*/
/*
- * Copyright (C) 2012, 2013 Eric Biggers
+ * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
*
- * This file is part of wimlib, a library for working with WIM files.
+ * This file is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the Free
+ * Software Foundation; either version 3 of the License, or (at your option) any
+ * later version.
*
- * wimlib is free software; you can redistribute it and/or modify it under the
- * terms of the GNU General Public License as published by the Free
- * Software Foundation; either version 3 of the License, or (at your option)
- * any later version.
- *
- * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
- * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
- * A PARTICULAR PURPOSE. See the GNU General Public License for more
+ * This file is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
* details.
*
- * You should have received a copy of the GNU General Public License
- * along with wimlib; if not, see http://www.gnu.org/licenses/.
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this file; if not, see http://www.gnu.org/licenses/.
*/
#ifdef HAVE_CONFIG_H
# include <sys/file.h>
#endif
+#include <errno.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include "wimlib/alloca.h"
+#include "wimlib/assert.h"
#include "wimlib/chunk_compressor.h"
#include "wimlib/endianness.h"
#include "wimlib/error.h"
#include "wimlib/integrity.h"
#include "wimlib/lookup_table.h"
#include "wimlib/metadata.h"
+#include "wimlib/paths.h"
#include "wimlib/progress.h"
#include "wimlib/resource.h"
+#include "wimlib/solid.h"
#ifdef __WIN32__
# include "wimlib/win32.h" /* win32_rename_replacement() */
#endif
#include "wimlib/write.h"
#include "wimlib/xml.h"
-#include <errno.h>
-#include <fcntl.h>
-#include <stdlib.h>
-#include <unistd.h>
-
-#ifdef HAVE_ALLOCA_H
-# include <alloca.h>
-#endif
/* wimlib internal flags used when writing resources. */
#define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001
#define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002
-#define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004
+#define WRITE_RESOURCE_FLAG_SOLID 0x00000004
+#define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008
+#define WRITE_RESOURCE_FLAG_SOLID_SORT 0x00000010
static inline int
write_flags_to_resource_flags(int write_flags)
write_resource_flags |= WRITE_RESOURCE_FLAG_RECOMPRESS;
if (write_flags & WIMLIB_WRITE_FLAG_PIPABLE)
write_resource_flags |= WRITE_RESOURCE_FLAG_PIPABLE;
- if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
- write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
+ if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
+ write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID;
+ if (write_flags & WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES)
+ write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE;
+ if ((write_flags & (WIMLIB_WRITE_FLAG_SOLID |
+ WIMLIB_WRITE_FLAG_NO_SOLID_SORT)) ==
+ WIMLIB_WRITE_FLAG_SOLID)
+ write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID_SORT;
return write_resource_flags;
}
stream_filtered(const struct wim_lookup_table_entry *lte,
const struct filter_context *ctx)
{
- int write_flags = ctx->write_flags;
- WIMStruct *wim = ctx->wim;
+ int write_flags;
+ WIMStruct *wim;
if (ctx == NULL)
return 0;
+ write_flags = ctx->write_flags;
+ wim = ctx->wim;
+
if (write_flags & WIMLIB_WRITE_FLAG_OVERWRITE &&
lte->resource_location == RESOURCE_IN_WIM &&
lte->rspec->wim == wim)
rspec->chunk_size == out_chunk_size);
}
- if ((rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
- (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
+ if ((rspec->flags & WIM_RESHDR_FLAG_SOLID) &&
+ (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
{
- /* Packed resource: Such resources may contain multiple streams,
+ /* Solid resource: Such resources may contain multiple streams,
* and in general only a subset of them need to be written. As
* a heuristic, re-use the raw data if more than two-thirds the
* uncompressed size is being written. */
- /* Note: packed resources contain a header that specifies the
+ /* Note: solid resources contain a header that specifies the
* compression type and chunk size; therefore we don't need to
* check if they are compatible with @out_ctype and
* @out_chunk_size. */
static u8
filter_resource_flags(u8 flags)
{
- return (flags & ~(WIM_RESHDR_FLAG_PACKED_STREAMS |
+ return (flags & ~(WIM_RESHDR_FLAG_SOLID |
WIM_RESHDR_FLAG_COMPRESSED |
WIM_RESHDR_FLAG_SPANNED |
WIM_RESHDR_FLAG_FREE));
wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
rspec = lte->rspec;
- if (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
+ if (rspec->flags & WIM_RESHDR_FLAG_SOLID) {
- wimlib_assert(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS);
+ wimlib_assert(lte->flags & WIM_RESHDR_FLAG_SOLID);
lte->out_reshdr.offset_in_wim = lte->offset_in_res;
lte->out_reshdr.uncompressed_size = 0;
lte->out_res_size_in_wim = rspec->size_in_wim;
lte->out_res_uncompressed_size = rspec->uncompressed_size;
} else {
- wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
+ wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_SOLID));
lte->out_reshdr.offset_in_wim = rspec->offset_in_wim;
lte->out_reshdr.uncompressed_size = rspec->uncompressed_size;
static int
do_write_streams_progress(struct write_streams_progress_data *progress_data,
- struct wim_lookup_table_entry *cur_stream,
u64 complete_size,
u32 complete_count,
bool discarded)
if (progress_data->next_progress == progress->write_streams.total_bytes) {
progress_data->next_progress = ~(uint64_t)0;
} else {
+ /* Handle rate-limiting of messages */
+
+ /* Send new message as soon as another 1/128 of the
+ * total has been written. (Arbitrary number.) */
progress_data->next_progress =
- min(progress->write_streams.total_bytes,
- progress->write_streams.completed_bytes +
- progress->write_streams.total_bytes / 100);
+ progress->write_streams.completed_bytes +
+ progress->write_streams.total_bytes / 128;
+
+ /* ... Unless that would be more than 5000000 bytes, in
+ * which case send the next after the next 5000000
+ * bytes. (Another arbitrary number.) */
+ if (progress->write_streams.completed_bytes + 5000000 <
+ progress_data->next_progress)
+ progress_data->next_progress =
+ progress->write_streams.completed_bytes + 5000000;
+
+ /* ... But always send a message as soon as we're
+ * completely done. */
+ if (progress->write_streams.total_bytes <
+ progress_data->next_progress)
+ progress_data->next_progress =
+ progress->write_streams.total_bytes;
}
}
return 0;
* uncompressed. */
struct chunk_compressor *compressor;
- /* Buffer for dividing the read data into chunks of size
- * @out_chunk_size. */
- u8 *chunk_buf;
+ /* A buffer of size @out_chunk_size that has been loaned out from the
+ * chunk compressor and is currently being filled with the uncompressed
+ * data of the next chunk. */
+ u8 *cur_chunk_buf;
- /* Number of bytes in @chunk_buf that are currently filled. */
- size_t chunk_buf_filled;
+ /* Number of bytes in @cur_chunk_buf that are currently filled. */
+ size_t cur_chunk_buf_filled;
/* List of streams that currently have chunks being compressed. */
struct list_head pending_streams;
- /* List of streams in the resource pack. Streams are moved here after
- * @pending_streams only when writing a packed resource. */
- struct list_head pack_streams;
-
- /* Set to true if the stream currently being read was a duplicate, and
- * therefore the corresponding stream entry needs to be freed once the
- * read finishes. (In this case we add the duplicate entry to
- * pending_streams rather than the entry being read.) */
- bool stream_was_duplicate;
+ /* List of streams in the solid resource. Streams are moved here after
+ * @pending_streams only when writing a solid resource. */
+ struct list_head solid_streams;
/* Current uncompressed offset in the stream being read. */
u64 cur_read_stream_offset;
/* Calculate the number of chunks and chunk entries that should be
* needed for the resource. These normally will be the final values,
- * but in PACKED_STREAMS mode some of the streams we're planning to
- * write into the resource may be duplicates, and therefore discarded,
- * potentially decreasing the number of chunk entries needed. */
+ * but in SOLID mode some of the streams we're planning to write into
+ * the resource may be duplicates, and therefore discarded, potentially
+ * decreasing the number of chunk entries needed. */
expected_num_chunks = DIV_ROUND_UP(res_expected_size, ctx->out_chunk_size);
expected_num_chunk_entries = expected_num_chunks;
- if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
+ if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
expected_num_chunk_entries--;
/* Make sure the chunk_csizes array is long enough to store the
if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)) {
/* Reserve space for the chunk table in the output file. In the
- * case of packed resources this reserves the upper bound for
- * the needed space, not necessarily the exact space which will
+ * case of solid resources this reserves the upper bound for the
+ * needed space, not necessarily the exact space which will
* prove to be needed. At this point, we just use @chunk_csizes
* for a buffer of 0's because the actual compressed chunk sizes
* are unknown. */
reserve_size = expected_num_chunk_entries *
get_chunk_entry_size(res_expected_size,
0 != (ctx->write_resource_flags &
- WRITE_RESOURCE_FLAG_PACK_STREAMS));
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
+ WRITE_RESOURCE_FLAG_SOLID));
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
reserve_size += sizeof(struct alt_chunk_table_header_disk);
memset(ctx->chunk_csizes, 0, reserve_size);
ret = full_write(ctx->out_fd, ctx->chunk_csizes, reserve_size);
actual_num_chunks = ctx->chunk_index;
actual_num_chunk_entries = actual_num_chunks;
- if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
+ if (!(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
actual_num_chunk_entries--;
chunk_entry_size = get_chunk_entry_size(res_actual_size,
0 != (ctx->write_resource_flags &
- WRITE_RESOURCE_FLAG_PACK_STREAMS));
+ WRITE_RESOURCE_FLAG_SOLID));
- typedef le64 __attribute__((may_alias)) aliased_le64_t;
- typedef le32 __attribute__((may_alias)) aliased_le32_t;
+ typedef le64 _may_alias_attribute aliased_le64_t;
+ typedef le32 _may_alias_attribute aliased_le32_t;
if (chunk_entry_size == 4) {
aliased_le32_t *entries = (aliased_le32_t*)ctx->chunk_csizes;
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
for (size_t i = 0; i < actual_num_chunk_entries; i++)
entries[i] = cpu_to_le32(ctx->chunk_csizes[i]);
} else {
} else {
aliased_le64_t *entries = (aliased_le64_t*)ctx->chunk_csizes;
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
for (size_t i = 0; i < actual_num_chunk_entries; i++)
entries[i] = cpu_to_le64(ctx->chunk_csizes[i]);
} else {
if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
ret = full_write(ctx->out_fd, ctx->chunk_csizes, chunk_table_size);
if (ret)
- goto error;
+ goto write_error;
res_end_offset = ctx->out_fd->offset;
res_start_offset = ctx->chunks_start_offset;
} else {
chunk_table_offset = ctx->chunks_start_offset - chunk_table_size;
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
struct alt_chunk_table_header_disk hdr;
hdr.res_usize = cpu_to_le64(res_actual_size);
ret = full_pwrite(ctx->out_fd, &hdr, sizeof(hdr),
chunk_table_offset - sizeof(hdr));
if (ret)
- goto error;
+ goto write_error;
res_start_offset = chunk_table_offset - sizeof(hdr);
} else {
res_start_offset = chunk_table_offset;
ret = full_pwrite(ctx->out_fd, ctx->chunk_csizes,
chunk_table_size, chunk_table_offset);
if (ret)
- goto error;
+ goto write_error;
}
*res_start_offset_ret = res_start_offset;
return 0;
-error:
+write_error:
ERROR_WITH_ERRNO("Write error");
return ret;
}
u64 res_offset_in_wim;
wimlib_assert(ctx->cur_write_stream_offset == ctx->cur_write_res_size ||
- (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS));
+ (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID));
res_uncompressed_size = ctx->cur_write_res_size;
if (ctx->compressor) {
return 0;
}
+/* No more data streams of the file at @path are needed. */
+static int
+done_with_file(const tchar *path, wimlib_progress_func_t progfunc, void *progctx)
+{
+ union wimlib_progress_info info;
+
+ info.done_with_file.path_to_file = path;
+
+ return call_progress(progfunc, WIMLIB_PROGRESS_MSG_DONE_WITH_FILE,
+ &info, progctx);
+}
+
+static inline bool
+is_file_stream(const struct wim_lookup_table_entry *lte)
+{
+ return lte->resource_location == RESOURCE_IN_FILE_ON_DISK
+#ifdef __WIN32__
+ || lte->resource_location == RESOURCE_IN_WINNT_FILE_ON_DISK
+ || lte->resource_location == RESOURCE_WIN32_ENCRYPTED
+#endif
+ ;
+}
+
+static int
+do_done_with_stream(struct wim_lookup_table_entry *lte,
+ wimlib_progress_func_t progfunc, void *progctx)
+{
+ int ret;
+ struct wim_inode *inode;
+
+ if (!lte->may_send_done_with_file)
+ return 0;
+
+ inode = lte->file_inode;
+
+ wimlib_assert(inode != NULL);
+ wimlib_assert(inode->num_remaining_streams > 0);
+ if (--inode->num_remaining_streams > 0)
+ return 0;
+
+#ifdef __WIN32__
+ /* XXX: This logic really should be somewhere else. */
+
+ /* We want the path to the file, but lte->file_on_disk might actually
+ * refer to a named data stream. Temporarily strip the named data
+ * stream from the path. */
+ wchar_t *p_colon = NULL;
+ wchar_t *p_question_mark = NULL;
+ const wchar_t *p_stream_name;
+
+ p_stream_name = path_stream_name(lte->file_on_disk);
+ if (unlikely(p_stream_name)) {
+ p_colon = (wchar_t *)(p_stream_name - 1);
+ wimlib_assert(*p_colon == L':');
+ *p_colon = L'\0';
+ }
+
+ /* We also should use a fake Win32 path instead of a NT path */
+ if (!wcsncmp(lte->file_on_disk, L"\\??\\", 4)) {
+ p_question_mark = <e->file_on_disk[1];
+ *p_question_mark = L'\\';
+ }
+#endif
+
+ ret = done_with_file(lte->file_on_disk, progfunc, progctx);
+
+#ifdef __WIN32__
+ if (p_colon)
+ *p_colon = L':';
+ if (p_question_mark)
+ *p_question_mark = L'?';
+#endif
+ return ret;
+}
+
+/* Handle WIMLIB_WRITE_FLAG_SEND_DONE_WITH_FILE_MESSAGES mode. */
+static inline int
+done_with_stream(struct wim_lookup_table_entry *lte,
+ struct write_streams_ctx *ctx)
+{
+ if (likely(!(ctx->write_resource_flags &
+ WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE)))
+ return 0;
+ return do_done_with_stream(lte, ctx->progress_data.progfunc,
+ ctx->progress_data.progctx);
+}
+
/* Begin processing a stream for writing. */
static int
-write_stream_begin_read(struct wim_lookup_table_entry *lte,
- u32 flags, void *_ctx)
+write_stream_begin_read(struct wim_lookup_table_entry *lte, void *_ctx)
{
struct write_streams_ctx *ctx = _ctx;
int ret;
* still provide the data again to write_stream_process_chunk(). This
* is okay because an unhashed stream cannot be in a WIM resource, which
* might be costly to decompress. */
- ctx->stream_was_duplicate = false;
if (ctx->lookup_table != NULL && lte->unhashed && !lte->unique_size) {
- wimlib_assert(!(flags & BEGIN_STREAM_FLAG_PARTIAL_RESOURCE));
-
struct wim_lookup_table_entry *lte_new;
ret = hash_unhashed_stream(lte, ctx->lookup_table, <e_new);
DEBUG("Discarding duplicate stream of "
"length %"PRIu64, lte->size);
ret = do_write_streams_progress(&ctx->progress_data,
- lte, lte->size,
+ lte->size,
1, true);
list_del(<e->write_streams_list);
list_del(<e->lookup_table_list);
if (lte_new->will_be_in_output_wim)
lte_new->out_refcnt += lte->out_refcnt;
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS)
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID)
ctx->cur_write_res_size -= lte->size;
+ if (!ret)
+ ret = done_with_stream(lte, ctx);
free_lookup_table_entry(lte);
if (ret)
return ret;
<e_new->write_streams_list);
list_replace(<e->lookup_table_list,
<e_new->lookup_table_list);
+ lte->will_be_in_output_wim = 0;
lte_new->out_refcnt = lte->out_refcnt;
lte_new->will_be_in_output_wim = 1;
- ctx->stream_was_duplicate = true;
+ lte_new->may_send_done_with_file = 0;
lte = lte_new;
}
}
lte->out_reshdr.size_in_wim = lte->size;
lte->out_reshdr.flags &= ~(WIM_RESHDR_FLAG_COMPRESSED |
- WIM_RESHDR_FLAG_PACKED_STREAMS);
+ WIM_RESHDR_FLAG_SOLID);
return 0;
}
+/* Returns true if the specified stream should be truncated from the WIM file
+ * and re-written as uncompressed. lte->out_reshdr must be filled in from the
+ * initial write of the stream. */
+static bool
+should_rewrite_stream_uncompressed(const struct write_streams_ctx *ctx,
+ const struct wim_lookup_table_entry *lte)
+{
+ /* If the compressed data is smaller than the uncompressed data, prefer
+ * the compressed data. */
+ if (lte->out_reshdr.size_in_wim < lte->out_reshdr.uncompressed_size)
+ return false;
+
+ /* If we're not actually writing compressed data, then there's no need
+ * for re-writing. */
+ if (!ctx->compressor)
+ return false;
+
+ /* If writing a pipable WIM, everything we write to the output is final
+ * (it might actually be a pipe!). */
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE)
+ return false;
+
+ /* If the stream that would need to be re-read is located in a solid
+ * resource in another WIM file, then re-reading it would be costly. So
+ * don't do it.
+ *
+ * Exception: if the compressed size happens to be *exactly* the same as
+ * the uncompressed size, then the stream *must* be written uncompressed
+ * in order to remain compatible with the Windows Overlay Filesystem
+ * Filter Driver (WOF).
+ *
+ * TODO: we are currently assuming that the optimization for
+ * single-chunk resources in maybe_rewrite_stream_uncompressed()
+ * prevents this case from being triggered too often. To fully prevent
+ * excessive decompressions in degenerate cases, we really should
+ * obtain the uncompressed data by decompressing the compressed data we
+ * wrote to the output file.
+ */
+ if ((lte->flags & WIM_RESHDR_FLAG_SOLID) &&
+ (lte->out_reshdr.size_in_wim != lte->out_reshdr.uncompressed_size))
+ return false;
+
+ return true;
+}
+
+static int
+maybe_rewrite_stream_uncompressed(struct write_streams_ctx *ctx,
+ struct wim_lookup_table_entry *lte)
+{
+ if (!should_rewrite_stream_uncompressed(ctx, lte))
+ return 0;
+
+ /* Regular (non-solid) WIM resources with exactly one chunk and
+ * compressed size equal to uncompressed size are exactly the same as
+ * the corresponding compressed data --- since there must be 0 entries
+ * in the chunk table and the only chunk must be stored uncompressed.
+ * In this case, there's no need to rewrite anything. */
+ if (ctx->chunk_index == 1 &&
+ lte->out_reshdr.size_in_wim == lte->out_reshdr.uncompressed_size)
+ {
+ lte->out_reshdr.flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
+ return 0;
+ }
+
+ return write_stream_uncompressed(lte, ctx->out_fd);
+}
+
/* Write the next chunk of (typically compressed) data to the output WIM,
* handling the writing of the chunk table. */
static int
struct wim_lookup_table_entry, write_streams_list);
if (ctx->cur_write_stream_offset == 0 &&
- !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS))
+ !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID))
{
- /* Starting to write a new stream in non-packed mode. */
+ /* Starting to write a new stream in non-solid mode. */
if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) {
int additional_reshdr_flags = 0;
ret = full_write(ctx->out_fd, &chunk_hdr,
sizeof(chunk_hdr));
if (ret)
- goto error;
+ goto write_error;
}
}
/* Write the chunk data. */
ret = full_write(ctx->out_fd, cchunk, csize);
if (ret)
- goto error;
+ goto write_error;
ctx->cur_write_stream_offset += usize;
completed_size = usize;
completed_stream_count = 0;
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
- /* Wrote chunk in packed mode. It may have finished multiple
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
+ /* Wrote chunk in solid mode. It may have finished multiple
* streams. */
- while (ctx->cur_write_stream_offset > lte->size) {
- struct wim_lookup_table_entry *next;
+ struct wim_lookup_table_entry *next_lte;
+
+ while (lte && ctx->cur_write_stream_offset >= lte->size) {
ctx->cur_write_stream_offset -= lte->size;
- wimlib_assert(!list_is_singular(&ctx->pending_streams) &&
- !list_empty(&ctx->pending_streams));
- next = list_entry(lte->write_streams_list.next,
- struct wim_lookup_table_entry,
- write_streams_list);
- list_move_tail(<e->write_streams_list,
- &ctx->pack_streams);
- lte = next;
- completed_stream_count++;
- }
- if (ctx->cur_write_stream_offset == lte->size) {
- ctx->cur_write_stream_offset = 0;
- list_move_tail(<e->write_streams_list,
- &ctx->pack_streams);
+ if (ctx->cur_write_stream_offset)
+ next_lte = list_entry(lte->write_streams_list.next,
+ struct wim_lookup_table_entry,
+ write_streams_list);
+ else
+ next_lte = NULL;
+
+ ret = done_with_stream(lte, ctx);
+ if (ret)
+ return ret;
+ list_move_tail(<e->write_streams_list, &ctx->solid_streams);
completed_stream_count++;
+
+ lte = next_lte;
}
} else {
- /* Wrote chunk in non-packed mode. It may have finished a
+ /* Wrote chunk in non-solid mode. It may have finished a
* stream. */
if (ctx->cur_write_stream_offset == lte->size) {
- completed_stream_count++;
-
- list_del(<e->write_streams_list);
-
wimlib_assert(ctx->cur_write_stream_offset ==
ctx->cur_write_res_size);
if (ctx->compressor != NULL)
lte->out_reshdr.flags |= WIM_RESHDR_FLAG_COMPRESSED;
- if (ctx->compressor != NULL &&
- lte->out_reshdr.size_in_wim >= lte->out_reshdr.uncompressed_size &&
- !(ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PIPABLE) &&
- !(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS))
- {
- /* Stream did not compress to less than its original
- * size. If we're not writing a pipable WIM (which
- * could mean the output file descriptor is
- * non-seekable), and the stream isn't located in a
- * resource pack (which would make reading it again
- * costly), truncate the file to the start of the stream
- * and write it uncompressed instead. */
- DEBUG("Stream of size %"PRIu64" did not compress to "
- "less than original size; writing uncompressed.",
- lte->size);
- ret = write_stream_uncompressed(lte, ctx->out_fd);
- if (ret)
- return ret;
- }
+ ret = maybe_rewrite_stream_uncompressed(ctx, lte);
+ if (ret)
+ return ret;
+
wimlib_assert(lte->out_reshdr.uncompressed_size == lte->size);
ctx->cur_write_stream_offset = 0;
+
+ ret = done_with_stream(lte, ctx);
+ if (ret)
+ return ret;
+ list_del(<e->write_streams_list);
+ completed_stream_count++;
}
}
- return do_write_streams_progress(&ctx->progress_data, lte,
+ return do_write_streams_progress(&ctx->progress_data,
completed_size, completed_stream_count,
false);
-error:
+write_error:
ERROR_WITH_ERRNO("Write error");
return ret;
}
static int
-submit_chunk_for_compression(struct write_streams_ctx *ctx,
- const void *chunk, size_t size)
+prepare_chunk_buffer(struct write_streams_ctx *ctx)
{
- /* While we are unable to submit the chunk for compression (due to too
- * many chunks already outstanding), retrieve and write the next
- * compressed chunk. */
- while (!ctx->compressor->submit_chunk(ctx->compressor, chunk, size)) {
+ /* While we are unable to get a new chunk buffer due to too many chunks
+ * already outstanding, retrieve and write the next compressed chunk. */
+ while (!(ctx->cur_chunk_buf =
+ ctx->compressor->get_chunk_buffer(ctx->compressor)))
+ {
const void *cchunk;
u32 csize;
u32 usize;
bool bret;
int ret;
- bret = ctx->compressor->get_chunk(ctx->compressor,
- &cchunk, &csize, &usize);
-
+ bret = ctx->compressor->get_compression_result(ctx->compressor,
+ &cchunk,
+ &csize,
+ &usize);
wimlib_assert(bret);
ret = write_chunk(ctx, cchunk, csize, usize);
chunkptr = chunk;
chunkend = chunkptr + size;
do {
- const u8 *resized_chunk;
size_t needed_chunk_size;
+ size_t bytes_consumed;
+
+ if (!ctx->cur_chunk_buf) {
+ ret = prepare_chunk_buffer(ctx);
+ if (ret)
+ return ret;
+ }
- if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
needed_chunk_size = ctx->out_chunk_size;
} else {
- u64 res_bytes_remaining;
-
- res_bytes_remaining = ctx->cur_read_stream_size -
- ctx->cur_read_stream_offset;
needed_chunk_size = min(ctx->out_chunk_size,
- ctx->chunk_buf_filled +
- res_bytes_remaining);
+ ctx->cur_chunk_buf_filled +
+ (ctx->cur_read_stream_size -
+ ctx->cur_read_stream_offset));
}
- if (ctx->chunk_buf_filled == 0 &&
- chunkend - chunkptr >= needed_chunk_size)
- {
- /* No intermediate buffering needed. */
- resized_chunk = chunkptr;
- chunkptr += needed_chunk_size;
- ctx->cur_read_stream_offset += needed_chunk_size;
- } else {
- /* Intermediate buffering needed. */
- size_t bytes_consumed;
-
- bytes_consumed = min(chunkend - chunkptr,
- needed_chunk_size - ctx->chunk_buf_filled);
+ bytes_consumed = min(chunkend - chunkptr,
+ needed_chunk_size - ctx->cur_chunk_buf_filled);
- memcpy(&ctx->chunk_buf[ctx->chunk_buf_filled],
- chunkptr, bytes_consumed);
+ memcpy(&ctx->cur_chunk_buf[ctx->cur_chunk_buf_filled],
+ chunkptr, bytes_consumed);
- chunkptr += bytes_consumed;
- ctx->cur_read_stream_offset += bytes_consumed;
- ctx->chunk_buf_filled += bytes_consumed;
- if (ctx->chunk_buf_filled == needed_chunk_size) {
- resized_chunk = ctx->chunk_buf;
- ctx->chunk_buf_filled = 0;
- } else {
- break;
- }
+ chunkptr += bytes_consumed;
+ ctx->cur_read_stream_offset += bytes_consumed;
+ ctx->cur_chunk_buf_filled += bytes_consumed;
+ if (ctx->cur_chunk_buf_filled == needed_chunk_size) {
+ ctx->compressor->signal_chunk_filled(ctx->compressor,
+ ctx->cur_chunk_buf_filled);
+ ctx->cur_chunk_buf = NULL;
+ ctx->cur_chunk_buf_filled = 0;
}
-
- ret = submit_chunk_for_compression(ctx, resized_chunk,
- needed_chunk_size);
- if (ret)
- return ret;
-
} while (chunkptr != chunkend);
return 0;
}
write_stream_end_read(struct wim_lookup_table_entry *lte, int status, void *_ctx)
{
struct write_streams_ctx *ctx = _ctx;
- if (status == 0)
- wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size);
- if (ctx->stream_was_duplicate) {
+
+ wimlib_assert(ctx->cur_read_stream_offset == ctx->cur_read_stream_size || status);
+
+ if (!lte->will_be_in_output_wim) {
+ /* The 'lte' stream was a duplicate. Now that its data has
+ * finished being read, it is being discarded in favor of the
+ * duplicate entry. It therefore is no longer needed, and we
+ * can fire the DONE_WITH_FILE callback because the file will
+ * not be read again.
+ *
+ * Note: we can't yet fire DONE_WITH_FILE for non-duplicate
+ * streams, since it needs to be possible to re-read the file if
+ * it does not compress to less than its original size. */
+ if (!status)
+ status = done_with_stream(lte, ctx);
free_lookup_table_entry(lte);
- } else if (lte->unhashed && ctx->lookup_table != NULL) {
+ } else if (!status && lte->unhashed && ctx->lookup_table != NULL) {
+ /* The 'lte' stream was not a duplicate and was previously
+ * unhashed. Since we passed COMPUTE_MISSING_STREAM_HASHES to
+ * read_stream_list(), lte->hash is now computed and valid. So
+ * turn this stream into a "hashed" stream. */
list_del(<e->unhashed_list);
lookup_table_insert(ctx->lookup_table, lte);
lte->unhashed = 0;
list_for_each_entry(lte, &in_rspec->stream_list, rspec_node) {
if (lte->will_be_in_output_wim) {
stream_set_out_reshdr_for_reuse(lte);
- if (in_rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
+ if (in_rspec->flags & WIM_RESHDR_FLAG_SOLID)
lte->out_res_offset_in_wim = out_offset_in_wim;
else
lte->out_reshdr.offset_in_wim = out_offset_in_wim;
list_for_each_entry(lte, raw_copy_streams, write_streams_list) {
if (lte->rspec->raw_copy_ok) {
- /* Write each packed resource only one time, no matter
+ /* Write each solid resource only one time, no matter
* how many streams reference it. */
ret = write_raw_copy_resource(lte->rspec, out_fd);
if (ret)
return ret;
lte->rspec->raw_copy_ok = 0;
}
- ret = do_write_streams_progress(progress_data, lte, lte->size,
+ ret = do_write_streams_progress(progress_data, lte->size,
1, false);
if (ret)
return ret;
if (ctx->compressor == NULL)
return 0;
- if (ctx->chunk_buf_filled != 0) {
- ret = submit_chunk_for_compression(ctx, ctx->chunk_buf,
- ctx->chunk_buf_filled);
- if (ret)
- return ret;
+ if (ctx->cur_chunk_buf_filled != 0) {
+ ctx->compressor->signal_chunk_filled(ctx->compressor,
+ ctx->cur_chunk_buf_filled);
}
- while (ctx->compressor->get_chunk(ctx->compressor, &cdata, &csize, &usize)) {
+ while (ctx->compressor->get_compression_result(ctx->compressor, &cdata,
+ &csize, &usize))
+ {
ret = write_chunk(ctx, cdata, csize, usize);
if (ret)
return ret;
}
}
+static void
+init_done_with_file_info(struct list_head *stream_list)
+{
+ struct wim_lookup_table_entry *lte;
+
+ list_for_each_entry(lte, stream_list, write_streams_list) {
+ if (is_file_stream(lte)) {
+ lte->file_inode->num_remaining_streams = 0;
+ lte->may_send_done_with_file = 1;
+ } else {
+ lte->may_send_done_with_file = 0;
+ }
+ }
+
+ list_for_each_entry(lte, stream_list, write_streams_list)
+ if (lte->may_send_done_with_file)
+ lte->file_inode->num_remaining_streams++;
+}
+
/*
* Write a list of streams to the output WIM file.
*
* furthermore do so in such a way that no seeking backwards in
* @out_fd will be performed (so it may be a pipe).
*
- * WRITE_RESOURCE_FLAG_PACK_STREAMS:
- * Pack all the streams into a single resource rather than writing
- * them in separate resources. This flag is only valid if the WIM
- * version number has been, or will be, set to
- * WIM_VERSION_PACKED_STREAMS. This flag may not be combined with
+ * WRITE_RESOURCE_FLAG_SOLID:
+ * Combine all the streams into a single resource rather than
+ * writing them in separate resources. This flag is only valid if
+ * the WIM version number has been, or will be, set to
+ * WIM_VERSION_SOLID. This flag may not be combined with
* WRITE_RESOURCE_FLAG_PIPABLE.
*
* @out_ctype
* Compression format to use to write the output streams, specified as one
- * of the WIMLIB_COMPRESSION_TYPE_* constants, excepting
- * WIMLIB_COMPRESSION_TYPE_INVALID but including
- * WIMLIB_COMPRESSION_TYPE_NONE.
+ * of the WIMLIB_COMPRESSION_TYPE_* constants.
+ * WIMLIB_COMPRESSION_TYPE_NONE is allowed.
*
* @out_chunk_size
* Chunk size to use to write the streams. It must be a valid chunk size
* can be NULL.
*
* This function will write the streams in @stream_list to resources in
- * consecutive positions in the output WIM file, or to a single packed resource
- * if WRITE_RESOURCE_FLAG_PACK_STREAMS was specified in @write_resource_flags.
- * In both cases, the @out_reshdr of the `struct wim_lookup_table_entry' for
- * each stream written will be updated to specify its location, size, and flags
- * in the output WIM. In the packed resource case,
- * WIM_RESHDR_FLAG_PACKED_STREAMS will be set in the @flags field of each
- * @out_reshdr, and furthermore @out_res_offset_in_wim and @out_res_size_in_wim
- * of each @out_reshdr will be set to the offset and size, respectively, in the
- * output WIM of the packed resource containing the corresponding stream.
+ * consecutive positions in the output WIM file, or to a single solid resource
+ * if WRITE_RESOURCE_FLAG_SOLID was specified in @write_resource_flags. In both
+ * cases, the @out_reshdr of the `struct wim_lookup_table_entry' for each stream
+ * written will be updated to specify its location, size, and flags in the
+ * output WIM. In the solid resource case, WIM_RESHDR_FLAG_SOLID will be set in
+ * the @flags field of each @out_reshdr, and furthermore @out_res_offset_in_wim
+ * and @out_res_size_in_wim of each @out_reshdr will be set to the offset and
+ * size, respectively, in the output WIM of the solid resource containing the
+ * corresponding stream.
*
* Each of the streams to write may be in any location supported by the
* resource-handling code (specifically, read_stream_list()), such as the
* @will_be_in_output_wim member be set to 1 on all streams in @stream_list as
* well as any other streams not in @stream_list that will be in the output WIM
* file, but set to 0 on any other streams in the output WIM's lookup table or
- * sharing a packed resource with a stream in @stream_list. Still furthermore,
+ * sharing a solid resource with a stream in @stream_list. Still furthermore,
* if on-the-fly deduplication of streams is possible, then all streams in
* @stream_list must also be linked by @lookup_table_list along with any other
* streams that have @will_be_in_output_wim set.
struct list_head raw_copy_streams;
wimlib_assert((write_resource_flags &
- (WRITE_RESOURCE_FLAG_PACK_STREAMS |
+ (WRITE_RESOURCE_FLAG_SOLID |
WRITE_RESOURCE_FLAG_PIPABLE)) !=
- (WRITE_RESOURCE_FLAG_PACK_STREAMS |
+ (WRITE_RESOURCE_FLAG_SOLID |
WRITE_RESOURCE_FLAG_PIPABLE));
remove_zero_length_streams(stream_list);
return 0;
}
- memset(&ctx, 0, sizeof(ctx));
+ /* If needed, set auxiliary information so that we can detect when the
+ * library has finished using each external file. */
+ if (unlikely(write_resource_flags & WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE))
+ init_done_with_file_info(stream_list);
- /* Pre-sorting the streams is required for compute_stream_list_stats().
- * Afterwards, read_stream_list() need not sort them again. */
- ret = sort_stream_list_by_sequential_order(stream_list,
- offsetof(struct wim_lookup_table_entry,
- write_streams_list));
- if (ret)
- return ret;
+ memset(&ctx, 0, sizeof(ctx));
ctx.out_fd = out_fd;
ctx.lookup_table = lookup_table;
ctx.write_resource_flags = write_resource_flags;
ctx.filter_ctx = filter_ctx;
- if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
- wimlib_assert(out_chunk_size != 0);
- if (out_chunk_size <= STACK_MAX) {
- ctx.chunk_buf = alloca(out_chunk_size);
- } else {
- ctx.chunk_buf = MALLOC(out_chunk_size);
- if (ctx.chunk_buf == NULL) {
- ret = WIMLIB_ERR_NOMEM;
- goto out_destroy_context;
- }
- }
- }
- ctx.chunk_buf_filled = 0;
+ /*
+ * We normally sort the streams to write by a "sequential" order that is
+ * optimized for reading. But when using solid compression, we instead
+ * sort the streams by file extension and file name (when applicable;
+ * and we don't do this for streams from solid resources) so that
+ * similar files are grouped together, which improves the compression
+ * ratio. This is somewhat of a hack since a stream does not
+ * necessarily correspond one-to-one with a filename, nor is there any
+ * guarantee that two files with similar names or extensions are
+ * actually similar in content. A potential TODO is to sort the streams
+ * based on some measure of similarity of their actual contents.
+ */
+
+ ret = sort_stream_list_by_sequential_order(stream_list,
+ offsetof(struct wim_lookup_table_entry,
+ write_streams_list));
+ if (ret)
+ return ret;
compute_stream_list_stats(stream_list, &ctx);
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID_SORT) {
+ ret = sort_stream_list_for_solid_compression(stream_list);
+ if (unlikely(ret))
+ WARNING("Failed to sort streams for solid compression. Continuing anyways.");
+ }
+
ctx.progress_data.progfunc = progfunc;
ctx.progress_data.progctx = progctx;
out_chunk_size,
num_threads, 0,
&ctx.compressor);
- if (ret) {
- DEBUG("Couldn't create parallel chunk compressor "
- "(status %d)", ret);
+ if (ret > 0) {
+ WARNING("Couldn't create parallel chunk compressor: %"TS".\n"
+ " Falling back to single-threaded compression.",
+ wimlib_get_error_string(ret));
}
}
#endif
ctx.progress_data.progress.write_streams.num_threads);
INIT_LIST_HEAD(&ctx.pending_streams);
- INIT_LIST_HEAD(&ctx.pack_streams);
+ INIT_LIST_HEAD(&ctx.solid_streams);
ret = call_progress(ctx.progress_data.progfunc,
WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
if (ret)
goto out_destroy_context;
- if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
ret = begin_write_resource(&ctx, ctx.num_bytes_to_compress);
if (ret)
goto out_destroy_context;
if (ret)
goto out_destroy_context;
- if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
struct wim_reshdr reshdr;
struct wim_lookup_table_entry *lte;
u64 offset_in_res;
if (ret)
goto out_destroy_context;
- DEBUG("Ending packed resource: %lu %lu %lu.",
+ DEBUG("Ending solid resource: %lu %lu %lu.",
reshdr.offset_in_wim,
reshdr.size_in_wim,
reshdr.uncompressed_size);
offset_in_res = 0;
- list_for_each_entry(lte, &ctx.pack_streams, write_streams_list) {
+ list_for_each_entry(lte, &ctx.solid_streams, write_streams_list) {
lte->out_reshdr.size_in_wim = lte->size;
lte->out_reshdr.flags = filter_resource_flags(lte->flags);
- lte->out_reshdr.flags |= WIM_RESHDR_FLAG_PACKED_STREAMS;
+ lte->out_reshdr.flags |= WIM_RESHDR_FLAG_SOLID;
lte->out_reshdr.uncompressed_size = 0;
lte->out_reshdr.offset_in_wim = offset_in_res;
lte->out_res_offset_in_wim = reshdr.offset_in_wim;
&ctx.progress_data);
out_destroy_context:
- if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && out_chunk_size > STACK_MAX)
- FREE(ctx.chunk_buf);
FREE(ctx.chunk_csizes);
if (ctx.compressor)
ctx.compressor->destroy(ctx.compressor);
}
static int
-is_stream_packed(struct wim_lookup_table_entry *lte, void *_ignore)
+is_stream_in_solid_resource(struct wim_lookup_table_entry *lte, void *_ignore)
{
return lte_is_partial(lte);
}
static bool
-wim_has_packed_streams(WIMStruct *wim)
+wim_has_solid_resources(WIMStruct *wim)
{
- return for_lookup_table_entry(wim->lookup_table, is_stream_packed, NULL);
+ return for_lookup_table_entry(wim->lookup_table,
+ is_stream_in_solid_resource, NULL);
}
static int
write_resource_flags = write_flags_to_resource_flags(write_flags);
- /* wimlib v1.7.0: pack streams by default if the WIM version has been
- * set to WIM_VERSION_PACKED_STREAMS and at least one stream in the
- * WIM's lookup table is located in a packed resource (may be the same
+ /* wimlib v1.7.0: create a solid WIM file by default if the WIM version
+ * has been set to WIM_VERSION_SOLID and at least one stream in the
+ * WIM's lookup table is located in a solid resource (may be the same
* WIM, or a different one in the case of export). */
- if (wim->hdr.wim_version == WIM_VERSION_PACKED_STREAMS &&
- wim_has_packed_streams(wim))
+ if (wim->hdr.wim_version == WIM_VERSION_SOLID &&
+ wim_has_solid_resources(wim))
{
- write_resource_flags |= WRITE_RESOURCE_FLAG_PACK_STREAMS;
+ write_resource_flags |= WRITE_RESOURCE_FLAG_SOLID;
}
- if (write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) {
- out_chunk_size = wim->out_pack_chunk_size;
- out_ctype = wim->out_pack_compression_type;
+ if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
+ out_chunk_size = wim->out_solid_chunk_size;
+ out_ctype = wim->out_solid_compression_type;
} else {
out_chunk_size = wim->out_chunk_size;
out_ctype = wim->out_compression_type;
lte->will_be_in_output_wim = 1;
return write_stream_list(&stream_list,
out_fd,
- write_resource_flags & ~WRITE_RESOURCE_FLAG_PACK_STREAMS,
+ write_resource_flags & ~WRITE_RESOURCE_FLAG_SOLID,
out_ctype,
out_chunk_size,
1,
write_resource_flags = write_flags_to_resource_flags(write_flags);
- write_resource_flags &= ~WRITE_RESOURCE_FLAG_PACK_STREAMS;
+ write_resource_flags &= ~WRITE_RESOURCE_FLAG_SOLID;
DEBUG("Writing metadata resources (offset=%"PRIu64")",
wim->out_fd.offset);
lte1 = *(const struct wim_lookup_table_entry**)p1;
lte2 = *(const struct wim_lookup_table_entry**)p2;
- if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
- if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
+ if (lte1->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
+ if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
if (lte1->out_res_offset_in_wim != lte2->out_res_offset_in_wim)
return cmp_u64(lte1->out_res_offset_in_wim,
lte2->out_res_offset_in_wim);
return 1;
}
} else {
- if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
+ if (lte2->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID)
return -1;
}
return cmp_u64(lte1->out_reshdr.offset_in_wim,
* (private) WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE:
* Don't write the lookup table.
*
- * (private) WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE:
- * When (if) writing the integrity table, re-use entries from the
- * existing integrity table, if possible.
- *
* (private) WIMLIB_WRITE_FLAG_CHECKPOINT_AFTER_XML:
* After writing the XML data but before writing the integrity
* table, write a temporary WIM header and flush the stream so that
* Use the existing <TOTALBYTES> stored in the in-memory XML
* information, rather than setting it to the offset of the XML
* data being written.
+ * (private) WIMLIB_WRITE_FLAG_OVERWRITE
+ * The existing WIM file is being updated in-place. The entries
+ * from its integrity table may be re-used.
*/
static int
finish_write(WIMStruct *wim, int image, int write_flags,
int ret;
off_t hdr_offset;
int write_resource_flags;
- off_t old_lookup_table_end;
+ off_t old_lookup_table_end = 0;
off_t new_lookup_table_end;
u64 xml_totalbytes;
+ struct integrity_table *old_integrity_table = NULL;
DEBUG("image=%d, write_flags=%08x", image, write_flags);
wim->hdr.boot_idx - 1]->metadata_lte->out_reshdr);
}
- /* Write lookup table. (Save old position first.) */
- old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
- wim->hdr.lookup_table_reshdr.size_in_wim;
+ /* If overwriting the WIM file containing an integrity table in-place,
+ * we'd like to re-use the information in the old integrity table
+ * instead of recalculating it. But we might overwrite the old
+ * integrity table when we expand the XML data. Read it into memory
+ * just in case. */
+ if ((write_flags & (WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)) ==
+ (WIMLIB_WRITE_FLAG_OVERWRITE |
+ WIMLIB_WRITE_FLAG_CHECK_INTEGRITY)
+ && wim_has_integrity_table(wim))
+ {
+ old_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
+ wim->hdr.lookup_table_reshdr.size_in_wim;
+ (void)read_integrity_table(wim,
+ old_lookup_table_end - WIM_HEADER_DISK_SIZE,
+ &old_integrity_table);
+ /* If we couldn't read the old integrity table, we can still
+ * re-calculate the full integrity table ourselves. Hence the
+ * ignoring of the return value. */
+ }
+
+ /* Write lookup table. */
if (!(write_flags & WIMLIB_WRITE_FLAG_NO_LOOKUP_TABLE)) {
ret = write_wim_lookup_table(wim, image, write_flags,
&wim->hdr.lookup_table_reshdr,
lookup_table_list);
- if (ret)
+ if (ret) {
+ free_integrity_table(old_integrity_table);
return ret;
+ }
}
/* Write XML data. */
ret = write_wim_xml_data(wim, image, xml_totalbytes,
&wim->hdr.xml_data_reshdr,
write_resource_flags);
- if (ret)
+ if (ret) {
+ free_integrity_table(old_integrity_table);
return ret;
+ }
/* Write integrity table (optional). */
if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
ret = write_wim_header_at_offset(&checkpoint_hdr,
&wim->out_fd, 0);
- if (ret)
+ if (ret) {
+ free_integrity_table(old_integrity_table);
return ret;
+ }
}
- if (!(write_flags & WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE))
- old_lookup_table_end = 0;
-
new_lookup_table_end = wim->hdr.lookup_table_reshdr.offset_in_wim +
wim->hdr.lookup_table_reshdr.size_in_wim;
ret = write_integrity_table(wim,
new_lookup_table_end,
- old_lookup_table_end);
+ old_lookup_table_end,
+ old_integrity_table);
+ free_integrity_table(old_integrity_table);
if (ret)
return ret;
} else {
if (write_flags & WIMLIB_WRITE_FLAG_RETAIN_GUID)
DEBUG("\tRETAIN_GUID");
- if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
- DEBUG("\tPACK_STREAMS");
+ if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
+ DEBUG("\tSOLID");
if (write_flags & WIMLIB_WRITE_FLAG_FILE_DESCRIPTOR)
DEBUG("\tFILE_DESCRIPTOR");
/* Save previous header, then start initializing the new one. */
memcpy(&hdr_save, &wim->hdr, sizeof(struct wim_header));
- /* Set default integrity, pipable, and packed stream flags. */
+ /* Set default integrity, pipable, and solid flags. */
if (!(write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
WIMLIB_WRITE_FLAG_NOT_PIPABLE)))
if (wim_is_pipable(wim)) {
}
if ((write_flags & (WIMLIB_WRITE_FLAG_PIPABLE |
- WIMLIB_WRITE_FLAG_PACK_STREAMS))
+ WIMLIB_WRITE_FLAG_SOLID))
== (WIMLIB_WRITE_FLAG_PIPABLE |
- WIMLIB_WRITE_FLAG_PACK_STREAMS))
+ WIMLIB_WRITE_FLAG_SOLID))
{
- ERROR("Cannot specify both PIPABLE and PACK_STREAMS!");
+ ERROR("Cannot specify both PIPABLE and SOLID!");
return WIMLIB_ERR_INVALID_PARAM;
}
wim->hdr.magic = WIM_MAGIC;
/* Set appropriate version number. */
- if ((write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS) ||
+ if ((write_flags & WIMLIB_WRITE_FLAG_SOLID) ||
wim->out_compression_type == WIMLIB_COMPRESSION_TYPE_LZMS)
- wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
+ wim->hdr.wim_version = WIM_VERSION_SOLID;
else
wim->hdr.wim_version = WIM_VERSION_DEFAULT;
if (wim_has_integrity_table(wim))
write_flags |= WIMLIB_WRITE_FLAG_CHECK_INTEGRITY;
- /* Set WIM version if adding packed streams. */
- if (write_flags & WIMLIB_WRITE_FLAG_PACK_STREAMS)
- wim->hdr.wim_version = WIM_VERSION_PACKED_STREAMS;
+ /* Set WIM version if writing solid resources. */
+ if (write_flags & WIMLIB_WRITE_FLAG_SOLID)
+ wim->hdr.wim_version = WIM_VERSION_SOLID;
/* Set additional flags for overwrite. */
write_flags |= WIMLIB_WRITE_FLAG_OVERWRITE |
* allow any file and metadata resources to appear without returning
* WIMLIB_ERR_RESOURCE_ORDER (due to the fact that we would otherwise
* overwrite these resources). */
- if (!wim->deletion_occurred && !any_images_modified(wim)) {
+ if (!wim->image_deletion_occurred && !any_images_modified(wim)) {
/* If no images have been modified and no images have been
* deleted, a new lookup table does not need to be written. We
* shall write the new XML data and optional integrity table
if (ret)
goto out_truncate;
- write_flags |= WIMLIB_WRITE_FLAG_REUSE_INTEGRITY_TABLE;
ret = finish_write(wim, WIMLIB_ALL_IMAGES, write_flags,
&lookup_table_list);
if (ret)
if (write_flags & WIMLIB_WRITE_FLAG_REBUILD)
return false;
- /* Deletions cause full rebuild by default. */
- if (wim->deletion_occurred && !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
+ /* Image deletions cause full rebuild by default. */
+ if (wim->image_deletion_occurred &&
+ !(write_flags & WIMLIB_WRITE_FLAG_SOFT_DELETE))
return false;
/* Pipable WIMs cannot be updated in place, nor can a non-pipable WIM be