From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sat, 2 Nov 2024 20:06:47 +0000 (-0700)
Subject: Improve sparse I/O performance
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=568919d77b346bfc833d476761204f3325868c0a;p=thirdparty%2Ftar.git

Improve sparse I/O performance

* src/sparse.c (sparse_dump_region, sparse_extract_region):
Don’t insist on reading and writing sparse files 512
bytes at a time.  This resulted in a 4× to 6× performance
improvement on my platform.
---

diff --git a/NEWS b/NEWS
index 56f59dbb..7a5ee22e 100644
--- a/NEWS
+++ b/NEWS
@@ -1,4 +1,4 @@
-GNU tar NEWS - User visible changes. 2023-09-12
+GNU tar NEWS - User visible changes. 2024-11-02
 Please send GNU tar bug reports to <bug-tar@gnu.org>
 
 version TBD
@@ -36,6 +36,10 @@ used, command output will be parsed using strptime(3).
 ** Transformations that change case (e.g., --transform='s/.*/\L&/')
    now work correctly with multi-byte characters.
 
+* Performance improvements
+
+** Sparse files are now read and written with larger blocksizes.
+
 
 version 1.35 - Sergey Poznyakoff, 2023-07-18
 
diff --git a/src/sparse.c b/src/sparse.c
index d04798b9..9518b3ce 100644
--- a/src/sparse.c
+++ b/src/sparse.c
@@ -415,7 +415,8 @@ sparse_dump_region (struct tar_sparse_file *file, idx_t i)
   while (bytes_left > 0)
     {
       union block *blk = find_next_block ();
-      idx_t bufsize = min (bytes_left, BLOCKSIZE);
+      idx_t avail = available_space_after (blk);
+      idx_t bufsize = min (bytes_left, avail);
       idx_t bytes_read = full_read (file->fd, blk->buffer, bufsize);
       if (bytes_read < BLOCKSIZE)
 	memset (blk->buffer + bytes_read, 0, BLOCKSIZE - bytes_read);
@@ -449,7 +450,7 @@ sparse_dump_region (struct tar_sparse_file *file, idx_t i)
 	  return false;
 	}
 
-      set_next_block_after (blk);
+      set_next_block_after (blk + (bufsize - 1) / BLOCKSIZE);
     }
 
   return true;
@@ -473,15 +474,16 @@ sparse_extract_region (struct tar_sparse_file *file, idx_t i)
     }
   else while (write_size > 0)
     {
-      idx_t wrbytes = min (write_size, BLOCKSIZE);
       union block *blk = find_next_block ();
       if (!blk)
 	{
 	  paxerror (0, _("Unexpected EOF in archive"));
 	  return false;
 	}
-      set_next_block_after (blk);
-      file->dumped_size += BLOCKSIZE;
+      idx_t avail = available_space_after (blk);
+      idx_t wrbytes = min (write_size, avail);
+      set_next_block_after (blk + (wrbytes - 1) / BLOCKSIZE);
+      file->dumped_size += avail;
       idx_t count = blocking_write (file->fd, blk->buffer, wrbytes);
       write_size -= count;
       mv_size_left (file->stat_info->archive_file_size - file->dumped_size);