]> wimlib.net Git - wimlib/blobdiff - src/resource.c
Make different threads use different FILE*'s
[wimlib] / src / resource.c
index ffcd7530a35f5ab590ad4ac9cc226701d7c389a1..1871cfcccb8617ba005124652507c555f933f480 100644 (file)
@@ -27,6 +27,8 @@
 #include <stdlib.h>
 #include <stdarg.h>
 
+#include "dentry.h"
+
 #ifdef WITH_NTFS_3G
 #include <ntfs-3g/attrib.h>
 #include <ntfs-3g/inode.h>
@@ -39,7 +41,6 @@
 #include "lzx.h"
 #include "xpress.h"
 #include "sha1.h"
-#include "dentry.h"
 #include <unistd.h>
 #include <errno.h>
 #ifdef HAVE_ALLOCA_H
@@ -420,6 +421,57 @@ u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
        return p;
 }
 
+static FILE *wim_get_fp(WIMStruct *w)
+{
+       pthread_mutex_lock(&w->fp_tab_mutex);
+       FILE *fp;
+
+       wimlib_assert(w->filename != NULL);
+
+       for (size_t i = 0; i < w->num_allocated_fps; i++) {
+               if (w->fp_tab[i]) {
+                       fp = w->fp_tab[i];
+                       w->fp_tab[i] = NULL;
+                       goto out;
+               }
+       }
+       DEBUG("Opening extra file descriptor to `%s'", w->filename);
+       fp = fopen(w->filename, "rb");
+       if (!fp)
+               ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
+out:
+       pthread_mutex_unlock(&w->fp_tab_mutex);
+       return fp;
+}
+
+static int wim_release_fp(WIMStruct *w, FILE *fp)
+{
+       int ret = 0;
+       FILE **fp_tab;
+
+       pthread_mutex_lock(&w->fp_tab_mutex);
+
+       for (size_t i = 0; i < w->num_allocated_fps; i++) {
+               if (w->fp_tab[i] == NULL) {
+                       w->fp_tab[i] = fp;
+                       goto out;
+               }
+       }
+
+       fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
+       if (!fp_tab) {
+               ret = WIMLIB_ERR_NOMEM;
+               goto out;
+       }
+       w->fp_tab = fp_tab;
+       memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
+       w->fp_tab[w->num_allocated_fps] = fp;
+       w->num_allocated_fps += 4;
+out:
+       pthread_mutex_unlock(&w->fp_tab_mutex);
+       return ret;
+}
+
 /*
  * Reads some data from the resource corresponding to a WIM lookup table entry.
  *
@@ -427,47 +479,61 @@ u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
  * @buf:       Buffer into which to write the data.
  * @size:      Number of bytes to read.
  * @offset:    Offset at which to start reading the resource.
- * @raw:       If %true, compressed data is read literally rather than being
- *                     decompressed first.
  *
  * Returns zero on success, nonzero on failure.
  */
 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
-                     size_t size, u64 offset, bool raw)
+                     size_t size, u64 offset, int flags)
 {
+       int ctype;
+       int ret = 0;
+       FILE *fp;
+
        /* We shouldn't be allowing read over-runs in any part of the library.
         * */
-       if (raw)
+       if (flags & WIMLIB_RESOURCE_FLAG_RAW)
                wimlib_assert(offset + size <= lte->resource_entry.size);
        else
                wimlib_assert(offset + size <= lte->resource_entry.original_size);
 
-       int ctype;
-       int ret;
-       FILE *fp;
        switch (lte->resource_location) {
        case RESOURCE_IN_WIM:
                /* The resource is in a WIM file, and its WIMStruct is given by
                 * the lte->wim member.  The resource may be either compressed
                 * or uncompressed. */
-               wimlib_assert(lte->wim);
-               wimlib_assert(lte->wim->fp);
+               wimlib_assert(lte->wim != NULL);
+
+               if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
+                       fp = wim_get_fp(lte->wim);
+                       if (!fp)
+                               return WIMLIB_ERR_OPEN;
+               } else {
+                       wimlib_assert(lte->wim->fp != NULL);
+                       fp = lte->wim->fp;
+               }
+
                ctype = wim_resource_compression_type(lte);
 
                wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
                              (lte->resource_entry.original_size ==
                               lte->resource_entry.size));
 
-               if (raw || ctype == WIM_COMPRESSION_TYPE_NONE)
-                       return read_uncompressed_resource(lte->wim->fp,
-                                                         lte->resource_entry.offset + offset,
-                                                         size, buf);
+               if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
+                   || ctype == WIM_COMPRESSION_TYPE_NONE)
+                       ret = read_uncompressed_resource(fp,
+                                                        lte->resource_entry.offset + offset,
+                                                        size, buf);
                else
-                       return read_compressed_resource(lte->wim->fp,
-                                                       lte->resource_entry.size,
-                                                       lte->resource_entry.original_size,
-                                                       lte->resource_entry.offset,
-                                                       ctype, size, offset, buf);
+                       ret = read_compressed_resource(fp,
+                                                      lte->resource_entry.size,
+                                                      lte->resource_entry.original_size,
+                                                      lte->resource_entry.offset,
+                                                      ctype, size, offset, buf);
+               if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
+                       int ret2 = wim_release_fp(lte->wim, fp);
+                       if (ret == 0)
+                               ret = ret2;
+               }
                break;
        case RESOURCE_IN_STAGING_FILE:
        case RESOURCE_IN_FILE_ON_DISK:
@@ -484,46 +550,42 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
                        if (!fp) {
                                ERROR_WITH_ERRNO("Failed to open the file "
                                                 "`%s'", lte->file_on_disk);
-                               return WIMLIB_ERR_OPEN;
+                               ret = WIMLIB_ERR_OPEN;
+                               break;
                        }
                }
                ret = read_uncompressed_resource(fp, offset, size, buf);
                if (fp != lte->file_on_disk_fp)
                        fclose(fp);
-               return ret;
                break;
        case RESOURCE_IN_ATTACHED_BUFFER:
                /* The resource is directly attached uncompressed in an
                 * in-memory buffer. */
-               wimlib_assert(lte->attached_buffer);
+               wimlib_assert(lte->attached_buffer != NULL);
                memcpy(buf, lte->attached_buffer + offset, size);
-               return 0;
                break;
 #ifdef WITH_NTFS_3G
        case RESOURCE_IN_NTFS_VOLUME:
-               wimlib_assert(lte->ntfs_loc);
-               if (lte->attr) {
-                       u64 adjusted_offset;
+               wimlib_assert(lte->ntfs_loc != NULL);
+               wimlib_assert(lte->attr != NULL);
+               {
                        if (lte->ntfs_loc->is_reparse_point)
-                               adjusted_offset = offset + 8;
-                       else
-                               adjusted_offset = offset;
-                       if (ntfs_attr_pread(lte->attr, offset, size, buf) == size) {
-                               return 0;
-                       } else {
+                               offset += 8;
+                       if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
                                ERROR_WITH_ERRNO("Error reading NTFS attribute "
                                                 "at `%s'",
                                                 lte->ntfs_loc->path_utf8);
-                               return WIMLIB_ERR_NTFS_3G;
+                               ret = WIMLIB_ERR_NTFS_3G;
                        }
-               } else {
-                       wimlib_assert(0);
+                       break;
                }
-               break;
 #endif
        default:
-               assert(0);
+               wimlib_assert(0);
+               ret = -1;
+               break;
        }
+       return ret;
 }
 
 /*
@@ -536,9 +598,10 @@ int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
  *
  * Returns 0 on success; nonzero on failure.
  */
-int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[])
+int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
+                          int flags)
 {
-       return read_wim_resource(lte, buf, wim_resource_size(lte), 0, false);
+       return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
 }
 
 /* Chunk table that's located at the beginning of each compressed resource in
@@ -571,7 +634,6 @@ begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
        struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
        int ret;
 
-
        if (!chunk_tab) {
                ERROR("Failed to allocate chunk table for %"PRIu64" byte "
                      "resource", size);
@@ -751,14 +813,15 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
  */
 static int write_wim_resource(struct lookup_table_entry *lte,
                              FILE *out_fp, int out_ctype,
-                             struct resource_entry *out_res_entry)
+                             struct resource_entry *out_res_entry,
+                             int flags)
 {
        u64 bytes_remaining;
        u64 original_size;
        u64 old_compressed_size;
        u64 new_compressed_size;
-       u64 offset = 0;
-       int ret = 0;
+       u64 offset;
+       int ret;
        struct chunk_table *chunk_tab = NULL;
        bool raw;
        off_t file_offset;
@@ -786,10 +849,14 @@ static int write_wim_resource(struct lookup_table_entry *lte,
         * without decompressing and recompressing the data). */
        raw = (wim_resource_compression_type(lte) == out_ctype
               && out_ctype != WIM_COMPRESSION_TYPE_NONE);
-       if (raw)
+
+       if (raw) {
+               flags |= WIMLIB_RESOURCE_FLAG_RAW;
                bytes_remaining = old_compressed_size;
-       else
+       } else {
+               flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
                bytes_remaining = original_size;
+       }
 
        /* Empty resource; nothing needs to be done, so just return success. */
        if (bytes_remaining == 0)
@@ -859,9 +926,10 @@ static int write_wim_resource(struct lookup_table_entry *lte,
        /* While there are still bytes remaining in the WIM resource, read a
         * chunk of the resource, update SHA1, then write that chunk using the
         * desired compression type. */
+       offset = 0;
        do {
                u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
-               ret = read_wim_resource(lte, buf, to_read, offset, raw);
+               ret = read_wim_resource(lte, buf, to_read, offset, flags);
                if (ret != 0)
                        goto out_fclose;
                if (!raw)
@@ -930,7 +998,7 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                        goto out_fclose;
                }
                ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
-                                        out_res_entry);
+                                        out_res_entry, flags);
                if (ret != 0)
                        goto out_fclose;
                if (fflush(out_fp) != 0) {
@@ -941,22 +1009,23 @@ static int write_wim_resource(struct lookup_table_entry *lte,
                if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) {
                        ERROR_WITH_ERRNO("Failed to truncate output WIM file");
                        ret = WIMLIB_ERR_WRITE;
+                       goto out_fclose;
+               }
+       } else {
+               if (out_res_entry) {
+                       out_res_entry->size          = new_compressed_size;
+                       out_res_entry->original_size = original_size;
+                       out_res_entry->offset        = file_offset;
+                       out_res_entry->flags         = lte->resource_entry.flags
+                                                       & ~WIM_RESHDR_FLAG_COMPRESSED;
+                       if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
+                               out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
                }
-               goto out_fclose;
-       }
-       wimlib_assert(new_compressed_size <= original_size || raw);
-       if (out_res_entry) {
-               out_res_entry->size          = new_compressed_size;
-               out_res_entry->original_size = original_size;
-               out_res_entry->offset        = file_offset;
-               out_res_entry->flags         = lte->resource_entry.flags
-                                               & ~WIM_RESHDR_FLAG_COMPRESSED;
-               if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
-                       out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
        }
+       ret = 0;
 out_fclose:
        if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
-            && lte->file_on_disk_fp) {
+           && lte->file_on_disk_fp) {
                fclose(lte->file_on_disk_fp);
                lte->file_on_disk_fp = NULL;
        }
@@ -965,9 +1034,9 @@ out_fclose:
                if (lte->attr) {
                        ntfs_attr_close(lte->attr);
                        lte->attr = NULL;
-               } if (ni) {
-                       ntfs_inode_close(ni);
                }
+               if (ni)
+                       ntfs_inode_close(ni);
        }
 #endif
 out:
@@ -983,7 +1052,7 @@ static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
                                          struct resource_entry *out_res_entry,
                                          u8 hash[SHA1_HASH_SIZE])
 {
-       /* Set up a temporary lookup table entry that we provide to
+       /* Set up a temporary lookup table entry to provide to
         * write_wim_resource(). */
        struct lookup_table_entry lte;
        int ret;
@@ -995,7 +1064,7 @@ static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
        lte.attached_buffer              = (u8*)buf;
 
        zero_out_hash(lte.hash);
-       ret = write_wim_resource(&lte, out_fp, out_ctype, out_res_entry);
+       ret = write_wim_resource(&lte, out_fp, out_ctype, out_res_entry, 0);
        if (ret != 0)
                return ret;
        copy_hash(hash, lte.hash);
@@ -1022,7 +1091,7 @@ int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
 
        while (bytes_remaining) {
                u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
-               ret = read_wim_resource(lte, buf, to_read, offset, false);
+               ret = read_wim_resource(lte, buf, to_read, offset, 0);
                if (ret != 0)
                        break;
                sha1_update(&ctx, buf, to_read);
@@ -1076,7 +1145,7 @@ int copy_resource(struct lookup_table_entry *lte, void *wim)
 
        ret = write_wim_resource(lte, w->out_fp,
                                 wim_resource_compression_type(lte),
-                                &lte->output_resource_entry);
+                                &lte->output_resource_entry, 0);
        if (ret != 0)
                return ret;
        lte->out_refcnt = lte->refcnt;
@@ -1109,7 +1178,7 @@ int write_dentry_resources(struct dentry *dentry, void *wim_p)
                lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table);
                if (lte && ++lte->out_refcnt == 1) {
                        ret = write_wim_resource(lte, w->out_fp, ctype,
-                                                &lte->output_resource_entry);
+                                                &lte->output_resource_entry, 0);
                        if (ret != 0)
                                break;
                }
@@ -1164,6 +1233,12 @@ int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
                return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
        }
 
+       if (sizeof(size_t) < 8 && metadata_len > 0xffffffff) {
+               ERROR("Metadata resource is too large (%"PRIu64" bytes",
+                     metadata_len);
+               return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
+       }
+
        /* Allocate memory for the uncompressed metadata resource. */
        buf = MALLOC(metadata_len);
 
@@ -1174,7 +1249,7 @@ int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
        }
 
        /* Read the metadata resource into memory.  (It may be compressed.) */
-       ret = read_full_wim_resource(metadata_lte, buf);
+       ret = read_full_wim_resource(metadata_lte, buf, 0);
        if (ret != 0)
                goto out_free_buf;
 
@@ -1191,14 +1266,18 @@ int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
         * and if successful, go ahead and calculate the offset in the metadata
         * resource of the root dentry. */
 
+       wimlib_assert(imd->security_data == NULL);
        ret = read_security_data(buf, metadata_len, &imd->security_data);
        if (ret != 0)
                goto out_free_buf;
 
-       get_u32(buf, &dentry_offset);
-       if (dentry_offset == 0)
-               dentry_offset = 8;
-       dentry_offset = (dentry_offset + 7) & ~7;
+       dentry_offset = (imd->security_data->total_length + 7) & ~7;
+
+       if (dentry_offset == 0) {
+               ERROR("Integer overflow while reading metadata resource");
+               ret = WIMLIB_ERR_INVALID_SECURITY_DATA;
+               goto out_free_security_data;
+       }
 
        /* Allocate memory for the root dentry and read it into memory */
        dentry = MALLOC(sizeof(struct dentry));
@@ -1211,10 +1290,9 @@ int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
 
        ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
 
-       /* This is the root dentry, so set its pointers correctly. */
+       /* This is the root dentry, so set its parent to itself. */
        dentry->parent = dentry;
-       dentry->next   = dentry;
-       dentry->prev   = dentry;
+
        if (ret != 0)
                goto out_free_dentry_tree;
        inode_add_dentry(dentry, dentry->d_inode);
@@ -1277,7 +1355,6 @@ int write_metadata_resource(WIMStruct *w)
        struct lookup_table_entry *lte;
        u64 metadata_original_size;
        const struct wim_security_data *sd;
-       const unsigned random_tail_len = 20;
 
        DEBUG("Writing metadata resource for image %d", w->current_image);
 
@@ -1286,7 +1363,8 @@ int write_metadata_resource(WIMStruct *w)
 
        /* We do not allow the security data pointer to be NULL, although it may
         * point to an empty security data with no entries. */
-       wimlib_assert(sd);
+       wimlib_assert(root != NULL);
+       wimlib_assert(sd != NULL);
 
        /* Offset of first child of the root dentry.  It's equal to:
         * - The total length of the security data, rounded to the next 8-byte
@@ -1302,7 +1380,7 @@ int write_metadata_resource(WIMStruct *w)
        calculate_subdir_offsets(root, &subdir_offset);
 
        /* Total length of the metadata resource (uncompressed) */
-       metadata_original_size = subdir_offset + random_tail_len;
+       metadata_original_size = subdir_offset;
 
        /* Allocate a buffer to contain the uncompressed metadata resource */
        buf = MALLOC(metadata_original_size);
@@ -1316,26 +1394,18 @@ int write_metadata_resource(WIMStruct *w)
        p = write_security_data(sd, buf);
 
        /* Write the dentry tree into the resource buffer */
-       DEBUG("Writing dentry tree.");
        p = write_dentry_tree(root, p);
 
-       /*
-        * Append 20 random bytes to the metadata resource so that we don't have
-        * identical metadata resources if we happen to append exactly the same
-        * image twice without any changes in timestamps.  If this were to
-        * happen, it would cause confusion about the number and order of images
-        * in the WIM.
-        */
-       randomize_byte_array(p, random_tail_len);
-
        /* We MUST have exactly filled the buffer; otherwise we calculated its
         * size incorrectly or wrote the data incorrectly. */
-       wimlib_assert(p - buf + random_tail_len == metadata_original_size);
+       wimlib_assert(p - buf == metadata_original_size);
 
        /* Get the lookup table entry for the metadata resource so we can update
         * it. */
        lte = wim_metadata_lookup_table_entry(w);
 
+       wimlib_assert(lte != NULL);
+
        /* Write the metadata resource to the output WIM using the proper
         * compression type.  The lookup table entry for the metadata resource
         * is updated. */
@@ -1349,14 +1419,17 @@ int write_metadata_resource(WIMStruct *w)
 
        /* It's very likely the SHA1 message digest of the metadata resource
         * changed, so re-insert the lookup table entry into the lookup table.
+        *
+        * We do not check for other lookup table entries having the same SHA1
+        * message digest.  It's possible for 2 absolutely identical images to
+        * be added, therefore causing 2 identical metadata resources to be in
+        * the WIM.  However, in this case, it's expected for 2 separate lookup
+        * table entries to be created, even though this doesn't make a whole
+        * lot of sense since they will share the same SHA1 message digest.
         * */
        lookup_table_unlink(w->lookup_table, lte);
        lookup_table_insert(w->lookup_table, lte);
 
-       /* We do not allow a metadata resource to be referenced multiple times,
-        * and the 20 random bytes appended to it should make it extremely
-        * likely for each metadata resource to be unique, even if the exact
-        * same image is captured. */
        wimlib_assert(lte->out_refcnt == 0);
        lte->out_refcnt = 1;