2 files changed, 49 insertions, 5 deletions

diff --git a/fs/xfs/linux-2.6/xfs_buf.c b/fs/xfs/linux-2.6/xfs_buf.c
index a7c7cb27fa5a..522cfaa70258 100644
--- a/fs/xfs/linux-2.6/xfs_buf.c
+++ b/fs/xfs/linux-2.6/xfs_buf.c
@@ -409,7 +409,6 @@ _xfs_buf_lookup_pages(
                         congestion_wait(WRITE, HZ/50);
                         goto retry;
                 }
-                unlock_page(page);
 
                 XFS_STATS_INC(xb_page_found);
 
@@ -419,7 +418,10 @@ _xfs_buf_lookup_pages(
                 ASSERT(!PagePrivate(page));
                 if (!PageUptodate(page)) {
                         page_count--;
-                        if (blocksize < PAGE_CACHE_SIZE && !PagePrivate(page)) {
+                        if (blocksize >= PAGE_CACHE_SIZE) {
+                                if (flags & XBF_READ)
+                                        bp->b_locked = 1;
+                        } else if (!PagePrivate(page)) {
                                 if (test_page_region(page, offset, nbytes))
                                         page_count++;
                         }
@@ -429,6 +431,11 @@ _xfs_buf_lookup_pages(
                 offset = 0;
         }
 
+        if (!bp->b_locked) {
+                for (i = 0; i < bp->b_page_count; i++)
+                        unlock_page(bp->b_pages[i]);
+        }
+
         if (page_count == bp->b_page_count)
                 bp->b_flags |= XBF_DONE;
 
@@ -745,6 +752,7 @@ xfs_buf_associate_memory(
                 bp->b_pages[i] = mem_to_page((void *)pageaddr);
                 pageaddr += PAGE_CACHE_SIZE;
         }
+        bp->b_locked = 0;
 
         bp->b_count_desired = len;
         bp->b_buffer_length = buflen;
@@ -1091,13 +1099,25 @@ xfs_buf_iostart(
         return status;
 }
 
+STATIC_INLINE int
+_xfs_buf_iolocked(
+        xfs_buf_t               *bp)
+{
+        ASSERT(bp->b_flags & (XBF_READ | XBF_WRITE));
+        if (bp->b_flags & XBF_READ)
+                return bp->b_locked;
+        return 0;
+}
+
 STATIC_INLINE void
 _xfs_buf_ioend(
         xfs_buf_t               *bp,
         int                     schedule)
 {
-        if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
+        if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
+                bp->b_locked = 0;
                 xfs_buf_ioend(bp, schedule);
+        }
 }
 
 STATIC void
@@ -1128,6 +1148,10 @@ xfs_buf_bio_end_io(
 
                 if (--bvec >= bio->bi_io_vec)
                         prefetchw(&bvec->bv_page->flags);
+
+                if (_xfs_buf_iolocked(bp)) {
+                        unlock_page(page);
+                }
         } while (bvec >= bio->bi_io_vec);
 
         _xfs_buf_ioend(bp, 1);
@@ -1138,12 +1162,13 @@ STATIC void
 _xfs_buf_ioapply(
         xfs_buf_t               *bp)
 {
-        int                     rw, map_i, total_nr_pages, nr_pages;
+        int                     i, rw, map_i, total_nr_pages, nr_pages;
         struct bio              *bio;
         int                     offset = bp->b_offset;
         int                     size = bp->b_count_desired;
         sector_t                sector = bp->b_bn;
         unsigned int            blocksize = bp->b_target->bt_bsize;
+        int                     locking = _xfs_buf_iolocked(bp);
 
         total_nr_pages = bp->b_page_count;
         map_i = 0;
@@ -1166,7 +1191,7 @@ _xfs_buf_ioapply(
          * filesystem block size is not smaller than the page size.
          */
         if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
-            (bp->b_flags & XBF_READ) &&
+            (bp->b_flags & XBF_READ) && locking &&
             (blocksize >= PAGE_CACHE_SIZE)) {
                 bio = bio_alloc(GFP_NOIO, 1);
 
@@ -1183,6 +1208,24 @@ _xfs_buf_ioapply(
                 goto submit_io;
         }
 
+        /* Lock down the pages which we need to for the request */
+        if (locking && (bp->b_flags & XBF_WRITE) && (bp->b_locked == 0)) {
+                for (i = 0; size; i++) {
+                        int             nbytes = PAGE_CACHE_SIZE - offset;
+                        struct page     *page = bp->b_pages[i];
+
+                        if (nbytes > size)
+                                nbytes = size;
+
+                        lock_page(page);
+
+                        size -= nbytes;
+                        offset = 0;
+                }
+                offset = bp->b_offset;
+                size = bp->b_count_desired;
+        }
+
 next_chunk:
         atomic_inc(&bp->b_io_remaining);
         nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
diff --git a/fs/xfs/linux-2.6/xfs_buf.h b/fs/xfs/linux-2.6/xfs_buf.h
index a3d207de48b8..b5908a34b15d 100644
--- a/fs/xfs/linux-2.6/xfs_buf.h
+++ b/fs/xfs/linux-2.6/xfs_buf.h
@@ -143,6 +143,7 @@ typedef struct xfs_buf {
         void                    *b_fspriv2;
         void                    *b_fspriv3;
         unsigned short          b_error;        /* error code on I/O */
+        unsigned short          b_locked;       /* page array is locked */
         unsigned int            b_page_count;   /* size of page array */
         unsigned int            b_offset;       /* page offset in first page */
         struct page             **b_pages;      /* array of page pointers */