[PATCH] readahead: ->prev_page can overrun the ahead window
authorOleg Nesterov <oleg@tv-sign.ru>
Wed, 22 Mar 2006 08:08:47 +0000 (00:08 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Wed, 22 Mar 2006 15:54:03 +0000 (07:54 -0800)
If get_next_ra_size() does not grow fast enough, ->prev_page can overrun
the ahead window.  This means the caller will read the pages from
->ahead_start + ->ahead_size to ->prev_page synchronously.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
mm/readahead.c

index 8d6eeaa..57557e2 100644 (file)
@@ -52,13 +52,24 @@ static inline unsigned long get_min_readahead(struct file_ra_state *ra)
        return (VM_MIN_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 }
 
+static inline void reset_ahead_window(struct file_ra_state *ra)
+{
+       /*
+        * ... but preserve ahead_start + ahead_size value,
+        * see 'recheck:' label in page_cache_readahead().
+        * Note: We never use ->ahead_size as rvalue without
+        * checking ->ahead_start != 0 first.
+        */
+       ra->ahead_size += ra->ahead_start;
+       ra->ahead_start = 0;
+}
+
 static inline void ra_off(struct file_ra_state *ra)
 {
        ra->start = 0;
        ra->flags = 0;
        ra->size = 0;
-       ra->ahead_start = 0;
-       ra->ahead_size = 0;
+       reset_ahead_window(ra);
        return;
 }
 
@@ -426,8 +437,7 @@ static int make_ahead_window(struct address_space *mapping, struct file *filp,
                 * congestion.  The ahead window will any way be closed
                 * in case we failed due to excessive page cache hits.
                 */
-               ra->ahead_start = 0;
-               ra->ahead_size = 0;
+               reset_ahead_window(ra);
        }
 
        return ret;
@@ -520,11 +530,11 @@ page_cache_readahead(struct address_space *mapping, struct file_ra_state *ra,
         * If we get here we are doing sequential IO and this was not the first
         * occurence (ie we have an existing window)
         */
-
        if (ra->ahead_start == 0) {      /* no ahead window yet */
                if (!make_ahead_window(mapping, filp, ra, 0))
-                       goto out;
+                       goto recheck;
        }
+
        /*
         * Already have an ahead window, check if we crossed into it.
         * If so, shift windows and issue a new ahead window.
@@ -536,6 +546,10 @@ page_cache_readahead(struct address_space *mapping, struct file_ra_state *ra,
                ra->start = ra->ahead_start;
                ra->size = ra->ahead_size;
                make_ahead_window(mapping, filp, ra, 0);
+recheck:
+               /* prev_page shouldn't overrun the ahead window */
+               ra->prev_page = min(ra->prev_page,
+                       ra->ahead_start + ra->ahead_size - 1);
        }
 
 out: