[XFS] use pagevec lookups This reduces the time spend in the radix tree
authorChristoph Hellwig <hch@sgi.com>
Wed, 11 Jan 2006 09:48:14 +0000 (20:48 +1100)
committerNathan Scott <nathans@sgi.com>
Wed, 11 Jan 2006 09:48:14 +0000 (20:48 +1100)
lookups and avoids unessecary look roundtrips.

SGI-PV: 947118
SGI-Modid: xfs-linux-melb:xfs-kern:203823a

Signed-off-by: Christoph Hellwig <hch@sgi.com>
Signed-off-by: Nathan Scott <nathans@sgi.com>
fs/xfs/linux-2.6/xfs_aops.c

index e99d04d3fe82214a17e141fd6b86f32f006ff666..e998009c0f52e25c21ae9b3a7deaa08793a80057 100644 (file)
@@ -40,6 +40,7 @@
 #include "xfs_rw.h"
 #include "xfs_iomap.h"
 #include <linux/mpage.h>
+#include <linux/pagevec.h>
 #include <linux/writeback.h>
 
 STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
@@ -501,18 +502,13 @@ xfs_map_at_offset(
  */
 STATIC unsigned int
 xfs_probe_unmapped_page(
-       struct address_space    *mapping,
-       pgoff_t                 index,
+       struct page             *page,
        unsigned int            pg_offset)
 {
-       struct page             *page;
        int                     ret = 0;
 
-       page = find_trylock_page(mapping, index);
-       if (!page)
-               return 0;
        if (PageWriteback(page))
-               goto out;
+               return 0;
 
        if (page->mapping && PageDirty(page)) {
                if (page_has_buffers(page)) {
@@ -530,8 +526,6 @@ xfs_probe_unmapped_page(
                        ret = PAGE_CACHE_SIZE;
        }
 
-out:
-       unlock_page(page);
        return ret;
 }
 
@@ -542,59 +536,75 @@ xfs_probe_unmapped_cluster(
        struct buffer_head      *bh,
        struct buffer_head      *head)
 {
-       size_t                  len, total = 0;
+       struct pagevec          pvec;
        pgoff_t                 tindex, tlast, tloff;
-       unsigned int            pg_offset;
-       struct address_space    *mapping = inode->i_mapping;
+       size_t                  total = 0;
+       int                     done = 0, i;
 
        /* First sum forwards in this page */
        do {
                if (buffer_mapped(bh))
-                       break;
+                       return total;
                total += bh->b_size;
        } while ((bh = bh->b_this_page) != head);
 
-       /* If we reached the end of the page, sum forwards in
-        * following pages.
-        */
-       if (bh == head) {
-               tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
-               /* Prune this back to avoid pathological behavior */
-               tloff = min(tlast, startpage->index + 64);
-               for (tindex = startpage->index + 1; tindex < tloff; tindex++) {
-                       len = xfs_probe_unmapped_page(mapping, tindex,
-                                                       PAGE_CACHE_SIZE);
-                       if (!len)
-                               return total;
+       /* if we reached the end of the page, sum forwards in following pages */
+       tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
+       tindex = startpage->index + 1;
+
+       /* Prune this back to avoid pathological behavior */
+       tloff = min(tlast, startpage->index + 64);
+
+       pagevec_init(&pvec, 0);
+       while (!done && tindex <= tloff) {
+               unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
+
+               if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
+                       break;
+
+               for (i = 0; i < pagevec_count(&pvec); i++) {
+                       struct page *page = pvec.pages[i];
+                       size_t pg_offset, len = 0;
+
+                       if (tindex == tlast) {
+                               pg_offset =
+                                   i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
+                               if (!pg_offset)
+                                       break;
+                       } else
+                               pg_offset = PAGE_CACHE_SIZE;
+
+                       if (page->index == tindex && !TestSetPageLocked(page)) {
+                               len = xfs_probe_unmapped_page(page, pg_offset);
+                               unlock_page(page);
+                       }
+
+                       if (!len) {
+                               done = 1;
+                               break;
+                       }
+
                        total += len;
                }
-               if (tindex == tlast &&
-                   (pg_offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
-                       total += xfs_probe_unmapped_page(mapping,
-                                                       tindex, pg_offset);
-               }
+
+               pagevec_release(&pvec);
+               cond_resched();
        }
+
        return total;
 }
 
 /*
- * Probe for a given page (index) in the inode and test if it is suitable
- * for writing as part of an unwritten or delayed allocate extent.
- * Returns page locked and with an extra reference count if so, else NULL.
+ * Test if a given page is suitable for writing as part of an unwritten
+ * or delayed allocate extent.
  */
-STATIC struct page *
-xfs_probe_delayed_page(
-       struct inode            *inode,
-       pgoff_t                 index,
+STATIC int
+xfs_is_delayed_page(
+       struct page             *page,
        unsigned int            type)
 {
-       struct page             *page;
-
-       page = find_trylock_page(inode->i_mapping, index);
-       if (!page)
-               return NULL;
        if (PageWriteback(page))
-               goto out;
+               return 0;
 
        if (page->mapping && page_has_buffers(page)) {
                struct buffer_head      *bh, *head;
@@ -611,12 +621,10 @@ xfs_probe_delayed_page(
                } while ((bh = bh->b_this_page) != head);
 
                if (acceptable)
-                       return page;
+                       return 1;
        }
 
-out:
-       unlock_page(page);
-       return NULL;
+       return 0;
 }
 
 /*
@@ -629,10 +637,10 @@ STATIC int
 xfs_convert_page(
        struct inode            *inode,
        struct page             *page,
+       loff_t                  tindex,
        xfs_iomap_t             *iomapp,
        xfs_ioend_t             **ioendp,
        struct writeback_control *wbc,
-       void                    *private,
        int                     startio,
        int                     all_bh)
 {
@@ -644,6 +652,17 @@ xfs_convert_page(
        int                     len, page_dirty;
        int                     count = 0, done = 0, uptodate = 1;
 
+       if (page->index != tindex)
+               goto fail;
+       if (TestSetPageLocked(page))
+               goto fail;
+       if (PageWriteback(page))
+               goto fail_unlock_page;
+       if (page->mapping != inode->i_mapping)
+               goto fail_unlock_page;
+       if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
+               goto fail_unlock_page;
+
        end_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1));
 
        /*
@@ -715,6 +734,10 @@ xfs_convert_page(
        }
 
        return done;
+ fail_unlock_page:
+       unlock_page(page);
+ fail:
+       return 1;
 }
 
 /*
@@ -732,16 +755,25 @@ xfs_cluster_write(
        int                     all_bh,
        pgoff_t                 tlast)
 {
-       struct page             *page;
-       unsigned int            type = (*ioendp)->io_type;
-       int                     done;
+       struct pagevec          pvec;
+       int                     done = 0, i;
 
-       for (done = 0; tindex <= tlast && !done; tindex++) {
-               page = xfs_probe_delayed_page(inode, tindex, type);
-               if (!page)
+       pagevec_init(&pvec, 0);
+       while (!done && tindex <= tlast) {
+               unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
+
+               if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
                        break;
-               done = xfs_convert_page(inode, page, iomapp, ioendp,
-                                               wbc, NULL, startio, all_bh);
+
+               for (i = 0; i < pagevec_count(&pvec); i++) {
+                       done = xfs_convert_page(inode, pvec.pages[i], tindex++,
+                                       iomapp, ioendp, wbc, startio, all_bh);
+                       if (done)
+                               break;
+               }
+
+               pagevec_release(&pvec);
+               cond_resched();
        }
 }