Re: [PATCH] Suppress deprecated f_maxcount in 'struct file'

[Peter Staubach <staubach@redhat.com>]

Eric Dumazet wrote:

> Andrew Morton a écrit :
>
> > Eric Dumazet <dada1@cosmosbay.com> wrote:
> >
> > > Considering :
> > >
> > > [root@dada1 linux-2.6.13-rc6]# find .|xargs grep f_maxcount
> > > ./fs/file_table.c:      f->f_maxcount = INT_MAX;
> > > ./fs/read_write.c:      if (unlikely(count > file->f_maxcount))
> > > ./include/linux/fs.h:   size_t                  f_maxcount;
> > >
> > >
> > > I was wondering if f_maxcount has a real use these days...
> >
> >
> >
> > No, I guess we can just stick a hard-wired INT_MAX in there.
>
> OK here is a patch doing the hard wiring then :)
>
> * struct file cleanup : f_maxcount has an unique value (INT_MAX). Just 
> use the hard-wired value.
>
> Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
>
> ------------------------------------------------------------------------
> diff -Nru linux-2.6.13-rc6/fs/read_write.c linux-2.6.13-rc6-ed/fs/read_write.c
> --- linux-2.6.13-rc6/fs/read_write.c	2005-08-07 20:18:56.000000000 +0200
> +++ linux-2.6.13-rc6-ed/fs/read_write.c	2005-08-19 23:51:20.000000000 +0200
> @@ -188,7 +188,7 @@
> 	struct inode *inode;
> 	loff_t pos;
>
> -	if (unlikely(count > file->f_maxcount))
> +	if (unlikely(count > INT_MAX))
> 		goto Einval;
> 	pos = *ppos;
> 	if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))

And depending upon how you feel about read(2) and write(2) returning larger
than can be represented by a ssize_t, you can get rid of this test too and
apply the attached patch to prevent failures occuring in the direct-io
subsystem.

Limiting i/o requests to INT_MAX is starting to seem a little small.

   Thanx...

      ps

Signed-off-by: Peter Staubach <staubach@redhat.com>

--- linux-2.6.12/fs/direct-io.c.org	2005-08-22 08:56:40.000000000 -0400
+++ linux-2.6.12/fs/direct-io.c	2005-08-22 08:59:49.000000000 -0400
@@ -1139,6 +1139,12 @@ direct_io_worker(int rw, struct kiocb *i
 }
 
 /*
+ * The maximum size of an i/o request which can handled by block
+ * devices.
+ */
+#define	MAX_DIO_SIZE	((u32)INT_MAX + 1)
+
+/*
  * This is a library function for use by filesystem drivers.
  * The locking rules are governed by the dio_lock_type parameter.
  *
@@ -1174,6 +1180,12 @@ __blockdev_direct_IO(int rw, struct kioc
 	loff_t end = offset;
 	struct dio *dio;
 	int reader_with_isem = (rw == READ && dio_lock_type == DIO_OWN_LOCKING);
+	int nseg;
+	unsigned long segs_reqd;
+	struct iovec *niov = NULL;
+	int asegs;
+	caddr_t nbase;
+	size_t nlen;
 
 	if (rw & WRITE)
 		current->flags |= PF_SYNCWRITE;
@@ -1189,6 +1201,44 @@ __blockdev_direct_IO(int rw, struct kioc
 			goto out;
 	}
 
+	/*
+	 * Check to see if any individual segment is larger than
+	 * 2G.  If so, then break it up into 2G sized chunks.
+	 */
+	segs_reqd = 0;
+	for (seg = 0; seg < nr_segs; seg++)
+		segs_reqd += ((iov[seg].iov_len - 1) / MAX_DIO_SIZE) + 1;
+	if (segs_reqd != nr_segs) {
+		if (segs_reqd > UIO_MAXIOV) {
+			retval = -EINVAL;
+			goto out;
+		}
+		niov = kmalloc(sizeof(*niov) * segs_reqd, GFP_KERNEL);
+		if (!niov) {
+			retval = -ENOMEM;
+			goto out;
+		}
+		nseg = 0;
+		for (seg = 0; seg < nr_segs; seg++) {
+			nbase = iov[seg].iov_base;
+			nlen = iov[seg].iov_len;
+			asegs = (iov[seg].iov_len - 1) / MAX_DIO_SIZE;
+			while (asegs > 0) {
+				niov[nseg].iov_base = nbase;
+				niov[nseg].iov_len = MAX_DIO_SIZE;
+				nbase += MAX_DIO_SIZE;
+				nlen -= MAX_DIO_SIZE;
+				nseg++;
+				asegs--;
+			}
+			niov[nseg].iov_base = nbase;
+			niov[nseg].iov_len = nlen;
+			nseg++;
+		}
+		iov = niov;
+		nr_segs = segs_reqd;
+	}
+
 	/* Check the memory alignment.  Blocks cannot straddle pages */
 	for (seg = 0; seg < nr_segs; seg++) {
 		addr = (unsigned long)iov[seg].iov_base;
@@ -1265,6 +1315,8 @@ __blockdev_direct_IO(int rw, struct kioc
 out:
 	if (reader_with_isem)
 		up(&inode->i_sem);
+	if (niov)
+		kfree(niov);
 	if (rw & WRITE)
 		current->flags &= ~PF_SYNCWRITE;
 	return retval;