]> git.karo-electronics.de Git - karo-tx-linux.git/blobdiff - mm/filemap.c
signalfd: fill in ssi_int for posix timers and message queues
[karo-tx-linux.git] / mm / filemap.c
index 42bbc6909ba4852ac41506b175584ba4d00e32eb..a4800c51d6c99c23a0eae9c53dc46a93b0cf808b 100644 (file)
@@ -209,7 +209,7 @@ int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
        int ret;
        struct writeback_control wbc = {
                .sync_mode = sync_mode,
-               .nr_to_write = mapping->nrpages * 2,
+               .nr_to_write = LONG_MAX,
                .range_start = start,
                .range_end = end,
        };
@@ -558,14 +558,14 @@ EXPORT_SYMBOL(wait_on_page_bit);
  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
  *
  * The first mb is necessary to safely close the critical section opened by the
- * TestSetPageLocked(), the second mb is necessary to enforce ordering between
- * the clear_bit and the read of the waitqueue (to avoid SMP races with a
- * parallel wait_on_page_locked()).
+ * test_and_set_bit() to lock the page; the second mb is necessary to enforce
+ * ordering between the clear_bit and the read of the waitqueue (to avoid SMP
+ * races with a parallel wait_on_page_locked()).
  */
 void unlock_page(struct page *page)
 {
        smp_mb__before_clear_bit();
-       if (!TestClearPageLocked(page))
+       if (!test_and_clear_bit(PG_locked, &page->flags))
                BUG();
        smp_mb__after_clear_bit(); 
        wake_up_page(page, PG_locked);
@@ -931,7 +931,7 @@ grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
        struct page *page = find_get_page(mapping, index);
 
        if (page) {
-               if (!TestSetPageLocked(page))
+               if (trylock_page(page))
                        return page;
                page_cache_release(page);
                return NULL;
@@ -1027,7 +1027,7 @@ find_page:
                        if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
                                        !mapping->a_ops->is_partially_uptodate)
                                goto page_not_up_to_date;
-                       if (TestSetPageLocked(page))
+                       if (!trylock_page(page))
                                goto page_not_up_to_date;
                        if (!mapping->a_ops->is_partially_uptodate(page,
                                                                desc, offset))
@@ -1118,6 +1118,12 @@ page_not_up_to_date_locked:
                }
 
 readpage:
+               /*
+                * A previous I/O error may have been due to temporary
+                * failures, eg. multipath errors.
+                * PG_error will be set again if readpage fails.
+                */
+               ClearPageError(page);
                /* Start the actual read. The read will unlock the page. */
                error = mapping->a_ops->readpage(filp, page);
 
@@ -1304,7 +1310,8 @@ generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
                        goto out; /* skip atime */
                size = i_size_read(inode);
                if (pos < size) {
-                       retval = filemap_write_and_wait(mapping);
+                       retval = filemap_write_and_wait_range(mapping, pos,
+                                       pos + iov_length(iov, nr_segs) - 1);
                        if (!retval) {
                                retval = mapping->a_ops->direct_IO(READ, iocb,
                                                        iov, pos, nr_segs);
@@ -1353,7 +1360,7 @@ do_readahead(struct address_space *mapping, struct file *filp,
        return 0;
 }
 
-asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
+SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
 {
        ssize_t ret;
        struct file *file;
@@ -1372,6 +1379,13 @@ asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
        }
        return ret;
 }
+#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
+asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
+{
+       return SYSC_readahead((int) fd, offset, (size_t) count);
+}
+SYSCALL_ALIAS(sys_readahead, SyS_readahead);
+#endif
 
 #ifdef CONFIG_MMU
 /**
@@ -1879,7 +1893,7 @@ void iov_iter_advance(struct iov_iter *i, size_t bytes)
                 * The !iov->iov_len check ensures we skip over unlikely
                 * zero-length segments (without overruning the iovec).
                 */
-               while (bytes || unlikely(!iov->iov_len && i->count)) {
+               while (bytes || unlikely(i->count && !iov->iov_len)) {
                        int copy;
 
                        copy = min(bytes, iov->iov_len - base);
@@ -2026,7 +2040,7 @@ int pagecache_write_begin(struct file *file, struct address_space *mapping,
                struct inode *inode = mapping->host;
                struct page *page;
 again:
-               page = __grab_cache_page(mapping, index);
+               page = grab_cache_page_write_begin(mapping, index, flags);
                *pagep = page;
                if (!page)
                        return -ENOMEM;
@@ -2110,18 +2124,10 @@ generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
        if (count != ocount)
                *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
 
-       /*
-        * Unmap all mmappings of the file up-front.
-        *
-        * This will cause any pte dirty bits to be propagated into the
-        * pageframes for the subsequent filemap_write_and_wait().
-        */
        write_len = iov_length(iov, *nr_segs);
        end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
-       if (mapping_mapped(mapping))
-               unmap_mapping_range(mapping, pos, write_len, 0);
 
-       written = filemap_write_and_wait(mapping);
+       written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
        if (written)
                goto out;
 
@@ -2129,13 +2135,20 @@ generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
         * After a write we want buffered reads to be sure to go to disk to get
         * the new data.  We invalidate clean cached page from the region we're
         * about to write.  We do this *before* the write so that we can return
-        * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
+        * without clobbering -EIOCBQUEUED from ->direct_IO().
         */
        if (mapping->nrpages) {
                written = invalidate_inode_pages2_range(mapping,
                                        pos >> PAGE_CACHE_SHIFT, end);
-               if (written)
+               /*
+                * If a page can not be invalidated, return 0 to fall back
+                * to buffered write.
+                */
+               if (written) {
+                       if (written == -EBUSY)
+                               return 0;
                        goto out;
+               }
        }
 
        written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
@@ -2183,19 +2196,24 @@ EXPORT_SYMBOL(generic_file_direct_write);
  * Find or create a page at the given pagecache position. Return the locked
  * page. This function is specifically for buffered writes.
  */
-struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
+struct page *grab_cache_page_write_begin(struct address_space *mapping,
+                                       pgoff_t index, unsigned flags)
 {
        int status;
        struct page *page;
+       gfp_t gfp_notmask = 0;
+       if (flags & AOP_FLAG_NOFS)
+               gfp_notmask = __GFP_FS;
 repeat:
        page = find_lock_page(mapping, index);
        if (likely(page))
                return page;
 
-       page = page_cache_alloc(mapping);
+       page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
        if (!page)
                return NULL;
-       status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
+       status = add_to_page_cache_lru(page, mapping, index,
+                                               GFP_KERNEL & ~gfp_notmask);
        if (unlikely(status)) {
                page_cache_release(page);
                if (status == -EEXIST)
@@ -2204,7 +2222,7 @@ repeat:
        }
        return page;
 }
-EXPORT_SYMBOL(__grab_cache_page);
+EXPORT_SYMBOL(grab_cache_page_write_begin);
 
 static ssize_t generic_perform_write_2copy(struct file *file,
                                struct iov_iter *i, loff_t pos)
@@ -2249,7 +2267,7 @@ static ssize_t generic_perform_write_2copy(struct file *file,
                        break;
                }
 
-               page = __grab_cache_page(mapping, index);
+               page = grab_cache_page_write_begin(mapping, index, 0);
                if (!page) {
                        status = -ENOMEM;
                        break;
@@ -2429,6 +2447,7 @@ again:
                pagefault_enable();
                flush_dcache_page(page);
 
+               mark_page_accessed(page);
                status = a_ops->write_end(file, mapping, pos, bytes, copied,
                                                page, fsdata);
                if (unlikely(status < 0))
@@ -2500,7 +2519,8 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
         * the file data here, to try to honour O_DIRECT expectations.
         */
        if (unlikely(file->f_flags & O_DIRECT) && written)
-               status = filemap_write_and_wait(mapping);
+               status = filemap_write_and_wait_range(mapping,
+                                       pos, pos + written - 1);
 
        return written ? written : status;
 }