X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=6c97c79604823415a831f7a7fecc303195f1ec9d;hp=abbe4c01fc4668bdaff71f1d1f3a647ea441a3cb;hb=71abd45bcc85ddb5cf60aba8bbeb53b40780f521;hpb=e8c3ca2d1d0cac3d64985b45a9f654d2029a7518 diff --git a/src/write.c b/src/write.c index abbe4c01..6c97c796 100644 --- a/src/write.c +++ b/src/write.c @@ -53,220 +53,222 @@ # include #endif -#include -#include #include - -#ifdef WITH_NTFS_3G -# include -# include -# include -# include -#endif +#include +#include +#include +#include #ifdef HAVE_ALLOCA_H # include -#else -# include #endif -#include #ifndef __WIN32__ # include /* for `struct iovec' */ #endif +static int +alloc_lzx_context(int write_resource_flags, struct wimlib_lzx_context **ctx_pp) +{ + struct wimlib_lzx_params params; + params.size_of_this = sizeof(params); + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_COMPRESS_SLOW) + params.algorithm = WIMLIB_LZX_ALGORITHM_SLOW; + else + params.algorithm = WIMLIB_LZX_ALGORITHM_FAST; + params.use_defaults = 1; + return wimlib_lzx_alloc_context(¶ms, ctx_pp); +} + +static unsigned +compress_chunk(const void * uncompressed_data, + unsigned uncompressed_len, + void *compressed_data, + int out_ctype, + struct wimlib_lzx_context *comp_ctx) +{ + switch (out_ctype) { + case WIMLIB_COMPRESSION_TYPE_XPRESS: + return wimlib_xpress_compress(uncompressed_data, + uncompressed_len, + compressed_data); + case WIMLIB_COMPRESSION_TYPE_LZX: + return wimlib_lzx_compress2(uncompressed_data, + uncompressed_len, + compressed_data, + comp_ctx); + default: + wimlib_assert(0); + return 0; + } +} + /* Chunk table that's located at the beginning of each compressed resource in * the WIM. (This is not the on-disk format; the on-disk format just has an * array of offsets.) */ struct chunk_table { - off_t file_offset; - u64 num_chunks; u64 original_resource_size; - u64 bytes_per_chunk_entry; + u64 num_chunks; u64 table_disk_size; - u64 cur_offset; - u64 *cur_offset_p; - u64 offsets[0]; + unsigned bytes_per_chunk_entry; + void *cur_offset_p; + union { + u32 cur_offset_u32; + u64 cur_offset_u64; + }; + /* Beginning of chunk offsets, in either 32-bit or 64-bit little endian + * integers, including the first offset of 0, which will not be written. + * */ + u8 offsets[] _aligned_attribute(8); }; -/* - * Allocates and initializes a chunk table, and reserves space for it in the - * output file. - */ +/* Allocate and initializes a chunk table, then reserve space for it in the + * output file unless writing a pipable resource. */ static int begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, - int out_fd, - off_t file_offset, - struct chunk_table **chunk_tab_ret) + struct filedes *out_fd, + struct chunk_table **chunk_tab_ret, + int resource_flags) { - u64 size = wim_resource_size(lte); - u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); - struct chunk_table *chunk_tab = CALLOC(1, alloc_size); + u64 size; + u64 num_chunks; + unsigned bytes_per_chunk_entry; + size_t alloc_size; + struct chunk_table *chunk_tab; + int ret; - DEBUG("Begin chunk table for stream with size %"PRIu64, size); + size = wim_resource_size(lte); + num_chunks = wim_resource_chunks(lte); + bytes_per_chunk_entry = (size > (1ULL << 32)) ? 8 : 4; + alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); + chunk_tab = CALLOC(1, alloc_size); if (!chunk_tab) { ERROR("Failed to allocate chunk table for %"PRIu64" byte " "resource", size); return WIMLIB_ERR_NOMEM; } - chunk_tab->file_offset = file_offset; chunk_tab->num_chunks = num_chunks; chunk_tab->original_resource_size = size; - chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4; + chunk_tab->bytes_per_chunk_entry = bytes_per_chunk_entry; chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry * (num_chunks - 1); - chunk_tab->cur_offset = 0; chunk_tab->cur_offset_p = chunk_tab->offsets; - if (full_write(out_fd, chunk_tab, - chunk_tab->table_disk_size) != chunk_tab->table_disk_size) - { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - FREE(chunk_tab); - return WIMLIB_ERR_WRITE; + /* We don't know the correct offsets yet; so just write zeroes to + * reserve space for the table, so we can go back to it later after + * we've written the compressed chunks following it. + * + * Special case: if writing a pipable WIM, compressed resources are in a + * modified format (see comment above write_pipable_wim()) and do not + * have a chunk table at the beginning, so don't reserve any space for + * one. */ + if (!(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { + ret = full_write(out_fd, chunk_tab->offsets, + chunk_tab->table_disk_size); + if (ret) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); + FREE(chunk_tab); + return ret; + } } *chunk_tab_ret = chunk_tab; return 0; } -/* - * compress_func_t- Pointer to a function to compresses a chunk - * of a WIM resource. This may be either - * wimlib_xpress_compress() (xpress-compress.c) or - * wimlib_lzx_compress() (lzx-compress.c). - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk, in bytes. - * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes. - * - * Returns the size of the compressed data written to @out in bytes, or 0 if the - * data could not be compressed to (@chunk_size - 1) bytes or fewer. - * - * As a special requirement, the compression code is optimized for the WIM - * format and therefore requires (@chunk_size <= 32768). - * - * As another special requirement, the compression code will read up to 8 bytes - * off the end of the @chunk array for performance reasons. The values of these - * bytes will not affect the output of the compression, but the calling code - * must make sure that the buffer holding the uncompressed chunk is actually at - * least (@chunk_size + 8) bytes, or at least that these extra bytes are in - * mapped memory that will not cause a memory access violation if accessed. - */ -typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, - void *out); - -static compress_func_t -get_compress_func(int out_ctype) +/* Add the offset for the next chunk to the chunk table being constructed for a + * compressed stream. */ +static void +chunk_tab_record_chunk(struct chunk_table *chunk_tab, unsigned out_chunk_size) { - if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - return wimlib_lzx_compress; - else - return wimlib_xpress_compress; + if (chunk_tab->bytes_per_chunk_entry == 4) { + *(le32*)chunk_tab->cur_offset_p = cpu_to_le32(chunk_tab->cur_offset_u32); + chunk_tab->cur_offset_p = (le32*)chunk_tab->cur_offset_p + 1; + chunk_tab->cur_offset_u32 += out_chunk_size; + } else { + *(le64*)chunk_tab->cur_offset_p = cpu_to_le64(chunk_tab->cur_offset_u64); + chunk_tab->cur_offset_p = (le64*)chunk_tab->cur_offset_p + 1; + chunk_tab->cur_offset_u64 += out_chunk_size; + } } -/* - * Writes a chunk of a WIM resource to an output file. - * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) - * @out_fd: File descriptor to write the chunk to. - * @compress: Compression function to use (NULL if writing uncompressed - * data). - * @chunk_tab: Pointer to chunk table being created. It is updated with the - * offset of the chunk we write. - * - * Returns 0 on success; nonzero on failure. - */ +/* Finishes a WIM chunk table and writes it to the output file at the correct + * offset. */ static int -write_wim_resource_chunk(const void * restrict chunk, - unsigned chunk_size, - int out_fd, - compress_func_t compress, - struct chunk_table * restrict chunk_tab) +finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, + struct filedes *out_fd, + off_t res_start_offset, + int write_resource_flags) { - const void *out_chunk; - unsigned out_chunk_size; - if (compress) { - void *compressed_chunk = alloca(chunk_size); + int ret; - out_chunk_size = (*compress)(chunk, chunk_size, compressed_chunk); - if (out_chunk_size) { - /* Write compressed */ - out_chunk = compressed_chunk; - } else { - /* Write uncompressed */ - out_chunk = chunk; - out_chunk_size = chunk_size; - } - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += out_chunk_size; + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + ret = full_write(out_fd, + chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + chunk_tab->table_disk_size); } else { - /* Write uncompressed */ - out_chunk = chunk; - out_chunk_size = chunk_size; + ret = full_pwrite(out_fd, + chunk_tab->offsets + + chunk_tab->bytes_per_chunk_entry, + chunk_tab->table_disk_size, + res_start_offset); } - if (full_write(out_fd, out_chunk, out_chunk_size) != out_chunk_size) { - ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); - return WIMLIB_ERR_WRITE; + if (ret) { + ERROR_WITH_ERRNO("Failed to write chunk table in compressed " + "file resource"); } - return 0; + return ret; } -/* - * Finishes a WIM chunk table and writes it to the output file at the correct - * offset. - * - * The final size of the full compressed resource is returned in the - * @compressed_size_p. +/* Write the header for a stream in a pipable WIM. */ static int -finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - int out_fd, u64 *compressed_size_p) +write_pwm_stream_header(const struct wim_lookup_table_entry *lte, + struct filedes *out_fd, + int additional_reshdr_flags) { - size_t bytes_written; + struct pwm_stream_hdr stream_hdr; + u32 reshdr_flags; + int ret; - if (chunk_tab->bytes_per_chunk_entry == 8) { - array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); + stream_hdr.magic = PWM_STREAM_MAGIC; + stream_hdr.uncompressed_size = cpu_to_le64(lte->resource_entry.original_size); + if (additional_reshdr_flags & PWM_RESHDR_FLAG_UNHASHED) { + zero_out_hash(stream_hdr.hash); } else { - for (u64 i = 0; i < chunk_tab->num_chunks; i++) - ((u32*)chunk_tab->offsets)[i] = - cpu_to_le32(chunk_tab->offsets[i]); - } - bytes_written = full_pwrite(out_fd, - (u8*)chunk_tab->offsets + chunk_tab->bytes_per_chunk_entry, - chunk_tab->table_disk_size, - chunk_tab->file_offset); - if (bytes_written != chunk_tab->table_disk_size) { - ERROR_WITH_ERRNO("Failed to write chunk table in compressed " - "file resource"); - return WIMLIB_ERR_WRITE; + wimlib_assert(!lte->unhashed); + copy_hash(stream_hdr.hash, lte->hash); } - *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; - return 0; + + reshdr_flags = lte->resource_entry.flags & ~WIM_RESHDR_FLAG_COMPRESSED; + reshdr_flags |= additional_reshdr_flags; + stream_hdr.flags = cpu_to_le32(reshdr_flags); + ret = full_write(out_fd, &stream_hdr, sizeof(stream_hdr)); + if (ret) + ERROR_WITH_ERRNO("Error writing stream header"); + return ret; } static int -seek_and_truncate(int out_fd, off_t offset) +seek_and_truncate(struct filedes *out_fd, off_t offset) { - if (lseek(out_fd, offset, SEEK_SET) == -1 || - ftruncate(out_fd, offset)) + if (filedes_seek(out_fd, offset) == -1 || + ftruncate(out_fd->fd, offset)) { ERROR_WITH_ERRNO("Failed to truncate output WIM file"); return WIMLIB_ERR_WRITE; - } else { - return 0; } + return 0; } static int -finalize_and_check_sha1(SHA_CTX * restrict sha_ctx, - struct wim_lookup_table_entry * restrict lte) +finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte) { u8 md[SHA1_HASH_SIZE]; + sha1_final(md, sha_ctx); if (lte->unhashed) { copy_hash(lte->hash, md); @@ -282,45 +284,108 @@ finalize_and_check_sha1(SHA_CTX * restrict sha_ctx, return 0; } - struct write_resource_ctx { - compress_func_t compress; + int out_ctype; + struct wimlib_lzx_context *comp_ctx; struct chunk_table *chunk_tab; - int out_fd; + struct filedes *out_fd; SHA_CTX sha_ctx; bool doing_sha; + int resource_flags; }; static int -write_resource_cb(const void *restrict chunk, size_t chunk_size, - void *restrict _ctx) +write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx) { struct write_resource_ctx *ctx = _ctx; + const void *out_chunk; + unsigned out_chunk_size; + int ret; if (ctx->doing_sha) sha1_update(&ctx->sha_ctx, chunk, chunk_size); - return write_wim_resource_chunk(chunk, chunk_size, - ctx->out_fd, ctx->compress, - ctx->chunk_tab); + + out_chunk = chunk; + out_chunk_size = chunk_size; + if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + void *compressed_chunk; + unsigned compressed_size; + + /* Compress the chunk. */ + compressed_chunk = alloca(chunk_size); + + compressed_size = compress_chunk(chunk, chunk_size, + compressed_chunk, + ctx->out_ctype, + ctx->comp_ctx); + /* Use compressed data if compression to less than input size + * was successful. */ + if (compressed_size) { + out_chunk = compressed_chunk; + out_chunk_size = compressed_size; + } + } + + if (ctx->chunk_tab) { + /* Update chunk table accounting. */ + chunk_tab_record_chunk(ctx->chunk_tab, out_chunk_size); + + /* If writing compressed chunks to a pipable WIM, before the + * chunk data write a chunk header that provides the compressed + * chunk size. */ + if (ctx->resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + struct pwm_chunk_hdr chunk_hdr = { + .compressed_size = cpu_to_le32(out_chunk_size), + }; + ret = full_write(ctx->out_fd, &chunk_hdr, + sizeof(chunk_hdr)); + if (ret) + goto error; + } + } + + /* Write the chunk data. */ + ret = full_write(ctx->out_fd, out_chunk, out_chunk_size); + if (ret) + goto error; + return 0; + +error: + ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); + return ret; } /* + * write_wim_resource()- + * * Write a resource to an output WIM. * - * @lte: Lookup table entry for the resource, which could be in another WIM, - * in an external file, or in another location. + * @lte: + * Lookup table entry for the resource, which could be in another WIM, in + * an external file, or in another location. * - * @out_fd: File descriptor opened to the output WIM. + * @out_fd: + * File descriptor opened to the output WIM. * - * @out_ctype: One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate - * which compression algorithm to use. + * @out_ctype: + * One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate which + * compression algorithm to use. * - * @out_res_entry: On success, this is filled in with the offset, flags, - * compressed size, and uncompressed size of the resource - * in the output WIM. + * @out_res_entry: + * On success, this is filled in with the offset, flags, compressed size, + * and uncompressed size of the resource in the output WIM. * - * @flags: WIMLIB_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed - * even if it could otherwise be copied directly from the input. + * @resource_flags: + * * WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed even + * if it could otherwise be copied directly from the input; + * * WIMLIB_WRITE_RESOURCE_FLAG_COMPRESS_SLOW to compress the data as much + * as possible; + * * WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE if writing a resource for a pipable WIM + * (and the output file descriptor may be a pipe). + * + * @comp_ctx: + * Location of LZX compression context pointer, which will be allocated or + * updated if needed. (Initialize to NULL.) * * Additional notes: The SHA1 message digest of the uncompressed data is * calculated (except when doing a raw copy --- see below). If the @unhashed @@ -330,36 +395,44 @@ write_resource_cb(const void *restrict chunk, size_t chunk_size, */ int write_wim_resource(struct wim_lookup_table_entry *lte, - int out_fd, int out_ctype, + struct filedes *out_fd, int out_ctype, struct resource_entry *out_res_entry, - int flags) + int resource_flags, + struct wimlib_lzx_context **comp_ctx) { struct write_resource_ctx write_ctx; + off_t res_start_offset; u64 read_size; - u64 new_size; - off_t offset; int ret; - flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS; + /* Mask out any irrelevant flags, since this function also uses this + * variable to store WIMLIB_READ_RESOURCE flags. */ + resource_flags &= WIMLIB_WRITE_RESOURCE_MASK; - /* Get current position in output WIM */ - offset = filedes_offset(out_fd); - if (offset == -1) { - ERROR_WITH_ERRNO("Can't get position in output WIM"); - return WIMLIB_ERR_WRITE; - } + /* Get current position in output WIM. */ + res_start_offset = out_fd->offset; /* If we are not forcing the data to be recompressed, and the input * resource is located in a WIM with the same compression type as that * desired other than no compression, we can simply copy the compressed * data without recompressing it. This also means we must skip - * calculating the SHA1, as we never will see the uncompressed data. */ - if (!(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && - lte->resource_location == RESOURCE_IN_WIM && + * calculating the SHA1, as we never will see the uncompressed data. */ + if (lte->resource_location == RESOURCE_IN_WIM && + out_ctype == wim_resource_compression_type(lte) && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && - wimlib_get_compression_type(lte->wim) == out_ctype) + !(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS)) { - flags |= WIMLIB_RESOURCE_FLAG_RAW; + /* Normally we can request a RAW_FULL read, but if we're reading + * from a pipable resource and writing a non-pipable resource or + * vice versa, then a RAW_CHUNKS read needs to be requested so + * that the written resource can be appropriately formatted. + * However, in neither case is any actual decompression needed. + */ + if (lte->is_pipable == !!(resource_flags & + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL; + else + resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS; write_ctx.doing_sha = false; read_size = lte->resource_entry.size; } else { @@ -368,27 +441,51 @@ write_wim_resource(struct wim_lookup_table_entry *lte, read_size = lte->resource_entry.original_size; } - /* Initialize the chunk table and set the compression function if - * compressing the resource. */ - if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || - (flags & WIMLIB_RESOURCE_FLAG_RAW)) { - write_ctx.compress = NULL; - write_ctx.chunk_tab = NULL; - } else { - write_ctx.compress = get_compress_func(out_ctype); - ret = begin_wim_resource_chunk_tab(lte, out_fd, - offset, - &write_ctx.chunk_tab); + + /* If the output resource is to be compressed, initialize the chunk + * table and set the function to use for chunk compression. Exceptions: + * no compression function is needed if doing a raw copy; also, no chunk + * table is needed if doing a *full* (not per-chunk) raw copy. */ + write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + write_ctx.chunk_tab = NULL; + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) { + write_ctx.out_ctype = out_ctype; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = alloc_lzx_context(resource_flags, comp_ctx); + if (ret) + goto out; + } + write_ctx.comp_ctx = *comp_ctx; + } + if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)) { + ret = begin_wim_resource_chunk_tab(lte, out_fd, + &write_ctx.chunk_tab, + resource_flags); + if (ret) + goto out; + } + } + + /* If writing a pipable resource, write the stream header and update + * @res_start_offset to be the end of the stream header. */ + if (resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + int reshdr_flags = 0; + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) + reshdr_flags |= WIM_RESHDR_FLAG_COMPRESSED; + ret = write_pwm_stream_header(lte, out_fd, reshdr_flags); if (ret) - return ret; + goto out_free_chunk_tab; + res_start_offset = out_fd->offset; } /* Write the entire resource by reading the entire resource and feeding * the data through the write_resource_cb function. */ write_ctx.out_fd = out_fd; + write_ctx.resource_flags = resource_flags; try_write_again: ret = read_resource_prefix(lte, read_size, - write_resource_cb, &write_ctx, flags); + write_resource_cb, &write_ctx, resource_flags); if (ret) goto out_free_chunk_tab; @@ -400,49 +497,108 @@ try_write_again: goto out_free_chunk_tab; } - out_res_entry->flags = lte->resource_entry.flags; - out_res_entry->original_size = wim_resource_size(lte); - out_res_entry->offset = offset; - if (flags & WIMLIB_RESOURCE_FLAG_RAW) { - /* Doing a raw write: The new compressed size is the same as - * the compressed size in the other WIM. */ - new_size = lte->resource_entry.size; - } else if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) { - /* Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size - * is the original size. */ - new_size = lte->resource_entry.original_size; - out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; - } else { - /* Using a different compression type: Call - * finish_wim_resource_chunk_tab() and it will provide the new - * compressed size. */ - ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fd, - &new_size); + /* Write chunk table if needed. */ + if (write_ctx.chunk_tab) { + ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, + out_fd, + res_start_offset, + resource_flags); if (ret) goto out_free_chunk_tab; - if (new_size >= wim_resource_size(lte)) { - /* Oops! We compressed the resource to larger than the original - * size. Write the resource uncompressed instead. */ - DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " - "writing uncompressed instead", - wim_resource_size(lte), new_size); - ret = seek_and_truncate(out_fd, offset); - if (ret) - goto out_free_chunk_tab; - write_ctx.compress = NULL; - write_ctx.doing_sha = false; - out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; - goto try_write_again; - } + } + + /* Fill in out_res_entry with information about the newly written + * resource. */ + out_res_entry->size = out_fd->offset - res_start_offset; + out_res_entry->flags = lte->resource_entry.flags; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) + out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED; + else out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; + out_res_entry->offset = res_start_offset; + out_res_entry->original_size = wim_resource_size(lte); + + /* Check for resources compressed to greater than their original size + * and write them uncompressed instead. (But never do this if writing + * to a pipe, and don't bother if we did a raw copy.) */ + if (out_res_entry->size > out_res_entry->original_size && + !(resource_flags & (WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE | + WIMLIB_READ_RESOURCE_FLAG_RAW))) + { + DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " + "writing uncompressed instead", + out_res_entry->original_size, out_res_entry->size); + ret = seek_and_truncate(out_fd, res_start_offset); + if (ret) + goto out_free_chunk_tab; + out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + FREE(write_ctx.chunk_tab); + write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + write_ctx.chunk_tab = NULL; + write_ctx.doing_sha = false; + goto try_write_again; + } + if (resource_flags & (WIMLIB_READ_RESOURCE_FLAG_RAW)) { + DEBUG("Copied raw compressed data " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_res_entry->original_size, out_res_entry->size, + out_res_entry->offset, out_res_entry->flags); + } else if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) { + DEBUG("Wrote compressed resource " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_res_entry->original_size, out_res_entry->size, + out_res_entry->offset, out_res_entry->flags); + } else { + DEBUG("Wrote uncompressed resource " + "(%"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + out_res_entry->original_size, + out_res_entry->offset, out_res_entry->flags); } - out_res_entry->size = new_size; ret = 0; out_free_chunk_tab: FREE(write_ctx.chunk_tab); +out: return ret; } +/* Like write_wim_resource(), but the resource is specified by a buffer of + * uncompressed data rather a lookup table entry; also writes the SHA1 hash of + * the buffer to @hash_ret. */ +int +write_wim_resource_from_buffer(const void *buf, size_t buf_size, + int reshdr_flags, struct filedes *out_fd, + int out_ctype, + struct resource_entry *out_res_entry, + u8 *hash_ret, int write_resource_flags, + struct wimlib_lzx_context **comp_ctx) +{ + /* Set up a temporary lookup table entry to provide to + * write_wim_resource(). */ + struct wim_lookup_table_entry lte; + int ret; + + lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte.attached_buffer = (void*)buf; + lte.resource_entry.original_size = buf_size; + lte.resource_entry.flags = reshdr_flags; + + if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) { + sha1_buffer(buf, buf_size, lte.hash); + lte.unhashed = 0; + } else { + lte.unhashed = 1; + } + + ret = write_wim_resource(<e, out_fd, out_ctype, out_res_entry, + write_resource_flags, comp_ctx); + if (ret) + return ret; + if (hash_ret) + copy_hash(hash_ret, lte.hash); + return 0; +} + + #ifdef ENABLE_MULTITHREADED_COMPRESSION /* Blocking shared queue (solves the producer-consumer problem) */ @@ -535,7 +691,8 @@ shared_queue_get(struct shared_queue *q) struct compressor_thread_params { struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; - compress_func_t compress; + int out_ctype; + struct wimlib_lzx_context *comp_ctx; }; #define MAX_CHUNKS_PER_MSG 2 @@ -546,7 +703,6 @@ struct message { u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; struct iovec out_chunks[MAX_CHUNKS_PER_MSG]; - size_t total_out_bytes; unsigned num_chunks; struct list_head list; bool complete; @@ -554,13 +710,18 @@ struct message { }; static void -compress_chunks(struct message *msg, compress_func_t compress) +compress_chunks(struct message *msg, int out_ctype, + struct wimlib_lzx_context *comp_ctx) { - msg->total_out_bytes = 0; for (unsigned i = 0; i < msg->num_chunks; i++) { - unsigned len = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i]); + unsigned len; + + len = compress_chunk(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i], + out_ctype, + comp_ctx); + void *out_chunk; unsigned out_len; if (len) { @@ -574,7 +735,6 @@ compress_chunks(struct message *msg, compress_func_t compress) } msg->out_chunks[i].iov_base = out_chunk; msg->out_chunks[i].iov_len = out_len; - msg->total_out_bytes += out_len; } } @@ -589,12 +749,11 @@ compressor_thread_proc(void *arg) struct compressor_thread_params *params = arg; struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; struct shared_queue *compressed_res_queue = params->compressed_res_queue; - compress_func_t compress = params->compress; struct message *msg; DEBUG("Compressor thread ready"); while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) { - compress_chunks(msg, compress); + compress_chunks(msg, params->out_ctype, params->comp_ctx); shared_queue_put(compressed_res_queue, msg); } DEBUG("Compressor thread terminating"); @@ -602,32 +761,52 @@ compressor_thread_proc(void *arg) } #endif /* ENABLE_MULTITHREADED_COMPRESSION */ +struct write_streams_progress_data { + wimlib_progress_func_t progress_func; + union wimlib_progress_info progress; + uint64_t next_progress; + WIMStruct *prev_wim_part; +}; + static void -do_write_streams_progress(union wimlib_progress_info *progress, - wimlib_progress_func_t progress_func, - uint64_t size_added, +do_write_streams_progress(struct write_streams_progress_data *progress_data, + struct wim_lookup_table_entry *lte, bool stream_discarded) { + union wimlib_progress_info *progress = &progress_data->progress; + bool new_wim_part; + if (stream_discarded) { - progress->write_streams.total_bytes -= size_added; - if (progress->write_streams._private != ~(uint64_t)0 && - progress->write_streams._private > progress->write_streams.total_bytes) + progress->write_streams.total_bytes -= wim_resource_size(lte); + if (progress_data->next_progress != ~(uint64_t)0 && + progress_data->next_progress > progress->write_streams.total_bytes) { - progress->write_streams._private = progress->write_streams.total_bytes; + progress_data->next_progress = progress->write_streams.total_bytes; } } else { - progress->write_streams.completed_bytes += size_added; + progress->write_streams.completed_bytes += wim_resource_size(lte); + } + new_wim_part = false; + if (lte->resource_location == RESOURCE_IN_WIM && + lte->wim != progress_data->prev_wim_part) + { + if (progress_data->prev_wim_part) { + new_wim_part = true; + progress->write_streams.completed_parts++; + } + progress_data->prev_wim_part = lte->wim; } progress->write_streams.completed_streams++; - if (progress_func && - progress->write_streams.completed_bytes >= progress->write_streams._private) + if (progress_data->progress_func + && (progress->write_streams.completed_bytes >= progress_data->next_progress + || new_wim_part)) { - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, - progress); - if (progress->write_streams._private == progress->write_streams.total_bytes) { - progress->write_streams._private = ~(uint64_t)0; + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + if (progress_data->next_progress == progress->write_streams.total_bytes) { + progress_data->next_progress = ~(uint64_t)0; } else { - progress->write_streams._private = + progress_data->next_progress = min(progress->write_streams.total_bytes, progress->write_streams.completed_bytes + progress->write_streams.total_bytes / 100); @@ -636,8 +815,9 @@ do_write_streams_progress(union wimlib_progress_info *progress, } struct serial_write_stream_ctx { - int out_fd; + struct filedes *out_fd; int out_ctype; + struct wimlib_lzx_context **comp_ctx; int write_resource_flags; }; @@ -647,9 +827,11 @@ serial_write_stream(struct wim_lookup_table_entry *lte, void *_ctx) struct serial_write_stream_ctx *ctx = _ctx; return write_wim_resource(lte, ctx->out_fd, ctx->out_ctype, <e->output_resource_entry, - ctx->write_resource_flags); + ctx->write_resource_flags, + ctx->comp_ctx); } + /* Write a list of streams, taking into account that some streams may be * duplicates that are checksummed and discarded on the fly, and also delegating * the actual writing of a stream to a function @write_stream_cb, which is @@ -659,8 +841,7 @@ do_write_stream_list(struct list_head *stream_list, struct wim_lookup_table *lookup_table, int (*write_stream_cb)(struct wim_lookup_table_entry *, void *), void *write_stream_ctx, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { int ret = 0; struct wim_lookup_table_entry *lte; @@ -678,19 +859,29 @@ do_write_stream_list(struct list_head *stream_list, * stream in the WIM we are writing. The stream must be * checksummed to know if we need to write it or not. */ struct wim_lookup_table_entry *tmp; - u32 orig_refcnt = lte->out_refcnt; + u32 orig_out_refcnt = lte->out_refcnt; ret = hash_unhashed_stream(lte, lookup_table, &tmp); if (ret) break; if (tmp != lte) { + /* We found a duplicate stream. 'lte' was + * freed, so replace it with the duplicate. */ lte = tmp; - /* We found a duplicate stream. */ - if (orig_refcnt != tmp->out_refcnt) { - /* We have already written, or are going - * to write, the duplicate stream. So - * just skip to the next stream. */ - DEBUG("Discarding duplicate stream of length %"PRIu64, + + /* 'out_refcnt' was transferred to the + * duplicate, and we can detect if the duplicate + * stream was already referenced for writing by + * checking if its 'out_refcnt' is higher than + * that of the original stream. In such cases, + * the current stream can be discarded. We can + * also discard the current stream if it was + * previously marked as filtered (e.g. already + * present in the WIM being written). */ + if (lte->out_refcnt > orig_out_refcnt || + lte->filtered) { + DEBUG("Discarding duplicate stream of " + "length %"PRIu64, wim_resource_size(lte)); lte->no_progress = 0; stream_discarded = true; @@ -723,10 +914,8 @@ do_write_stream_list(struct list_head *stream_list, } skip_to_progress: if (!lte->no_progress) { - do_write_streams_progress(progress, - progress_func, - wim_resource_size(lte), - stream_discarded); + do_write_streams_progress(progress_data, + lte, stream_discarded); } } return ret; @@ -735,23 +924,23 @@ do_write_stream_list(struct list_head *stream_list, static int do_write_stream_list_serial(struct list_head *stream_list, struct wim_lookup_table *lookup_table, - int out_fd, + struct filedes *out_fd, int out_ctype, + struct wimlib_lzx_context **comp_ctx, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { struct serial_write_stream_ctx ctx = { .out_fd = out_fd, .out_ctype = out_ctype, .write_resource_flags = write_resource_flags, + .comp_ctx = comp_ctx, }; return do_write_stream_list(stream_list, lookup_table, serial_write_stream, &ctx, - progress_func, - progress); + progress_data); } static inline int @@ -760,61 +949,91 @@ write_flags_to_resource_flags(int write_flags) int resource_flags = 0; if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; + resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS; + if (write_flags & WIMLIB_WRITE_FLAG_COMPRESS_SLOW) + resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_COMPRESS_SLOW; + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + resource_flags |= WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE; return resource_flags; } static int write_stream_list_serial(struct list_head *stream_list, struct wim_lookup_table *lookup_table, - int out_fd, + struct filedes *out_fd, int out_ctype, + struct wimlib_lzx_context **comp_ctx, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress) + struct write_streams_progress_data *progress_data) { - DEBUG("Writing stream list (serial version)"); + union wimlib_progress_info *progress = &progress_data->progress; + DEBUG("Writing stream list of size %"PRIu64" (serial version)", + progress->write_streams.total_streams); progress->write_streams.num_threads = 1; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } return do_write_stream_list_serial(stream_list, lookup_table, out_fd, out_ctype, + comp_ctx, write_resource_flags, - progress_func, - progress); + progress_data); } #ifdef ENABLE_MULTITHREADED_COMPRESSION static int -write_wim_chunks(struct message *msg, int out_fd, - struct chunk_table *chunk_tab) +write_wim_chunks(struct message *msg, struct filedes *out_fd, + struct chunk_table *chunk_tab, + int write_resource_flags) { - for (unsigned i = 0; i < msg->num_chunks; i++) { - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += msg->out_chunks[i].iov_len; + struct iovec *vecs; + struct pwm_chunk_hdr *chunk_hdrs; + unsigned nvecs; + int ret; + + for (unsigned i = 0; i < msg->num_chunks; i++) + chunk_tab_record_chunk(chunk_tab, msg->out_chunks[i].iov_len); + + if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) { + nvecs = msg->num_chunks; + vecs = msg->out_chunks; + } else { + /* Special case: If writing a compressed resource to a pipable + * WIM, prefix each compressed chunk with a header that gives + * its compressed size. */ + nvecs = msg->num_chunks * 2; + vecs = alloca(nvecs * sizeof(vecs[0])); + chunk_hdrs = alloca(msg->num_chunks * sizeof(chunk_hdrs[0])); + + for (unsigned i = 0; i < msg->num_chunks; i++) { + chunk_hdrs[i].compressed_size = cpu_to_le32(msg->out_chunks[i].iov_len); + vecs[i * 2].iov_base = &chunk_hdrs[i]; + vecs[i * 2].iov_len = sizeof(chunk_hdrs[i]); + vecs[i * 2 + 1].iov_base = msg->out_chunks[i].iov_base; + vecs[i * 2 + 1].iov_len = msg->out_chunks[i].iov_len; + } } - if (full_writev(out_fd, msg->out_chunks, - msg->num_chunks) != msg->total_out_bytes) - { + ret = full_writev(out_fd, vecs, nvecs); + if (ret) ERROR_WITH_ERRNO("Failed to write WIM chunks"); - return WIMLIB_ERR_WRITE; - } - return 0; + return ret; } struct main_writer_thread_ctx { struct list_head *stream_list; struct wim_lookup_table *lookup_table; - int out_fd; + struct filedes *out_fd; + off_t res_start_offset; int out_ctype; + struct wimlib_lzx_context **comp_ctx; int write_resource_flags; struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; size_t num_messages; - wimlib_progress_func_t progress_func; - union wimlib_progress_info *progress; + struct write_streams_progress_data *progress_data; struct list_head available_msgs; struct list_head outstanding_streams; @@ -961,21 +1180,35 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) { list_move(&msg->list, &ctx->available_msgs); if (msg->begin_chunk == 0) { - /* This is the first set of chunks. Leave space - * for the chunk table in the output file. */ - off_t cur_offset = filedes_offset(ctx->out_fd); - if (cur_offset == -1) - return WIMLIB_ERR_WRITE; + /* First set of chunks. */ + + /* Write pipable WIM stream header if needed. */ + if (ctx->write_resource_flags & + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) + { + ret = write_pwm_stream_header(cur_lte, ctx->out_fd, + WIM_RESHDR_FLAG_COMPRESSED); + if (ret) + return ret; + } + + /* Save current offset. */ + ctx->res_start_offset = ctx->out_fd->offset; + + /* Begin building the chunk table, and leave space for + * it if needed. */ ret = begin_wim_resource_chunk_tab(cur_lte, ctx->out_fd, - cur_offset, - &ctx->cur_chunk_tab); + &ctx->cur_chunk_tab, + ctx->write_resource_flags); if (ret) return ret; + } /* Write the compressed chunks from the message. */ - ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab); + ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab, + ctx->write_resource_flags); if (ret) return ret; @@ -985,38 +1218,40 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) msg->begin_chunk + msg->num_chunks == ctx->cur_chunk_tab->num_chunks) { u64 res_csize; - off_t offset; ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab, ctx->out_fd, - &res_csize); + ctx->res_start_offset, + ctx->write_resource_flags); if (ret) return ret; list_del(&cur_lte->being_compressed_list); - /* Grab the offset of this stream in the output file - * from the chunk table before we free it. */ - offset = ctx->cur_chunk_tab->file_offset; + res_csize = ctx->out_fd->offset - ctx->res_start_offset; FREE(ctx->cur_chunk_tab); ctx->cur_chunk_tab = NULL; - if (res_csize >= wim_resource_size(cur_lte)) { - /* Oops! We compressed the resource to - * larger than the original size. Write - * the resource uncompressed instead. */ + /* Check for resources compressed to greater than or + * equal to their original size and write them + * uncompressed instead. (But never do this if writing + * to a pipe.) */ + if (res_csize >= wim_resource_size(cur_lte) && + !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) + { DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " "writing uncompressed instead", wim_resource_size(cur_lte), res_csize); - ret = seek_and_truncate(ctx->out_fd, offset); + ret = seek_and_truncate(ctx->out_fd, ctx->res_start_offset); if (ret) return ret; ret = write_wim_resource(cur_lte, ctx->out_fd, WIMLIB_COMPRESSION_TYPE_NONE, &cur_lte->output_resource_entry, - ctx->write_resource_flags); + ctx->write_resource_flags, + ctx->comp_ctx); if (ret) return ret; } else { @@ -1027,17 +1262,22 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) cur_lte->resource_entry.original_size; cur_lte->output_resource_entry.offset = - offset; + ctx->res_start_offset; cur_lte->output_resource_entry.flags = cur_lte->resource_entry.flags | WIM_RESHDR_FLAG_COMPRESSED; + + DEBUG("Wrote compressed resource " + "(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)", + cur_lte->output_resource_entry.original_size, + cur_lte->output_resource_entry.size, + cur_lte->output_resource_entry.offset, + cur_lte->output_resource_entry.flags); } - do_write_streams_progress(ctx->progress, - ctx->progress_func, - wim_resource_size(cur_lte), - false); + do_write_streams_progress(ctx->progress_data, + cur_lte, false); /* Since we just finished writing a stream, write any * streams that have been added to the serial_streams @@ -1050,9 +1290,9 @@ receive_compressed_chunks(struct main_writer_thread_ctx *ctx) ctx->lookup_table, ctx->out_fd, ctx->out_ctype, + ctx->comp_ctx, ctx->write_resource_flags, - ctx->progress_func, - ctx->progress); + ctx->progress_data); if (ret) return ret; } @@ -1149,9 +1389,9 @@ main_writer_thread_finish(void *_ctx) ctx->lookup_table, ctx->out_fd, ctx->out_ctype, + ctx->comp_ctx, ctx->write_resource_flags, - ctx->progress_func, - ctx->progress); + ctx->progress_data); } static int @@ -1172,11 +1412,10 @@ submit_stream_for_compression(struct wim_lookup_table_entry *lte, list_add_tail(<e->being_compressed_list, &ctx->outstanding_streams); ret = read_resource_prefix(lte, wim_resource_size(lte), main_writer_thread_cb, ctx, 0); - if (ret == 0) { - wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); - ret = finalize_and_check_sha1(&ctx->next_sha_ctx, lte); - } - return ret; + if (ret) + return ret; + wimlib_assert(ctx->next_chunk == ctx->next_num_chunks); + return finalize_and_check_sha1(&ctx->next_sha_ctx, lte); } static int @@ -1188,8 +1427,8 @@ main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) if (wim_resource_size(lte) < 1000 || ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE || (lte->resource_location == RESOURCE_IN_WIM && - !(ctx->write_resource_flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) && - wimlib_get_compression_type(lte->wim) == ctx->out_ctype)) + !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) && + lte->wim->compression_type == ctx->out_ctype)) { /* Stream is too small or isn't being compressed. Process it by * the main thread when we have a chance. We can't necessarily @@ -1244,17 +1483,18 @@ get_default_num_threads(void) static int write_stream_list_parallel(struct list_head *stream_list, struct wim_lookup_table *lookup_table, - int out_fd, + struct filedes *out_fd, int out_ctype, + struct wimlib_lzx_context **comp_ctx, int write_resource_flags, - wimlib_progress_func_t progress_func, - union wimlib_progress_info *progress, + struct write_streams_progress_data *progress_data, unsigned num_threads) { int ret; struct shared_queue res_to_compress_queue; struct shared_queue compressed_res_queue; pthread_t *compressor_threads = NULL; + union wimlib_progress_info *progress = &progress_data->progress; if (num_threads == 0) { long nthreads = get_default_num_threads(); @@ -1268,8 +1508,9 @@ write_stream_list_parallel(struct list_head *stream_list, } } - DEBUG("Writing stream list (parallel version, num_threads=%u)", - num_threads); + DEBUG("Writing stream list of size %"PRIu64" " + "(parallel version, num_threads=%u)", + progress->write_streams.total_streams, num_threads); progress->write_streams.num_threads = num_threads; @@ -1286,21 +1527,36 @@ write_stream_list_parallel(struct list_head *stream_list, if (ret) goto out_destroy_res_to_compress_queue; - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); + struct compressor_thread_params *params; + + params = CALLOC(num_threads, sizeof(params[0])); + if (params == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } + + for (unsigned i = 0; i < num_threads; i++) { + params[i].res_to_compress_queue = &res_to_compress_queue; + params[i].compressed_res_queue = &compressed_res_queue; + params[i].out_ctype = out_ctype; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) { + ret = alloc_lzx_context(write_resource_flags, + ¶ms[i].comp_ctx); + if (ret) + goto out_free_params; + } + } compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); if (!compressor_threads) { ret = WIMLIB_ERR_NOMEM; - goto out_destroy_compressed_res_queue; + goto out_free_params; } for (unsigned i = 0; i < num_threads; i++) { DEBUG("pthread_create thread %u of %u", i + 1, num_threads); ret = pthread_create(&compressor_threads[i], NULL, - compressor_thread_proc, ¶ms); + compressor_thread_proc, ¶ms[i]); if (ret != 0) { ret = -1; ERROR_WITH_ERRNO("Failed to create compressor " @@ -1311,26 +1567,28 @@ write_stream_list_parallel(struct list_head *stream_list, } } - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + if (progress_data->progress_func) { + progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } struct main_writer_thread_ctx ctx; ctx.stream_list = stream_list; ctx.lookup_table = lookup_table; ctx.out_fd = out_fd; ctx.out_ctype = out_ctype; + ctx.comp_ctx = comp_ctx; ctx.res_to_compress_queue = &res_to_compress_queue; ctx.compressed_res_queue = &compressed_res_queue; ctx.num_messages = queue_size; ctx.write_resource_flags = write_resource_flags; - ctx.progress_func = progress_func; - ctx.progress = progress; + ctx.progress_data = progress_data; ret = main_writer_thread_init_ctx(&ctx); if (ret) goto out_join; ret = do_write_stream_list(stream_list, lookup_table, main_thread_process_next_stream, - &ctx, progress_func, progress); + &ctx, progress_data); if (ret) goto out_destroy_ctx; @@ -1355,6 +1613,10 @@ out_join: } } FREE(compressor_threads); +out_free_params: + for (unsigned i = 0; i < num_threads; i++) + wimlib_lzx_free_context(params[i].comp_ctx); + FREE(params); out_destroy_compressed_res_queue: shared_queue_destroy(&compressed_res_queue); out_destroy_res_to_compress_queue: @@ -1368,9 +1630,9 @@ out_serial_quiet: lookup_table, out_fd, out_ctype, + comp_ctx, write_resource_flags, - progress_func, - progress); + progress_data); } #endif @@ -1382,22 +1644,35 @@ out_serial_quiet: static int write_stream_list(struct list_head *stream_list, struct wim_lookup_table *lookup_table, - int out_fd, int out_ctype, int write_flags, + struct filedes *out_fd, int out_ctype, + struct wimlib_lzx_context **comp_ctx, + int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { struct wim_lookup_table_entry *lte; size_t num_streams = 0; u64 total_bytes = 0; u64 total_compression_bytes = 0; - union wimlib_progress_info progress; + struct write_streams_progress_data progress_data; int ret; int write_resource_flags; + unsigned total_parts = 0; + WIMStruct *prev_wim_part = NULL; - if (list_empty(stream_list)) + if (list_empty(stream_list)) { + DEBUG("No streams to write."); return 0; + } write_resource_flags = write_flags_to_resource_flags(write_flags); + DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)", + out_fd->offset, write_resource_flags); + + sort_stream_list_by_sequential_order(stream_list, + offsetof(struct wim_lookup_table_entry, + write_streams_list)); + /* Calculate the total size of the streams to be written. Note: this * will be the uncompressed size, as we may not know the compressed size * yet, and also this will assume that every unhashed stream will be @@ -1407,18 +1682,32 @@ write_stream_list(struct list_head *stream_list, total_bytes += wim_resource_size(lte); if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && (wim_resource_compression_type(lte) != out_ctype || - (write_resource_flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS))) + (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS))) { total_compression_bytes += wim_resource_size(lte); } + if (lte->resource_location == RESOURCE_IN_WIM) { + if (prev_wim_part != lte->wim) { + prev_wim_part = lte->wim; + total_parts++; + } + } } - progress.write_streams.total_bytes = total_bytes; - progress.write_streams.total_streams = num_streams; - progress.write_streams.completed_bytes = 0; - progress.write_streams.completed_streams = 0; - progress.write_streams.num_threads = num_threads; - progress.write_streams.compression_type = out_ctype; - progress.write_streams._private = 0; + + memset(&progress_data, 0, sizeof(progress_data)); + progress_data.progress_func = progress_func; + + progress_data.progress.write_streams.total_bytes = total_bytes; + progress_data.progress.write_streams.total_streams = num_streams; + progress_data.progress.write_streams.completed_bytes = 0; + progress_data.progress.write_streams.completed_streams = 0; + progress_data.progress.write_streams.num_threads = num_threads; + progress_data.progress.write_streams.compression_type = out_ctype; + progress_data.progress.write_streams.total_parts = total_parts; + progress_data.progress.write_streams.completed_parts = 0; + + progress_data.next_progress = 0; + progress_data.prev_wim_part = NULL; #ifdef ENABLE_MULTITHREADED_COMPRESSION if (total_compression_bytes >= 2000000 && num_threads != 1) @@ -1426,9 +1715,9 @@ write_stream_list(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + comp_ctx, write_resource_flags, - progress_func, - &progress, + &progress_data, num_threads); else #endif @@ -1436,9 +1725,13 @@ write_stream_list(struct list_head *stream_list, lookup_table, out_fd, out_ctype, + comp_ctx, write_resource_flags, - progress_func, - &progress); + &progress_data); + if (ret == 0) + DEBUG("Successfully wrote stream list."); + else + DEBUG("Failed to write stream list."); return ret; } @@ -1473,7 +1766,7 @@ stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) struct wim_lookup_table_entry *same_size_lte; struct hlist_node *tmp; - pos = hash_u64(wim_resource_size(lte)) % tab->capacity; + pos = hash_u64(wim_resource_size(lte)) % tab->capacity; lte->unique_size = 1; hlist_for_each_entry(same_size_lte, tmp, &tab->array[pos], hash_list_2) { if (wim_resource_size(same_size_lte) == wim_resource_size(lte)) { @@ -1488,148 +1781,65 @@ stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) return 0; } - -struct lte_overwrite_prepare_args { +struct find_streams_ctx { WIMStruct *wim; - off_t end_offset; + int write_flags; struct list_head stream_list; struct stream_size_table stream_size_tab; }; -/* First phase of preparing streams for an in-place overwrite. This is called - * on all streams, both hashed and unhashed, except the metadata resources. */ -static int -lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *_args) +static void +lte_reference_for_logical_write(struct wim_lookup_table_entry *lte, + struct find_streams_ctx *ctx, + unsigned nref) { - struct lte_overwrite_prepare_args *args = _args; - - wimlib_assert(!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)); - if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) - list_add_tail(<e->write_streams_list, &args->stream_list); - lte->out_refcnt = lte->refcnt; - stream_size_table_insert(lte, &args->stream_size_tab); - return 0; + if (lte->out_refcnt == 0) { + stream_size_table_insert(lte, &ctx->stream_size_tab); + list_add_tail(<e->write_streams_list, &ctx->stream_list); + } + lte->out_refcnt += nref; } -/* Second phase of preparing streams for an in-place overwrite. This is called - * on existing metadata resources and hashed streams, but not unhashed streams. - * - * NOTE: lte->output_resource_entry is in union with lte->hash_list_2, so - * lte_overwrite_prepare_2() must be called after lte_overwrite_prepare(), as - * the latter uses lte->hash_list_2, while the former expects to set - * lte->output_resource_entry. */ static int -lte_overwrite_prepare_2(struct wim_lookup_table_entry *lte, void *_args) +do_lte_full_reference_for_logical_write(struct wim_lookup_table_entry *lte, + void *_ctx) { - struct lte_overwrite_prepare_args *args = _args; - - if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == args->wim) { - /* We can't do an in place overwrite on the WIM if there are - * streams after the XML data. */ - if (lte->resource_entry.offset + - lte->resource_entry.size > args->end_offset) - { - #ifdef ENABLE_ERROR_MESSAGES - ERROR("The following resource is after the XML data:"); - print_lookup_table_entry(lte, stderr); - #endif - return WIMLIB_ERR_RESOURCE_ORDER; - } - copy_resource_entry(<e->output_resource_entry, - <e->resource_entry); - } + struct find_streams_ctx *ctx = _ctx; + lte->out_refcnt = 0; + lte_reference_for_logical_write(lte, ctx, + (lte->refcnt ? lte->refcnt : 1)); return 0; } -/* Given a WIM that we are going to overwrite in place with zero or more - * additional streams added, construct a list the list of new unique streams - * ('struct wim_lookup_table_entry's) that must be written, plus any unhashed - * streams that need to be added but may be identical to other hashed or - * unhashed streams. These unhashed streams are checksummed while the streams - * are being written. To aid this process, the member @unique_size is set to 1 - * on streams that have a unique size and therefore must be written. - * - * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to - * indicate the number of times the stream is referenced in only the streams - * that are being written; this may still be adjusted later when unhashed - * streams are being resolved. - */ static int -prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset, - struct list_head *stream_list) -{ - int ret; - struct lte_overwrite_prepare_args args; - unsigned i; - - args.wim = wim; - args.end_offset = end_offset; - ret = init_stream_size_table(&args.stream_size_tab, - wim->lookup_table->capacity); - if (ret) - return ret; - - INIT_LIST_HEAD(&args.stream_list); - for (i = 0; i < wim->hdr.image_count; i++) { - struct wim_image_metadata *imd; - struct wim_lookup_table_entry *lte; - - imd = wim->image_metadata[i]; - image_for_each_unhashed_stream(lte, imd) - lte_overwrite_prepare(lte, &args); - } - for_lookup_table_entry(wim->lookup_table, lte_overwrite_prepare, &args); - list_transfer(&args.stream_list, stream_list); - - for (i = 0; i < wim->hdr.image_count; i++) { - ret = lte_overwrite_prepare_2(wim->image_metadata[i]->metadata_lte, - &args); - if (ret) - goto out_destroy_stream_size_table; - } - ret = for_lookup_table_entry(wim->lookup_table, - lte_overwrite_prepare_2, &args); -out_destroy_stream_size_table: - destroy_stream_size_table(&args.stream_size_tab); - return ret; -} - - -struct find_streams_ctx { - struct list_head stream_list; - struct stream_size_table stream_size_tab; -}; - -static void inode_find_streams_to_write(struct wim_inode *inode, struct wim_lookup_table *table, - struct list_head *stream_list, - struct stream_size_table *tab) + struct find_streams_ctx *ctx) { struct wim_lookup_table_entry *lte; - for (unsigned i = 0; i <= inode->i_num_ads; i++) { + unsigned i; + + for (i = 0; i <= inode->i_num_ads; i++) { lte = inode_stream_lte(inode, i, table); - if (lte) { - if (lte->out_refcnt == 0) { - if (lte->unhashed) - stream_size_table_insert(lte, tab); - list_add_tail(<e->write_streams_list, stream_list); - } - lte->out_refcnt += inode->i_nlink; - } + if (lte) + lte_reference_for_logical_write(lte, ctx, inode->i_nlink); + else if (!is_zero_hash(inode_stream_hash(inode, i))) + return WIMLIB_ERR_RESOURCE_NOT_FOUND; } + return 0; } static int -image_find_streams_to_write(WIMStruct *w) +image_find_streams_to_write(WIMStruct *wim) { struct find_streams_ctx *ctx; struct wim_image_metadata *imd; struct wim_inode *inode; struct wim_lookup_table_entry *lte; + int ret; - ctx = w->private; - imd = wim_get_current_image_metadata(w); + ctx = wim->private; + imd = wim_get_current_image_metadata(wim); image_for_each_unhashed_stream(lte, imd) lte->out_refcnt = 0; @@ -1637,328 +1847,933 @@ image_find_streams_to_write(WIMStruct *w) /* Go through this image's inodes to find any streams that have not been * found yet. */ image_for_each_inode(inode, imd) { - inode_find_streams_to_write(inode, w->lookup_table, - &ctx->stream_list, - &ctx->stream_size_tab); + ret = inode_find_streams_to_write(inode, wim->lookup_table, ctx); + if (ret) + return ret; } return 0; } -/* Given a WIM that from which one or all of the images is being written, build - * the list of unique streams ('struct wim_lookup_table_entry's) that must be - * written, plus any unhashed streams that need to be written but may be - * identical to other hashed or unhashed streams being written. These unhashed - * streams are checksummed while the streams are being written. To aid this - * process, the member @unique_size is set to 1 on streams that have a unique - * size and therefore must be written. +/* + * Build a list of streams (via `struct wim_lookup_table_entry's) included in + * the "logical write" of the WIM, meaning all streams that are referenced at + * least once by dentries in the the image(s) being written. 'out_refcnt' on + * each stream being included in the logical write is set to the number of + * references from dentries in the image(s). Furthermore, 'unique_size' on each + * stream being included in the logical write is set to indicate whether that + * stream has a unique size relative to the streams being included in the + * logical write. Still furthermore, 'part_number' on each stream being + * included in the logical write is set to the part number given in the + * in-memory header of @p wim. * - * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to - * indicate the number of times the stream is referenced in only the streams - * that are being written; this may still be adjusted later when unhashed - * streams are being resolved. + * This is considered a "logical write" because it does not take into account + * filtering out streams already present in the WIM (in the case of an in place + * overwrite) or present in other WIMs (in case of creating delta WIM). */ static int -prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) +prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok, + struct find_streams_ctx *ctx) { int ret; + struct wim_lookup_table_entry *lte; + + if (streams_ok && (image == WIMLIB_ALL_IMAGES || + (image == 1 && wim->hdr.image_count == 1))) + { + /* Fast case: Assume that all streams are being written and + * that the reference counts are correct. */ + struct wim_lookup_table_entry *lte; + struct wim_image_metadata *imd; + unsigned i; + + for_lookup_table_entry(wim->lookup_table, + do_lte_full_reference_for_logical_write, ctx); + for (i = 0; i < wim->hdr.image_count; i++) { + imd = wim->image_metadata[i]; + image_for_each_unhashed_stream(lte, imd) + do_lte_full_reference_for_logical_write(lte, ctx); + } + } else { + /* Slow case: Walk through the images being written and + * determine the streams referenced. */ + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + wim->private = ctx; + ret = for_image(wim, image, image_find_streams_to_write); + if (ret) + return ret; + } + + list_for_each_entry(lte, &ctx->stream_list, write_streams_list) + lte->part_number = wim->hdr.part_number; + return 0; +} + +static int +process_filtered_stream(struct wim_lookup_table_entry *lte, void *_ctx) +{ + struct find_streams_ctx *ctx = _ctx; + u16 filtered = 0; + + /* Calculate and set lte->filtered. */ + if (lte->resource_location == RESOURCE_IN_WIM) { + if (lte->wim == ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE)) + filtered |= FILTERED_SAME_WIM; + if (lte->wim != ctx->wim && + (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)) + filtered |= FILTERED_EXTERNAL_WIM; + } + lte->filtered = filtered; + + /* Filtered streams get inserted into the stream size table too, unless + * they already were. This is because streams that are checksummed + * on-the-fly during the write should not be written if they are + * duplicates of filtered stream. */ + if (lte->filtered && lte->out_refcnt == 0) + stream_size_table_insert(lte, &ctx->stream_size_tab); + return 0; +} + +static int +mark_stream_not_filtered(struct wim_lookup_table_entry *lte, void *_ignore) +{ + lte->filtered = 0; + return 0; +} + +/* Given the list of streams to include in a logical write of a WIM, handle + * filtering out streams already present in the WIM or already present in + * external WIMs, depending on the write flags provided. */ +static void +handle_stream_filtering(struct find_streams_ctx *ctx) +{ + struct wim_lookup_table_entry *lte, *tmp; + + if (!(ctx->write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE | + WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS))) + { + for_lookup_table_entry(ctx->wim->lookup_table, + mark_stream_not_filtered, ctx); + return; + } + + for_lookup_table_entry(ctx->wim->lookup_table, + process_filtered_stream, ctx); + + /* Streams in logical write list that were filtered can be removed. */ + list_for_each_entry_safe(lte, tmp, &ctx->stream_list, + write_streams_list) + if (lte->filtered) + list_del(<e->write_streams_list); +} + +/* Prepares list of streams to write for the specified WIM image(s). This wraps + * around prepare_logical_stream_list() to handle filtering out streams already + * present in the WIM or already present in external WIMs, depending on the + * write flags provided. + * + * Note: some additional data is stored in each `struct wim_lookup_table_entry': + * + * - 'out_refcnt' is set to the number of references found for the logical write. + * This will be nonzero on all streams in the list returned by this function, + * but will also be nonzero on streams not in the list that were included in + * the logical write list, but filtered out from the returned list. + * - 'filtered' is set to nonzero if the stream was filtered. Filtered streams + * are not included in the list of streams returned by this function. + * - 'unique_size' is set if the stream has a unique size among all streams in + * the logical write plus any filtered streams in the entire WIM that could + * potentially turn out to have the same checksum as a yet-to-be-checksummed + * stream being written. + */ +static int +prepare_stream_list(WIMStruct *wim, int image, int write_flags, + struct list_head *stream_list) +{ + int ret; + bool streams_ok; struct find_streams_ctx ctx; - for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + INIT_LIST_HEAD(&ctx.stream_list); ret = init_stream_size_table(&ctx.stream_size_tab, wim->lookup_table->capacity); if (ret) return ret; - for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, - &ctx.stream_size_tab); - INIT_LIST_HEAD(&ctx.stream_list); - wim->private = &ctx; - ret = for_image(wim, image, image_find_streams_to_write); + ctx.write_flags = write_flags; + ctx.wim = wim; + + streams_ok = ((write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) != 0); + + ret = prepare_logical_stream_list(wim, image, streams_ok, &ctx); + if (ret) + goto out_destroy_table; + + handle_stream_filtering(&ctx); + list_transfer(&ctx.stream_list, stream_list); + ret = 0; +out_destroy_table: destroy_stream_size_table(&ctx.stream_size_tab); - if (ret == 0) - list_transfer(&ctx.stream_list, stream_list); return ret; } -/* Writes the streams for the specified @image in @wim to @wim->out_fd. - */ static int write_wim_streams(WIMStruct *wim, int image, int write_flags, unsigned num_threads, - wimlib_progress_func_t progress_func) + wimlib_progress_func_t progress_func, + struct list_head *stream_list_override) { int ret; - struct list_head stream_list; + struct list_head _stream_list; + struct list_head *stream_list; + struct wim_lookup_table_entry *lte; - ret = prepare_stream_list(wim, image, &stream_list); - if (ret) - return ret; - return write_stream_list(&stream_list, + if (stream_list_override == NULL) { + /* Normal case: prepare stream list from image(s) being written. + */ + stream_list = &_stream_list; + ret = prepare_stream_list(wim, image, write_flags, stream_list); + if (ret) + return ret; + } else { + /* Currently only as a result of wimlib_split() being called: + * use stream list already explicitly provided. Use existing + * reference counts. */ + stream_list = stream_list_override; + list_for_each_entry(lte, stream_list, write_streams_list) { + lte->out_refcnt = (lte->refcnt ? lte->refcnt : 1); + lte->part_number = wim->hdr.part_number; + } + } + + return write_stream_list(stream_list, wim->lookup_table, - wim->out_fd, - wimlib_get_compression_type(wim), + &wim->out_fd, + wim->compression_type, + &wim->lzx_context, write_flags, num_threads, progress_func); } +static int +write_wim_metadata_resources(WIMStruct *wim, int image, int write_flags, + wimlib_progress_func_t progress_func) +{ + 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."); + return 0; + } + + write_resource_flags = write_flags_to_resource_flags(write_flags); + + DEBUG("Writing metadata resources (offset=%"PRIu64")", + wim->out_fd.offset); + + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); + + if (image == WIMLIB_ALL_IMAGES) { + start_image = 1; + end_image = wim->hdr.image_count; + } else { + start_image = image; + end_image = image; + } + + for (int i = start_image; i <= end_image; i++) { + struct wim_image_metadata *imd; + + imd = wim->image_metadata[i - 1]; + /* Build a new metadata resource only if image was modified from + * the original (or was newly added). Otherwise just copy the + * existing one. */ + if (imd->modified) { + DEBUG("Image %u was modified; building and writing new " + "metadata resource", i); + ret = write_metadata_resource(wim, i, + write_resource_flags); + } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) { + DEBUG("Image %u was not modified; re-using existing " + "metadata resource.", i); + copy_resource_entry(&imd->metadata_lte->output_resource_entry, + &imd->metadata_lte->resource_entry); + ret = 0; + } else { + DEBUG("Image %u was not modified; copying existing " + "metadata resource.", i); + ret = write_wim_resource(imd->metadata_lte, + &wim->out_fd, + wim->compression_type, + &imd->metadata_lte->output_resource_entry, + write_resource_flags, + &wim->lzx_context); + } + if (ret) + return ret; + } + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + return 0; +} + +static int +open_wim_writable(WIMStruct *wim, const tchar *path, int open_flags) +{ + int raw_fd; + DEBUG("Opening \"%"TS"\" for writing.", path); + + raw_fd = topen(path, open_flags | O_BINARY, 0644); + if (raw_fd < 0) { + ERROR_WITH_ERRNO("Failed to open \"%"TS"\" for writing", path); + return WIMLIB_ERR_OPEN; + } + filedes_init(&wim->out_fd, raw_fd); + return 0; +} + +static int +close_wim_writable(WIMStruct *wim, int write_flags) +{ + int ret = 0; + + if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) + if (filedes_valid(&wim->out_fd)) + if (filedes_close(&wim->out_fd)) + ret = WIMLIB_ERR_WRITE; + filedes_invalidate(&wim->out_fd); + return ret; +} + /* + * finish_write(): + * * Finish writing a WIM file: write the lookup table, xml data, and integrity - * table (optional), then overwrite the WIM header. + * 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_CHECK_INTEGRITY: + * Include an integrity table. * - * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: - * Don't write the lookup table. + * (public) WIMLIB_WRITE_FLAG_FSYNC: + * fsync() the output file before closing it. * - * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: - * When (if) writing the integrity table, re-use entries from the - * existing integrity table, if possible. + * (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_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_NO_LOOKUP_TABLE: + * Don't write the lookup table. * - * (private) WIMLIB_WRITE_FLAG_FSYNC: - * fsync() the output file before closing it. + * (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. */ -int -finish_write(WIMStruct *w, int image, int write_flags, - wimlib_progress_func_t progress_func) +static int +finish_write(WIMStruct *wim, int image, int write_flags, + wimlib_progress_func_t progress_func, + struct list_head *stream_list_override) { int ret; - struct wim_header hdr; - - /* @hdr will be the header for the new WIM. First copy all the data - * from the header in the WIMStruct; then set all the fields that may - * have changed, including the resource entries, boot index, and image - * count. */ - memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); + off_t hdr_offset; + int write_resource_flags; + off_t old_lookup_table_end; + off_t new_lookup_table_end; + u64 xml_totalbytes; - /* Set image count and boot index correctly for single image writes */ - if (image != WIMLIB_ALL_IMAGES) { - hdr.image_count = 1; - if (hdr.boot_idx == image) - hdr.boot_idx = 1; - else - hdr.boot_idx = 0; - } + write_resource_flags = write_flags_to_resource_flags(write_flags); /* In the WIM header, there is room for the resource entry for a * metadata resource labeled as the "boot metadata". This entry should * be zeroed out if there is no bootable image (boot_idx 0). Otherwise, * it should be a copy of the resource entry for the image that is * marked as bootable. This is not well documented... */ - if (hdr.boot_idx == 0) { - zero_resource_entry(&hdr.boot_metadata_res_entry); + if (wim->hdr.boot_idx == 0) { + zero_resource_entry(&wim->hdr.boot_metadata_res_entry); } else { - copy_resource_entry(&hdr.boot_metadata_res_entry, - &w->image_metadata[ hdr.boot_idx- 1 + copy_resource_entry(&wim->hdr.boot_metadata_res_entry, + &wim->image_metadata[wim->hdr.boot_idx- 1 ]->metadata_lte->output_resource_entry); } + /* Write lookup table. (Save old position first.) */ + old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + + wim->hdr.lookup_table_res_entry.size; if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); + ret = write_wim_lookup_table(wim, image, write_flags, + &wim->hdr.lookup_table_res_entry, + stream_list_override); if (ret) - goto out_close_wim; + return ret; } - ret = write_xml_data(w->wim_info, image, w->out_fd, - (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? - wim_info_get_total_bytes(w->wim_info) : 0, - &hdr.xml_res_entry); + /* Write XML data. */ + xml_totalbytes = wim->out_fd.offset; + if (write_flags & WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES) + xml_totalbytes = WIM_TOTALBYTES_USE_EXISTING; + ret = write_wim_xml_data(wim, image, xml_totalbytes, + &wim->hdr.xml_res_entry, + write_resource_flags); if (ret) - goto out_close_wim; + return ret; + /* Write integrity table (optional). */ if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { struct wim_header checkpoint_hdr; - memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); + memcpy(&checkpoint_hdr, &wim->hdr, sizeof(struct wim_header)); zero_resource_entry(&checkpoint_hdr.integrity); - ret = write_header(&checkpoint_hdr, w->out_fd); + checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS; + ret = write_wim_header_at_offset(&checkpoint_hdr, + &wim->out_fd, 0); if (ret) - goto out_close_wim; + return ret; } - off_t old_lookup_table_end; - off_t new_lookup_table_end; - if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { - old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; - } else { + if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE)) old_lookup_table_end = 0; - } - new_lookup_table_end = hdr.lookup_table_res_entry.offset + - hdr.lookup_table_res_entry.size; - ret = write_integrity_table(w->out_fd, - &hdr.integrity, + new_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + + wim->hdr.lookup_table_res_entry.size; + + ret = write_integrity_table(wim, new_lookup_table_end, old_lookup_table_end, progress_func); if (ret) - goto out_close_wim; + return ret; } else { - zero_resource_entry(&hdr.integrity); - } - - ret = write_header(&hdr, w->out_fd); + /* No integrity table. */ + zero_resource_entry(&wim->hdr.integrity); + } + + /* 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; + ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset); if (ret) - goto out_close_wim; + return ret; + /* Possibly sync file data to disk before closing. On POSIX systems, it + * is necessary to do this before using rename() to overwrite an + * existing file with a new file. Otherwise, data loss would occur if + * the system is abruptly terminated when the metadata for the rename + * operation has been written to disk, but the new file data has not. + */ if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { - if (fsync(w->out_fd)) { + if (fsync(wim->out_fd.fd)) { ERROR_WITH_ERRNO("Error syncing data to WIM file"); - ret = WIMLIB_ERR_WRITE; + return WIMLIB_ERR_WRITE; } } -out_close_wim: - if (close(w->out_fd)) { + + if (close_wim_writable(wim, write_flags)) { ERROR_WITH_ERRNO("Failed to close the output WIM file"); - if (ret == 0) - ret = WIMLIB_ERR_WRITE; + return WIMLIB_ERR_WRITE; } - w->out_fd = -1; - return ret; + + return 0; } #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) int -lock_wim(WIMStruct *w, int fd) +lock_wim(WIMStruct *wim, int fd) { int ret = 0; - if (fd != -1 && !w->wim_locked) { + 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!", w->filename); + " by another process!", wim->filename); ret = WIMLIB_ERR_ALREADY_LOCKED; } else { WARNING_WITH_ERRNO("Failed to lock `%"TS"'", - w->filename); + wim->filename); ret = 0; } } else { - w->wim_locked = 1; + wim->wim_locked = 1; } } return ret; } #endif +/* + * write_pipable_wim(): + * + * Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM + * capable of being applied from a pipe). + * + * Pipable WIMs are a wimlib-specific modification of the WIM format such that + * images can be applied from them sequentially when the file data is sent over + * a pipe. In addition, a pipable WIM can be written sequentially to a pipe. + * The modifications made to the WIM format for pipable WIMs are: + * + * - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead + * of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also + * stops other software from trying to read the file as a normal WIM. + * + * - The header at the beginning of the file does not contain all the normal + * information; in particular it will have all 0's for the 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. + * Consequently, wimlib will write the full header at the very end of the + * file. The header at the end, however, is only used when reading the WIM + * from a seekable file (not a pipe). + * + * - An extra copy of the XML data is placed directly after the header. This + * allows image names and sizes to be determined at an appropriate time when + * reading the WIM from a pipe. This copy of the XML data is ignored if the + * WIM is read from a seekable file (not a pipe). + * + * - 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 + * written until all chunks have been compressed. Instead, each compressed + * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size. + * Furthermore, the chunk table is written at the end of the resource instead + * of the start. Note: chunk offsets are given in the chunk table as if the + * `struct pwm_chunk_hdr's were not present; also, the chunk table is only + * used if the WIM is being read from a seekable file (not a pipe). + * + * - 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.) + * + * - 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 + * 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 + * 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 | + * ---------+----------+--------------------+----------------+--------------+-----------+--------+ + * + * Layout of normal WIM: + * + * +--------+-----------------------------+-------------------------+ + * | Header | File and metadata resources | Lookup table | XML data | + * +--------+-----------------------------+-------------------------+ + * + * An optional integrity table can follow the final XML data in both normal and + * pipable WIMs. However, due to implementation details, wimlib currently can + * only include an integrity table in a pipable WIM when writing it to a + * seekable file (not a pipe). + * + * Do note that since pipable WIMs are not supported by Microsoft's software, + * wimlib does not create them unless explicitly requested (with + * WIMLIB_WRITE_FLAG_PIPABLE) and as stated above they use different magic + * characters to identify the file. + */ static int -open_wim_writable(WIMStruct *w, const tchar *path, int open_flags) +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) { - w->out_fd = topen(path, open_flags | O_BINARY, 0644); - if (w->out_fd == -1) { - ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path); - return WIMLIB_ERR_OPEN; - } - return 0; -} + int ret; + struct resource_entry xml_res_entry; + + WARNING("Creating a pipable WIM, which will " + "be incompatible\n" + " with Microsoft's software (wimgapi/imagex/Dism)."); + + /* At this point, the header at the beginning of the file has already + * been written. */ + + /* For efficiency, when wimlib adds an image to the WIM with + * wimlib_add_image(), the SHA1 message digests of files is 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); + if (ret) + return ret; + /* Write extra copy of the XML data. */ + ret = write_wim_xml_data(wim, image, WIM_TOTALBYTES_OMIT, + &xml_res_entry, + WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE); + if (ret) + return ret; -void -close_wim_writable(WIMStruct *w) -{ - if (w->out_fd != -1) { - if (close(w->out_fd)) - WARNING_WITH_ERRNO("Failed to close output WIM"); - w->out_fd = -1; - } + /* Write metadata resources for the image(s) being included in the + * output WIM. */ + ret = write_wim_metadata_resources(wim, image, write_flags, + progress_func); + 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); + + /* The lookup table, XML data, and header at end are handled by + * finish_write(). */ } -/* Open file stream and write dummy header for WIM. */ +/* Write a standalone WIM or split WIM (SWM) part to a new file or to a file + * descriptor. */ int -begin_write(WIMStruct *w, const tchar *path, int write_flags) +write_wim_part(WIMStruct *wim, + const void *path_or_fd, + 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, + const u8 *guid) { int ret; - int open_flags = O_TRUNC | O_CREAT; + struct wim_header hdr_save; + struct list_head lt_stream_list_override; + + 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) - open_flags |= O_RDWR; + 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_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 - open_flags |= O_WRONLY; - ret = open_wim_writable(w, path, open_flags); - if (ret) - return ret; - /* Write dummy header. It will be overwritten later. */ - ret = write_header(&w->hdr, w->out_fd); + 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")); + + /* Internally, this is always called with a valid part number and total + * parts. */ + wimlib_assert(total_parts >= 1); + wimlib_assert(part_number >= 1 && part_number <= total_parts); + + /* A valid image (or all images) must be specified. */ + if (image != WIMLIB_ALL_IMAGES && + (image < 1 || image > wim->hdr.image_count)) + return WIMLIB_ERR_INVALID_IMAGE; + + /* If we need to write metadata resources, make sure the ::WIMStruct has + * the needed information attached (e.g. is not a resource-only WIM, + * such as a non-first part of a split WIM). */ + if (!wim_has_metadata(wim) && + !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)) + return WIMLIB_ERR_METADATA_NOT_FOUND; + + /* Check for contradictory flags. */ + if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + == (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY)) + return WIMLIB_ERR_INVALID_PARAM; + + if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + == (WIMLIB_WRITE_FLAG_PIPABLE | + WIMLIB_WRITE_FLAG_NOT_PIPABLE)) + return WIMLIB_ERR_INVALID_PARAM; + + /* Save previous header, then start initializing the new one. */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); + + /* Set default integrity and pipable flags. */ + 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_CHECK_INTEGRITY | + WIMLIB_WRITE_FLAG_NO_CHECK_INTEGRITY))) + if (wim_has_integrity_table(wim)) + write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY; + + /* Set appropriate magic number. */ + if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE) + wim->hdr.magic = PWM_MAGIC; + else + wim->hdr.magic = WIM_MAGIC; + + /* 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); + + /* Set SPANNED header flag if writing part of a split WIM. */ + 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; + + /* Use GUID if specified; otherwise generate a new one. */ + 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_resource_entry(&wim->hdr.lookup_table_res_entry); + zero_resource_entry(&wim->hdr.xml_res_entry); + zero_resource_entry(&wim->hdr.boot_metadata_res_entry); + zero_resource_entry(&wim->hdr.integrity); + + /* 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; + } + + /* Split WIMs can't be bootable. */ + if (total_parts != 1) + wim->hdr.boot_idx = 0; + + /* Initialize 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; + if (!filedes_is_seekable(&wim->out_fd)) { + ret = WIMLIB_ERR_INVALID_PARAM; + if (!(write_flags & WIMLIB_WRITE_FLAG_PIPABLE)) + goto out_restore_hdr; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + ERROR("Can't include integrity check when " + "writing pipable WIM to pipe!"); + goto out_restore_hdr; + } + } + + } 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; + } + + /* 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). */ + 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; if (ret) - return ret; - if (lseek(w->out_fd, WIM_HEADER_DISK_SIZE, SEEK_SET) == -1) { - ERROR_WITH_ERRNO("Failed to seek to end of WIM header"); - return WIMLIB_ERR_WRITE; + goto out_restore_hdr; + + if (stream_list_override) { + struct wim_lookup_table_entry *lte; + INIT_LIST_HEAD(<_stream_list_override); + list_for_each_entry(lte, stream_list_override, + write_streams_list) + { + list_add_tail(<e->lookup_table_list, + <_stream_list_override); + } } - return 0; + + /* Write metadata resources and streams. */ + 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); + if (ret) + goto out_restore_hdr; + + ret = write_wim_metadata_resources(wim, image, write_flags, + progress_func); + if (ret) + goto out_restore_hdr; + } else { + /* Non-default case: create pipable WIM. */ + ret = write_pipable_wim(wim, image, write_flags, num_threads, + progress_func, stream_list_override); + if (ret) + goto out_restore_hdr; + write_flags |= WIMLIB_WRITE_FLAG_HEADER_AT_END; + } + + if (stream_list_override) + stream_list_override = <_stream_list_override; + + /* Write lookup table, XML data, and (optional) integrity table. */ + ret = finish_write(wim, image, write_flags, progress_func, + stream_list_override); +out_restore_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); + (void)close_wim_writable(wim, write_flags); + 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) +{ + return write_wim_part(wim, path_or_fd, image, write_flags, + num_threads, progress_func, 1, 1, NULL, NULL); } -/* Writes a stand-alone WIM to a file. */ +/* API function documented in wimlib.h */ WIMLIBAPI int -wimlib_write(WIMStruct *w, const tchar *path, +wimlib_write(WIMStruct *wim, const tchar *path, int image, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { - int ret; - if (!path) return WIMLIB_ERR_INVALID_PARAM; write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - if (image != WIMLIB_ALL_IMAGES && - (image < 1 || image > w->hdr.image_count)) - return WIMLIB_ERR_INVALID_IMAGE; - - if (w->hdr.total_parts != 1) { - ERROR("Cannot call wimlib_write() on part of a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; - } - - ret = begin_write(w, path, write_flags); - if (ret) - goto out_close_wim; - - ret = write_wim_streams(w, image, write_flags, num_threads, - progress_func); - if (ret) - goto out_close_wim; - - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); + return write_standalone_wim(wim, path, image, write_flags, + num_threads, progress_func); +} - ret = for_image(w, image, write_metadata_resource); - if (ret) - goto out_close_wim; +/* 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) +{ + if (fd < 0) + return WIMLIB_ERR_INVALID_PARAM; - if (progress_func) - progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + write_flags |= WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR; - ret = finish_write(w, image, write_flags, progress_func); - /* finish_write() closed the WIM for us */ - goto out; -out_close_wim: - close_wim_writable(w); -out: - DEBUG("wimlib_write(path=%"TS") = %d", path, ret); - return ret; + return write_standalone_wim(wim, &fd, image, write_flags, + num_threads, progress_func); } static bool -any_images_modified(WIMStruct *w) +any_images_modified(WIMStruct *wim) { - for (int i = 0; i < w->hdr.image_count; i++) - if (w->image_metadata[i]->modified) + for (int i = 0; i < wim->hdr.image_count; i++) + if (wim->image_metadata[i]->modified) return true; return false; } +static int +check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim) +{ + const WIMStruct *wim = _wim; + off_t end_offset = *(const off_t*)wim->private; + + if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == wim && + lte->resource_entry.offset + lte->resource_entry.size > end_offset) + return WIMLIB_ERR_RESOURCE_ORDER; + return 0; +} + +/* Make sure no file or metadata resources are located after the XML data (or + * integrity table if present)--- otherwise we can't safely overwrite the WIM in + * place and we return WIMLIB_ERR_RESOURCE_ORDER. */ +static int +check_resource_offsets(WIMStruct *wim, off_t end_offset) +{ + int ret; + unsigned i; + + wim->private = &end_offset; + ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim); + if (ret) + return ret; + + for (i = 0; i < wim->hdr.image_count; i++) { + ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim); + if (ret) + return ret; + } + return 0; +} + /* * Overwrite a WIM, possibly appending streams to it. * @@ -2017,7 +2832,7 @@ any_images_modified(WIMStruct *w) * small amount of space compared to the streams, however.) */ static int -overwrite_wim_inplace(WIMStruct *w, int write_flags, +overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { @@ -2025,25 +2840,35 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, struct list_head stream_list; off_t old_wim_end; u64 old_lookup_table_end, old_xml_begin, old_xml_end; - int open_flags; + struct wim_header hdr_save; + + DEBUG("Overwriting `%"TS"' in-place", wim->filename); - DEBUG("Overwriting `%"TS"' in-place", w->filename); + /* Set default integrity flag. */ + 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 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 = w->hdr.xml_res_entry.offset; - old_xml_end = old_xml_begin + w->hdr.xml_res_entry.size; - old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; - if (w->hdr.integrity.offset != 0 && w->hdr.integrity.offset < old_xml_end) { - ERROR("Didn't expect the integrity table to be before the XML data"); + old_xml_begin = wim->hdr.xml_res_entry.offset; + old_xml_end = old_xml_begin + wim->hdr.xml_res_entry.size; + old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset + + wim->hdr.lookup_table_res_entry.size; + if (wim->hdr.integrity.offset != 0 && wim->hdr.integrity.offset < old_xml_end) { + WARNING("Didn't expect the integrity table to be before the XML data"); return WIMLIB_ERR_RESOURCE_ORDER; } if (old_lookup_table_end > old_xml_begin) { - ERROR("Didn't expect the lookup table to be after the XML data"); + WARNING("Didn't expect the lookup table to be after the XML data"); return WIMLIB_ERR_RESOURCE_ORDER; } @@ -2051,7 +2876,7 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, * 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 (!w->deletion_occurred && !any_images_modified(w)) { + 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 @@ -2062,159 +2887,183 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags, old_wim_end = old_lookup_table_end; write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; - } else if (w->hdr.integrity.offset) { + } else if (wim->hdr.integrity.offset) { /* Old WIM has an integrity table; begin writing new streams * after it. */ - old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size; + old_wim_end = wim->hdr.integrity.offset + wim->hdr.integrity.size; } else { /* No existing integrity table; begin writing new streams after * the old XML data. */ old_wim_end = old_xml_end; } - ret = prepare_streams_for_overwrite(w, old_wim_end, &stream_list); + ret = check_resource_offsets(wim, old_wim_end); if (ret) return ret; - open_flags = 0; - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - open_flags |= O_RDWR; - else - open_flags |= O_WRONLY; - ret = open_wim_writable(w, w->filename, open_flags); + ret = prepare_stream_list(wim, WIMLIB_ALL_IMAGES, write_flags, + &stream_list); if (ret) return ret; - ret = lock_wim(w, w->out_fd); - if (ret) { - close_wim_writable(w); + ret = open_wim_writable(wim, wim->filename, O_RDWR); + if (ret) return ret; + + ret = lock_wim(wim, wim->out_fd.fd); + if (ret) + goto out_close_wim; + + /* Save original header so it can be restored in case of error */ + memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header)); + + /* 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); + if (ret) { + ERROR_WITH_ERRNO("Error updating WIM header flags"); + goto out_restore_memory_hdr; } - if (lseek(w->out_fd, old_wim_end, SEEK_SET) == -1) { + if (filedes_seek(&wim->out_fd, old_wim_end) == -1) { ERROR_WITH_ERRNO("Can't seek to end of WIM"); - close_wim_writable(w); - w->wim_locked = 0; - return WIMLIB_ERR_WRITE; + ret = WIMLIB_ERR_WRITE; + goto out_restore_physical_hdr; } - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); ret = write_stream_list(&stream_list, - w->lookup_table, - w->out_fd, - wimlib_get_compression_type(w), + wim->lookup_table, + &wim->out_fd, + wim->compression_type, + &wim->lzx_context, write_flags, num_threads, progress_func); if (ret) goto out_truncate; - for (int i = 0; i < w->hdr.image_count; i++) { - if (w->image_metadata[i]->modified) { - select_wim_image(w, i + 1); - ret = write_metadata_resource(w); - if (ret) - goto out_truncate; - } - } + ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES, + write_flags, progress_func); + if (ret) + goto out_truncate; + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; - ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, - progress_func); + ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags, + progress_func, NULL); + if (ret) + goto out_truncate; + + goto out_unlock_wim; + out_truncate: - close_wim_writable(w); - if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", - w->filename, old_wim_end); - /* Return value of truncate() is ignored because this is already - * an error path. */ - (void)ttruncate(w->filename, old_wim_end); - } - w->wim_locked = 0; + 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_memory_hdr: + memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header)); +out_close_wim: + (void)close_wim_writable(wim, write_flags); +out_unlock_wim: + wim->wim_locked = 0; return ret; } static int -overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, +overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { size_t wim_name_len; int ret; - DEBUG("Overwriting `%"TS"' via a temporary file", w->filename); + 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(w->filename); + wim_name_len = tstrlen(wim->filename); tchar tmpfile[wim_name_len + 10]; - tmemcpy(tmpfile, w->filename, wim_name_len); + tmemcpy(tmpfile, wim->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); tmpfile[wim_name_len + 9] = T('\0'); - ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, + ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES, write_flags | WIMLIB_WRITE_FLAG_FSYNC, num_threads, progress_func); if (ret) { - ERROR("Failed to write the WIM file `%"TS"'", tmpfile); - goto out_unlink; + tunlink(tmpfile); + return ret; } - close_wim(w); + close_wim(wim); - DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); - /* Rename the new file to the old file .*/ - if (trename(tmpfile, w->filename) != 0) { + /* 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"'", - tmpfile, w->filename); - ret = WIMLIB_ERR_RENAME; - goto out_unlink; + tmpfile, wim->filename); + #ifdef __WIN32__ + if (ret < 0) + #endif + { + tunlink(tmpfile); + } + return WIMLIB_ERR_RENAME; } if (progress_func) { union wimlib_progress_info progress; progress.rename.from = tmpfile; - progress.rename.to = w->filename; + progress.rename.to = wim->filename; progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } - goto out; -out_unlink: - /* Remove temporary file. */ - if (tunlink(tmpfile) != 0) - WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); -out: - return ret; + return 0; } -/* - * Writes a WIM file to the original file that it was read from, overwriting it. - */ +/* API function documented in wimlib.h */ WIMLIBAPI int -wimlib_overwrite(WIMStruct *w, int write_flags, +wimlib_overwrite(WIMStruct *wim, int write_flags, unsigned num_threads, wimlib_progress_func_t progress_func) { + int ret; + u32 orig_hdr_flags; + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - if (!w->filename) + if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR) + return WIMLIB_ERR_INVALID_PARAM; + + if (!wim->filename) return WIMLIB_ERR_NO_FILENAME; - if (w->hdr.total_parts != 1) { - ERROR("Cannot modify a split WIM"); - return WIMLIB_ERR_SPLIT_UNSUPPORTED; - } + orig_hdr_flags = wim->hdr.flags; + if (write_flags & WIMLIB_WRITE_FLAG_IGNORE_READONLY_FLAG) + wim->hdr.flags &= ~WIM_HDR_FLAG_READONLY; + ret = can_modify_wim(wim); + wim->hdr.flags = orig_hdr_flags; + if (ret) + return ret; - if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) - && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) + if ((!wim->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + && !(write_flags & (WIMLIB_WRITE_FLAG_REBUILD | + WIMLIB_WRITE_FLAG_PIPABLE)) + && !(wim_is_pipable(wim))) { - int ret; - ret = overwrite_wim_inplace(w, write_flags, num_threads, + ret = overwrite_wim_inplace(wim, write_flags, num_threads, progress_func); - if (ret == WIMLIB_ERR_RESOURCE_ORDER) - WARNING("Falling back to re-building entire WIM"); - else + if (ret != WIMLIB_ERR_RESOURCE_ORDER) return ret; + WARNING("Falling back to re-building entire WIM"); } - return overwrite_wim_via_tmpfile(w, write_flags, num_threads, + return overwrite_wim_via_tmpfile(wim, write_flags, num_threads, progress_func); }