t msg_avail_cond; - pthread_cond_t space_avail_cond; -}; -static int shared_queue_init(struct shared_queue *q, unsigned size) +static int +write_file_data_blobs(WIMStruct *wim, + struct list_head *blob_list, + int write_flags, + unsigned num_threads, + struct filter_context *filter_ctx) { - q->array = CALLOC(sizeof(q->array[0]), size); - if (!q->array) - return WIMLIB_ERR_NOMEM; - 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; + int out_ctype; + u32 out_chunk_size; + int write_resource_flags; + + write_resource_flags = write_flags_to_resource_flags(write_flags); + + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { + out_chunk_size = wim->out_solid_chunk_size; + out_ctype = wim->out_solid_compression_type; + } else { + out_chunk_size = wim->out_chunk_size; + out_ctype = wim->out_compression_type; + } + + return write_blob_list(blob_list, + &wim->out_fd, + write_resource_flags, + out_ctype, + out_chunk_size, + num_threads, + wim->blob_table, + filter_ctx, + wim->progfunc, + wim->progctx); } -static void shared_queue_destroy(struct shared_queue *q) +/* Write the contents of the specified blob as a WIM resource. */ +static int +write_wim_resource(struct blob_descriptor *blob, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + int write_resource_flags) { - FREE(q->array); - pthread_mutex_destroy(&q->lock); - pthread_cond_destroy(&q->msg_avail_cond); - pthread_cond_destroy(&q->space_avail_cond); + LIST_HEAD(blob_list); + list_add(&blob->write_blobs_list, &blob_list); + blob->will_be_in_output_wim = 1; + return write_blob_list(&blob_list, + out_fd, + write_resource_flags & ~WRITE_RESOURCE_FLAG_SOLID, + out_ctype, + out_chunk_size, + 1, + NULL, + NULL, + NULL, + NULL); } -static void shared_queue_put(struct shared_queue *q, void *obj) +/* Write the contents of the specified buffer as a WIM resource. */ +int +write_wim_resource_from_buffer(const void *buf, + size_t buf_size, + bool is_metadata, + struct filedes *out_fd, + int out_ctype, + u32 out_chunk_size, + struct wim_reshdr *out_reshdr, + u8 *hash_ret, + int write_resource_flags) { - pthread_mutex_lock(&q->lock); - while (q->filled_slots == q->size) - pthread_cond_wait(&q->space_avail_cond, &q->lock); + int ret; + struct blob_descriptor blob; + + if (unlikely(buf_size == 0)) { + zero_reshdr(out_reshdr); + if (hash_ret) + copy_hash(hash_ret, zero_hash); + return 0; + } + + blob_set_is_located_in_attached_buffer(&blob, (void *)buf, buf_size); + sha1_buffer(buf, buf_size, blob.hash); + blob.unhashed = 0; + blob.is_metadata = is_metadata; + + ret = write_wim_resource(&blob, out_fd, out_ctype, out_chunk_size, + write_resource_flags); + if (ret) + return ret; - q->back = (q->back + 1) % q->size; - q->array[q->back] = obj; - q->filled_slots++; + copy_reshdr(out_reshdr, &blob.out_reshdr); - pthread_cond_broadcast(&q->msg_avail_cond); - pthread_mutex_unlock(&q->lock); + if (hash_ret) + copy_hash(hash_ret, blob.hash); + return 0; } -static void *shared_queue_get(struct shared_queue *q) +struct blob_size_table { + struct hlist_head *array; + size_t num_entries; + size_t capacity; +}; + +static int +init_blob_size_table(struct blob_size_table *tab, size_t capacity) { - void *obj; + tab->array = CALLOC(capacity, sizeof(tab->array[0])); + if (tab->array == NULL) + return WIMLIB_ERR_NOMEM; + tab->num_entries = 0; + tab->capacity = capacity; + return 0; +} - pthread_mutex_lock(&q->lock); - while (q->filled_slots == 0) - pthread_cond_wait(&q->msg_avail_cond, &q->lock); +static void +destroy_blob_size_table(struct blob_size_table *tab) +{ + FREE(tab->array); +} - obj = q->array[q->front]; - q->array[q->front] = NULL; - q->front = (q->front + 1) % q->size; - q->filled_slots--; +static int +blob_size_table_insert(struct blob_descriptor *blob, void *_tab) +{ + struct blob_size_table *tab = _tab; + size_t pos; + struct blob_descriptor *same_size_blob; + + pos = hash_u64(blob->size) % tab->capacity; + blob->unique_size = 1; + hlist_for_each_entry(same_size_blob, &tab->array[pos], hash_list_2) { + if (same_size_blob->size == blob->size) { + blob->unique_size = 0; + same_size_blob->unique_size = 0; + break; + } + } - pthread_cond_broadcast(&q->space_avail_cond); - pthread_mutex_unlock(&q->lock); - return obj; + hlist_add_head(&blob->hash_list_2, &tab->array[pos]); + tab->num_entries++; + return 0; } -struct compressor_thread_params { - struct shared_queue *res_to_compress_queue; - struct shared_queue *compressed_res_queue; - compress_func_t compress; +struct find_blobs_ctx { + WIMStruct *wim; + int write_flags; + struct list_head blob_list; + struct blob_size_table blob_size_tab; }; -#define MAX_CHUNKS_PER_MSG 2 - -struct message { - struct lookup_table_entry *lte; - u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; - u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG]; - u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; - unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; - unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; - unsigned num_chunks; - struct list_head list; - bool complete; - u64 begin_chunk; -}; +static void +reference_blob_for_write(struct blob_descriptor *blob, + struct list_head *blob_list, u32 nref) +{ + if (!blob->will_be_in_output_wim) { + blob->out_refcnt = 0; + list_add_tail(&blob->write_blobs_list, blob_list); + blob->will_be_in_output_wim = 1; + } + blob->out_refcnt += nref; +} + +static int +fully_reference_blob_for_write(struct blob_descriptor *blob, void *_blob_list) +{ + struct list_head *blob_list = _blob_list; + blob->will_be_in_output_wim = 0; + reference_blob_for_write(blob, blob_list, blob->refcnt); + return 0; +} -static void compress_chunks(struct message *msg, compress_func_t compress) +static int +inode_find_blobs_to_reference(const struct wim_inode *inode, + const struct blob_table *table, + struct list_head *blob_list) { - for (unsigned i = 0; i < msg->num_chunks; i++) { - DEBUG2("compress chunk %u of %u", i, msg->num_chunks); - int ret = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i], - &msg->compressed_chunk_sizes[i]); - if (ret == 0) { - msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + wimlib_assert(inode->i_nlink > 0); + + for (unsigned i = 0; i < inode->i_num_streams; i++) { + struct blob_descriptor *blob; + const u8 *hash; + + blob = stream_blob(&inode->i_streams[i], table); + if (blob) { + reference_blob_for_write(blob, blob_list, inode->i_nlink); } else { - msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; - msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; + hash = stream_hash(&inode->i_streams[i]); + if (!is_zero_hash(hash)) + return blob_not_found_error(inode, hash); } } + return 0; +} + +static int +do_blob_set_not_in_output_wim(struct blob_descriptor *blob, void *_ignore) +{ + blob->will_be_in_output_wim = 0; + return 0; } -static void *compressor_thread_proc(void *arg) +static int +image_find_blobs_to_reference(WIMStruct *wim) { - struct compressor_thread_params *params = arg; - struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; - struct shared_queue *compressed_res_queue = params->compressed_res_queue; - compress_func_t compress = params->compress; - struct message *msg; + struct wim_image_metadata *imd; + struct wim_inode *inode; + struct blob_descriptor *blob; + struct list_head *blob_list; + int ret; + + imd = wim_get_current_image_metadata(wim); + + image_for_each_unhashed_blob(blob, imd) + blob->will_be_in_output_wim = 0; - DEBUG("Compressor thread ready"); - while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { - compress_chunks(msg, compress); - shared_queue_put(compressed_res_queue, msg); + blob_list = wim->private; + image_for_each_inode(inode, imd) { + ret = inode_find_blobs_to_reference(inode, + wim->blob_table, + blob_list); + if (ret) + return ret; } - DEBUG("Compressor thread terminating"); - return NULL; + return 0; } -#endif -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 +prepare_unfiltered_list_of_blobs_in_output_wim(WIMStruct *wim, + int image, + int blobs_ok, + struct list_head *blob_list_ret) { int ret; - struct lookup_table_entry *lte, *tmp; - - 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) - 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); + + INIT_LIST_HEAD(blob_list_ret); + + if (blobs_ok && (image == WIMLIB_ALL_IMAGES || + (image == 1 && wim->hdr.image_count == 1))) + { + /* Fast case: Assume that all blobs are being written and that + * the reference counts are correct. */ + struct blob_descriptor *blob; + struct wim_image_metadata *imd; + unsigned i; + + for_blob_in_table(wim->blob_table, + fully_reference_blob_for_write, + blob_list_ret); + + for (i = 0; i < wim->hdr.image_count; i++) { + imd = wim->image_metadata[i]; + image_for_each_unhashed_blob(blob, imd) + fully_reference_blob_for_write(blob, blob_list_ret); } + } else { + /* Slow case: Walk through the images being written and + * determine the blobs referenced. */ + for_blob_in_table(wim->blob_table, + do_blob_set_not_in_output_wim, NULL); + wim->private = blob_list_ret; + ret = for_image(wim, image, image_find_blobs_to_reference); + if (ret) + return ret; } + 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) +struct insert_other_if_hard_filtered_ctx { + struct blob_size_table *tab; + struct filter_context *filter_ctx; +}; + +static int +insert_other_if_hard_filtered(struct blob_descriptor *blob, void *_ctx) { - int write_resource_flags; + struct insert_other_if_hard_filtered_ctx *ctx = _ctx; - if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS; - else - write_resource_flags = 0; - 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, - out_ctype, progress_func, - progress, write_resource_flags); + if (!blob->will_be_in_output_wim && + blob_hard_filtered(blob, ctx->filter_ctx)) + blob_size_table_insert(blob, ctx->tab); + return 0; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION -static int write_wim_chunks(struct message *msg, FILE *out_fp, - struct chunk_table *chunk_tab) +static int +determine_blob_size_uniquity(struct list_head *blob_list, + struct blob_table *lt, + struct filter_context *filter_ctx) { - for (unsigned i = 0; i < msg->num_chunks; i++) { - unsigned chunk_csize = msg->compressed_chunk_sizes[i]; - - DEBUG2("Write wim chunk %u of %u (csize = %u)", - i, msg->num_chunks, chunk_csize); + int ret; + struct blob_size_table tab; + struct blob_descriptor *blob; - if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) - != chunk_csize) - { - ERROR_WITH_ERRNO("Failed to write WIM chunk"); - return WIMLIB_ERR_WRITE; - } + ret = init_blob_size_table(&tab, 9001); + if (ret) + return ret; - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += chunk_csize; + if (may_hard_filter_blobs(filter_ctx)) { + struct insert_other_if_hard_filtered_ctx ctx = { + .tab = &tab, + .filter_ctx = filter_ctx, + }; + for_blob_in_table(lt, insert_other_if_hard_filtered, &ctx); } + + list_for_each_entry(blob, blob_list, write_blobs_list) + blob_size_table_insert(blob, &tab); + + destroy_blob_size_table(&tab); return 0; } -/* - * This function is executed by the main thread when the resources are being - * compressed in parallel. The main thread is in change of all reading of the - * uncompressed data and writing of the compressed data. The compressor threads - * *only* do compression from/to in-memory buffers. - * - * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG - * chunks of compressed data to compress, represented in a `struct message'. - * Each message is passed from the main thread to a worker thread through the - * res_to_compress_queue, and it is passed back through the - * compressed_res_queue. - */ -static int 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 void +filter_blob_list_for_write(struct list_head *blob_list, + struct filter_context *filter_ctx) { - int ret; + struct blob_descriptor *blob, *tmp; - struct message msgs[queue_size]; - ZERO_ARRAY(msgs); - - // Initially, all the messages are available to use. - LIST_HEAD(available_msgs); - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) - list_add(&msgs[i].list, &available_msgs); - - // outstanding_resources is the list of resources that currently have - // had chunks sent off for compression. - // - // The first stream in outstanding_resources is the stream that is - // currently being written (cur_lte). - // - // The last stream in outstanding_resources is the stream that is - // currently being read and chunks fed to the compressor threads - // (next_lte). - // - // Depending on the number of threads and the sizes of the resource, - // the outstanding streams list may contain streams between cur_lte and - // next_lte that have all their chunks compressed or being compressed, - // but haven't been written yet. - // - LIST_HEAD(outstanding_resources); - struct list_head *next_resource = stream_list->next; - struct lookup_table_entry *next_lte = NULL; - u64 next_chunk = 0; - u64 next_num_chunks = 0; - - // As in write_wim_resource(), each resource we read is checksummed. - SHA_CTX 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 = NULL; - struct chunk_table *cur_chunk_tab = NULL; - struct message *msg; - -#ifdef WITH_NTFS_3G - ntfs_inode *ni = NULL; -#endif + list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) { + int status = blob_filtered(blob, filter_ctx); - DEBUG("Initializing buffers for uncompressed " - "and compressed data (%zu bytes needed)", - queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2); - - // Pre-allocate all the buffers that will be needed to do the chunk - // compression. - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { - for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { - msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); - - // 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; + if (status == 0) { + /* Not filtered. */ + continue; + } else { + if (status > 0) { + /* Soft filtered. */ + } else { + /* Hard filtered. */ + blob->will_be_in_output_wim = 0; + list_del(&blob->blob_table_list); } + list_del(&blob->write_blobs_list); } } +} - // This loop is executed until all resources have been written, except - // possibly a few that have been added to the @my_resources list for - // writing later. - while (1) { - // Send chunks to the compressor threads until either (a) there - // are no more messages available since they were all sent off, - // or (b) there are no more resources that need to be - // compressed. - while (!list_empty(&available_msgs)) { - 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; - } - } +/* + * prepare_blob_list_for_write() - + * + * Prepare the list of blobs to write for writing a WIM containing the specified + * image(s) with the specified write flags. + * + * @wim + * The WIMStruct on whose behalf the write is occurring. + * + * @image + * Image(s) from the WIM to write; may be WIMLIB_ALL_IMAGES. + * + * @write_flags + * WIMLIB_WRITE_FLAG_* flags for the write operation: + * + * STREAMS_OK: For writes of all images, assume that all blobs in the blob + * table of @wim and the per-image lists of unhashed blobs should be taken + * as-is, and image metadata should not be searched for references. This + * does not exclude filtering with APPEND and SKIP_EXTERNAL_WIMS, below. + * + * APPEND: Blobs already present in @wim shall not be returned in + * @blob_list_ret. + * + * SKIP_EXTERNAL_WIMS: Blobs already present in a WIM file, but not @wim, + * shall be returned in neither @blob_list_ret nor @blob_table_list_ret. + * + * @blob_list_ret + * List of blobs, linked by write_blobs_list, that need to be written will + * be returned here. + * + * Note that this function assumes that unhashed blobs will be written; it + * does not take into account that they may become duplicates when actually + * hashed. + * + * @blob_table_list_ret + * List of blobs, linked by blob_table_list, that need to be included in + * the WIM's blob table will be returned here. This will be a superset of + * the blobs in @blob_list_ret. + * + * This list will be a proper superset of @blob_list_ret if and only if + * WIMLIB_WRITE_FLAG_APPEND was specified in @write_flags and some of the + * blobs that would otherwise need to be written were already located in + * the WIM file. + * + * All blobs in this list will have @out_refcnt set to the number of + * references to the blob in the output WIM. If + * WIMLIB_WRITE_FLAG_STREAMS_OK was specified in @write_flags, @out_refcnt + * may be as low as 0. + * + * @filter_ctx_ret + * A context for queries of blob filter status with blob_filtered() is + * returned in this location. + * + * In addition, @will_be_in_output_wim will be set to 1 in all blobs inserted + * into @blob_table_list_ret and to 0 in all blobs in the blob table of @wim not + * inserted into @blob_table_list_ret. + * + * Still furthermore, @unique_size will be set to 1 on all blobs in + * @blob_list_ret that have unique size among all blobs in @blob_list_ret and + * among all blobs in the blob table of @wim that are ineligible for being + * written due to filtering. + * + * Returns 0 on success; nonzero on read error, memory allocation error, or + * otherwise. + */ +static int +prepare_blob_list_for_write(WIMStruct *wim, int image, + int write_flags, + struct list_head *blob_list_ret, + struct list_head *blob_table_list_ret, + struct filter_context *filter_ctx_ret) +{ + int ret; + struct blob_descriptor *blob; - // 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; - } - next_lte = container_of(next_resource, - struct 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) - cur_lte = next_lte; - break; - } - } - } + filter_ctx_ret->write_flags = write_flags; + filter_ctx_ret->wim = wim; - if (next_lte == NULL) - break; + ret = prepare_unfiltered_list_of_blobs_in_output_wim( + wim, + image, + write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK, + blob_list_ret); + if (ret) + return ret; - // Get a message from the available messages - // list - msg = container_of(available_msgs.next, - struct message, - list); + INIT_LIST_HEAD(blob_table_list_ret); + list_for_each_entry(blob, blob_list_ret, write_blobs_list) + list_add_tail(&blob->blob_table_list, blob_table_list_ret); - // ... and delete it from the available messages - // list - list_del(&msg->list); + ret = determine_blob_size_uniquity(blob_list_ret, wim->blob_table, + filter_ctx_ret); + if (ret) + return ret; - // Initialize the message with the chunks to - // compress. - msg->num_chunks = min(next_num_chunks - next_chunk, - MAX_CHUNKS_PER_MSG); - msg->lte = next_lte; - msg->complete = false; - msg->begin_chunk = next_chunk; + if (may_filter_blobs(filter_ctx_ret)) + filter_blob_list_for_write(blob_list_ret, filter_ctx_ret); - unsigned size = WIM_CHUNK_SIZE; - for (unsigned i = 0; i < msg->num_chunks; i++) { + return 0; +} - // Read chunk @next_chunk of the stream into the - // message so that a compressor thread can - // compress it. +static int +write_file_data(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + struct list_head *blob_list_override, + struct list_head *blob_table_list_ret) +{ + int ret; + struct list_head _blob_list; + struct list_head *blob_list; + struct blob_descriptor *blob; + struct filter_context _filter_ctx; + struct filter_context *filter_ctx; + + if (blob_list_override == NULL) { + /* Normal case: prepare blob list from image(s) being written. + */ + blob_list = &_blob_list; + filter_ctx = &_filter_ctx; + ret = prepare_blob_list_for_write(wim, image, write_flags, + blob_list, + blob_table_list_ret, + filter_ctx); + if (ret) + return ret; + } else { + /* Currently only as a result of wimlib_split() being called: + * use blob list already explicitly provided. Use existing + * reference counts. */ + blob_list = blob_list_override; + filter_ctx = NULL; + INIT_LIST_HEAD(blob_table_list_ret); + list_for_each_entry(blob, blob_list, write_blobs_list) { + blob->out_refcnt = blob->refcnt; + blob->will_be_in_output_wim = 1; + blob->unique_size = 0; + list_add_tail(&blob->blob_table_list, blob_table_list_ret); + } + } - 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; - } + return write_file_data_blobs(wim, + blob_list, + write_flags, + num_threads, + filter_ctx); +} +static int +write_metadata_resources(WIMStruct *wim, int image, int write_flags) +{ + int ret; + int start_image; + int end_image; + int write_resource_flags; - DEBUG2("Read resource (size=%u, offset=%zu)", - size, next_chunk * WIM_CHUNK_SIZE); + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) + return 0; - msg->uncompressed_chunk_sizes[i] = size; + write_resource_flags = write_flags_to_resource_flags(write_flags); - ret = read_wim_resource(next_lte, - msg->uncompressed_chunks[i], - size, - next_chunk * WIM_CHUNK_SIZE, - 0); - if (ret != 0) - goto out; - sha1_update(&next_sha_ctx, - msg->uncompressed_chunks[i], size); - next_chunk++; - } + write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID; - // Send the compression request - list_add_tail(&msg->list, &next_lte->msg_list); - shared_queue_put(res_to_compress_queue, msg); - DEBUG2("Compression request sent"); - } + ret = call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, + NULL, wim->progctx); + if (ret) + return ret; - // If there are no outstanding resources, there are no more - // resources that need to be written. - if (list_empty(&outstanding_resources)) { + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; + } else { + start_image = image; + end_image = image; + } + + for (int i = start_image; i <= end_image; i++) { + struct wim_image_metadata *imd; + + imd = wim->image_metadata[i - 1]; + if (is_image_dirty(imd)) { + /* The image was modified from the original, or was + * newly added, so we have to build and write a new + * metadata resource. */ + ret = write_metadata_resource(wim, i, + write_resource_flags); + } else if (is_image_unchanged_from_wim(imd, wim) && + (write_flags & (WIMLIB_WRITE_FLAG_UNSAFE_COMPACT | + WIMLIB_WRITE_FLAG_APPEND))) + { + /* The metadata resource is already in the WIM file. + * For appends, we don't need to write it at all. For + * compactions, we re-write existing metadata resources + * along with the existing file resources, not here. */ + if (write_flags & WIMLIB_WRITE_FLAG_APPEND) + blob_set_out_reshdr_for_reuse(imd->metadata_blob); ret = 0; - goto out; + } else { + /* The metadata resource is in a WIM file other than the + * one being written to. We need to rewrite it, + * possibly compressed differently; but rebuilding the + * metadata itself isn't necessary. */ + ret = write_wim_resource(imd->metadata_blob, + &wim->out_fd, + wim->out_compression_type, + wim->out_chunk_size, + write_resource_flags); } + if (ret) + return ret; + } - // Get the next message from the queue and process it. - // The message will contain 1 or more data chunks that have been - // compressed. - msg = shared_queue_get(compressed_res_queue); - msg->complete = true; - - // Is this the next chunk in the current resource? If it's not - // (i.e., an earlier chunk in a same or different resource - // hasn't been compressed yet), do nothing, and keep this - // message around until all earlier chunks are received. - // - // Otherwise, write all the chunks we can. - while (cur_lte != NULL && !list_empty(&cur_lte->msg_list) - && (msg = container_of(cur_lte->msg_list.next, - struct message, - list))->complete) - { - DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk); - if (msg->begin_chunk == 0) { - DEBUG2("Begin chunk tab"); - - // This is the first set of chunks. Leave space - // for the chunk table in the output file. - off_t cur_offset = ftello(out_fp); - if (cur_offset == -1) { - ret = WIMLIB_ERR_WRITE; - goto out; - } - ret = begin_wim_resource_chunk_tab(cur_lte, - out_fp, - cur_offset, - &cur_chunk_tab); - if (ret != 0) - goto out; - } + return call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, + NULL, wim->progctx); +} - // Write the compressed chunks from the message. - ret = write_wim_chunks(msg, out_fp, cur_chunk_tab); - if (ret != 0) - goto out; +static int +open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) +{ + int 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; +} - list_del(&msg->list); +static int +close_wim_writable(WIMStruct *wim, int write_flags) +{ + int ret = 0; - // This message is available to use for different chunks - // now. - list_add(&msg->list, &available_msgs); + 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; +} - // Was this the last chunk of the stream? If so, - // finish it. - if (list_empty(&cur_lte->msg_list) && - msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks) - { - DEBUG2("Finish wim chunk tab"); - u64 res_csize; - ret = finish_wim_resource_chunk_tab(cur_chunk_tab, - out_fp, - &res_csize); - if (ret != 0) - goto out; - - 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; - } +static int +cmp_blobs_by_out_rdesc(const void *p1, const void *p2) +{ + const struct blob_descriptor *blob1, *blob2; - progress->write_streams.completed_bytes += - wim_resource_size(cur_lte); - progress->write_streams.completed_streams++; + blob1 = *(const struct blob_descriptor**)p1; + blob2 = *(const struct blob_descriptor**)p2; - if (progress_func) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - } + if (blob1->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) { + if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) { + if (blob1->out_res_offset_in_wim != blob2->out_res_offset_in_wim) + return cmp_u64(blob1->out_res_offset_in_wim, + blob2->out_res_offset_in_wim); + } else { + return 1; + } + } else { + if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) + return -1; + } + return cmp_u64(blob1->out_reshdr.offset_in_wim, + blob2->out_reshdr.offset_in_wim); +} - 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) { - if (next_lte == NULL) { - DEBUG("No more outstanding resources"); - ret = 0; - goto out; - } else { - DEBUG("No more outstanding resources---" - "but still more to compress!"); - cur_lte = NULL; - } - } else { - cur_lte = container_of(cur_lte->staging_list.next, - struct lookup_table_entry, - staging_list); - } +static int +write_blob_table(WIMStruct *wim, int image, int write_flags, + struct list_head *blob_table_list) +{ + int ret; - // Since we just finished writing a stream, - // write any streams that have been added to the - // my_resources list for direct writing by the - // main thread (e.g. resources that don't need - // to be compressed because the desired - // compression type is the same as the previous - // compression type). - ret = do_write_stream_list(&my_resources, - out_fp, - out_ctype, - progress_func, - progress, - 0); - if (ret != 0) - goto out; + /* Set output resource metadata for blobs already present in WIM. */ + if (write_flags & WIMLIB_WRITE_FLAG_APPEND) { + struct blob_descriptor *blob; + list_for_each_entry(blob, blob_table_list, blob_table_list) { + if (blob->blob_location == BLOB_IN_WIM && + blob->rdesc->wim == wim) + { + blob_set_out_reshdr_for_reuse(blob); } } } -out: - if (next_lte) { -#ifdef WITH_NTFS_3G - end_wim_resource_read(next_lte, ni); -#else - end_wim_resource_read(next_lte); -#endif - } + ret = sort_blob_list(blob_table_list, + offsetof(struct blob_descriptor, blob_table_list), + cmp_blobs_by_out_rdesc); + if (ret) + return ret; - 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; + /* Add entries for metadata resources. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) { + int start_image; + int end_image; - list_for_each(cur, &available_msgs) { - num_available_msgs++; + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; + } else { + start_image = image; + end_image = image; } - while (num_available_msgs < ARRAY_LEN(msgs)) { - shared_queue_get(compressed_res_queue); - num_available_msgs++; - } - } + /* Push metadata blob table entries onto the front of the list + * in reverse order, so that they're written in order. + */ + for (int i = end_image; i >= start_image; i--) { + struct blob_descriptor *metadata_blob; - 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]); + metadata_blob = wim->image_metadata[i - 1]->metadata_blob; + wimlib_assert(metadata_blob->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA); + metadata_blob->out_refcnt = 1; + list_add(&metadata_blob->blob_table_list, blob_table_list); } } - if (cur_chunk_tab != NULL) - FREE(cur_chunk_tab); - return ret; + return write_blob_table_from_blob_list(blob_table_list, + &wim->out_fd, + wim->out_hdr.part_number, + &wim->out_hdr.blob_table_reshdr, + write_flags_to_resource_flags(write_flags)); } - -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) +/* + * Finish writing a WIM file: write the blob table, xml data, and integrity + * table, then overwrite the WIM header. + * + * The output file descriptor is closed on success, except when writing to a + * user-specified file descriptor (WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR set). + */ +static int +finish_write(WIMStruct *wim, int image, int write_flags, + struct list_head *blob_table_list) { + int write_resource_flags; + off_t old_blob_table_end = 0; + struct integrity_table *old_integrity_table = NULL; + off_t new_blob_table_end; + u64 xml_totalbytes; int ret; - struct shared_queue res_to_compress_queue; - struct shared_queue compressed_res_queue; - pthread_t *compressor_threads = NULL; - - if (num_threads == 0) { - long nthreads = sysconf(_SC_NPROCESSORS_ONLN); - if (nthreads < 1) { - WARNING("Could not determine number of processors! Assuming 1"); - goto out_serial; - } else { - num_threads = nthreads; - } - } - - progress->write_streams.num_threads = num_threads; - wimlib_assert(stream_list->next != stream_list); - - static const double MESSAGES_PER_THREAD = 2.0; - size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); + write_resource_flags = write_flags_to_resource_flags(write_flags); - ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret != 0) - goto out_serial; + /* In the WIM header, there is room for the resource entry for a + * metadata resource labeled as the "boot metadata". This entry should + * be zeroed out if there is no bootable image (boot_idx 0). Otherwise, + * it should be a copy of the resource entry for the image that is + * marked as bootable. */ + if (wim->out_hdr.boot_idx == 0) { + zero_reshdr(&wim->out_hdr.boot_metadata_reshdr); + } else { + copy_reshdr(&wim->out_hdr.boot_metadata_reshdr, + &wim->image_metadata[ + wim->out_hdr.boot_idx - 1]->metadata_blob->out_reshdr); + } + + /* If appending to a WIM file containing an integrity table, we'd like + * to re-use the information in the old integrity table instead of + * recalculating it. But we might overwrite the old integrity table + * when we expand the XML data. Read it into memory just in case. */ + if ((write_flags & (WIMLIB_WRITE_FLAG_APPEND | + WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) == + (WIMLIB_WRITE_FLAG_APPEND | + WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) + && wim_has_integrity_table(wim)) + { + old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim + + wim->hdr.blob_table_reshdr.size_in_wim; + (void)read_integrity_table(wim, + old_blob_table_end - WIM_HEADER_DISK_SIZE, + &old_integrity_table); + /* If we couldn't read the old integrity table, we can still + * re-calculate the full integrity table ourselves. Hence the + * ignoring of the return value. */ + } + + /* Write blob table if needed. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)) { + ret = write_blob_table(wim, image, write_flags, + blob_table_list); + if (ret) + goto out; + } - ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret != 0) - goto out_destroy_res_to_compress_queue; + /* Write XML data. */ + xml_totalbytes = wim->out_fd.offset; + if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES) + xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING; + ret = write_wim_xml_data(wim, image, xml_totalbytes, + &wim->out_hdr.xml_data_reshdr, + write_resource_flags); + if (ret) + goto out; - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); + /* Write integrity table if needed. */ + if ((write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) && + wim->out_hdr.blob_table_reshdr.offset_in_wim != 0) + { + if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS) { + /* The XML data we wrote may have overwritten part of + * the old integrity table, so while calculating the new + * integrity table we should temporarily update the WIM + * header to remove the integrity table reference. */ + struct wim_header checkpoint_hdr; + memcpy(&checkpoint_hdr, &wim->out_hdr, sizeof(struct wim_header)); + zero_reshdr(&checkpoint_hdr.integrity_table_reshdr); + checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header(&checkpoint_hdr, &wim->out_fd, 0); + if (ret) + goto out; + } - compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + new_blob_table_end = wim->out_hdr.blob_table_reshdr.offset_in_wim + + wim->out_hdr.blob_table_reshdr.size_in_wim; - for (unsigned i = 0; i < num_threads; i++) { - DEBUG("pthread_create thread %u", i); - ret = pthread_create(&compressor_threads[i], NULL, - compressor_thread_proc, ¶ms); - if (ret != 0) { - ret = -1; - ERROR_WITH_ERRNO("Failed to create compressor " - "thread %u", i); - num_threads = i; - goto out_join; - } + ret = write_integrity_table(wim, + new_blob_table_end, + old_blob_table_end, + old_integrity_table); + if (ret) + goto out; + } else { + /* No integrity table. */ + zero_reshdr(&wim->out_hdr.integrity_table_reshdr); } - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); - - ret = main_writer_thread_proc(stream_list, - out_fp, - out_ctype, - &res_to_compress_queue, - &compressed_res_queue, - queue_size, - write_flags, - progress_func, - progress); -out_join: - for (unsigned i = 0; i < num_threads; i++) - shared_queue_put(&res_to_compress_queue, NULL); - - for (unsigned i = 0; i < num_threads; i++) { - if (pthread_join(compressor_threads[i], NULL)) { - WARNING("Failed to join compressor thread %u: %s", - i, strerror(errno)); + /* Now that all information in the WIM header has been determined, the + * preliminary header written earlier can be overwritten, the header of + * the existing WIM file can be overwritten, or the final header can be + * written to the end of the pipable WIM. */ + wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset); + else + ret = write_wim_header(&wim->out_hdr, &wim->out_fd, 0); + if (ret) + goto out; + + ret = WIMLIB_ERR_WRITE; + if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) { + /* Truncate any data the compaction freed up. */ + if (ftruncate(wim->out_fd.fd, wim->out_fd.offset)) { + ERROR_WITH_ERRNO("Failed to truncate the output WIM file"); + goto out; } } - FREE(compressor_threads); - shared_queue_destroy(&compressed_res_queue); -out_destroy_res_to_compress_queue: - shared_queue_destroy(&res_to_compress_queue); - if (ret >= 0 && ret != WIMLIB_ERR_NOMEM) - return ret; -out_serial: - WARNING("Falling back to single-threaded compression"); - return write_stream_list_serial(stream_list, - out_fp, - out_ctype, - write_flags, - progress_func, - progress); - -} -#endif - -/* - * 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) -{ - struct lookup_table_entry *lte; - size_t num_streams = 0; - u64 total_bytes = 0; - 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 (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); + /* Possibly sync file data to disk before closing. On POSIX systems, it + * is necessary to do this before using rename() to overwrite an + * existing file with a new file. Otherwise, data loss would occur if + * the system is abruptly terminated when the metadata for the rename + * operation has been written to disk, but the new file data has not. + */ + ret = WIMLIB_ERR_WRITE; + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + if (fsync(wim->out_fd.fd)) { + ERROR_WITH_ERRNO("Error syncing data to WIM file"); + goto out; } } - 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; - if (num_streams == 0) { - ret = 0; + ret = WIMLIB_ERR_WRITE; + if (close_wim_writable(wim, write_flags)) { + ERROR_WITH_ERRNO("Failed to close the output WIM file"); goto out; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION - if (total_compression_bytes >= 1000000 && num_threads != 1) { - ret = write_stream_list_parallel(stream_list, - out_fp, - out_ctype, - write_flags, - num_threads, - progress_func, - &progress); - } - else -#endif - { - ret = write_stream_list_serial(stream_list, - out_fp, - out_ctype, - write_flags, - progress_func, - &progress); - } + ret = 0; out: + free_integrity_table(old_integrity_table); return ret; } +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) -static int dentry_find_streams_to_write(struct dentry *dentry, - void *wim) +/* Set advisory lock on WIM file (if not already done so) */ +int +lock_wim_for_append(WIMStruct *wim) { - WIMStruct *w = wim; - struct list_head *stream_list = w->private; - struct lookup_table_entry *lte; - for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) { - lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table); - if (lte && ++lte->out_refcnt == 1) - list_add_tail(<e->staging_list, stream_list); + if (wim->locked_for_append) + return 0; + if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) { + wim->locked_for_append = 1; + return 0; } - return 0; -} - -static int find_streams_to_write(WIMStruct *w) -{ - return for_dentry_in_tree(wim_root_dentry(w), - dentry_find_streams_to_write, w); + if (errno != EWOULDBLOCK) + return 0; + return WIMLIB_ERR_ALREADY_LOCKED; } -static int write_wim_streams(WIMStruct *w, int image, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +/* Remove advisory lock on WIM file (if present) */ +void +unlock_wim_for_append(WIMStruct *wim) { - - 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); + if (wim->locked_for_append) { + flock(wim->in_fd.fd, LOCK_UN); + wim->locked_for_append = 0; + } } +#endif /* - * Finish writing a WIM file: write the lookup table, xml data, and integrity - * table (optional), then overwrite the WIM header. + * write_pipable_wim(): + * + * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM + * capable of being applied from a pipe). + * + * Pipable WIMs are a wimlib-specific modification of the WIM format such that + * images can be applied from them sequentially when the file data is sent over + * a pipe. In addition, a pipable WIM can be written sequentially to a pipe. + * The modifications made to the WIM format for pipable WIMs are: + * + * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead + * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also + * stops other software from trying to read the file as a normal WIM. + * + * - The header at the beginning of the file does not contain all the normal + * information; in particular it will have all 0's for the blob table and XML + * data resource entries. This is because this information cannot be + * determined until the blob table and XML data have been written. + * Consequently, wimlib will write the full header at the very end of the + * file. The header at the end, however, is only used when reading the WIM + * from a seekable file (not a pipe). + * + * - An extra copy of the XML data is placed directly after the header. This + * allows image names and sizes to be determined at an appropriate time when + * reading the WIM from a pipe. This copy of the XML data is ignored if the + * WIM is read from a seekable file (not a pipe). * - * write_flags is a bitwise OR of the following: + * - Solid resources are not allowed. Each blob is always stored in its own + * resource. * - * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: - * Include an integrity table. + * - The format of resources, or blobs, has been modified to allow them to be + * used before the "blob table" has been read. Each blob is prefixed with a + * `struct pwm_blob_hdr' that is basically an abbreviated form of `struct + * blob_descriptor_disk' that only contains the SHA-1 message digest, + * uncompressed blob size, and flags that indicate whether the blob is + * compressed. The data of uncompressed blobs then follows literally, while + * the data of compressed blobs follows in a modified format. Compressed + * blobs do not begin with a chunk table, since the chunk table cannot be + * written until all chunks have been compressed. Instead, each compressed + * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size. + * Furthermore, the chunk table is written at the end of the resource instead + * of the start. Note: chunk offsets are given in the chunk table as if the + * `struct pwm_chunk_hdr's were not present; also, the chunk table is only + * used if the WIM is being read from a seekable file (not a pipe). * - * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS: - * Show progress information when (if) writing the integrity table. + * - Metadata blobs always come before non-metadata blobs. (This does not by + * itself constitute an incompatibility with normal WIMs, since this is valid + * in normal WIMs.) * - * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: - * Don't write the lookup table. + * - At least up to the end of the blobs, all components must be packed as + * tightly as possible; there cannot be any "holes" in the WIM. (This does + * not by itself consititute an incompatibility with normal WIMs, since this + * is valid in normal WIMs.) * - * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: - * When (if) writing the integrity table, re-use entries from the - * existing integrity table, if possible. + * Note: the blob table, XML data, and header at the end are not used when + * applying from a pipe. They exist to support functionality such as image + * application and export when the WIM is *not* read from a pipe. * - * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: - * After writing the XML data but before writing the integrity - * table, write a temporary WIM header and flush the stream so that - * the WIM is less likely to become corrupted upon abrupt program - * termination. + * Layout of pipable WIM: * - * (private) WIMLIB_WRITE_FLAG_FSYNC: - * fsync() the output file before closing it. + * ---------+----------+--------------------+----------------+--------------+-----------+--------+ + * | Header | XML data | Metadata resources | File resources | Blob table | XML data | Header | + * ---------+----------+--------------------+----------------+--------------+-----------+--------+ * + * Layout of normal WIM: + * + * +--------+-----------------------------+-------------------------+ + * | Header | File and metadata resources | Blob table | XML data | + * +--------+-----------------------------+-------------------------+ + * + * An optional integrity table can follow the final XML data in both normal and + * pipable WIMs. However, due to implementation details, wimlib currently can + * only include an integrity table in a pipable WIM when writing it to a + * seekable file (not a pipe). + * + * Do note that since pipable WIMs are not supported by Microsoft's software, + * wimlib does not create them unless explicitly requested (with + * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic + * characters to identify the file. */ -int finish_write(WIMStruct *w, int image, int write_flags, - wimlib_progress_func_t progress_func) +static int +write_pipable_wim(WIMStruct *wim, int image, int write_flags, + unsigned num_threads, + struct list_head *blob_list_override, + struct list_head *blob_table_list_ret) { int ret; - struct wim_header hdr; - FILE *out = w->out_fp; - - /* @hdr will be the header for the new WIM. First copy all the data - * from the header in the WIMStruct; then set all the fields that may - * have changed, including the resource entries, boot index, and image - * count. */ - memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); - - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); - if (ret != 0) - goto out; - } + struct wim_reshdr xml_reshdr; + + WARNING("Creating a pipable WIM, which will " + "be incompatible\n" + " with Microsoft's software (WIMGAPI/ImageX/DISM)."); + + /* At this point, the header at the beginning of the file has already + * been written. */ + + /* For efficiency, when wimlib adds an image to the WIM with + * wimlib_add_image(), the SHA-1 message digests of files are not + * calculated; instead, they are calculated while the files are being + * written. However, this does not work when writing a pipable WIM, + * since when writing a blob to a pipable WIM, its SHA-1 message digest + * needs to be known before the blob data is written. Therefore, before + * getting much farther, we need to pre-calculate the SHA-1 message + * digests of all blobs that will be written. */ + ret = wim_checksum_unhashed_blobs(wim); + if (ret) + return ret; - ret = write_xml_data(w->wim_info, image, out, - (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? - wim_info_get_total_bytes(w->wim_info) : 0, - &hdr.xml_res_entry); - if (ret != 0) - goto out; + /* Write extra copy of the XML data. */ + ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, + &xml_reshdr, WRITE_RESOURCE_FLAG_PIPABLE); + if (ret) + return ret; - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { - struct wim_header checkpoint_hdr; - memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); - memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); - if (fseeko(out, 0, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to beginning " - "of WIM being written"); - ret = WIMLIB_ERR_WRITE; - goto out; - } - ret = write_header(&checkpoint_hdr, out); - if (ret != 0) - goto out; + /* Write metadata resources for the image(s) being included in the + * output WIM. */ + ret = write_metadata_resources(wim, image, write_flags); + if (ret) + return ret; - if (fflush(out) != 0) { - ERROR_WITH_ERRNO("Can't write data to WIM"); - ret = WIMLIB_ERR_WRITE; - goto out; - } + /* Write file data needed for the image(s) being included in the output + * WIM, or file data needed for the split WIM part. */ + return write_file_data(wim, image, write_flags, + num_threads, blob_list_override, + blob_table_list_ret); - 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; - } - } + /* The blob table, XML data, and header at end are handled by + * finish_write(). */ +} - off_t old_lookup_table_end; - off_t new_lookup_table_end; - if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { - old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; - } else { - old_lookup_table_end = 0; +static bool +should_default_to_solid_compression(WIMStruct *wim, int write_flags) +{ + return wim->out_hdr.wim_version == WIM_VERSION_SOLID && + !(write_flags & (WIMLIB_WRITE_FLAG_SOLID | + WIMLIB_WRITE_FLAG_PIPABLE)) && + wim_has_solid_resources(wim); +} + +/* Update the images' filecount/bytecount stats (in the XML info) to take into + * account any recent modifications. */ +static int +update_image_stats(WIMStruct *wim) +{ + if (!wim_has_metadata(wim)) + return 0; + for (int i = 0; i < wim->hdr.image_count; i++) { + struct wim_image_metadata *imd = wim->image_metadata[i]; + if (imd->stats_outdated) { + int ret = xml_update_image_info(wim, i + 1); + if (ret) + return ret; + imd->stats_outdated = false; } - new_lookup_table_end = hdr.lookup_table_res_entry.offset + - hdr.lookup_table_res_entry.size; - - ret = write_integrity_table(out, - &hdr.integrity, - new_lookup_table_end, - old_lookup_table_end, - progress_func); - if (ret != 0) - goto out; - } else { - memset(&hdr.integrity, 0, sizeof(struct resource_entry)); } + return 0; +} - /* - * In the WIM header, there is room for the resource entry for a - * metadata resource labeled as the "boot metadata". This entry should - * be zeroed out if there is no bootable image (boot_idx 0). Otherwise, - * it should be a copy of the resource entry for the image that is - * marked as bootable. This is not well documented... - */ +/* Write a standalone WIM or split WIM (SWM) part to a new file or to a file + * descriptor. */ +int +write_wim_part(WIMStruct *wim, + const void *path_or_fd, + int image, + int write_flags, + unsigned num_threads, + unsigned part_number, + unsigned total_parts, + struct list_head *blob_list_override, + const u8 *guid) +{ + int ret; + struct list_head blob_table_list; + + /* Internally, this is always called with a valid part number and total + * parts. */ + wimlib_assert(total_parts >= 1); + wimlib_assert(part_number >= 1 && part_number <= total_parts); + + /* A valid image (or all images) must be specified. */ + if (image != WIMLIB_ALL_IMAGES && + (image < 1 || image > wim->hdr.image_count)) + return WIMLIB_ERR_INVALID_IMAGE; + + /* If we need to write metadata resources, make sure the ::WIMStruct has + * the needed information attached (e.g. is not a resource-only WIM, + * such as a non-first part of a split WIM). */ + if (!wim_has_metadata(wim) && + !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) + return WIMLIB_ERR_METADATA_NOT_FOUND; + + /* Check for contradictory flags. */ + if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + return WIMLIB_ERR_INVALID_PARAM; - /* Set image count and boot index correctly for single image writes */ - if (image != WIMLIB_ALL_IMAGES) { - hdr.image_count = 1; - if (hdr.boot_idx == image) - hdr.boot_idx = 1; - else - hdr.boot_idx = 0; + if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + return WIMLIB_ERR_INVALID_PARAM; + + /* Only wimlib_overwrite() accepts UNSAFE_COMPACT. */ + if (write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT) + return WIMLIB_ERR_INVALID_PARAM; + + /* Include an integrity table by default if no preference was given and + * the WIM already had an integrity table. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) { + if (wim_has_integrity_table(wim)) + write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; } - 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)); + /* Write a pipable WIM by default if no preference was given and the WIM + * was already pipable. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE))) { + if (wim_is_pipable(wim)) + write_flags |= WIMLIB_WRITE_FLAG_PIPABLE; } - 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; + if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_SOLID)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_SOLID)) + { + ERROR("Solid compression is unsupported in pipable WIMs"); + return WIMLIB_ERR_INVALID_PARAM; } - ret = write_header(&hdr, out); - if (ret != 0) - goto out; + /* Start initializing the new file header. */ + memset(&wim->out_hdr, 0, sizeof(wim->out_hdr)); - if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { - if (fflush(out) != 0 - || fsync(fileno(out)) != 0) - { - ERROR_WITH_ERRNO("Error flushing data to WIM file"); - ret = WIMLIB_ERR_WRITE; - } - } -out: - if (fclose(out) != 0) { - ERROR_WITH_ERRNO("Failed to close the WIM file"); - if (ret == 0) - ret = WIMLIB_ERR_WRITE; - } - w->out_fp = NULL; - return ret; -} + /* Set the magic number. */ + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + wim->out_hdr.magic = PWM_MAGIC; + else + wim->out_hdr.magic = WIM_MAGIC; -#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) -int lock_wim(FILE *fp, const char *path) -{ - int ret = 0; - if (fp) { - 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); - ret = WIMLIB_ERR_ALREADY_LOCKED; - } else { - WARNING("Failed to lock `%s': %s", - path, strerror(errno)); - ret = 0; - } + /* Set the version number. */ + if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) || + wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS) + wim->out_hdr.wim_version = WIM_VERSION_SOLID; + else + wim->out_hdr.wim_version = WIM_VERSION_DEFAULT; + + /* Default to solid compression if it is valid in the chosen WIM file + * format and the WIMStruct references any solid resources. This is + * useful when exporting an image from a solid WIM. */ + if (should_default_to_solid_compression(wim, write_flags)) + write_flags |= WIMLIB_WRITE_FLAG_SOLID; + + /* Set the header flags. */ + wim->out_hdr.flags = (wim->hdr.flags & (WIM_HDR_FLAG_RP_FIX | + WIM_HDR_FLAG_READONLY)); + if (total_parts != 1) + wim->out_hdr.flags |= WIM_HDR_FLAG_SPANNED; + if (wim->out_compression_type != WIMLIB_COMPRESSION_TYPE_NONE) { + wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESSION; + switch (wim->out_compression_type) { + case WIMLIB_COMPRESSION_TYPE_XPRESS: + wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_XPRESS; + break; + case WIMLIB_COMPRESSION_TYPE_LZX: + wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZX; + break; + case WIMLIB_COMPRESSION_TYPE_LZMS: + wim->out_hdr.flags |= WIM_HDR_FLAG_COMPRESS_LZMS; + break; } } - return ret; -} -#endif -static int open_wim_writable(WIMStruct *w, const char *path, - bool trunc, bool readable) -{ - const char *mode; - if (trunc) - if (readable) - mode = "w+b"; - else - mode = "wb"; + /* Set the chunk size. */ + wim->out_hdr.chunk_size = wim->out_chunk_size; + + /* Set the GUID. */ + if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID) + guid = wim->hdr.guid; + if (guid) + copy_guid(wim->out_hdr.guid, guid); else - mode = "r+b"; + generate_guid(wim->out_hdr.guid); - wimlib_assert(w->out_fp == NULL); - w->out_fp = fopen(path, mode); - if (w->out_fp) { - return 0; - } else { - ERROR_WITH_ERRNO("Failed to open `%s' for writing", path); - return WIMLIB_ERR_OPEN; + /* Set the part number and total parts. */ + wim->out_hdr.part_number = part_number; + wim->out_hdr.total_parts = total_parts; + + /* Set the image count. */ + if (image == WIMLIB_ALL_IMAGES) + wim->out_hdr.image_count = wim->hdr.image_count; + else + wim->out_hdr.image_count = 1; + + /* Set the boot index. */ + wim->out_hdr.boot_idx = 0; + if (total_parts == 1) { + if (image == WIMLIB_ALL_IMAGES) + wim->out_hdr.boot_idx = wim->hdr.boot_idx; + else if (image == wim->hdr.boot_idx) + wim->out_hdr.boot_idx = 1; } -} + /* Update image stats if needed. */ + ret = update_image_stats(wim); + if (ret) + return ret; -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)); + /* Set up the output file descriptor. */ + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) { + /* File descriptor was explicitly provided. */ + filedes_init(&wim->out_fd, *(const int *)path_or_fd); + if (!filedes_is_seekable(&wim->out_fd)) { + /* The file descriptor is a pipe. */ + ret = WIMLIB_ERR_INVALID_PARAM; + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + goto out_cleanup; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + ERROR("Can't include integrity check when " + "writing pipable WIM to pipe!"); + goto out_cleanup; + } } - w->out_fp = NULL; - } + } else { + /* Filename of WIM to write was provided; open file descriptor + * to it. */ + ret = open_wim_writable(wim, (const tchar*)path_or_fd, + O_TRUNC | O_CREAT | O_RDWR); + if (ret) + goto out_cleanup; + } + + /* Write initial header. This is merely a "dummy" header since it + * doesn't have resource entries filled in yet, so it will be + * overwritten later (unless writing a pipable WIM). */ + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + wim->out_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header(&wim->out_hdr, &wim->out_fd, wim->out_fd.offset); + wim->out_hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; + if (ret) + goto out_cleanup; + + /* Write file data and metadata resources. */ + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) { + /* Default case: create a normal (non-pipable) WIM. */ + ret = write_file_data(wim, image, write_flags, + num_threads, + blob_list_override, + &blob_table_list); + if (ret) + goto out_cleanup; + + ret = write_metadata_resources(wim, image, write_flags); + if (ret) + goto out_cleanup; + } else { + /* Non-default case: create pipable WIM. */ + ret = write_pipable_wim(wim, image, write_flags, num_threads, + blob_list_override, + &blob_table_list); + if (ret) + goto out_cleanup; + } + + /* Write blob table, XML data, and (optional) integrity table. */ + ret = finish_write(wim, image, write_flags, &blob_table_list); +out_cleanup: + (void)close_wim_writable(wim, write_flags); + return ret; } -/* Open file stream and write dummy header for WIM. */ -int begin_write(WIMStruct *w, const char *path, int write_flags) +/* Write a standalone WIM to a file or file descriptor. */ +static int +write_standalone_wim(WIMStruct *wim, const void *path_or_fd, + int image, int write_flags, unsigned num_threads) { - int ret; - 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. */ - return write_header(&w->hdr, w->out_fp); + return write_wim_part(wim, path_or_fd, image, write_flags, + num_threads, 1, 1, NULL, NULL); } -/* 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) +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write(WIMStruct *wim, const tchar *path, + int image, int write_flags, unsigned num_threads) { - int ret; - - if (!path) + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) return WIMLIB_ERR_INVALID_PARAM; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - - if (image != WIMLIB_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) - return WIMLIB_ERR_INVALID_IMAGE; - - if (w->hdr.total_parts != 1) { - ERROR("Cannot call wimlib_write() on part of a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; - } + if (path == NULL || path[0] == T('\0')) + return WIMLIB_ERR_INVALID_PARAM; - ret = begin_write(w, path, write_flags); - if (ret != 0) - goto out; + return write_standalone_wim(wim, path, image, write_flags, num_threads); +} - ret = write_wim_streams(w, image, write_flags, num_threads, - progress_func); - if (ret != 0) - goto out; +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_write_to_fd(WIMStruct *wim, int fd, + int image, int write_flags, unsigned num_threads) +{ + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); + if (fd < 0) + return WIMLIB_ERR_INVALID_PARAM; - ret = for_image(w, image, write_metadata_resource); - if (ret != 0) - goto out; + write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + return write_standalone_wim(wim, &fd, image, write_flags, num_threads); +} - ret = finish_write(w, image, write_flags, progress_func); -out: - close_wim_writable(w); - return ret; +/* Have there been any changes to images in the specified WIM, including updates + * as well as deletions and additions of entire images, but excluding changes to + * the XML document? */ +static bool +any_images_changed(WIMStruct *wim) +{ + if (wim->image_deletion_occurred) + return true; + for (int i = 0; i < wim->hdr.image_count; i++) + if (!is_image_unchanged_from_wim(wim->image_metadata[i], wim)) + return true; + return false; } -static int lte_overwrite_prepare(struct lookup_table_entry *lte, - void *ignore) +static int +check_resource_offset(struct blob_descriptor *blob, void *_wim) { - memcpy(<e->output_resource_entry, <e->resource_entry, - sizeof(struct resource_entry)); - lte->out_refcnt = 0; + const WIMStruct *wim = _wim; + off_t end_offset = *(const off_t*)wim->private; + + if (blob->blob_location == BLOB_IN_WIM && + blob->rdesc->wim == wim && + blob->rdesc->offset_in_wim + blob->rdesc->size_in_wim > end_offset) + return WIMLIB_ERR_RESOURCE_ORDER; return 0; } -static int check_resource_offset(struct lookup_table_entry *lte, void *arg) +/* Make sure no file or metadata resources are located after the XML data (or + * integrity table if present)--- otherwise we can't safely append to the WIM + * file and we return WIMLIB_ERR_RESOURCE_ORDER. */ +static int +check_resource_offsets(WIMStruct *wim, off_t end_offset) { - if (lte->out_refcnt > lte->refcnt) { - WARNING("Detected invalid stream reference count. " - "Forcing re-build of entire WIM."); - return WIMLIB_ERR_RESOURCE_ORDER; - } else if (lte->out_refcnt < lte->refcnt) { - off_t end_offset = *(u64*)arg; - 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; - } + int ret; + unsigned i; + + wim->private = &end_offset; + ret = for_blob_in_table(wim->blob_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_blob, wim); + if (ret) + return ret; } return 0; } -static int find_new_streams(struct lookup_table_entry *lte, void *arg) +static int +free_blob_if_invalidated(struct blob_descriptor *blob, void *_wim) { - if (lte->out_refcnt == lte->refcnt) - list_add(<e->staging_list, (struct list_head*)arg); - else - lte->out_refcnt = lte->refcnt; + const WIMStruct *wim = _wim; + + if (!blob->will_be_in_output_wim && + blob->blob_location == BLOB_IN_WIM && blob->rdesc->wim == wim) + { + blob_table_unlink(wim->blob_table, blob); + free_blob_descriptor(blob); + } return 0; } /* - * Overwrite a WIM, possibly appending streams to it. + * Overwrite a WIM, possibly appending new resources to it. * * A WIM looks like (or is supposed to look like) the following: * * Header (212 bytes) - * Streams and metadata resources (variable size) - * Lookup table (variable size) + * Resources for metadata and files (variable size) + * Blob table (variable size) * XML data (variable size) * Integrity table (optional) (variable size) * - * If we are not adding any streams or metadata resources, the lookup table is + * If we are not adding any new files or metadata, then the blob table is * unchanged--- so we only need to overwrite the XML data, integrity table, and * header. This operation is potentially unsafe if the program is abruptly * terminated while the XML data or integrity table are being overwritten, but * before the new header has been written. To partially alleviate this problem, - * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to - * finish_write() to cause a temporary WIM header to be written after the XML - * data has been written. This may prevent the WIM from becoming corrupted if - * the program is terminated while the integrity table is being calculated (but - * no guarantees, due to write re-ordering...). - * - * If we are adding new streams or images (metadata resources), the lookup table - * needs to be changed, and those streams need to be written. In this case, we - * try to perform a safe update of the WIM file by writing the streams *after* - * the end of the previous WIM, then writing the new lookup table, XML data, and - * (optionally) integrity table following the new streams. This will produce a - * layout like the following: + * we write a temporary header after the XML data has been written. This may + * prevent the WIM from becoming corrupted if the program is terminated while + * the integrity table is being calculated (but no guarantees, due to write + * re-ordering...). + * + * If we are adding new blobs, including new file data as well as any metadata + * for any new images, then the blob table needs to be changed, and those blobs + * need to be written. In this case, we try to perform a safe update of the WIM + * file by writing the blobs *after* the end of the previous WIM, then writing + * the new blob table, XML data, and (optionally) integrity table following the + * new blobs. This will produce a layout like the following: * * Header (212 bytes) - * (OLD) Streams and metadata resources (variable size) - * (OLD) Lookup table (variable size) + * (OLD) Resources for metadata and files (variable size) + * (OLD) Blob table (variable size) * (OLD) XML data (variable size) * (OLD) Integrity table (optional) (variable size) - * (NEW) Streams and metadata resources (variable size) - * (NEW) Lookup table (variable size) + * (NEW) Resources for metadata and files (variable size) + * (NEW) Blob table (variable size) * (NEW) XML data (variable size) * (NEW) Integrity table (optional) (variable size) * * At all points, the WIM is valid as nothing points to the new data yet. Then, - * the header is overwritten to point to the new lookup table, XML data, and + * the header is overwritten to point to the new blob table, XML data, and * integrity table, to produce the following layout: * * Header (212 bytes) - * Streams and metadata resources (variable size) + * Resources for metadata and files (variable size) * Nothing (variable size) - * More Streams and metadata resources (variable size) - * Lookup table (variable size) + * Resources for metadata and files (variable size) + * Blob table (variable size) * XML data (variable size) * Integrity table (optional) (variable size) * - * This method allows an image to be appended to a large WIM very quickly, and - * is is crash-safe except in the case of write re-ordering, but the - * disadvantage is that a small hole is left in the WIM where the old lookup - * table, xml data, and integrity table were. (These usually only take up a - * small amount of space compared to the streams, however.) + * This function allows an image to be appended to a large WIM very quickly, and + * 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 blob table, xml data, + * and integrity table were. (These usually only take up a small amount of + * space compared to the blobs, however.) + * + * Finally, this function also supports "compaction" overwrites as an + * alternative to the normal "append" overwrites described above. In a + * compaction, data is written starting immediately from the end of the header. + * All existing resources are written first, in order by file offset. New + * resources are written afterwards, and at the end any extra data is truncated + * from the file. The advantage of this approach is that is that the WIM file + * ends up fully optimized, without any holes remaining. The main disadavantage + * is that this operation is fundamentally unsafe and cannot be interrupted + * without data corruption. Consequently, compactions are only ever done when + * explicitly requested by the library user with the flag + * WIMLIB_WRITE_FLAG_UNSAFE_COMPACT. (Another disadvantage is that a compaction + * can be much slower than an append.) */ -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 *wim, int write_flags, unsigned num_threads) { int ret; - struct list_head stream_list; off_t old_wim_end; + struct list_head blob_list; + struct list_head blob_table_list; + struct filter_context filter_ctx; + + /* Include an integrity table by default if no preference was given and + * the WIM already had an integrity table. */ + if (!(write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) + if (wim_has_integrity_table(wim)) + write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + + /* Start preparing the updated file header. */ + memcpy(&wim->out_hdr, &wim->hdr, sizeof(wim->out_hdr)); + + /* If using solid compression, the version number must be set to + * WIM_VERSION_SOLID. */ + if (write_flags & WIMLIB_WRITE_FLAG_SOLID) + wim->out_hdr.wim_version = WIM_VERSION_SOLID; + + /* Default to solid compression if it is valid in the chosen WIM file + * format and the WIMStruct references any solid resources. This is + * useful when updating a solid WIM. */ + if (should_default_to_solid_compression(wim, write_flags)) + write_flags |= WIMLIB_WRITE_FLAG_SOLID; + + if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) { + + /* In-place compaction */ + + WARNING("The WIM file \"%"TS"\" is being compacted in place.\n" + " Do *not* interrupt the operation, or else " + "the WIM file will be\n" + " corrupted!", wim->filename); + wim->being_compacted = 1; + old_wim_end = WIM_HEADER_DISK_SIZE; + + ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES, + write_flags, &blob_list, + &blob_table_list, &filter_ctx); + if (ret) + goto out; - DEBUG("Overwriting `%s' in-place", w->filename); - - /* Make sure that the integrity table (if present) is after the XML - * data, and that there are no stream resources, metadata resources, or - * lookup tables after the XML data. Otherwise, these data would be - * overwritten. */ - if (w->hdr.integrity.offset != 0 && - w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the integrity table to be before the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } - - if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the lookup table to be after the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } + /* Prevent new files from being deduplicated with existing blobs + * in the WIM that we haven't decided to write. Such blobs will + * be overwritten during the compaction. */ + for_blob_in_table(wim->blob_table, free_blob_if_invalidated, wim); + + if (wim_has_metadata(wim)) { + /* Add existing metadata resources to be compacted along + * with the file resources. */ + for (int i = 0; i < wim->hdr.image_count; i++) { + struct wim_image_metadata *imd = wim->image_metadata[i]; + if (is_image_unchanged_from_wim(imd, wim)) { + fully_reference_blob_for_write(imd->metadata_blob, + &blob_list); + } + } + } + } else { + u64 old_blob_table_end, old_xml_begin, old_xml_end; + + /* Set additional flags for append. */ + write_flags |= WIMLIB_WRITE_FLAG_APPEND | + WIMLIB_WRITE_FLAG_STREAMS_OK; + + /* Make sure there is no data after the XML data, except + * possibily an integrity table. If this were the case, then + * this data would be overwritten. */ + old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim; + old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim; + old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim + + wim->hdr.blob_table_reshdr.size_in_wim; + if (wim_has_integrity_table(wim) && + wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) { + WARNING("Didn't expect the integrity table to be " + "before the XML data"); + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out; + } - DEBUG("Identifying newly added streams"); - for_lookup_table_entry(w->lookup_table, lte_overwrite_prepare, NULL); - INIT_LIST_HEAD(&stream_list); - for (int i = modified_image_idx; i < w->hdr.image_count; i++) { - DEBUG("Identifiying streams in image %d", i + 1); - w->private = &stream_list; - for_dentry_in_tree(w->image_metadata[i].root_dentry, - dentry_find_streams_to_write, w); - } + if (old_blob_table_end > old_xml_begin) { + WARNING("Didn't expect the blob table to be after " + "the XML data"); + ret = WIMLIB_ERR_RESOURCE_ORDER; + goto out; + } + /* Set @old_wim_end, which indicates the point beyond which we + * don't allow any file and metadata resources to appear without + * returning WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we + * would otherwise overwrite these resources). */ + if (!any_images_changed(wim)) { + /* If no images have been modified, added, or deleted, + * then a new blob table does not need to be written. + * We shall write the new XML data and optional + * integrity table immediately after the blob table. + * Note that this may overwrite an existing integrity + * table. */ + old_wim_end = old_blob_table_end; + write_flags |= WIMLIB_WRITE_FLAG_NO_NEW_BLOBS; + } else if (wim_has_integrity_table(wim)) { + /* Old WIM has an integrity table; begin writing new + * blobs after it. */ + old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim + + wim->hdr.integrity_table_reshdr.size_in_wim; + } else { + /* No existing integrity table; begin writing new blobs + * after the old XML data. */ + old_wim_end = old_xml_end; + } - 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 = check_resource_offsets(wim, old_wim_end); + if (ret) + goto out; - ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, - &old_wim_end); - if (ret != 0) - return ret; + ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES, + write_flags, &blob_list, + &blob_table_list, &filter_ctx); + if (ret) + goto out; - if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) { - /* If no images have been modified and no images have been - * deleted, a new lookup table does not need to be written. */ - 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; + if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS) + wimlib_assert(list_empty(&blob_list)); } - INIT_LIST_HEAD(&stream_list); - for_lookup_table_entry(w->lookup_table, find_new_streams, - &stream_list); + /* Update image stats if needed. */ + ret = update_image_stats(wim); + if (ret) + goto out; + + ret = open_wim_writable(wim, wim->filename, O_RDWR); + if (ret) + goto out; - ret = open_wim_writable(w, w->filename, false, - (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); - if (ret != 0) - return ret; + ret = lock_wim_for_append(wim); + if (ret) + goto out_close_wim; - ret = lock_wim(w->out_fp, w->filename); - if (ret != 0) { - fclose(w->out_fp); - w->out_fp = NULL; - return ret; + /* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */ + wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd); + wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; + if (ret) { + ERROR_WITH_ERRNO("Error updating WIM header flags"); + goto out_unlock_wim; } - if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { + if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - fclose(w->out_fp); - w->out_fp = NULL; - return WIMLIB_ERR_WRITE; - } - - if (!list_empty(&stream_list)) { - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); - ret = write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), - write_flags, num_threads, - progress_func); - if (ret != 0) - goto out_ftruncate; - } else { - DEBUG("No new streams were added"); + ret = WIMLIB_ERR_WRITE; + goto out_restore_hdr; } - for (int i = modified_image_idx; i < w->hdr.image_count; i++) { - select_wim_image(w, i + 1); - ret = write_metadata_resource(w); - if (ret != 0) - goto out_ftruncate; - } - write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; - ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, - progress_func); -out_ftruncate: - close_wim_writable(w); - 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); - } + ret = write_file_data_blobs(wim, &blob_list, write_flags, + num_threads, &filter_ctx); + if (ret) + goto out_truncate; + + ret = write_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags); + if (ret) + goto out_truncate; + + ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, + &blob_table_list); + if (ret) + goto out_truncate; + + unlock_wim_for_append(wim); + return 0; + +out_truncate: + if (!(write_flags & (WIMLIB_WRITE_FLAG_NO_NEW_BLOBS | + WIMLIB_WRITE_FLAG_UNSAFE_COMPACT))) { + WARNING("Truncating \"%"TS"\" to its original size " + "(%"PRIu64" bytes)", wim->filename, old_wim_end); + /* Return value of ftruncate() is ignored because this is + * already an error path. */ + (void)ftruncate(wim->out_fd.fd, old_wim_end); + } +out_restore_hdr: + (void)write_wim_header_flags(wim->hdr.flags, &wim->out_fd); +out_unlock_wim: + unlock_wim_for_append(wim); +out_close_wim: + (void)close_wim_writable(wim, write_flags); +out: + wim->being_compacted = 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 *wim, int write_flags, unsigned num_threads) { size_t wim_name_len; int ret; - DEBUG("Overwriting `%s' via a temporary file", w->filename); - /* Write the WIM to a temporary file in the same directory as the * original WIM. */ - wim_name_len = strlen(w->filename); - char tmpfile[wim_name_len + 10]; - memcpy(tmpfile, w->filename, wim_name_len); + wim_name_len = tstrlen(wim->filename); + tchar tmpfile[wim_name_len + 10]; + tmemcpy(tmpfile, wim->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\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); - goto err; + tmpfile[wim_name_len + 9] = T('\0'); + + ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, + write_flags | + WIMLIB_WRITE_FLAG_FSYNC | + WIMLIB_WRITE_FLAG_RETAIN_GUID, + num_threads); + if (ret) { + tunlink(tmpfile); + return ret; } - DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); - - /* Rename the new file to the old file .*/ - if (rename(tmpfile, w->filename) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", - tmpfile, w->filename); - ret = WIMLIB_ERR_RENAME; - goto err; + if (filedes_valid(&wim->in_fd)) { + filedes_close(&wim->in_fd); + filedes_invalidate(&wim->in_fd); } - if (progress_func) { - union wimlib_progress_info progress; - progress.rename.from = tmpfile; - progress.rename.to = w->filename; - progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); + /* Rename the new WIM file to the original WIM file. Note: on Windows + * this actually calls win32_rename_replacement(), not _wrename(), so + * that removing the existing destination file can be handled. */ + ret = trename(tmpfile, wim->filename); + if (ret) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", + tmpfile, wim->filename); + #ifdef __WIN32__ + if (ret < 0) + #endif + { + tunlink(tmpfile); + } + return WIMLIB_ERR_RENAME; } - /* Close the original WIM file that was opened for reading. */ - if (w->fp != NULL) { - fclose(w->fp); - w->fp = NULL; - } + union wimlib_progress_info progress; + progress.rename.from = tmpfile; + progress.rename.to = wim->filename; + return call_progress(wim->progfunc, WIMLIB_PROGRESS_MSG_RENAME, + &progress, wim->progctx); +} - /* Re-open the WIM read-only. */ - w->fp = fopen(w->filename, "rb"); - if (w->fp == NULL) { - ret = WIMLIB_ERR_REOPEN; - WARNING("Failed to re-open `%s' read-only: %s", - w->filename, strerror(errno)); - 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)); - return ret; +/* Determine if the specified WIM file may be updated in-place rather than by + * writing and replacing it with an entirely new file. */ +static bool +can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) +{ + /* REBUILD flag forces full rebuild. */ + if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) + return false; + + /* Image deletions cause full rebuild by default. */ + if (wim->image_deletion_occurred && + !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + return false; + + /* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be + * turned into a pipable WIM in-place. */ + if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + return false; + + /* The default compression type and compression chunk size selected for + * the output WIM must be the same as those currently used for the WIM. + */ + if (wim->compression_type != wim->out_compression_type) + return false; + if (wim->chunk_size != wim->out_chunk_size) + return false; + + return true; } -/* - * 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) +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads) { - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + int ret; + u32 orig_hdr_flags; - if (!w->filename) - return WIMLIB_ERR_NO_FILENAME; + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; - if (w->hdr.total_parts != 1) { - ERROR("Cannot modify a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; - } + if (!wim->filename) + return WIMLIB_ERR_NO_FILENAME; - if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) - && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) - { - int i; - int modified_image_idx; - int ret; + if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) { + /* + * In UNSAFE_COMPACT mode: + * - RECOMPRESS is forbidden + * - REBUILD is ignored + * - SOFT_DELETE and NO_SOLID_SORT are implied + */ + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + return WIMLIB_ERR_COMPACTION_NOT_POSSIBLE; + write_flags &= ~WIMLIB_WRITE_FLAG_REBUILD; + write_flags |= WIMLIB_WRITE_FLAG_SOFT_DELETE; + write_flags |= WIMLIB_WRITE_FLAG_NO_SOLID_SORT; + } + + orig_hdr_flags = wim->hdr.flags; + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) + wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY; + ret = can_modify_wim(wim); + wim->hdr.flags = orig_hdr_flags; + if (ret) + return ret; - 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) { - ret = overwrite_wim_inplace(w, write_flags, num_threads, - progress_func, - modified_image_idx); - if (ret == WIMLIB_ERR_RESOURCE_ORDER) - WARNING("Falling back to re-building entire WIM"); - else - return ret; - } + if (can_overwrite_wim_inplace(wim, write_flags)) { + ret = overwrite_wim_inplace(wim, write_flags, num_threads); + if (ret != WIMLIB_ERR_RESOURCE_ORDER) + return ret; + WARNING("Falling back to re-building entire WIM"); } - return overwrite_wim_via_tmpfile(w, write_flags, num_threads, - progress_func); + if (write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT) + return WIMLIB_ERR_COMPACTION_NOT_POSSIBLE; + return overwrite_wim_via_tmpfile(wim, write_flags, num_threads); }