#include "dentry.h"
#include "lookup_table.h"
#include "xml.h"
-#include "lzx.h"
-#include "xpress.h"
-
#ifdef ENABLE_MULTITHREADED_COMPRESSION
# include <pthread.h>
# define INVALID_HANDLE_VALUE ((HANDLE)(-1))
#endif
-static int
-fflush_and_ftruncate(FILE *fp, off_t size)
-{
- int ret;
-
- ret = fflush(fp);
- if (ret != 0) {
- ERROR_WITH_ERRNO("Failed to flush data to output WIM file");
- return WIMLIB_ERR_WRITE;
- }
- 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;
-}
-
/* 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.) */
chunk_tab->table_disk_size) {
ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
"file resource");
+ FREE(chunk_tab);
ret = WIMLIB_ERR_WRITE;
goto out;
}
ret = 0;
-out:
*chunk_tab_ret = chunk_tab;
+out:
return ret;
}
/*
* compress_func_t- Pointer to a function to compresses a chunk
- * of a WIM resource. This may be either xpress_compress()
- * (xpress-compress.c) or lzx_compress() (lzx-compress.c).
+ * 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.
get_compress_func(int out_ctype)
{
if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
- return lzx_compress;
+ return wimlib_lzx_compress;
else
- return xpress_compress;
+ return wimlib_xpress_compress;
}
/*
*
* @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_fp: FILE * 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,
+write_wim_resource_chunk(const void *chunk, unsigned chunk_size,
FILE *out_fp, compress_func_t compress,
struct chunk_table *chunk_tab)
{
const u8 *out_chunk;
unsigned out_chunk_size;
- if (chunk_tab) {
+ if (compress) {
u8 *compressed_chunk = alloca(chunk_size);
out_chunk_size = compress(chunk, chunk_size, compressed_chunk);
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 wim_lookup_table_entry *lte
-
- #ifdef WITH_NTFS_3G
- , ntfs_inode **ni_ret
- #endif
- )
+finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte)
{
- switch (lte->resource_location) {
- case RESOURCE_IN_FILE_ON_DISK:
- if (!lte->file_on_disk_fp) {
- lte->file_on_disk_fp = tfopen(lte->file_on_disk, T("rb"));
- if (!lte->file_on_disk_fp) {
- ERROR_WITH_ERRNO("Failed to open the file "
- "`%"TS"'", lte->file_on_disk);
- return WIMLIB_ERR_OPEN;
- }
+ 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);
}
- break;
-#ifdef WITH_NTFS_3G
- case RESOURCE_IN_NTFS_VOLUME:
- if (!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);
- if (!ni) {
- ERROR_WITH_ERRNO("Failed to open inode `%"TS"' 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 `%"TS"' in "
- "NTFS volume", loc->path);
- ntfs_inode_close(ni);
- return WIMLIB_ERR_NTFS_3G;
- }
- *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_data_only(lte->file_on_disk);
- if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
- ERROR("Win32 API: Can't open %"TS, lte->file_on_disk);
- win32_error_last();
- return WIMLIB_ERR_OPEN;
- }
- }
- break;
-#endif
- default:
- break;
+ return WIMLIB_ERR_INVALID_RESOURCE_HASH;
}
return 0;
}
-/* 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)
- {
- 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
-#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
-}
+
+struct write_resource_ctx {
+ compress_func_t compress;
+ struct chunk_table *chunk_tab;
+ FILE *out_fp;
+ SHA_CTX sha_ctx;
+ bool doing_sha;
+};
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)
+write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx)
{
- int ret;
- if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
- ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of "
- "output WIM file", file_offset);
- return WIMLIB_ERR_WRITE;
- }
- ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
- out_res_entry, 0);
- if (ret != 0)
- return ret;
+ struct write_resource_ctx *ctx = _ctx;
- return fflush_and_ftruncate(out_fp,
- file_offset + wim_resource_size(lte));
+ if (ctx->doing_sha)
+ sha1_update(&ctx->sha_ctx, chunk, chunk_size);
+
+ if (ctx->compress) {
+ return write_wim_resource_chunk(chunk, chunk_size,
+ ctx->out_fp, ctx->compress,
+ ctx->chunk_tab);
+ } else {
+ if (fwrite(chunk, 1, chunk_size, ctx->out_fp) != chunk_size) {
+ ERROR_WITH_ERRNO("Error writing to output WIM");
+ return WIMLIB_ERR_WRITE;
+ } else {
+ return 0;
+ }
+ }
}
/*
- * 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_fp: FILE * 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 wim_lookup_table_entry *lte,
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 = NULL;
-#ifdef WITH_NTFS_3G
- ntfs_inode *ni = NULL;
-#endif
-
- wimlib_assert(lte);
- /* Original size of the resource */
- original_size = wim_resource_size(lte);
+ flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS;
- /* Compressed size of the resource (as it exists now) */
- old_compressed_size = wim_resource_compressed_size(lte);
+ if (wim_resource_size(lte) == 0) {
+ /* Empty resource; nothing needs to be done, so just return
+ * success. */
+ return 0;
+ }
- /* 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 = ftello(out_fp);
+ 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 != WIMLIB_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 != WIMLIB_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 WIMLIB_COMPRESSION_TYPE_NONE: The new compressed size is the
- * original size
- *
- * Using a different compression type: Call
- * finish_wim_resource_chunk_tab() and it will provide the new
- * compressed size.
- */
- if (raw) {
- new_compressed_size = old_compressed_size;
+ /* 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 == WIMLIB_COMPRESSION_TYPE_NONE)
- new_compressed_size = original_size;
- else {
- ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
- &new_compressed_size);
- if (ret != 0)
- goto out_fclose;
- }
+ write_ctx.compress = get_compress_func(out_ctype);
+ ret = begin_wim_resource_chunk_tab(lte, out_fp,
+ offset,
+ &write_ctx.chunk_tab);
+ if (ret)
+ return ret;
}
- /* 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 `%"TS"'; maybe "
- "it changed while we were reading it.",
- lte->file_on_disk);
- }
- ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
- 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_fp = out_fp;
+try_write_again:
+ ret = read_resource_prefix(lte, read_size,
+ write_resource_cb, &write_ctx, flags);
+ if (ret)
+ goto out_free_chunk_tab;
- if (!raw && new_compressed_size >= original_size &&
- out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
- {
- /* Oops! We compressed the resource to larger than the original
- * size. Write the resource uncompressed instead. */
- ret = write_uncompressed_resource_and_truncate(lte,
- out_fp,
- file_offset,
- out_res_entry);
- if (ret != 0)
- goto out_fclose;
+ /* 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 != WIMLIB_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_fp,
+ &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. */
+ if (fseeko(out_fp, offset, SEEK_SET) ||
+ fflush(out_fp) ||
+ ftruncate(fileno(out_fp),
+ offset + wim_resource_size(lte)))
+ {
+ ERROR_WITH_ERRNO("Failed to flush and/or truncate "
+ "output WIM file");
+ ret = WIMLIB_ERR_WRITE;
+ goto out_free_chunk_tab;
+ }
+ DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; "
+ "writing uncompressed instead",
+ wim_resource_size(lte), new_size);
+ 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;
}
}
static int
-do_write_stream_list(struct list_head *my_resources,
+do_write_stream_list(struct list_head *stream_list,
+ struct wim_lookup_table *lookup_table,
FILE *out_fp,
int out_ctype,
wimlib_progress_func_t progress_func,
int write_resource_flags)
{
int ret;
- struct wim_lookup_table_entry *lte, *tmp;
+ struct wim_lookup_table_entry *lte;
+
+ /* For each stream in @stream_list ... */
+ while (!list_empty(stream_list)) {
+ 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)
+ return ret;
+ 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));
+ goto skip_to_progress;
+ }
+ }
+ }
- list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
+ /* 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);
ret = write_wim_resource(lte,
out_fp,
out_ctype,
<e->output_resource_entry,
write_resource_flags);
- if (ret != 0)
+ if (ret)
return ret;
- list_del(<e->staging_list);
-
+ if (lte->unhashed) {
+ list_del(<e->unhashed_list);
+ lookup_table_insert(lookup_table, lte);
+ lte->unhashed = 0;
+ }
+ skip_to_progress:
do_write_streams_progress(progress,
progress_func,
wim_resource_size(lte));
static int
write_stream_list_serial(struct list_head *stream_list,
+ struct wim_lookup_table *lookup_table,
FILE *out_fp,
int out_ctype,
int write_flags,
wimlib_progress_func_t progress_func,
union wimlib_progress_info *progress)
{
- int write_resource_flags;
-
+ int write_resource_flags = 0;
if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
- write_resource_flags = WIMLIB_RESOURCE_FLAG_RECOMPRESS;
- else
- write_resource_flags = 0;
+ write_resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS;
+
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,
+ return do_write_stream_list(stream_list,
+ lookup_table,
+ out_fp,
out_ctype, progress_func,
progress, write_resource_flags);
}
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.
- */
+struct main_writer_thread_ctx {
+ struct list_head *stream_list;
+ struct wim_lookup_table *lookup_table;
+ 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;
+
+ struct list_head available_msgs;
+ struct list_head outstanding_streams;
+ struct list_head serial_streams;
+ u64 next_chunk;
+ u64 next_num_chunks;
+ struct message *msgs;
+ struct message *next_msg;
+ size_t next_chunk_in_msg;
+ struct wim_lookup_table_entry *cur_lte;
+ struct chunk_table *cur_chunk_tab;
+ struct wim_lookup_table_entry *next_lte;
+ SHA_CTX sha_ctx;
+ u8 next_hash[20];
+};
+
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)
+init_message(struct message *msg)
{
- int ret;
- 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);
-
- if (!msgs) {
- ret = WIMLIB_ERR_NOMEM;
- goto out;
+ 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;
+}
- 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
- // 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;
- u64 next_chunk = 0;
- u64 next_num_chunks = 0;
-
- // As in write_wim_resource(), each resource we read is checksummed.
- SHA_CTX next_sha_ctx;
- u8 next_hash[SHA1_HASH_SIZE];
-
- // Resources that don't need any chunks compressed are added to this
- // list and written directly by the main thread.
- LIST_HEAD(my_resources);
-
- struct wim_lookup_table_entry *cur_lte = NULL;
- struct message *msg;
+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);
+ }
+}
- DEBUG("Initializing buffers for uncompressed "
- "and compressed data (%zu bytes needed)",
- num_messages * MAX_CHUNKS_PER_MSG * WIM_CHUNK_SIZE * 2);
+static struct message *
+allocate_messages(size_t num_messages)
+{
+ struct message *msgs;
- // Pre-allocate all the buffers that will be needed to do the chunk
- // compression.
+ msgs = CALLOC(num_messages, sizeof(struct message));
+ if (!msgs)
+ return NULL;
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);
-
- // 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)
- {
- ret = WIMLIB_ERR_NOMEM;
- goto out;
- }
+ 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)) {
- 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 `%"TS"'; "
- "maybe it changed while we were "
- "reading it.",
- next_lte->file_on_disk);
- }
- ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
- goto out;
- }
- }
+static void
+main_writer_thread_destroy_ctx(struct main_writer_thread_ctx *ctx)
+{
+ free_messages(ctx->msgs, ctx->num_messages);
+ FREE(ctx->cur_chunk_tab);
+}
- // 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;
- }
+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;
- // 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) {
- size = MODULO_NONZERO(wim_resource_size(next_lte),
- WIM_CHUNK_SIZE);
- }
+ /* 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);
- DEBUG2("Read resource (size=%u, offset=%zu)",
- size, next_chunk * WIM_CHUNK_SIZE);
+ /* 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 (cur_lte).
+ *
+ * The last stream in outstanding_streams is the stream that is
+ * currently being read and chunks fed to the compressor threads. */
+ INIT_LIST_HEAD(&ctx->outstanding_streams);
- msg->uncompressed_chunk_sizes[i] = size;
+ /* 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);
- 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++;
- }
+ ctx->cur_lte = NULL;
+ return 0;
+}
- // 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");
- }
+static int
+receive_compressed_chunks(struct main_writer_thread_ctx *ctx)
+{
+ struct message *msg;
+ struct wim_lookup_table_entry *cur_lte;
+ int ret;
- // If there are no outstanding resources, there are no more
- // resources that need to be written.
- if (list_empty(&outstanding_resources)) {
- ret = 0;
- goto out;
- }
+ wimlib_assert(!list_empty(&ctx->outstanding_streams));
- // Get the next message from the queue and process it.
- // The message will contain 1 or more data chunks that have been
- // compressed.
- msg = shared_queue_get(compressed_res_queue);
- msg->complete = true;
-
- // Is this the next chunk in the current resource? If it's not
- // (i.e., an earlier chunk in a same or different resource
- // hasn't been compressed yet), do nothing, and keep this
- // message around until all earlier chunks are received.
- //
- // Otherwise, write all the chunks we can.
- while (cur_lte != NULL &&
- !list_empty(&cur_lte->msg_list) &&
- (msg = container_of(cur_lte->msg_list.next,
- struct message,
- list))->complete)
- {
- DEBUG2("Complete msg (begin_chunk=%"PRIu64")", msg->begin_chunk);
- if (msg->begin_chunk == 0) {
- DEBUG2("Begin chunk tab");
-
- // This is the first set of chunks. Leave space
- // for the chunk table in the output file.
- off_t cur_offset = ftello(out_fp);
- if (cur_offset == -1) {
- ret = WIMLIB_ERR_WRITE;
- goto out;
- }
- ret = begin_wim_resource_chunk_tab(cur_lte,
- out_fp,
- cur_offset,
- &cur_chunk_tab);
- if (ret != 0)
- goto out;
- }
+ /* 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;
+ cur_lte = ctx->cur_lte;
- // Write the compressed chunks from the message.
- ret = write_wim_chunks(msg, out_fp, cur_chunk_tab);
- if (ret != 0)
+ /* 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)
+ {
+ 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 = ftello(ctx->out_fp);
+ if (cur_offset == -1) {
+ ret = WIMLIB_ERR_WRITE;
+ goto out;
+ }
+ ret = begin_wim_resource_chunk_tab(cur_lte,
+ ctx->out_fp,
+ cur_offset,
+ &ctx->cur_chunk_tab);
+ if (ret)
goto out;
+ }
- list_del(&msg->list);
+ /* Write the compressed chunks from the message. */
+ ret = write_wim_chunks(msg, ctx->out_fp, ctx->cur_chunk_tab);
+ if (ret)
+ goto out;
- // This message is available to use for different chunks
- // now.
- list_add(&msg->list, &available_msgs);
+ list_del(&msg->list);
- // 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;
+ /* This message is available to use for different chunks
+ * now. */
+ list_add(&msg->list, &ctx->available_msgs);
- if (res_csize >= wim_resource_size(cur_lte)) {
- /* Oops! We compressed the resource to
- * larger than the original size. Write
- * the resource uncompressed instead. */
- ret = write_uncompressed_resource_and_truncate(
- cur_lte,
- out_fp,
- cur_chunk_tab->file_offset,
- &cur_lte->output_resource_entry);
- if (ret != 0)
- goto out;
- } else {
- cur_lte->output_resource_entry.size =
- res_csize;
-
- cur_lte->output_resource_entry.original_size =
- cur_lte->resource_entry.original_size;
-
- cur_lte->output_resource_entry.offset =
- cur_chunk_tab->file_offset;
-
- cur_lte->output_resource_entry.flags =
- cur_lte->resource_entry.flags |
- WIM_RESHDR_FLAG_COMPRESSED;
- }
+ /* 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)
+ {
+ DEBUG2("Finish wim chunk tab");
+ u64 res_csize;
+ ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab,
+ ctx->out_fp,
+ &res_csize);
+ if (ret)
+ goto out;
- do_write_streams_progress(progress, progress_func,
- wim_resource_size(cur_lte));
-
- 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)
- cur_lte = NULL;
- else
- cur_lte = container_of(cur_lte->staging_list.next,
- struct wim_lookup_table_entry,
- staging_list);
-
- // Since we just finished writing a stream,
- // write any streams that have been added to the
- // my_resources list for direct writing by the
- // main thread (e.g. resources that don't need
- // to be compressed because the desired
- // compression type is the same as the previous
- // compression type).
- ret = do_write_stream_list(&my_resources,
- out_fp,
- out_ctype,
- progress_func,
- progress,
- 0);
- if (ret != 0)
+#if 0
+ 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,
+ ctx->out_fp,
+ ctx->cur_chunk_tab->file_offset,
+ &cur_lte->output_resource_entry);
+ if (ret)
goto out;
+ } else
+#endif
+ {
+ cur_lte->output_resource_entry.size =
+ res_csize;
+
+ cur_lte->output_resource_entry.original_size =
+ cur_lte->resource_entry.original_size;
+
+ cur_lte->output_resource_entry.offset =
+ ctx->cur_chunk_tab->file_offset;
+
+ cur_lte->output_resource_entry.flags =
+ cur_lte->resource_entry.flags |
+ WIM_RESHDR_FLAG_COMPRESSED;
}
+ do_write_streams_progress(ctx->progress, ctx->progress_func,
+ wim_resource_size(cur_lte));
+ FREE(ctx->cur_chunk_tab);
+ ctx->cur_chunk_tab = NULL;
+
+ struct list_head *next = cur_lte->write_streams_list.next;
+ list_del(&cur_lte->write_streams_list);
+
+ if (next == &ctx->outstanding_streams)
+ cur_lte = NULL;
+ else
+ cur_lte = container_of(cur_lte->write_streams_list.next,
+ struct wim_lookup_table_entry,
+ write_streams_list);
+
+ /* 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). */
+ ret = do_write_stream_list(&ctx->serial_streams,
+ ctx->lookup_table,
+ ctx->out_fp,
+ ctx->out_ctype,
+ ctx->progress_func,
+ ctx->progress,
+ 0);
+ if (ret)
+ goto out;
}
}
-
out:
- if (ret == WIMLIB_ERR_NOMEM) {
- ERROR("Could not allocate enough memory for "
- "multi-threaded compression");
+ ctx->cur_lte = cur_lte;
+ return ret;
+}
+
+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;
+
+ next_msg = ctx->next_msg;
+
+ sha1_update(&ctx->sha_ctx, chunk, chunk_size);
+
+ if (!next_msg) {
+ if (list_empty(&ctx->available_msgs)) {
+ ret = receive_compressed_chunks(ctx);
+ if (ret)
+ return ret;
+ }
+
+ wimlib_assert(!list_empty(&ctx->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_chunk_in_msg = 0;
}
- if (next_lte) {
-#ifdef WITH_NTFS_3G
- end_wim_resource_read(next_lte, ni);
-#else
- end_wim_resource_read(next_lte);
-#endif
+ wimlib_assert(next_msg != NULL);
+ wimlib_assert(ctx->next_chunk_in_msg < next_msg->num_chunks);
+
+ next_msg->uncompressed_chunk_sizes[ctx->next_chunk_in_msg] = chunk_size;
+ memcpy(next_msg->uncompressed_chunks[ctx->next_chunk_in_msg],
+ chunk, chunk_size);
+
+ if (++ctx->next_chunk_in_msg == next_msg->num_chunks) {
+ shared_queue_put(ctx->res_to_compress_queue,
+ next_msg);
+ ctx->next_msg = NULL;
}
+ return 0;
+}
- if (ret == 0) {
- ret = do_write_stream_list(&my_resources, out_fp,
- out_ctype, progress_func,
- progress, 0);
- } else {
- if (msgs) {
- size_t num_available_msgs = 0;
- struct list_head *cur;
+static int
+submit_stream_for_compression(struct wim_lookup_table_entry *lte,
+ struct main_writer_thread_ctx *ctx)
+{
+ int ret;
- list_for_each(cur, &available_msgs) {
- num_available_msgs++;
- }
+ sha1_init(&ctx->sha_ctx);
+ ctx->next_num_chunks = wim_resource_chunks(lte);
+ ret = read_resource_prefix(lte, wim_resource_size(lte),
+ main_writer_thread_cb, ctx, 0);
+ if (ret)
+ return ret;
+ ret = finalize_and_check_sha1(&ctx->sha_ctx, lte);
+ if (ret)
+ return ret;
+}
+
+/*
+ * 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 main_writer_thread_ctx *ctx)
+{
+ int ret;
+ struct list_head *stream_list;
+ struct wim_lookup_table_entry *lte;
- while (num_available_msgs < num_messages) {
- shared_queue_get(compressed_res_queue);
- num_available_msgs++;
+ ret = main_writer_thread_init_ctx(ctx);
+ if (ret)
+ goto out_destroy_ctx;
+
+ stream_list = ctx->stream_list;
+ while (!list_empty(stream_list)) {
+ 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) {
+ struct wim_lookup_table_entry *tmp;
+ u32 orig_refcnt = lte->out_refcnt;
+
+ ret = hash_unhashed_stream(lte, ctx->lookup_table, &tmp);
+ if (ret)
+ goto out_destroy_ctx;
+ if (tmp != lte) {
+ lte = tmp;
+ if (orig_refcnt != tmp->out_refcnt) {
+ DEBUG("Discarding duplicate stream of length %"PRIu64,
+ wim_resource_size(lte));
+ goto skip_to_progress;
+ }
}
}
- }
- 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]);
+ if (wim_resource_size(lte) < 1000 ||
+ ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE ||
+ (lte->resource_location == RESOURCE_IN_WIM &&
+ wimlib_get_compression_type(lte->wim) == ctx->out_ctype))
+ {
+ list_add(<e->write_streams_list,
+ &ctx->serial_streams);
+ } else {
+ ret = submit_stream_for_compression(lte, ctx);
+ if (ret)
+ goto out_destroy_ctx;
+ if (lte->unhashed) {
+ list_del(<e->unhashed_list);
+ lookup_table_insert(ctx->lookup_table, lte);
+ lte->unhashed = 0;
}
}
- FREE(msgs);
+ skip_to_progress:
+ do_write_streams_progress(ctx->progress,
+ ctx->progress_func,
+ wim_resource_size(lte));
}
- FREE(cur_chunk_tab);
+ while (!list_empty(&ctx->outstanding_streams)) {
+ ret = receive_compressed_chunks(ctx);
+ if (ret)
+ goto out_destroy_ctx;
+ }
+ ret = 0;
+out_destroy_ctx:
+ main_writer_thread_destroy_ctx(ctx);
return ret;
}
static int
write_stream_list_parallel(struct list_head *stream_list,
+ struct wim_lookup_table *lookup_table,
FILE *out_fp,
int out_ctype,
int write_flags,
}
progress->write_streams.num_threads = num_threads;
- wimlib_assert(stream_list->next != stream_list);
static const double MESSAGES_PER_THREAD = 2.0;
size_t queue_size = (size_t)(num_threads * MESSAGES_PER_THREAD);
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);
+ struct main_writer_thread_ctx ctx;
+ memset(&ctx, 0, sizeof(ctx));
+ ctx.stream_list = stream_list;
+ ctx.lookup_table = lookup_table;
+ ctx.out_fp = out_fp;
+ 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_flags = write_flags;
+ ctx.progress_func = progress_func;
+ ctx.progress = progress;
+ ret = main_writer_thread_proc(&ctx);
out_join:
for (unsigned i = 0; i < num_threads; i++)
shared_queue_put(&res_to_compress_queue, NULL);
out_serial:
WARNING("Falling back to single-threaded compression");
return write_stream_list_serial(stream_list,
+ lookup_table,
out_fp,
out_ctype,
write_flags,
* @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)
+write_stream_list(struct list_head *stream_list,
+ struct wim_lookup_table *lookup_table,
+ FILE *out_fp, int out_ctype, int write_flags,
+ unsigned num_threads, wimlib_progress_func_t progress_func)
{
struct wim_lookup_table_entry *lte;
size_t num_streams = 0;
u64 total_bytes = 0;
u64 total_compression_bytes = 0;
union wimlib_progress_info progress;
+ int ret;
+
+ if (list_empty(stream_list))
+ return 0;
- list_for_each_entry(lte, stream_list, staging_list) {
+ 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
#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);
+ ret = write_stream_list_parallel(stream_list,
+ lookup_table,
+ 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);
+ ret = write_stream_list_serial(stream_list,
+ lookup_table,
+ out_fp,
+ out_ctype,
+ write_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 *hashed_lte;
+ struct hlist_node *tmp;
+
+ pos = hash_u64(wim_resource_size(lte)) % tab->capacity;
+ lte->unique_size = 1;
+ hlist_for_each_entry(hashed_lte, tmp, &tab->array[pos], hash_list_2) {
+ if (wim_resource_size(hashed_lte) == wim_resource_size(lte)) {
+ lte->unique_size = 0;
+ hashed_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 list_head stream_list;
+ struct stream_size_table stream_size_tab;
};
static int
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)
+ lte->wim == args->wim)
{
- #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;
+ /* 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;
+ }
+ } else {
+ wimlib_assert(!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA));
+ list_add_tail(<e->write_streams_list, &args->stream_list);
}
-
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))
- if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
- list_add(<e->staging_list, args->stream_list);
+ stream_size_table_insert(lte, &args->stream_size_tab);
return 0;
}
static int
-wim_prepare_streams(WIMStruct *wim, off_t end_offset,
- struct list_head *stream_list)
+lte_set_output_res_entry(struct wim_lookup_table_entry *lte, void *_wim)
+{
+ if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == _wim) {
+ 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)
{
- struct lte_overwrite_prepare_args args = {
- .wim = wim,
- .end_offset = end_offset,
- .stream_list = stream_list,
- };
int ret;
+ struct lte_overwrite_prepare_args args;
+
+ 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 (int i = 0; i < wim->hdr.image_count; i++) {
- ret = lte_overwrite_prepare(wim->image_metadata[i].metadata_lte,
- &args);
- if (ret)
- return ret;
+ struct wim_image_metadata *imd;
+ struct wim_lookup_table_entry *lte;
+
+ imd = wim->image_metadata[i];
+ image_for_each_unhashed_stream(lte, imd) {
+ ret = lte_overwrite_prepare(lte, &args);
+ if (ret)
+ goto out_destroy_stream_size_table;
+ }
}
- return for_lookup_table_entry(wim->lookup_table,
- lte_overwrite_prepare, &args);
+ ret = for_lookup_table_entry(wim->lookup_table,
+ lte_overwrite_prepare, &args);
+ if (ret)
+ goto out_destroy_stream_size_table;
+
+ for (int i = 0; i < wim->hdr.image_count; i++)
+ lte_set_output_res_entry(wim->image_metadata[i]->metadata_lte,
+ wim);
+ for_lookup_table_entry(wim->lookup_table, lte_set_output_res_entry, wim);
+ INIT_LIST_HEAD(stream_list);
+ list_splice(&args.stream_list, stream_list);
+out_destroy_stream_size_table:
+ destroy_stream_size_table(&args.stream_size_tab);
+ return ret;
}
-static int
+
+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 list_head *stream_list,
+ struct stream_size_table *tab)
{
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);
+ 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;
}
}
- return 0;
}
static int
image_find_streams_to_write(WIMStruct *w)
{
+ struct wim_image_metadata *imd;
+ struct find_streams_ctx *ctx;
struct wim_inode *inode;
- struct hlist_node *cur;
- struct hlist_head *inode_list;
+ struct wim_lookup_table_entry *lte;
- inode_list = &wim_get_current_image_metadata(w)->inode_list;
- hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
+ 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,
- (struct list_head*)w->private);
+ &ctx->stream_list,
+ &ctx->stream_size_tab);
}
return 0;
}
+/* Given a WIM that from which one or all of the images is being written, build
+ * the list of unique streams ('struct wim_lookup_table_entry's) that must be
+ * written, plus any unhashed streams that need to be written but may be
+ * identical to other hashed or unhashed streams being written. These unhashed
+ * streams are checksummed while the streams are being written. To aid this
+ * process, the member @unique_size is set to 1 on streams that have a unique
+ * size and therefore must be written.
+ *
+ * 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)
+{
+ 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) {
+ INIT_LIST_HEAD(stream_list);
+ list_splice(&ctx.stream_list, stream_list);
+ }
+ return ret;
+}
+
+/* Writes the streams for the specified @image in @wim to @wim->out_fp.
+ */
static int
-write_wim_streams(WIMStruct *w, int image, int write_flags,
- unsigned num_threads,
- wimlib_progress_func_t progress_func)
+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, image_find_streams_to_write);
- return write_stream_list(&stream_list, w->out_fp,
- wimlib_get_compression_type(w), write_flags,
- num_threads, progress_func);
+ ret = prepare_stream_list(wim, image, &stream_list);
+ if (ret)
+ return ret;
+ return write_stream_list(&stream_list,
+ wim->lookup_table,
+ wim->out_fp,
+ wimlib_get_compression_type(wim),
+ write_flags,
+ num_threads,
+ progress_func);
}
/*
if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
ret = write_lookup_table(w, image, &hdr.lookup_table_res_entry);
- if (ret != 0)
- goto out;
+ if (ret)
+ goto out_close_wim;
}
ret = write_xml_data(w->wim_info, image, out,
(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
wim_info_get_total_bytes(w->wim_info) : 0,
&hdr.xml_res_entry);
- if (ret != 0)
- goto out;
+ 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) {
+ if (fseeko(out, 0, SEEK_SET)) {
ERROR_WITH_ERRNO("Failed to seek to beginning "
"of WIM being written");
ret = WIMLIB_ERR_WRITE;
- goto out;
+ goto out_close_wim;
}
ret = write_header(&checkpoint_hdr, out);
- if (ret != 0)
- goto out;
+ if (ret)
+ goto out_close_wim;
if (fflush(out) != 0) {
ERROR_WITH_ERRNO("Can't write data to WIM");
ret = WIMLIB_ERR_WRITE;
- goto out;
+ goto out_close_wim;
}
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;
+ goto out_close_wim;
}
}
new_lookup_table_end,
old_lookup_table_end,
progress_func);
- if (ret != 0)
- goto out;
+ if (ret)
+ goto out_close_wim;
} else {
memset(&hdr.integrity, 0, sizeof(struct resource_entry));
}
} else {
memcpy(&hdr.boot_metadata_res_entry,
&w->image_metadata[
- hdr.boot_idx - 1].metadata_lte->output_resource_entry,
+ hdr.boot_idx - 1]->metadata_lte->output_resource_entry,
sizeof(struct resource_entry));
}
ERROR_WITH_ERRNO("Failed to seek to beginning of WIM "
"being written");
ret = WIMLIB_ERR_WRITE;
- goto out;
+ goto out_close_wim;
}
ret = write_header(&hdr, out);
- if (ret != 0)
- goto out;
+ if (ret)
+ goto out_close_wim;
if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
if (fflush(out) != 0
ret = WIMLIB_ERR_WRITE;
}
}
-out:
+out_close_wim:
if (fclose(out) != 0) {
ERROR_WITH_ERRNO("Failed to close the WIM file");
if (ret == 0)
int ret;
ret = open_wim_writable(w, path, true,
(write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
- if (ret != 0)
+ if (ret)
return ret;
/* Write dummy header. It will be overwritten later. */
return write_header(&w->hdr, w->out_fp);
}
ret = begin_write(w, path, write_flags);
- if (ret != 0)
+ if (ret)
goto out;
ret = write_wim_streams(w, image, write_flags, num_threads,
progress_func);
- if (ret != 0)
+ if (ret)
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)
+ if (ret)
goto out;
if (progress_func)
any_images_modified(WIMStruct *w)
{
for (int i = 0; i < w->hdr.image_count; i++)
- if (w->image_metadata[i].modified)
+ if (w->image_metadata[i]->modified)
return true;
return false;
}
int ret;
struct list_head stream_list;
off_t old_wim_end;
- bool found_modified_image;
DEBUG("Overwriting `%"TS"' in-place", w->filename);
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. */
+ DEBUG("Skipping writing lookup table "
+ "(no images modified or deleted)");
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_prepare_streams(w, old_wim_end, &stream_list);
- if (ret != 0)
+ 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)
+ if (ret)
return ret;
ret = lock_wim(w, w->out_fp);
- if (ret != 0) {
+ if (ret) {
fclose(w->out_fp);
w->out_fp = NULL;
return ret;
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");
- }
+ DEBUG("Writing newly added streams (offset = %"PRIu64")",
+ old_wim_end);
+ ret = write_stream_list(&stream_list,
+ w->lookup_table,
+ w->out_fp,
+ wimlib_get_compression_type(w),
+ write_flags,
+ num_threads,
+ progress_func);
+ if (ret)
+ goto out_ftruncate;
- 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) {
+ if (w->image_metadata[i]->modified) {
select_wim_image(w, i + 1);
ret = write_metadata_resource(w);
- if (ret != 0)
+ if (ret)
goto out_ftruncate;
}
}