[PATCH] Bug #3054 madvise(MADV_WILLNEED,...) fix for exceeding rlimitrss

[Matt Keenan]

Here is an updated patch for 2.6.12, can it be included in -mm for 
testing? I have been banging on the code for an hour or so now; no probs.

Matt

--- linux-2.6.11.7/mm/madvise.c 2005-04-12 15:58:30.000000000 +0100
+++ linux/mm/madvise.c  2005-06-19 17:20:56.000000000 +0100
@@ -61,6 +61,7 @@ static long madvise_willneed(struct vm_a
                            unsigned long start, unsigned long end)
{
       struct file *file = vma->vm_file;
+       struct task_struct *tsk = current;

       if (!file)
               return -EBADF;
@@ -70,6 +71,28 @@ static long madvise_willneed(struct vm_a
               end = vma->vm_end;
       end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;

+       /*
+        * This code below checks to see if mapping the requested
+        * readahead would make the task's rss exceed the task's
+        * rlimit rss.
+        *
+        * This doesn't account for pages that may already be mapped
+        * due to readahead, but since this is merely a hint to the
+        * kernel no harm should be done, it won't unmap anything
+        * already mapped if it fails. N.B. This won't affect the
+        * kernel's internal automatic readahead which doesn't check
+        * (or honour) rlimit rss.
+        */
+
+       spin_lock(&tsk->mm->page_table_lock);
+       if (((max_sane_readahead(end-start) << PAGE_SHIFT) +
+           tsk->mm->_rss) > tsk->signal->rlim[RLIMIT_RSS].rlim_cur)
+       {
+               spin_unlock(&tsk->mm->page_table_lock);
+               return -EIO;
+       }
+       spin_unlock(&tsk->mm->page_table_lock);
+
       force_page_cache_readahead(file->f_mapping,
                       file, start, max_sane_readahead(end - start));
       return 0;
@@ -170,6 +193,8 @@ static long madvise_vma(struct vm_area_s
 *  -ENOMEM - addresses in the specified range are not currently
 *             mapped, or are outside the AS of the process.
 *  -EIO    - an I/O error occurred while paging in data.
+ *          - MADV_WILLNEED would map in pages that would make the task's
+ *              rss exceed rlimit rss.
 *  -EBADF  - map exists, but area maps something that isn't a file.
 *  -EAGAIN - a kernel resource was temporarily unavailable.
 */

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/