X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=90dcaed4201d0d92ce6cf74dfc9691dd4bf11025;hp=1eb695d5edca2dbd03a1404e3c7e3b6585f0a0e2;hb=77b2c74599190197074c82937787f24f83ece0af;hpb=9368708ff94a8da723874c098d24fefcd5866207 diff --git a/src/write.c b/src/write.c index 1eb695d5..90dcaed4 100644 --- a/src/write.c +++ b/src/write.c @@ -6,8 +6,7 @@ */ /* - * Copyright (C) 2010 Carl Thijssen - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -25,73 +24,76 @@ * along with wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "wimlib_internal.h" -#include "io.h" -#include "dentry.h" -#include "lookup_table.h" -#include "xml.h" -#include "lzx.h" -#include "xpress.h" -#include +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +/* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can + * overwrite the LIST_HEAD macro. */ +# include +#endif + +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/header.h" +#include "wimlib/integrity.h" +#include "wimlib/lookup_table.h" +#include "wimlib/metadata.h" +#include "wimlib/resource.h" +#include "wimlib/write.h" +#include "wimlib/xml.h" + +#ifdef __WIN32__ +# include "wimlib/win32.h" /* win32_get_number_of_processors() */ +#endif #ifdef ENABLE_MULTITHREADED_COMPRESSION -#include -#include -#include +# include #endif +#include +#include +#include + #ifdef WITH_NTFS_3G -#include -#include -#include -#include +# include +# include +# include +# include #endif - #ifdef HAVE_ALLOCA_H -#include +# include #else -#include +# include #endif -static int do_fflush(FILE *fp) -{ - int ret = fflush(fp); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to flush data to output WIM file"); - return WIMLIB_ERR_WRITE; - } - return 0; -} - -static int fflush_and_ftruncate(FILE *fp, off_t size) -{ - int ret; +#include - ret = do_fflush(fp); - if (ret != 0) - return ret; - ret = ftruncate(fileno(fp), size); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to truncate output WIM file to " - "%"PRIu64" bytes", size); - return WIMLIB_ERR_WRITE; - } - return 0; -} +#ifndef __WIN32__ +# include /* for `struct iovec' */ +#endif /* 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); }; /* @@ -99,69 +101,96 @@ struct chunk_table { * output file. */ static int -begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, - FILE *out_fp, +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) { u64 size = wim_resource_size(lte); - u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; + u64 num_chunks = wim_resource_chunks(lte); + unsigned bytes_per_chunk_entry = (size > (1ULL << 32)) ? 8 : 4; size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64); struct chunk_table *chunk_tab = CALLOC(1, alloc_size); - int ret; + + DEBUG("Beginning chunk table for stream with size %"PRIu64, size); if (!chunk_tab) { ERROR("Failed to allocate chunk table for %"PRIu64" byte " "resource", size); - ret = WIMLIB_ERR_NOMEM; - goto out; + 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 (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) != - chunk_tab->table_disk_size) { + /* We don't know the correct offsets yet; this just writes zeroes to + * reserve space for the table, so we can go back to it later after + * we've written the compressed chunks following it. */ + if (full_write(out_fd, chunk_tab->offsets, + chunk_tab->table_disk_size) != chunk_tab->table_disk_size) + { ERROR_WITH_ERRNO("Failed to write chunk table in compressed " "file resource"); - ret = WIMLIB_ERR_WRITE; - goto out; + FREE(chunk_tab); + return WIMLIB_ERR_WRITE; } - - ret = 0; -out: *chunk_tab_ret = chunk_tab; - return ret; + return 0; +} + +/* 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 (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; + } } /* - * Pointer to function to compresses a chunk of a WIM resource. + * 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. * - * @chunk: Uncompressed data of the chunk. - * @chunk_size: Size of the uncompressed chunk in bytes. - * @compressed_chunk: Pointer to output buffer of size at least - * (@chunk_size - 1) bytes. - * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size - * of the compressed chunk will be - * returned. + * As a special requirement, the compression code is optimized for the WIM + * format and therefore requires (@chunk_size <= 32768). * - * Returns zero if compressed succeeded, and nonzero if the chunk could not be - * compressed to any smaller than @chunk_size. This function cannot fail for - * any other reasons. + * 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 int (*compress_func_t)(const void *, unsigned, void *, unsigned *); +typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size, + void *out); -compress_func_t get_compress_func(int out_ctype) +static compress_func_t +get_compress_func(int out_ctype) { - if (out_ctype == WIM_COMPRESSION_TYPE_LZX) - return lzx_compress; + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) + return wimlib_lzx_compress; else - return xpress_compress; + return wimlib_xpress_compress; } /* @@ -169,39 +198,42 @@ compress_func_t get_compress_func(int out_ctype) * * @chunk: Uncompressed data of the chunk. * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE) - * @out_fp: FILE * to write tho chunk to. - * @out_ctype: Compression type to use when writing the chunk (ignored if no - * chunk table provided) + * @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. */ -static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, - FILE *out_fp, compress_func_t compress, - struct chunk_table *chunk_tab) +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) { - const u8 *out_chunk; + const void *out_chunk; unsigned out_chunk_size; - if (chunk_tab) { - u8 *compressed_chunk = alloca(chunk_size); - int ret; + if (compress) { + void *compressed_chunk = alloca(chunk_size); - ret = compress(chunk, chunk_size, compressed_chunk, - &out_chunk_size); - if (ret == 0) { + 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; + chunk_tab_record_chunk(chunk_tab, out_chunk_size); } else { + /* Write uncompressed */ out_chunk = chunk; out_chunk_size = chunk_size; } - if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { + 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; } @@ -209,7 +241,7 @@ static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, } /* - * Finishes a WIM chunk tale and writes it to the output file at the correct + * Finishes a WIM chunk table and writes it to the output file at the correct * offset. * * The final size of the full compressed resource is returned in the @@ -217,380 +249,305 @@ static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, */ static int finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, - FILE *out_fp, u64 *compressed_size_p) + int out_fd, u64 *compressed_size_p) { size_t bytes_written; - if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output " - "WIM file", chunk_tab->file_offset); - return WIMLIB_ERR_WRITE; - } - if (chunk_tab->bytes_per_chunk_entry == 8) { - array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks); - } else { - for (u64 i = 0; i < chunk_tab->num_chunks; i++) - ((u32*)chunk_tab->offsets)[i] = - cpu_to_le32(chunk_tab->offsets[i]); - } - bytes_written = fwrite((u8*)chunk_tab->offsets + - chunk_tab->bytes_per_chunk_entry, - 1, chunk_tab->table_disk_size, out_fp); + bytes_written = full_pwrite(out_fd, + 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; } - if (fseeko(out_fp, 0, SEEK_END) != 0) { - ERROR_WITH_ERRNO("Failed to seek to end of output WIM file"); - return WIMLIB_ERR_WRITE; - } - *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size; + if (chunk_tab->bytes_per_chunk_entry == 4) + *compressed_size_p = chunk_tab->cur_offset_u32 + chunk_tab->table_disk_size; + else + *compressed_size_p = chunk_tab->cur_offset_u64 + chunk_tab->table_disk_size; return 0; } -/* Prepare for multiple reads to a resource by caching a FILE * or NTFS - * attribute pointer in the lookup table entry. */ -static int prepare_resource_for_read(struct lookup_table_entry *lte - - #ifdef WITH_NTFS_3G - , ntfs_inode **ni_ret - #endif - ) +static int +seek_and_truncate(int out_fd, off_t offset) { - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && !lte->file_on_disk_fp) + if (lseek(out_fd, offset, SEEK_SET) == -1 || + ftruncate(out_fd, offset)) { - wimlib_assert(lte->file_on_disk); - lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb"); - if (!lte->file_on_disk_fp) { - ERROR_WITH_ERRNO("Failed to open the file `%s' for " - "reading", lte->file_on_disk); - return WIMLIB_ERR_OPEN; - } + ERROR_WITH_ERRNO("Failed to truncate output WIM file"); + return WIMLIB_ERR_WRITE; + } else { + return 0; } -#ifdef WITH_NTFS_3G - else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME - && !lte->attr) - { - struct ntfs_location *loc = lte->ntfs_loc; - ntfs_inode *ni; - wimlib_assert(loc); - ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8); - if (!ni) { - ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS " - "volume", loc->path_utf8); - return WIMLIB_ERR_NTFS_3G; - } - lte->attr = ntfs_attr_open(ni, - loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, - (ntfschar*)loc->stream_name_utf16, - loc->stream_name_utf16_num_chars); - if (!lte->attr) { - ERROR_WITH_ERRNO("Failed to open attribute of `%s' in " - "NTFS volume", loc->path_utf8); - ntfs_inode_close(ni); - return WIMLIB_ERR_NTFS_3G; +} + +static int +finalize_and_check_sha1(SHA_CTX * restrict sha_ctx, + struct wim_lookup_table_entry * restrict lte) +{ + u8 md[SHA1_HASH_SIZE]; + sha1_final(md, sha_ctx); + if (lte->unhashed) { + copy_hash(lte->hash, md); + } else if (!hashes_equal(md, lte->hash)) { + ERROR("WIM resource has incorrect hash!"); + if (lte_filename_valid(lte)) { + ERROR("We were reading it from \"%"TS"\"; maybe " + "it changed while we were reading it.", + lte->file_on_disk); } - *ni_ret = ni; + return WIMLIB_ERR_INVALID_RESOURCE_HASH; } -#endif return 0; } -/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS - * attribute. */ -static void end_wim_resource_read(struct lookup_table_entry *lte - #ifdef WITH_NTFS_3G - , ntfs_inode *ni - #endif - ) + +struct write_resource_ctx { + compress_func_t compress; + struct chunk_table *chunk_tab; + int out_fd; + SHA_CTX sha_ctx; + bool doing_sha; +}; + +static int +write_resource_cb(const void *restrict chunk, size_t chunk_size, + void *restrict _ctx) { - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && lte->file_on_disk_fp) { - fclose(lte->file_on_disk_fp); - lte->file_on_disk_fp = NULL; - } -#ifdef WITH_NTFS_3G - else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) { - if (lte->attr) { - ntfs_attr_close(lte->attr); - lte->attr = NULL; - } - if (ni) - ntfs_inode_close(ni); - } -#endif + struct write_resource_ctx *ctx = _ctx; + + 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); } /* - * Writes a WIM resource to a FILE * opened for writing. The resource may be - * written uncompressed or compressed depending on the @out_ctype parameter. + * Write a resource to an output WIM. * - * If by chance the resource compresses to more than the original size (this may - * happen with random data or files than are pre-compressed), the resource is - * instead written uncompressed (and this is reflected in the @out_res_entry by - * removing the WIM_RESHDR_FLAG_COMPRESSED flag). + * @lte: Lookup table entry for the resource, which could be in another WIM, + * in an external file, or in another location. * - * @lte: The lookup table entry for the WIM resource. - * @out_fp: The FILE * to write the resource to. - * @out_ctype: The compression type of the resource to write. Note: if this is - * the same as the compression type of the WIM resource we - * need to read, we simply copy the data (i.e. we do not - * uncompress it, then compress it again). - * @out_res_entry: If non-NULL, a resource entry that is filled in with the - * offset, original size, compressed size, and compression flag - * of the output resource. + * @out_fd: File descriptor opened to the output WIM. * - * Returns 0 on success; nonzero on failure. + * @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. + * + * @flags: WIMLIB_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed + * even if it could otherwise be copied directly from the input. + * + * Additional notes: The SHA1 message digest of the uncompressed data is + * calculated (except when doing a raw copy --- see below). If the @unhashed + * flag is set on the lookup table entry, this message digest is simply copied + * to it; otherwise, the message digest is compared with the existing one, and + * the function will fail if they do not match. */ -int write_wim_resource(struct lookup_table_entry *lte, - FILE *out_fp, int out_ctype, - struct resource_entry *out_res_entry, - int flags) +int +write_wim_resource(struct wim_lookup_table_entry *lte, + int out_fd, int out_ctype, + struct resource_entry *out_res_entry, + int flags) { - u64 bytes_remaining; - u64 original_size; - u64 old_compressed_size; - u64 new_compressed_size; - u64 offset; + struct write_resource_ctx write_ctx; + u64 read_size; + u64 new_size; + off_t offset; int ret; - struct chunk_table *chunk_tab = NULL; - bool raw; - off_t file_offset; - compress_func_t compress; -#ifdef WITH_NTFS_3G - ntfs_inode *ni = NULL; -#endif - wimlib_assert(lte); + flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS; - /* Original size of the resource */ - original_size = wim_resource_size(lte); - - /* Compressed size of the resource (as it exists now) */ - old_compressed_size = wim_resource_compressed_size(lte); - - /* Current offset in output file */ - file_offset = ftello(out_fp); - if (file_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset in output " - "stream"); + /* 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; } - /* Are the compression types the same? If so, do a raw copy (copy - * without decompressing and recompressing the data). */ - raw = (wim_resource_compression_type(lte) == out_ctype - && out_ctype != WIM_COMPRESSION_TYPE_NONE - && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); - - if (raw) { + /* 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 && + out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && + wimlib_get_compression_type(lte->wim) == out_ctype) + { flags |= WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = old_compressed_size; + write_ctx.doing_sha = false; + read_size = lte->resource_entry.size; } else { - flags &= ~WIMLIB_RESOURCE_FLAG_RAW; - bytes_remaining = original_size; + write_ctx.doing_sha = true; + sha1_init(&write_ctx.sha_ctx); + read_size = lte->resource_entry.original_size; } - /* Empty resource; nothing needs to be done, so just return success. */ - if (bytes_remaining == 0) - return 0; - - /* Buffer for reading chunks for the resource */ - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - - /* If we are writing a compressed resource and not doing a raw copy, we - * need to initialize the chunk table */ - if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) { - ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, - &chunk_tab); - if (ret != 0) - goto out; - } - - /* If the WIM resource is in an external file, open a FILE * to it so we - * don't have to open a temporary one in read_wim_resource() for each - * chunk. */ -#ifdef WITH_NTFS_3G - ret = prepare_resource_for_read(lte, &ni); -#else - ret = prepare_resource_for_read(lte); -#endif - if (ret != 0) - goto out; - - /* If we aren't doing a raw copy, we will compute the SHA1 message - * digest of the resource as we read it, and verify it's the same as the - * hash given in the lookup table entry once we've finished reading the - * resource. */ - SHA_CTX ctx; - if (!raw) { - sha1_init(&ctx); - compress = get_compress_func(out_ctype); - } - offset = 0; - - /* While there are still bytes remaining in the WIM resource, read a - * chunk of the resource, update SHA1, then write that chunk using the - * desired compression type. */ - do { - u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, flags); - if (ret != 0) - goto out_fclose; - if (!raw) - sha1_update(&ctx, buf, to_read); - ret = write_wim_resource_chunk(buf, to_read, out_fp, - compress, chunk_tab); - if (ret != 0) - goto out_fclose; - bytes_remaining -= to_read; - offset += to_read; - } while (bytes_remaining); - - /* Raw copy: The new compressed size is the same as the old compressed - * size - * - * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the - * original size - * - * Using a different compression type: Call - * finish_wim_resource_chunk_tab() and it will provide the new - * compressed size. - */ - if (raw) { - new_compressed_size = old_compressed_size; + /* 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 { - if (out_ctype == WIM_COMPRESSION_TYPE_NONE) - new_compressed_size = original_size; - else { - ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp, - &new_compressed_size); - if (ret != 0) - goto out_fclose; - } - } - - /* Verify SHA1 message digest of the resource, unless we are doing a raw - * write (in which case we never even saw the uncompressed data). Or, - * if the hash we had before is all 0's, just re-set it to be the new - * hash. */ - if (!raw) { - u8 md[SHA1_HASH_SIZE]; - sha1_final(md, &ctx); - if (is_zero_hash(lte->hash)) { - copy_hash(lte->hash, md); - } else if (!hashes_equal(md, lte->hash)) { - ERROR("WIM resource has incorrect hash!"); - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; maybe it changed " - "while we were reading it.", - lte->file_on_disk); - } - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out_fclose; - } + write_ctx.compress = get_compress_func(out_ctype); + ret = begin_wim_resource_chunk_tab(lte, out_fd, + offset, + &write_ctx.chunk_tab); + if (ret) + return ret; } - if (!raw && new_compressed_size >= original_size && - out_ctype != WIM_COMPRESSION_TYPE_NONE) - { - /* Oops! We compressed the resource to larger than the original - * size. Write the resource uncompressed instead. */ - if (fseeko(out_fp, file_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" " - "of output WIM file", file_offset); - ret = WIMLIB_ERR_WRITE; - goto out_fclose; - } - ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE, - out_res_entry, flags); - if (ret != 0) - goto out_fclose; - - ret = fflush_and_ftruncate(out_fp, file_offset + out_res_entry->size); - if (ret != 0) - goto out_fclose; + /* 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; +try_write_again: + ret = read_resource_prefix(lte, read_size, + write_resource_cb, &write_ctx, flags); + if (ret) + goto out_free_chunk_tab; + + /* Verify SHA1 message digest of the resource, or set the hash for the + * first time. */ + if (write_ctx.doing_sha) { + ret = finalize_and_check_sha1(&write_ctx.sha_ctx, lte); + if (ret) + 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 { - if (out_res_entry) { - out_res_entry->size = new_compressed_size; - out_res_entry->original_size = original_size; - out_res_entry->offset = file_offset; - out_res_entry->flags = lte->resource_entry.flags - & ~WIM_RESHDR_FLAG_COMPRESSED; - if (out_ctype != WIM_COMPRESSION_TYPE_NONE) - out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; + /* 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); + 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; } + out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } + out_res_entry->size = new_size; ret = 0; -out_fclose: -#ifdef WITH_NTFS_3G - end_wim_resource_read(lte, ni); -#else - end_wim_resource_read(lte); -#endif -out: - FREE(chunk_tab); +out_free_chunk_tab: + FREE(write_ctx.chunk_tab); return ret; } #ifdef ENABLE_MULTITHREADED_COMPRESSION + +/* Blocking shared queue (solves the producer-consumer problem) */ struct shared_queue { - sem_t filled_slots; - sem_t empty_slots; - pthread_mutex_t lock; + unsigned size; unsigned front; unsigned back; + unsigned filled_slots; void **array; - unsigned size; + pthread_mutex_t lock; + pthread_cond_t msg_avail_cond; + pthread_cond_t space_avail_cond; }; -static int shared_queue_init(struct shared_queue *q, unsigned size) +static int +shared_queue_init(struct shared_queue *q, unsigned size) { + wimlib_assert(size != 0); q->array = CALLOC(sizeof(q->array[0]), size); if (!q->array) - return WIMLIB_ERR_NOMEM; - - sem_init(&q->filled_slots, 0, 0); - sem_init(&q->empty_slots, 0, size); - pthread_mutex_init(&q->lock, NULL); + goto err; + q->filled_slots = 0; q->front = 0; q->back = size - 1; q->size = size; + if (pthread_mutex_init(&q->lock, NULL)) { + ERROR_WITH_ERRNO("Failed to initialize mutex"); + goto err; + } + if (pthread_cond_init(&q->msg_avail_cond, NULL)) { + ERROR_WITH_ERRNO("Failed to initialize condition variable"); + goto err_destroy_lock; + } + if (pthread_cond_init(&q->space_avail_cond, NULL)) { + ERROR_WITH_ERRNO("Failed to initialize condition variable"); + goto err_destroy_msg_avail_cond; + } return 0; +err_destroy_msg_avail_cond: + pthread_cond_destroy(&q->msg_avail_cond); +err_destroy_lock: + pthread_mutex_destroy(&q->lock); +err: + return WIMLIB_ERR_NOMEM; } -static void shared_queue_destroy(struct shared_queue *q) +static void +shared_queue_destroy(struct shared_queue *q) { - sem_destroy(&q->filled_slots); - sem_destroy(&q->empty_slots); - pthread_mutex_destroy(&q->lock); FREE(q->array); + pthread_mutex_destroy(&q->lock); + pthread_cond_destroy(&q->msg_avail_cond); + pthread_cond_destroy(&q->space_avail_cond); } -static void shared_queue_put(struct shared_queue *q, void *obj) +static void +shared_queue_put(struct shared_queue *q, void *obj) { - sem_wait(&q->empty_slots); pthread_mutex_lock(&q->lock); + while (q->filled_slots == q->size) + pthread_cond_wait(&q->space_avail_cond, &q->lock); q->back = (q->back + 1) % q->size; q->array[q->back] = obj; + q->filled_slots++; - sem_post(&q->filled_slots); + pthread_cond_broadcast(&q->msg_avail_cond); pthread_mutex_unlock(&q->lock); } -static void *shared_queue_get(struct shared_queue *q) +static void * +shared_queue_get(struct shared_queue *q) { - sem_wait(&q->filled_slots); + void *obj; + pthread_mutex_lock(&q->lock); + while (q->filled_slots == 0) + pthread_cond_wait(&q->msg_avail_cond, &q->lock); - void *obj = q->array[q->front]; + obj = q->array[q->front]; q->array[q->front] = NULL; q->front = (q->front + 1) % q->size; + q->filled_slots--; - sem_post(&q->empty_slots); + pthread_cond_broadcast(&q->space_avail_cond); pthread_mutex_unlock(&q->lock); return obj; } @@ -604,36 +561,50 @@ struct compressor_thread_params { #define MAX_CHUNKS_PER_MSG 2 struct message { - struct lookup_table_entry *lte; + struct wim_lookup_table_entry *lte; u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG]; - u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG]; u8 *compressed_chunks[MAX_CHUNKS_PER_MSG]; unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG]; - unsigned compressed_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; u64 begin_chunk; }; -static void compress_chunks(struct message *msg, compress_func_t compress) +static void +compress_chunks(struct message *msg, compress_func_t compress) { + msg->total_out_bytes = 0; for (unsigned i = 0; i < msg->num_chunks; i++) { - DEBUG2("compress chunk %u of %u", i, msg->num_chunks); - int ret = compress(msg->uncompressed_chunks[i], - msg->uncompressed_chunk_sizes[i], - msg->compressed_chunks[i], - &msg->compressed_chunk_sizes[i]); - if (ret == 0) { - msg->out_compressed_chunks[i] = msg->compressed_chunks[i]; + unsigned len = compress(msg->uncompressed_chunks[i], + msg->uncompressed_chunk_sizes[i], + msg->compressed_chunks[i]); + void *out_chunk; + unsigned out_len; + if (len) { + /* To be written compressed */ + out_chunk = msg->compressed_chunks[i]; + out_len = len; } else { - msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i]; - msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i]; + /* To be written uncompressed */ + out_chunk = msg->uncompressed_chunks[i]; + out_len = msg->uncompressed_chunk_sizes[i]; } + msg->out_chunks[i].iov_base = out_chunk; + msg->out_chunks[i].iov_len = out_len; + msg->total_out_bytes += out_len; } } -static void *compressor_thread_proc(void *arg) +/* Compressor thread routine. This is a lot simpler than the main thread + * routine: just repeatedly get a group of chunks from the + * res_to_compress_queue, compress them, and put them in the + * compressed_res_queue. A NULL pointer indicates that the thread should stop. + * */ +static void * +compressor_thread_proc(void *arg) { struct compressor_thread_params *params = arg; struct shared_queue *res_to_compress_queue = params->res_to_compress_queue; @@ -647,417 +618,426 @@ static void *compressor_thread_proc(void *arg) shared_queue_put(compressed_res_queue, msg); } DEBUG("Compressor thread terminating"); + return NULL; } -#endif +#endif /* ENABLE_MULTITHREADED_COMPRESSION */ -static void show_stream_write_progress(u64 *cur_size, u64 *next_size, - u64 total_size, u64 one_percent, - unsigned *cur_percent, - const struct lookup_table_entry *cur_lte) +static void +do_write_streams_progress(union wimlib_progress_info *progress, + wimlib_progress_func_t progress_func, + uint64_t size_added, + bool stream_discarded) { - if (*cur_size >= *next_size) { - printf("\r%"PRIu64" MiB of %"PRIu64" MiB " - "(uncompressed) written (%u%% done)", - *cur_size >> 20, - total_size >> 20, *cur_percent); - fflush(stdout); - *next_size += one_percent; - (*cur_percent)++; - } - *cur_size += wim_resource_size(cur_lte); + 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._private = progress->write_streams.total_bytes; + } + } else { + progress->write_streams.completed_bytes += size_added; + } + progress->write_streams.completed_streams++; + if (progress_func && + progress->write_streams.completed_bytes >= progress->write_streams._private) + { + 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; + } else { + progress->write_streams._private = + min(progress->write_streams.total_bytes, + progress->write_streams.completed_bytes + + progress->write_streams.total_bytes / 100); + } + } } -static void finish_stream_write_progress(u64 total_size) +struct serial_write_stream_ctx { + int out_fd; + int out_ctype; + int write_resource_flags; +}; + +static int +serial_write_stream(struct wim_lookup_table_entry *lte, void *_ctx) { - printf("\r%"PRIu64" MiB of %"PRIu64" MiB " - "(uncompressed) written (100%% done)\n", - total_size >> 20, total_size >> 20); - fflush(stdout); + 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); } -static int write_stream_list_serial(struct list_head *stream_list, - FILE *out_fp, int out_ctype, - int write_flags, u64 total_size) +/* 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 + * passed the context @write_stream_ctx. */ +static int +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 lookup_table_entry *lte; - int ret; - - u64 one_percent = total_size / 100; - u64 cur_size = 0; - u64 next_size = 0; - unsigned cur_percent = 0; - int write_resource_flags = 0; - - if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; + int ret = 0; + struct wim_lookup_table_entry *lte; + bool stream_discarded; + + /* For each stream in @stream_list ... */ + while (!list_empty(stream_list)) { + stream_discarded = false; + lte = container_of(stream_list->next, + struct wim_lookup_table_entry, + write_streams_list); + list_del(<e->write_streams_list); + if (lte->unhashed && !lte->unique_size) { + /* Unhashed stream that shares a size with some other + * 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; + + ret = hash_unhashed_stream(lte, lookup_table, &tmp); + if (ret) + break; + if (tmp != lte) { + 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, + wim_resource_size(lte)); + lte->no_progress = 0; + stream_discarded = true; + goto skip_to_progress; + } + } + } - list_for_each_entry(lte, stream_list, staging_list) { - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, &next_size, - total_size, one_percent, - &cur_percent, lte); + /* Here, @lte is either a hashed stream or an unhashed stream + * with a unique size. In either case we know that the stream + * has to be written. In either case the SHA1 message digest + * will be calculated over the stream while writing it; however, + * in the former case this is done merely to check the data, + * while in the latter case this is done because we do not have + * the SHA1 message digest yet. */ + wimlib_assert(lte->out_refcnt != 0); + lte->deferred = 0; + lte->no_progress = 0; + ret = (*write_stream_cb)(lte, write_stream_ctx); + if (ret) + break; + /* In parallel mode, some streams are deferred for later, + * serialized processing; ignore them here. */ + if (lte->deferred) + continue; + if (lte->unhashed) { + list_del(<e->unhashed_list); + lookup_table_insert(lookup_table, lte); + lte->unhashed = 0; + } + skip_to_progress: + if (!lte->no_progress) { + do_write_streams_progress(progress, + progress_func, + wim_resource_size(lte), + stream_discarded); } - ret = write_wim_resource(lte, out_fp, out_ctype, - <e->output_resource_entry, - write_resource_flags); - if (ret != 0) - return ret; } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - finish_stream_write_progress(total_size); - return 0; + return ret; } -#ifdef ENABLE_MULTITHREADED_COMPRESSION -static int write_wim_chunks(struct message *msg, FILE *out_fp, - struct chunk_table *chunk_tab) +static int +do_write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + int out_fd, + int out_ctype, + int write_resource_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { - for (unsigned i = 0; i < msg->num_chunks; i++) { - unsigned chunk_csize = msg->compressed_chunk_sizes[i]; + struct serial_write_stream_ctx ctx = { + .out_fd = out_fd, + .out_ctype = out_ctype, + .write_resource_flags = write_resource_flags, + }; + return do_write_stream_list(stream_list, + lookup_table, + serial_write_stream, + &ctx, + progress_func, + progress); +} - DEBUG2("Write wim chunk %u of %u (csize = %u)", - i, msg->num_chunks, chunk_csize); +static inline int +write_flags_to_resource_flags(int write_flags) +{ + int resource_flags = 0; - if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) - != chunk_csize) - { - ERROR_WITH_ERRNO("Failed to write WIM chunk"); - return WIMLIB_ERR_WRITE; - } + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS; + return resource_flags; +} - *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; - chunk_tab->cur_offset += chunk_csize; +static int +write_stream_list_serial(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + int out_fd, + int out_ctype, + int write_resource_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) +{ + DEBUG("Writing stream list (serial version)"); + progress->write_streams.num_threads = 1; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + return do_write_stream_list_serial(stream_list, + lookup_table, + out_fd, + out_ctype, + write_resource_flags, + progress_func, + progress); +} + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +static int +write_wim_chunks(struct message *msg, int out_fd, + struct chunk_table *chunk_tab) +{ + for (unsigned i = 0; i < msg->num_chunks; i++) + chunk_tab_record_chunk(chunk_tab, msg->out_chunks[i].iov_len); + if (full_writev(out_fd, msg->out_chunks, + msg->num_chunks) != msg->total_out_bytes) + { + ERROR_WITH_ERRNO("Failed to write WIM chunks"); + return WIMLIB_ERR_WRITE; } return 0; } -/* - * This function is executed by the main thread when the resources are being - * compressed in parallel. The main thread is in change of all reading of the - * uncompressed data and writing of the compressed data. The compressor threads - * *only* do compression from/to in-memory buffers. - * - * Each unit of work given to a compressor thread is up to MAX_CHUNKS_PER_MSG - * chunks of compressed data to compress, represented in a `struct message'. - * Each message is passed from the main thread to a worker thread through the - * res_to_compress_queue, and it is passed back through the - * compressed_res_queue. - */ -static int main_writer_thread_proc(struct list_head *stream_list, - FILE *out_fp, - int out_ctype, - struct shared_queue *res_to_compress_queue, - struct shared_queue *compressed_res_queue, - size_t queue_size, - int write_flags, - u64 total_size) -{ - int ret; +struct main_writer_thread_ctx { + struct list_head *stream_list; + struct wim_lookup_table *lookup_table; + int out_fd; + int out_ctype; + 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 list_head available_msgs; + struct list_head outstanding_streams; + struct list_head serial_streams; + size_t num_outstanding_messages; - struct message msgs[queue_size]; - ZERO_ARRAY(msgs); - - // Initially, all the messages are available to use. - LIST_HEAD(available_msgs); - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) - list_add(&msgs[i].list, &available_msgs); - - // outstanding_resources is the list of resources that currently have - // had chunks sent off for compression. - // - // The first stream in outstanding_resources is the stream that is - // currently being written (cur_lte). - // - // The last stream in outstanding_resources is the stream that is - // currently being read and chunks fed to the compressor threads - // (next_lte). - // - // Depending on the number of threads and the sizes of the resource, - // the outstanding streams list may contain streams between cur_lte and - // next_lte that have all their chunks compressed or being compressed, - // but haven't been written yet. - // - LIST_HEAD(outstanding_resources); - struct list_head *next_resource = stream_list->next; - struct lookup_table_entry *next_lte = container_of(next_resource, - struct lookup_table_entry, - staging_list); - next_resource = next_resource->next; - u64 next_chunk = 0; - u64 next_num_chunks = wim_resource_chunks(next_lte); - INIT_LIST_HEAD(&next_lte->msg_list); - list_add_tail(&next_lte->staging_list, &outstanding_resources); - - // As in write_wim_resource(), each resource we read is checksummed. SHA_CTX next_sha_ctx; - sha1_init(&next_sha_ctx); - u8 next_hash[SHA1_HASH_SIZE]; + u64 next_chunk; + u64 next_num_chunks; + struct wim_lookup_table_entry *next_lte; - // Resources that don't need any chunks compressed are added to this - // list and written directly by the main thread. - LIST_HEAD(my_resources); + struct message *msgs; + struct message *next_msg; + struct chunk_table *cur_chunk_tab; +}; - struct lookup_table_entry *cur_lte = next_lte; - struct chunk_table *cur_chunk_tab = NULL; - struct lookup_table_entry *lte; - struct message *msg; +static int +init_message(struct message *msg) +{ + for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { + msg->compressed_chunks[i] = MALLOC(WIM_CHUNK_SIZE); + msg->uncompressed_chunks[i] = MALLOC(WIM_CHUNK_SIZE); + if (msg->compressed_chunks[i] == NULL || + msg->uncompressed_chunks[i] == NULL) + return WIMLIB_ERR_NOMEM; + } + return 0; +} - u64 one_percent = total_size / 100; - u64 cur_size = 0; - u64 next_size = 0; - unsigned cur_percent = 0; +static void +destroy_message(struct message *msg) +{ + for (size_t i = 0; i < MAX_CHUNKS_PER_MSG; i++) { + FREE(msg->compressed_chunks[i]); + FREE(msg->uncompressed_chunks[i]); + } +} -#ifdef WITH_NTFS_3G - ntfs_inode *ni = NULL; -#endif +static void +free_messages(struct message *msgs, size_t num_messages) +{ + if (msgs) { + for (size_t i = 0; i < num_messages; i++) + destroy_message(&msgs[i]); + FREE(msgs); + } +} -#ifdef WITH_NTFS_3G - ret = prepare_resource_for_read(next_lte, &ni); -#else - ret = prepare_resource_for_read(next_lte); -#endif - if (ret != 0) - goto out; - - DEBUG("Initializing buffers for uncompressed " - "and compressed data (%zu bytes needed)", - queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2); - - // Pre-allocate all the buffers that will be needed to do the chunk - // compression. - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { - for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { - msgs[i].compressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); - msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE); - if (msgs[i].compressed_chunks[j] == NULL || - msgs[i].uncompressed_chunks[j] == NULL) - { - ERROR("Could not allocate enough memory for " - "multi-threaded compression"); - ret = WIMLIB_ERR_NOMEM; - goto out; - } +static struct message * +allocate_messages(size_t num_messages) +{ + struct message *msgs; + + msgs = CALLOC(num_messages, sizeof(struct message)); + if (!msgs) + return NULL; + for (size_t i = 0; i < num_messages; i++) { + if (init_message(&msgs[i])) { + free_messages(msgs, num_messages); + return NULL; } } + return msgs; +} - // This loop is executed until all resources have been written, except - // possibly a few that have been added to the @my_resources list for - // writing later. - while (1) { - // Send chunks to the compressor threads until either (a) there - // are no more messages available since they were all sent off, - // or (b) there are no more resources that need to be - // compressed. - while (!list_empty(&available_msgs) && next_lte != NULL) { - - // Get a message from the available messages - // list - msg = container_of(available_msgs.next, - struct message, - list); - - // ... and delete it from the available messages - // list - list_del(&msg->list); - - // Initialize the message with the chunks to - // compress. - msg->num_chunks = min(next_num_chunks - next_chunk, - MAX_CHUNKS_PER_MSG); - msg->lte = next_lte; - msg->complete = false; - msg->begin_chunk = next_chunk; - - unsigned size = WIM_CHUNK_SIZE; - for (unsigned i = 0; i < msg->num_chunks; i++) { - - // Read chunk @next_chunk of the stream into the - // message so that a compressor thread can - // compress it. - - if (next_chunk == next_num_chunks - 1 && - wim_resource_size(next_lte) % WIM_CHUNK_SIZE != 0) - { - size = wim_resource_size(next_lte) % WIM_CHUNK_SIZE; - } - - - DEBUG2("Read resource (size=%u, offset=%zu)", - size, next_chunk * WIM_CHUNK_SIZE); - - msg->uncompressed_chunk_sizes[i] = size; - - ret = read_wim_resource(next_lte, - msg->uncompressed_chunks[i], - size, - next_chunk * WIM_CHUNK_SIZE, - 0); - if (ret != 0) - goto out; - sha1_update(&next_sha_ctx, - msg->uncompressed_chunks[i], size); - next_chunk++; - } +static void +main_writer_thread_destroy_ctx(struct main_writer_thread_ctx *ctx) +{ + while (ctx->num_outstanding_messages--) + shared_queue_get(ctx->compressed_res_queue); + free_messages(ctx->msgs, ctx->num_messages); + FREE(ctx->cur_chunk_tab); +} - // Send the compression request - list_add_tail(&msg->list, &next_lte->msg_list); - shared_queue_put(res_to_compress_queue, msg); - DEBUG2("Compression request sent"); - - if (next_chunk != next_num_chunks) - // More chunks to send for this resource - continue; - - // Done sending compression requests for a resource! - // Check the SHA1 message digest. - DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)", next_num_chunks); - sha1_final(next_hash, &next_sha_ctx); - if (!hashes_equal(next_lte->hash, next_hash)) { - ERROR("WIM resource has incorrect hash!"); - if (next_lte->resource_location == RESOURCE_IN_FILE_ON_DISK) { - ERROR("We were reading it from `%s'; maybe it changed " - "while we were reading it.", - next_lte->file_on_disk); - } - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - goto out; - } +static int +main_writer_thread_init_ctx(struct main_writer_thread_ctx *ctx) +{ + /* Pre-allocate all the buffers that will be needed to do the chunk + * compression. */ + ctx->msgs = allocate_messages(ctx->num_messages); + if (!ctx->msgs) + return WIMLIB_ERR_NOMEM; - // Advance to the next resource. - // - // If the next resource needs no compression, just write - // it with this thread (not now though--- we could be in - // the middle of writing another resource.) Keep doing - // this until we either get to the end of the resources - // list, or we get to a resource that needs compression. - - while (1) { - if (next_resource == stream_list) { - next_lte = NULL; - break; - } - #ifdef WITH_NTFS_3G - end_wim_resource_read(next_lte, ni); - ni = NULL; - #else - end_wim_resource_read(next_lte); - #endif - - next_lte = container_of(next_resource, - struct lookup_table_entry, - staging_list); - next_resource = next_resource->next; - if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) - && next_lte->resource_location == RESOURCE_IN_WIM - && wimlib_get_compression_type(next_lte->wim) == out_ctype) - || wim_resource_size(next_lte) == 0) - { - list_add_tail(&next_lte->staging_list, - &my_resources); - } else { - list_add_tail(&next_lte->staging_list, - &outstanding_resources); - next_chunk = 0; - next_num_chunks = wim_resource_chunks(next_lte); - sha1_init(&next_sha_ctx); - INIT_LIST_HEAD(&next_lte->msg_list); - #ifdef WITH_NTFS_3G - ret = prepare_resource_for_read(next_lte, &ni); - #else - ret = prepare_resource_for_read(next_lte); - #endif - if (ret != 0) - goto out; - DEBUG2("Updated next_lte"); - break; - } - } - } + /* Initially, all the messages are available to use. */ + INIT_LIST_HEAD(&ctx->available_msgs); + for (size_t i = 0; i < ctx->num_messages; i++) + list_add_tail(&ctx->msgs[i].list, &ctx->available_msgs); - // If there are no outstanding resources, there are no more - // resources that need to be written. - if (list_empty(&outstanding_resources)) { - DEBUG("No outstanding resources! Done"); - ret = 0; - goto out; - } + /* outstanding_streams is the list of streams that currently have had + * chunks sent off for compression. + * + * The first stream in outstanding_streams is the stream that is + * currently being written. + * + * The last stream in outstanding_streams is the stream that is + * currently being read and having chunks fed to the compressor threads. + * */ + INIT_LIST_HEAD(&ctx->outstanding_streams); + ctx->num_outstanding_messages = 0; - // Get the next message from the queue and process it. - // The message will contain 1 or more data chunks that have been - // compressed. - DEBUG2("Waiting for message"); - msg = shared_queue_get(compressed_res_queue); - msg->complete = true; + ctx->next_msg = NULL; - DEBUG2("Received msg (begin_chunk=%"PRIu64")", msg->begin_chunk); + /* Resources that don't need any chunks compressed are added to this + * list and written directly by the main thread. */ + INIT_LIST_HEAD(&ctx->serial_streams); - list_for_each_entry(msg, &cur_lte->msg_list, list) { - DEBUG2("complete=%d", msg->complete); - } + ctx->cur_chunk_tab = NULL; - // Is this the next chunk in the current resource? If it's not - // (i.e., an earlier chunk in a same or different resource - // hasn't been compressed yet), do nothing, and keep this - // message around until all earlier chunks are received. - // - // Otherwise, write all the chunks we can. - while (!list_empty(&cur_lte->msg_list) - && (msg = container_of(cur_lte->msg_list.next, - struct message, - list))->complete) - { - DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk); - if (msg->begin_chunk == 0) { - DEBUG2("Begin chunk tab"); - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, - &next_size, - total_size, - one_percent, - &cur_percent, - cur_lte); - } + return 0; +} - // This is the first set of chunks. Leave space - // for the chunk table in the output file. - off_t cur_offset = ftello(out_fp); - if (cur_offset == -1) { - ret = WIMLIB_ERR_WRITE; - goto out; - } - ret = begin_wim_resource_chunk_tab(cur_lte, - out_fp, - cur_offset, - &cur_chunk_tab); - if (ret != 0) - goto out; - } +static int +receive_compressed_chunks(struct main_writer_thread_ctx *ctx) +{ + struct message *msg; + struct wim_lookup_table_entry *cur_lte; + int ret; - // Write the compressed chunks from the message. - ret = write_wim_chunks(msg, out_fp, cur_chunk_tab); - if (ret != 0) - goto out; + wimlib_assert(!list_empty(&ctx->outstanding_streams)); + wimlib_assert(ctx->num_outstanding_messages != 0); - list_del(&msg->list); + cur_lte = container_of(ctx->outstanding_streams.next, + struct wim_lookup_table_entry, + being_compressed_list); - // This message is available to use for different chunks - // now. - list_add(&msg->list, &available_msgs); + /* Get the next message from the queue and process it. + * The message will contain 1 or more data chunks that have been + * compressed. */ + msg = shared_queue_get(ctx->compressed_res_queue); + msg->complete = true; + --ctx->num_outstanding_messages; - // Was this the last chunk of the stream? If so, - // finish it. - if (list_empty(&cur_lte->msg_list) && - msg->begin_chunk + msg->num_chunks == cur_chunk_tab->num_chunks) - { - DEBUG2("Finish wim chunk tab"); - u64 res_csize; - ret = finish_wim_resource_chunk_tab(cur_chunk_tab, - out_fp, - &res_csize); - if (ret != 0) - goto out; + /* Is this the next chunk in the current resource? If it's not + * (i.e., an earlier chunk in a same or different resource + * hasn't been compressed yet), do nothing, and keep this + * message around until all earlier chunks are received. + * + * Otherwise, write all the chunks we can. */ + while (cur_lte != NULL && + !list_empty(&cur_lte->msg_list) + && (msg = container_of(cur_lte->msg_list.next, + struct message, + list))->complete) + { + 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; + ret = begin_wim_resource_chunk_tab(cur_lte, + ctx->out_fd, + cur_offset, + &ctx->cur_chunk_tab); + if (ret) + return ret; + } + /* Write the compressed chunks from the message. */ + ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab); + if (ret) + return ret; + /* Was this the last chunk of the stream? If so, finish + * it. */ + if (list_empty(&cur_lte->msg_list) && + msg->begin_chunk + msg->num_chunks == 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); + 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; + + 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. */ + DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; " + "writing uncompressed instead", + wim_resource_size(cur_lte), res_csize); + ret = seek_and_truncate(ctx->out_fd, 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); + if (ret) + return ret; + } else { cur_lte->output_resource_entry.size = res_csize; @@ -1065,133 +1045,229 @@ static int main_writer_thread_proc(struct list_head *stream_list, cur_lte->resource_entry.original_size; cur_lte->output_resource_entry.offset = - cur_chunk_tab->file_offset; + offset; cur_lte->output_resource_entry.flags = cur_lte->resource_entry.flags | WIM_RESHDR_FLAG_COMPRESSED; + } - FREE(cur_chunk_tab); - cur_chunk_tab = NULL; - - struct list_head *next = cur_lte->staging_list.next; - list_del(&cur_lte->staging_list); - - if (next == &outstanding_resources) { - DEBUG("No more outstanding resources"); - ret = 0; - goto out; - } else { - cur_lte = container_of(cur_lte->staging_list.next, - struct lookup_table_entry, - staging_list); - } + do_write_streams_progress(ctx->progress, + ctx->progress_func, + wim_resource_size(cur_lte), + false); + + /* Since we just finished writing a stream, write any + * streams that have been added to the serial_streams + * list for direct writing by the main thread (e.g. + * resources that don't need to be compressed because + * the desired compression type is the same as the + * previous compression type). */ + if (!list_empty(&ctx->serial_streams)) { + ret = do_write_stream_list_serial(&ctx->serial_streams, + ctx->lookup_table, + ctx->out_fd, + ctx->out_ctype, + ctx->write_resource_flags, + ctx->progress_func, + ctx->progress); + if (ret) + return ret; + } - // Since we just finished writing a stream, - // write any streams that have been added to the - // my_resources list for direct writing by the - // main thread (e.g. resources that don't need - // to be compressed because the desired - // compression type is the same as the previous - // compression type). - struct lookup_table_entry *tmp; - list_for_each_entry_safe(lte, - tmp, - &my_resources, - staging_list) - { - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, - &next_size, - total_size, - one_percent, - &cur_percent, - lte); - } - - ret = write_wim_resource(lte, - out_fp, - out_ctype, - <e->output_resource_entry, - 0); - list_del(<e->staging_list); - if (ret != 0) - goto out; - } + /* Advance to the next stream to write. */ + if (list_empty(&ctx->outstanding_streams)) { + cur_lte = NULL; + } else { + cur_lte = container_of(ctx->outstanding_streams.next, + struct wim_lookup_table_entry, + being_compressed_list); } } } + return 0; +} -out: -#ifdef WITH_NTFS_3G - end_wim_resource_read(cur_lte, ni); -#else - end_wim_resource_read(cur_lte); -#endif - if (ret == 0) { - list_for_each_entry(lte, &my_resources, staging_list) { - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, - &next_size, - total_size, - one_percent, - &cur_percent, - lte); - } - ret = write_wim_resource(lte, out_fp, - out_ctype, - <e->output_resource_entry, - 0); - if (ret != 0) - break; +/* Called when the main thread has read a new chunk of data. */ +static int +main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx) +{ + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + struct message *next_msg; + u64 next_chunk_in_msg; + + /* Update SHA1 message digest for the stream currently being read by the + * main thread. */ + sha1_update(&ctx->next_sha_ctx, chunk, chunk_size); + + /* We send chunks of data to the compressor chunks in batches which we + * refer to as "messages". @next_msg is the message that is currently + * being prepared to send off. If it is NULL, that indicates that we + * need to start a new message. */ + next_msg = ctx->next_msg; + if (!next_msg) { + /* We need to start a new message. First check to see if there + * is a message available in the list of available messages. If + * so, we can just take one. If not, all the messages (there is + * a fixed number of them, proportional to the number of + * threads) have been sent off to the compressor threads, so we + * receive messages from the compressor threads containing + * compressed chunks of data. + * + * We may need to receive multiple messages before one is + * actually available to use because messages received that are + * *not* for the very next set of chunks to compress must be + * buffered until it's time to write those chunks. */ + while (list_empty(&ctx->available_msgs)) { + ret = receive_compressed_chunks(ctx); + if (ret) + return ret; } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - finish_stream_write_progress(total_size); - } else { - size_t num_available_msgs = 0; - struct list_head *cur; - list_for_each(cur, &available_msgs) { - num_available_msgs++; - } + next_msg = container_of(ctx->available_msgs.next, + struct message, list); + list_del(&next_msg->list); + next_msg->complete = false; + next_msg->begin_chunk = ctx->next_chunk; + next_msg->num_chunks = min(MAX_CHUNKS_PER_MSG, + ctx->next_num_chunks - ctx->next_chunk); + ctx->next_msg = next_msg; + } + + /* Fill in the next chunk to compress */ + next_chunk_in_msg = ctx->next_chunk - next_msg->begin_chunk; + + next_msg->uncompressed_chunk_sizes[next_chunk_in_msg] = chunk_size; + memcpy(next_msg->uncompressed_chunks[next_chunk_in_msg], + chunk, chunk_size); + ctx->next_chunk++; + if (++next_chunk_in_msg == next_msg->num_chunks) { + /* Send off an array of chunks to compress */ + list_add_tail(&next_msg->list, &ctx->next_lte->msg_list); + shared_queue_put(ctx->res_to_compress_queue, next_msg); + ++ctx->num_outstanding_messages; + ctx->next_msg = NULL; + } + return 0; +} - while (num_available_msgs < ARRAY_LEN(msgs)) { - shared_queue_get(compressed_res_queue); - num_available_msgs++; - } +static int +main_writer_thread_finish(void *_ctx) +{ + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + while (ctx->num_outstanding_messages != 0) { + ret = receive_compressed_chunks(ctx); + if (ret) + return ret; } + wimlib_assert(list_empty(&ctx->outstanding_streams)); + return do_write_stream_list_serial(&ctx->serial_streams, + ctx->lookup_table, + ctx->out_fd, + ctx->out_ctype, + ctx->write_resource_flags, + ctx->progress_func, + ctx->progress); +} - DEBUG("Freeing messages"); +static int +submit_stream_for_compression(struct wim_lookup_table_entry *lte, + struct main_writer_thread_ctx *ctx) +{ + int ret; - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) { - for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { - FREE(msgs[i].compressed_chunks[j]); - FREE(msgs[i].uncompressed_chunks[j]); - } + /* Read the entire stream @lte, feeding its data chunks to the + * compressor threads. Also SHA1-sum the stream; this is required in + * the case that @lte is unhashed, and a nice additional verification + * when @lte is already hashed. */ + sha1_init(&ctx->next_sha_ctx); + ctx->next_chunk = 0; + ctx->next_num_chunks = wim_resource_chunks(lte); + ctx->next_lte = lte; + INIT_LIST_HEAD(<e->msg_list); + 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); } - - if (cur_chunk_tab != NULL) - FREE(cur_chunk_tab); return ret; } - -static const char *get_data_type(int ctype) +static int +main_thread_process_next_stream(struct wim_lookup_table_entry *lte, void *_ctx) { - switch (ctype) { - case WIM_COMPRESSION_TYPE_NONE: - return "uncompressed"; - case WIM_COMPRESSION_TYPE_LZX: - return "LZX-compressed"; - case WIM_COMPRESSION_TYPE_XPRESS: - return "XPRESS-compressed"; + struct main_writer_thread_ctx *ctx = _ctx; + int ret; + + 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)) + { + /* Stream is too small or isn't being compressed. Process it by + * the main thread when we have a chance. We can't necessarily + * process it right here, as the main thread could be in the + * middle of writing a different stream. */ + list_add_tail(<e->write_streams_list, &ctx->serial_streams); + lte->deferred = 1; + ret = 0; + } else { + ret = submit_stream_for_compression(lte, ctx); } + lte->no_progress = 1; + return ret; +} + +static long +get_default_num_threads(void) +{ +#ifdef __WIN32__ + return win32_get_number_of_processors(); +#else + return sysconf(_SC_NPROCESSORS_ONLN); +#endif } -static int write_stream_list_parallel(struct list_head *stream_list, - FILE *out_fp, int out_ctype, - int write_flags, u64 total_size, - unsigned num_threads) +/* Equivalent to write_stream_list_serial(), except this takes a @num_threads + * parameter and will perform compression using that many threads. Falls + * back to write_stream_list_serial() on certain errors, such as a failure to + * create the number of threads requested. + * + * High level description of the algorithm for writing compressed streams in + * parallel: We perform compression on chunks of size WIM_CHUNK_SIZE bytes + * rather than on full files. The currently executing thread becomes the main + * thread and is entirely in charge of reading the data to compress (which may + * be in any location understood by the resource code--- such as in an external + * file being captured, or in another WIM file from which an image is being + * exported) and actually writing the compressed data to the output file. + * Additional threads are "compressor threads" and all execute the + * compressor_thread_proc, where they repeatedly retrieve buffers of data from + * the main thread, compress them, and hand them back to the main thread. + * + * Certain streams, such as streams that do not need to be compressed (e.g. + * input compression type same as output compression type) or streams of very + * small size are placed in a list (main_writer_thread_ctx.serial_list) and + * handled entirely by the main thread at an appropriate time. + * + * At any given point in time, multiple streams may be having chunks compressed + * concurrently. The stream that the main thread is currently *reading* may be + * later in the list that the stream that the main thread is currently + * *writing*. + */ +static int +write_stream_list_parallel(struct list_head *stream_list, + struct wim_lookup_table *lookup_table, + int out_fd, + int out_ctype, + int write_resource_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress, + unsigned num_threads) { int ret; struct shared_queue res_to_compress_queue; @@ -1199,28 +1275,33 @@ static int write_stream_list_parallel(struct list_head *stream_list, pthread_t *compressor_threads = NULL; if (num_threads == 0) { - long nthreads = sysconf(_SC_NPROCESSORS_ONLN); - if (nthreads < 1) { + long nthreads = get_default_num_threads(); + if (nthreads < 1 || nthreads > UINT_MAX) { WARNING("Could not determine number of processors! Assuming 1"); goto out_serial; + } else if (nthreads == 1) { + goto out_serial_quiet; } else { num_threads = nthreads; } } - wimlib_assert(stream_list->next != stream_list); + DEBUG("Writing stream list (parallel version, num_threads=%u)", + num_threads); - static const double MESSAGES_PER_THREAD = 2.0; + progress->write_streams.num_threads = num_threads; + + static const size_t MESSAGES_PER_THREAD = 2; size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret != 0) + if (ret) goto out_serial; ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret != 0) + if (ret) goto out_destroy_res_to_compress_queue; struct compressor_thread_params params; @@ -1229,45 +1310,70 @@ static int write_stream_list_parallel(struct list_head *stream_list, params.compress = get_compress_func(out_ctype); compressor_threads = MALLOC(num_threads * sizeof(pthread_t)); + if (!compressor_threads) { + ret = WIMLIB_ERR_NOMEM; + goto out_destroy_compressed_res_queue; + } for (unsigned i = 0; i < num_threads; i++) { - DEBUG("pthread_create thread %u", i); + DEBUG("pthread_create thread %u of %u", i + 1, num_threads); ret = pthread_create(&compressor_threads[i], NULL, compressor_thread_proc, ¶ms); if (ret != 0) { ret = -1; ERROR_WITH_ERRNO("Failed to create compressor " - "thread %u", i); + "thread %u of %u", + i + 1, num_threads); num_threads = i; goto out_join; } } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - printf("Writing %s data using %u threads...\n", - get_data_type(out_ctype), num_threads); - } - - ret = main_writer_thread_proc(stream_list, - out_fp, - out_ctype, - &res_to_compress_queue, - &compressed_res_queue, - queue_size, - write_flags, - total_size); - + if (progress_func) + 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.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; + 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); + if (ret) + goto out_destroy_ctx; + + /* The main thread has finished reading all streams that are going to be + * compressed in parallel, and it now needs to wait for all remaining + * chunks to be compressed so that the remaining streams can actually be + * written to the output file. Furthermore, any remaining streams that + * had processing deferred to the main thread need to be handled. These + * tasks are done by the main_writer_thread_finish() function. */ + ret = main_writer_thread_finish(&ctx); +out_destroy_ctx: + main_writer_thread_destroy_ctx(&ctx); out_join: for (unsigned i = 0; i < num_threads; i++) shared_queue_put(&res_to_compress_queue, NULL); for (unsigned i = 0; i < num_threads; i++) { if (pthread_join(compressor_threads[i], NULL)) { - WARNING("Failed to join compressor thread %u: %s", - i, strerror(errno)); + WARNING_WITH_ERRNO("Failed to join compressor " + "thread %u of %u", + i + 1, num_threads); } } FREE(compressor_threads); +out_destroy_compressed_res_queue: shared_queue_destroy(&compressed_res_queue); out_destroy_res_to_compress_queue: shared_queue_destroy(&res_to_compress_queue); @@ -1275,106 +1381,342 @@ out_destroy_res_to_compress_queue: return ret; out_serial: WARNING("Falling back to single-threaded compression"); - return write_stream_list_serial(stream_list, out_fp, - out_ctype, write_flags, total_size); +out_serial_quiet: + return write_stream_list_serial(stream_list, + lookup_table, + out_fd, + out_ctype, + write_resource_flags, + progress_func, + progress); + } #endif /* - * Write a list of streams to a WIM (@out_fp) using the compression type + * Write a list of streams to a WIM (@out_fd) using the compression type * @out_ctype and up to @num_threads compressor threads. */ -static int write_stream_list(struct list_head *stream_list, FILE *out_fp, - int out_ctype, int write_flags, - unsigned num_threads) +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, + unsigned num_threads, wimlib_progress_func_t progress_func) { - struct lookup_table_entry *lte; + struct wim_lookup_table_entry *lte; size_t num_streams = 0; - u64 total_size = 0; - bool compression_needed = false; + u64 total_bytes = 0; + u64 total_compression_bytes = 0; + union wimlib_progress_info progress; + int ret; + int write_resource_flags; - list_for_each_entry(lte, stream_list, staging_list) { - num_streams++; - total_size += wim_resource_size(lte); - if (!compression_needed - && - (out_ctype != WIM_COMPRESSION_TYPE_NONE - && (lte->resource_location != RESOURCE_IN_WIM - || wimlib_get_compression_type(lte->wim) != out_ctype - || (write_flags & WIMLIB_WRITE_FLAG_REBUILD))) - && wim_resource_size(lte) != 0) - compression_needed = true; - } - - if (num_streams == 0) { - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) - printf("No streams to write\n"); + if (list_empty(stream_list)) return 0; - } - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) { - printf("Preparing to write %zu streams " - "(%"PRIu64" total bytes uncompressed)\n", - num_streams, total_size); - printf("Using compression type %s\n", - wimlib_get_compression_type_string(out_ctype)); + write_resource_flags = write_flags_to_resource_flags(write_flags); + + /* 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 + * written (which will not necessarily be the case). */ + list_for_each_entry(lte, stream_list, write_streams_list) { + num_streams++; + 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))) + { + total_compression_bytes += wim_resource_size(lte); + } } + 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; #ifdef ENABLE_MULTITHREADED_COMPRESSION - if (compression_needed && total_size >= 1000000 && num_threads != 1) { - return write_stream_list_parallel(stream_list, out_fp, - out_ctype, write_flags, - total_size, num_threads); - } + if (total_compression_bytes >= 2000000 && num_threads != 1) + ret = write_stream_list_parallel(stream_list, + lookup_table, + out_fd, + out_ctype, + write_resource_flags, + progress_func, + &progress, + num_threads); else #endif - { - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - const char *reason = ""; - if (!compression_needed) - reason = " (no compression needed)"; - printf("Writing %s data using 1 thread%s\n", - get_data_type(out_ctype), reason); + ret = write_stream_list_serial(stream_list, + lookup_table, + out_fd, + out_ctype, + write_resource_flags, + progress_func, + &progress); + return ret; +} + +struct stream_size_table { + struct hlist_head *array; + size_t num_entries; + size_t capacity; +}; + +static int +init_stream_size_table(struct stream_size_table *tab, size_t capacity) +{ + tab->array = CALLOC(capacity, sizeof(tab->array[0])); + if (!tab->array) + return WIMLIB_ERR_NOMEM; + tab->num_entries = 0; + tab->capacity = capacity; + return 0; +} + +static void +destroy_stream_size_table(struct stream_size_table *tab) +{ + FREE(tab->array); +} + +static int +stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab) +{ + struct stream_size_table *tab = _tab; + size_t pos; + struct wim_lookup_table_entry *same_size_lte; + struct hlist_node *tmp; + + 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)) { + lte->unique_size = 0; + same_size_lte->unique_size = 0; + break; } + } + + hlist_add_head(<e->hash_list_2, &tab->array[pos]); + tab->num_entries++; + return 0; +} + + +struct lte_overwrite_prepare_args { + WIMStruct *wim; + off_t end_offset; + 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) +{ + 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; +} - return write_stream_list_serial(stream_list, out_fp, - out_ctype, write_flags, - total_size); +/* 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) +{ + 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); } + 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; } -static int dentry_find_streams_to_write(struct dentry *dentry, - void *wim) +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) { - WIMStruct *w = wim; - struct list_head *stream_list = w->private; - struct lookup_table_entry *lte; - for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) { - lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table); - if (lte && ++lte->out_refcnt == 1) - list_add_tail(<e->staging_list, stream_list); + struct wim_lookup_table_entry *lte; + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + lte = inode_stream_lte(inode, i, table); + if (lte) { + if (lte->out_refcnt == 0) { + if (lte->unhashed) + stream_size_table_insert(lte, tab); + list_add_tail(<e->write_streams_list, stream_list); + } + lte->out_refcnt += inode->i_nlink; + } + } +} + +static int +image_find_streams_to_write(WIMStruct *w) +{ + struct find_streams_ctx *ctx; + struct wim_image_metadata *imd; + struct wim_inode *inode; + struct wim_lookup_table_entry *lte; + + ctx = w->private; + imd = wim_get_current_image_metadata(w); + + image_for_each_unhashed_stream(lte, imd) + lte->out_refcnt = 0; + + /* 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); } return 0; } -static int find_streams_to_write(WIMStruct *w) +/* 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. + * + * 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_stream_list(WIMStruct *wim, int image, struct list_head *stream_list) { - return for_dentry_in_tree(wim_root_dentry(w), - dentry_find_streams_to_write, w); + int ret; + struct find_streams_ctx ctx; + + for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL); + ret = init_stream_size_table(&ctx.stream_size_tab, + wim->lookup_table->capacity); + if (ret) + return ret; + for_lookup_table_entry(wim->lookup_table, stream_size_table_insert, + &ctx.stream_size_tab); + INIT_LIST_HEAD(&ctx.stream_list); + wim->private = &ctx; + ret = for_image(wim, image, image_find_streams_to_write); + destroy_stream_size_table(&ctx.stream_size_tab); + if (ret == 0) + list_transfer(&ctx.stream_list, stream_list); + return ret; } -static int write_wim_streams(WIMStruct *w, int image, int write_flags, - unsigned num_threads) +/* 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) { + int ret; + struct list_head stream_list; - for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); - LIST_HEAD(stream_list); - w->private = &stream_list; - for_image(w, image, find_streams_to_write); - return write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), write_flags, - num_threads); + ret = prepare_stream_list(wim, image, &stream_list); + if (ret) + return ret; + return write_stream_list(&stream_list, + wim->lookup_table, + wim->out_fd, + wimlib_get_compression_type(wim), + write_flags, + num_threads, + progress_func); } /* @@ -1386,9 +1728,6 @@ static int write_wim_streams(WIMStruct *w, int image, int write_flags, * (public) WIMLIB_WRITE_FLAG_CHECK_INTEGRITY: * Include an integrity table. * - * (public) WIMLIB_WRITE_FLAG_SHOW_PROGRESS: - * Show progress information when (if) writing the integrity table. - * * (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE: * Don't write the lookup table. * @@ -1406,11 +1745,12 @@ static int write_wim_streams(WIMStruct *w, int image, int write_flags, * fsync() the output file before closing it. * */ -int finish_write(WIMStruct *w, int image, int write_flags) +int +finish_write(WIMStruct *w, int image, int write_flags, + wimlib_progress_func_t progress_func) { int ret; struct wim_header hdr; - FILE *out = w->out_fp; /* @hdr will be the header for the new WIM. First copy all the data * from the header in the WIMStruct; then set all the fields that may @@ -1418,47 +1758,53 @@ int finish_write(WIMStruct *w, int image, int write_flags) * count. */ memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); + /* 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; + } + + /* 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); + } else { + copy_resource_entry(&hdr.boot_metadata_res_entry, + &w->image_metadata[ hdr.boot_idx- 1 + ]->metadata_lte->output_resource_entry); + } + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); - if (ret != 0) - goto out; + ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry); + if (ret) + goto out_close_wim; } - ret = write_xml_data(w->wim_info, image, out, + 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); - if (ret != 0) - goto out; + if (ret) + goto out_close_wim; if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { struct wim_header checkpoint_hdr; memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); - memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); - if (fseeko(out, 0, SEEK_SET) != 0) { - ret = WIMLIB_ERR_WRITE; - goto out; - } - ret = write_header(&checkpoint_hdr, out); - if (ret != 0) - goto out; - - if (fflush(out) != 0) { - ERROR_WITH_ERRNO("Can't write data to WIM"); - ret = WIMLIB_ERR_WRITE; - goto out; - } - - if (fseeko(out, 0, SEEK_END) != 0) { - ret = WIMLIB_ERR_WRITE; - goto out; - } + zero_resource_entry(&checkpoint_hdr.integrity); + ret = write_header(&checkpoint_hdr, w->out_fd); + if (ret) + goto out_close_wim; } off_t old_lookup_table_end; off_t new_lookup_table_end; - bool show_progress; 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; @@ -1467,112 +1813,124 @@ int finish_write(WIMStruct *w, int image, int write_flags) } new_lookup_table_end = hdr.lookup_table_res_entry.offset + hdr.lookup_table_res_entry.size; - show_progress = ((write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) != 0); - ret = write_integrity_table(out, + ret = write_integrity_table(w->out_fd, &hdr.integrity, new_lookup_table_end, old_lookup_table_end, - show_progress); - if (ret != 0) - goto out; + progress_func); + if (ret) + goto out_close_wim; } else { - memset(&hdr.integrity, 0, sizeof(struct resource_entry)); + zero_resource_entry(&hdr.integrity); } - /* - * 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 || !w->image_metadata - || (image != WIM_ALL_IMAGES && image != hdr.boot_idx)) { - memset(&hdr.boot_metadata_res_entry, 0, - sizeof(struct resource_entry)); - } else { - memcpy(&hdr.boot_metadata_res_entry, - &w->image_metadata[ - hdr.boot_idx - 1].metadata_lte->output_resource_entry, - sizeof(struct resource_entry)); - } - - /* Set image count and boot index correctly for single image writes */ - if (image != WIM_ALL_IMAGES) { - hdr.image_count = 1; - if (hdr.boot_idx == image) - hdr.boot_idx = 1; - else - hdr.boot_idx = 0; - } - - if (fseeko(out, 0, SEEK_SET) != 0) { - ret = WIMLIB_ERR_WRITE; - goto out; - } - - ret = write_header(&hdr, out); - if (ret != 0) - goto out; + ret = write_header(&hdr, w->out_fd); + if (ret) + goto out_close_wim; if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { - if (fflush(out) != 0 - || fsync(fileno(out)) != 0) - { - ERROR_WITH_ERRNO("Error flushing data to WIM file"); + if (fsync(w->out_fd)) { + ERROR_WITH_ERRNO("Error syncing data to WIM file"); ret = WIMLIB_ERR_WRITE; } } -out: - if (fclose(out) != 0) { - ERROR_WITH_ERRNO("Failed to close the WIM file"); +out_close_wim: + if (close(w->out_fd)) { + ERROR_WITH_ERRNO("Failed to close the output WIM file"); if (ret == 0) ret = WIMLIB_ERR_WRITE; } - w->out_fp = NULL; + w->out_fd = -1; return ret; } -static void close_wim_writable(WIMStruct *w) +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +int +lock_wim(WIMStruct *w, int fd) { - if (w->out_fp) { - if (fclose(w->out_fp) != 0) { - WARNING("Failed to close output WIM: %s", - strerror(errno)); + int ret = 0; + if (fd != -1 && !w->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); + ret = WIMLIB_ERR_ALREADY_LOCKED; + } else { + WARNING_WITH_ERRNO("Failed to lock `%"TS"'", + w->filename); + ret = 0; + } + } else { + w->wim_locked = 1; } - w->out_fp = NULL; + } + return ret; +} +#endif + +static int +open_wim_writable(WIMStruct *w, const tchar *path, int open_flags) +{ + 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; +} + + +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; } } /* Open file stream and write dummy header for WIM. */ -int begin_write(WIMStruct *w, const char *path, int write_flags) +int +begin_write(WIMStruct *w, const tchar *path, int write_flags) { int ret; - bool need_readable = false; - bool trunc = true; + int open_flags = O_TRUNC | O_CREAT; if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - need_readable = true; - - ret = open_wim_writable(w, path, trunc, need_readable); - if (ret != 0) + open_flags |= O_RDWR; + 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. */ - return write_header(&w->hdr, w->out_fp); + ret = write_header(&w->hdr, w->out_fd); + 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; + } + return 0; } /* Writes a stand-alone WIM to a file. */ -WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, - int image, int write_flags, unsigned num_threads) +WIMLIBAPI int +wimlib_write(WIMStruct *w, const tchar *path, + int image, int write_flags, unsigned num_threads, + wimlib_progress_func_t progress_func) { int ret; - if (!w || !path) + if (!path) return WIMLIB_ERR_INVALID_PARAM; write_flags &= WIMLIB_WRITE_MASK_PUBLIC; - if (image != WIM_ALL_IMAGES && + if (image != WIMLIB_ALL_IMAGES && (image < 1 || image > w->hdr.image_count)) return WIMLIB_ERR_INVALID_IMAGE; @@ -1581,65 +1939,42 @@ WIMLIBAPI int wimlib_write(WIMStruct *w, const char *path, return WIMLIB_ERR_SPLIT_UNSUPPORTED; } - if (image == WIM_ALL_IMAGES) - DEBUG("Writing all images to `%s'.", path); - else - DEBUG("Writing image %d to `%s'.", image, path); - ret = begin_write(w, path, write_flags); - if (ret != 0) - goto out; + if (ret) + goto out_close_wim; - ret = write_wim_streams(w, image, write_flags, num_threads); - if (ret != 0) - goto out; + ret = write_wim_streams(w, image, write_flags, num_threads, + progress_func); + if (ret) + goto out_close_wim; - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - printf("Writing image metadata...\n"); + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL); ret = for_image(w, image, write_metadata_resource); - if (ret != 0) - goto out; + if (ret) + goto out_close_wim; - ret = finish_write(w, image, write_flags); - if (ret == 0 && (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS)) - printf("Successfully wrote `%s'\n", path); -out: + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + + 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; } -static int lte_overwrite_prepare(struct lookup_table_entry *lte, - void *ignore) -{ - memcpy(<e->output_resource_entry, <e->resource_entry, - sizeof(struct resource_entry)); - lte->out_refcnt = 0; - return 0; -} - -static int check_resource_offset(struct lookup_table_entry *lte, void *arg) -{ - off_t end_offset = *(u64*)arg; - - wimlib_assert(lte->out_refcnt <= lte->refcnt); - if (lte->out_refcnt < lte->refcnt) { - if (lte->resource_entry.offset + lte->resource_entry.size > end_offset) { - ERROR("The following resource is after the XML data:"); - print_lookup_table_entry(lte); - return WIMLIB_ERR_RESOURCE_ORDER; - } - } - return 0; -} - -static int find_new_streams(struct lookup_table_entry *lte, void *arg) +static bool +any_images_modified(WIMStruct *w) { - if (lte->out_refcnt == lte->refcnt) - list_add(<e->staging_list, (struct list_head*)arg); - else - lte->out_refcnt = lte->refcnt; - return 0; + for (int i = 0; i < w->hdr.image_count; i++) + if (w->image_metadata[i]->modified) + return true; + return false; } /* @@ -1697,178 +2032,186 @@ static int find_new_streams(struct lookup_table_entry *lte, void *arg) * is is crash-safe except in the case of write re-ordering, but the * disadvantage is that a small hole is left in the WIM where the old lookup * table, xml data, and integrity table were. (These usually only take up a - * small amount of space compared to the streams, however. + * small amount of space compared to the streams, however.) */ -static int overwrite_wim_inplace(WIMStruct *w, int write_flags, - unsigned num_threads, - int modified_image_idx) +static int +overwrite_wim_inplace(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { int ret; struct list_head stream_list; off_t old_wim_end; + u64 old_lookup_table_end, old_xml_begin, old_xml_end; + int open_flags; - DEBUG("Overwriting `%s' in-place", w->filename); + DEBUG("Overwriting `%"TS"' in-place", w->filename); /* Make sure that the integrity table (if present) is after the XML * data, and that there are no stream resources, metadata resources, or * lookup tables after the XML data. Otherwise, these data would be * overwritten. */ - if (w->hdr.integrity.offset != 0 && - w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { + 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"); return WIMLIB_ERR_RESOURCE_ORDER; } - if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) { + if (old_lookup_table_end > old_xml_begin) { ERROR("Didn't expect the lookup table to be after the XML data"); return WIMLIB_ERR_RESOURCE_ORDER; } - DEBUG("Identifying newly added streams"); - for_lookup_table_entry(w->lookup_table, lte_overwrite_prepare, NULL); - INIT_LIST_HEAD(&stream_list); - for (int i = modified_image_idx; i < w->hdr.image_count; i++) { - DEBUG("Identifiying streams in image %d", i + 1); - wimlib_assert(w->image_metadata[i].modified); - wimlib_assert(!w->image_metadata[i].has_been_mounted_rw); - wimlib_assert(w->image_metadata[i].root_dentry != NULL); - wimlib_assert(w->image_metadata[i].metadata_lte != NULL); - w->private = &stream_list; - for_dentry_in_tree(w->image_metadata[i].root_dentry, - dentry_find_streams_to_write, w); - } - - if (w->hdr.integrity.offset) - old_wim_end = w->hdr.integrity.offset + w->hdr.integrity.size; - else - old_wim_end = w->hdr.xml_res_entry.offset + w->hdr.xml_res_entry.size; - - ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, - &old_wim_end); - if (ret != 0) - return ret; - - if (modified_image_idx == w->hdr.image_count && !w->deletion_occurred) { + /* Set @old_wim_end, which indicates the point beyond which we don't + * allow any file and metadata resources to appear without returning + * WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise + * overwrite these resources). */ + if (!w->deletion_occurred && !any_images_modified(w)) { /* If no images have been modified and no images have been - * deleted, a new lookup table does not need to be written. */ - wimlib_assert(list_empty(&stream_list)); - old_wim_end = w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size; + * deleted, a new lookup table does not need to be written. We + * shall write the new XML data and optional integrity table + * immediately after the lookup table. Note that this may + * overwrite an existing integrity table. */ + DEBUG("Skipping writing lookup table " + "(no images modified or deleted)"); + old_wim_end = old_lookup_table_end; write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; + } else if (w->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; + } else { + /* No existing integrity table; begin writing new streams after + * the old XML data. */ + old_wim_end = old_xml_end; } - INIT_LIST_HEAD(&stream_list); - for_lookup_table_entry(w->lookup_table, find_new_streams, - &stream_list); + ret = prepare_streams_for_overwrite(w, old_wim_end, &stream_list); + if (ret) + return ret; - ret = open_wim_writable(w, w->filename, false, - (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); - if (ret != 0) + 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); + if (ret) return ret; - if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Can't seek to end of WIM"); - return WIMLIB_ERR_WRITE; + ret = lock_wim(w, w->out_fd); + if (ret) { + close_wim_writable(w); + return ret; } - if (!list_empty(&stream_list)) { - DEBUG("Writing newly added streams (offset = %"PRIu64")", - old_wim_end); - ret = write_stream_list(&stream_list, w->out_fp, - wimlib_get_compression_type(w), - write_flags, num_threads); - if (ret != 0) - goto out_ftruncate; - } else { - DEBUG("No new streams were added"); + if (lseek(w->out_fd, old_wim_end, SEEK_SET) == -1) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + close_wim_writable(w); + w->wim_locked = 0; + return WIMLIB_ERR_WRITE; } - for (int i = modified_image_idx; i < w->hdr.image_count; i++) { - select_wim_image(w, i + 1); - ret = write_metadata_resource(w); - if (ret != 0) - goto out_ftruncate; + 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), + 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; + } } write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; - ret = finish_write(w, WIM_ALL_IMAGES, write_flags); -out_ftruncate: + ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, + progress_func); +out_truncate: close_wim_writable(w); - if (ret != 0) { - WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)", w->filename, old_wim_end); - truncate(w->filename, old_wim_end); + /* Return value of truncate() is ignored because this is already + * an error path. */ + (void)ttruncate(w->filename, old_wim_end); } + w->wim_locked = 0; return ret; } -static int overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, - unsigned num_threads) +static int +overwrite_wim_via_tmpfile(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { size_t wim_name_len; int ret; - DEBUG("Overwrining `%s' via a temporary file", w->filename); + DEBUG("Overwriting `%"TS"' via a temporary file", w->filename); /* Write the WIM to a temporary file in the same directory as the * original WIM. */ - wim_name_len = strlen(w->filename); - char tmpfile[wim_name_len + 10]; - memcpy(tmpfile, w->filename, wim_name_len); + wim_name_len = tstrlen(w->filename); + tchar tmpfile[wim_name_len + 10]; + tmemcpy(tmpfile, w->filename, wim_name_len); randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; + tmpfile[wim_name_len + 9] = T('\0'); - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, + ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, write_flags | WIMLIB_WRITE_FLAG_FSYNC, - num_threads); - if (ret != 0) { - ERROR("Failed to write the WIM file `%s'", tmpfile); - goto err; + num_threads, progress_func); + if (ret) { + ERROR("Failed to write the WIM file `%"TS"'", tmpfile); + goto out_unlink; } - /* Close the original WIM file that was opened for reading. */ - if (w->fp != NULL) { - fclose(w->fp); - w->fp = NULL; - } - - DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); + close_wim(w); + DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename); /* Rename the new file to the old file .*/ - if (rename(tmpfile, w->filename) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + if (trename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'", tmpfile, w->filename); ret = WIMLIB_ERR_RENAME; - goto err; + goto out_unlink; } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - printf("Successfully renamed `%s' to `%s'\n", tmpfile, w->filename); - - /* Re-open the WIM read-only. */ - w->fp = fopen(w->filename, "rb"); - if (w->fp == NULL) { - ret = WIMLIB_ERR_REOPEN; - WARNING("Failed to re-open `%s' read-only: %s", - w->filename, strerror(errno)); + if (progress_func) { + union wimlib_progress_info progress; + progress.rename.from = tmpfile; + progress.rename.to = w->filename; + progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); } - return ret; -err: + goto out; +out_unlink: /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - WARNING("Failed to remove `%s': %s", tmpfile, strerror(errno)); + if (tunlink(tmpfile) != 0) + WARNING_WITH_ERRNO("Failed to remove `%"TS"'", tmpfile); +out: return ret; } /* * Writes a WIM file to the original file that it was read from, overwriting it. */ -WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags, - unsigned num_threads) +WIMLIBAPI int +wimlib_overwrite(WIMStruct *w, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { - if (!w) - return WIMLIB_ERR_INVALID_PARAM; - write_flags &= WIMLIB_WRITE_MASK_PUBLIC; if (!w->filename) @@ -1882,23 +2225,14 @@ WIMLIBAPI int wimlib_overwrite(WIMStruct *w, int write_flags, if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) { - int i, modified_image_idx; - for (i = 0; i < w->hdr.image_count && !w->image_metadata[i].modified; i++) - ; - modified_image_idx = i; - for (; i < w->hdr.image_count && w->image_metadata[i].modified && - !w->image_metadata[i].has_been_mounted_rw; i++) - ; - if (i == w->hdr.image_count) { - return overwrite_wim_inplace(w, write_flags, num_threads, - modified_image_idx); - } + int ret; + ret = overwrite_wim_inplace(w, write_flags, num_threads, + progress_func); + if (ret == WIMLIB_ERR_RESOURCE_ORDER) + WARNING("Falling back to re-building entire WIM"); + else + return ret; } - return overwrite_wim_via_tmpfile(w, write_flags, num_threads); -} - -/* Deprecated */ -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *wim, int write_flags) -{ - return wimlib_overwrite(wim, write_flags, 1); + return overwrite_wim_via_tmpfile(w, write_flags, num_threads, + progress_func); }