* along with wimlib; if not, see http://www.gnu.org/licenses/.
*/
-#include "config.h"
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
-/* On BSD, this should be included before "list.h" so that "list.h" can
+/* On BSD, this should be included before "wimlib/list.h" so that "wimlib/list.h" can
* overwrite the LIST_HEAD macro. */
# include <sys/file.h>
#endif
+#include "wimlib/endianness.h"
+#include "wimlib/error.h"
+#include "wimlib/file_io.h"
+#include "wimlib/header.h"
+#include "wimlib/integrity.h"
+#include "wimlib/lookup_table.h"
+#include "wimlib/metadata.h"
+#include "wimlib/resource.h"
+#include "wimlib/write.h"
+#include "wimlib/xml.h"
+
#ifdef __WIN32__
-# include "win32.h"
+# include "wimlib/win32.h" /* win32_get_number_of_processors() */
#endif
-#include "list.h"
-#include "wimlib_internal.h"
-#include "buffer_io.h"
-#include "dentry.h"
-#include "lookup_table.h"
-#include "xml.h"
-
#ifdef ENABLE_MULTITHREADED_COMPRESSION
# include <pthread.h>
#endif
#include <unistd.h>
+#include <fcntl.h>
#include <errno.h>
#ifdef WITH_NTFS_3G
#include <limits.h>
-#if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
-# define INVALID_HANDLE_VALUE ((HANDLE)(-1))
+#ifndef __WIN32__
+# include <sys/uio.h> /* for `struct iovec' */
#endif
/* Chunk table that's located at the beginning of each compressed resource in
*/
static int
begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
- FILE *out_fp,
+ int out_fd,
off_t file_offset,
struct chunk_table **chunk_tab_ret)
{
size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
+ DEBUG("Begin chunk table for stream with size %"PRIu64, size);
+
if (!chunk_tab) {
ERROR("Failed to allocate chunk table for %"PRIu64" byte "
"resource", size);
chunk_tab->cur_offset = 0;
chunk_tab->cur_offset_p = chunk_tab->offsets;
- if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
- chunk_tab->table_disk_size) {
+ if (full_write(out_fd, chunk_tab,
+ chunk_tab->table_disk_size) != chunk_tab->table_disk_size)
+ {
ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
"file resource");
FREE(chunk_tab);
*
* @chunk: Uncompressed data of the chunk.
* @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
- * @out_fp: FILE * to write the chunk to.
+ * @out_fd: File descriptor to write the chunk to.
* @compress: Compression function to use (NULL if writing uncompressed
* data).
* @chunk_tab: Pointer to chunk table being created. It is updated with the
* Returns 0 on success; nonzero on failure.
*/
static int
-write_wim_resource_chunk(const void *chunk, unsigned chunk_size,
- FILE *out_fp, compress_func_t compress,
- struct chunk_table *chunk_tab)
+write_wim_resource_chunk(const void * restrict chunk,
+ unsigned chunk_size,
+ int out_fd,
+ compress_func_t compress,
+ struct chunk_table * restrict chunk_tab)
{
- const u8 *out_chunk;
+ const void *out_chunk;
unsigned out_chunk_size;
if (compress) {
- u8 *compressed_chunk = alloca(chunk_size);
+ void *compressed_chunk = alloca(chunk_size);
- out_chunk_size = compress(chunk, chunk_size, compressed_chunk);
+ out_chunk_size = (*compress)(chunk, chunk_size, compressed_chunk);
if (out_chunk_size) {
/* Write compressed */
out_chunk = compressed_chunk;
out_chunk = chunk;
out_chunk_size = chunk_size;
}
- if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
+ if (full_write(out_fd, out_chunk, out_chunk_size) != out_chunk_size) {
ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
return WIMLIB_ERR_WRITE;
}
*/
static int
finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
- FILE *out_fp, u64 *compressed_size_p)
+ int out_fd, u64 *compressed_size_p)
{
size_t bytes_written;
- if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
- ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
- "WIM file", chunk_tab->file_offset);
- return WIMLIB_ERR_WRITE;
- }
if (chunk_tab->bytes_per_chunk_entry == 8) {
array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
((u32*)chunk_tab->offsets)[i] =
cpu_to_le32(chunk_tab->offsets[i]);
}
- bytes_written = fwrite((u8*)chunk_tab->offsets +
- chunk_tab->bytes_per_chunk_entry,
- 1, chunk_tab->table_disk_size, out_fp);
+ bytes_written = full_pwrite(out_fd,
+ (u8*)chunk_tab->offsets + chunk_tab->bytes_per_chunk_entry,
+ chunk_tab->table_disk_size,
+ chunk_tab->file_offset);
if (bytes_written != chunk_tab->table_disk_size) {
ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
"file resource");
return WIMLIB_ERR_WRITE;
}
- if (fseeko(out_fp, 0, SEEK_END) != 0) {
- ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
- return WIMLIB_ERR_WRITE;
- }
*compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
return 0;
}
static int
-finalize_and_check_sha1(SHA_CTX *sha_ctx, struct wim_lookup_table_entry *lte)
+seek_and_truncate(int out_fd, off_t offset)
+{
+ if (lseek(out_fd, offset, SEEK_SET) == -1 ||
+ ftruncate(out_fd, offset))
+ {
+ ERROR_WITH_ERRNO("Failed to truncate output WIM file");
+ return WIMLIB_ERR_WRITE;
+ } else {
+ return 0;
+ }
+}
+
+static int
+finalize_and_check_sha1(SHA_CTX * restrict sha_ctx,
+ struct wim_lookup_table_entry * restrict lte)
{
u8 md[SHA1_HASH_SIZE];
sha1_final(md, sha_ctx);
struct write_resource_ctx {
compress_func_t compress;
struct chunk_table *chunk_tab;
- FILE *out_fp;
+ int out_fd;
SHA_CTX sha_ctx;
bool doing_sha;
};
static int
-write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx)
+write_resource_cb(const void *restrict chunk, size_t chunk_size,
+ void *restrict _ctx)
{
struct write_resource_ctx *ctx = _ctx;
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;
- }
- }
+ return write_wim_resource_chunk(chunk, chunk_size,
+ ctx->out_fd, ctx->compress,
+ ctx->chunk_tab);
}
/*
* @lte: Lookup table entry for the resource, which could be in another WIM,
* in an external file, or in another location.
*
- * @out_fp: FILE * opened to the output WIM.
+ * @out_fd: File descriptor opened to the output WIM.
*
* @out_ctype: One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate
* which compression algorithm to use.
*/
int
write_wim_resource(struct wim_lookup_table_entry *lte,
- FILE *out_fp, int out_ctype,
+ int out_fd, int out_ctype,
struct resource_entry *out_res_entry,
int flags)
{
off_t offset;
int ret;
- DEBUG2("wim_resource_size(lte)=%"PRIu64, wim_resource_size(lte));
-
flags &= ~WIMLIB_RESOURCE_FLAG_RECOMPRESS;
/* Get current position in output WIM */
- offset = ftello(out_fp);
+ offset = filedes_offset(out_fd);
if (offset == -1) {
ERROR_WITH_ERRNO("Can't get position in output WIM");
return WIMLIB_ERR_WRITE;
write_ctx.chunk_tab = NULL;
} else {
write_ctx.compress = get_compress_func(out_ctype);
- ret = begin_wim_resource_chunk_tab(lte, out_fp,
+ ret = begin_wim_resource_chunk_tab(lte, out_fd,
offset,
&write_ctx.chunk_tab);
if (ret)
/* 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;
+ write_ctx.out_fd = out_fd;
try_write_again:
ret = read_resource_prefix(lte, read_size,
write_resource_cb, &write_ctx, flags);
/* 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,
+ ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fd,
&new_size);
if (ret)
goto out_free_chunk_tab;
if (new_size >= wim_resource_size(lte)) {
/* Oops! We compressed the resource to larger than the original
* size. Write the resource uncompressed instead. */
- 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);
+ ret = seek_and_truncate(out_fd, offset);
+ if (ret)
+ goto out_free_chunk_tab;
write_ctx.compress = NULL;
write_ctx.doing_sha = false;
out_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
struct message {
struct wim_lookup_table_entry *lte;
u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
- u8 *out_compressed_chunks[MAX_CHUNKS_PER_MSG];
u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
- unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
+ struct iovec out_chunks[MAX_CHUNKS_PER_MSG];
+ size_t total_out_bytes;
unsigned num_chunks;
struct list_head list;
bool complete;
static void
compress_chunks(struct message *msg, compress_func_t compress)
{
+ msg->total_out_bytes = 0;
for (unsigned i = 0; i < msg->num_chunks; i++) {
- DEBUG2("compress chunk %u of %u", i, msg->num_chunks);
unsigned len = compress(msg->uncompressed_chunks[i],
msg->uncompressed_chunk_sizes[i],
msg->compressed_chunks[i]);
+ void *out_chunk;
+ unsigned out_len;
if (len) {
/* To be written compressed */
- msg->out_compressed_chunks[i] = msg->compressed_chunks[i];
- msg->compressed_chunk_sizes[i] = len;
+ out_chunk = msg->compressed_chunks[i];
+ out_len = len;
} else {
/* To be written uncompressed */
- msg->out_compressed_chunks[i] = msg->uncompressed_chunks[i];
- msg->compressed_chunk_sizes[i] = msg->uncompressed_chunk_sizes[i];
-
+ out_chunk = msg->uncompressed_chunks[i];
+ out_len = msg->uncompressed_chunk_sizes[i];
}
+ msg->out_chunks[i].iov_base = out_chunk;
+ msg->out_chunks[i].iov_len = out_len;
+ msg->total_out_bytes += out_len;
}
}
static void
do_write_streams_progress(union wimlib_progress_info *progress,
wimlib_progress_func_t progress_func,
- uint64_t size_added)
+ uint64_t size_added,
+ bool stream_discarded)
{
- progress->write_streams.completed_bytes += size_added;
+ if (stream_discarded) {
+ progress->write_streams.total_bytes -= size_added;
+ if (progress->write_streams._private != ~(uint64_t)0 &&
+ progress->write_streams._private > progress->write_streams.total_bytes)
+ {
+ progress->write_streams._private = progress->write_streams.total_bytes;
+ }
+ } else {
+ progress->write_streams.completed_bytes += size_added;
+ }
progress->write_streams.completed_streams++;
if (progress_func &&
progress->write_streams.completed_bytes >= progress->write_streams._private)
progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
progress);
if (progress->write_streams._private == progress->write_streams.total_bytes) {
- progress->write_streams._private = ~0;
+ progress->write_streams._private = ~(uint64_t)0;
} else {
progress->write_streams._private =
min(progress->write_streams.total_bytes,
}
struct serial_write_stream_ctx {
- FILE *out_fp;
+ int out_fd;
int out_ctype;
int write_resource_flags;
};
serial_write_stream(struct wim_lookup_table_entry *lte, void *_ctx)
{
struct serial_write_stream_ctx *ctx = _ctx;
- return write_wim_resource(lte, ctx->out_fp,
+ return write_wim_resource(lte, ctx->out_fd,
ctx->out_ctype, <e->output_resource_entry,
ctx->write_resource_flags);
}
+/* Write a list of streams, taking into account that some streams may be
+ * duplicates that are checksummed and discarded on the fly, and also delegating
+ * the actual writing of a stream to a function @write_stream_cb, which is
+ * passed the context @write_stream_ctx. */
static int
do_write_stream_list(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
{
int ret = 0;
struct wim_lookup_table_entry *lte;
+ bool stream_discarded;
/* For each stream in @stream_list ... */
while (!list_empty(stream_list)) {
+ stream_discarded = false;
lte = container_of(stream_list->next,
struct wim_lookup_table_entry,
write_streams_list);
* just skip to the next stream. */
DEBUG("Discarding duplicate stream of length %"PRIu64,
wim_resource_size(lte));
+ lte->no_progress = 0;
+ stream_discarded = true;
goto skip_to_progress;
}
}
* the SHA1 message digest yet. */
wimlib_assert(lte->out_refcnt != 0);
lte->deferred = 0;
+ lte->no_progress = 0;
ret = (*write_stream_cb)(lte, write_stream_ctx);
if (ret)
break;
lte->unhashed = 0;
}
skip_to_progress:
- if (progress_func) {
+ if (!lte->no_progress) {
do_write_streams_progress(progress,
progress_func,
- wim_resource_size(lte));
+ wim_resource_size(lte),
+ stream_discarded);
}
}
return ret;
static int
do_write_stream_list_serial(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
- FILE *out_fp,
+ int out_fd,
int out_ctype,
int write_resource_flags,
wimlib_progress_func_t progress_func,
union wimlib_progress_info *progress)
{
struct serial_write_stream_ctx ctx = {
- .out_fp = out_fp,
+ .out_fd = out_fd,
.out_ctype = out_ctype,
.write_resource_flags = write_resource_flags,
};
static inline int
write_flags_to_resource_flags(int write_flags)
{
- return (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS) ?
- WIMLIB_RESOURCE_FLAG_RECOMPRESS : 0;
+ int resource_flags = 0;
+
+ if (write_flags & WIMLIB_WRITE_FLAG_RECOMPRESS)
+ resource_flags |= WIMLIB_RESOURCE_FLAG_RECOMPRESS;
+ return resource_flags;
}
static int
write_stream_list_serial(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
- FILE *out_fp,
+ int out_fd,
int out_ctype,
int write_resource_flags,
wimlib_progress_func_t progress_func,
union wimlib_progress_info *progress)
{
+ DEBUG("Writing stream list (serial version)");
progress->write_streams.num_threads = 1;
if (progress_func)
progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
return do_write_stream_list_serial(stream_list,
lookup_table,
- out_fp,
+ out_fd,
out_ctype,
write_resource_flags,
progress_func,
#ifdef ENABLE_MULTITHREADED_COMPRESSION
static int
-write_wim_chunks(struct message *msg, FILE *out_fp,
+write_wim_chunks(struct message *msg, int out_fd,
struct chunk_table *chunk_tab)
{
for (unsigned i = 0; i < msg->num_chunks; i++) {
- unsigned chunk_csize = msg->compressed_chunk_sizes[i];
-
- DEBUG2("Write wim chunk %u of %u (csize = %u)",
- i, msg->num_chunks, chunk_csize);
-
- if (fwrite(msg->out_compressed_chunks[i], 1, chunk_csize, out_fp)
- != chunk_csize)
- {
- ERROR_WITH_ERRNO("Failed to write WIM chunk");
- return WIMLIB_ERR_WRITE;
- }
-
*chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
- chunk_tab->cur_offset += chunk_csize;
+ chunk_tab->cur_offset += msg->out_chunks[i].iov_len;
+ }
+ if (full_writev(out_fd, msg->out_chunks,
+ msg->num_chunks) != msg->total_out_bytes)
+ {
+ ERROR_WITH_ERRNO("Failed to write WIM chunks");
+ return WIMLIB_ERR_WRITE;
}
return 0;
}
struct main_writer_thread_ctx {
struct list_head *stream_list;
struct wim_lookup_table *lookup_table;
- FILE *out_fp;
+ int out_fd;
int out_ctype;
int write_resource_flags;
struct shared_queue *res_to_compress_queue;
wimlib_assert(!list_empty(&ctx->outstanding_streams));
wimlib_assert(ctx->num_outstanding_messages != 0);
- DEBUG2("Receiving more compressed chunks");
cur_lte = container_of(ctx->outstanding_streams.next,
struct wim_lookup_table_entry,
being_compressed_list);
msg->complete = true;
--ctx->num_outstanding_messages;
- DEBUG2("recved msg %p", msg);
-
/* 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
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);
+ off_t cur_offset = filedes_offset(ctx->out_fd);
if (cur_offset == -1)
return WIMLIB_ERR_WRITE;
ret = begin_wim_resource_chunk_tab(cur_lte,
- ctx->out_fp,
+ ctx->out_fd,
cur_offset,
&ctx->cur_chunk_tab);
if (ret)
}
/* Write the compressed chunks from the message. */
- ret = write_wim_chunks(msg, ctx->out_fp, ctx->cur_chunk_tab);
+ ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab);
if (ret)
return ret;
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;
+ off_t offset;
+
ret = finish_wim_resource_chunk_tab(ctx->cur_chunk_tab,
- ctx->out_fp,
+ ctx->out_fd,
&res_csize);
if (ret)
return ret;
list_del(&cur_lte->being_compressed_list);
-#if 0
+
+ /* Grab the offset of this stream in the output file
+ * from the chunk table before we free it. */
+ offset = ctx->cur_chunk_tab->file_offset;
+
+ FREE(ctx->cur_chunk_tab);
+ ctx->cur_chunk_tab = NULL;
+
if (res_csize >= wim_resource_size(cur_lte)) {
/* Oops! We compressed the resource to
* larger than the original size. Write
* the resource uncompressed instead. */
- ret = write_uncompressed_resource_and_truncate(
- cur_lte,
- ctx->out_fp,
- ctx->cur_chunk_tab->file_offset,
- &cur_lte->output_resource_entry);
+ DEBUG("Compressed %"PRIu64" => %"PRIu64" bytes; "
+ "writing uncompressed instead",
+ wim_resource_size(cur_lte), res_csize);
+ ret = seek_and_truncate(ctx->out_fd, offset);
if (ret)
- goto out;
- } else
-#endif
- {
+ return ret;
+ ret = write_wim_resource(cur_lte,
+ ctx->out_fd,
+ WIMLIB_COMPRESSION_TYPE_NONE,
+ &cur_lte->output_resource_entry,
+ ctx->write_resource_flags);
+ if (ret)
+ return ret;
+ } else {
cur_lte->output_resource_entry.size =
res_csize;
cur_lte->resource_entry.original_size;
cur_lte->output_resource_entry.offset =
- ctx->cur_chunk_tab->file_offset;
+ 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;
+
+ do_write_streams_progress(ctx->progress,
+ ctx->progress_func,
+ wim_resource_size(cur_lte),
+ false);
/* Since we just finished writing a stream, write any
* streams that have been added to the serial_streams
* 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_serial(&ctx->serial_streams,
- ctx->lookup_table,
- ctx->out_fp,
- ctx->out_ctype,
- ctx->write_resource_flags,
- ctx->progress_func,
- ctx->progress);
- if (ret)
- return ret;
+ if (!list_empty(&ctx->serial_streams)) {
+ ret = do_write_stream_list_serial(&ctx->serial_streams,
+ ctx->lookup_table,
+ ctx->out_fd,
+ ctx->out_ctype,
+ ctx->write_resource_flags,
+ ctx->progress_func,
+ ctx->progress);
+ if (ret)
+ return ret;
+ }
+
+ /* Advance to the next stream to write. */
if (list_empty(&ctx->outstanding_streams)) {
cur_lte = NULL;
} else {
cur_lte = container_of(ctx->outstanding_streams.next,
struct wim_lookup_table_entry,
being_compressed_list);
- #ifdef ENABLE_MORE_DEBUG
- DEBUG2("Advance to stream:");
- print_lookup_table_entry(cur_lte, stderr);
- #endif
}
}
}
return 0;
}
+/* Called when the main thread has read a new chunk of data. */
static int
main_writer_thread_cb(const void *chunk, size_t chunk_size, void *_ctx)
{
struct message *next_msg;
u64 next_chunk_in_msg;
- DEBUG2("chunk_size=%zu, wim_resource_size(next_lte)=%"PRIu64,
- chunk_size, wim_resource_size(ctx->next_lte));
-
+ /* Update SHA1 message digest for the stream currently being read by the
+ * main thread. */
sha1_update(&ctx->next_sha_ctx, chunk, chunk_size);
+
+ /* We send chunks of data to the compressor chunks in batches which we
+ * refer to as "messages". @next_msg is the message that is currently
+ * being prepared to send off. If it is NULL, that indicates that we
+ * need to start a new message. */
next_msg = ctx->next_msg;
if (!next_msg) {
- /* Start filling in a new message */
-
- DEBUG2("Start new msg");
-
+ /* We need to start a new message. First check to see if there
+ * is a message available in the list of available messages. If
+ * so, we can just take one. If not, all the messages (there is
+ * a fixed number of them, proportional to the number of
+ * threads) have been sent off to the compressor threads, so we
+ * receive messages from the compressor threads containing
+ * compressed chunks of data.
+ *
+ * We may need to receive multiple messages before one is
+ * actually available to use because messages received that are
+ * *not* for the very next set of chunks to compress must be
+ * buffered until it's time to write those chunks. */
while (list_empty(&ctx->available_msgs)) {
- /* No message available; receive messages, writing
- * compressed data. */
- DEBUG2("No msgs available!");
ret = receive_compressed_chunks(ctx);
if (ret)
return ret;
next_msg->begin_chunk = ctx->next_chunk;
next_msg->num_chunks = min(MAX_CHUNKS_PER_MSG,
ctx->next_num_chunks - ctx->next_chunk);
- DEBUG2("next_msg {begin_chunk=%"PRIu64", num_chunks=%"PRIu64"}",
- next_msg->begin_chunk, next_msg->num_chunks);
ctx->next_msg = next_msg;
}
+ /* Fill in the next chunk to compress */
next_chunk_in_msg = ctx->next_chunk - next_msg->begin_chunk;
- /* Fill in the next chunk to compress */
next_msg->uncompressed_chunk_sizes[next_chunk_in_msg] = chunk_size;
memcpy(next_msg->uncompressed_chunks[next_chunk_in_msg],
chunk, chunk_size);
ctx->next_chunk++;
if (++next_chunk_in_msg == next_msg->num_chunks) {
- DEBUG2("Sending message %p", next_msg);
/* Send off an array of chunks to compress */
list_add_tail(&next_msg->list, &ctx->next_lte->msg_list);
shared_queue_put(ctx->res_to_compress_queue, next_msg);
{
struct main_writer_thread_ctx *ctx = _ctx;
int ret;
- DEBUG2("finishing");
while (ctx->num_outstanding_messages != 0) {
ret = receive_compressed_chunks(ctx);
if (ret)
wimlib_assert(list_empty(&ctx->outstanding_streams));
return do_write_stream_list_serial(&ctx->serial_streams,
ctx->lookup_table,
- ctx->out_fp,
+ ctx->out_fd,
ctx->out_ctype,
ctx->write_resource_flags,
ctx->progress_func,
{
int ret;
-#ifdef ENABLE_MORE_DEBUG
- DEBUG2("Submit for compression:");
- print_lookup_table_entry(lte, stderr);
-#endif
-
+ /* Read the entire stream @lte, feeding its data chunks to the
+ * compressor threads. Also SHA1-sum the stream; this is required in
+ * the case that @lte is unhashed, and a nice additional verification
+ * when @lte is already hashed. */
sha1_init(&ctx->next_sha_ctx);
ctx->next_chunk = 0;
ctx->next_num_chunks = wim_resource_chunks(lte);
!(ctx->write_resource_flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) &&
wimlib_get_compression_type(lte->wim) == ctx->out_ctype))
{
+ /* Stream is too small or isn't being compressed. Process it by
+ * the main thread when we have a chance. We can't necessarily
+ * process it right here, as the main thread could be in the
+ * middle of writing a different stream. */
list_add_tail(<e->write_streams_list, &ctx->serial_streams);
lte->deferred = 1;
ret = 0;
} else {
ret = submit_stream_for_compression(lte, ctx);
}
+ lte->no_progress = 1;
return ret;
}
static long
-get_default_num_threads()
+get_default_num_threads(void)
{
#ifdef __WIN32__
return win32_get_number_of_processors();
#endif
}
+/* Equivalent to write_stream_list_serial(), except this takes a @num_threads
+ * parameter and will perform compression using that many threads. Falls
+ * back to write_stream_list_serial() on certain errors, such as a failure to
+ * create the number of threads requested.
+ *
+ * High level description of the algorithm for writing compressed streams in
+ * parallel: We perform compression on chunks of size WIM_CHUNK_SIZE bytes
+ * rather than on full files. The currently executing thread becomes the main
+ * thread and is entirely in charge of reading the data to compress (which may
+ * be in any location understood by the resource code--- such as in an external
+ * file being captured, or in another WIM file from which an image is being
+ * exported) and actually writing the compressed data to the output file.
+ * Additional threads are "compressor threads" and all execute the
+ * compressor_thread_proc, where they repeatedly retrieve buffers of data from
+ * the main thread, compress them, and hand them back to the main thread.
+ *
+ * Certain streams, such as streams that do not need to be compressed (e.g.
+ * input compression type same as output compression type) or streams of very
+ * small size are placed in a list (main_writer_thread_ctx.serial_list) and
+ * handled entirely by the main thread at an appropriate time.
+ *
+ * At any given point in time, multiple streams may be having chunks compressed
+ * concurrently. The stream that the main thread is currently *reading* may be
+ * later in the list that the stream that the main thread is currently
+ * *writing*.
+ */
static int
write_stream_list_parallel(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
- FILE *out_fp,
+ int out_fd,
int out_ctype,
int write_resource_flags,
wimlib_progress_func_t progress_func,
if (nthreads < 1 || nthreads > UINT_MAX) {
WARNING("Could not determine number of processors! Assuming 1");
goto out_serial;
+ } else if (nthreads == 1) {
+ goto out_serial_quiet;
} else {
num_threads = nthreads;
}
}
+ DEBUG("Writing stream list (parallel version, num_threads=%u)",
+ num_threads);
+
progress->write_streams.num_threads = num_threads;
static const size_t MESSAGES_PER_THREAD = 2;
struct main_writer_thread_ctx ctx;
ctx.stream_list = stream_list;
ctx.lookup_table = lookup_table;
- ctx.out_fp = out_fp;
+ ctx.out_fd = out_fd;
ctx.out_ctype = out_ctype;
ctx.res_to_compress_queue = &res_to_compress_queue;
ctx.compressed_res_queue = &compressed_res_queue;
ctx.num_messages = queue_size;
- ctx.write_resource_flags = write_resource_flags | WIMLIB_RESOURCE_FLAG_THREADSAFE_READ;
+ ctx.write_resource_flags = write_resource_flags;
ctx.progress_func = progress_func;
ctx.progress = progress;
ret = main_writer_thread_init_ctx(&ctx);
goto out_join;
ret = do_write_stream_list(stream_list, lookup_table,
main_thread_process_next_stream,
- &ctx, NULL, NULL);
+ &ctx, progress_func, progress);
if (ret)
goto out_destroy_ctx;
+
+ /* The main thread has finished reading all streams that are going to be
+ * compressed in parallel, and it now needs to wait for all remaining
+ * chunks to be compressed so that the remaining streams can actually be
+ * written to the output file. Furthermore, any remaining streams that
+ * had processing deferred to the main thread need to be handled. These
+ * tasks are done by the main_writer_thread_finish() function. */
ret = main_writer_thread_finish(&ctx);
out_destroy_ctx:
main_writer_thread_destroy_ctx(&ctx);
return ret;
out_serial:
WARNING("Falling back to single-threaded compression");
+out_serial_quiet:
return write_stream_list_serial(stream_list,
lookup_table,
- out_fp,
+ out_fd,
out_ctype,
write_resource_flags,
progress_func,
#endif
/*
- * Write a list of streams to a WIM (@out_fp) using the compression type
+ * Write a list of streams to a WIM (@out_fd) using the compression type
* @out_ctype and up to @num_threads compressor threads.
*/
static int
write_stream_list(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
- FILE *out_fp, int out_ctype, int write_flags,
+ int out_fd, int out_ctype, int write_flags,
unsigned num_threads, wimlib_progress_func_t progress_func)
{
struct wim_lookup_table_entry *lte;
progress.write_streams._private = 0;
#ifdef ENABLE_MULTITHREADED_COMPRESSION
- if (total_compression_bytes >= 1000000 && num_threads != 1)
+ if (total_compression_bytes >= 2000000 && num_threads != 1)
ret = write_stream_list_parallel(stream_list,
lookup_table,
- out_fp,
+ out_fd,
out_ctype,
write_resource_flags,
progress_func,
#endif
ret = write_stream_list_serial(stream_list,
lookup_table,
- out_fp,
+ out_fd,
out_ctype,
write_resource_flags,
progress_func,
return ret;
}
-/* Writes the streams for the specified @image in @wim to @wim->out_fp.
+/* Writes the streams for the specified @image in @wim to @wim->out_fd.
*/
static int
write_wim_streams(WIMStruct *wim, int image, int write_flags,
return ret;
return write_stream_list(&stream_list,
wim->lookup_table,
- wim->out_fp,
+ wim->out_fd,
wimlib_get_compression_type(wim),
write_flags,
num_threads,
* (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.
*
{
int ret;
struct wim_header hdr;
- FILE *out = w->out_fp;
/* @hdr will be the header for the new WIM. First copy all the data
* from the header in the WIMStruct; then set all the fields that may
goto out_close_wim;
}
- ret = write_xml_data(w->wim_info, image, out,
+ ret = write_xml_data(w->wim_info, image, w->out_fd,
(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE) ?
wim_info_get_total_bytes(w->wim_info) : 0,
&hdr.xml_res_entry);
struct wim_header checkpoint_hdr;
memcpy(&checkpoint_hdr, &hdr, sizeof(struct wim_header));
zero_resource_entry(&checkpoint_hdr.integrity);
- if (fseeko(out, 0, SEEK_SET)) {
- ERROR_WITH_ERRNO("Failed to seek to beginning "
- "of WIM being written");
- ret = WIMLIB_ERR_WRITE;
- goto out_close_wim;
- }
- ret = write_header(&checkpoint_hdr, out);
+ ret = write_header(&checkpoint_hdr, w->out_fd);
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_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_close_wim;
- }
}
off_t old_lookup_table_end;
new_lookup_table_end = hdr.lookup_table_res_entry.offset +
hdr.lookup_table_res_entry.size;
- ret = write_integrity_table(out,
+ ret = write_integrity_table(w->out_fd,
&hdr.integrity,
new_lookup_table_end,
old_lookup_table_end,
zero_resource_entry(&hdr.integrity);
}
- 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_close_wim;
- }
-
- ret = write_header(&hdr, out);
+ ret = write_header(&hdr, w->out_fd);
if (ret)
goto out_close_wim;
if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
- if (fflush(out) != 0
- || fsync(fileno(out)) != 0)
- {
- ERROR_WITH_ERRNO("Error flushing data to WIM file");
+ if (fsync(w->out_fd)) {
+ ERROR_WITH_ERRNO("Error syncing data to WIM file");
ret = WIMLIB_ERR_WRITE;
}
}
out_close_wim:
- if (fclose(out) != 0) {
+ if (close(w->out_fd)) {
ERROR_WITH_ERRNO("Failed to close the output WIM file");
if (ret == 0)
ret = WIMLIB_ERR_WRITE;
}
- w->out_fp = NULL;
+ w->out_fd = -1;
return ret;
}
#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
int
-lock_wim(WIMStruct *w, FILE *fp)
+lock_wim(WIMStruct *w, int fd)
{
int ret = 0;
- if (fp && !w->wim_locked) {
- ret = flock(fileno(fp), LOCK_EX | LOCK_NB);
+ if (fd != -1 && !w->wim_locked) {
+ ret = flock(fd, LOCK_EX | LOCK_NB);
if (ret != 0) {
if (errno == EWOULDBLOCK) {
ERROR("`%"TS"' is already being modified or has been "
#endif
static int
-open_wim_writable(WIMStruct *w, const tchar *path,
- bool trunc, bool also_readable)
+open_wim_writable(WIMStruct *w, const tchar *path, int open_flags)
{
- const tchar *mode;
- if (trunc)
- if (also_readable)
- mode = T("w+b");
- else
- mode = T("wb");
- else
- mode = T("r+b");
-
- wimlib_assert(w->out_fp == NULL);
- w->out_fp = tfopen(path, mode);
- if (w->out_fp) {
- return 0;
- } else {
+ w->out_fd = topen(path, open_flags | O_BINARY, 0644);
+ if (w->out_fd == -1) {
ERROR_WITH_ERRNO("Failed to open `%"TS"' for writing", path);
return WIMLIB_ERR_OPEN;
}
+ return 0;
}
void
close_wim_writable(WIMStruct *w)
{
- if (w->out_fp) {
- if (fclose(w->out_fp) != 0) {
+ if (w->out_fd != -1) {
+ if (close(w->out_fd))
WARNING_WITH_ERRNO("Failed to close output WIM");
- }
- w->out_fp = NULL;
+ w->out_fd = -1;
}
}
begin_write(WIMStruct *w, const tchar *path, int write_flags)
{
int ret;
- ret = open_wim_writable(w, path, true,
- (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
+ int open_flags = O_TRUNC | O_CREAT;
+ if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
+ open_flags |= O_RDWR;
+ else
+ open_flags |= O_WRONLY;
+ ret = open_wim_writable(w, path, open_flags);
if (ret)
return ret;
/* Write dummy header. It will be overwritten later. */
- return write_header(&w->hdr, w->out_fp);
+ ret = write_header(&w->hdr, w->out_fd);
+ if (ret)
+ return ret;
+ if (lseek(w->out_fd, WIM_HEADER_DISK_SIZE, SEEK_SET) == -1) {
+ ERROR_WITH_ERRNO("Failed to seek to end of WIM header");
+ return WIMLIB_ERR_WRITE;
+ }
+ return 0;
}
/* Writes a stand-alone WIM to a file. */
struct list_head stream_list;
off_t old_wim_end;
u64 old_lookup_table_end, old_xml_begin, old_xml_end;
+ int open_flags;
DEBUG("Overwriting `%"TS"' in-place", w->filename);
if (ret)
return ret;
- ret = open_wim_writable(w, w->filename, false,
- (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) != 0);
+ open_flags = 0;
+ if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
+ open_flags |= O_RDWR;
+ else
+ open_flags |= O_WRONLY;
+ ret = open_wim_writable(w, w->filename, open_flags);
if (ret)
return ret;
- ret = lock_wim(w, w->out_fp);
+ ret = lock_wim(w, w->out_fd);
if (ret) {
close_wim_writable(w);
return ret;
}
- if (fseeko(w->out_fp, old_wim_end, SEEK_SET) != 0) {
+ if (lseek(w->out_fd, old_wim_end, SEEK_SET) == -1) {
ERROR_WITH_ERRNO("Can't seek to end of WIM");
close_wim_writable(w);
w->wim_locked = 0;
old_wim_end);
ret = write_stream_list(&stream_list,
w->lookup_table,
- w->out_fp,
+ w->out_fd,
wimlib_get_compression_type(w),
write_flags,
num_threads,
goto out_unlink;
}
- DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename);
-
-#ifdef __WIN32__
- /* Windows won't let you delete open files unless FILE_SHARE_DELETE was
- * specified to CreateFile(). The WIM was opened with fopen(), which
- * didn't provided this flag to CreateFile, so the handle must be closed
- * before executing the rename(). */
- if (w->fp != NULL) {
- fclose(w->fp);
- w->fp = NULL;
- }
-#endif
+ close_wim(w);
+ DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, w->filename);
/* Rename the new file to the old file .*/
if (trename(tmpfile, w->filename) != 0) {
ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
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 = tfopen(w->filename, T("rb"));
- if (w->fp == NULL) {
- ret = WIMLIB_ERR_REOPEN;
- WARNING_WITH_ERRNO("Failed to re-open `%"TS"' read-only",
- w->filename);
- FREE(w->filename);
- w->filename = NULL;
- }
goto out;
out_unlink:
/* Remove temporary file. */