# include <pthread.h>
#endif
-#include <unistd.h>
-#include <fcntl.h>
#include <errno.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <stdlib.h>
+#include <unistd.h>
#ifdef HAVE_ALLOCA_H
# include <alloca.h>
-#else
-# include <stdlib.h>
#endif
-#include <limits.h>
#ifndef __WIN32__
# include <sys/uio.h> /* for `struct iovec' */
#endif
+/* Return true if the specified resource is compressed and the compressed data
+ * can be reused with the specified output parameters. */
+static bool
+can_raw_copy(const struct wim_lookup_table_entry *lte,
+ int write_resource_flags, int out_ctype, u32 out_chunk_size)
+{
+ return (out_ctype == wim_resource_compression_type(lte)
+ && out_chunk_size == wim_resource_chunk_size(lte)
+ && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE);
+}
+
+
+/* Return true if the specified resource must be recompressed when the specified
+ * output parameters are used. */
+static bool
+must_compress_stream(const struct wim_lookup_table_entry *lte,
+ int write_resource_flags, int out_ctype, u32 out_chunk_size)
+{
+ return (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
+ && ((write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS)
+ || !can_raw_copy(lte, write_resource_flags,
+ out_ctype, out_chunk_size)));
+}
+
+static unsigned
+compress_chunk(const void * uncompressed_data,
+ unsigned uncompressed_len,
+ void *compressed_data,
+ int out_ctype,
+ struct wimlib_lzx_context *comp_ctx)
+{
+ switch (out_ctype) {
+ case WIMLIB_COMPRESSION_TYPE_XPRESS:
+ return wimlib_xpress_compress(uncompressed_data,
+ uncompressed_len,
+ compressed_data);
+ case WIMLIB_COMPRESSION_TYPE_LZX:
+ return wimlib_lzx_compress2(uncompressed_data,
+ uncompressed_len,
+ compressed_data,
+ comp_ctx);
+ default:
+ wimlib_assert(0);
+ return 0;
+ }
+}
+
/* Chunk table that's located at the beginning of each compressed resource in
* the WIM. (This is not the on-disk format; the on-disk format just has an
* array of offsets.) */
static int
begin_wim_resource_chunk_tab(const struct wim_lookup_table_entry *lte,
struct filedes *out_fd,
+ u32 out_chunk_size,
struct chunk_table **chunk_tab_ret,
int resource_flags)
{
int ret;
size = wim_resource_size(lte);
- num_chunks = wim_resource_chunks(lte);
+ num_chunks = DIV_ROUND_UP(size, out_chunk_size);
bytes_per_chunk_entry = (size > (1ULL << 32)) ? 8 : 4;
alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
chunk_tab = CALLOC(1, alloc_size);
}
}
-/*
- * compress_func_t- Pointer to a function to compresses a chunk
- * of a WIM resource. This may be either
- * wimlib_xpress_compress() (xpress-compress.c) or
- * wimlib_lzx_compress() (lzx-compress.c).
- *
- * @chunk: Uncompressed data of the chunk.
- * @chunk_size: Size of the uncompressed chunk, in bytes.
- * @out: Pointer to output buffer of size at least (@chunk_size - 1) bytes.
- *
- * Returns the size of the compressed data written to @out in bytes, or 0 if the
- * data could not be compressed to (@chunk_size - 1) bytes or fewer.
- *
- * As a special requirement, the compression code is optimized for the WIM
- * format and therefore requires (@chunk_size <= 32768).
- *
- * As another special requirement, the compression code will read up to 8 bytes
- * off the end of the @chunk array for performance reasons. The values of these
- * bytes will not affect the output of the compression, but the calling code
- * must make sure that the buffer holding the uncompressed chunk is actually at
- * least (@chunk_size + 8) bytes, or at least that these extra bytes are in
- * mapped memory that will not cause a memory access violation if accessed.
- */
-typedef unsigned (*compress_func_t)(const void *chunk, unsigned chunk_size,
- void *out);
-
-static compress_func_t
-get_compress_func(int out_ctype)
-{
- if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
- return wimlib_lzx_compress;
- else
- return wimlib_xpress_compress;
-}
-
/* Finishes a WIM chunk table and writes it to the output file at the correct
* offset. */
static int
chunk_tab->bytes_per_chunk_entry,
chunk_tab->table_disk_size);
} else {
- ret = full_pwrite(out_fd,
- chunk_tab->offsets +
- chunk_tab->bytes_per_chunk_entry,
- chunk_tab->table_disk_size,
- res_start_offset);
- }
- if (ret) {
- ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
- "file resource");
+ ret = full_pwrite(out_fd,
+ chunk_tab->offsets +
+ chunk_tab->bytes_per_chunk_entry,
+ chunk_tab->table_disk_size,
+ res_start_offset);
}
+ if (ret)
+ ERROR_WITH_ERRNO("Write error");
return ret;
}
}
struct write_resource_ctx {
- compress_func_t compress;
+ int out_ctype;
+ u32 out_chunk_size;
+ struct wimlib_lzx_context *comp_ctx;
struct chunk_table *chunk_tab;
struct filedes *out_fd;
SHA_CTX sha_ctx;
const void *out_chunk;
unsigned out_chunk_size;
int ret;
+ void *compressed_chunk = NULL;
+ unsigned compressed_size;
+ bool compressed_chunk_malloced = false;
+ size_t stack_max = 32768;
if (ctx->doing_sha)
sha1_update(&ctx->sha_ctx, chunk, chunk_size);
out_chunk = chunk;
out_chunk_size = chunk_size;
- if (ctx->compress) {
- void *compressed_chunk;
- unsigned compressed_size;
+ if (ctx->out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
/* Compress the chunk. */
- compressed_chunk = alloca(chunk_size);
- compressed_size = (*ctx->compress)(chunk, chunk_size,
- compressed_chunk);
+ if (chunk_size <= stack_max) {
+ compressed_chunk = alloca(chunk_size);
+ } else {
+ compressed_chunk = MALLOC(chunk_size);
+ if (compressed_chunk == NULL)
+ return WIMLIB_ERR_NOMEM;
+ compressed_chunk_malloced = true;
+ }
+ compressed_size = compress_chunk(chunk, chunk_size,
+ compressed_chunk,
+ ctx->out_ctype,
+ ctx->comp_ctx);
/* Use compressed data if compression to less than input size
* was successful. */
if (compressed_size) {
ret = full_write(ctx->out_fd, out_chunk, out_chunk_size);
if (ret)
goto error;
- return 0;
+
+out_free_memory:
+ if (compressed_chunk_malloced)
+ FREE(compressed_chunk);
+ return ret;
error:
ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
- return ret;
+ goto out_free_memory;
}
/*
* One of the WIMLIB_COMPRESSION_TYPE_* constants to indicate which
* compression algorithm to use.
*
+ * @out_chunk_size:
+ * Compressed chunk size 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.
*
- * @write_resource_flags:
+ * @resource_flags:
* * WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS to force data to be recompressed even
* if it could otherwise be copied directly from the input;
* * WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE if writing a resource for a pipable WIM
* (and the output file descriptor may be a pipe).
*
+ * @comp_ctx:
+ * Location of LZX compression context pointer, which will be allocated or
+ * updated if needed. (Initialize to NULL.)
+ *
* Additional notes: The SHA1 message digest of the uncompressed data is
* calculated (except when doing a raw copy --- see below). If the @unhashed
* 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.
+ * this function will fail if they do not match.
*/
-int
+static int
write_wim_resource(struct wim_lookup_table_entry *lte,
struct filedes *out_fd, int out_ctype,
+ u32 out_chunk_size,
struct resource_entry *out_res_entry,
- int resource_flags)
+ int resource_flags,
+ struct wimlib_lzx_context **comp_ctx)
{
struct write_resource_ctx write_ctx;
off_t res_start_offset;
+ u32 in_chunk_size;
u64 read_size;
int ret;
res_start_offset = out_fd->offset;
/* If we are not forcing the data to be recompressed, and the input
- * resource is located in a WIM with the same compression type as that
- * desired other than no compression, we can simply copy the compressed
- * data without recompressing it. This also means we must skip
- * calculating the SHA1, as we never will see the uncompressed data. */
- if (lte->resource_location == RESOURCE_IN_WIM &&
- out_ctype == wim_resource_compression_type(lte) &&
- out_ctype != WIMLIB_COMPRESSION_TYPE_NONE &&
- !(resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS))
- {
- /* Normally we can request a RAW_FULL read, but if we're reading
- * from a pipable resource and writing a non-pipable resource or
- * vice versa, then a RAW_CHUNKS read needs to be requested so
- * that the written resource can be appropriately formatted.
- * However, in neither case is any actual decompression needed.
- */
+ * resource is located in a WIM with a compression mode compatible with
+ * the output, 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 (can_raw_copy(lte, resource_flags, out_ctype, out_chunk_size)) {
+ /* Normally, for raw copies we can request a RAW_FULL read, but
+ * if we're reading from a pipable resource and writing a
+ * non-pipable resource or vice versa, then a RAW_CHUNKS read
+ * needs to be requested so that the written resource can be
+ * appropriately formatted. However, in neither case is any
+ * actual decompression needed. */
if (lte->is_pipable == !!(resource_flags &
WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE))
+ {
resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_FULL;
- else
+ read_size = lte->resource_entry.size;
+ } else {
resource_flags |= WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS;
+ read_size = lte->resource_entry.original_size;
+ }
write_ctx.doing_sha = false;
- read_size = lte->resource_entry.size;
} else {
write_ctx.doing_sha = true;
sha1_init(&write_ctx.sha_ctx);
read_size = lte->resource_entry.original_size;
}
- /* If the output resource is to be compressed, initialize the chunk
- * table and set the function to use for chunk compression. Exceptions:
- * no compression function is needed if doing a raw copy; also, no chunk
- * table is needed if doing a *full* (not per-chunk) raw copy. */
- write_ctx.compress = NULL;
+ /* Set the output compression mode and initialize chunk table if needed.
+ */
+ write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
+ write_ctx.out_chunk_size = out_chunk_size;
write_ctx.chunk_tab = NULL;
if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
- if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW))
- write_ctx.compress = get_compress_func(out_ctype);
+ wimlib_assert(out_chunk_size > 0);
+ if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW)) {
+ /* Compression needed. */
+ write_ctx.out_ctype = out_ctype;
+ if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) {
+ ret = wimlib_lzx_alloc_context(out_chunk_size,
+ NULL, comp_ctx);
+ if (ret)
+ goto out;
+ }
+ write_ctx.comp_ctx = *comp_ctx;
+ }
if (!(resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)) {
+ /* Chunk table needed. */
ret = begin_wim_resource_chunk_tab(lte, out_fd,
+ out_chunk_size,
&write_ctx.chunk_tab,
resource_flags);
if (ret)
}
/* Write the entire resource by reading the entire resource and feeding
- * the data through the write_resource_cb function. */
+ * the data through write_resource_cb(). */
write_ctx.out_fd = out_fd;
write_ctx.resource_flags = resource_flags;
try_write_again:
+ if (write_ctx.out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
+ in_chunk_size = wim_resource_chunk_size(lte);
+ else
+ in_chunk_size = out_chunk_size;
ret = read_resource_prefix(lte, read_size,
- write_resource_cb, &write_ctx, resource_flags);
+ write_resource_cb,
+ in_chunk_size, &write_ctx, resource_flags);
if (ret)
goto out_free_chunk_tab;
goto out_free_chunk_tab;
out_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
FREE(write_ctx.chunk_tab);
- write_ctx.compress = NULL;
+ write_ctx.out_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
write_ctx.chunk_tab = NULL;
write_ctx.doing_sha = false;
goto try_write_again;
}
- if (resource_flags & (WIMLIB_READ_RESOURCE_FLAG_RAW)) {
+ if (resource_flags & WIMLIB_READ_RESOURCE_FLAG_RAW) {
DEBUG("Copied raw compressed data "
"(%"PRIu64" => %"PRIu64" bytes @ +%"PRIu64", flags=0x%02x)",
out_res_entry->original_size, out_res_entry->size,
}
/* Like write_wim_resource(), but the resource is specified by a buffer of
- * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
- * the buffer to @hash_ret. */
+ * uncompressed data rather a lookup table entry. Also writes the SHA1 message
+ * digest of the buffer to @hash_ret if it is non-NULL. */
int
write_wim_resource_from_buffer(const void *buf, size_t buf_size,
int reshdr_flags, struct filedes *out_fd,
int out_ctype,
+ u32 out_chunk_size,
struct resource_entry *out_res_entry,
- u8 *hash_ret, int write_resource_flags)
+ u8 *hash_ret, int write_resource_flags,
+ struct wimlib_lzx_context **comp_ctx)
{
- /* Set up a temporary lookup table entry to provide to
- * write_wim_resource(). */
- struct wim_lookup_table_entry lte;
int ret;
+ struct wim_lookup_table_entry *lte;
+
+ /* Set up a temporary lookup table entry to provide to
+ * write_wim_resource(). */
+
+ lte = new_lookup_table_entry();
+ if (lte == NULL)
+ return WIMLIB_ERR_NOMEM;
- lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
- lte.attached_buffer = (void*)buf;
- lte.resource_entry.original_size = buf_size;
- lte.resource_entry.flags = reshdr_flags;
+ lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
+ lte->attached_buffer = (void*)buf;
+ lte->resource_entry.original_size = buf_size;
+ lte->resource_entry.flags = reshdr_flags;
if (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE) {
- sha1_buffer(buf, buf_size, lte.hash);
- lte.unhashed = 0;
+ sha1_buffer(buf, buf_size, lte->hash);
+ lte->unhashed = 0;
} else {
- lte.unhashed = 1;
+ lte->unhashed = 1;
}
- ret = write_wim_resource(<e, out_fd, out_ctype, out_res_entry,
- write_resource_flags);
+ ret = write_wim_resource(lte, out_fd, out_ctype, out_chunk_size,
+ out_res_entry, write_resource_flags, comp_ctx);
if (ret)
- return ret;
+ goto out_free_lte;
if (hash_ret)
- copy_hash(hash_ret, lte.hash);
- return 0;
+ copy_hash(hash_ret, lte->hash);
+ ret = 0;
+out_free_lte:
+ lte->resource_location = RESOURCE_NONEXISTENT;
+ free_lookup_table_entry(lte);
+ return ret;
}
struct compressor_thread_params {
struct shared_queue *res_to_compress_queue;
struct shared_queue *compressed_res_queue;
- compress_func_t compress;
+ int out_ctype;
+ struct wimlib_lzx_context *comp_ctx;
};
#define MAX_CHUNKS_PER_MSG 2
struct message {
struct wim_lookup_table_entry *lte;
+ u32 out_chunk_size;
u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
struct iovec out_chunks[MAX_CHUNKS_PER_MSG];
- size_t total_out_bytes;
unsigned num_chunks;
struct list_head list;
bool complete;
};
static void
-compress_chunks(struct message *msg, compress_func_t compress)
+compress_chunks(struct message *msg, int out_ctype,
+ struct wimlib_lzx_context *comp_ctx)
{
- msg->total_out_bytes = 0;
for (unsigned i = 0; i < msg->num_chunks; i++) {
- unsigned len = compress(msg->uncompressed_chunks[i],
- msg->uncompressed_chunk_sizes[i],
- msg->compressed_chunks[i]);
+ unsigned len;
+
+ len = compress_chunk(msg->uncompressed_chunks[i],
+ msg->uncompressed_chunk_sizes[i],
+ msg->compressed_chunks[i],
+ out_ctype,
+ comp_ctx);
+
void *out_chunk;
unsigned out_len;
if (len) {
}
msg->out_chunks[i].iov_base = out_chunk;
msg->out_chunks[i].iov_len = out_len;
- msg->total_out_bytes += out_len;
}
}
struct compressor_thread_params *params = arg;
struct shared_queue *res_to_compress_queue = params->res_to_compress_queue;
struct shared_queue *compressed_res_queue = params->compressed_res_queue;
- compress_func_t compress = params->compress;
struct message *msg;
DEBUG("Compressor thread ready");
while ((msg = shared_queue_get(res_to_compress_queue)) != NULL) {
- compress_chunks(msg, compress);
+ compress_chunks(msg, params->out_ctype, params->comp_ctx);
shared_queue_put(compressed_res_queue, msg);
}
DEBUG("Compressor thread terminating");
}
#endif /* ENABLE_MULTITHREADED_COMPRESSION */
+struct write_streams_progress_data {
+ wimlib_progress_func_t progress_func;
+ union wimlib_progress_info progress;
+ uint64_t next_progress;
+ WIMStruct *prev_wim_part;
+};
+
static void
-do_write_streams_progress(union wimlib_progress_info *progress,
- wimlib_progress_func_t progress_func,
- uint64_t size_added,
+do_write_streams_progress(struct write_streams_progress_data *progress_data,
+ struct wim_lookup_table_entry *lte,
bool stream_discarded)
{
+ union wimlib_progress_info *progress = &progress_data->progress;
+ bool new_wim_part;
+
if (stream_discarded) {
- progress->write_streams.total_bytes -= size_added;
- if (progress->write_streams._private != ~(uint64_t)0 &&
- progress->write_streams._private > progress->write_streams.total_bytes)
+ progress->write_streams.total_bytes -= wim_resource_size(lte);
+ if (progress_data->next_progress != ~(uint64_t)0 &&
+ progress_data->next_progress > progress->write_streams.total_bytes)
{
- progress->write_streams._private = progress->write_streams.total_bytes;
+ progress_data->next_progress = progress->write_streams.total_bytes;
}
} else {
- progress->write_streams.completed_bytes += size_added;
+ progress->write_streams.completed_bytes += wim_resource_size(lte);
+ }
+ new_wim_part = false;
+ if (lte->resource_location == RESOURCE_IN_WIM &&
+ lte->wim != progress_data->prev_wim_part)
+ {
+ if (progress_data->prev_wim_part) {
+ new_wim_part = true;
+ progress->write_streams.completed_parts++;
+ }
+ progress_data->prev_wim_part = lte->wim;
}
progress->write_streams.completed_streams++;
- if (progress_func &&
- progress->write_streams.completed_bytes >= progress->write_streams._private)
+ if (progress_data->progress_func
+ && (progress->write_streams.completed_bytes >= progress_data->next_progress
+ || new_wim_part))
{
- progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
- progress);
- if (progress->write_streams._private == progress->write_streams.total_bytes) {
- progress->write_streams._private = ~(uint64_t)0;
+ progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
+ progress);
+ if (progress_data->next_progress == progress->write_streams.total_bytes) {
+ progress_data->next_progress = ~(uint64_t)0;
} else {
- progress->write_streams._private =
+ progress_data->next_progress =
min(progress->write_streams.total_bytes,
progress->write_streams.completed_bytes +
progress->write_streams.total_bytes / 100);
struct serial_write_stream_ctx {
struct filedes *out_fd;
int out_ctype;
+ u32 out_chunk_size;
+ struct wimlib_lzx_context **comp_ctx;
int write_resource_flags;
};
{
struct serial_write_stream_ctx *ctx = _ctx;
return write_wim_resource(lte, ctx->out_fd,
- ctx->out_ctype, <e->output_resource_entry,
- ctx->write_resource_flags);
+ ctx->out_ctype,
+ ctx->out_chunk_size,
+ <e->output_resource_entry,
+ ctx->write_resource_flags,
+ ctx->comp_ctx);
}
+
/* Write a list of streams, taking into account that some streams may be
* duplicates that are checksummed and discarded on the fly, and also delegating
* the actual writing of a stream to a function @write_stream_cb, which is
struct wim_lookup_table *lookup_table,
int (*write_stream_cb)(struct wim_lookup_table_entry *, void *),
void *write_stream_ctx,
- wimlib_progress_func_t progress_func,
- union wimlib_progress_info *progress)
+ struct write_streams_progress_data *progress_data)
{
int ret = 0;
struct wim_lookup_table_entry *lte;
* stream in the WIM we are writing. The stream must be
* checksummed to know if we need to write it or not. */
struct wim_lookup_table_entry *tmp;
- u32 orig_refcnt = lte->out_refcnt;
+ u32 orig_out_refcnt = lte->out_refcnt;
ret = hash_unhashed_stream(lte, lookup_table, &tmp);
if (ret)
break;
if (tmp != lte) {
+ /* We found a duplicate stream. 'lte' was
+ * freed, so replace it with the duplicate. */
lte = tmp;
- /* We found a duplicate stream. */
- if (orig_refcnt != tmp->out_refcnt) {
- /* We have already written, or are going
- * to write, the duplicate stream. So
- * just skip to the next stream. */
- DEBUG("Discarding duplicate stream of length %"PRIu64,
+
+ /* 'out_refcnt' was transferred to the
+ * duplicate, and we can detect if the duplicate
+ * stream was already referenced for writing by
+ * checking if its 'out_refcnt' is higher than
+ * that of the original stream. In such cases,
+ * the current stream can be discarded. We can
+ * also discard the current stream if it was
+ * previously marked as filtered (e.g. already
+ * present in the WIM being written). */
+ if (lte->out_refcnt > orig_out_refcnt ||
+ lte->filtered) {
+ DEBUG("Discarding duplicate stream of "
+ "length %"PRIu64,
wim_resource_size(lte));
lte->no_progress = 0;
stream_discarded = true;
if (ret)
break;
/* In parallel mode, some streams are deferred for later,
- * serialized processing; ignore them here. */
+ * serialized processing; ignore them here. */
if (lte->deferred)
continue;
if (lte->unhashed) {
}
skip_to_progress:
if (!lte->no_progress) {
- do_write_streams_progress(progress,
- progress_func,
- wim_resource_size(lte),
- stream_discarded);
+ do_write_streams_progress(progress_data,
+ lte, stream_discarded);
}
}
return ret;
struct wim_lookup_table *lookup_table,
struct filedes *out_fd,
int out_ctype,
+ u32 out_chunk_size,
+ struct wimlib_lzx_context **comp_ctx,
int write_resource_flags,
- wimlib_progress_func_t progress_func,
- union wimlib_progress_info *progress)
+ struct write_streams_progress_data *progress_data)
{
struct serial_write_stream_ctx ctx = {
.out_fd = out_fd,
.out_ctype = out_ctype,
+ .out_chunk_size = out_chunk_size,
.write_resource_flags = write_resource_flags,
+ .comp_ctx = comp_ctx,
};
return do_write_stream_list(stream_list,
lookup_table,
serial_write_stream,
&ctx,
- progress_func,
- progress);
+ progress_data);
}
static inline int
struct wim_lookup_table *lookup_table,
struct filedes *out_fd,
int out_ctype,
+ u32 out_chunk_size,
+ struct wimlib_lzx_context **comp_ctx,
int write_resource_flags,
- wimlib_progress_func_t progress_func,
- union wimlib_progress_info *progress)
+ struct write_streams_progress_data *progress_data)
{
+ union wimlib_progress_info *progress = &progress_data->progress;
DEBUG("Writing stream list of size %"PRIu64" (serial version)",
progress->write_streams.total_streams);
progress->write_streams.num_threads = 1;
- if (progress_func)
- progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
+ if (progress_data->progress_func) {
+ progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
+ progress);
+ }
return do_write_stream_list_serial(stream_list,
lookup_table,
out_fd,
out_ctype,
+ out_chunk_size,
+ comp_ctx,
write_resource_flags,
- progress_func,
- progress);
+ progress_data);
}
#ifdef ENABLE_MULTITHREADED_COMPRESSION
struct iovec *vecs;
struct pwm_chunk_hdr *chunk_hdrs;
unsigned nvecs;
- size_t nbytes;
int ret;
for (unsigned i = 0; i < msg->num_chunks; i++)
if (!(write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_PIPABLE)) {
nvecs = msg->num_chunks;
vecs = msg->out_chunks;
- nbytes = msg->total_out_bytes;
} else {
/* Special case: If writing a compressed resource to a pipable
* WIM, prefix each compressed chunk with a header that gives
vecs[i * 2 + 1].iov_base = msg->out_chunks[i].iov_base;
vecs[i * 2 + 1].iov_len = msg->out_chunks[i].iov_len;
}
- nbytes = msg->total_out_bytes + msg->num_chunks * sizeof(chunk_hdrs[0]);
}
ret = full_writev(out_fd, vecs, nvecs);
if (ret)
- ERROR_WITH_ERRNO("Failed to write WIM chunks");
+ ERROR_WITH_ERRNO("Write error");
return ret;
}
struct filedes *out_fd;
off_t res_start_offset;
int out_ctype;
+ u32 out_chunk_size;
+ struct wimlib_lzx_context **comp_ctx;
int write_resource_flags;
struct shared_queue *res_to_compress_queue;
struct shared_queue *compressed_res_queue;
size_t num_messages;
- wimlib_progress_func_t progress_func;
- union wimlib_progress_info *progress;
+ struct write_streams_progress_data *progress_data;
struct list_head available_msgs;
struct list_head outstanding_streams;
};
static int
-init_message(struct message *msg)
+init_message(struct message *msg, u32 out_chunk_size)
{
+ msg->out_chunk_size = out_chunk_size;
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);
+ msg->compressed_chunks[i] = MALLOC(out_chunk_size);
+ msg->uncompressed_chunks[i] = MALLOC(out_chunk_size);
if (msg->compressed_chunks[i] == NULL ||
msg->uncompressed_chunks[i] == NULL)
return WIMLIB_ERR_NOMEM;
}
static struct message *
-allocate_messages(size_t num_messages)
+allocate_messages(size_t num_messages, u32 out_chunk_size)
{
struct message *msgs;
msgs = CALLOC(num_messages, sizeof(struct message));
- if (!msgs)
+ if (msgs == NULL)
return NULL;
for (size_t i = 0; i < num_messages; i++) {
- if (init_message(&msgs[i])) {
+ if (init_message(&msgs[i], out_chunk_size)) {
free_messages(msgs, num_messages);
return NULL;
}
{
/* Pre-allocate all the buffers that will be needed to do the chunk
* compression. */
- ctx->msgs = allocate_messages(ctx->num_messages);
- if (!ctx->msgs)
+ ctx->msgs = allocate_messages(ctx->num_messages, ctx->out_chunk_size);
+ if (ctx->msgs == NULL)
return WIMLIB_ERR_NOMEM;
/* Initially, all the messages are available to use. */
INIT_LIST_HEAD(&ctx->outstanding_streams);
ctx->num_outstanding_messages = 0;
+ /* Message currently being prepared. */
ctx->next_msg = NULL;
/* Resources that don't need any chunks compressed are added to this
- * list and written directly by the main thread. */
+ * list and written directly by the main thread. */
INIT_LIST_HEAD(&ctx->serial_streams);
+ /* Pointer to chunk table for stream currently being written. */
ctx->cur_chunk_tab = NULL;
return 0;
/* Get the next message from the queue and process it.
* The message will contain 1 or more data chunks that have been
- * compressed. */
+ * compressed. */
msg = shared_queue_get(ctx->compressed_res_queue);
msg->complete = true;
--ctx->num_outstanding_messages;
- /* 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.
+ /* 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. */
+ * 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,
* it if needed. */
ret = begin_wim_resource_chunk_tab(cur_lte,
ctx->out_fd,
+ ctx->out_chunk_size,
&ctx->cur_chunk_tab,
ctx->write_resource_flags);
if (ret)
return ret;
-
}
- /* Write the compressed chunks from the message. */
+ /* Write the compressed chunks from the message. */
ret = write_wim_chunks(msg, ctx->out_fd, ctx->cur_chunk_tab,
ctx->write_resource_flags);
if (ret)
return ret;
- /* Was this the last chunk of the stream? If so, finish
- * it. */
+ /* Was this the last chunk of the stream? If so, finish the
+ * stream by writing the chunk table. */
if (list_empty(&cur_lte->msg_list) &&
msg->begin_chunk + msg->num_chunks == ctx->cur_chunk_tab->num_chunks)
{
ret = write_wim_resource(cur_lte,
ctx->out_fd,
WIMLIB_COMPRESSION_TYPE_NONE,
+ 0,
&cur_lte->output_resource_entry,
- ctx->write_resource_flags);
+ ctx->write_resource_flags,
+ ctx->comp_ctx);
if (ret)
return ret;
} else {
cur_lte->output_resource_entry.flags);
}
- do_write_streams_progress(ctx->progress,
- ctx->progress_func,
- wim_resource_size(cur_lte),
- false);
+ do_write_streams_progress(ctx->progress_data,
+ cur_lte, false);
/* Since we just finished writing a stream, write any
* streams that have been added to the serial_streams
* 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). */
+ * previous compression type). */
if (!list_empty(&ctx->serial_streams)) {
ret = do_write_stream_list_serial(&ctx->serial_streams,
ctx->lookup_table,
ctx->out_fd,
ctx->out_ctype,
+ ctx->out_chunk_size,
+ ctx->comp_ctx,
ctx->write_resource_flags,
- ctx->progress_func,
- ctx->progress);
+ ctx->progress_data);
if (ret)
return ret;
}
- /* Advance to the next stream to write. */
+ /* Advance to the next stream to write. */
if (list_empty(&ctx->outstanding_streams)) {
cur_lte = NULL;
} else {
return 0;
}
-/* Called when the main thread has read a new chunk of data. */
+/* 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)
{
ctx->lookup_table,
ctx->out_fd,
ctx->out_ctype,
+ ctx->out_chunk_size,
+ ctx->comp_ctx,
ctx->write_resource_flags,
- ctx->progress_func,
- ctx->progress);
+ ctx->progress_data);
}
static int
/* 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. */
+ * when @lte is already hashed. */
sha1_init(&ctx->next_sha_ctx);
ctx->next_chunk = 0;
- ctx->next_num_chunks = wim_resource_chunks(lte);
+ ctx->next_num_chunks = DIV_ROUND_UP(wim_resource_size(lte),
+ ctx->out_chunk_size);
ctx->next_lte = lte;
INIT_LIST_HEAD(<e->msg_list);
list_add_tail(<e->being_compressed_list, &ctx->outstanding_streams);
ret = read_resource_prefix(lte, wim_resource_size(lte),
- main_writer_thread_cb, ctx, 0);
+ main_writer_thread_cb,
+ ctx->out_chunk_size, ctx, 0);
if (ret)
return ret;
wimlib_assert(ctx->next_chunk == ctx->next_num_chunks);
int ret;
if (wim_resource_size(lte) < 1000 ||
- ctx->out_ctype == WIMLIB_COMPRESSION_TYPE_NONE ||
- (lte->resource_location == RESOURCE_IN_WIM &&
- !(ctx->write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS) &&
- lte->wim->compression_type == ctx->out_ctype))
+ !must_compress_stream(lte, ctx->write_resource_flags,
+ ctx->out_ctype, ctx->out_chunk_size))
{
/* 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. */
+ * middle of writing a different stream. */
list_add_tail(<e->write_streams_list, &ctx->serial_streams);
lte->deferred = 1;
ret = 0;
* 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.
+ * parallel: We perform compression on chunks 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
* 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*.
- */
+ * *writing*. */
static int
write_stream_list_parallel(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
struct filedes *out_fd,
int out_ctype,
+ u32 out_chunk_size,
+ struct wimlib_lzx_context **comp_ctx,
int write_resource_flags,
- wimlib_progress_func_t progress_func,
- union wimlib_progress_info *progress,
+ struct write_streams_progress_data *progress_data,
unsigned num_threads)
{
int ret;
struct shared_queue res_to_compress_queue;
struct shared_queue compressed_res_queue;
pthread_t *compressor_threads = NULL;
+ union wimlib_progress_info *progress = &progress_data->progress;
+ unsigned num_started_threads;
+ bool can_retry = true;
if (num_threads == 0) {
long nthreads = get_default_num_threads();
if (nthreads < 1 || nthreads > UINT_MAX) {
WARNING("Could not determine number of processors! Assuming 1");
- goto out_serial;
+ goto out_serial_quiet;
} else if (nthreads == 1) {
goto out_serial_quiet;
} else {
if (ret)
goto out_destroy_res_to_compress_queue;
- struct compressor_thread_params params;
- params.res_to_compress_queue = &res_to_compress_queue;
- params.compressed_res_queue = &compressed_res_queue;
- params.compress = get_compress_func(out_ctype);
+ struct compressor_thread_params *params;
- compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
- if (!compressor_threads) {
+ params = CALLOC(num_threads, sizeof(params[0]));
+ if (params == NULL) {
ret = WIMLIB_ERR_NOMEM;
goto out_destroy_compressed_res_queue;
}
+ for (unsigned i = 0; i < num_threads; i++) {
+ params[i].res_to_compress_queue = &res_to_compress_queue;
+ params[i].compressed_res_queue = &compressed_res_queue;
+ params[i].out_ctype = out_ctype;
+ if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX) {
+ ret = wimlib_lzx_alloc_context(out_chunk_size,
+ NULL, ¶ms[i].comp_ctx);
+ if (ret)
+ goto out_free_params;
+ }
+ }
+
+ compressor_threads = MALLOC(num_threads * sizeof(pthread_t));
+ if (compressor_threads == NULL) {
+ ret = WIMLIB_ERR_NOMEM;
+ goto out_free_params;
+ }
+
for (unsigned i = 0; i < num_threads; i++) {
DEBUG("pthread_create thread %u of %u", i + 1, num_threads);
ret = pthread_create(&compressor_threads[i], NULL,
- compressor_thread_proc, ¶ms);
- if (ret != 0) {
+ compressor_thread_proc, ¶ms[i]);
+ if (ret) {
+ errno = ret;
ret = -1;
ERROR_WITH_ERRNO("Failed to create compressor "
"thread %u of %u",
i + 1, num_threads);
- num_threads = i;
+ num_started_threads = i;
goto out_join;
}
}
+ num_started_threads = num_threads;
- if (progress_func)
- progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
+ if (progress_data->progress_func) {
+ progress_data->progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
+ progress);
+ }
struct main_writer_thread_ctx ctx;
+
+ memset(&ctx, 0, sizeof(ctx));
+
ctx.stream_list = stream_list;
ctx.lookup_table = lookup_table;
ctx.out_fd = out_fd;
ctx.out_ctype = out_ctype;
+ ctx.out_chunk_size = out_chunk_size;
+ ctx.comp_ctx = comp_ctx;
ctx.res_to_compress_queue = &res_to_compress_queue;
ctx.compressed_res_queue = &compressed_res_queue;
ctx.num_messages = queue_size;
ctx.write_resource_flags = write_resource_flags;
- ctx.progress_func = progress_func;
- ctx.progress = progress;
+ ctx.progress_data = progress_data;
ret = main_writer_thread_init_ctx(&ctx);
if (ret)
goto out_join;
+
+ can_retry = false;
ret = do_write_stream_list(stream_list, lookup_table,
main_thread_process_next_stream,
- &ctx, progress_func, progress);
+ &ctx, progress_data);
if (ret)
goto out_destroy_ctx;
* 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. */
+ * tasks are done by the main_writer_thread_finish() function. */
ret = main_writer_thread_finish(&ctx);
out_destroy_ctx:
main_writer_thread_destroy_ctx(&ctx);
out_join:
- for (unsigned i = 0; i < num_threads; i++)
+ for (unsigned i = 0; i < num_started_threads; i++)
shared_queue_put(&res_to_compress_queue, NULL);
- for (unsigned i = 0; i < num_threads; i++) {
+ for (unsigned i = 0; i < num_started_threads; i++) {
if (pthread_join(compressor_threads[i], NULL)) {
WARNING_WITH_ERRNO("Failed to join compressor "
"thread %u of %u",
}
}
FREE(compressor_threads);
+out_free_params:
+ for (unsigned i = 0; i < num_threads; i++)
+ wimlib_lzx_free_context(params[i].comp_ctx);
+ FREE(params);
out_destroy_compressed_res_queue:
shared_queue_destroy(&compressed_res_queue);
out_destroy_res_to_compress_queue:
shared_queue_destroy(&res_to_compress_queue);
- if (ret >= 0 && ret != WIMLIB_ERR_NOMEM)
+ if (!can_retry || (ret >= 0 && ret != WIMLIB_ERR_NOMEM))
return ret;
out_serial:
WARNING("Falling back to single-threaded compression");
lookup_table,
out_fd,
out_ctype,
+ out_chunk_size,
+ comp_ctx,
write_resource_flags,
- progress_func,
- progress);
+ progress_data);
}
#endif
-/*
- * Write a list of streams to a WIM (@out_fd) using the compression type
- * @out_ctype and up to @num_threads compressor threads.
- */
+/* Write a list of streams to a WIM (@out_fd) using the compression type
+ * @out_ctype, chunk size @out_chunk_size, and up to @num_threads compressor
+ * threads. */
static int
write_stream_list(struct list_head *stream_list,
struct wim_lookup_table *lookup_table,
- struct filedes *out_fd, int out_ctype, int write_flags,
+ struct filedes *out_fd, int out_ctype,
+ u32 out_chunk_size,
+ struct wimlib_lzx_context **comp_ctx,
+ int write_flags,
unsigned num_threads, wimlib_progress_func_t progress_func)
{
- struct wim_lookup_table_entry *lte;
- size_t num_streams = 0;
- u64 total_bytes = 0;
- u64 total_compression_bytes = 0;
- union wimlib_progress_info progress;
int ret;
int write_resource_flags;
+ u64 total_bytes;
+ u64 total_compression_bytes;
+ unsigned total_parts;
+ WIMStruct *prev_wim_part;
+ size_t num_streams;
+ struct wim_lookup_table_entry *lte;
+ struct write_streams_progress_data progress_data;
- if (list_empty(stream_list))
+ if (list_empty(stream_list)) {
+ DEBUG("No streams to write.");
return 0;
+ }
write_resource_flags = write_flags_to_resource_flags(write_flags);
- DEBUG("write_resource_flags=0x%08x", write_resource_flags);
+ DEBUG("Writing stream list (offset = %"PRIu64", write_resource_flags=0x%08x)",
+ out_fd->offset, write_resource_flags);
+
+ /* Sort the stream list into a good order for reading. */
+ ret = sort_stream_list_by_sequential_order(stream_list,
+ offsetof(struct wim_lookup_table_entry,
+ write_streams_list));
+ if (ret)
+ return ret;
/* Calculate the total size of the streams to be written. Note: this
* will be the uncompressed size, as we may not know the compressed size
* yet, and also this will assume that every unhashed stream will be
- * written (which will not necessarily be the case). */
+ * written (which will not necessarily be the case). */
+ total_bytes = 0;
+ total_compression_bytes = 0;
+ num_streams = 0;
+ total_parts = 0;
+ prev_wim_part = NULL;
list_for_each_entry(lte, stream_list, write_streams_list) {
num_streams++;
total_bytes += wim_resource_size(lte);
- if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
- && (wim_resource_compression_type(lte) != out_ctype ||
- (write_resource_flags & WIMLIB_WRITE_RESOURCE_FLAG_RECOMPRESS)))
- {
+ if (must_compress_stream(lte, write_resource_flags,
+ out_ctype, out_chunk_size))
total_compression_bytes += wim_resource_size(lte);
+ if (lte->resource_location == RESOURCE_IN_WIM) {
+ if (prev_wim_part != lte->wim) {
+ prev_wim_part = lte->wim;
+ total_parts++;
+ }
}
}
- progress.write_streams.total_bytes = total_bytes;
- progress.write_streams.total_streams = num_streams;
- progress.write_streams.completed_bytes = 0;
- progress.write_streams.completed_streams = 0;
- progress.write_streams.num_threads = num_threads;
- progress.write_streams.compression_type = out_ctype;
- progress.write_streams._private = 0;
+
+ memset(&progress_data, 0, sizeof(progress_data));
+ progress_data.progress_func = progress_func;
+
+ progress_data.progress.write_streams.total_bytes = total_bytes;
+ progress_data.progress.write_streams.total_streams = num_streams;
+ progress_data.progress.write_streams.completed_bytes = 0;
+ progress_data.progress.write_streams.completed_streams = 0;
+ progress_data.progress.write_streams.num_threads = num_threads;
+ progress_data.progress.write_streams.compression_type = out_ctype;
+ progress_data.progress.write_streams.total_parts = total_parts;
+ progress_data.progress.write_streams.completed_parts = 0;
+
+ progress_data.next_progress = 0;
+ progress_data.prev_wim_part = NULL;
#ifdef ENABLE_MULTITHREADED_COMPRESSION
if (total_compression_bytes >= 2000000 && num_threads != 1)
lookup_table,
out_fd,
out_ctype,
+ out_chunk_size,
+ comp_ctx,
write_resource_flags,
- progress_func,
- &progress,
+ &progress_data,
num_threads);
else
#endif
lookup_table,
out_fd,
out_ctype,
+ out_chunk_size,
+ comp_ctx,
write_resource_flags,
- progress_func,
- &progress);
+ &progress_data);
if (ret == 0)
DEBUG("Successfully wrote stream list.");
else
- DEBUG("Failed to write stream list.");
+ DEBUG("Failed to write stream list (ret=%d).", ret);
return ret;
}
return 0;
}
-
-struct lte_overwrite_prepare_args {
+struct find_streams_ctx {
WIMStruct *wim;
- off_t end_offset;
+ int write_flags;
struct list_head stream_list;
struct stream_size_table stream_size_tab;
};
-/* First phase of preparing streams for an in-place overwrite. This is called
- * on all streams, both hashed and unhashed, except the metadata resources. */
-static int
-lte_overwrite_prepare(struct wim_lookup_table_entry *lte, void *_args)
+static void
+lte_reference_for_logical_write(struct wim_lookup_table_entry *lte,
+ struct find_streams_ctx *ctx,
+ unsigned nref)
{
- struct lte_overwrite_prepare_args *args = _args;
-
- wimlib_assert(!(lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA));
- if (lte->resource_location != RESOURCE_IN_WIM || lte->wim != args->wim)
- list_add_tail(<e->write_streams_list, &args->stream_list);
- lte->out_refcnt = lte->refcnt;
- stream_size_table_insert(lte, &args->stream_size_tab);
- return 0;
+ if (lte->out_refcnt == 0) {
+ stream_size_table_insert(lte, &ctx->stream_size_tab);
+ list_add_tail(<e->write_streams_list, &ctx->stream_list);
+ }
+ lte->out_refcnt += nref;
}
-/* Second phase of preparing streams for an in-place overwrite. This is called
- * on existing metadata resources and hashed streams, but not unhashed streams.
- *
- * NOTE: lte->output_resource_entry is in union with lte->hash_list_2, so
- * lte_overwrite_prepare_2() must be called after lte_overwrite_prepare(), as
- * the latter uses lte->hash_list_2, while the former expects to set
- * lte->output_resource_entry. */
static int
-lte_overwrite_prepare_2(struct wim_lookup_table_entry *lte, void *_args)
+do_lte_full_reference_for_logical_write(struct wim_lookup_table_entry *lte,
+ void *_ctx)
{
- struct lte_overwrite_prepare_args *args = _args;
-
- if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == args->wim) {
- /* We can't do an in place overwrite on the WIM if there are
- * streams after the XML data. */
- if (lte->resource_entry.offset +
- lte->resource_entry.size > args->end_offset)
- {
- if (wimlib_print_errors) {
- ERROR("The following resource is after the XML data:");
- print_lookup_table_entry(lte, stderr);
- }
- return WIMLIB_ERR_RESOURCE_ORDER;
- }
- copy_resource_entry(<e->output_resource_entry,
- <e->resource_entry);
- }
+ struct find_streams_ctx *ctx = _ctx;
+ lte->out_refcnt = 0;
+ lte_reference_for_logical_write(lte, ctx,
+ (lte->refcnt ? lte->refcnt : 1));
return 0;
}
-/* Given a WIM that we are going to overwrite in place with zero or more
- * additional streams added, construct a list the list of new unique streams
- * ('struct wim_lookup_table_entry's) that must be written, plus any unhashed
- * streams that need to be added but may be identical to other hashed or
- * unhashed streams. These unhashed streams are checksummed while the streams
- * are being written. To aid this process, the member @unique_size is set to 1
- * on streams that have a unique size and therefore must be written.
- *
- * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to
- * indicate the number of times the stream is referenced in only the streams
- * that are being written; this may still be adjusted later when unhashed
- * streams are being resolved.
- */
static int
-prepare_streams_for_overwrite(WIMStruct *wim, off_t end_offset,
- struct list_head *stream_list)
-{
- int ret;
- struct lte_overwrite_prepare_args args;
- unsigned i;
-
- args.wim = wim;
- args.end_offset = end_offset;
- ret = init_stream_size_table(&args.stream_size_tab,
- wim->lookup_table->capacity);
- if (ret)
- return ret;
-
- INIT_LIST_HEAD(&args.stream_list);
- for (i = 0; i < wim->hdr.image_count; i++) {
- struct wim_image_metadata *imd;
- struct wim_lookup_table_entry *lte;
-
- imd = wim->image_metadata[i];
- image_for_each_unhashed_stream(lte, imd)
- lte_overwrite_prepare(lte, &args);
- }
- for_lookup_table_entry(wim->lookup_table, lte_overwrite_prepare, &args);
- list_transfer(&args.stream_list, stream_list);
-
- for (i = 0; i < wim->hdr.image_count; i++) {
- ret = lte_overwrite_prepare_2(wim->image_metadata[i]->metadata_lte,
- &args);
- if (ret)
- goto out_destroy_stream_size_table;
- }
- ret = for_lookup_table_entry(wim->lookup_table,
- lte_overwrite_prepare_2, &args);
-out_destroy_stream_size_table:
- destroy_stream_size_table(&args.stream_size_tab);
- return ret;
-}
-
-
-struct find_streams_ctx {
- struct list_head stream_list;
- struct stream_size_table stream_size_tab;
-};
-
-static void
inode_find_streams_to_write(struct wim_inode *inode,
struct wim_lookup_table *table,
- struct list_head *stream_list,
- struct stream_size_table *tab)
+ struct find_streams_ctx *ctx)
{
struct wim_lookup_table_entry *lte;
- for (unsigned i = 0; i <= inode->i_num_ads; i++) {
+ unsigned i;
+
+ for (i = 0; i <= inode->i_num_ads; i++) {
lte = inode_stream_lte(inode, i, table);
- if (lte) {
- if (lte->out_refcnt == 0) {
- if (lte->unhashed)
- stream_size_table_insert(lte, tab);
- list_add_tail(<e->write_streams_list, stream_list);
- }
- lte->out_refcnt += inode->i_nlink;
- }
+ if (lte)
+ lte_reference_for_logical_write(lte, ctx, inode->i_nlink);
+ else if (!is_zero_hash(inode_stream_hash(inode, i)))
+ return WIMLIB_ERR_RESOURCE_NOT_FOUND;
}
+ return 0;
}
static int
struct wim_image_metadata *imd;
struct wim_inode *inode;
struct wim_lookup_table_entry *lte;
+ int ret;
ctx = wim->private;
imd = wim_get_current_image_metadata(wim);
/* 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, wim->lookup_table,
- &ctx->stream_list,
- &ctx->stream_size_tab);
+ ret = inode_find_streams_to_write(inode, wim->lookup_table, ctx);
+ if (ret)
+ return ret;
}
return 0;
}
-/* Given a WIM that from which one or all of the images is being written, build
- * the list of unique streams ('struct wim_lookup_table_entry's) that must be
- * written, plus any unhashed streams that need to be written but may be
- * identical to other hashed or unhashed streams being written. These unhashed
- * streams are checksummed while the streams are being written. To aid this
- * process, the member @unique_size is set to 1 on streams that have a unique
- * size and therefore must be written.
+/*
+ * Build a list of streams (via `struct wim_lookup_table_entry's) included in
+ * the "logical write" of the WIM, meaning all streams that are referenced at
+ * least once by dentries in the the image(s) being written. 'out_refcnt' on
+ * each stream being included in the logical write is set to the number of
+ * references from dentries in the image(s). Furthermore, 'unique_size' on each
+ * stream being included in the logical write is set to indicate whether that
+ * stream has a unique size relative to the streams being included in the
+ * logical write. Still furthermore, 'part_number' on each stream being
+ * included in the logical write is set to the part number given in the
+ * in-memory header of @p wim.
*
- * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to
- * indicate the number of times the stream is referenced in only the streams
- * that are being written; this may still be adjusted later when unhashed
- * streams are being resolved.
+ * This is considered a "logical write" because it does not take into account
+ * filtering out streams already present in the WIM (in the case of an in place
+ * overwrite) or present in other WIMs (in case of creating delta WIM).
*/
static int
-prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list)
+prepare_logical_stream_list(WIMStruct *wim, int image, bool streams_ok,
+ struct find_streams_ctx *ctx)
{
int ret;
- struct find_streams_ctx ctx;
+ struct wim_lookup_table_entry *lte;
+
+ if (streams_ok && (image == WIMLIB_ALL_IMAGES ||
+ (image == 1 && wim->hdr.image_count == 1)))
+ {
+ /* Fast case: Assume that all streams are being written and
+ * that the reference counts are correct. */
+ struct wim_lookup_table_entry *lte;
+ struct wim_image_metadata *imd;
+ unsigned i;
+
+ for_lookup_table_entry(wim->lookup_table,
+ do_lte_full_reference_for_logical_write, ctx);
+ for (i = 0; i < wim->hdr.image_count; i++) {
+ imd = wim->image_metadata[i];
+ image_for_each_unhashed_stream(lte, imd)
+ do_lte_full_reference_for_logical_write(lte, ctx);
+ }
+ } else {
+ /* Slow case: Walk through the images being written and
+ * determine the streams referenced. */
+ for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL);
+ wim->private = ctx;
+ ret = for_image(wim, image, image_find_streams_to_write);
+ if (ret)
+ return ret;
+ }
+
+ list_for_each_entry(lte, &ctx->stream_list, write_streams_list)
+ lte->part_number = wim->hdr.part_number;
+ return 0;
+}
+
+static int
+process_filtered_stream(struct wim_lookup_table_entry *lte, void *_ctx)
+{
+ struct find_streams_ctx *ctx = _ctx;
+ u16 filtered = 0;
+
+ /* Calculate and set lte->filtered. */
+ if (lte->resource_location == RESOURCE_IN_WIM) {
+ if (lte->wim == ctx->wim &&
+ (ctx->write_flags & WIMLIB_WRITE_FLAG_OVERWRITE))
+ filtered |= FILTERED_SAME_WIM;
+ if (lte->wim != ctx->wim &&
+ (ctx->write_flags & WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS))
+ filtered |= FILTERED_EXTERNAL_WIM;
+ }
+ lte->filtered = filtered;
+
+ /* Filtered streams get inserted into the stream size table too, unless
+ * they already were. This is because streams that are checksummed
+ * on-the-fly during the write should not be written if they are
+ * duplicates of filtered stream. */
+ if (lte->filtered && lte->out_refcnt == 0)
+ stream_size_table_insert(lte, &ctx->stream_size_tab);
+ return 0;
+}
+
+static int
+mark_stream_not_filtered(struct wim_lookup_table_entry *lte, void *_ignore)
+{
+ lte->filtered = 0;
+ return 0;
+}
+
+/* Given the list of streams to include in a logical write of a WIM, handle
+ * filtering out streams already present in the WIM or already present in
+ * external WIMs, depending on the write flags provided. */
+static void
+handle_stream_filtering(struct find_streams_ctx *ctx)
+{
+ struct wim_lookup_table_entry *lte, *tmp;
+
+ if (!(ctx->write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_SKIP_EXTERNAL_WIMS)))
+ {
+ for_lookup_table_entry(ctx->wim->lookup_table,
+ mark_stream_not_filtered, ctx);
+ return;
+ }
+
+ for_lookup_table_entry(ctx->wim->lookup_table,
+ process_filtered_stream, ctx);
- DEBUG("Preparing list of streams to write for image %d.", image);
+ /* Streams in logical write list that were filtered can be removed. */
+ list_for_each_entry_safe(lte, tmp, &ctx->stream_list,
+ write_streams_list)
+ if (lte->filtered)
+ list_del(<e->write_streams_list);
+}
+
+/* Prepares list of streams to write for the specified WIM image(s). This wraps
+ * around prepare_logical_stream_list() to handle filtering out streams already
+ * present in the WIM or already present in external WIMs, depending on the
+ * write flags provided.
+ *
+ * Note: some additional data is stored in each `struct wim_lookup_table_entry':
+ *
+ * - 'out_refcnt' is set to the number of references found for the logical write.
+ * This will be nonzero on all streams in the list returned by this function,
+ * but will also be nonzero on streams not in the list that were included in
+ * the logical write list, but filtered out from the returned list.
+ * - 'filtered' is set to nonzero if the stream was filtered. Filtered streams
+ * are not included in the list of streams returned by this function.
+ * - 'unique_size' is set if the stream has a unique size among all streams in
+ * the logical write plus any filtered streams in the entire WIM that could
+ * potentially turn out to have the same checksum as a yet-to-be-checksummed
+ * stream being written.
+ */
+static int
+prepare_stream_list(WIMStruct *wim, int image, int write_flags,
+ struct list_head *stream_list)
+{
+ int ret;
+ bool streams_ok;
+ struct find_streams_ctx ctx;
- for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL);
+ INIT_LIST_HEAD(&ctx.stream_list);
ret = init_stream_size_table(&ctx.stream_size_tab,
wim->lookup_table->capacity);
if (ret)
return ret;
- for_lookup_table_entry(wim->lookup_table, stream_size_table_insert,
- &ctx.stream_size_tab);
- INIT_LIST_HEAD(&ctx.stream_list);
- wim->private = &ctx;
- ret = for_image(wim, image, image_find_streams_to_write);
- destroy_stream_size_table(&ctx.stream_size_tab);
+ ctx.write_flags = write_flags;
+ ctx.wim = wim;
+
+ streams_ok = ((write_flags & WIMLIB_WRITE_FLAG_STREAMS_OK) != 0);
+
+ ret = prepare_logical_stream_list(wim, image, streams_ok, &ctx);
if (ret)
- return ret;
+ goto out_destroy_table;
+
+ handle_stream_filtering(&ctx);
list_transfer(&ctx.stream_list, stream_list);
- return 0;
+ ret = 0;
+out_destroy_table:
+ destroy_stream_size_table(&ctx.stream_size_tab);
+ return ret;
}
-/* Writes the streams for the specified @image in @wim to @wim->out_fd.
- * Alternatively, if @stream_list_override is specified, it is taken to be the
- * list of streams to write (connected with 'write_streams_list') and @image is
- * ignored. */
static int
write_wim_streams(WIMStruct *wim, int image, int write_flags,
unsigned num_threads,
struct list_head *stream_list;
struct wim_lookup_table_entry *lte;
- if (stream_list_override) {
- stream_list = stream_list_override;
- list_for_each_entry(lte, stream_list, write_streams_list) {
- if (lte->refcnt)
- lte->out_refcnt = lte->refcnt;
- else
- lte->out_refcnt = 1;
- }
- } else {
+ if (stream_list_override == NULL) {
+ /* Normal case: prepare stream list from image(s) being written.
+ */
stream_list = &_stream_list;
- ret = prepare_stream_list(wim, image, stream_list);
+ ret = prepare_stream_list(wim, image, write_flags, stream_list);
if (ret)
return ret;
+ } else {
+ /* Currently only as a result of wimlib_split() being called:
+ * use stream list already explicitly provided. Use existing
+ * reference counts. */
+ stream_list = stream_list_override;
+ list_for_each_entry(lte, stream_list, write_streams_list) {
+ lte->out_refcnt = (lte->refcnt ? lte->refcnt : 1);
+ lte->part_number = wim->hdr.part_number;
+ }
}
- list_for_each_entry(lte, stream_list, write_streams_list)
- lte->part_number = wim->hdr.part_number;
+
return write_stream_list(stream_list,
wim->lookup_table,
&wim->out_fd,
- wim->compression_type,
+ wim->out_compression_type,
+ wim->out_chunk_size,
+ &wim->lzx_context,
write_flags,
num_threads,
progress_func);
int end_image;
int write_resource_flags;
- if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA)
+ if (write_flags & WIMLIB_WRITE_FLAG_NO_METADATA) {
+ DEBUG("Not writing any metadata resources.");
return 0;
+ }
write_resource_flags = write_flags_to_resource_flags(write_flags);
struct wim_image_metadata *imd;
imd = wim->image_metadata[i - 1];
+ /* Build a new metadata resource only if image was modified from
+ * the original (or was newly added). Otherwise just copy the
+ * existing one. */
if (imd->modified) {
+ DEBUG("Image %u was modified; building and writing new "
+ "metadata resource", i);
ret = write_metadata_resource(wim, i,
write_resource_flags);
+ } else if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE) {
+ DEBUG("Image %u was not modified; re-using existing "
+ "metadata resource.", i);
+ copy_resource_entry(&imd->metadata_lte->output_resource_entry,
+ &imd->metadata_lte->resource_entry);
+ ret = 0;
} else {
+ DEBUG("Image %u was not modified; copying existing "
+ "metadata resource.", i);
ret = write_wim_resource(imd->metadata_lte,
&wim->out_fd,
- wim->compression_type,
+ wim->out_compression_type,
+ wim->out_chunk_size,
&imd->metadata_lte->output_resource_entry,
- write_resource_flags);
+ write_resource_flags,
+ &wim->lzx_context);
}
if (ret)
return ret;
{
int ret = 0;
- if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR))
+ if (!(write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)) {
+ DEBUG("Closing WIM file.");
if (filedes_valid(&wim->out_fd))
if (filedes_close(&wim->out_fd))
ret = WIMLIB_ERR_WRITE;
+ }
filedes_invalidate(&wim->out_fd);
return ret;
}
/*
+ * finish_write():
+ *
* Finish writing a WIM file: write the lookup table, xml data, and integrity
- * table, then overwrite the WIM header. Always closes the WIM file descriptor
- * (wim->out_fd).
+ * table, then overwrite the WIM header. By default, closes the WIM file
+ * descriptor (@wim->out_fd) if successful.
*
* write_flags is a bitwise OR of the following:
*
* (public) WIMLIB_WRITE_FLAG_FSYNC:
* fsync() the output file before closing it.
*
- * (public) WIMLIB_WRITE_FLAG_PIPABLE:
+ * (public) WIMLIB_WRITE_FLAG_PIPABLE:
* Writing a pipable WIM, possibly to a pipe; include pipable WIM
* stream headers before the lookup table and XML data, and also
* write the WIM header at the end instead of seeking to the
* Instead of overwriting the WIM header at the beginning of the
* file, simply append it to the end of the file. (Used when
* writing to pipe.)
+ * (private) WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR:
+ * Do not close the file descriptor @wim->out_fd on either success
+ * on failure.
* (private) WIMLIB_WRITE_FLAG_USE_EXISTING_TOTALBYTES:
* Use the existing <TOTALBYTES> stored in the in-memory XML
* information, rather than setting it to the offset of the XML
off_t new_lookup_table_end;
u64 xml_totalbytes;
+ DEBUG("image=%d, write_flags=%08x", image, write_flags);
+
write_resource_flags = write_flags_to_resource_flags(write_flags);
/* In the WIM header, there is room for the resource entry for a
&wim->hdr.lookup_table_res_entry,
stream_list_override);
if (ret)
- goto out_close_wim;
+ return ret;
}
/* Write XML data. */
&wim->hdr.xml_res_entry,
write_resource_flags);
if (ret)
- goto out_close_wim;
+ return ret;
+ /* Write integrity table (optional). */
if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
if (write_flags & WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML) {
struct wim_header checkpoint_hdr;
ret = write_wim_header_at_offset(&checkpoint_hdr,
&wim->out_fd, 0);
if (ret)
- goto out_close_wim;
+ return ret;
}
if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
old_lookup_table_end,
progress_func);
if (ret)
- goto out_close_wim;
+ return ret;
} else {
+ /* No integrity table. */
zero_resource_entry(&wim->hdr.integrity);
}
+ /* Now that all information in the WIM header has been determined, the
+ * preliminary header written earlier can be overwritten, the header of
+ * the existing WIM file can be overwritten, or the final header can be
+ * written to the end of the pipable WIM. */
wim->hdr.flags &= ~WIM_HDR_FLAG_WRITE_IN_PROGRESS;
hdr_offset = 0;
if (write_flags & WIMLIB_WRITE_FLAG_HEADER_AT_END)
hdr_offset = wim->out_fd.offset;
+ DEBUG("Writing new header @ %"PRIu64".", hdr_offset);
ret = write_wim_header_at_offset(&wim->hdr, &wim->out_fd, hdr_offset);
if (ret)
- goto out_close_wim;
+ return ret;
+ /* Possibly sync file data to disk before closing. On POSIX systems, it
+ * is necessary to do this before using rename() to overwrite an
+ * existing file with a new file. Otherwise, data loss would occur if
+ * the system is abruptly terminated when the metadata for the rename
+ * operation has been written to disk, but the new file data has not.
+ */
if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
+ DEBUG("Syncing WIM file.");
if (fsync(wim->out_fd.fd)) {
ERROR_WITH_ERRNO("Error syncing data to WIM file");
- ret = WIMLIB_ERR_WRITE;
- goto out_close_wim;
+ return WIMLIB_ERR_WRITE;
}
}
- ret = 0;
-out_close_wim:
if (close_wim_writable(wim, write_flags)) {
- if (ret == 0) {
- ERROR_WITH_ERRNO("Failed to close the output WIM file");
- ret = WIMLIB_ERR_WRITE;
- }
+ ERROR_WITH_ERRNO("Failed to close the output WIM file");
+ return WIMLIB_ERR_WRITE;
}
- return ret;
+
+ return 0;
}
#if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
#endif
/*
+ * write_pipable_wim():
+ *
* Perform the intermediate stages of creating a "pipable" WIM (i.e. a WIM
- * capable of being applied from a pipe). Such a WIM looks like:
+ * capable of being applied from a pipe).
*
* Pipable WIMs are a wimlib-specific modification of the WIM format such that
* images can be applied from them sequentially when the file data is sent over
*
* - Magic characters in header are "WLPWM\0\0\0" (wimlib pipable WIM) instead
* of "MSWIM\0\0\0". This lets wimlib know that the WIM is pipable and also
- * should stop other software from trying to read the file as a normal WIM.
+ * stops other software from trying to read the file as a normal WIM.
*
* - The header at the beginning of the file does not contain all the normal
* information; in particular it will have all 0's for the lookup table and
* uncompressed stream size, and flags that indicate whether the stream is
* compressed. The data of uncompressed streams then follows literally, while
* the data of compressed streams follows in a modified format. Compressed
- * streams have no chunk table, since the chunk table cannot be written until
- * all chunks have been compressed; instead, each compressed chunk is prefixed
- * by a `struct pwm_chunk_hdr' that gives its size. However, the offsets are
- * given in the chunk table as if these chunk headers were not present.
+ * streams do not begin with a chunk table, since the chunk table cannot be
+ * written until all chunks have been compressed. Instead, each compressed
+ * chunk is prefixed by a `struct pwm_chunk_hdr' that gives its size.
+ * Furthermore, the chunk table is written at the end of the resource instead
+ * of the start. Note: chunk offsets are given in the chunk table as if the
+ * `struct pwm_chunk_hdr's were not present; also, the chunk table is only
+ * used if the WIM is being read from a seekable file (not a pipe).
*
* - Metadata resources always come before other file resources (streams).
* (This does not by itself constitute an incompatibility with normal WIMs,
*
* Layout of pipable WIM:
*
- * ----------+----------+--------------------+----------------+--------------+------------+--------+
- * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
- * ----------+----------+--------------------+----------------+--------------+------------+--------+
+ * ---------+----------+--------------------+----------------+--------------+-----------+--------+
+ * | Header | XML data | Metadata resources | File resources | Lookup table | XML data | Header |
+ * ---------+----------+--------------------+----------------+--------------+-----------+--------+
*
* Layout of normal WIM:
*
- * +---------+--------------------+----------------+--------------+----------+
- * | Header | Metadata resources | File resources | Lookup table | XML data |
- * +---------+--------------------+----------------+--------------+----------+
+ * +--------+-----------------------------+-------------------------+
+ * | Header | File and metadata resources | Lookup table | XML data |
+ * +--------+-----------------------------+-------------------------+
+ *
+ * An optional integrity table can follow the final XML data in both normal and
+ * pipable WIMs. However, due to implementation details, wimlib currently can
+ * only include an integrity table in a pipable WIM when writing it to a
+ * seekable file (not a pipe).
*
* Do note that since pipable WIMs are not supported by Microsoft's software,
* wimlib does not create them unless explicitly requested (with
DEBUG("Number of threads: %u", num_threads);
DEBUG("Progress function: %s", (progress_func ? "yes" : "no"));
DEBUG("Stream list: %s", (stream_list_override ? "specified" : "autodetect"));
- DEBUG("GUID: %s", (guid ? "specified" : "generate new"));
+ DEBUG("GUID: %s", ((guid || wim->guid_set_explicitly) ?
+ "specified" : "generate new"));
/* Internally, this is always called with a valid part number and total
* parts. */
(image < 1 || image > wim->hdr.image_count))
return WIMLIB_ERR_INVALID_IMAGE;
- /* @wim must specify a standalone WIM. */
- if (wim->hdr.total_parts != 1)
- return WIMLIB_ERR_SPLIT_UNSUPPORTED;
+ /* If we need to write metadata resources, make sure the ::WIMStruct has
+ * the needed information attached (e.g. is not a resource-only WIM,
+ * such as a non-first part of a split WIM). */
+ if (!wim_has_metadata(wim) &&
+ !(write_flags & WIMLIB_WRITE_FLAG_NO_METADATA))
+ return WIMLIB_ERR_METADATA_NOT_FOUND;
/* Check for contradictory flags. */
if ((write_flags & (WIMLIB_WRITE_FLAG_CHECK_INTEGRITY |
wim->hdr.part_number = part_number;
wim->hdr.total_parts = total_parts;
+ /* Set compression type if different. */
+ if (wim->compression_type != wim->out_compression_type)
+ wim->hdr.flags = get_wim_hdr_cflags(wim->out_compression_type);
+
+ /* Set chunk size if different. */
+ wim->hdr.chunk_size = wim->out_chunk_size;
+
/* Use GUID if specified; otherwise generate a new one. */
if (guid)
memcpy(wim->hdr.guid, guid, WIMLIB_GUID_LEN);
- else
+ else if (!wim->guid_set_explicitly)
randomize_byte_array(wim->hdr.guid, WIMLIB_GUID_LEN);
/* Clear references to resources that have not been written yet. */
stream_list_override);
out_restore_hdr:
memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
- close_wim_writable(wim, write_flags);
+ (void)close_wim_writable(wim, write_flags);
+ DEBUG("ret=%d", ret);
return ret;
}
return false;
}
+static int
+check_resource_offset(struct wim_lookup_table_entry *lte, void *_wim)
+{
+ const WIMStruct *wim = _wim;
+ off_t end_offset = *(const off_t*)wim->private;
+
+ if (lte->resource_location == RESOURCE_IN_WIM && lte->wim == wim &&
+ lte->resource_entry.offset + lte->resource_entry.size > end_offset)
+ return WIMLIB_ERR_RESOURCE_ORDER;
+ return 0;
+}
+
+/* Make sure no file or metadata resources are located after the XML data (or
+ * integrity table if present)--- otherwise we can't safely overwrite the WIM in
+ * place and we return WIMLIB_ERR_RESOURCE_ORDER. */
+static int
+check_resource_offsets(WIMStruct *wim, off_t end_offset)
+{
+ int ret;
+ unsigned i;
+
+ wim->private = &end_offset;
+ ret = for_lookup_table_entry(wim->lookup_table, check_resource_offset, wim);
+ if (ret)
+ return ret;
+
+ for (i = 0; i < wim->hdr.image_count; i++) {
+ ret = check_resource_offset(wim->image_metadata[i]->metadata_lte, wim);
+ if (ret)
+ return ret;
+ }
+ return 0;
+}
+
/*
* Overwrite a WIM, possibly appending streams to it.
*
struct list_head stream_list;
off_t old_wim_end;
u64 old_lookup_table_end, old_xml_begin, old_xml_end;
-
+ struct wim_header hdr_save;
DEBUG("Overwriting `%"TS"' in-place", wim->filename);
if (wim_has_integrity_table(wim))
write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
+ /* Set additional flags for overwrite. */
+ write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_STREAMS_OK;
+
/* Make sure that the integrity table (if present) is after the XML
* data, and that there are no stream resources, metadata resources, or
* lookup tables after the XML data. Otherwise, these data would be
old_lookup_table_end = wim->hdr.lookup_table_res_entry.offset +
wim->hdr.lookup_table_res_entry.size;
if (wim->hdr.integrity.offset != 0 && wim->hdr.integrity.offset < old_xml_end) {
- ERROR("Didn't expect the integrity table to be before the XML data");
+ WARNING("Didn't expect the integrity table to be before the XML data");
return WIMLIB_ERR_RESOURCE_ORDER;
}
if (old_lookup_table_end > old_xml_begin) {
- ERROR("Didn't expect the lookup table to be after the XML data");
+ WARNING("Didn't expect the lookup table to be after the XML data");
return WIMLIB_ERR_RESOURCE_ORDER;
}
old_wim_end = old_xml_end;
}
- ret = prepare_streams_for_overwrite(wim, old_wim_end, &stream_list);
+ ret = check_resource_offsets(wim, old_wim_end);
+ if (ret)
+ return ret;
+
+ ret = prepare_stream_list(wim, WIMLIB_ALL_IMAGES, write_flags,
+ &stream_list);
if (ret)
return ret;
return ret;
ret = lock_wim(wim, wim->out_fd.fd);
- if (ret) {
- close_wim_writable(wim, write_flags);
- return ret;
- }
+ if (ret)
+ goto out_close_wim;
+
+ /* Save original header so it can be restored in case of error */
+ memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
/* Set WIM_HDR_FLAG_WRITE_IN_PROGRESS flag in header. */
- ret = write_wim_header_flags(wim->hdr.flags | WIM_HDR_FLAG_WRITE_IN_PROGRESS,
- &wim->out_fd);
+ wim->hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
+ ret = write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
if (ret) {
ERROR_WITH_ERRNO("Error updating WIM header flags");
- close_wim_writable(wim, write_flags);
- goto out_unlock_wim;
+ goto out_restore_memory_hdr;
}
if (filedes_seek(&wim->out_fd, old_wim_end) == -1) {
ERROR_WITH_ERRNO("Can't seek to end of WIM");
- close_wim_writable(wim, write_flags);
ret = WIMLIB_ERR_WRITE;
- goto out_unlock_wim;
+ goto out_restore_physical_hdr;
}
- DEBUG("Writing newly added streams (offset = %"PRIu64")",
- old_wim_end);
ret = write_stream_list(&stream_list,
wim->lookup_table,
&wim->out_fd,
wim->compression_type,
+ wim->chunk_size,
+ &wim->lzx_context,
write_flags,
num_threads,
progress_func);
if (ret)
goto out_truncate;
- for (unsigned i = 1; i <= wim->hdr.image_count; i++) {
- if (wim->image_metadata[i - 1]->modified) {
- ret = write_metadata_resource(wim, i, 0);
- if (ret)
- goto out_truncate;
- }
- }
+ ret = write_wim_metadata_resources(wim, WIMLIB_ALL_IMAGES,
+ write_flags, progress_func);
+ if (ret)
+ goto out_truncate;
+
write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
progress_func, NULL);
+ if (ret)
+ goto out_truncate;
+
+ goto out_unlock_wim;
+
out_truncate:
- close_wim_writable(wim, write_flags);
- if (ret && !(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
+ if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
WARNING("Truncating `%"TS"' to its original size (%"PRIu64" bytes)",
wim->filename, old_wim_end);
- /* Return value of truncate() is ignored because this is already
- * an error path. */
- (void)ttruncate(wim->filename, old_wim_end);
+ /* Return value of ftruncate() is ignored because this is
+ * already an error path. */
+ (void)ftruncate(wim->out_fd.fd, old_wim_end);
}
+out_restore_physical_hdr:
+ (void)write_wim_header_flags(hdr_save.flags, &wim->out_fd);
+out_restore_memory_hdr:
+ memcpy(&wim->hdr, &hdr_save, sizeof(struct wim_header));
+out_close_wim:
+ (void)close_wim_writable(wim, write_flags);
out_unlock_wim:
wim->wim_locked = 0;
return ret;
ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,
write_flags | WIMLIB_WRITE_FLAG_FSYNC,
num_threads, progress_func);
- if (ret)
- goto out_unlink;
+ if (ret) {
+ tunlink(tmpfile);
+ return ret;
+ }
close_wim(wim);
+ /* Rename the new WIM file to the original WIM file. Note: on Windows
+ * this actually calls win32_rename_replacement(), not _wrename(), so
+ * that removing the existing destination file can be handled. */
DEBUG("Renaming `%"TS"' to `%"TS"'", tmpfile, wim->filename);
- /* Rename the new file to the old file .*/
- if (trename(tmpfile, wim->filename) != 0) {
+ ret = trename(tmpfile, wim->filename);
+ if (ret) {
ERROR_WITH_ERRNO("Failed to rename `%"TS"' to `%"TS"'",
tmpfile, wim->filename);
- ret = WIMLIB_ERR_RENAME;
- goto out_unlink;
+ #ifdef __WIN32__
+ if (ret < 0)
+ #endif
+ {
+ tunlink(tmpfile);
+ }
+ return WIMLIB_ERR_RENAME;
}
if (progress_func) {
progress_func(WIMLIB_PROGRESS_MSG_RENAME, &progress);
}
return 0;
-
-out_unlink:
- /* Remove temporary file. */
- tunlink(tmpfile);
- return ret;
}
/* API function documented in wimlib.h */
if ((!wim->deletion_occurred || (write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
&& !(write_flags & (WIMLIB_WRITE_FLAG_REBUILD |
WIMLIB_WRITE_FLAG_PIPABLE))
- && !(wim_is_pipable(wim)))
+ && !(wim_is_pipable(wim))
+ && wim->compression_type == wim->out_compression_type
+ && wim->chunk_size == wim->out_chunk_size)
{
ret = overwrite_wim_inplace(wim, write_flags, num_threads,
progress_func);