X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=7f13630cceec991aea303afd27b74f0b1b67f888;hp=42f8ecd91e8b753b34a651948c678c741c518cb1;hb=05bbdd83d65782a9872b579505e1299587cd2009;hpb=a267f3e6b9df128bd4bdf5d39b951868dd4528ab diff --git a/src/write.c b/src/write.c index 42f8ecd9..7f13630c 100644 --- a/src/write.c +++ b/src/write.c @@ -6,22 +6,20 @@ */ /* - * Copyright (C) 2012, 2013 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/. */ #ifdef HAVE_CONFIG_H @@ -30,50 +28,64 @@ #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 - * overwrite the LIST_HEAD macro. */ + * override the LIST_HEAD macro. */ # include #endif +#include +#include +#include +#include + +#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/lookup_table.h" #include "wimlib/metadata.h" +#include "wimlib/paths.h" +#include "wimlib/progress.h" #include "wimlib/resource.h" -#ifdef __WIN32__ -# include "wimlib/win32.h" /* win32_rename_replacement() */ -#endif +#include "wimlib/solid.h" +#include "wimlib/win32.h" /* win32_rename_replacement() */ #include "wimlib/write.h" #include "wimlib/xml.h" -#include -#include -#include -#include - -#ifdef HAVE_ALLOCA_H -# include -#endif /* wimlib internal flags used when writing resources. */ #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002 -#define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004 +#define WRITE_RESOURCE_FLAG_SOLID 0x00000004 +#define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008 +#define WRITE_RESOURCE_FLAG_SOLID_SORT 0x00000010 -static inline int +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_PACK_STREAMS) - write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS; + + 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; } @@ -82,258 +94,254 @@ struct filter_context { WIMStruct *wim; }; -/* Determine specified stream should be filtered out from the write. +/* + * Determine whether the specified blob should be filtered out from the write. * * Return values: * - * < 0 : The stream should be hard-filtered; that is, not included in the - * output WIM at all. - * 0 : The stream should not be filtered out. - * > 0 : The stream should be soft-filtered; that is, it already exists in the + * < 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 -stream_filtered(const struct wim_lookup_table_entry *lte, - const struct filter_context *ctx) +blob_filtered(const struct blob_descriptor *blob, + const struct filter_context *ctx) { - int write_flags = ctx->write_flags; - WIMStruct *wim = ctx->wim; + int write_flags; + WIMStruct *wim; if (ctx == NULL) return 0; - if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE && - lte->resource_location == RESOURCE_IN_WIM && - lte->rspec->wim == wim) + 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 && - lte->resource_location == RESOURCE_IN_WIM && - lte->rspec->wim != wim) + blob->blob_location == BLOB_IN_WIM && + blob->rdesc->wim != wim) return -1; return 0; } static bool -stream_hard_filtered(const struct wim_lookup_table_entry *lte, - struct filter_context *ctx) +blob_hard_filtered(const struct blob_descriptor *blob, + struct filter_context *ctx) { - return stream_filtered(lte, ctx) < 0; + return blob_filtered(blob, ctx) < 0; } -static inline int -may_soft_filter_streams(const struct filter_context *ctx) +static inline bool +may_soft_filter_blobs(const struct filter_context *ctx) { - if (ctx == NULL) - return 0; - return ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE; + return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_APPEND); } -static inline int -may_hard_filter_streams(const struct filter_context *ctx) +static inline bool +may_hard_filter_blobs(const struct filter_context *ctx) { - if (ctx == NULL) - return 0; - return ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS; + return ctx && (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS); } -static inline int -may_filter_streams(const struct filter_context *ctx) +static inline bool +may_filter_blobs(const struct filter_context *ctx) { - return (may_soft_filter_streams(ctx) || - may_hard_filter_streams(ctx)); + return (may_soft_filter_blobs(ctx) || may_hard_filter_blobs(ctx)); } - -/* Return true if the specified resource is compressed and the compressed data - * can be reused with the specified output parameters. */ +/* 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 wim_lookup_table_entry *lte, - int write_resource_flags, int out_ctype, u32 out_chunk_size) +can_raw_copy(const struct blob_descriptor *blob, int write_resource_flags, + int out_ctype, u32 out_chunk_size) { - const struct wim_resource_spec *rspec; + const struct wim_resource_descriptor *rdesc; + /* Recompress everything if requested. */ if (write_resource_flags & WRITE_RESOURCE_FLAG_RECOMPRESS) return false; - if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + /* A blob not located in a WIM resource cannot be reused. */ + if (blob->blob_location != BLOB_IN_WIM) return false; - if (lte->resource_location != RESOURCE_IN_WIM) - return false; + rdesc = blob->rdesc; - rspec = lte->rspec; + /* In the case of an in-place compaction, always reuse resources located + * in the WIM being compacted. */ + if (rdesc->wim->being_compacted) + return true; - if (rspec->is_pipable != !!(write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) + /* 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; - if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) { - /* Normal compressed resource: Must use same compression type - * and chunk size. */ - return (rspec->wim->compression_type == out_ctype && - rspec->wim->chunk_size == out_chunk_size); - } + /* 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; - /* XXX: For compatibility, we can't allow multiple packed resources per - * WIM. */ -#if 0 - if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && - (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) - { - /* Packed resource: Such resources may contain multiple streams, + /* 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 at least half the + * a heuristic, re-use the raw data if more than two-thirds the * uncompressed size is being written. */ - /* Note: packed resources contain a header that specifies the + /* 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. */ - struct wim_lookup_table_entry *res_stream; + /* 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_stream, &rspec->stream_list, rspec_node) - if (res_stream->will_be_in_output_wim) - write_size += res_stream->size; + 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 > rspec->uncompressed_size / 2); + return (write_size > rdesc->uncompressed_size * 2 / 3); } -#endif - - return false; } -static u8 -filter_resource_flags(u8 flags) +static u32 +reshdr_flags_for_blob(const struct blob_descriptor *blob) { - return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS | - WIM_RESHDR_FLAG_COMPRESSED | - WIM_RESHDR_FLAG_SPANNED | - WIM_RESHDR_FLAG_FREE)); + u32 reshdr_flags = 0; + if (blob->is_metadata) + reshdr_flags |= WIM_RESHDR_FLAG_METADATA; + return reshdr_flags; } static void -stream_set_out_reshdr_for_reuse(struct wim_lookup_table_entry *lte) +blob_set_out_reshdr_for_reuse(struct blob_descriptor *blob) { - const struct wim_resource_spec *rspec; - - wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); - rspec = lte->rspec; - - if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { + const struct wim_resource_descriptor *rdesc; - wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS); + wimlib_assert(blob->blob_location == BLOB_IN_WIM); + rdesc = blob->rdesc; - lte->out_reshdr.offset_in_wim = lte->offset_in_res; - lte->out_reshdr.uncompressed_size = 0; - lte->out_reshdr.size_in_wim = lte->size; + 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; - lte->out_res_offset_in_wim = rspec->offset_in_wim; - lte->out_res_size_in_wim = rspec->size_in_wim; - /*lte->out_res_uncompressed_size = rspec->uncompressed_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 { - wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); - - lte->out_reshdr.offset_in_wim = rspec->offset_in_wim; - lte->out_reshdr.uncompressed_size = rspec->uncompressed_size; - lte->out_reshdr.size_in_wim = rspec->size_in_wim; + 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; } - lte->out_reshdr.flags = lte->flags; + blob->out_reshdr.flags = rdesc->flags; } -/* Write the header for a stream in a pipable WIM. */ +/* Write the header for a blob in a pipable WIM. */ static int -write_pwm_stream_header(const struct wim_lookup_table_entry *lte, - struct filedes *out_fd, - int additional_reshdr_flags) +write_pwm_blob_header(const struct blob_descriptor *blob, + struct filedes *out_fd, bool compressed) { - struct pwm_stream_hdr stream_hdr; + struct pwm_blob_hdr blob_hdr; u32 reshdr_flags; int ret; - stream_hdr.magic = cpu_to_le64(PWM_STREAM_MAGIC); - stream_hdr.uncompressed_size = cpu_to_le64(lte->size); - if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) { - zero_out_hash(stream_hdr.hash); - } else { - wimlib_assert(!lte->unhashed); - copy_hash(stream_hdr.hash, lte->hash); - } + wimlib_assert(!blob->unhashed); - reshdr_flags = filter_resource_flags(lte->flags); - reshdr_flags |= additional_reshdr_flags; - stream_hdr.flags = cpu_to_le32(reshdr_flags); - ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr)); + 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_streams_progress_data { - wimlib_progress_func_t progress_func; +struct write_blobs_progress_data { + wimlib_progress_func_t progfunc; + void *progctx; union wimlib_progress_info progress; - uint64_t next_progress; - WIMStruct *prev_wim_part; + u64 next_progress; }; -static void -do_write_streams_progress(struct write_streams_progress_data *progress_data, - u64 size, - bool discarded, - struct wim_lookup_table_entry *cur_stream) +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; - bool new_wim_part; + int ret; if (discarded) { - progress->write_streams.total_bytes -= size; - if (progress_data->next_progress != ~(uint64_t)0 && + 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 += size; - } - new_wim_part = false; - if (cur_stream->resource_location == RESOURCE_IN_WIM && - cur_stream->rspec->wim != progress_data->prev_wim_part) - { - if (progress_data->prev_wim_part) { - new_wim_part = true; - progress->write_streams.completed_parts++; - } - progress_data->prev_wim_part = cur_stream->rspec->wim; + progress->write_streams.completed_bytes += complete_size; + progress->write_streams.completed_streams += complete_count; } - progress->write_streams.completed_streams++; - if (progress_data->progress_func - && (progress->write_streams.completed_bytes >= progress_data->next_progress - || new_wim_part)) - { - progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - if (progress_data->next_progress == progress->write_streams.total_bytes) { - progress_data->next_progress = ~(uint64_t)0; - } else { - progress_data->next_progress = - min(progress->write_streams.total_bytes, - progress->write_streams.completed_bytes + - progress->write_streams.total_bytes / 100); - } + + 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; } -struct write_streams_ctx { - /* File descriptor the streams are being written to. */ +struct write_blobs_ctx { + /* File descriptor to which the blobs are being written. */ struct filedes *out_fd; - /* Lookup table for the WIMStruct on whose behalf the streams are being + /* Blob table for the WIMStruct on whose behalf the blobs are being * written. */ - struct wim_lookup_table *lookup_table; + struct blob_table *blob_table; /* Compression format to use. */ int out_ctype; @@ -341,47 +349,42 @@ struct write_streams_ctx { /* Maximum uncompressed chunk size in compressed resources to use. */ u32 out_chunk_size; - /* Flags that affect how the streams will be written. */ + /* Flags that affect how the blobs will be written. */ int write_resource_flags; /* Data used for issuing WRITE_STREAMS progress. */ - struct write_streams_progress_data progress_data; + struct write_blobs_progress_data progress_data; struct filter_context *filter_ctx; - /* Upper bound on the total number of bytes that need to be compressed. - * */ - u64 num_bytes_to_compress; - /* 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; - /* Buffer for dividing the read data into chunks of size - * @out_chunk_size. */ - u8 *chunk_buf; + /* 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 @chunk_buf that are currently filled. */ - size_t chunk_buf_filled; + /* Number of bytes in @cur_chunk_buf that are currently filled. */ + size_t cur_chunk_buf_filled; - /* List of streams that currently have chunks being compressed. */ - struct list_head pending_streams; + /* List of blobs that currently have chunks being compressed. */ + struct list_head blobs_being_compressed; - /* Set to true if the stream currently being read was a duplicate, and - * therefore the corresponding stream entry needs to be freed once the - * read finishes. (In this case we add the duplicate entry to - * pending_streams rather than the entry being read.) */ - bool stream_was_duplicate; + /* 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 resource being read. */ - u64 cur_read_res_offset; + /* Current uncompressed offset in the blob being read. */ + u64 cur_read_blob_offset; - /* Uncompressed size of the resource currently being read. */ - u64 cur_read_res_size; + /* Uncompressed size of the blob currently being read. */ + u64 cur_read_blob_size; - /* Current uncompressed offset in the resource being written. */ - u64 cur_write_res_offset; + /* 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; @@ -401,20 +404,10 @@ struct write_streams_ctx { u64 chunks_start_offset; }; -static u64 -get_chunk_entry_size(u64 res_size, int write_resource_flags) -{ - if (res_size <= UINT32_MAX || - (write_resource_flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) - return 4; - else - return 8; -} - /* Reserve space for the chunk table and prepare to accumulate the chunk table * in memory. */ static int -begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) +begin_chunk_table(struct write_blobs_ctx *ctx, u64 res_expected_size) { u64 expected_num_chunks; u64 expected_num_chunk_entries; @@ -423,12 +416,12 @@ begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) /* Calculate the number of chunks and chunk entries that should be * needed for the resource. These normally will be the final values, - * but in PACKED_STREAMS mode some of the streams we're planning to - * write into the resource may be duplicates, and therefore discarded, - * potentially decreasing the number of chunk entries needed. */ + * 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_PACK_STREAMS)) + 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 @@ -455,15 +448,16 @@ begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) { /* Reserve space for the chunk table in the output file. In the - * case of packed resources this reserves the upper bound for - * the needed space, not necessarily the exact space which will + * 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, - ctx->write_resource_flags); - if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) + 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); @@ -474,7 +468,7 @@ begin_chunk_table(struct write_streams_ctx *ctx, u64 res_expected_size) } static int -begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size) +begin_write_resource(struct write_blobs_ctx *ctx, u64 res_expected_size) { int ret; @@ -489,13 +483,13 @@ begin_write_resource(struct write_streams_ctx *ctx, u64 res_expected_size) /* 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_res_offset = 0; + ctx->cur_write_blob_offset = 0; ctx->cur_write_res_size = res_expected_size; return 0; } static int -end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, +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; @@ -505,19 +499,20 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, actual_num_chunks = ctx->chunk_index; actual_num_chunk_entries = actual_num_chunks; - if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) + if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) actual_num_chunk_entries--; chunk_entry_size = get_chunk_entry_size(res_actual_size, - ctx->write_resource_flags); + 0 != (ctx->write_resource_flags & + WRITE_RESOURCE_FLAG_SOLID)); - typedef le64 __attribute__((may_alias)) aliased_le64_t; - typedef le32 __attribute__((may_alias)) aliased_le32_t; + 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_PACK_STREAMS) { + 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 { @@ -531,7 +526,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, } else { aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes; - if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + 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 { @@ -551,7 +546,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size); if (ret) - goto error; + goto write_error; res_end_offset = ctx->out_fd->offset; res_start_offset = ctx->chunks_start_offset; } else { @@ -561,21 +556,21 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, chunk_table_offset = ctx->chunks_start_offset - chunk_table_size; - if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + 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); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2); - BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3); + 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 error; + goto write_error; res_start_offset = chunk_table_offset - sizeof(hdr); } else { res_start_offset = chunk_table_offset; @@ -584,7 +579,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes, chunk_table_size, chunk_table_offset); if (ret) - goto error; + goto write_error; } *res_start_offset_ret = res_start_offset; @@ -592,7 +587,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, return 0; -error: +write_error: ERROR_WITH_ERRNO("Write error"); return ret; } @@ -600,14 +595,15 @@ error: /* 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_streams_ctx *ctx, struct wim_reshdr *out_reshdr) +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_res_size == ctx->cur_write_res_offset); + 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) { @@ -622,108 +618,154 @@ end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr) 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; - DEBUG("Finished writing resource: %"PRIu64" => %"PRIu64" @ %"PRIu64"", - res_uncompressed_size, res_size_in_wim, res_offset_in_wim); return 0; } -/* Begin processing a stream for writing. */ +/* 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) +{ + 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); +} + static int -write_stream_begin_read(struct wim_lookup_table_entry *lte, - bool is_partial_res, void *_ctx) +do_done_with_blob(struct blob_descriptor *blob, + wimlib_progress_func_t progfunc, void *progctx) { - struct write_streams_ctx *ctx = _ctx; int ret; + struct wim_inode *inode; + tchar *cookie1; + tchar *cookie2; + + if (!blob->may_send_done_with_file) + return 0; - wimlib_assert(lte->size > 0); + inode = blob->file_inode; - ctx->cur_read_res_offset = 0; - ctx->cur_read_res_size = lte->size; + wimlib_assert(inode != NULL); + wimlib_assert(inode->i_num_remaining_streams > 0); + if (--inode->i_num_remaining_streams > 0) + return 0; - /* As an optimization, we allow some streams to be "unhashed", meaning - * their SHA1 message digests are unknown. This is the case with - * streams that are added by scanning a directry tree with - * wimlib_add_image(), for example. Since WIM uses single-instance - * streams, we don't know whether such each such stream really need to - * written until it is actually checksummed, unless it has a unique - * size. In such cases we read and checksum the stream in this - * function, thereby advancing ahead of read_stream_list(), which will - * still provide the data again to write_stream_process_chunk(). This - * is okay because an unhashed stream cannot be in a WIM resource, which - * might be costly to decompress. */ - ctx->stream_was_duplicate = false; - if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) { + cookie1 = progress_get_streamless_path(blob->file_on_disk); + cookie2 = progress_get_win32_path(blob->file_on_disk); - wimlib_assert(!is_partial_res); + ret = done_with_file(blob->file_on_disk, progfunc, progctx); - struct wim_lookup_table_entry *lte_new; + progress_put_win32_path(cookie2); + progress_put_streamless_path(cookie1); + + return ret; +} - ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new); +/* 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 (lte_new != lte) { - /* Duplicate stream detected. */ + if (new_blob != blob) { + /* Duplicate blob detected. */ - if (lte_new->will_be_in_output_wim || - stream_filtered(lte_new, ctx->filter_ctx)) + if (new_blob->will_be_in_output_wim || + blob_filtered(new_blob, ctx->filter_ctx)) { - /* The duplicate stream is already being - * included in the output WIM, or it would be - * filtered out if it had been. Skip writing - * this stream (and reading it again) entirely, - * passing its output reference count to the - * duplicate stream in the former case. */ - DEBUG("Discarding duplicate stream of " - "length %"PRIu64, lte->size); - do_write_streams_progress(&ctx->progress_data, - lte->size, true, lte); - list_del(<e->write_streams_list); - list_del(<e->lookup_table_list); - if (lte_new->will_be_in_output_wim) - lte_new->out_refcnt += lte->out_refcnt; - if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) - ctx->cur_write_res_size -= lte->size; - free_lookup_table_entry(lte); - return BEGIN_STREAM_STATUS_SKIP_STREAM; + /* 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 stream can validly be written, + /* The duplicate blob can validly be written, * but was not marked as such. Discard the - * current stream entry and use the duplicate, - * but actually freeing the current entry must - * wait until read_stream_list() has finished - * reading its data. */ - DEBUG("Stream duplicate, but not already " - "selected for writing."); - list_replace(<e->write_streams_list, - <e_new->write_streams_list); - list_replace(<e->lookup_table_list, - <e_new->lookup_table_list); - lte_new->out_refcnt = lte->out_refcnt; - lte_new->will_be_in_output_wim = 1; - ctx->stream_was_duplicate = true; - lte = lte_new; + * 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(<e->write_streams_list, &ctx->pending_streams); + list_move_tail(&blob->write_blobs_list, &ctx->blobs_being_compressed); return 0; } -/* Rewrite a stream that was just written compressed as uncompressed instead. - * This function is optional, but if a stream did not compress to less than its - * original size, it might as well be written uncompressed. */ +/* Rewrite a blob that was just written compressed (as a non-solid WIM resource) + * as uncompressed instead. */ static int -write_stream_uncompressed(struct wim_lookup_table_entry *lte, - struct filedes *out_fd) +write_blob_uncompressed(struct blob_descriptor *blob, struct filedes *out_fd) { int ret; - u64 begin_offset = lte->out_reshdr.offset_in_wim; + 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_full_stream_to_fd(lte, out_fd); + ret = extract_blob_to_fd(blob, out_fd); if (ret) { /* Error reading the uncompressed data. */ if (out_fd->offset == begin_offset && @@ -733,63 +775,120 @@ write_stream_uncompressed(struct wim_lookup_table_entry *lte, * seeked to the end of the compressed resource, so * don't issue a hard error; just keep the compressed * resource instead. */ - WARNING("Recovered compressed stream of " - "size %"PRIu64", continuing on.", - lte->size); + WARNING("Recovered compressed resource of " + "size %"PRIu64", continuing on.", blob->size); return 0; } return ret; } - wimlib_assert(out_fd->offset - begin_offset == lte->size); + 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. */ - if (out_fd->offset < end_offset && - 0 != ftruncate(out_fd->fd, out_fd->offset)) + 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) { - ERROR_WITH_ERRNO("Can't truncate output file to " - "offset %"PRIu64, out_fd->offset); - return WIMLIB_ERR_WRITE; + blob->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED; + return 0; } - lte->out_reshdr.size_in_wim = lte->size; - lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED | - WIM_RESHDR_FLAG_PACKED_STREAMS); - 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_streams_ctx *ctx, const void *cchunk, +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; - struct wim_lookup_table_entry *lte; - - lte = list_entry(ctx->pending_streams.next, - struct wim_lookup_table_entry, write_streams_list); + blob = list_entry(ctx->blobs_being_compressed.next, + struct blob_descriptor, write_blobs_list); - if (ctx->cur_write_res_offset == 0 && - !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) + if (ctx->cur_write_blob_offset == 0 && + !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)) { - /* Starting to write a new stream in non-packed mode. */ + /* Starting to write a new blob in non-solid mode. */ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { - int additional_reshdr_flags = 0; - if (ctx->compressor != NULL) - additional_reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; - - DEBUG("Writing pipable WIM stream header " - "(offset=%"PRIu64")", ctx->out_fd->offset); - - ret = write_pwm_stream_header(lte, ctx->out_fd, - additional_reshdr_flags); + ret = write_pwm_blob_header(blob, ctx->out_fd, + ctx->compressor != NULL); if (ret) return ret; } - ret = begin_write_resource(ctx, lte->size); + ret = begin_write_resource(ctx, blob->size); if (ret) return ret; } @@ -808,85 +907,101 @@ write_chunk(struct write_streams_ctx *ctx, const void *cchunk, ret = full_write(ctx->out_fd, &chunk_hdr, sizeof(chunk_hdr)); if (ret) - goto error; + goto write_error; } } /* Write the chunk data. */ ret = full_write(ctx->out_fd, cchunk, csize); if (ret) - goto error; + goto write_error; - ctx->cur_write_res_offset += usize; + ctx->cur_write_blob_offset += usize; - do_write_streams_progress(&ctx->progress_data, - usize, false, lte); + 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; - if (ctx->cur_write_res_offset == ctx->cur_write_res_size && - !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)) - { - wimlib_assert(ctx->cur_write_res_offset == lte->size); - - /* Finished writing a stream in non-packed mode. */ + while (blob && ctx->cur_write_blob_offset >= blob->size) { - ret = end_write_resource(ctx, <e->out_reshdr); - if (ret) - return ret; + ctx->cur_write_blob_offset -= blob->size; - lte->out_reshdr.flags = filter_resource_flags(lte->flags); - if (ctx->compressor != NULL) - lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED; + 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; - if (ctx->compressor != NULL && - lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size && - !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) && - !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) - { - /* Stream did not compress to less than its original - * size. If we're not writing a pipable WIM (which - * could mean the output file descriptor is - * non-seekable), and the stream isn't located in a - * resource pack (which would make reading it again - * costly), truncate the file to the start of the stream - * and write it uncompressed instead. */ - DEBUG("Stream of size %"PRIu64" did not compress to " - "less than original size; writing uncompressed.", - lte->size); - ret = write_stream_uncompressed(lte, ctx->out_fd); + 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; } + } else { + /* 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); - wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size); + ret = end_write_resource(ctx, &blob->out_reshdr); + if (ret) + return ret; - list_del(<e->write_streams_list); - ctx->cur_write_res_offset = 0; + 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++; + } } - return 0; + return do_write_blobs_progress(&ctx->progress_data, completed_size, + completed_blob_count, false); -error: +write_error: ERROR_WITH_ERRNO("Write error"); return ret; } static int -submit_chunk_for_compression(struct write_streams_ctx *ctx, - const void *chunk, size_t size) +prepare_chunk_buffer(struct write_blobs_ctx *ctx) { - /* While we are unable to submit the chunk for compression (due to too - * many chunks already outstanding), retrieve and write the next - * compressed chunk. */ - while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) { + /* 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; - unsigned csize; - unsigned usize; + u32 csize; + u32 usize; bool bret; int ret; - bret = ctx->compressor->get_chunk(ctx->compressor, - &cchunk, &csize, &usize); - + bret = ctx->compressor->get_compression_result(ctx->compressor, + &cchunk, + &csize, + &usize); wimlib_assert(bret); ret = write_chunk(ctx, cchunk, csize, usize); @@ -898,9 +1013,9 @@ submit_chunk_for_compression(struct write_streams_ctx *ctx, /* Process the next chunk of data to be written to a WIM resource. */ static int -write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) +write_blob_process_chunk(const void *chunk, size_t size, void *_ctx) { - struct write_streams_ctx *ctx = _ctx; + struct write_blobs_ctx *ctx = _ctx; int ret; const u8 *chunkptr, *chunkend; @@ -911,7 +1026,7 @@ write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) ret = write_chunk(ctx, chunk, size, size); if (ret) return ret; - ctx->cur_read_res_offset += size; + ctx->cur_read_blob_offset += size; return 0; } @@ -921,159 +1036,173 @@ write_stream_process_chunk(const void *chunk, size_t size, void *_ctx) chunkptr = chunk; chunkend = chunkptr + size; do { - const u8 *resized_chunk; 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_PACK_STREAMS) { + if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { needed_chunk_size = ctx->out_chunk_size; } else { - u64 res_bytes_remaining; - - res_bytes_remaining = ctx->cur_read_res_size - - ctx->cur_read_res_offset; needed_chunk_size = min(ctx->out_chunk_size, - ctx->chunk_buf_filled + - res_bytes_remaining); + ctx->cur_chunk_buf_filled + + (ctx->cur_read_blob_size - + ctx->cur_read_blob_offset)); } - if (ctx->chunk_buf_filled == 0 && - chunkend - chunkptr >= needed_chunk_size) - { - /* No intermediate buffering needed. */ - resized_chunk = chunkptr; - chunkptr += needed_chunk_size; - ctx->cur_read_res_offset += needed_chunk_size; - } else { - /* Intermediate buffering needed. */ - size_t bytes_consumed; - - bytes_consumed = min(chunkend - chunkptr, - needed_chunk_size - ctx->chunk_buf_filled); + bytes_consumed = min(chunkend - chunkptr, + needed_chunk_size - ctx->cur_chunk_buf_filled); - memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled], - chunkptr, bytes_consumed); + memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled], + chunkptr, bytes_consumed); - resized_chunk = ctx->chunk_buf; - - chunkptr += bytes_consumed; - ctx->cur_read_res_offset += bytes_consumed; - ctx->chunk_buf_filled += bytes_consumed; - if (ctx->chunk_buf_filled == needed_chunk_size) { - resized_chunk = ctx->chunk_buf; - ctx->chunk_buf_filled = 0; - } else { - break; - } + 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; } - - ret = submit_chunk_for_compression(ctx, resized_chunk, - needed_chunk_size); - if (ret) - return ret; - } while (chunkptr != chunkend); return 0; } -/* Finish processing a stream for writing. It may not have been completely +/* 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_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx) -{ - struct write_streams_ctx *ctx = _ctx; - if (status == 0) - wimlib_assert(ctx->cur_read_res_offset == ctx->cur_read_res_size); - if (ctx->stream_was_duplicate) { - free_lookup_table_entry(lte); - } else if (lte->unhashed && ctx->lookup_table != NULL) { - list_del(<e->unhashed_list); - lookup_table_insert(ctx->lookup_table, lte); - lte->unhashed = 0; +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; } -/* Compute statistics about a list of streams that will be written. +/* + * Compute statistics about a list of blobs that will be written. * - * Assumes the streams are sorted such that all streams located in each distinct - * WIM (specified by WIMStruct) are together. */ -static void -compute_stream_list_stats(struct list_head *stream_list, - struct write_streams_ctx *ctx) + * Assumes the blobs are sorted such that all blobs located in each distinct WIM + * (specified by WIMStruct) are together. + * + * 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 +compute_blob_list_stats(struct list_head *blob_list, + struct write_blobs_ctx *ctx) { - struct wim_lookup_table_entry *lte; + struct blob_descriptor *blob; u64 total_bytes = 0; - u64 num_streams = 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; - list_for_each_entry(lte, stream_list, write_streams_list) { - num_streams++; - total_bytes += lte->size; - if (lte->resource_location == RESOURCE_IN_WIM) { - if (prev_wim_part != lte->rspec->wim) { - prev_wim_part = lte->rspec->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; + } } } ctx->progress_data.progress.write_streams.total_bytes = total_bytes; - ctx->progress_data.progress.write_streams.total_streams = num_streams; + 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; - ctx->progress_data.prev_wim_part = NULL; + return 0; } -/* Find streams in @stream_list that can be copied to the output WIM in raw form - * rather than compressed. Delete these streams from @stream_list, and move one - * per resource to @raw_copy_resources. Return the total uncompressed size of - * the streams that need to be compressed. */ +/* 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_resources(struct list_head *stream_list, - int write_resource_flags, - int out_ctype, - u32 out_chunk_size, - struct list_head *raw_copy_resources) +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) { - struct wim_lookup_table_entry *lte, *tmp; - u64 num_bytes_to_compress = 0; + struct blob_descriptor *blob, *tmp; + u64 num_nonraw_bytes = 0; - INIT_LIST_HEAD(raw_copy_resources); + INIT_LIST_HEAD(raw_copy_blobs); /* Initialize temporary raw_copy_ok flag. */ - list_for_each_entry(lte, stream_list, write_streams_list) - if (lte->resource_location == RESOURCE_IN_WIM) - lte->rspec->raw_copy_ok = 0; + 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(lte, tmp, stream_list, write_streams_list) { - if (lte->resource_location == RESOURCE_IN_WIM && - lte->rspec->raw_copy_ok) - { - list_del(<e->write_streams_list); - } else if (can_raw_copy(lte, write_resource_flags, + 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)) { - lte->rspec->raw_copy_ok = 1; - list_move_tail(<e->write_streams_list, - raw_copy_resources); + blob->rdesc->raw_copy_ok = 1; + list_move_tail(&blob->write_blobs_list, raw_copy_blobs); } else { - num_bytes_to_compress += lte->size; + num_nonraw_bytes += blob->size; } } - return num_bytes_to_compress; + 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_spec *in_rspec, +write_raw_copy_resource(struct wim_resource_descriptor *in_rdesc, struct filedes *out_fd) { u64 cur_read_offset; @@ -1082,95 +1211,117 @@ write_raw_copy_resource(struct wim_resource_spec *in_rspec, size_t bytes_to_read; int ret; struct filedes *in_fd; - struct wim_lookup_table_entry *lte; + struct blob_descriptor *blob; u64 out_offset_in_wim; - DEBUG("Copying raw compressed data (size_in_wim=%"PRIu64", " - "uncompressed_size=%"PRIu64")", - in_rspec->size_in_wim, in_rspec->uncompressed_size); - /* Copy the raw data. */ - cur_read_offset = in_rspec->offset_in_wim; - end_read_offset = cur_read_offset + in_rspec->size_in_wim; + 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_rspec->is_pipable) { - if (cur_read_offset < sizeof(struct pwm_stream_hdr)) + 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_stream_hdr); - out_offset_in_wim += sizeof(struct pwm_stream_hdr); + cur_read_offset -= sizeof(struct pwm_blob_hdr); + out_offset_in_wim += sizeof(struct pwm_blob_hdr); } - in_fd = &in_rspec->wim->in_fd; + in_fd = &in_rdesc->wim->in_fd; wimlib_assert(cur_read_offset != end_read_offset); - do { - bytes_to_read = min(sizeof(buf), end_read_offset - cur_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_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; + ret = full_write(out_fd, buf, bytes_to_read); + if (ret) + return ret; - cur_read_offset += bytes_to_read; + cur_read_offset += bytes_to_read; - } while (cur_read_offset != end_read_offset); + } 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. */ - list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) { - if (lte->will_be_in_output_wim) { - stream_set_out_reshdr_for_reuse(lte); - if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) - lte->out_res_offset_in_wim = out_offset_in_wim; + /* 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 - lte->out_reshdr.offset_in_wim = out_offset_in_wim; + blob->out_reshdr.offset_in_wim = out_offset_in_wim; } } return 0; } -/* Copy a list of raw compressed resources located other WIM file(s) to the WIM - * file being written. */ +/* 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_resources, +write_raw_copy_resources(struct list_head *raw_copy_blobs, struct filedes *out_fd, - struct write_streams_progress_data *progress_data) + struct write_blobs_progress_data *progress_data) { - struct wim_lookup_table_entry *lte; + struct blob_descriptor *blob; int ret; - list_for_each_entry(lte, raw_copy_resources, write_streams_list) { - ret = write_raw_copy_resource(lte->rspec, out_fd); + 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; + } + ret = do_write_blobs_progress(progress_data, blob->size, + 1, false); if (ret) return ret; - do_write_streams_progress(progress_data, lte->size, false, lte); } return 0; } /* Wait for and write all chunks pending in the compressor. */ static int -finish_remaining_chunks(struct write_streams_ctx *ctx) +finish_remaining_chunks(struct write_blobs_ctx *ctx) { const void *cdata; - unsigned csize; - unsigned usize; + u32 csize; + u32 usize; int ret; if (ctx->compressor == NULL) return 0; - if (ctx->chunk_buf_filled != 0) { - ret = submit_chunk_for_compression(ctx, ctx->chunk_buf, - ctx->chunk_buf_filled); - if (ret) - return ret; + if (ctx->cur_chunk_buf_filled != 0) { + ctx->compressor->signal_chunk_filled(ctx->compressor, + ctx->cur_chunk_buf_filled); } - while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) { + while (ctx->compressor->get_compression_result(ctx->compressor, &cdata, + &csize, &usize)) + { ret = write_chunk(ctx, cdata, csize, usize); if (ret) return ret; @@ -1179,235 +1330,232 @@ finish_remaining_chunks(struct write_streams_ctx *ctx) } static void -remove_zero_length_streams(struct list_head *stream_list) -{ - struct wim_lookup_table_entry *lte, *tmp; - - list_for_each_entry_safe(lte, tmp, stream_list, write_streams_list) { - wimlib_assert(lte->will_be_in_output_wim); - if (lte->size == 0) { - list_del(<e->write_streams_list); - lte->out_reshdr.offset_in_wim = 0; - lte->out_reshdr.size_in_wim = 0; - lte->out_reshdr.uncompressed_size = 0; - lte->out_reshdr.flags = filter_resource_flags(lte->flags); +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); + } +} + +static void +init_done_with_file_info(struct list_head *blob_list) +{ + 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; } } + + list_for_each_entry(blob, blob_list, write_blobs_list) + if (blob->may_send_done_with_file) + blob->file_inode->i_num_remaining_streams++; } /* - * Write a list of streams to the output WIM file. + * Write a list of blobs to the output WIM file. * - * @stream_list - * The list of streams to write, specifies a list of `struct - * wim_lookup_table_entry's linked by the 'write_streams_list' member. + * @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 streams. + * The file descriptor, opened for writing, to which to write the blobs. * * @write_resource_flags - * Flags to modify how the streams are written: + * 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 caping the raw data, due to being located in a WIM + * 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, contrary to the - * default behavior). + * @out_fd will be performed (so it may be a pipe). * - * WRITE_RESOURCE_FLAG_PACK_STREAMS: - * Pack all the streams into a single resource rather than writing - * them in separate resources. This format is only valid if the - * WIM version number is WIM_VERSION_PACKED_STREAMS. This flag - * currently may not be combined with WRITE_RESOURCE_FLAG_PIPABLE. + * 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 to write the output streams, specified as one - * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting - * WIMLIB_COMPRESSION_TYPE_INVALID but including - * WIMLIB_COMPRESSION_TYPE_NONE. + * 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 - * Chunk size to use to write the streams. 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. + * 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. * - * @lookup_table - * If on-the-fly deduplication of unhashed streams is desired, this - * parameter must be pointer to the lookup table for the WIMStruct on whose - * behalf the streams are being written. Otherwise, this parameter can be - * NULL. + * @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 streams is desired, this - * parameter can be a pointer to a context for stream filtering used to - * detect whether the duplicate stream has been hard-filtered or not. If - * no streams are hard-filtered or no streams are unhashed, this parameter - * can be NULL. - * - * @progress_func - * If non-NULL, a progress function that will be called periodically with - * WIMLIB_PROGRESS_MSG_WRITE_STREAMS messages. Note that on-the-fly - * deduplication of unhashed streams may result in the total bytes provided - * in the progress data to decrease from one message to the next. - * - * This function will write the streams in @stream_list to resources in - * consecutive positions in the output WIM file, or to a single packed resource - * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags. - * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for - * each stream written will be updated to specify its location, size, and flags - * in the output WIM. In the packed resource case, - * WIM_RESHDR_FLAG_PACKED_STREAMS shall be set in the @flags field of the - * @out_reshdr, and @out_res_offset_in_wim and @out_res_size_in_wim will also - * be set to the offset and size, respectively, in the output WIM of the full - * packed resource containing the corresponding stream. - * - * Each of the streams to write may be in any location supported by the - * resource-handling code (specifically, read_stream_list()), such as the - * contents of external file that has been logically added to the output WIM, or - * a stream in another WIM file that has been imported, or even stream in the - * "same" WIM file of which a modified copy is being written. In the case that - * a stream 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 + * 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 on all streams in @stream_list as well - * as any other streams not in @stream_list that will be in the output WIM file, - * but not on any other streams in the output WIM's lookup table or sharing a - * packed resource with a stream in @stream_list. Still furthermore, if - * on-the-fly deduplication of streams is possible, then all streams in - * @stream_list must also be linked by @lookup_table_list along with any other - * streams that have @will_be_in_output_wim set. - * - * This function handles on-the-fly deduplication of streams for which SHA1 - * message digests have not yet been calculated and it is therefore known - * whether such streams are already in @stream_list or in the WIM's lookup table - * at all. If @lookup_table is non-NULL, then each stream in @stream_list that + * @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 stream already being written or one that would be - * filtered out of the output WIM using stream_filtered() with the context - * @filter_ctx. Each such duplicate stream will be removed from @stream_list, its - * reference count transfered to the pre-existing duplicate stream, its memory - * freed, and will not be written. Alternatively, if a stream in @stream_list - * is a duplicate with any stream in @lookup_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 @lookup_table_list. - * - * Returns 0 if all streams were written successfully (or did not need to be - * written); otherwise a non-zero error code. + * 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. */ static int -write_stream_list(struct list_head *stream_list, - struct filedes *out_fd, - int write_resource_flags, - int out_ctype, - u32 out_chunk_size, - unsigned num_threads, - struct wim_lookup_table *lookup_table, - struct filter_context *filter_ctx, - wimlib_progress_func_t progress_func) +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 write_streams_ctx ctx; - struct list_head raw_copy_resources; + struct write_blobs_ctx ctx; + struct list_head raw_copy_blobs; + u64 num_nonraw_bytes; wimlib_assert((write_resource_flags & - (WRITE_RESOURCE_FLAG_PACK_STREAMS | + (WRITE_RESOURCE_FLAG_SOLID | WRITE_RESOURCE_FLAG_PIPABLE)) != - (WRITE_RESOURCE_FLAG_PACK_STREAMS | + (WRITE_RESOURCE_FLAG_SOLID | WRITE_RESOURCE_FLAG_PIPABLE)); - remove_zero_length_streams(stream_list); + validate_blob_list(blob_list); - if (list_empty(stream_list)) { - DEBUG("No streams to write."); + if (list_empty(blob_list)) return 0; - } - memset(&ctx, 0, sizeof(ctx)); + /* 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); - /* Pre-sorting the streams is required for compute_stream_list_stats(). - * Afterwards, read_stream_list() need not sort them again. */ - ret = sort_stream_list_by_sequential_order(stream_list, - offsetof(struct wim_lookup_table_entry, - write_streams_list)); - if (ret) - return ret; + memset(&ctx, 0, sizeof(ctx)); ctx.out_fd = out_fd; - ctx.lookup_table = lookup_table; + 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 (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - wimlib_assert(out_chunk_size != 0); - if (out_chunk_size <= STACK_MAX) { - ctx.chunk_buf = alloca(out_chunk_size); - } else { - ctx.chunk_buf = MALLOC(out_chunk_size); - if (ctx.chunk_buf == NULL) { - ret = WIMLIB_ERR_NOMEM; - goto out_destroy_context; - } - } - } - ctx.chunk_buf_filled = 0; + /* + * 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. + */ - compute_stream_list_stats(stream_list, &ctx); + ret = sort_blob_list_by_sequential_order(blob_list, + offsetof(struct blob_descriptor, + write_blobs_list)); + if (ret) + return ret; - ctx.progress_data.progress_func = progress_func; + ret = compute_blob_list_stats(blob_list, &ctx); + if (ret) + return ret; - ctx.num_bytes_to_compress = find_raw_copy_resources(stream_list, - write_resource_flags, - out_ctype, - out_chunk_size, - &raw_copy_resources); + 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."); + } - DEBUG("Writing stream list " - "(offset = %"PRIu64", write_resource_flags=0x%08x, " - "out_ctype=%d, out_chunk_size=%u, num_threads=%u, " - "total_bytes=%"PRIu64", num_bytes_to_compress=%"PRIu64")", - out_fd->offset, write_resource_flags, - out_ctype, out_chunk_size, num_threads, - ctx.progress_data.progress.write_streams.total_bytes, - ctx.num_bytes_to_compress); + ctx.progress_data.progfunc = progfunc; + ctx.progress_data.progctx = progctx; - if (ctx.num_bytes_to_compress == 0) { - DEBUG("No compression needed; skipping to raw copy!"); - goto out_write_raw_copy_resources; - } + num_nonraw_bytes = find_raw_copy_blobs(blob_list, write_resource_flags, + out_ctype, out_chunk_size, + &raw_copy_blobs); - /* Unless uncompressed output was required, allocate a chunk_compressor - * to do compression. There are serial and parallel implementations of - * the chunk_compressor interface. We default to parallel using the + /* 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 2000000 (heuristic value). */ - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { - + * 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 (ctx.num_bytes_to_compress >= 2000000) { + 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) { - DEBUG("Couldn't create parallel chunk compressor " - "(status %d)", ret); + if (ret > 0) { + WARNING("Couldn't create parallel chunk compressor: %"TS".\n" + " Falling back to single-threaded compression.", + wimlib_get_error_string(ret)); } } #endif @@ -1425,40 +1573,48 @@ write_stream_list(struct list_head *stream_list, else ctx.progress_data.progress.write_streams.num_threads = 1; - DEBUG("Actually using %u threads", - ctx.progress_data.progress.write_streams.num_threads); + 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); - INIT_LIST_HEAD(&ctx.pending_streams); + if (ret || num_nonraw_bytes == 0) + goto out_destroy_context; - if (ctx.progress_data.progress_func) { - (*ctx.progress_data.progress_func)(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - &ctx.progress_data.progress); - } + INIT_LIST_HEAD(&ctx.blobs_being_compressed); + + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { - if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { - ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress); + 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 streams needing to be compressed, using the - * specified callbacks to execute processing of the data. */ + /* Read the list of blobs needing to be compressed, using the specified + * callbacks to execute processing of the data. */ - struct read_stream_list_callbacks cbs = { - .begin_stream = write_stream_begin_read, - .begin_stream_ctx = &ctx, - .consume_chunk = write_stream_process_chunk, - .consume_chunk_ctx = &ctx, - .end_stream = write_stream_end_read, - .end_stream_ctx = &ctx, + 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_stream_list(stream_list, - offsetof(struct wim_lookup_table_entry, write_streams_list), - &cbs, - STREAM_LIST_ALREADY_SORTED | - VERIFY_STREAM_HASHES | - COMPUTE_MISSING_STREAM_HASHES); + 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; @@ -1467,58 +1623,44 @@ write_stream_list(struct list_head *stream_list, if (ret) goto out_destroy_context; - if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { + if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) { struct wim_reshdr reshdr; - struct wim_lookup_table_entry *lte; + struct blob_descriptor *blob; u64 offset_in_res; ret = end_write_resource(&ctx, &reshdr); if (ret) goto out_destroy_context; - DEBUG("Ending packed resource: %lu %lu %lu.", - reshdr.offset_in_wim, - reshdr.size_in_wim, - reshdr.uncompressed_size); - offset_in_res = 0; - list_for_each_entry(lte, &ctx.pending_streams, write_streams_list) { - lte->out_reshdr.size_in_wim = lte->size; - lte->out_reshdr.flags = filter_resource_flags(lte->flags); - lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS; - lte->out_reshdr.uncompressed_size = 0; - lte->out_reshdr.offset_in_wim = offset_in_res; - lte->out_res_offset_in_wim = reshdr.offset_in_wim; - lte->out_res_size_in_wim = reshdr.size_in_wim; - /*lte->out_res_uncompressed_size = reshdr.uncompressed_size;*/ - offset_in_res += lte->size; + 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); } -out_write_raw_copy_resources: - /* Copy any compressed resources for which the raw data can be reused - * without decompression. */ - ret = write_raw_copy_resources(&raw_copy_resources, ctx.out_fd, - &ctx.progress_data); - out_destroy_context: - if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX) - FREE(ctx.chunk_buf); FREE(ctx.chunk_csizes); if (ctx.compressor) ctx.compressor->destroy(ctx.compressor); - DEBUG("Done (ret=%d)", ret); return ret; } + static int -wim_write_stream_list(WIMStruct *wim, - struct list_head *stream_list, +write_file_data_blobs(WIMStruct *wim, + struct list_head *blob_list, int write_flags, unsigned num_threads, - struct filter_context *filter_ctx, - wimlib_progress_func_t progress_func) + struct filter_context *filter_ctx) { int out_ctype; u32 out_chunk_size; @@ -1526,101 +1668,96 @@ wim_write_stream_list(WIMStruct *wim, write_resource_flags = write_flags_to_resource_flags(write_flags); - if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { - out_chunk_size = wim->out_pack_chunk_size; - out_ctype = wim->out_pack_compression_type; + 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_stream_list(stream_list, - &wim->out_fd, - write_resource_flags, - out_ctype, - out_chunk_size, - num_threads, - wim->lookup_table, - filter_ctx, - progress_func); + 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); } +/* Write the contents of the specified blob as a WIM resource. */ static int -write_wim_resource(struct wim_lookup_table_entry *lte, +write_wim_resource(struct blob_descriptor *blob, struct filedes *out_fd, int out_ctype, u32 out_chunk_size, int write_resource_flags) { - LIST_HEAD(stream_list); - list_add(<e->write_streams_list, &stream_list); - lte->will_be_in_output_wim = 1; - return write_stream_list(&stream_list, - out_fd, - write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS, - out_ctype, - out_chunk_size, - 1, - NULL, - NULL, - NULL); -} - + 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); +} + +/* Write the contents of the specified buffer as a WIM resource. */ int -write_wim_resource_from_buffer(const void *buf, size_t buf_size, - int reshdr_flags, struct filedes *out_fd, +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, + u8 *hash_ret, int write_resource_flags) { int ret; - struct wim_lookup_table_entry *lte; + struct blob_descriptor blob; - /* Set up a temporary lookup table entry to provide to - * write_wim_resource(). */ - - lte = new_lookup_table_entry(); - if (lte == NULL) - return WIMLIB_ERR_NOMEM; - - lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte->attached_buffer = (void*)buf; - lte->size = buf_size; - lte->flags = reshdr_flags; - - if (write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) { - sha1_buffer(buf, buf_size, lte->hash); - lte->unhashed = 0; - } else { - lte->unhashed = 1; + if (unlikely(buf_size == 0)) { + zero_reshdr(out_reshdr); + if (hash_ret) + copy_hash(hash_ret, zero_hash); + return 0; } - ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size, + 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) - goto out_free_lte; + return ret; - copy_reshdr(out_reshdr, <e->out_reshdr); + copy_reshdr(out_reshdr, &blob.out_reshdr); - if (hash) - copy_hash(hash, lte->hash); - ret = 0; -out_free_lte: - lte->resource_location = RESOURCE_NONEXISTENT; - free_lookup_table_entry(lte); - return ret; + if (hash_ret) + copy_hash(hash_ret, blob.hash); + return 0; } -struct stream_size_table { +struct blob_size_table { struct hlist_head *array; size_t num_entries; size_t capacity; }; static int -init_stream_size_table(struct stream_size_table *tab, size_t capacity) +init_blob_size_table(struct blob_size_table *tab, size_t capacity) { tab->array = CALLOC(capacity, sizeof(tab->array[0])); if (tab->array == NULL) @@ -1631,110 +1768,110 @@ init_stream_size_table(struct stream_size_table *tab, size_t capacity) } static void -destroy_stream_size_table(struct stream_size_table *tab) +destroy_blob_size_table(struct blob_size_table *tab) { FREE(tab->array); } static int -stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) +blob_size_table_insert(struct blob_descriptor *blob, void *_tab) { - struct stream_size_table *tab = _tab; + struct blob_size_table *tab = _tab; size_t pos; - struct wim_lookup_table_entry *same_size_lte; - struct hlist_node *tmp; - - pos = hash_u64(lte->size) % tab->capacity; - lte->unique_size = 1; - hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) { - if (same_size_lte->size == lte->size) { - lte->unique_size = 0; - same_size_lte->unique_size = 0; + 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; } } - hlist_add_head(<e->hash_list_2, &tab->array[pos]); + hlist_add_head(&blob->hash_list_2, &tab->array[pos]); tab->num_entries++; return 0; } -struct find_streams_ctx { +struct find_blobs_ctx { WIMStruct *wim; int write_flags; - struct list_head stream_list; - struct stream_size_table stream_size_tab; + struct list_head blob_list; + struct blob_size_table blob_size_tab; }; static void -reference_stream_for_write(struct wim_lookup_table_entry *lte, - struct list_head *stream_list, u32 nref) +reference_blob_for_write(struct blob_descriptor *blob, + struct list_head *blob_list, u32 nref) { - if (!lte->will_be_in_output_wim) { - lte->out_refcnt = 0; - list_add_tail(<e->write_streams_list, stream_list); - lte->will_be_in_output_wim = 1; + 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; } - lte->out_refcnt += nref; + blob->out_refcnt += nref; } static int -fully_reference_stream_for_write(struct wim_lookup_table_entry *lte, - void *_stream_list) +fully_reference_blob_for_write(struct blob_descriptor *blob, void *_blob_list) { - struct list_head *stream_list = _stream_list; - lte->will_be_in_output_wim = 0; - reference_stream_for_write(lte, stream_list, lte->refcnt); + 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 int -inode_find_streams_to_reference(const struct wim_inode *inode, - const struct wim_lookup_table *table, - struct list_head *stream_list) +inode_find_blobs_to_reference(const struct wim_inode *inode, + const struct blob_table *table, + struct list_head *blob_list) { - struct wim_lookup_table_entry *lte; - unsigned i; - wimlib_assert(inode->i_nlink > 0); - for (i = 0; i <= inode->i_num_ads; i++) { - lte = inode_stream_lte(inode, i, table); - if (lte) - reference_stream_for_write(lte, stream_list, - inode->i_nlink); - else if (!is_zero_hash(inode_stream_hash(inode, i))) - return WIMLIB_ERR_RESOURCE_NOT_FOUND; + 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 { + hash = stream_hash(&inode->i_streams[i]); + if (!is_zero_hash(hash)) + return blob_not_found_error(inode, hash); + } } return 0; } static int -do_stream_set_not_in_output_wim(struct wim_lookup_table_entry *lte, void *_ignore) +do_blob_set_not_in_output_wim(struct blob_descriptor *blob, void *_ignore) { - lte->will_be_in_output_wim = 0; + blob->will_be_in_output_wim = 0; return 0; } static int -image_find_streams_to_reference(WIMStruct *wim) +image_find_blobs_to_reference(WIMStruct *wim) { struct wim_image_metadata *imd; struct wim_inode *inode; - struct wim_lookup_table_entry *lte; - struct list_head *stream_list; + struct blob_descriptor *blob; + struct list_head *blob_list; int ret; imd = wim_get_current_image_metadata(wim); - image_for_each_unhashed_stream(lte, imd) - lte->will_be_in_output_wim = 0; + image_for_each_unhashed_blob(blob, imd) + blob->will_be_in_output_wim = 0; - stream_list = wim->private; + blob_list = wim->private; image_for_each_inode(inode, imd) { - ret = inode_find_streams_to_reference(inode, - wim->lookup_table, - stream_list); + ret = inode_find_blobs_to_reference(inode, + wim->blob_table, + blob_list); if (ret) return ret; } @@ -1742,40 +1879,40 @@ image_find_streams_to_reference(WIMStruct *wim) } static int -prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim, - int image, - int streams_ok, - struct list_head *stream_list_ret) +prepare_unfiltered_list_of_blobs_in_output_wim(WIMStruct *wim, + int image, + int blobs_ok, + struct list_head *blob_list_ret) { int ret; - INIT_LIST_HEAD(stream_list_ret); + INIT_LIST_HEAD(blob_list_ret); - if (streams_ok && (image == WIMLIB_ALL_IMAGES || - (image == 1 && wim->hdr.image_count == 1))) + if (blobs_ok && (image == WIMLIB_ALL_IMAGES || + (image == 1 && wim->hdr.image_count == 1))) { - /* Fast case: Assume that all streams are being written and - * that the reference counts are correct. */ - struct wim_lookup_table_entry *lte; + /* 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_lookup_table_entry(wim->lookup_table, - fully_reference_stream_for_write, - stream_list_ret); + 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_stream(lte, imd) - fully_reference_stream_for_write(lte, stream_list_ret); + 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 streams referenced. */ - for_lookup_table_entry(wim->lookup_table, - do_stream_set_not_in_output_wim, NULL); - wim->private = stream_list_ret; - ret = for_image(wim, image, image_find_streams_to_reference); + * 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; } @@ -1784,59 +1921,57 @@ prepare_unfiltered_list_of_streams_in_output_wim(WIMStruct *wim, } struct insert_other_if_hard_filtered_ctx { - struct stream_size_table *tab; + struct blob_size_table *tab; struct filter_context *filter_ctx; }; static int -insert_other_if_hard_filtered(struct wim_lookup_table_entry *lte, void *_ctx) +insert_other_if_hard_filtered(struct blob_descriptor *blob, void *_ctx) { struct insert_other_if_hard_filtered_ctx *ctx = _ctx; - if (!lte->will_be_in_output_wim && - stream_hard_filtered(lte, ctx->filter_ctx)) - stream_size_table_insert(lte, ctx->tab); + if (!blob->will_be_in_output_wim && + blob_hard_filtered(blob, ctx->filter_ctx)) + blob_size_table_insert(blob, ctx->tab); return 0; } static int -determine_stream_size_uniquity(struct list_head *stream_list, - struct wim_lookup_table *lt, - struct filter_context *filter_ctx) +determine_blob_size_uniquity(struct list_head *blob_list, + struct blob_table *lt, + struct filter_context *filter_ctx) { int ret; - struct stream_size_table tab; - struct wim_lookup_table_entry *lte; + struct blob_size_table tab; + struct blob_descriptor *blob; - ret = init_stream_size_table(&tab, lt->capacity); + ret = init_blob_size_table(&tab, 9001); if (ret) return ret; - if (may_hard_filter_streams(filter_ctx)) { + if (may_hard_filter_blobs(filter_ctx)) { struct insert_other_if_hard_filtered_ctx ctx = { .tab = &tab, .filter_ctx = filter_ctx, }; - for_lookup_table_entry(lt, insert_other_if_hard_filtered, &ctx); + for_blob_in_table(lt, insert_other_if_hard_filtered, &ctx); } - list_for_each_entry(lte, stream_list, write_streams_list) - stream_size_table_insert(lte, &tab); + list_for_each_entry(blob, blob_list, write_blobs_list) + blob_size_table_insert(blob, &tab); - destroy_stream_size_table(&tab); + destroy_blob_size_table(&tab); return 0; } static void -filter_stream_list_for_write(struct list_head *stream_list, - struct filter_context *filter_ctx) +filter_blob_list_for_write(struct list_head *blob_list, + struct filter_context *filter_ctx) { - struct wim_lookup_table_entry *lte, *tmp; + struct blob_descriptor *blob, *tmp; - list_for_each_entry_safe(lte, tmp, - stream_list, write_streams_list) - { - int status = stream_filtered(lte, filter_ctx); + list_for_each_entry_safe(blob, tmp, blob_list, write_blobs_list) { + int status = blob_filtered(blob, filter_ctx); if (status == 0) { /* Not filtered. */ @@ -1846,19 +1981,19 @@ filter_stream_list_for_write(struct list_head *stream_list, /* Soft filtered. */ } else { /* Hard filtered. */ - lte->will_be_in_output_wim = 0; - list_del(<e->lookup_table_list); + blob->will_be_in_output_wim = 0; + list_del(&blob->blob_table_list); } - list_del(<e->write_streams_list); + list_del(&blob->write_blobs_list); } } } /* - * prepare_stream_list_for_write() - + * prepare_blob_list_for_write() - * - * Prepare the list of streams to write for writing a WIM containing the - * specified image(s) with the specified write flags. + * 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. @@ -1869,165 +2004,158 @@ filter_stream_list_for_write(struct list_head *stream_list, * @write_flags * WIMLIB_WRITE_FLAG_* flags for the write operation: * - * STREAMS_OK: For writes of all images, assume that all streams in the - * lookup table of @wim and the per-image lists of unhashed streams should - * be taken as-is, and image metadata should not be searched for - * references. This does not exclude filtering with OVERWRITE and - * SKIP_EXTERNAL_WIMS, below. + * 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. * - * OVERWRITE: Streams already present in @wim shall not be returned in - * @stream_list_ret. + * APPEND: Blobs already present in @wim shall not be returned in + * @blob_list_ret. * - * SKIP_EXTERNAL_WIMS: Streams already present in a WIM file, but not - * @wim, shall be be returned in neither @stream_list_ret nor - * @lookup_table_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. * - * @stream_list_ret - * List of streams, linked by write_streams_list, that need to be written - * will be returned here. + * @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 streams will be written; - * it does not take into account that they may become duplicates when - * actually hashed. + * 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. * - * @lookup_table_list_ret - * List of streams, linked by lookup_table_list, that need to be included - * in the WIM's lookup table will be returned here. This will be a - * superset of the streams in @stream_list_ret. + * @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 @stream_list_ret if and only if - * WIMLIB_WRITE_FLAG_OVERWRITE was specified in @write_flags and some of - * the streams that would otherwise need to be written were already located - * in the WIM file. + * 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 streams in this list will have @out_refcnt set to the number of - * references to the stream in the output WIM. If + * 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 stream filter status with stream_filtered() is + * 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 stream entries - * inserted into @lookup_table_list_ret and to 0 in all stream entries in the - * lookup table of @wim not inserted into @lookup_table_list_ret. + * 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 stream entries in - * @stream_list_ret that have unique size among all stream entries in - * @stream_list_ret and among all stream entries in the lookup table of @wim - * that are ineligible for being written due to filtering. + * 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_stream_list_for_write(WIMStruct *wim, int image, - int write_flags, - struct list_head *stream_list_ret, - struct list_head *lookup_table_list_ret, - struct filter_context *filter_ctx_ret) +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 wim_lookup_table_entry *lte; + struct blob_descriptor *blob; filter_ctx_ret->write_flags = write_flags; filter_ctx_ret->wim = wim; - ret = prepare_unfiltered_list_of_streams_in_output_wim( + ret = prepare_unfiltered_list_of_blobs_in_output_wim( wim, image, write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK, - stream_list_ret); + blob_list_ret); if (ret) return ret; - INIT_LIST_HEAD(lookup_table_list_ret); - list_for_each_entry(lte, stream_list_ret, write_streams_list) - list_add_tail(<e->lookup_table_list, lookup_table_list_ret); + 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); - ret = determine_stream_size_uniquity(stream_list_ret, wim->lookup_table, - filter_ctx_ret); + ret = determine_blob_size_uniquity(blob_list_ret, wim->blob_table, + filter_ctx_ret); if (ret) return ret; - if (may_filter_streams(filter_ctx_ret)) - filter_stream_list_for_write(stream_list_ret, filter_ctx_ret); + if (may_filter_blobs(filter_ctx_ret)) + filter_blob_list_for_write(blob_list_ret, filter_ctx_ret); return 0; } static int -write_wim_streams(WIMStruct *wim, int image, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func, - struct list_head *stream_list_override, - struct list_head *lookup_table_list_ret) +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 _stream_list; - struct list_head *stream_list; - struct wim_lookup_table_entry *lte; + 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 (stream_list_override == NULL) { - /* Normal case: prepare stream list from image(s) being written. + if (blob_list_override == NULL) { + /* Normal case: prepare blob list from image(s) being written. */ - stream_list = &_stream_list; + blob_list = &_blob_list; filter_ctx = &_filter_ctx; - ret = prepare_stream_list_for_write(wim, image, write_flags, - stream_list, - lookup_table_list_ret, - 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 stream list already explicitly provided. Use existing + * use blob list already explicitly provided. Use existing * reference counts. */ - stream_list = stream_list_override; + blob_list = blob_list_override; filter_ctx = NULL; - INIT_LIST_HEAD(lookup_table_list_ret); - list_for_each_entry(lte, stream_list, write_streams_list) { - lte->out_refcnt = lte->refcnt; - lte->will_be_in_output_wim = 1; - lte->unique_size = 0; - list_add_tail(<e->lookup_table_list, lookup_table_list_ret); + 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); } } - return wim_write_stream_list(wim, - stream_list, + return write_file_data_blobs(wim, + blob_list, write_flags, num_threads, - filter_ctx, - progress_func); + filter_ctx); } static int -write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, - wimlib_progress_func_t progress_func) +write_metadata_resources(WIMStruct *wim, int image, int write_flags) { int ret; int start_image; int end_image; int write_resource_flags; - if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) { - DEBUG("Not writing any metadata resources."); + if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) return 0; - } write_resource_flags = write_flags_to_resource_flags(write_flags); - write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS; - - DEBUG("Writing metadata resources (offset=%"PRIu64")", - wim->out_fd.offset); + write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); + ret = call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, + NULL, wim->progctx); + if (ret) + return ret; if (image == WIMLIB_ALL_IMAGES) { start_image = 1; @@ -2041,23 +2169,29 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, struct wim_image_metadata *imd; imd = wim->image_metadata[i - 1]; - /* Build a new metadata resource only if image was modified from - * the original (or was newly added). Otherwise just copy the - * existing one. */ - if (imd->modified) { - DEBUG("Image %u was modified; building and writing new " - "metadata resource", i); + 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 (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { - DEBUG("Image %u was not modified; re-using existing " - "metadata resource.", i); - stream_set_out_reshdr_for_reuse(imd->metadata_lte); + } 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; } else { - DEBUG("Image %u was not modified; copying existing " - "metadata resource.", i); - ret = write_wim_resource(imd->metadata_lte, + /* 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, @@ -2066,18 +2200,16 @@ write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, if (ret) return ret; } - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); - return 0; + + return call_progress(wim->progfunc, + WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, + NULL, wim->progctx); } static int open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) { - int raw_fd; - DEBUG("Opening \"%"TS"\" for writing.", path); - - raw_fd = topen(path, open_flags | O_BINARY, 0644); + 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; @@ -2091,63 +2223,59 @@ close_wim_writable(WIMStruct *wim, int write_flags) { int ret = 0; - if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) { - DEBUG("Closing WIM file."); + 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; } static int -cmp_streams_by_out_rspec(const void *p1, const void *p2) +cmp_blobs_by_out_rdesc(const void *p1, const void *p2) { - const struct wim_lookup_table_entry *lte1, *lte2; + const struct blob_descriptor *blob1, *blob2; - lte1 = *(const struct wim_lookup_table_entry**)p1; - lte2 = *(const struct wim_lookup_table_entry**)p2; + blob1 = *(const struct blob_descriptor**)p1; + blob2 = *(const struct blob_descriptor**)p2; - if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { - if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) { - if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim) - return cmp_u64(lte1->out_res_offset_in_wim, - lte2->out_res_offset_in_wim); + 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 (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + if (blob2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) return -1; } - return cmp_u64(lte1->out_reshdr.offset_in_wim, - lte2->out_reshdr.offset_in_wim); + return cmp_u64(blob1->out_reshdr.offset_in_wim, + blob2->out_reshdr.offset_in_wim); } static int -write_wim_lookup_table(WIMStruct *wim, int image, int write_flags, - struct wim_reshdr *out_reshdr, - struct list_head *lookup_table_list) +write_blob_table(WIMStruct *wim, int image, int write_flags, + struct list_head *blob_table_list) { int ret; - /* Set output resource metadata for streams already present in WIM. */ - if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { - struct wim_lookup_table_entry *lte; - list_for_each_entry(lte, lookup_table_list, lookup_table_list) - { - if (lte->resource_location == RESOURCE_IN_WIM && - lte->rspec->wim == wim) + /* 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) { - stream_set_out_reshdr_for_reuse(lte); + blob_set_out_reshdr_for_reuse(blob); } } } - ret = sort_stream_list(lookup_table_list, - offsetof(struct wim_lookup_table_entry, lookup_table_list), - cmp_streams_by_out_rspec); + ret = sort_blob_list(blob_table_list, + offsetof(struct blob_descriptor, blob_table_list), + cmp_blobs_by_out_rdesc); if (ret) return ret; @@ -2164,85 +2292,43 @@ write_wim_lookup_table(WIMStruct *wim, int image, int write_flags, end_image = image; } - /* Push metadata resource lookup table entries onto the front of - * the list in reverse order, so that they're written in order. + /* 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 wim_lookup_table_entry *metadata_lte; + struct blob_descriptor *metadata_blob; - metadata_lte = wim->image_metadata[i - 1]->metadata_lte; - wimlib_assert(metadata_lte->out_reshdr.flags & WIM_RESHDR_FLAG_METADATA); - metadata_lte->out_refcnt = 1; - list_add(&metadata_lte->lookup_table_list, lookup_table_list); + 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); } } - return write_wim_lookup_table_from_stream_list(lookup_table_list, - &wim->out_fd, - wim->hdr.part_number, - out_reshdr, - write_flags_to_resource_flags(write_flags)); + 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)); } /* - * finish_write(): - * - * Finish writing a WIM file: write the lookup table, xml data, and integrity - * table, then overwrite the WIM header. By default, closes the WIM file - * descriptor (@wim->out_fd) if successful. - * - * write_flags is a bitwise OR of the following: - * - * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: - * Include an integrity table. - * - * (public) WIMLIB_WRITE_FLAG_FSYNC: - * fsync() the output file before closing it. - * - * (public) WIMLIB_WRITE_FLAG_PIPABLE: - * Writing a pipable WIM, possibly to a pipe; include pipable WIM - * stream headers before the lookup table and XML data, and also - * write the WIM header at the end instead of seeking to the - * beginning. Can't be combined with - * WIMLIB_WRITE_FLAG_CHECK_INTEGRITY. - * - * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: - * Don't write the lookup table. - * - * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: - * When (if) writing the integrity table, re-use entries from the - * existing integrity table, if possible. - * - * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: - * After writing the XML data but before writing the integrity - * table, write a temporary WIM header and flush the stream so that - * the WIM is less likely to become corrupted upon abrupt program - * termination. - * (private) WIMLIB_WRITE_FLAG_HEADER_AT_END: - * Instead of overwriting the WIM header at the beginning of the - * file, simply append it to the end of the file. (Used when - * writing to pipe.) - * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR: - * Do not close the file descriptor @wim->out_fd on either success - * on failure. - * (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES: - * Use the existing stored in the in-memory XML - * information, rather than setting it to the offset of the XML - * data being written. + * 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, - wimlib_progress_func_t progress_func, - struct list_head *lookup_table_list) + struct list_head *blob_table_list) { - int ret; - off_t hdr_offset; int write_resource_flags; - off_t old_lookup_table_end; - off_t new_lookup_table_end; + off_t old_blob_table_end = 0; + struct integrity_table *old_integrity_table = NULL; + off_t new_blob_table_end; u64 xml_totalbytes; - - DEBUG("image=%d, write_flags=%08x", image, write_flags); + int ret; write_resource_flags = write_flags_to_resource_flags(write_flags); @@ -2250,24 +2336,41 @@ finish_write(WIMStruct *wim, int image, int write_flags, * 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... */ - if (wim->hdr.boot_idx == 0) { - zero_reshdr(&wim->hdr.boot_metadata_reshdr); + * marked as bootable. */ + if (wim->out_hdr.boot_idx == 0) { + zero_reshdr(&wim->out_hdr.boot_metadata_reshdr); } else { - copy_reshdr(&wim->hdr.boot_metadata_reshdr, + copy_reshdr(&wim->out_hdr.boot_metadata_reshdr, &wim->image_metadata[ - wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr); - } - - /* Write lookup table. (Save old position first.) */ - old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + - wim->hdr.lookup_table_reshdr.size_in_wim; - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_wim_lookup_table(wim, image, write_flags, - &wim->hdr.lookup_table_reshdr, - lookup_table_list); + 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) - return ret; + goto out; } /* Write XML data. */ @@ -2275,53 +2378,66 @@ finish_write(WIMStruct *wim, int image, int write_flags, 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->hdr.xml_data_reshdr, + &wim->out_hdr.xml_data_reshdr, write_resource_flags); if (ret) - return ret; + goto out; - /* Write integrity table (optional). */ - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { + /* 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->hdr, sizeof(struct wim_header)); + 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_at_offset(&checkpoint_hdr, - &wim->out_fd, 0); + ret = write_wim_header(&checkpoint_hdr, &wim->out_fd, 0); if (ret) - return ret; + goto out; } - if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) - old_lookup_table_end = 0; - - new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + - wim->hdr.lookup_table_reshdr.size_in_wim; + new_blob_table_end = wim->out_hdr.blob_table_reshdr.offset_in_wim + + wim->out_hdr.blob_table_reshdr.size_in_wim; ret = write_integrity_table(wim, - new_lookup_table_end, - old_lookup_table_end, - progress_func); + new_blob_table_end, + old_blob_table_end, + old_integrity_table); if (ret) - return ret; + goto out; } else { /* No integrity table. */ - zero_reshdr(&wim->hdr.integrity_table_reshdr); + zero_reshdr(&wim->out_hdr.integrity_table_reshdr); } /* 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->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; - hdr_offset = 0; - if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END) - hdr_offset = wim->out_fd.offset; - DEBUG("Writing new header @ %"PRIu64".", hdr_offset); - ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset); + 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) - return 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) && + errno != EINVAL) /* allow compaction on untruncatable files, + e.g. block devices */ + { + ERROR_WITH_ERRNO("Failed to truncate the output WIM file"); + goto out; + } + } /* Possibly sync file data to disk before closing. On POSIX systems, it * is necessary to do this before using rename() to overwrite an @@ -2329,45 +2445,51 @@ finish_write(WIMStruct *wim, int image, int write_flags, * 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) { - DEBUG("Syncing WIM file."); if (fsync(wim->out_fd.fd)) { ERROR_WITH_ERRNO("Error syncing data to WIM file"); - return WIMLIB_ERR_WRITE; + goto out; } } + ret = WIMLIB_ERR_WRITE; if (close_wim_writable(wim, write_flags)) { ERROR_WITH_ERRNO("Failed to close the output WIM file"); - return WIMLIB_ERR_WRITE; + goto out; } - return 0; + ret = 0; +out: + free_integrity_table(old_integrity_table); + return ret; } #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) + +/* Set advisory lock on WIM file (if not already done so) */ int -lock_wim(WIMStruct *wim, int fd) +lock_wim_for_append(WIMStruct *wim) { - int ret = 0; - if (fd != -1 && !wim->wim_locked) { - ret = flock(fd, LOCK_EX | LOCK_NB); - if (ret != 0) { - if (errno == EWOULDBLOCK) { - ERROR("`%"TS"' is already being modified or has been " - "mounted read-write\n" - " by another process!", wim->filename); - ret = WIMLIB_ERR_ALREADY_LOCKED; - } else { - WARNING_WITH_ERRNO("Failed to lock `%"TS"'", - wim->filename); - ret = 0; - } - } else { - wim->wim_locked = 1; - } + if (wim->locked_for_append) + return 0; + if (!flock(wim->in_fd.fd, LOCK_EX | LOCK_NB)) { + wim->locked_for_append = 1; + return 0; + } + if (errno != EWOULDBLOCK) + return 0; + return WIMLIB_ERR_ALREADY_LOCKED; +} + +/* Remove advisory lock on WIM file (if present) */ +void +unlock_wim_for_append(WIMStruct *wim) +{ + if (wim->locked_for_append) { + flock(wim->in_fd.fd, LOCK_UN); + wim->locked_for_append = 0; } - return ret; } #endif @@ -2387,9 +2509,9 @@ lock_wim(WIMStruct *wim, int fd) * 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 lookup table and - * XML data resource entries. This is because this information cannot be - * determined until the lookup table and XML data have been written. + * 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). @@ -2399,14 +2521,17 @@ lock_wim(WIMStruct *wim, int fd) * 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). * - * - The format of resources, or streams, has been modified to allow them to be - * used before the "lookup table" has been read. Each stream is prefixed with - * a `struct pwm_stream_hdr' that is basically an abbreviated form of `struct - * wim_lookup_table_entry_disk' that only contains the SHA1 message digest, - * uncompressed stream size, and flags that indicate whether the stream is - * compressed. The data of uncompressed streams then follows literally, while - * the data of compressed streams follows in a modified format. Compressed - * streams do not begin with a chunk table, since the chunk table cannot be + * - Solid resources are not allowed. Each blob is always stored in its own + * resource. + * + * - 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 @@ -2414,29 +2539,29 @@ lock_wim(WIMStruct *wim, int fd) * `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). * - * - Metadata resources always come before other file resources (streams). - * (This does not by itself constitute an incompatibility with normal WIMs, - * since this is valid in normal WIMs.) + * - 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.) * - * - At least up to the end of the file resources, all components must be packed - * as tightly as possible; there cannot be any "holes" in the WIM. (This does + * - 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.) * - * Note: the lookup table, XML data, and header at the end are not used when + * 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. * * Layout of pipable WIM: * * ---------+----------+--------------------+----------------+--------------+-----------+--------+ - * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header | + * | Header | XML data | Metadata resources | File resources | Blob table | XML data | Header | * ---------+----------+--------------------+----------------+--------------+-----------+--------+ * * Layout of normal WIM: * * +--------+-----------------------------+-------------------------+ - * | Header | File and metadata resources | Lookup table | XML data | + * | Header | File and metadata resources | Blob table | XML data | * +--------+-----------------------------+-------------------------+ * * An optional integrity table can follow the final XML data in both normal and @@ -2451,56 +2576,82 @@ lock_wim(WIMStruct *wim, int fd) */ static int write_pipable_wim(WIMStruct *wim, int image, int write_flags, - unsigned num_threads, wimlib_progress_func_t progress_func, - struct list_head *stream_list_override, - struct list_head *lookup_table_list_ret) + unsigned num_threads, + struct list_head *blob_list_override, + struct list_head *blob_table_list_ret) { int ret; struct wim_reshdr xml_reshdr; WARNING("Creating a pipable WIM, which will " "be incompatible\n" - " with Microsoft's software (wimgapi/imagex/Dism)."); + " 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 SHA1 message digests of files is not + * 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 stream to a pipable WIM, its SHA1 message digest - * needs to be known before the stream data is written. Therefore, - * before getting much farther, we need to pre-calculate the SHA1 - * message digests of all streams that will be written. */ - ret = wim_checksum_unhashed_streams(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; /* Write extra copy of the XML data. */ ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, - &xml_reshdr, - WRITE_RESOURCE_FLAG_PIPABLE); + &xml_reshdr, WRITE_RESOURCE_FLAG_PIPABLE); if (ret) return ret; /* Write metadata resources for the image(s) being included in the * output WIM. */ - ret = write_wim_metadata_resources(wim, image, write_flags, - progress_func); + ret = write_metadata_resources(wim, image, write_flags); if (ret) return ret; - /* Write streams needed for the image(s) being included in the output - * WIM, or streams needed for the split WIM part. */ - return write_wim_streams(wim, image, write_flags, num_threads, - progress_func, stream_list_override, - lookup_table_list_ret); + /* 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); - /* The lookup table, XML data, and header at end are handled by + /* The blob table, XML data, and header at end are handled by * finish_write(). */ } +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; + } + } + return 0; +} + /* Write a standalone WIM or split WIM (SWM) part to a new file or to a file * descriptor. */ int @@ -2509,74 +2660,13 @@ write_wim_part(WIMStruct *wim, int image, int write_flags, unsigned num_threads, - wimlib_progress_func_t progress_func, unsigned part_number, unsigned total_parts, - struct list_head *stream_list_override, + struct list_head *blob_list_override, const u8 *guid) { int ret; - struct wim_header hdr_save; - struct list_head lookup_table_list; - - if (total_parts == 1) - DEBUG("Writing standalone WIM."); - else - DEBUG("Writing split WIM part %u/%u", part_number, total_parts); - if (image == WIMLIB_ALL_IMAGES) - DEBUG("Including all images."); - else - DEBUG("Including image %d only.", image); - if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) - DEBUG("File descriptor: %d", *(const int*)path_or_fd); - else - DEBUG("Path: \"%"TS"\"", (const tchar*)path_or_fd); - DEBUG("Write flags: 0x%08x", write_flags); - - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - DEBUG("\tCHECK_INTEGRITY"); - - if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) - DEBUG("\tREBUILD"); - - if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - DEBUG("\tRECOMPRESS"); - - if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) - DEBUG("\tFSYNC"); - - if (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE) - DEBUG("\tFSYNC"); - - if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) - DEBUG("\tIGNORE_READONLY_FLAG"); - - if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) - DEBUG("\tPIPABLE"); - - if (write_flags & WIMLIB_WRITE_FLAG_NOT_PIPABLE) - DEBUG("\tNOT_PIPABLE"); - - if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) - DEBUG("\tPACK_STREAMS"); - - if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) - DEBUG("\tFILE_DESCRIPTOR"); - - if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) - DEBUG("\tNO_METADATA"); - - if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES) - DEBUG("\tUSE_EXISTING_TOTALBYTES"); - - if (num_threads == 0) - DEBUG("Number of threads: autodetect"); - else - DEBUG("Number of threads: %u", num_threads); - DEBUG("Progress function: %s", (progress_func ? "yes" : "no")); - DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect")); - DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ? - "specified" : "generate new")); + struct list_head blob_table_list; /* Internally, this is always called with a valid part number and total * parts. */ @@ -2608,231 +2698,244 @@ write_wim_part(WIMStruct *wim, WIMLIB_WRITE_FLAG_NOT_PIPABLE)) return WIMLIB_ERR_INVALID_PARAM; - /* Save previous header, then start initializing the new one. */ - memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); - - /* Set default integrity, pipable, and packed stream flags. */ - if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | - WIMLIB_WRITE_FLAG_NOT_PIPABLE))) - if (wim_is_pipable(wim)) { - DEBUG("WIM is pipable; default to PIPABLE."); - write_flags |= WIMLIB_WRITE_FLAG_PIPABLE; - } + /* 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)) { - DEBUG("Integrity table present; default to CHECK_INTEGRITY."); + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) { + if (wim_has_integrity_table(wim)) write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; - } + } + + /* 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 ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | - WIMLIB_WRITE_FLAG_PACK_STREAMS)) + WIMLIB_WRITE_FLAG_SOLID)) == (WIMLIB_WRITE_FLAG_PIPABLE | - WIMLIB_WRITE_FLAG_PACK_STREAMS)) + WIMLIB_WRITE_FLAG_SOLID)) { - ERROR("Cannot specify both PIPABLE and PACK_STREAMS!"); + ERROR("Solid compression is unsupported in pipable WIMs"); return WIMLIB_ERR_INVALID_PARAM; } - /* Set appropriate magic number. */ + /* Start initializing the new file header. */ + memset(&wim->out_hdr, 0, sizeof(wim->out_hdr)); + + /* Set the magic number. */ if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) - wim->hdr.magic = PWM_MAGIC; + wim->out_hdr.magic = PWM_MAGIC; else - wim->hdr.magic = WIM_MAGIC; + wim->out_hdr.magic = WIM_MAGIC; - /* Set appropriate version number. */ - if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) || + /* Set the version number. */ + if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) || wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS) - wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS; + wim->out_hdr.wim_version = WIM_VERSION_SOLID; else - wim->hdr.wim_version = WIM_VERSION_DEFAULT; + wim->out_hdr.wim_version = WIM_VERSION_DEFAULT; - /* Clear header flags that will be set automatically. */ - wim->hdr.flags &= ~(WIM_HDR_FLAG_METADATA_ONLY | - WIM_HDR_FLAG_RESOURCE_ONLY | - WIM_HDR_FLAG_SPANNED | - WIM_HDR_FLAG_WRITE_IN_PROGRESS); + /* 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 SPANNED header flag if writing part of a split WIM. */ + /* 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->hdr.flags |= WIM_HDR_FLAG_SPANNED; - - /* Set part number and total parts of split WIM. This will be 1 and 1 - * if the WIM is standalone. */ - wim->hdr.part_number = part_number; - wim->hdr.total_parts = total_parts; - - /* Set compression type if different. */ - if (wim->compression_type != wim->out_compression_type) { - ret = set_wim_hdr_cflags(wim->out_compression_type, &wim->hdr); - wimlib_assert(ret == 0); + 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; + } } - /* Set chunk size if different. */ - wim->hdr.chunk_size = wim->out_chunk_size; + /* Set the chunk size. */ + wim->out_hdr.chunk_size = wim->out_chunk_size; - /* Use GUID if specified; otherwise generate a new one. */ + /* Set the GUID. */ + if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID) + guid = wim->hdr.guid; if (guid) - memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN); - else if (!wim->guid_set_explicitly) - randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN); - - /* Clear references to resources that have not been written yet. */ - zero_reshdr(&wim->hdr.lookup_table_reshdr); - zero_reshdr(&wim->hdr.xml_data_reshdr); - zero_reshdr(&wim->hdr.boot_metadata_reshdr); - zero_reshdr(&wim->hdr.integrity_table_reshdr); - - /* Set image count and boot index correctly for single image writes. */ - if (image != WIMLIB_ALL_IMAGES) { - wim->hdr.image_count = 1; - if (wim->hdr.boot_idx == image) - wim->hdr.boot_idx = 1; - else - wim->hdr.boot_idx = 0; + copy_guid(wim->out_hdr.guid, guid); + else + generate_guid(wim->out_hdr.guid); + + /* 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; } - /* Split WIMs can't be bootable. */ - if (total_parts != 1) - wim->hdr.boot_idx = 0; + /* Update image stats if needed. */ + ret = update_image_stats(wim); + if (ret) + return ret; - /* Initialize output file descriptor. */ + /* Set up the output file descriptor. */ if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) { - /* File descriptor was explicitly provided. Return error if - * file descriptor is not seekable, unless writing a pipable WIM - * was requested. */ - wim->out_fd.fd = *(const int*)path_or_fd; - wim->out_fd.offset = 0; + /* 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_restore_hdr; + 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_restore_hdr; + goto out_cleanup; } } - } 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_restore_hdr; + goto out_cleanup; } /* Write initial header. This is merely a "dummy" header since it - * doesn't have all the information yet, so it will be overwritten later - * (unless writing a pipable WIM). */ + * 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->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; - ret = write_wim_header(&wim->hdr, &wim->out_fd); - wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS; + 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_restore_hdr; + goto out_cleanup; - /* Write metadata resources and streams. */ + /* Write file data and metadata resources. */ if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) { /* Default case: create a normal (non-pipable) WIM. */ - ret = write_wim_streams(wim, image, write_flags, num_threads, - progress_func, stream_list_override, - &lookup_table_list); + ret = write_file_data(wim, image, write_flags, + num_threads, + blob_list_override, + &blob_table_list); if (ret) - goto out_restore_hdr; + goto out_cleanup; - ret = write_wim_metadata_resources(wim, image, write_flags, - progress_func); + ret = write_metadata_resources(wim, image, write_flags); if (ret) - goto out_restore_hdr; + goto out_cleanup; } else { /* Non-default case: create pipable WIM. */ ret = write_pipable_wim(wim, image, write_flags, num_threads, - progress_func, stream_list_override, - &lookup_table_list); + blob_list_override, + &blob_table_list); if (ret) - goto out_restore_hdr; - write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END; + goto out_cleanup; } - - /* Write lookup table, XML data, and (optional) integrity table. */ - ret = finish_write(wim, image, write_flags, progress_func, - &lookup_table_list); -out_restore_hdr: - memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); + /* 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); - DEBUG("ret=%d", ret); return ret; } /* 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, - wimlib_progress_func_t progress_func) + int image, int write_flags, unsigned num_threads) { return write_wim_part(wim, path_or_fd, image, write_flags, - num_threads, progress_func, 1, 1, NULL, NULL); + num_threads, 1, 1, NULL, NULL); } /* API function documented in wimlib.h */ WIMLIBAPI int wimlib_write(WIMStruct *wim, const tchar *path, - int image, int write_flags, unsigned num_threads, - wimlib_progress_func_t progress_func) + int image, int write_flags, unsigned num_threads) { - if (!path) + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) return WIMLIB_ERR_INVALID_PARAM; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + if (path == NULL || path[0] == T('\0')) + return WIMLIB_ERR_INVALID_PARAM; - return write_standalone_wim(wim, path, image, write_flags, - num_threads, progress_func); + return write_standalone_wim(wim, path, image, write_flags, num_threads); } /* API function documented in wimlib.h */ WIMLIBAPI int wimlib_write_to_fd(WIMStruct *wim, int fd, - int image, int write_flags, unsigned num_threads, - wimlib_progress_func_t progress_func) + int image, int write_flags, unsigned num_threads) { + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) + return WIMLIB_ERR_INVALID_PARAM; + if (fd < 0) return WIMLIB_ERR_INVALID_PARAM; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR; - return write_standalone_wim(wim, &fd, image, write_flags, - num_threads, progress_func); + return write_standalone_wim(wim, &fd, image, write_flags, num_threads); } +/* 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_modified(WIMStruct *wim) +any_images_changed(WIMStruct *wim) { + if (wim->image_deletion_occurred) + return true; for (int i = 0; i < wim->hdr.image_count; i++) - if (wim->image_metadata[i]->modified) + if (!is_image_unchanged_from_wim(wim->image_metadata[i], wim)) return true; return false; } static int -check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim) +check_resource_offset(struct blob_descriptor *blob, void *_wim) { const WIMStruct *wim = _wim; off_t end_offset = *(const off_t*)wim->private; - if (lte->resource_location == RESOURCE_IN_WIM && lte->rspec->wim == wim && - lte->rspec->offset_in_wim + lte->rspec->size_in_wim > end_offset) + 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; } /* Make sure no file or metadata resources are located after the XML data (or - * integrity table if present)--- otherwise we can't safely overwrite the WIM in - * place and we return WIMLIB_ERR_RESOURCE_ORDER. */ + * 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) { @@ -2840,175 +2943,250 @@ check_resource_offsets(WIMStruct *wim, off_t end_offset) unsigned i; wim->private = &end_offset; - ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim); + 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_lte, wim); + ret = check_resource_offset(wim->image_metadata[i]->metadata_blob, wim); if (ret) return ret; } return 0; } +static int +free_blob_if_invalidated(struct blob_descriptor *blob, void *_wim) +{ + 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 *wim, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads) { int ret; off_t old_wim_end; - u64 old_lookup_table_end, old_xml_begin, old_xml_end; - struct wim_header hdr_save; - struct list_head stream_list; - struct list_head lookup_table_list; + struct list_head blob_list; + struct list_head blob_table_list; struct filter_context filter_ctx; - DEBUG("Overwriting `%"TS"' in-place", wim->filename); - - /* Save original header so it can be restored in case of error */ - memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); - - /* Set default integrity flag. */ + /* 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; - /* Set WIM version if adding packed streams. */ - if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) - wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS; - - /* Set additional flags for overwrite. */ - write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE | - WIMLIB_WRITE_FLAG_STREAMS_OK; - - /* Make sure that the integrity table (if present) is after the XML - * data, and that there are no stream resources, metadata resources, or - * lookup tables after the XML data. Otherwise, these data would be - * 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_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim + - wim->hdr.lookup_table_reshdr.size_in_wim; - if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0 && - 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_restore_memory_hdr; - } - - if (old_lookup_table_end > old_xml_begin) { - WARNING("Didn't expect the lookup table to be after the XML data"); - ret = WIMLIB_ERR_RESOURCE_ORDER; - goto out_restore_memory_hdr; - } - - /* 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 (!wim->deletion_occurred && !any_images_modified(wim)) { - /* If no images have been modified and no images have been - * deleted, a new lookup table does not need to be written. We - * shall write the new XML data and optional integrity table - * immediately after the lookup table. Note that this may - * overwrite an existing integrity table. */ - DEBUG("Skipping writing lookup table " - "(no images modified or deleted)"); - old_wim_end = old_lookup_table_end; - write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | - WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; - } else if (wim->hdr.integrity_table_reshdr.offset_in_wim != 0) { - /* Old WIM has an integrity table; begin writing new streams - * after it. */ - old_wim_end = wim->hdr.integrity_table_reshdr.offset_in_wim + - wim->hdr.integrity_table_reshdr.size_in_wim; + /* 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; + + /* 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 { - /* No existing integrity table; begin writing new streams after - * the old XML data. */ - old_wim_end = old_xml_end; - } + 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; + if (wim->hdr.blob_table_reshdr.offset_in_wim == 0) + old_blob_table_end = WIM_HEADER_DISK_SIZE; + else + 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; + } - ret = check_resource_offsets(wim, old_wim_end); - if (ret) - goto out_restore_memory_hdr; + 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; + } - ret = prepare_stream_list_for_write(wim, WIMLIB_ALL_IMAGES, write_flags, - &stream_list, &lookup_table_list, - &filter_ctx); + ret = check_resource_offsets(wim, old_wim_end); + if (ret) + goto out; + + ret = prepare_blob_list_for_write(wim, WIMLIB_ALL_IMAGES, + write_flags, &blob_list, + &blob_table_list, &filter_ctx); + if (ret) + goto out; + + if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS) + wimlib_assert(list_empty(&blob_list)); + } + + /* Update image stats if needed. */ + ret = update_image_stats(wim); if (ret) - goto out_restore_memory_hdr; + goto out; ret = open_wim_writable(wim, wim->filename, O_RDWR); if (ret) - goto out_restore_memory_hdr; + goto out; - ret = lock_wim(wim, wim->out_fd.fd); + ret = lock_wim_for_append(wim); if (ret) goto out_close_wim; /* 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; @@ -3017,61 +3195,52 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); ret = WIMLIB_ERR_WRITE; - goto out_restore_physical_hdr; + goto out_restore_hdr; } - ret = wim_write_stream_list(wim, - &stream_list, - write_flags, - num_threads, - &filter_ctx, - progress_func); + ret = write_file_data_blobs(wim, &blob_list, write_flags, + num_threads, &filter_ctx); if (ret) goto out_truncate; - ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, - write_flags, progress_func); + ret = write_metadata_resources(wim, WIMLIB_ALL_IMAGES, write_flags); if (ret) goto out_truncate; - write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, - progress_func, &lookup_table_list); + &blob_table_list); if (ret) goto out_truncate; - wim->wim_locked = 0; + unlock_wim_for_append(wim); return 0; out_truncate: - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", - wim->filename, old_wim_end); + 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_physical_hdr: - (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd); +out_restore_hdr: + (void)write_wim_header_flags(wim->hdr.flags, &wim->out_fd); out_unlock_wim: - wim->wim_locked = 0; + unlock_wim_for_append(wim); out_close_wim: (void)close_wim_writable(wim, write_flags); -out_restore_memory_hdr: - memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); +out: + wim->being_compacted = 0; return ret; } static int -overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads) { size_t wim_name_len; int ret; - DEBUG("Overwriting `%"TS"' via a temporary file", wim->filename); - /* Write the WIM to a temporary file in the same directory as the * original WIM. */ wim_name_len = tstrlen(wim->filename); @@ -3081,19 +3250,23 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, tmpfile[wim_name_len + 9] = T('\0'); ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, - write_flags | WIMLIB_WRITE_FLAG_FSYNC, - num_threads, progress_func); + write_flags | + WIMLIB_WRITE_FLAG_FSYNC | + WIMLIB_WRITE_FLAG_RETAIN_GUID, + num_threads); if (ret) { tunlink(tmpfile); return ret; } - close_wim(wim); + if (filedes_valid(&wim->in_fd)) { + filedes_close(&wim->in_fd); + filedes_invalidate(&wim->in_fd); + } /* 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. */ - DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename); ret = trename(tmpfile, wim->filename); if (ret) { ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", @@ -3107,17 +3280,15 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, return WIMLIB_ERR_RENAME; } - if (progress_func) { - union wimlib_progress_info progress; - progress.rename.from = tmpfile; - progress.rename.to = wim->filename; - progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); - } - return 0; + 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); } -/* Determine if the specified WIM file may be updated by appending in-place - * rather than writing and replacing it with an entirely new file. */ +/* 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) { @@ -3125,8 +3296,9 @@ can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) if (write_flags & WIMLIB_WRITE_FLAG_REBUILD) return false; - /* Deletions cause full rebuild by default. */ - if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + /* 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 @@ -3134,14 +3306,6 @@ can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) if (wim_is_pipable(wim) || (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) return false; - /* wimlib allows multiple packs in a single WIM, but they don't seem to - * be compatible with WIMGAPI, so force all streams to be repacked if - * the WIM already may have contained a pack and PACK_STREAMS was - * requested. */ - if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS && - wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS) - 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. */ @@ -3155,21 +3319,31 @@ can_overwrite_wim_inplace(const WIMStruct *wim, int write_flags) /* API function documented in wimlib.h */ WIMLIBAPI int -wimlib_overwrite(WIMStruct *wim, int write_flags, - unsigned num_threads, - wimlib_progress_func_t progress_func) +wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads) { int ret; u32 orig_hdr_flags; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - - if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) + if (write_flags & ~WIMLIB_WRITE_MASK_PUBLIC) return WIMLIB_ERR_INVALID_PARAM; if (!wim->filename) return WIMLIB_ERR_NO_FILENAME; + 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; @@ -3179,12 +3353,12 @@ wimlib_overwrite(WIMStruct *wim, int write_flags, return ret; if (can_overwrite_wim_inplace(wim, write_flags)) { - ret = overwrite_wim_inplace(wim, write_flags, num_threads, - progress_func); + 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(wim, 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); }