]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
splice: fix offset mangling with direct splicing (sendfile)
authorJens Axboe <jens.axboe@oracle.com>
Fri, 13 Jul 2007 12:11:43 +0000 (14:11 +0200)
committerJens Axboe <jens.axboe@oracle.com>
Fri, 13 Jul 2007 12:14:31 +0000 (14:14 +0200)
If the output actor doesn't transfer the full amount of data, we will
increment ppos too much. Two related bugs in there:

- We need to break out and return actor() retval if it is shorted than
  what we spliced into the pipe.

- Adjust ppos only according to actor() return.

Also fix loop problem in generic_file_splice_read(), it should not keep
going when data has already been transferred.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
fs/splice.c

index ef808227bc11248325407a29bbbbab6d7cf88744..6c9828651e6f5cdec78ce99d5169bb238c61f0ea 100644 (file)
@@ -492,7 +492,7 @@ ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
 
        ret = 0;
        spliced = 0;
-       while (len) {
+       while (len && !spliced) {
                ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
 
                if (ret < 0)
@@ -1060,15 +1060,10 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
        sd->flags &= ~SPLICE_F_NONBLOCK;
 
        while (len) {
-               size_t read_len, max_read_len;
-
-               /*
-                * Do at most PIPE_BUFFERS pages worth of transfer:
-                */
-               max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));
+               size_t read_len;
 
-               ret = do_splice_to(in, &sd->pos, pipe, max_read_len, flags);
-               if (unlikely(ret < 0))
+               ret = do_splice_to(in, &sd->pos, pipe, len, flags);
+               if (unlikely(ret <= 0))
                        goto out_release;
 
                read_len = ret;
@@ -1080,26 +1075,17 @@ ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
                 * could get stuck data in the internal pipe:
                 */
                ret = actor(pipe, sd);
-               if (unlikely(ret < 0))
+               if (unlikely(ret <= 0))
                        goto out_release;
 
                bytes += ret;
                len -= ret;
 
-               /*
-                * In nonblocking mode, if we got back a short read then
-                * that was due to either an IO error or due to the
-                * pagecache entry not being there. In the IO error case
-                * the _next_ splice attempt will produce a clean IO error
-                * return value (not a short read), so in both cases it's
-                * correct to break out of the loop here:
-                */
-               if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
-                       break;
+               if (ret < read_len)
+                       goto out_release;
        }
 
        pipe->nrbufs = pipe->curbuf = 0;
-
        return bytes;
 
 out_release:
@@ -1161,10 +1147,12 @@ long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
                .pos            = *ppos,
                .u.file         = out,
        };
-       size_t ret;
+       long ret;
 
        ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
-       *ppos = sd.pos;
+       if (ret > 0)
+               *ppos += ret;
+
        return ret;
 }