X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwrite.c;h=139ecdabc73dd74da20fc45cfaaa772dd8478e59;hp=a84c16b0c91e560ded0b13f6ea95b812899e9e70;hb=9fb3aaca115429b0af2a623bf20bfceef74f047f;hpb=213d64619675a59db71fed619bf47c1870b0e2e2 diff --git a/src/write.c b/src/write.c index a84c16b0..139ecdab 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. * @@ -25,294 +24,72 @@ * along with wimlib; if not, see http://www.gnu.org/licenses/. */ +#include "config.h" + +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +/* On BSD, this should be included before "list.h" so that "list.h" can + * overwrite the LIST_HEAD macro. */ +# include +#endif + +#ifdef __WIN32__ +# include +#endif + +#include "list.h" #include "wimlib_internal.h" -#include "io.h" +#include "buffer_io.h" #include "dentry.h" #include "lookup_table.h" #include "xml.h" #include "lzx.h" #include "xpress.h" + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +# include +#endif + #include -#include -#include #include #ifdef WITH_NTFS_3G -#include -#include -#include -#include +# include +# include +# include +# include #endif - #ifdef HAVE_ALLOCA_H -#include +# include +#else +# include #endif +#include -/* Reopens the FILE* for a WIM read-write. */ -static int reopen_rw(WIMStruct *w) -{ - FILE *fp; - - if (fclose(w->fp) != 0) - ERROR_WITH_ERRNO("Failed to close the file `%s'", w->filename); - w->fp = NULL; - fp = fopen(w->filename, "r+b"); - if (!fp) { - ERROR_WITH_ERRNO("Failed to open `%s' for reading and writing", - w->filename); - return WIMLIB_ERR_OPEN; - } - w->fp = fp; - return 0; -} - - +#if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE) +# define INVALID_HANDLE_VALUE ((HANDLE)(-1)) +#endif -/* - * 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) +static int +fflush_and_ftruncate(FILE *fp, off_t size) { - const char *wimfile_name; - size_t wim_name_len; int ret; - if (!w) - return WIMLIB_ERR_INVALID_PARAM; - - write_flags &= ~WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE; - - wimfile_name = w->filename; - - DEBUG("Replacing WIM file `%s'.", wimfile_name); - - if (!wimfile_name) - return WIMLIB_ERR_NO_FILENAME; - - /* Write the WIM to a temporary file. */ - /* XXX should the temporary file be somewhere else? */ - wim_name_len = strlen(wimfile_name); - char tmpfile[wim_name_len + 10]; - memcpy(tmpfile, wimfile_name, wim_name_len); - randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); - tmpfile[wim_name_len + 9] = '\0'; - - ret = wimlib_write(w, tmpfile, WIM_ALL_IMAGES, write_flags, - num_threads); + ret = fflush(fp); if (ret != 0) { - ERROR("Failed to write the WIM file `%s'", tmpfile); - if (unlink(tmpfile) != 0) - WARNING("Failed to remove `%s'", tmpfile); - return ret; - } - - DEBUG("Closing original WIM file."); - /* Close the original WIM file that was opened for reading. */ - if (w->fp) { - if (fclose(w->fp) != 0) { - WARNING("Failed to close the file `%s'", wimfile_name); - } - w->fp = NULL; - } - - DEBUG("Renaming `%s' to `%s'", tmpfile, wimfile_name); - - /* Rename the new file to the old file .*/ - if (rename(tmpfile, wimfile_name) != 0) { - ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", - tmpfile, wimfile_name); - /* Remove temporary file. */ - if (unlink(tmpfile) != 0) - ERROR_WITH_ERRNO("Failed to remove `%s'", tmpfile); - return WIMLIB_ERR_RENAME; - } - - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) - printf("Successfully renamed `%s' to `%s'\n", tmpfile, wimfile_name); - - return 0; -} - -static int check_resource_offset(struct lookup_table_entry *lte, void *arg) -{ - u64 xml_data_offset = *(u64*)arg; - if (lte->resource_entry.offset > xml_data_offset) { - ERROR("The following resource is *after* the XML data:"); - print_lookup_table_entry(lte); - return WIMLIB_ERR_RESOURCE_ORDER; - } - return 0; -} - -WIMLIBAPI int wimlib_overwrite_xml_and_header(WIMStruct *w, int write_flags) -{ - int ret; - FILE *fp; - u8 *integrity_table = NULL; - off_t xml_end; - off_t xml_size; - size_t bytes_written; - - DEBUG("Overwriting XML and header of `%s', write_flags = %#x", - w->filename, write_flags); - - if (!w->filename) - return WIMLIB_ERR_NO_FILENAME; - - write_flags &= ~WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE; - - /* 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 - * destroyed by this function. */ - if (w->hdr.integrity.offset != 0 && - w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the integrity table to be before the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } - - if (w->hdr.lookup_table_res_entry.offset > - w->hdr.xml_res_entry.offset) { - ERROR("Didn't expect the lookup table to be after the XML data"); - return WIMLIB_ERR_RESOURCE_ORDER; - } - - ret = for_lookup_table_entry(w->lookup_table, check_resource_offset, - &w->hdr.xml_res_entry.offset); - if (ret != 0) - return ret; - - ret = reopen_rw(w); - if (ret != 0) - return ret; - - fp = w->fp; - - /* The old integrity table is still OK, as the SHA1 message digests in - * the integrity table include neither the header nor the XML data. - * Save it for later if it exists and an integrity table was required. - * */ - if ((write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - && w->hdr.integrity.offset != 0) - { - DEBUG("Reading existing integrity table."); - integrity_table = MALLOC(w->hdr.integrity.size); - if (!integrity_table) - return WIMLIB_ERR_NOMEM; - - ret = read_uncompressed_resource(fp, w->hdr.integrity.offset, - w->hdr.integrity.original_size, - integrity_table); - if (ret != 0) - goto err; - DEBUG("Done reading existing integrity table."); - } - - DEBUG("Overwriting XML data."); - /* Overwrite the XML data. */ - if (fseeko(fp, w->hdr.xml_res_entry.offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" " - "for XML data", w->hdr.xml_res_entry.offset); - ret = WIMLIB_ERR_WRITE; - goto err; - } - ret = write_xml_data(w->wim_info, WIM_ALL_IMAGES, fp, 0); - if (ret != 0) - goto err; - - DEBUG("Updating XML resource entry."); - /* Update the XML resource entry in the WIM header. */ - xml_end = ftello(fp); - if (xml_end == -1) { - ret = WIMLIB_ERR_WRITE; - goto err; - } - xml_size = xml_end - w->hdr.xml_res_entry.offset; - w->hdr.xml_res_entry.size = xml_size; - w->hdr.xml_res_entry.original_size = xml_size; - /* XML data offset is unchanged. */ - - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - DEBUG("Writing integrity table."); - w->hdr.integrity.offset = xml_end; - if (integrity_table) { - /* The existing integrity table was saved. */ - bytes_written = fwrite(integrity_table, 1, - w->hdr.integrity.size, fp); - if (bytes_written != w->hdr.integrity.size) { - ERROR_WITH_ERRNO("Failed to write integrity " - "table"); - ret = WIMLIB_ERR_WRITE; - goto err; - } - FREE(integrity_table); - } else { - /* There was no existing integrity table, so a new one - * must be calculated. */ - ret = write_integrity_table(fp, WIM_HEADER_DISK_SIZE, - w->hdr.lookup_table_res_entry.offset + - w->hdr.lookup_table_res_entry.size, - write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); - if (ret != 0) - return ret; - - off_t end_integrity = ftello(fp); - if (end_integrity == -1) - return WIMLIB_ERR_WRITE; - - off_t integrity_size = end_integrity - xml_end; - w->hdr.integrity.size = integrity_size; - w->hdr.integrity.original_size = integrity_size; - w->hdr.integrity.flags = 0; - } - } else { - DEBUG("Truncating file to end of XML data."); - /* No integrity table to write. The file should be truncated - * because it's possible that the old file was longer (due to it - * including an integrity table, or due to its XML data being - * longer) */ - if (fflush(fp) != 0) { - ERROR_WITH_ERRNO("Failed to flush stream for file `%s'", - w->filename); - return WIMLIB_ERR_WRITE; - } - if (ftruncate(fileno(fp), xml_end) != 0) { - ERROR_WITH_ERRNO("Failed to truncate `%s' to %"PRIu64" " - "bytes", w->filename, xml_end); - return WIMLIB_ERR_WRITE; - } - memset(&w->hdr.integrity, 0, sizeof(struct resource_entry)); - } - - DEBUG("Overwriting header."); - /* Overwrite the header. */ - if (fseeko(fp, 0, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to beginning of `%s'", - w->filename); + ERROR_WITH_ERRNO("Failed to flush data to output WIM file"); return WIMLIB_ERR_WRITE; } - - ret = write_header(&w->hdr, fp); - if (ret != 0) - return ret; - - DEBUG("Closing `%s'.", w->filename); - if (fclose(fp) != 0) { - ERROR_WITH_ERRNO("Failed to close `%s'", w->filename); + 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; } - w->fp = NULL; - DEBUG("Done."); return 0; -err: - FREE(integrity_table); - return ret; } - /* 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.) */ @@ -332,7 +109,7 @@ struct chunk_table { * output file. */ static int -begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte, +begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte, FILE *out_fp, off_t file_offset, struct chunk_table **chunk_tab_ret) @@ -372,19 +149,8 @@ out: return ret; } -typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); - -compress_func_t get_compress_func(int out_ctype) -{ - if (out_ctype == WIM_COMPRESSION_TYPE_LZX) - return lzx_compress; - else - return xpress_compress; -} - - /* - * Compresses a chunk of a WIM resource. + * Pointer to function to compresses a chunk of a WIM resource. * * @chunk: Uncompressed data of the chunk. * @chunk_size: Size of the uncompressed chunk in bytes. @@ -393,21 +159,20 @@ compress_func_t get_compress_func(int out_ctype) * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size * of the compressed chunk will be * returned. - * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX - * or WIM_COMPRESSION_TYPE_XPRESS. * * 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. */ -static int compress_chunk(const u8 chunk[], unsigned chunk_size, - u8 compressed_chunk[], - unsigned *compressed_chunk_len_ret, - int ctype) +typedef int (*compress_func_t)(const void *, unsigned, void *, unsigned *); + +compress_func_t +get_compress_func(int out_ctype) { - compress_func_t compress = get_compress_func(ctype); - return (*compress)(chunk, chunk_size, compressed_chunk, - compressed_chunk_len_ret); + if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) + return lzx_compress; + else + return xpress_compress; } /* @@ -423,24 +188,19 @@ static int compress_chunk(const u8 chunk[], unsigned chunk_size, * * Returns 0 on success; nonzero on failure. */ -static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, - FILE *out_fp, int out_ctype, - struct chunk_table *chunk_tab) +static int +write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, + FILE *out_fp, compress_func_t compress, + struct chunk_table *chunk_tab) { const u8 *out_chunk; unsigned out_chunk_size; - - wimlib_assert(chunk_size <= WIM_CHUNK_SIZE); - - if (!chunk_tab) { - out_chunk = chunk; - out_chunk_size = chunk_size; - } else { + if (chunk_tab) { u8 *compressed_chunk = alloca(chunk_size); int ret; - ret = compress_chunk(chunk, chunk_size, compressed_chunk, - &out_chunk_size, out_ctype); + ret = compress(chunk, chunk_size, compressed_chunk, + &out_chunk_size); if (ret == 0) { out_chunk = compressed_chunk; } else { @@ -449,8 +209,10 @@ static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size, } *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset; chunk_tab->cur_offset += out_chunk_size; + } else { + out_chunk = chunk; + out_chunk_size = chunk_size; } - if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) { ERROR_WITH_ERRNO("Failed to write WIM resource chunk"); return WIMLIB_ERR_WRITE; @@ -459,7 +221,7 @@ 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 @@ -499,61 +261,84 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab, return 0; } -static int prepare_resource_for_read(struct lookup_table_entry *lte +/* 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 wim_lookup_table_entry *lte - #ifdef WITH_NTFS_3G - , ntfs_inode **ni_ret - #endif - ) + #ifdef WITH_NTFS_3G + , ntfs_inode **ni_ret + #endif + ) { - if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && !lte->file_on_disk_fp) - { - wimlib_assert(lte->file_on_disk); - lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb"); + switch (lte->resource_location) { + case RESOURCE_IN_FILE_ON_DISK: 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; + 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'", lte->file_on_disk); + return WIMLIB_ERR_OPEN; + } } - } + break; #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); + case RESOURCE_IN_NTFS_VOLUME: 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; + 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); + if (!ni) { + ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS " + "volume", loc->path); + return WIMLIB_ERR_NTFS_3G; + } + lte->attr = ntfs_attr_open(ni, + loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, + loc->stream_name, + loc->stream_name_nchars); + if (!lte->attr) { + ERROR_WITH_ERRNO("Failed to open attribute of `%s' in " + "NTFS volume", loc->path); + ntfs_inode_close(ni); + return WIMLIB_ERR_NTFS_3G; + } + *ni_ret = ni; } - *ni_ret = ni; - } + break; +#endif +#ifdef __WIN32__ + case RESOURCE_WIN32: + if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) { + lte->win32_file_on_disk_fp = + win32_open_file_readonly(lte->win32_file_on_disk, true); + if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) { + ERROR("Win32 API: Can't open %ls", lte->win32_file_on_disk); + win32_error_last(); + return WIMLIB_ERR_OPEN; + } + } + break; #endif + default: + break; + } return 0; } -static void end_wim_resource_read(struct lookup_table_entry *lte - #ifdef WITH_NTFS_3G - , ntfs_inode *ni - #endif - ) +/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS + * attribute. */ +static void +end_wim_resource_read(struct wim_lookup_table_entry *lte + #ifdef WITH_NTFS_3G + , ntfs_inode *ni + #endif + ) { if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK - && lte->file_on_disk_fp) { + && lte->file_on_disk_fp) + { fclose(lte->file_on_disk_fp); lte->file_on_disk_fp = NULL; } @@ -567,6 +352,35 @@ static void end_wim_resource_read(struct lookup_table_entry *lte ntfs_inode_close(ni); } #endif +#ifdef __WIN32__ + else if (lte->resource_location == RESOURCE_WIN32 + && lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE) + { + win32_close_file(lte->win32_file_on_disk_fp); + lte->win32_file_on_disk_fp = INVALID_HANDLE_VALUE; + } +#endif +} + +static int +write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte, + FILE *out_fp, + off_t file_offset, + struct resource_entry *out_res_entry) +{ + int ret; + if (fseeko(out_fp, file_offset, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of " + "output WIM file", file_offset); + return WIMLIB_ERR_WRITE; + } + ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE, + out_res_entry, 0); + if (ret != 0) + return ret; + + return fflush_and_ftruncate(out_fp, + file_offset + wim_resource_size(lte)); } /* @@ -590,10 +404,11 @@ static void end_wim_resource_read(struct lookup_table_entry *lte * * Returns 0 on success; nonzero on failure. */ -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, + FILE *out_fp, int out_ctype, + struct resource_entry *out_res_entry, + int flags) { u64 bytes_remaining; u64 original_size; @@ -604,6 +419,7 @@ int write_wim_resource(struct lookup_table_entry *lte, struct chunk_table *chunk_tab = NULL; bool raw; off_t file_offset; + compress_func_t compress = NULL; #ifdef WITH_NTFS_3G ntfs_inode *ni = NULL; #endif @@ -627,7 +443,8 @@ int write_wim_resource(struct lookup_table_entry *lte, /* 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); + && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE + && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS)); if (raw) { flags |= WIMLIB_RESOURCE_FLAG_RAW; @@ -646,7 +463,7 @@ int write_wim_resource(struct lookup_table_entry *lte, /* 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) { + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) { ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset, &chunk_tab); if (ret != 0) @@ -669,13 +486,15 @@ int write_wim_resource(struct lookup_table_entry *lte, * hash given in the lookup table entry once we've finished reading the * resource. */ SHA_CTX ctx; - if (!raw) + 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. */ - offset = 0; do { u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); ret = read_wim_resource(lte, buf, to_read, offset, flags); @@ -684,7 +503,7 @@ int write_wim_resource(struct lookup_table_entry *lte, if (!raw) sha1_update(&ctx, buf, to_read); ret = write_wim_resource_chunk(buf, to_read, out_fp, - out_ctype, chunk_tab); + compress, chunk_tab); if (ret != 0) goto out_fclose; bytes_remaining -= to_read; @@ -694,7 +513,7 @@ int write_wim_resource(struct lookup_table_entry *lte, /* 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 + * Using WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the * original size * * Using a different compression type: Call @@ -704,7 +523,7 @@ int write_wim_resource(struct lookup_table_entry *lte, if (raw) { new_compressed_size = old_compressed_size; } else { - if (out_ctype == WIM_COMPRESSION_TYPE_NONE) + if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) new_compressed_size = original_size; else { ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp, @@ -736,30 +555,16 @@ int write_wim_resource(struct lookup_table_entry *lte, } if (!raw && new_compressed_size >= original_size && - out_ctype != WIM_COMPRESSION_TYPE_NONE) + out_ctype != WIMLIB_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); + ret = write_uncompressed_resource_and_truncate(lte, + out_fp, + file_offset, + out_res_entry); if (ret != 0) goto out_fclose; - if (fflush(out_fp) != 0) { - ERROR_WITH_ERRNO("Failed to flush output WIM file"); - ret = WIMLIB_ERR_WRITE; - goto out_fclose; - } - if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) { - ERROR_WITH_ERRNO("Failed to truncate output WIM file"); - ret = WIMLIB_ERR_WRITE; - goto out_fclose; - } } else { if (out_res_entry) { out_res_entry->size = new_compressed_size; @@ -767,7 +572,7 @@ int write_wim_resource(struct lookup_table_entry *lte, 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) + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; } } @@ -783,73 +588,80 @@ out: 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); + q->filled_slots = 0; q->front = 0; q->back = size - 1; q->size = size; + pthread_mutex_init(&q->lock, NULL); + pthread_cond_init(&q->msg_avail_cond, NULL); + pthread_cond_init(&q->space_avail_cond, NULL); return 0; } -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; } -static inline int shared_queue_get_filled(struct shared_queue *q) -{ - int sval; - sem_getvalue(&q->filled_slots, &sval); - return sval; -} - struct compressor_thread_params { struct shared_queue *res_to_compress_queue; struct shared_queue *compressed_res_queue; @@ -859,7 +671,7 @@ 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]; @@ -871,7 +683,8 @@ struct message { 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) { for (unsigned i = 0; i < msg->num_chunks; i++) { DEBUG2("compress chunk %u of %u", i, msg->num_chunks); @@ -888,7 +701,13 @@ static void compress_chunks(struct message *msg, compress_func_t compress) } } -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; @@ -902,63 +721,67 @@ static void *compressor_thread_proc(void *arg) shared_queue_put(compressed_res_queue, msg); } DEBUG("Compressor thread terminating"); + return NULL; } +#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) -{ - 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); -} - -static void finish_stream_write_progress(u64 total_size) -{ - printf("\r%"PRIu64" MiB of %"PRIu64" MiB " - "(uncompressed) written (100%% done)\n", - total_size >> 20, total_size >> 20); - fflush(stdout); -} - -static int write_stream_list_serial(struct list_head *stream_list, - FILE *out_fp, int out_ctype, - int write_flags, u64 total_size) +static int +do_write_stream_list(struct list_head *my_resources, + FILE *out_fp, + int out_ctype, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress, + int write_resource_flags) { - 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; - - 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); - } - ret = write_wim_resource(lte, out_fp, out_ctype, - <e->output_resource_entry, 0); + struct wim_lookup_table_entry *lte, *tmp; + + list_for_each_entry_safe(lte, tmp, my_resources, staging_list) { + ret = write_wim_resource(lte, + out_fp, + out_ctype, + <e->output_resource_entry, + write_resource_flags); if (ret != 0) return ret; + list_del(<e->staging_list); + progress->write_streams.completed_bytes += + wim_resource_size(lte); + progress->write_streams.completed_streams++; + if (progress_func) { + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - finish_stream_write_progress(total_size); return 0; } -static int write_wim_chunks(struct message *msg, FILE *out_fp, - struct chunk_table *chunk_tab) +static int +write_stream_list_serial(struct list_head *stream_list, + FILE *out_fp, + int out_ctype, + int write_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) +{ + int write_resource_flags; + + if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS; + else + write_resource_flags = 0; + progress->write_streams.num_threads = 1; + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + return do_write_stream_list(stream_list, out_fp, + out_ctype, progress_func, + progress, write_resource_flags); +} + +#ifdef ENABLE_MULTITHREADED_COMPRESSION +static int +write_wim_chunks(struct message *msg, FILE *out_fp, + struct chunk_table *chunk_tab) { for (unsigned i = 0; i < msg->num_chunks; i++) { unsigned chunk_csize = msg->compressed_chunk_sizes[i]; @@ -969,7 +792,7 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp) != chunk_csize) { - ERROR_WITH_ERRNO("Failed to write WIM"); + ERROR_WITH_ERRNO("Failed to write WIM chunk"); return WIMLIB_ERR_WRITE; } @@ -979,7 +802,6 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, 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 @@ -992,24 +814,31 @@ static int write_wim_chunks(struct message *msg, FILE *out_fp, * 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) +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 num_messages, + int write_flags, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { int ret; - - - struct message msgs[queue_size]; - ZERO_ARRAY(msgs); + struct chunk_table *cur_chunk_tab = NULL; + struct message *msgs = CALLOC(num_messages, sizeof(struct message)); + struct wim_lookup_table_entry *next_lte = NULL; // Initially, all the messages are available to use. LIST_HEAD(available_msgs); - for (size_t i = 0; i < ARRAY_LEN(msgs); i++) + + if (!msgs) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } + + for (size_t i = 0; i < num_messages; i++) list_add(&msgs[i].list, &available_msgs); // outstanding_resources is the list of resources that currently have @@ -1029,74 +858,149 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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); + u64 next_num_chunks = 0; // 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]; // Resources that don't need any chunks compressed are added to this // list and written directly by the main thread. LIST_HEAD(my_resources); - struct lookup_table_entry *cur_lte = next_lte; - struct chunk_table *cur_chunk_tab = NULL; - struct lookup_table_entry *lte; + struct wim_lookup_table_entry *cur_lte = NULL; struct message *msg; - u64 one_percent = total_size / 100; - u64 cur_size = 0; - u64 next_size = 0; - unsigned cur_percent = 0; - #ifdef WITH_NTFS_3G ntfs_inode *ni = NULL; #endif -#ifdef WITH_NTFS_3G - ret = prepare_resource_for_read(next_lte, &ni); -#else - ret = prepare_resource_for_read(next_lte); -#endif - DEBUG("Initializing buffers for uncompressed " "and compressed data (%zu bytes needed)", - queue_size * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2); + num_messages * 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 i = 0; i < num_messages; 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); + + // The extra 8 bytes is because longest_match() in + // lz77.c may read a little bit off the end of the + // uncompressed data. It doesn't need to be + // initialized--- we really just need to avoid accessing + // an unmapped page. + msgs[i].uncompressed_chunks[j] = MALLOC(WIM_CHUNK_SIZE + 8); if (msgs[i].compressed_chunks[j] == NULL || msgs[i].uncompressed_chunks[j] == NULL) { - ERROR("Could not allocate enough memory for " - "multi-threaded compression"); ret = WIMLIB_ERR_NOMEM; goto out; } } } + // 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) { + while (!list_empty(&available_msgs)) { + if (next_chunk == next_num_chunks) { + // If next_chunk == next_num_chunks, there are + // no more chunks to write in the current + // stream. So, check the SHA1 message digest of + // the stream that was just finished (unless + // next_lte == NULL, which is the case the very + // first time this loop is entered, and also + // near the very end of the compression when + // there are no more streams.) Then, advance to + // the next stream (if there is one). + if (next_lte != NULL) { + #ifdef WITH_NTFS_3G + end_wim_resource_read(next_lte, ni); + ni = NULL; + #else + end_wim_resource_read(next_lte); + #endif + DEBUG2("Finalize SHA1 md (next_num_chunks=%zu)", + next_num_chunks); + sha1_final(next_hash, &next_sha_ctx); + if (!hashes_equal(next_lte->hash, next_hash)) { + ERROR("WIM resource has incorrect hash!"); + if (next_lte->resource_location == + RESOURCE_IN_FILE_ON_DISK) + { + ERROR("We were reading it from `%s'; " + "maybe it changed while we were " + "reading it.", + next_lte->file_on_disk); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + goto out; + } + } - // Get a message from the available messages - // list + // 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) { + // No more resources to send for + // compression + next_lte = NULL; + break; + } + next_lte = container_of(next_resource, + struct wim_lookup_table_entry, + staging_list); + next_resource = next_resource->next; + if ((!(write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) + && wim_resource_compression_type(next_lte) == out_ctype) + || wim_resource_size(next_lte) == 0) + { + list_add_tail(&next_lte->staging_list, + &my_resources); + } else { + list_add_tail(&next_lte->staging_list, + &outstanding_resources); + next_chunk = 0; + next_num_chunks = wim_resource_chunks(next_lte); + sha1_init(&next_sha_ctx); + INIT_LIST_HEAD(&next_lte->msg_list); + #ifdef WITH_NTFS_3G + ret = prepare_resource_for_read(next_lte, &ni); + #else + ret = prepare_resource_for_read(next_lte); + #endif + + if (ret != 0) + goto out; + if (cur_lte == NULL) { + // Set cur_lte for the + // first time + cur_lte = next_lte; + } + break; + } + } + } + + if (next_lte == NULL) { + // No more resources to send for compression + break; + } + + // Get a message from the available messages + // list msg = container_of(available_msgs.next, struct message, list); @@ -1120,13 +1024,11 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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; + if (next_chunk == next_num_chunks - 1) { + size = MODULO_NONZERO(wim_resource_size(next_lte), + WIM_CHUNK_SIZE); } - DEBUG2("Read resource (size=%u, offset=%zu)", size, next_chunk * WIM_CHUNK_SIZE); @@ -1148,80 +1050,11 @@ static int main_writer_thread_proc(struct list_head *stream_list, 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; - } - - // 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 ((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; - } - } } // 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; } @@ -1229,42 +1062,25 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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; - DEBUG2("Received msg (begin_chunk=%"PRIu64")", msg->begin_chunk); - - list_for_each_entry(msg, &cur_lte->msg_list, list) { - DEBUG2("complete=%d", msg->complete); - } - // 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) + while (cur_lte != NULL && + !list_empty(&cur_lte->msg_list) && + (msg = container_of(cur_lte->msg_list.next, + struct message, + list))->complete) { DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk); if (msg->begin_chunk == 0) { DEBUG2("Begin chunk tab"); - - - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, - &next_size, - total_size, - one_percent, - &cur_percent, - cur_lte); - } - // This is the first set of chunks. Leave space // for the chunk table in the output file. off_t cur_offset = ftello(out_fp); @@ -1291,8 +1107,8 @@ static int main_writer_thread_proc(struct list_head *stream_list, // now. list_add(&msg->list, &available_msgs); - // Was this the last chunk of the stream? If so, - // finish it. + // 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) { @@ -1304,19 +1120,40 @@ static int main_writer_thread_proc(struct list_head *stream_list, if (ret != 0) goto out; + if (res_csize >= wim_resource_size(cur_lte)) { + /* Oops! We compressed the resource to + * larger than the original size. Write + * the resource uncompressed instead. */ + ret = write_uncompressed_resource_and_truncate( + cur_lte, + out_fp, + cur_chunk_tab->file_offset, + &cur_lte->output_resource_entry); + if (ret != 0) + goto out; + } else { + cur_lte->output_resource_entry.size = + res_csize; - cur_lte->output_resource_entry.size = - res_csize; + cur_lte->output_resource_entry.original_size = + cur_lte->resource_entry.original_size; - cur_lte->output_resource_entry.original_size = - cur_lte->resource_entry.original_size; + cur_lte->output_resource_entry.offset = + cur_chunk_tab->file_offset; - cur_lte->output_resource_entry.offset = - cur_chunk_tab->file_offset; + cur_lte->output_resource_entry.flags = + cur_lte->resource_entry.flags | + WIM_RESHDR_FLAG_COMPRESSED; + } - cur_lte->output_resource_entry.flags = - cur_lte->resource_entry.flags | - WIM_RESHDR_FLAG_COMPRESSED; + progress->write_streams.completed_bytes += + wim_resource_size(cur_lte); + progress->write_streams.completed_streams++; + + if (progress_func) { + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, + progress); + } FREE(cur_chunk_tab); cur_chunk_tab = NULL; @@ -1324,15 +1161,12 @@ static int main_writer_thread_proc(struct list_head *stream_list, 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 { + if (next == &outstanding_resources) + cur_lte = NULL; + else cur_lte = container_of(cur_lte->staging_list.next, - struct lookup_table_entry, + struct wim_lookup_table_entry, staging_list); - } // Since we just finished writing a stream, // write any streams that have been added to the @@ -1341,99 +1175,93 @@ static int main_writer_thread_proc(struct list_head *stream_list, // 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; - } + ret = do_write_stream_list(&my_resources, + out_fp, + out_ctype, + progress_func, + progress, + 0); + if (ret != 0) + goto out; } } } out: + if (ret == WIMLIB_ERR_NOMEM) { + ERROR("Could not allocate enough memory for " + "multi-threaded compression"); + } + + if (next_lte) { #ifdef WITH_NTFS_3G - end_wim_resource_read(cur_lte, ni); + end_wim_resource_read(next_lte, ni); #else - end_wim_resource_read(cur_lte); + end_wim_resource_read(next_lte); #endif + } + if (ret == 0) { - list_for_each_entry(lte, &my_resources, staging_list) { - ret = write_wim_resource(lte, out_fp, - out_ctype, - <e->output_resource_entry, - 0); - if (ret != 0) - break; - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - show_stream_write_progress(&cur_size, - &next_size, - total_size, - one_percent, - &cur_percent, - lte); - } - } - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) - finish_stream_write_progress(total_size); + ret = do_write_stream_list(&my_resources, out_fp, + out_ctype, progress_func, + progress, 0); } else { - size_t num_available_msgs = 0; - struct list_head *cur; + if (msgs) { + size_t num_available_msgs = 0; + struct list_head *cur; - list_for_each(cur, &available_msgs) { - num_available_msgs++; - } + list_for_each(cur, &available_msgs) { + num_available_msgs++; + } - while (num_available_msgs < ARRAY_LEN(msgs)) { - shared_queue_get(compressed_res_queue); - num_available_msgs++; + while (num_available_msgs < num_messages) { + shared_queue_get(compressed_res_queue); + num_available_msgs++; + } } } - DEBUG("Freeing messages"); - - 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]); + if (msgs) { + for (size_t i = 0; i < num_messages; i++) { + for (size_t j = 0; j < MAX_CHUNKS_PER_MSG; j++) { + FREE(msgs[i].compressed_chunks[j]); + FREE(msgs[i].uncompressed_chunks[j]); + } } + FREE(msgs); } - if (cur_chunk_tab != NULL) - FREE(cur_chunk_tab); + FREE(cur_chunk_tab); return ret; } -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) +static long +get_default_num_threads() +{ +#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, + unsigned num_threads, + wimlib_progress_func_t progress_func, + union wimlib_progress_info *progress) { int ret; struct shared_queue res_to_compress_queue; struct shared_queue compressed_res_queue; + pthread_t *compressor_threads = NULL; if (num_threads == 0) { - long nthreads = sysconf(_SC_NPROCESSORS_ONLN); - if (nthreads < 1) { + 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 { @@ -1441,66 +1269,70 @@ static int write_stream_list_parallel(struct list_head *stream_list, } } + progress->write_streams.num_threads = num_threads; wimlib_assert(stream_list->next != stream_list); - { - pthread_t compressor_threads[num_threads]; + static const double MESSAGES_PER_THREAD = 2.0; + size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); - static const double MESSAGES_PER_THREAD = 2.0; - size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD); + DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); - DEBUG("Initializing shared queues (queue_size=%zu)", queue_size); + ret = shared_queue_init(&res_to_compress_queue, queue_size); + if (ret != 0) + goto out_serial; - ret = shared_queue_init(&res_to_compress_queue, queue_size); - if (ret != 0) - goto out_serial; + ret = shared_queue_init(&compressed_res_queue, queue_size); + if (ret != 0) + goto out_destroy_res_to_compress_queue; - ret = shared_queue_init(&compressed_res_queue, queue_size); - if (ret != 0) - goto out_destroy_res_to_compress_queue; - - struct compressor_thread_params params; - params.res_to_compress_queue = &res_to_compress_queue; - params.compressed_res_queue = &compressed_res_queue; - params.compress = get_compress_func(out_ctype); - - for (unsigned i = 0; i < num_threads; i++) { - DEBUG("pthread_create thread %u", i); - ret = pthread_create(&compressor_threads[i], NULL, - compressor_thread_proc, ¶ms); - if (ret != 0) { - ERROR_WITH_ERRNO("Failed to create compressor " - "thread %u", i); - num_threads = i; - goto out_join; - } - } + struct compressor_thread_params params; + params.res_to_compress_queue = &res_to_compress_queue; + params.compressed_res_queue = &compressed_res_queue; + params.compress = get_compress_func(out_ctype); - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - printf("Writing compressed data using %u threads...\n", - num_threads); + 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); + ret = pthread_create(&compressor_threads[i], NULL, + compressor_thread_proc, ¶ms); + if (ret != 0) { + ret = -1; + ERROR_WITH_ERRNO("Failed to create compressor " + "thread %u", i); + num_threads = i; + goto out_join; } + } - ret = main_writer_thread_proc(stream_list, - out_fp, - out_ctype, - &res_to_compress_queue, - &compressed_res_queue, - queue_size, - write_flags, - total_size); - - 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)); - } + if (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress); + + ret = main_writer_thread_proc(stream_list, + out_fp, + out_ctype, + &res_to_compress_queue, + &compressed_res_queue, + queue_size, + write_flags, + progress_func, + progress); +out_join: + for (unsigned i = 0; i < num_threads; i++) + shared_queue_put(&res_to_compress_queue, NULL); + + for (unsigned i = 0; i < num_threads; i++) { + if (pthread_join(compressor_threads[i], NULL)) { + WARNING_WITH_ERRNO("Failed to join compressor " + "thread %u", i); } } + 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); @@ -1508,175 +1340,268 @@ 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); + return write_stream_list_serial(stream_list, + out_fp, + out_ctype, + write_flags, + progress_func, + progress); + } +#endif -static int write_stream_list(struct list_head *stream_list, FILE *out_fp, - int out_ctype, int write_flags, - unsigned num_threads) +/* + * Write a list of streams to a WIM (@out_fp) using the compression type + * @out_ctype and up to @num_threads compressor threads. + */ +static int +write_stream_list(struct list_head *stream_list, FILE *out_fp, + int out_ctype, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { - struct lookup_table_entry *lte; + 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; 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) - && wim_resource_size(lte) != 0) - compression_needed = true; - } - - 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)); - } - - 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); - } else { - if (write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS) { - const char *reason = ""; - if (num_threads != 1) - reason = " (no compression needed)"; - printf("Writing data using 1 thread%s\n", reason); + total_bytes += wim_resource_size(lte); + if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE + && (wim_resource_compression_type(lte) != out_ctype || + (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS))) + { + total_compression_bytes += wim_resource_size(lte); } + } + 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; + +#ifdef ENABLE_MULTITHREADED_COMPRESSION + if (total_compression_bytes >= 1000000 && num_threads != 1) + return write_stream_list_parallel(stream_list, + out_fp, + out_ctype, + write_flags, + num_threads, + progress_func, + &progress); + else +#endif + return write_stream_list_serial(stream_list, + out_fp, + out_ctype, + write_flags, + progress_func, + &progress); +} + +struct lte_overwrite_prepare_args { + WIMStruct *wim; + off_t end_offset; + struct list_head *stream_list; +}; + +static int +lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *arg) +{ + struct lte_overwrite_prepare_args *args = arg; + + if (lte->resource_location == RESOURCE_IN_WIM && + lte->wim == args->wim && + 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; + } - return write_stream_list_serial(stream_list, out_fp, - out_ctype, write_flags, - total_size); + lte->out_refcnt = lte->refcnt; + memcpy(<e->output_resource_entry, <e->resource_entry, + sizeof(struct resource_entry)); + if (!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)) { + wimlib_assert(lte->resource_location != RESOURCE_NONEXISTENT); + if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim) + list_add(<e->staging_list, args->stream_list); } + return 0; } +static int +wim_find_new_streams(WIMStruct *wim, off_t end_offset, + struct list_head *stream_list) +{ + struct lte_overwrite_prepare_args args = { + .wim = wim, + .end_offset = end_offset, + .stream_list = stream_list, + }; + + return for_lookup_table_entry(wim->lookup_table, + lte_overwrite_prepare, &args); +} -static int dentry_find_streams_to_write(struct dentry *dentry, - void *wim) +static int +inode_find_streams_to_write(struct wim_inode *inode, + struct wim_lookup_table *table, + struct list_head *stream_list) { - 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(<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) + list_add_tail(<e->staging_list, stream_list); + lte->out_refcnt += inode->i_nlink; + } } return 0; } -static int find_streams_to_write(WIMStruct *w) +static int +image_find_streams_to_write(WIMStruct *w) { - return for_dentry_in_tree(wim_root_dentry(w), - dentry_find_streams_to_write, w); + struct wim_inode *inode; + struct hlist_node *cur; + struct hlist_head *inode_list; + + inode_list = &wim_get_current_image_metadata(w)->inode_list; + hlist_for_each_entry(inode, cur, inode_list, i_hlist) { + inode_find_streams_to_write(inode, w->lookup_table, + (struct list_head*)w->private); + } + return 0; } -static int write_wim_streams(WIMStruct *w, int image, int write_flags, - unsigned num_threads) +static int +write_wim_streams(WIMStruct *w, int image, int write_flags, + unsigned num_threads, + wimlib_progress_func_t progress_func) { + 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); + for_image(w, image, image_find_streams_to_write); return write_stream_list(&stream_list, w->out_fp, wimlib_get_compression_type(w), write_flags, - num_threads); + num_threads, progress_func); } /* - * Write the lookup table, xml data, and integrity table, then overwrite the WIM - * header. + * Finish writing a WIM file: write the lookup table, xml data, and integrity + * table (optional), then overwrite the WIM header. + * + * write_flags is a bitwise OR of the following: + * + * (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. + * + * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE: + * When (if) writing the integrity table, re-use entries from the + * existing integrity table, if possible. + * + * (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML: + * After writing the XML data but before writing the integrity + * table, write a temporary WIM header and flush the stream so that + * the WIM is less likely to become corrupted upon abrupt program + * termination. + * + * (private) WIMLIB_WRITE_FLAG_FSYNC: + * 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) { - off_t lookup_table_offset; - off_t xml_data_offset; - off_t lookup_table_size; - off_t integrity_offset; - off_t xml_data_size; - off_t end_offset; - off_t integrity_size; int ret; struct wim_header hdr; FILE *out = w->out_fp; - if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - /* Write the lookup table. */ - lookup_table_offset = ftello(out); - if (lookup_table_offset == -1) - return WIMLIB_ERR_WRITE; - - DEBUG("Writing lookup table (offset %"PRIu64")", - lookup_table_offset); - ret = write_lookup_table(w->lookup_table, out); - if (ret != 0) - return ret; - } - - xml_data_offset = ftello(out); - if (xml_data_offset == -1) - return WIMLIB_ERR_WRITE; - /* @hdr will be the header for the new WIM. First copy all the data * from the header in the WIMStruct; then set all the fields that may * have changed, including the resource entries, boot index, and image * count. */ memcpy(&hdr, &w->hdr, sizeof(struct wim_header)); + if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { - lookup_table_size = xml_data_offset - lookup_table_offset; - hdr.lookup_table_res_entry.offset = lookup_table_offset; - hdr.lookup_table_res_entry.size = lookup_table_size; + ret = write_lookup_table(w->lookup_table, out, &hdr.lookup_table_res_entry); + if (ret != 0) + goto out; } - hdr.lookup_table_res_entry.original_size = hdr.lookup_table_res_entry.size; - hdr.lookup_table_res_entry.flags = WIM_RESHDR_FLAG_METADATA; - DEBUG("Writing XML data (offset %"PRIu64")", xml_data_offset); ret = write_xml_data(w->wim_info, image, out, (write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ? - wim_info_get_total_bytes(w->wim_info) : 0); + wim_info_get_total_bytes(w->wim_info) : 0, + &hdr.xml_res_entry); if (ret != 0) - return ret; + goto out; - integrity_offset = ftello(out); - if (integrity_offset == -1) - return WIMLIB_ERR_WRITE; - xml_data_size = integrity_offset - xml_data_offset; + if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { + if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) { + struct wim_header checkpoint_hdr; + memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header)); + memset(&checkpoint_hdr.integrity, 0, sizeof(struct resource_entry)); + if (fseeko(out, 0, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to beginning " + "of WIM being written"); + ret = WIMLIB_ERR_WRITE; + goto out; + } + ret = write_header(&checkpoint_hdr, out); + if (ret != 0) + goto out; - hdr.xml_res_entry.offset = xml_data_offset; - hdr.xml_res_entry.size = xml_data_size; - hdr.xml_res_entry.original_size = xml_data_size; - hdr.xml_res_entry.flags = 0; + if (fflush(out) != 0) { + ERROR_WITH_ERRNO("Can't write data to WIM"); + ret = WIMLIB_ERR_WRITE; + goto out; + } - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) { - ret = write_integrity_table(out, WIM_HEADER_DISK_SIZE, - xml_data_offset, - write_flags & WIMLIB_WRITE_FLAG_SHOW_PROGRESS); + if (fseeko(out, 0, SEEK_END) != 0) { + ERROR_WITH_ERRNO("Failed to seek to end " + "of WIM being written"); + ret = WIMLIB_ERR_WRITE; + goto out; + } + } + + off_t old_lookup_table_end; + off_t new_lookup_table_end; + if (write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE) { + old_lookup_table_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + } else { + old_lookup_table_end = 0; + } + new_lookup_table_end = hdr.lookup_table_res_entry.offset + + hdr.lookup_table_res_entry.size; + + ret = write_integrity_table(out, + &hdr.integrity, + new_lookup_table_end, + old_lookup_table_end, + progress_func); if (ret != 0) - return ret; - end_offset = ftello(out); - if (end_offset == -1) - return WIMLIB_ERR_WRITE; - integrity_size = end_offset - integrity_offset; - hdr.integrity.offset = integrity_offset; - hdr.integrity.size = integrity_size; - hdr.integrity.original_size = integrity_size; + goto out; } else { - hdr.integrity.offset = 0; - hdr.integrity.size = 0; - hdr.integrity.original_size = 0; + memset(&hdr.integrity, 0, sizeof(struct resource_entry)); } - hdr.integrity.flags = 0; - - DEBUG("Updating WIM header."); /* * In the WIM header, there is room for the resource entry for a @@ -1685,19 +1610,9 @@ int finish_write(WIMStruct *w, int image, int write_flags) * 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) { + if (image != WIMLIB_ALL_IMAGES) { hdr.image_count = 1; if (hdr.boot_idx == image) hdr.boot_idx = 1; @@ -1705,104 +1620,434 @@ int finish_write(WIMStruct *w, int image, int write_flags) hdr.boot_idx = 0; } + if (hdr.boot_idx == 0) { + memset(&hdr.boot_metadata_res_entry, 0, + sizeof(struct resource_entry)); + } else { + memcpy(&hdr.boot_metadata_res_entry, + &w->image_metadata[ + hdr.boot_idx - 1].metadata_lte->output_resource_entry, + sizeof(struct resource_entry)); + } - if (fseeko(out, 0, SEEK_SET) != 0) - return WIMLIB_ERR_WRITE; + if (fseeko(out, 0, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Failed to seek to beginning of WIM " + "being written"); + ret = WIMLIB_ERR_WRITE; + goto out; + } ret = write_header(&hdr, out); if (ret != 0) - return ret; + goto out; - DEBUG("Closing output file."); - wimlib_assert(w->out_fp != NULL); - if (fclose(w->out_fp) != 0) { + if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) { + if (fflush(out) != 0 + || fsync(fileno(out)) != 0) + { + ERROR_WITH_ERRNO("Error flushing data to WIM file"); + ret = WIMLIB_ERR_WRITE; + } + } +out: + if (fclose(out) != 0) { ERROR_WITH_ERRNO("Failed to close the WIM file"); - ret = WIMLIB_ERR_WRITE; + if (ret == 0) + ret = WIMLIB_ERR_WRITE; } w->out_fp = NULL; return ret; } -/* Open file stream and write dummy header for WIM. */ -int begin_write(WIMStruct *w, const char *path, int write_flags) +#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK) +int +lock_wim(WIMStruct *w, FILE *fp) +{ + int ret = 0; + if (fp && !w->wim_locked) { + ret = flock(fileno(fp), LOCK_EX | LOCK_NB); + if (ret != 0) { + if (errno == EWOULDBLOCK) { + ERROR("`%s' is already being modified or has been " + "mounted read-write\n" + " by another process!", w->filename); + ret = WIMLIB_ERR_ALREADY_LOCKED; + } else { + WARNING_WITH_ERRNO("Failed to lock `%s'", + w->filename); + ret = 0; + } + } else { + w->wim_locked = 1; + } + } + return ret; +} +#endif + +static int +open_wim_writable(WIMStruct *w, const mbchar *path, + bool trunc, bool readable) { const char *mode; - DEBUG("Opening `%s' for new WIM", path); - - /* checking the integrity requires going back over the file to read it. - * XXX - * (It also would be possible to keep a running sha1sum as the file is - * written-- this would be faster, but a bit more complicated) */ - if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) - mode = "w+b"; + if (trunc) + if (readable) + mode = "w+b"; + else + mode = "wb"; else - mode = "wb"; - - if (w->out_fp) - fclose(w->out_fp); + mode = "r+b"; + wimlib_assert(w->out_fp == NULL); w->out_fp = fopen(path, mode); - if (!w->out_fp) { - ERROR_WITH_ERRNO("Failed to open the file `%s' for writing", - path); + if (w->out_fp) { + return 0; + } else { + ERROR_WITH_ERRNO("Failed to open `%s' for writing", path); return WIMLIB_ERR_OPEN; } +} + + +void +close_wim_writable(WIMStruct *w) +{ + if (w->out_fp) { + if (fclose(w->out_fp) != 0) { + WARNING_WITH_ERRNO("Failed to close output WIM"); + } + w->out_fp = NULL; + } +} +/* Open file stream and write dummy header for WIM. */ +int +begin_write(WIMStruct *w, const mbchar *path, int write_flags) +{ + int ret; + ret = open_wim_writable(w, path, true, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); + if (ret != 0) + return ret; /* Write dummy header. It will be overwritten later. */ return write_header(&w->hdr, w->out_fp); } /* 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 mbchar *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_FLAG_NO_LOOKUP_TABLE; + 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; - if (w->hdr.total_parts != 1) { ERROR("Cannot call wimlib_write() on part of a split WIM"); 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) - return ret; + goto out; - for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL); + ret = write_wim_streams(w, image, write_flags, num_threads, + progress_func); + if (ret != 0) + goto out; + + 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 (progress_func) + progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_END, NULL); + + ret = finish_write(w, image, write_flags, progress_func); +out: + close_wim_writable(w); + DEBUG("wimlib_write(path=%s) = %d", path, ret); + return ret; +} + +static bool +any_images_modified(WIMStruct *w) +{ + for (int i = 0; i < w->hdr.image_count; i++) + if (w->image_metadata[i].modified) + return true; + return false; +} + +/* + * Overwrite a WIM, possibly appending streams to it. + * + * A WIM looks like (or is supposed to look like) the following: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * If we are not adding any streams or metadata resources, the lookup table is + * unchanged--- so we only need to overwrite the XML data, integrity table, and + * header. This operation is potentially unsafe if the program is abruptly + * terminated while the XML data or integrity table are being overwritten, but + * before the new header has been written. To partially alleviate this problem, + * a special flag (WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) is passed to + * finish_write() to cause a temporary WIM header to be written after the XML + * data has been written. This may prevent the WIM from becoming corrupted if + * the program is terminated while the integrity table is being calculated (but + * no guarantees, due to write re-ordering...). + * + * If we are adding new streams or images (metadata resources), the lookup table + * needs to be changed, and those streams need to be written. In this case, we + * try to perform a safe update of the WIM file by writing the streams *after* + * the end of the previous WIM, then writing the new lookup table, XML data, and + * (optionally) integrity table following the new streams. This will produce a + * layout like the following: + * + * Header (212 bytes) + * (OLD) Streams and metadata resources (variable size) + * (OLD) Lookup table (variable size) + * (OLD) XML data (variable size) + * (OLD) Integrity table (optional) (variable size) + * (NEW) Streams and metadata resources (variable size) + * (NEW) Lookup table (variable size) + * (NEW) XML data (variable size) + * (NEW) Integrity table (optional) (variable size) + * + * At all points, the WIM is valid as nothing points to the new data yet. Then, + * the header is overwritten to point to the new lookup table, XML data, and + * integrity table, to produce the following layout: + * + * Header (212 bytes) + * Streams and metadata resources (variable size) + * Nothing (variable size) + * More Streams and metadata resources (variable size) + * Lookup table (variable size) + * XML data (variable size) + * Integrity table (optional) (variable size) + * + * This method allows an image to be appended to a large WIM very quickly, and + * is is crash-safe except in the case of write re-ordering, but the + * disadvantage is that a small hole is left in the WIM where the old lookup + * table, xml data, and integrity table were. (These usually only take up a + * small amount of space compared to the streams, however.) + */ +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; + bool found_modified_image; + + DEBUG("Overwriting `%s' in-place", w->filename); + + /* Make sure that the integrity table (if present) is after the XML + * data, and that there are no stream resources, metadata resources, or + * lookup tables after the XML data. Otherwise, these data would be + * overwritten. */ + if (w->hdr.integrity.offset != 0 && + w->hdr.integrity.offset < w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the integrity table to be before the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } + + if (w->hdr.lookup_table_res_entry.offset > w->hdr.xml_res_entry.offset) { + ERROR("Didn't expect the lookup table to be after the XML data"); + return WIMLIB_ERR_RESOURCE_ORDER; + } - ret = write_wim_streams(w, image, write_flags, num_threads); + 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; + + 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. */ + old_wim_end = w->hdr.lookup_table_res_entry.offset + + w->hdr.lookup_table_res_entry.size; + write_flags |= WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE | + WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML; + } + INIT_LIST_HEAD(&stream_list); + ret = wim_find_new_streams(w, old_wim_end, &stream_list); + if (ret != 0) + return ret; + + ret = open_wim_writable(w, w->filename, false, + (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0); + if (ret != 0) + return ret; + + ret = lock_wim(w, w->out_fp); if (ret != 0) { - /*ERROR("Failed to write WIM file resources to `%s'", path);*/ + fclose(w->out_fp); + w->out_fp = NULL; return ret; } - ret = for_image(w, image, write_metadata_resource); + if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) { + ERROR_WITH_ERRNO("Can't seek to end of WIM"); + fclose(w->out_fp); + w->out_fp = NULL; + w->wim_locked = 0; + return WIMLIB_ERR_WRITE; + } + + if (!list_empty(&stream_list)) { + DEBUG("Writing newly added streams (offset = %"PRIu64")", + old_wim_end); + ret = write_stream_list(&stream_list, w->out_fp, + wimlib_get_compression_type(w), + write_flags, num_threads, + progress_func); + if (ret != 0) + goto out_ftruncate; + } else { + DEBUG("No new streams were added"); + } + + found_modified_image = false; + for (int i = 0; i < w->hdr.image_count; i++) { + if (!found_modified_image) + found_modified_image = w->image_metadata[i].modified; + if (found_modified_image) { + select_wim_image(w, i + 1); + ret = write_metadata_resource(w); + if (ret != 0) + goto out_ftruncate; + } + } + write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE; + ret = finish_write(w, WIMLIB_ALL_IMAGES, write_flags, + progress_func); +out_ftruncate: + close_wim_writable(w); + if (ret != 0 && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) { + WARNING("Truncating `%s' to its original size (%"PRIu64" bytes)", + w->filename, old_wim_end); + /* Return value of truncate() is ignored because this is already + * an error path. */ + (void)truncate(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, + wimlib_progress_func_t progress_func) +{ + size_t wim_name_len; + int ret; + + DEBUG("Overwriting `%s' via a temporary file", w->filename); + /* Write the WIM to a temporary file in the same directory as the + * original WIM. */ + wim_name_len = strlen(w->filename); + mbchar tmpfile[wim_name_len + 10]; + memcpy(tmpfile, w->filename, wim_name_len); + randomize_char_array_with_alnum(tmpfile + wim_name_len, 9); + tmpfile[wim_name_len + 9] = '\0'; + + ret = wimlib_write(w, tmpfile, WIMLIB_ALL_IMAGES, + write_flags | WIMLIB_WRITE_FLAG_FSYNC, + num_threads, progress_func); if (ret != 0) { - /*ERROR("Failed to write WIM image metadata to `%s'", path);*/ - return ret; + ERROR("Failed to write the WIM file `%s'", tmpfile); + goto err; } - ret = finish_write(w, image, write_flags); - if (ret != 0) - return ret; + DEBUG("Renaming `%s' to `%s'", tmpfile, w->filename); - if (write_flags & WIMLIB_WRITE_FLAG_VERBOSE) - printf("Successfully wrote `%s'\n", path); - return 0; + /* Rename the new file to the old file .*/ + if (rename(tmpfile, w->filename) != 0) { + ERROR_WITH_ERRNO("Failed to rename `%s' to `%s'", + tmpfile, w->filename); + ret = WIMLIB_ERR_RENAME; + goto err; + } + + if (progress_func) { + union wimlib_progress_info progress; + progress.rename.from = tmpfile; + progress.rename.to = w->filename; + progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress); + } + + /* Close the original WIM file that was opened for reading. */ + if (w->fp != NULL) { + fclose(w->fp); + w->fp = NULL; + } + + /* Re-open the WIM read-only. */ + w->fp = fopen(w->filename, "rb"); + if (w->fp == NULL) { + ret = WIMLIB_ERR_REOPEN; + WARNING_WITH_ERRNO("Failed to re-open `%s' read-only", + w->filename); + FREE(w->filename); + w->filename = NULL; + } + return ret; +err: + /* Remove temporary file. */ + if (unlink(tmpfile) != 0) + WARNING_WITH_ERRNO("Failed to remove `%s'", tmpfile); + 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, + wimlib_progress_func_t progress_func) +{ + write_flags &= WIMLIB_WRITE_MASK_PUBLIC; + + if (!w->filename) + return WIMLIB_ERR_NO_FILENAME; + + if (w->hdr.total_parts != 1) { + ERROR("Cannot modify a split WIM"); + return WIMLIB_ERR_SPLIT_UNSUPPORTED; + } + + if ((!w->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE)) + && !(write_flags & WIMLIB_WRITE_FLAG_REBUILD)) + { + 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, + progress_func); }