]> wimlib.net Git - wimlib/blobdiff - src/extract.c
Ensure WIM has filename before doing WIMBoot extraction
[wimlib] / src / extract.c
index 138623b630da38bfb86331c932132d8a94f8073e..edac626e4e6eb94331d46435caf9d1360785ce39 100644 (file)
@@ -67,9 +67,8 @@
 #include <sys/stat.h>
 #include <unistd.h>
 
-#define WIMLIB_EXTRACT_FLAG_MULTI_IMAGE 0x80000000
-#define WIMLIB_EXTRACT_FLAG_FROM_PIPE   0x40000000
-#define WIMLIB_EXTRACT_FLAG_IMAGEMODE   0x20000000
+#define WIMLIB_EXTRACT_FLAG_FROM_PIPE   0x80000000
+#define WIMLIB_EXTRACT_FLAG_IMAGEMODE   0x40000000
 
 /* Keep in sync with wimlib.h  */
 #define WIMLIB_EXTRACT_MASK_PUBLIC                             \
@@ -114,7 +113,6 @@ dentry_is_supported(struct wim_dentry *dentry,
 
 
 #define PWM_ALLOW_WIM_HDR 0x00001
-#define PWM_SILENT_EOF   0x00002
 
 /* Read the header from a stream in a pipable WIM.  */
 static int
@@ -133,7 +131,9 @@ read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
        if (ret)
                goto read_error;
 
-       if ((flags & PWM_ALLOW_WIM_HDR) && buf.stream_hdr.magic == PWM_MAGIC) {
+       if ((flags & PWM_ALLOW_WIM_HDR) &&
+           le64_to_cpu(buf.stream_hdr.magic) == PWM_MAGIC)
+       {
                BUILD_BUG_ON(sizeof(buf.pwm_hdr) < sizeof(buf.stream_hdr));
                ret = full_read(&pwm->in_fd, &buf.stream_hdr + 1,
                                sizeof(buf.pwm_hdr) - sizeof(buf.stream_hdr));
@@ -164,8 +164,7 @@ read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
        return 0;
 
 read_error:
-       if (ret != WIMLIB_ERR_UNEXPECTED_END_OF_FILE || !(flags & PWM_SILENT_EOF))
-               ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
+       ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
        return ret;
 }
 
@@ -218,7 +217,7 @@ load_streams_from_pipe(struct apply_ctx *ctx,
                        lte_unbind_wim_resource_spec(found_lte);
                        lte_bind_wim_resource_spec(needed_lte, rspec);
 
-                       ret = (*cbs->begin_stream)(needed_lte, 0,
+                       ret = (*cbs->begin_stream)(needed_lte,
                                                   cbs->begin_stream_ctx);
                        if (ret) {
                                lte_unbind_wim_resource_spec(needed_lte);
@@ -267,28 +266,73 @@ out:
        return ret;
 }
 
+/* Creates a temporary file opened for writing.  The open file descriptor is
+ * returned in @fd_ret and its name is returned in @name_ret (dynamically
+ * allocated).  */
+static int
+create_temporary_file(struct filedes *fd_ret, tchar **name_ret)
+{
+       tchar *name;
+       int open_flags;
+       int raw_fd;
+
+retry:
+       name = ttempnam(NULL, T("wimlib"));
+       if (!name) {
+               ERROR_WITH_ERRNO("Failed to create temporary filename");
+               return WIMLIB_ERR_NOMEM;
+       }
+
+       open_flags = O_WRONLY | O_CREAT | O_EXCL | O_BINARY;
+#ifdef __WIN32__
+       open_flags |= _O_SHORT_LIVED;
+#endif
+       raw_fd = topen(name, open_flags, 0600);
+
+       if (raw_fd < 0) {
+               if (errno == EEXIST) {
+                       FREE(name);
+                       goto retry;
+               }
+               ERROR_WITH_ERRNO("Failed to create temporary file "
+                                "\"%"TS"\"", name);
+               FREE(name);
+               return WIMLIB_ERR_OPEN;
+       }
+
+       filedes_init(fd_ret, raw_fd);
+       *name_ret = name;
+       return 0;
+}
+
 static int
-begin_extract_stream_with_progress(struct wim_lookup_table_entry *lte,
-                                  u32 flags, void *_ctx)
+begin_extract_stream_wrapper(struct wim_lookup_table_entry *lte, void *_ctx)
 {
        struct apply_ctx *ctx = _ctx;
 
        ctx->cur_stream = lte;
+       ctx->cur_stream_offset = 0;
 
-       return (*ctx->saved_cbs->begin_stream)(lte, flags,
-                                              ctx->saved_cbs->begin_stream_ctx);
+       if (unlikely(lte->out_refcnt > MAX_OPEN_STREAMS))
+               return create_temporary_file(&ctx->tmpfile_fd, &ctx->tmpfile_name);
+       else
+               return (*ctx->saved_cbs->begin_stream)(lte, ctx->saved_cbs->begin_stream_ctx);
 }
 
 static int
-consume_chunk_with_progress(const void *chunk, size_t size, void *_ctx)
+extract_chunk_wrapper(const void *chunk, size_t size, void *_ctx)
 {
        struct apply_ctx *ctx = _ctx;
        union wimlib_progress_info *progress = &ctx->progress;
        int ret;
 
+       ctx->cur_stream_offset += size;
+
        if (likely(ctx->supported_features.hard_links)) {
                progress->extract.completed_bytes +=
                        (u64)size * ctx->cur_stream->out_refcnt;
+               if (ctx->cur_stream_offset == ctx->cur_stream->size)
+                       progress->extract.completed_streams += ctx->cur_stream->out_refcnt;
        } else {
                const struct stream_owner *owners = stream_owners(ctx->cur_stream);
                for (u32 i = 0; i < ctx->cur_stream->out_refcnt; i++) {
@@ -300,6 +344,8 @@ consume_chunk_with_progress(const void *chunk, size_t size, void *_ctx)
                                            d_extraction_alias_node)
                        {
                                progress->extract.completed_bytes += size;
+                               if (ctx->cur_stream_offset == ctx->cur_stream->size)
+                                       progress->extract.completed_streams++;
                        }
                }
        }
@@ -314,13 +360,109 @@ consume_chunk_with_progress(const void *chunk, size_t size, void *_ctx)
                {
                        ctx->next_progress = UINT64_MAX;
                } else {
-                       ctx->next_progress += progress->extract.total_bytes / 128;
-                       if (ctx->next_progress > progress->extract.total_bytes)
+                       /* Send new message as soon as another 1/128 of the
+                        * total has been extracted.  (Arbitrary number.)  */
+                       ctx->next_progress =
+                               progress->extract.completed_bytes +
+                                       progress->extract.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->extract.completed_bytes + 5000000 <
+                           ctx->next_progress)
+                               ctx->next_progress =
+                                       progress->extract.completed_bytes + 5000000;
+
+                       /* ... But always send a message as soon as we're
+                        * completely done.  */
+                       if (progress->extract.total_bytes < ctx->next_progress)
                                ctx->next_progress = progress->extract.total_bytes;
                }
        }
-       return (*ctx->saved_cbs->consume_chunk)(chunk, size,
-                                               ctx->saved_cbs->consume_chunk_ctx);
+
+       if (unlikely(filedes_valid(&ctx->tmpfile_fd))) {
+               /* Just extracting to temporary file for now.  */
+               ret = full_write(&ctx->tmpfile_fd, chunk, size);
+               if (ret) {
+                       ERROR_WITH_ERRNO("Error writing data to "
+                                        "temporary file \"%"TS"\"",
+                                        ctx->tmpfile_name);
+               }
+               return ret;
+       } else {
+               return (*ctx->saved_cbs->consume_chunk)(chunk, size,
+                                                       ctx->saved_cbs->consume_chunk_ctx);
+       }
+}
+
+static int
+extract_from_tmpfile(const tchar *tmpfile_name, struct apply_ctx *ctx)
+{
+       struct wim_lookup_table_entry tmpfile_lte;
+       struct wim_lookup_table_entry *orig_lte = ctx->cur_stream;
+       const struct stream_owner *owners = stream_owners(orig_lte);
+       const struct read_stream_list_callbacks *cbs = ctx->saved_cbs;
+       int ret;
+
+       /* Copy the stream's data from the temporary file to each of its
+        * destinations.
+        *
+        * This is executed only in the very uncommon case that a
+        * single-instance stream is being extracted to more than
+        * MAX_OPEN_STREAMS locations!  */
+
+       memcpy(&tmpfile_lte, orig_lte, sizeof(struct wim_lookup_table_entry));
+       tmpfile_lte.resource_location = RESOURCE_IN_FILE_ON_DISK;
+       tmpfile_lte.file_on_disk = ctx->tmpfile_name;
+       tmpfile_lte.out_refcnt = 1;
+
+       for (u32 i = 0; i < orig_lte->out_refcnt; i++) {
+               tmpfile_lte.inline_stream_owners[0] = owners[i];
+
+               /* Note: it usually doesn't matter whether we pass the original
+                * stream entry to callbacks provided by the extraction backend
+                * as opposed to the tmpfile stream entry, since they shouldn't
+                * actually read data from the stream other than through the
+                * read_stream_prefix() call below.  But for
+                * WIMLIB_EXTRACT_FLAG_WIMBOOT mode on Windows it does matter
+                * because it needs the original stream location in order to
+                * create the external backing reference.  */
+
+               ret = (*cbs->begin_stream)(orig_lte, cbs->begin_stream_ctx);
+               if (ret)
+                       return ret;
+
+               /* Extra SHA-1 isn't necessary here, but it shouldn't hurt as
+                * this case is very rare anyway.  */
+               ret = extract_stream(&tmpfile_lte, tmpfile_lte.size,
+                                    cbs->consume_chunk,
+                                    cbs->consume_chunk_ctx);
+
+               return (*cbs->end_stream)(orig_lte, ret,
+                                         cbs->end_stream_ctx);
+       }
+       return 0;
+}
+
+static int
+end_extract_stream_wrapper(struct wim_lookup_table_entry *stream,
+                          int status, void *_ctx)
+{
+       struct apply_ctx *ctx = _ctx;
+
+       if (unlikely(filedes_valid(&ctx->tmpfile_fd))) {
+               filedes_close(&ctx->tmpfile_fd);
+               if (!status)
+                       status = extract_from_tmpfile(ctx->tmpfile_name, ctx);
+               filedes_invalidate(&ctx->tmpfile_fd);
+               tunlink(ctx->tmpfile_name);
+               FREE(ctx->tmpfile_name);
+               return status;
+       } else {
+               return (*ctx->saved_cbs->end_stream)(stream, status,
+                                                    ctx->saved_cbs->end_stream_ctx);
+       }
 }
 
 /*
@@ -333,30 +475,34 @@ consume_chunk_with_progress(const void *chunk, size_t size, void *_ctx)
  *
  * This also works if the WIM is being read from a pipe, whereas attempting to
  * read streams directly (e.g. with read_full_stream_into_buf()) will not.
+ *
+ * This also will split up streams that will need to be extracted to more than
+ * MAX_OPEN_STREAMS locations, as measured by the 'out_refcnt' of each stream.
+ * Therefore, the apply_operations implementation need not worry about running
+ * out of file descriptors, unless it might open more than one file descriptor
+ * per nominal destination (e.g. Win32 currently might because the destination
+ * file system might not support hard links).
  */
 int
 extract_stream_list(struct apply_ctx *ctx,
                    const struct read_stream_list_callbacks *cbs)
 {
        struct read_stream_list_callbacks wrapper_cbs = {
-               .begin_stream      = begin_extract_stream_with_progress,
+               .begin_stream      = begin_extract_stream_wrapper,
                .begin_stream_ctx  = ctx,
-               .consume_chunk     = consume_chunk_with_progress,
+               .consume_chunk     = extract_chunk_wrapper,
                .consume_chunk_ctx = ctx,
-               .end_stream        = cbs->end_stream,
-               .end_stream_ctx    = cbs->end_stream_ctx,
+               .end_stream        = end_extract_stream_wrapper,
+               .end_stream_ctx    = ctx,
        };
-       if (ctx->progfunc) {
-               ctx->saved_cbs = cbs;
-               cbs = &wrapper_cbs;
-       }
+       ctx->saved_cbs = cbs;
        if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
-               return load_streams_from_pipe(ctx, cbs);
+               return load_streams_from_pipe(ctx, &wrapper_cbs);
        } else {
                return read_stream_list(&ctx->stream_list,
                                        offsetof(struct wim_lookup_table_entry,
                                                 extraction_list),
-                                       cbs, VERIFY_STREAM_HASHES);
+                                       &wrapper_cbs, VERIFY_STREAM_HASHES);
        }
 }
 
@@ -442,7 +588,7 @@ remove_contained_trees(struct wim_dentry **trees, size_t num_trees)
        for (i = 0; i < num_trees; i++) {
                struct wim_dentry *d = trees[i];
                while (!dentry_is_root(d)) {
-                       d = d->parent;
+                       d = d->d_parent;
                        if (d->tmp_flag)
                                goto tree_contained;
                }
@@ -509,7 +655,7 @@ build_dentry_list(struct list_head *dentry_list, struct wim_dentry **trees,
                        place_after = dentry_list;
                        ancestor = dentry;
                        do {
-                               ancestor = ancestor->parent;
+                               ancestor = ancestor->d_parent;
                                if (will_extract_dentry(ancestor)) {
                                        place_after = &ancestor->d_extraction_list_node;
                                        break;
@@ -518,7 +664,7 @@ build_dentry_list(struct list_head *dentry_list, struct wim_dentry **trees,
 
                        ancestor = dentry;
                        do {
-                               ancestor = ancestor->parent;
+                               ancestor = ancestor->d_parent;
                                if (will_extract_dentry(ancestor))
                                        break;
                                list_add(&ancestor->d_extraction_list_node, place_after);
@@ -613,12 +759,14 @@ dentry_calculate_extraction_name(struct wim_dentry *dentry,
        if (dentry_is_root(dentry))
                return 0;
 
+#ifdef WITH_NTFS_3G
        if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
                dentry->d_extraction_name = dentry->file_name;
                dentry->d_extraction_name_nchars = dentry->file_name_nbytes /
                                                   sizeof(utf16lechar);
                return 0;
        }
+#endif
 
        if (!ctx->supported_features.case_sensitive_filenames) {
                struct wim_dentry *other;
@@ -820,7 +968,7 @@ ref_stream(struct wim_lookup_table_entry *lte, u32 stream_idx,
                return 0;
 
        ctx->progress.extract.total_bytes += lte->size;
-       ctx->progress.extract.num_streams++;
+       ctx->progress.extract.total_streams++;
 
        if (inode->i_visited)
                return 0;
@@ -1134,6 +1282,13 @@ do_feature_check(const struct wim_features *required_features,
                return WIMLIB_ERR_UNSUPPORTED;
        }
 
+       if (required_features->unix_data &&
+           !(extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA))
+       {
+               WARNING("Ignoring UNIX metadata of %lu files",
+                       required_features->unix_data);
+       }
+
        /* DOS Names.  */
        if (required_features->short_names &&
            !supported_features->short_names)
@@ -1221,6 +1376,7 @@ extract_trees(WIMStruct *wim, struct wim_dentry **trees, size_t num_trees,
                ctx->progress.extract.target = target;
        }
        INIT_LIST_HEAD(&ctx->stream_list);
+       filedes_invalidate(&ctx->tmpfile_fd);
 
        ret = (*ops->get_supported_features)(target, &ctx->supported_features);
        if (ret)
@@ -1355,12 +1511,16 @@ check_extract_flags(const WIMStruct *wim, int *extract_flags_p)
        }
 #endif
 
-#ifndef __WIN32__
        if (extract_flags & WIMLIB_EXTRACT_FLAG_WIMBOOT) {
+#ifdef __WIN32__
+               if (!wim->filename)
+                       return WIMLIB_ERR_NO_FILENAME;
+#else
                ERROR("WIMBoot extraction is only supported on Windows!");
                return WIMLIB_ERR_UNSUPPORTED;
-       }
 #endif
+       }
+
 
        if ((extract_flags & (WIMLIB_EXTRACT_FLAG_RPFIX |
                              WIMLIB_EXTRACT_FLAG_NORPFIX |
@@ -1563,8 +1723,6 @@ extract_all_images(WIMStruct *wim, const tchar *target, int extract_flags)
        int image;
        const tchar *image_name;
 
-       extract_flags |= WIMLIB_EXTRACT_FLAG_MULTI_IMAGE;
-
        if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
                ERROR("Cannot extract multiple images in NTFS extraction mode.");
                return WIMLIB_ERR_INVALID_PARAM;