]> wimlib.net Git - wimlib/blobdiff - src/write.c
Improve random number generation
[wimlib] / src / write.c
index 23dda573a3aff33e63836ae843ea2629eaf543c5..4edf64e3350bebfc0a80f28612290a6bcef43fdd 100644 (file)
@@ -6,7 +6,7 @@
  */
 
 /*
- * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
+ * Copyright (C) 2012-2016 Eric Biggers
  *
  * 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
@@ -288,7 +288,7 @@ write_pwm_blob_header(const struct blob_descriptor *blob,
        blob_hdr.flags = cpu_to_le32(reshdr_flags);
        ret = full_write(out_fd, &blob_hdr, sizeof(blob_hdr));
        if (ret)
-               ERROR_WITH_ERRNO("Write error");
+               ERROR_WITH_ERRNO("Error writing blob header to WIM file");
        return ret;
 }
 
@@ -377,12 +377,6 @@ struct write_blobs_ctx {
         * @blobs_being_compressed only when writing a solid resource.  */
        struct list_head blobs_in_solid_resource;
 
-       /* Current uncompressed offset in the blob being read.  */
-       u64 cur_read_blob_offset;
-
-       /* Uncompressed size of the blob currently being read.  */
-       u64 cur_read_blob_size;
-
        /* Current uncompressed offset in the blob being written.  */
        u64 cur_write_blob_offset;
 
@@ -461,8 +455,11 @@ begin_chunk_table(struct write_blobs_ctx *ctx, u64 res_expected_size)
                        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);
-               if (ret)
+               if (ret) {
+                       ERROR_WITH_ERRNO("Error reserving space for chunk "
+                                        "table in WIM file");
                        return ret;
+               }
        }
        return 0;
 }
@@ -588,7 +585,7 @@ end_chunk_table(struct write_blobs_ctx *ctx, u64 res_actual_size,
        return 0;
 
 write_error:
-       ERROR_WITH_ERRNO("Write error");
+       ERROR_WITH_ERRNO("Error writing chunk table to WIM file");
        return ret;
 }
 
@@ -639,6 +636,7 @@ do_done_with_blob(struct blob_descriptor *blob,
 {
        int ret;
        struct wim_inode *inode;
+       const tchar *path;
        tchar *cookie1;
        tchar *cookie2;
 
@@ -652,10 +650,12 @@ do_done_with_blob(struct blob_descriptor *blob,
        if (--inode->i_num_remaining_streams > 0)
                return 0;
 
-       cookie1 = progress_get_streamless_path(blob->file_on_disk);
-       cookie2 = progress_get_win32_path(blob->file_on_disk);
+       path = blob_file_path(blob);
+
+       cookie1 = progress_get_streamless_path(path);
+       cookie2 = progress_get_win32_path(path);
 
-       ret = done_with_file(blob->file_on_disk, progfunc, progctx);
+       ret = done_with_file(path, progfunc, progctx);
 
        progress_put_win32_path(cookie2);
        progress_put_streamless_path(cookie1);
@@ -683,9 +683,6 @@ write_blob_begin_read(struct blob_descriptor *blob, void *_ctx)
 
        wimlib_assert(blob->size > 0);
 
-       ctx->cur_read_blob_offset = 0;
-       ctx->cur_read_blob_size = blob->size;
-
        /* As an optimization, we allow some blobs to be "unhashed", meaning
         * their SHA-1 message digests are unknown.  This is the case with blobs
         * that are added by scanning a directory tree with wimlib_add_image(),
@@ -823,7 +820,7 @@ should_rewrite_blob_uncompressed(const struct write_blobs_ctx *ctx,
         * Exception: if the compressed size happens to be *exactly* the same as
         * the uncompressed size, then the blob *must* be written uncompressed
         * in order to remain compatible with the Windows Overlay Filesystem
-        * Filter Driver (WOF).
+        * filter driver (WOF).
         *
         * TODO: we are currently assuming that the optimization for
         * single-chunk resources in maybe_rewrite_blob_uncompressed() prevents
@@ -980,7 +977,7 @@ write_chunk(struct write_blobs_ctx *ctx, const void *cchunk,
                                       completed_blob_count, false);
 
 write_error:
-       ERROR_WITH_ERRNO("Write error");
+       ERROR_WITH_ERRNO("Error writing chunk data to WIM file");
        return ret;
 }
 
@@ -1013,7 +1010,8 @@ prepare_chunk_buffer(struct write_blobs_ctx *ctx)
 
 /* Process the next chunk of data to be written to a WIM resource.  */
 static int
-write_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
+write_blob_process_chunk(const struct blob_descriptor *blob, u64 offset,
+                        const void *chunk, size_t size, void *_ctx)
 {
        struct write_blobs_ctx *ctx = _ctx;
        int ret;
@@ -1026,7 +1024,6 @@ write_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
                 ret = write_chunk(ctx, chunk, size, size);
                 if (ret)
                         return ret;
-                ctx->cur_read_blob_offset += size;
                 return 0;
        }
 
@@ -1050,8 +1047,7 @@ write_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
                } else {
                        needed_chunk_size = min(ctx->out_chunk_size,
                                                ctx->cur_chunk_buf_filled +
-                                                       (ctx->cur_read_blob_size -
-                                                        ctx->cur_read_blob_offset));
+                                                       (blob->size - offset));
                }
 
                bytes_consumed = min(chunkend - chunkptr,
@@ -1061,7 +1057,7 @@ write_blob_process_chunk(const void *chunk, size_t size, void *_ctx)
                       chunkptr, bytes_consumed);
 
                chunkptr += bytes_consumed;
-               ctx->cur_read_blob_offset += bytes_consumed;
+               offset += bytes_consumed;
                ctx->cur_chunk_buf_filled += bytes_consumed;
 
                if (ctx->cur_chunk_buf_filled == needed_chunk_size) {
@@ -1082,8 +1078,6 @@ write_blob_end_read(struct blob_descriptor *blob, int status, void *_ctx)
 {
        struct write_blobs_ctx *ctx = _ctx;
 
-       wimlib_assert(ctx->cur_read_blob_offset == ctx->cur_read_blob_size || status);
-
        if (!blob->will_be_in_output_wim) {
                /* The blob was a duplicate.  Now that its data has finished
                 * being read, it is being discarded in favor of the duplicate
@@ -1237,12 +1231,18 @@ write_raw_copy_resource(struct wim_resource_descriptor *in_rdesc,
 
                        ret = full_pread(in_fd, buf, bytes_to_read,
                                         cur_read_offset);
-                       if (ret)
+                       if (ret) {
+                               ERROR_WITH_ERRNO("Error reading raw data "
+                                                "from WIM file");
                                return ret;
+                       }
 
                        ret = full_write(out_fd, buf, bytes_to_read);
-                       if (ret)
+                       if (ret) {
+                               ERROR_WITH_ERRNO("Error writing raw data "
+                                                "to WIM file");
                                return ret;
+                       }
 
                        cur_read_offset += bytes_to_read;
 
@@ -1340,17 +1340,6 @@ validate_blob_list(struct list_head *blob_list)
        }
 }
 
-static inline bool
-blob_is_in_file(const struct blob_descriptor *blob)
-{
-       return blob->blob_location == BLOB_IN_FILE_ON_DISK
-#ifdef __WIN32__
-           || blob->blob_location == BLOB_IN_WINNT_FILE_ON_DISK
-           || blob->blob_location == BLOB_WIN32_ENCRYPTED
-#endif
-          ;
-}
-
 static void
 init_done_with_file_info(struct list_head *blob_list)
 {
@@ -1551,21 +1540,12 @@ write_blob_list(struct list_head *blob_list,
                                               out_ctype, out_chunk_size,
                                               &raw_copy_blobs);
 
-       /* Copy any compressed resources for which the raw data can be reused
-        * without decompression.  */
-       ret = write_raw_copy_resources(&raw_copy_blobs, ctx.out_fd,
-                                      &ctx.progress_data);
-
-       if (ret || num_nonraw_bytes == 0)
-               goto out_destroy_context;
-
-       /* Unless uncompressed output was required, allocate a chunk_compressor
-        * to do compression.  There are serial and parallel implementations of
-        * the chunk_compressor interface.  We default to parallel using the
+       /* Unless no data needs to be compressed, allocate a chunk_compressor to
+        * do compression.  There are serial and parallel implementations of the
+        * chunk_compressor interface.  We default to parallel using the
         * specified number of threads, unless the upper bound on the number
         * bytes needing to be compressed is less than a heuristic value.  */
-       if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
-
+       if (num_nonraw_bytes != 0 && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE) {
        #ifdef ENABLE_MULTITHREADED_COMPRESSION
                if (num_nonraw_bytes > max(2000000, out_chunk_size)) {
                        ret = new_parallel_chunk_compressor(out_ctype,
@@ -1593,9 +1573,6 @@ write_blob_list(struct list_head *blob_list,
        else
                ctx.progress_data.progress.write_streams.num_threads = 1;
 
-       INIT_LIST_HEAD(&ctx.blobs_being_compressed);
-       INIT_LIST_HEAD(&ctx.blobs_in_solid_resource);
-
        ret = call_progress(ctx.progress_data.progfunc,
                            WIMLIB_PROGRESS_MSG_WRITE_STREAMS,
                            &ctx.progress_data.progress,
@@ -1603,7 +1580,20 @@ write_blob_list(struct list_head *blob_list,
        if (ret)
                goto out_destroy_context;
 
+       /* Copy any compressed resources for which the raw data can be reused
+        * without decompression.  */
+       ret = write_raw_copy_resources(&raw_copy_blobs, ctx.out_fd,
+                                      &ctx.progress_data);
+
+       if (ret || num_nonraw_bytes == 0)
+               goto out_destroy_context;
+
+       INIT_LIST_HEAD(&ctx.blobs_being_compressed);
+
        if (write_resource_flags & WRITE_RESOURCE_FLAG_SOLID) {
+
+               INIT_LIST_HEAD(&ctx.blobs_in_solid_resource);
+
                ret = begin_write_resource(&ctx, num_nonraw_bytes);
                if (ret)
                        goto out_destroy_context;
@@ -1614,7 +1604,7 @@ write_blob_list(struct list_head *blob_list,
 
        struct read_blob_callbacks cbs = {
                .begin_blob     = write_blob_begin_read,
-               .consume_chunk  = write_blob_process_chunk,
+               .continue_blob  = write_blob_process_chunk,
                .end_blob       = write_blob_end_read,
                .ctx            = &ctx,
        };
@@ -2379,10 +2369,8 @@ finish_write(WIMStruct *wim, int image, int write_flags,
        if (!(write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS)) {
                ret = write_blob_table(wim, image, write_flags,
                                       blob_table_list);
-               if (ret) {
-                       free_integrity_table(old_integrity_table);
-                       return ret;
-               }
+               if (ret)
+                       goto out;
        }
 
        /* Write XML data.  */
@@ -2392,13 +2380,13 @@ finish_write(WIMStruct *wim, int image, int write_flags,
        ret = write_wim_xml_data(wim, image, xml_totalbytes,
                                 &wim->out_hdr.xml_data_reshdr,
                                 write_resource_flags);
-       if (ret) {
-               free_integrity_table(old_integrity_table);
-               return ret;
-       }
+       if (ret)
+               goto out;
 
        /* Write integrity table if needed.  */
-       if (write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) {
+       if ((write_flags & WIMLIB_WRITE_FLAG_CHECK_INTEGRITY) &&
+           wim->out_hdr.blob_table_reshdr.offset_in_wim != 0)
+       {
                if (write_flags & WIMLIB_WRITE_FLAG_NO_NEW_BLOBS) {
                        /* The XML data we wrote may have overwritten part of
                         * the old integrity table, so while calculating the new
@@ -2409,10 +2397,8 @@ finish_write(WIMStruct *wim, int image, int write_flags,
                        zero_reshdr(&checkpoint_hdr.integrity_table_reshdr);
                        checkpoint_hdr.flags |= WIM_HDR_FLAG_WRITE_IN_PROGRESS;
                        ret = write_wim_header(&checkpoint_hdr, &wim->out_fd, 0);
-                       if (ret) {
-                               free_integrity_table(old_integrity_table);
-                               return ret;
-                       }
+                       if (ret)
+                               goto out;
                }
 
                new_blob_table_end = wim->out_hdr.blob_table_reshdr.offset_in_wim +
@@ -2422,9 +2408,8 @@ finish_write(WIMStruct *wim, int image, int write_flags,
                                            new_blob_table_end,
                                            old_blob_table_end,
                                            old_integrity_table);
-               free_integrity_table(old_integrity_table);
                if (ret)
-                       return ret;
+                       goto out;
        } else {
                /* No integrity table.  */
                zero_reshdr(&wim->out_hdr.integrity_table_reshdr);
@@ -2440,13 +2425,17 @@ finish_write(WIMStruct *wim, int image, int write_flags,
        else
                ret = write_wim_header(&wim->out_hdr, &wim->out_fd, 0);
        if (ret)
-               return ret;
+               goto out;
 
+       ret = WIMLIB_ERR_WRITE;
        if (unlikely(write_flags & WIMLIB_WRITE_FLAG_UNSAFE_COMPACT)) {
                /* Truncate any data the compaction freed up.  */
-               if (ftruncate(wim->out_fd.fd, wim->out_fd.offset)) {
+               if (ftruncate(wim->out_fd.fd, wim->out_fd.offset) &&
+                   errno != EINVAL) /* allow compaction on untruncatable files,
+                                       e.g. block devices  */
+               {
                        ERROR_WITH_ERRNO("Failed to truncate the output WIM file");
-                       return WIMLIB_ERR_WRITE;
+                       goto out;
                }
        }
 
@@ -2456,19 +2445,24 @@ finish_write(WIMStruct *wim, int image, int write_flags,
         * the system is abruptly terminated when the metadata for the rename
         * operation has been written to disk, but the new file data has not.
         */
+       ret = WIMLIB_ERR_WRITE;
        if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
                if (fsync(wim->out_fd.fd)) {
                        ERROR_WITH_ERRNO("Error syncing data to WIM file");
-                       return WIMLIB_ERR_WRITE;
+                       goto out;
                }
        }
 
+       ret = WIMLIB_ERR_WRITE;
        if (close_wim_writable(wim, write_flags)) {
                ERROR_WITH_ERRNO("Failed to close the output WIM file");
-               return WIMLIB_ERR_WRITE;
+               goto out;
        }
 
-       return 0;
+       ret = 0;
+out:
+       free_integrity_table(old_integrity_table);
+       return ret;
 }
 
 #if defined(HAVE_SYS_FILE_H) && defined(HAVE_FLOCK)
@@ -2961,6 +2955,20 @@ check_resource_offsets(WIMStruct *wim, off_t end_offset)
        return 0;
 }
 
+static int
+free_blob_if_invalidated(struct blob_descriptor *blob, void *_wim)
+{
+       const WIMStruct *wim = _wim;
+
+       if (!blob->will_be_in_output_wim &&
+           blob->blob_location == BLOB_IN_WIM && blob->rdesc->wim == wim)
+       {
+               blob_table_unlink(wim->blob_table, blob);
+               free_blob_descriptor(blob);
+       }
+       return 0;
+}
+
 /*
  * Overwrite a WIM, possibly appending new resources to it.
  *
@@ -3077,6 +3085,11 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
                if (ret)
                        goto out;
 
+               /* Prevent new files from being deduplicated with existing blobs
+                * in the WIM that we haven't decided to write.  Such blobs will
+                * be overwritten during the compaction.  */
+               for_blob_in_table(wim->blob_table, free_blob_if_invalidated, wim);
+
                if (wim_has_metadata(wim)) {
                        /* Add existing metadata resources to be compacted along
                         * with the file resources.  */
@@ -3100,8 +3113,11 @@ overwrite_wim_inplace(WIMStruct *wim, int write_flags, unsigned num_threads)
                 * this data would be overwritten.  */
                old_xml_begin = wim->hdr.xml_data_reshdr.offset_in_wim;
                old_xml_end = old_xml_begin + wim->hdr.xml_data_reshdr.size_in_wim;
-               old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
-                                    wim->hdr.blob_table_reshdr.size_in_wim;
+               if (wim->hdr.blob_table_reshdr.offset_in_wim == 0)
+                       old_blob_table_end = WIM_HEADER_DISK_SIZE;
+               else
+                       old_blob_table_end = wim->hdr.blob_table_reshdr.offset_in_wim +
+                                            wim->hdr.blob_table_reshdr.size_in_wim;
                if (wim_has_integrity_table(wim) &&
                    wim->hdr.integrity_table_reshdr.offset_in_wim < old_xml_end) {
                        WARNING("Didn't expect the integrity table to be "
@@ -3204,9 +3220,8 @@ out_truncate:
                             WIMLIB_WRITE_FLAG_UNSAFE_COMPACT))) {
                WARNING("Truncating \"%"TS"\" to its original size "
                        "(%"PRIu64" bytes)", wim->filename, old_wim_end);
-               /* Return value of ftruncate() is ignored because this is
-                * already an error path.  */
-               (void)ftruncate(wim->out_fd.fd, old_wim_end);
+               if (ftruncate(wim->out_fd.fd, old_wim_end))
+                       WARNING_WITH_ERRNO("Failed to truncate WIM file!");
        }
 out_restore_hdr:
        (void)write_wim_header_flags(wim->hdr.flags, &wim->out_fd);
@@ -3230,7 +3245,7 @@ overwrite_wim_via_tmpfile(WIMStruct *wim, int write_flags, unsigned num_threads)
        wim_name_len = tstrlen(wim->filename);
        tchar tmpfile[wim_name_len + 10];
        tmemcpy(tmpfile, wim->filename, wim_name_len);
-       randomize_char_array_with_alnum(tmpfile + wim_name_len, 9);
+       get_random_alnum_chars(tmpfile + wim_name_len, 9);
        tmpfile[wim_name_len + 9] = T('\0');
 
        ret = wimlib_write(wim, tmpfile, WIMLIB_ALL_IMAGES,