X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fwrite.c;h=0e15da71bffb7b3b7f3ff5f18e4a62b26831fed5;hb=29ac4319aa9c75811cd5629cd3471a681fbeb552;hp=2baba967390bd4cdda51c079c3cb8f66f409449b;hpb=b98b25b85877d1bccdc8673a23576b1fac0ab1c6;p=wimlib diff --git a/src/write.c b/src/write.c index 2baba967..0e15da71 100644 --- a/src/write.c +++ b/src/write.c @@ -8,20 +8,18 @@ /* * Copyright (C) 2012, 2013, 2014 Eric Biggers * - * This file is part of wimlib, a library for working with WIM files. - * - * 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 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. + * + * 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 @@ -43,6 +41,7 @@ #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" #ifdef __WIN32__ @@ -64,6 +63,7 @@ #define WRITE_RESOURCE_FLAG_RECOMPRESS 0x00000001 #define WRITE_RESOURCE_FLAG_PIPABLE 0x00000002 #define WRITE_RESOURCE_FLAG_PACK_STREAMS 0x00000004 +#define WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE 0x00000008 static inline int write_flags_to_resource_flags(int write_flags) @@ -76,6 +76,8 @@ write_flags_to_resource_flags(int write_flags) 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_SEND_DONE_WITH_FILE_MESSAGES) + write_resource_flags |= WRITE_RESOURCE_FLAG_SEND_DONE_WITH_FILE; return write_resource_flags; } @@ -98,12 +100,15 @@ static int 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) @@ -279,7 +284,6 @@ struct write_streams_progress_data { 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) @@ -312,10 +316,28 @@ do_write_streams_progress(struct write_streams_progress_data *progress_data, 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; @@ -366,12 +388,6 @@ struct write_streams_ctx { * @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; - /* Current uncompressed offset in the stream being read. */ u64 cur_read_stream_offset; @@ -541,7 +557,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, 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 { @@ -565,7 +581,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 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; @@ -574,7 +590,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, 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; @@ -582,7 +598,7 @@ end_chunk_table(struct write_streams_ctx *ctx, u64 res_actual_size, return 0; -error: +write_error: ERROR_WITH_ERRNO("Write error"); return ret; } @@ -618,10 +634,96 @@ end_write_resource(struct write_streams_ctx *ctx, struct wim_reshdr *out_reshdr) 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; @@ -642,11 +744,8 @@ write_stream_begin_read(struct wim_lookup_table_entry *lte, * 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); @@ -667,7 +766,7 @@ write_stream_begin_read(struct wim_lookup_table_entry *lte, 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); @@ -675,6 +774,8 @@ write_stream_begin_read(struct wim_lookup_table_entry *lte, lte_new->out_refcnt += lte->out_refcnt; if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) ctx->cur_write_res_size -= lte->size; + if (!ret) + ret = done_with_stream(lte, ctx); free_lookup_table_entry(lte); if (ret) return ret; @@ -692,9 +793,10 @@ write_stream_begin_read(struct wim_lookup_table_entry *lte, <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; } } @@ -751,6 +853,43 @@ write_stream_uncompressed(struct wim_lookup_table_entry *lte, 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 + * block 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). */ + if ((lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) && + (lte->out_reshdr.size_in_wim != lte->out_reshdr.uncompressed_size)) + return false; + + return true; +} + /* Write the next chunk of (typically compressed) data to the output WIM, * handling the writing of the chunk table. */ static int @@ -804,14 +943,14 @@ write_chunk(struct write_streams_ctx *ctx, const void *cchunk, 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; @@ -820,36 +959,32 @@ write_chunk(struct write_streams_ctx *ctx, const void *cchunk, if (ctx->write_resource_flags & WRITE_RESOURCE_FLAG_PACK_STREAMS) { /* Wrote chunk in packed 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->pack_streams); completed_stream_count++; + + lte = next_lte; } } else { /* Wrote chunk in non-packed 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); @@ -861,18 +996,7 @@ write_chunk(struct write_streams_ctx *ctx, const void *cchunk, 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. */ + if (should_rewrite_stream_uncompressed(ctx, lte)) { DEBUG("Stream of size %"PRIu64" did not compress to " "less than original size; writing uncompressed.", lte->size); @@ -883,14 +1007,20 @@ write_chunk(struct write_streams_ctx *ctx, const void *cchunk, 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; } @@ -1006,11 +1136,27 @@ static int 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; @@ -1177,7 +1323,7 @@ write_raw_copy_resources(struct list_head *raw_copy_streams, 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; @@ -1229,6 +1375,25 @@ remove_zero_length_streams(struct list_head *stream_list) } } +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. * @@ -1364,6 +1529,11 @@ write_stream_list(struct list_head *stream_list, return 0; } + /* 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); + memset(&ctx, 0, sizeof(ctx)); /* Pre-sorting the streams is required for compute_stream_list_stats().