mm: trim more holes
[linux-2.6.git] / mm / filemap.c
index c504db1..73b98c6 100644 (file)
@@ -1838,16 +1838,15 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
                size_t count, ssize_t written)
 {
        struct file *file = iocb->ki_filp;
-       struct address_space * mapping = file->f_mapping;
+       struct address_space *mapping = file->f_mapping;
        const struct address_space_operations *a_ops = mapping->a_ops;
        struct inode    *inode = mapping->host;
        long            status = 0;
        struct page     *page;
        struct page     *cached_page = NULL;
-       size_t          bytes;
        struct pagevec  lru_pvec;
        const struct iovec *cur_iov = iov; /* current iovec */
-       size_t          iov_base = 0;      /* offset in the current iovec */
+       size_t          iov_offset = 0;    /* offset in the current iovec */
        char __user     *buf;
 
        pagevec_init(&lru_pvec, 0);
@@ -1858,36 +1857,36 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
        if (likely(nr_segs == 1))
                buf = iov->iov_base + written;
        else {
-               filemap_set_next_iovec(&cur_iov, &iov_base, written);
-               buf = cur_iov->iov_base + iov_base;
+               filemap_set_next_iovec(&cur_iov, &iov_offset, written);
+               buf = cur_iov->iov_base + iov_offset;
        }
 
        do {
-               unsigned long index;
-               unsigned long offset;
-               size_t copied;
+               pgoff_t index;          /* Pagecache index for current page */
+               unsigned long offset;   /* Offset into pagecache page */
+               unsigned long maxlen;   /* Bytes remaining in current iovec */
+               size_t bytes;           /* Bytes to write to page */
+               size_t copied;          /* Bytes copied from user */
 
-               offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
+               offset = (pos & (PAGE_CACHE_SIZE - 1));
                index = pos >> PAGE_CACHE_SHIFT;
                bytes = PAGE_CACHE_SIZE - offset;
+               if (bytes > count)
+                       bytes = count;
 
-               /* Limit the size of the copy to the caller's write size */
-               bytes = min(bytes, count);
-
-               /*
-                * Limit the size of the copy to that of the current segment,
-                * because fault_in_pages_readable() doesn't know how to walk
-                * segments.
-                */
-               bytes = min(bytes, cur_iov->iov_len - iov_base);
+               maxlen = cur_iov->iov_len - iov_offset;
+               if (maxlen > bytes)
+                       maxlen = bytes;
 
+#ifndef CONFIG_DEBUG_VM
                /*
                 * Bring in the user page that we will copy from _first_.
                 * Otherwise there's a nasty deadlock on copying from the
                 * same page as we're writing to, without it being marked
                 * up-to-date.
                 */
-               fault_in_pages_readable(buf, bytes);
+               fault_in_pages_readable(buf, maxlen);
+#endif
 
                page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
                if (!page) {
@@ -1895,72 +1894,65 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
                        break;
                }
 
-               if (unlikely(bytes == 0)) {
-                       status = 0;
-                       copied = 0;
-                       goto zero_length_segment;
-               }
-
                status = a_ops->prepare_write(file, page, offset, offset+bytes);
-               if (unlikely(status)) {
-                       loff_t isize = i_size_read(inode);
+               if (unlikely(status))
+                       goto fs_write_aop_error;
 
-                       if (status != AOP_TRUNCATED_PAGE)
-                               unlock_page(page);
-                       page_cache_release(page);
-                       if (status == AOP_TRUNCATED_PAGE)
-                               continue;
-                       /*
-                        * prepare_write() may have instantiated a few blocks
-                        * outside i_size.  Trim these off again.
-                        */
-                       if (pos + bytes > isize)
-                               vmtruncate(inode, isize);
-                       break;
-               }
                if (likely(nr_segs == 1))
                        copied = filemap_copy_from_user(page, offset,
                                                        buf, bytes);
                else
                        copied = filemap_copy_from_user_iovec(page, offset,
-                                               cur_iov, iov_base, bytes);
+                                               cur_iov, iov_offset, bytes);
                flush_dcache_page(page);
                status = a_ops->commit_write(file, page, offset, offset+bytes);
-               if (status == AOP_TRUNCATED_PAGE) {
-                       page_cache_release(page);
-                       continue;
+               if (unlikely(status < 0 || status == AOP_TRUNCATED_PAGE))
+                       goto fs_write_aop_error;
+               if (unlikely(copied != bytes)) {
+                       status = -EFAULT;
+                       goto fs_write_aop_error;
                }
-zero_length_segment:
-               if (likely(copied >= 0)) {
-                       if (!status)
-                               status = copied;
-
-                       if (status >= 0) {
-                               written += status;
-                               count -= status;
-                               pos += status;
-                               buf += status;
-                               if (unlikely(nr_segs > 1)) {
-                                       filemap_set_next_iovec(&cur_iov,
-                                                       &iov_base, status);
-                                       if (count)
-                                               buf = cur_iov->iov_base +
-                                                       iov_base;
-                               } else {
-                                       iov_base += status;
-                               }
+               if (unlikely(status > 0)) /* filesystem did partial write */
+                       copied = status;
+
+               if (likely(copied > 0)) {
+                       written += copied;
+                       count -= copied;
+                       pos += copied;
+                       buf += copied;
+                       if (unlikely(nr_segs > 1)) {
+                               filemap_set_next_iovec(&cur_iov,
+                                               &iov_offset, copied);
+                               if (count)
+                                       buf = cur_iov->iov_base + iov_offset;
+                       } else {
+                               iov_offset += copied;
                        }
                }
-               if (unlikely(copied != bytes))
-                       if (status >= 0)
-                               status = -EFAULT;
                unlock_page(page);
                mark_page_accessed(page);
                page_cache_release(page);
-               if (status < 0)
-                       break;
                balance_dirty_pages_ratelimited(mapping);
                cond_resched();
+               continue;
+
+fs_write_aop_error:
+               if (status != AOP_TRUNCATED_PAGE)
+                       unlock_page(page);
+               page_cache_release(page);
+
+               /*
+                * prepare_write() may have instantiated a few blocks
+                * outside i_size.  Trim these off again. Don't need
+                * i_size_read because we hold i_mutex.
+                */
+               if (pos + bytes > inode->i_size)
+                       vmtruncate(inode, inode->i_size);
+               if (status == AOP_TRUNCATED_PAGE)
+                       continue;
+               else
+                       break;
+
        } while (count);
        *ppos = pos;