xfs: simplify xfs_vm_writepage
authorChristoph Hellwig <hch@infradead.org>
Wed, 23 Jun 2010 23:46:01 +0000 (09:46 +1000)
committerAlex Elder <aelder@sgi.com>
Mon, 26 Jul 2010 18:16:41 +0000 (13:16 -0500)
The writepage implementation in XFS still tries to deal with dirty but
unmapped buffers which used to caused by writes through shared mmaps.  Since
the introduction of ->page_mkwrite these can't happen anymore, so remove the
code dealing with them.

Note that the all_bh variable which causes us to start I/O on all buffers on
the pages was controlled by the count of unmapped buffers, which also
included those not actually dirty.  It's now unconditionally initialized to
0 but set to 1 for the case of small file size extensions.  It probably can
be removed entirely, but that's left for another patch.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
fs/xfs/linux-2.6/xfs_aops.c
fs/xfs/linux-2.6/xfs_aops.h
fs/xfs/linux-2.6/xfs_trace.h

index 7744a3b630e0ad8efbcc7f83229c1d1d17a20588..1776cdd944b532d8b1205408dd33d36e0ec5c4f4 100644 (file)
@@ -85,18 +85,15 @@ void
 xfs_count_page_state(
        struct page             *page,
        int                     *delalloc,
-       int                     *unmapped,
        int                     *unwritten)
 {
        struct buffer_head      *bh, *head;
 
-       *delalloc = *unmapped = *unwritten = 0;
+       *delalloc = *unwritten = 0;
 
        bh = head = page_buffers(page);
        do {
-               if (buffer_uptodate(bh) && !buffer_mapped(bh))
-                       (*unmapped) = 1;
-               else if (buffer_unwritten(bh))
+               if (buffer_unwritten(bh))
                        (*unwritten) = 1;
                else if (buffer_delay(bh))
                        (*delalloc) = 1;
@@ -607,31 +604,30 @@ xfs_map_at_offset(
 STATIC unsigned int
 xfs_probe_page(
        struct page             *page,
-       unsigned int            pg_offset,
-       int                     mapped)
+       unsigned int            pg_offset)
 {
+       struct buffer_head      *bh, *head;
        int                     ret = 0;
 
        if (PageWriteback(page))
                return 0;
+       if (!PageDirty(page))
+               return 0;
+       if (!page->mapping)
+               return 0;
+       if (!page_has_buffers(page))
+               return 0;
 
-       if (page->mapping && PageDirty(page)) {
-               if (page_has_buffers(page)) {
-                       struct buffer_head      *bh, *head;
-
-                       bh = head = page_buffers(page);
-                       do {
-                               if (!buffer_uptodate(bh))
-                                       break;
-                               if (mapped != buffer_mapped(bh))
-                                       break;
-                               ret += bh->b_size;
-                               if (ret >= pg_offset)
-                                       break;
-                       } while ((bh = bh->b_this_page) != head);
-               } else
-                       ret = mapped ? 0 : PAGE_CACHE_SIZE;
-       }
+       bh = head = page_buffers(page);
+       do {
+               if (!buffer_uptodate(bh))
+                       break;
+               if (!buffer_mapped(bh))
+                       break;
+               ret += bh->b_size;
+               if (ret >= pg_offset)
+                       break;
+       } while ((bh = bh->b_this_page) != head);
 
        return ret;
 }
@@ -641,8 +637,7 @@ xfs_probe_cluster(
        struct inode            *inode,
        struct page             *startpage,
        struct buffer_head      *bh,
-       struct buffer_head      *head,
-       int                     mapped)
+       struct buffer_head      *head)
 {
        struct pagevec          pvec;
        pgoff_t                 tindex, tlast, tloff;
@@ -651,7 +646,7 @@ xfs_probe_cluster(
 
        /* First sum forwards in this page */
        do {
-               if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
+               if (!buffer_uptodate(bh) || !buffer_mapped(bh))
                        return total;
                total += bh->b_size;
        } while ((bh = bh->b_this_page) != head);
@@ -685,7 +680,7 @@ xfs_probe_cluster(
                                pg_offset = PAGE_CACHE_SIZE;
 
                        if (page->index == tindex && trylock_page(page)) {
-                               pg_len = xfs_probe_page(page, pg_offset, mapped);
+                               pg_len = xfs_probe_page(page, pg_offset);
                                unlock_page(page);
                        }
 
@@ -1021,18 +1016,12 @@ out_invalidate:
  * For delalloc space on the page we need to allocate space and flush it.
  * For unwritten space on the page we need to start the conversion to
  * regular allocated space.
- * For unmapped buffer heads on the page we should allocate space if the
- * page is uptodate.
  * For any other dirty buffer heads on the page we should flush them.
  *
  * If we detect that a transaction would be required to flush the page, we
  * have to check the process flags first, if we are already in a transaction
  * or disk I/O during allocations is off, we need to fail the writepage and
  * redirty the page.
- *
- * The bh->b_state's cannot know if any of the blocks or which block for that
- * matter are dirty due to mmap writes, and therefore bh uptodate is only
- * valid if the page itself isn't completely uptodate.
  */
 STATIC int
 xfs_vm_writepage(
@@ -1040,8 +1029,7 @@ xfs_vm_writepage(
        struct writeback_control *wbc)
 {
        struct inode            *inode = page->mapping->host;
-       int                     need_trans;
-       int                     delalloc, unmapped, unwritten;
+       int                     delalloc, unwritten;
        struct buffer_head      *bh, *head;
        struct xfs_bmbt_irec    imap;
        xfs_ioend_t             *ioend = NULL, *iohead = NULL;
@@ -1052,10 +1040,12 @@ xfs_vm_writepage(
        ssize_t                 size, len;
        int                     flags, err, imap_valid = 0, uptodate = 1;
        int                     count = 0;
-       int                     all_bh;
+       int                     all_bh = 0;
 
        trace_xfs_writepage(inode, page, 0);
 
+       ASSERT(page_has_buffers(page));
+
        /*
         * Refuse to write the page out if we are called from reclaim context.
         *
@@ -1072,29 +1062,15 @@ xfs_vm_writepage(
                goto out_fail;
 
        /*
-        * We need a transaction if:
-        *  1. There are delalloc buffers on the page
-        *  2. The page is uptodate and we have unmapped buffers
-        *  3. The page is uptodate and we have no buffers
-        *  4. There are unwritten buffers on the page
-        */
-       if (!page_has_buffers(page)) {
-               unmapped = 1;
-               need_trans = 1;
-       } else {
-               xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
-               if (!PageUptodate(page))
-                       unmapped = 0;
-               need_trans = delalloc + unmapped + unwritten;
-       }
-
-       /*
-        * If we need a transaction and the process flags say
-        * we are already in a transaction, or no IO is allowed
-        * then mark the page dirty again and leave the page
-        * as is.
+        * We need a transaction if there are delalloc or unwritten buffers
+        * on the page.
+        *
+        * If we need a transaction and the process flags say we are already
+        * in a transaction, or no IO is allowed then mark the page dirty
+        * again and leave the page as is.
         */
-       if (current_test_flags(PF_FSTRANS) && need_trans)
+       xfs_count_page_state(page, &delalloc, &unwritten);
+       if ((current->flags & PF_FSTRANS) && (delalloc || unwritten))
                goto out_fail;
 
        /*
@@ -1117,7 +1093,8 @@ xfs_vm_writepage(
        }
 
        end_offset = min_t(unsigned long long,
-                       (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
+                       (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
+                       offset);
        len = 1 << inode->i_blkbits;
 
        bh = head = page_buffers(page);
@@ -1125,8 +1102,6 @@ xfs_vm_writepage(
        flags = BMAPI_READ;
        type = IO_NEW;
 
-       all_bh = unmapped;
-
        do {
                if (offset >= end_offset)
                        break;
@@ -1146,19 +1121,7 @@ xfs_vm_writepage(
                if (imap_valid)
                        imap_valid = xfs_imap_valid(inode, &imap, offset);
 
-               /*
-                * First case, map an unwritten extent and prepare for
-                * extent state conversion transaction on completion.
-                *
-                * Second case, allocate space for a delalloc buffer.
-                * We can return EAGAIN here in the release page case.
-                *
-                * Third case, an unmapped buffer was found, and we are
-                * in a path where we need to write the whole page out.
-                */
-               if (buffer_unwritten(bh) || buffer_delay(bh) ||
-                   ((buffer_uptodate(bh) || PageUptodate(page)) &&
-                    !buffer_mapped(bh))) {
+               if (buffer_unwritten(bh) || buffer_delay(bh)) {
                        int new_ioend = 0;
 
                        /*
@@ -1177,14 +1140,11 @@ xfs_vm_writepage(
                                if (wbc->sync_mode == WB_SYNC_NONE &&
                                    wbc->nonblocking)
                                        flags |= BMAPI_TRYLOCK;
-                       } else {
-                               type = IO_NEW;
-                               flags = BMAPI_WRITE | BMAPI_MMAP;
                        }
 
                        if (!imap_valid) {
                                /*
-                                * if we didn't have a valid mapping then we
+                                * If we didn't have a valid mapping then we
                                 * need to ensure that we put the new mapping
                                 * in a new ioend structure. This needs to be
                                 * done to ensure that the ioends correctly
@@ -1192,14 +1152,7 @@ xfs_vm_writepage(
                                 * for unwritten extent conversion.
                                 */
                                new_ioend = 1;
-                               if (type == IO_NEW) {
-                                       size = xfs_probe_cluster(inode,
-                                                       page, bh, head, 0);
-                               } else {
-                                       size = len;
-                               }
-
-                               err = xfs_map_blocks(inode, offset, size,
+                               err = xfs_map_blocks(inode, offset, len,
                                                &imap, flags);
                                if (err)
                                        goto error;
@@ -1220,8 +1173,7 @@ xfs_vm_writepage(
                         */
                        if (!imap_valid || flags != BMAPI_READ) {
                                flags = BMAPI_READ;
-                               size = xfs_probe_cluster(inode, page, bh,
-                                                               head, 1);
+                               size = xfs_probe_cluster(inode, page, bh, head);
                                err = xfs_map_blocks(inode, offset, size,
                                                &imap, flags);
                                if (err)
@@ -1240,7 +1192,6 @@ xfs_vm_writepage(
                         */
                        type = IO_NEW;
                        if (trylock_buffer(bh)) {
-                               ASSERT(buffer_mapped(bh));
                                if (imap_valid)
                                        all_bh = 1;
                                xfs_add_to_ioend(inode, bh, offset, type,
@@ -1250,6 +1201,7 @@ xfs_vm_writepage(
                                imap_valid = 0;
                        }
                } else if (PageUptodate(page)) {
+                       ASSERT(buffer_mapped(bh));
                        imap_valid = 0;
                }
 
@@ -1291,8 +1243,7 @@ error:
        if (iohead)
                xfs_cancel_ioend(iohead);
 
-       if (!unmapped)
-               xfs_aops_discard_page(page);
+       xfs_aops_discard_page(page);
        ClearPageUptodate(page);
        unlock_page(page);
        return err;
@@ -1324,11 +1275,11 @@ xfs_vm_releasepage(
        struct page             *page,
        gfp_t                   gfp_mask)
 {
-       int                     delalloc, unmapped, unwritten;
+       int                     delalloc, unwritten;
 
        trace_xfs_releasepage(page->mapping->host, page, 0);
 
-       xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
+       xfs_count_page_state(page, &delalloc, &unwritten);
 
        if (WARN_ON(delalloc))
                return 0;
index 4cfc6ea87df8c4c34564da66ee31aba497dad902..319da173cc1aa61a0cf67f43bf73d0d6a8ba9087 100644 (file)
@@ -45,6 +45,6 @@ extern int xfs_get_blocks(struct inode *, sector_t, struct buffer_head *, int);
 extern void xfs_ioend_init(void);
 extern void xfs_ioend_wait(struct xfs_inode *);
 
-extern void xfs_count_page_state(struct page *, int *, int *, int *);
+extern void xfs_count_page_state(struct page *, int *, int *);
 
 #endif /* __XFS_AOPS_H__ */
index 0aea6d5f705a9caca2a2c80786c75700333e527c..3f2eec28c70dd7e9daa55648ffbf0ac907e386c2 100644 (file)
@@ -832,33 +832,29 @@ DECLARE_EVENT_CLASS(xfs_page_class,
                __field(loff_t, size)
                __field(unsigned long, offset)
                __field(int, delalloc)
-               __field(int, unmapped)
                __field(int, unwritten)
        ),
        TP_fast_assign(
-               int delalloc = -1, unmapped = -1, unwritten = -1;
+               int delalloc = -1, unwritten = -1;
 
                if (page_has_buffers(page))
-                       xfs_count_page_state(page, &delalloc,
-                                            &unmapped, &unwritten);
+                       xfs_count_page_state(page, &delalloc, &unwritten);
                __entry->dev = inode->i_sb->s_dev;
                __entry->ino = XFS_I(inode)->i_ino;
                __entry->pgoff = page_offset(page);
                __entry->size = i_size_read(inode);
                __entry->offset = off;
                __entry->delalloc = delalloc;
-               __entry->unmapped = unmapped;
                __entry->unwritten = unwritten;
        ),
        TP_printk("dev %d:%d ino 0x%llx pgoff 0x%lx size 0x%llx offset %lx "
-                 "delalloc %d unmapped %d unwritten %d",
+                 "delalloc %d unwritten %d",
                  MAJOR(__entry->dev), MINOR(__entry->dev),
                  __entry->ino,
                  __entry->pgoff,
                  __entry->size,
                  __entry->offset,
                  __entry->delalloc,
-                 __entry->unmapped,
                  __entry->unwritten)
 )