1 /* handling of writes to regular files and writing back to the server
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 #include <linux/backing-dev.h>
12 #include <linux/slab.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/pagevec.h>
19 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
23 * mark a page as having been made dirty and thus needing writeback
25 int afs_set_page_dirty(struct page *page)
28 return __set_page_dirty_nobuffers(page);
32 * unlink a writeback record because its usage has reached zero
33 * - must be called with the wb->vnode->writeback_lock held
35 static void afs_unlink_writeback(struct afs_writeback *wb)
37 struct afs_writeback *front;
38 struct afs_vnode *vnode = wb->vnode;
40 list_del_init(&wb->link);
41 if (!list_empty(&vnode->writebacks)) {
42 /* if an fsync rises to the front of the queue then wake it
44 front = list_entry(vnode->writebacks.next,
45 struct afs_writeback, link);
46 if (front->state == AFS_WBACK_SYNCING) {
47 _debug("wake up sync");
48 front->state = AFS_WBACK_COMPLETE;
49 wake_up(&front->waitq);
55 * free a writeback record
57 static void afs_free_writeback(struct afs_writeback *wb)
65 * dispose of a reference to a writeback record
67 void afs_put_writeback(struct afs_writeback *wb)
69 struct afs_vnode *vnode = wb->vnode;
71 _enter("{%d}", wb->usage);
73 spin_lock(&vnode->writeback_lock);
75 afs_unlink_writeback(wb);
78 spin_unlock(&vnode->writeback_lock);
80 afs_free_writeback(wb);
84 * partly or wholly fill a page that's under preparation for writing
86 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
87 loff_t pos, unsigned int len, struct page *page)
92 _enter(",,%llu", (unsigned long long)pos);
94 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
99 atomic_set(&req->usage, 1);
103 req->pages[0] = page;
106 ret = afs_vnode_fetch_data(vnode, key, req);
109 if (ret == -ENOENT) {
110 _debug("got NOENT from server"
111 " - marking file deleted and stale");
112 set_bit(AFS_VNODE_DELETED, &vnode->flags);
117 _leave(" = %d", ret);
122 * prepare to perform part of a write to a page
124 int afs_write_begin(struct file *file, struct address_space *mapping,
125 loff_t pos, unsigned len, unsigned flags,
126 struct page **pagep, void **fsdata)
128 struct afs_writeback *candidate, *wb;
129 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
131 struct key *key = file->private_data;
132 unsigned from = pos & (PAGE_SIZE - 1);
133 unsigned to = from + len;
134 pgoff_t index = pos >> PAGE_SHIFT;
137 _enter("{%x:%u},{%lx},%u,%u",
138 vnode->fid.vid, vnode->fid.vnode, index, from, to);
140 candidate = kzalloc(sizeof(*candidate), GFP_KERNEL);
143 candidate->vnode = vnode;
144 candidate->first = candidate->last = index;
145 candidate->offset_first = from;
146 candidate->to_last = to;
147 INIT_LIST_HEAD(&candidate->link);
148 candidate->usage = 1;
149 candidate->state = AFS_WBACK_PENDING;
150 init_waitqueue_head(&candidate->waitq);
152 page = grab_cache_page_write_begin(mapping, index, flags);
158 if (!PageUptodate(page) && len != PAGE_SIZE) {
159 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
164 _leave(" = %d [prep]", ret);
167 SetPageUptodate(page);
170 /* page won't leak in error case: it eventually gets cleaned off LRU */
174 spin_lock(&vnode->writeback_lock);
176 /* see if this page is already pending a writeback under a suitable key
177 * - if so we can just join onto that one */
178 wb = (struct afs_writeback *) page_private(page);
180 if (wb->key == key && wb->state == AFS_WBACK_PENDING)
181 goto subsume_in_current_wb;
182 goto flush_conflicting_wb;
186 /* see if we can find an already pending writeback that we can
187 * append this page to */
188 list_for_each_entry(wb, &vnode->writebacks, link) {
189 if (wb->last == index - 1 && wb->key == key &&
190 wb->state == AFS_WBACK_PENDING)
191 goto append_to_previous_wb;
195 list_add_tail(&candidate->link, &vnode->writebacks);
196 candidate->key = key_get(key);
197 spin_unlock(&vnode->writeback_lock);
198 SetPagePrivate(page);
199 set_page_private(page, (unsigned long) candidate);
200 _leave(" = 0 [new]");
203 subsume_in_current_wb:
205 ASSERTRANGE(wb->first, <=, index, <=, wb->last);
206 if (index == wb->first && from < wb->offset_first)
207 wb->offset_first = from;
208 if (index == wb->last && to > wb->to_last)
210 spin_unlock(&vnode->writeback_lock);
212 _leave(" = 0 [sub]");
215 append_to_previous_wb:
216 _debug("append into %lx-%lx", wb->first, wb->last);
220 spin_unlock(&vnode->writeback_lock);
221 SetPagePrivate(page);
222 set_page_private(page, (unsigned long) wb);
224 _leave(" = 0 [app]");
227 /* the page is currently bound to another context, so if it's dirty we
228 * need to flush it before we can use the new context */
229 flush_conflicting_wb:
230 _debug("flush conflict");
231 if (wb->state == AFS_WBACK_PENDING)
232 wb->state = AFS_WBACK_CONFLICTING;
233 spin_unlock(&vnode->writeback_lock);
234 if (clear_page_dirty_for_io(page)) {
235 ret = afs_write_back_from_locked_page(wb, page);
237 afs_put_writeback(candidate);
238 _leave(" = %d", ret);
243 /* the page holds a ref on the writeback record */
244 afs_put_writeback(wb);
245 set_page_private(page, 0);
246 ClearPagePrivate(page);
251 * finalise part of a write to a page
253 int afs_write_end(struct file *file, struct address_space *mapping,
254 loff_t pos, unsigned len, unsigned copied,
255 struct page *page, void *fsdata)
257 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
258 struct key *key = file->private_data;
259 loff_t i_size, maybe_i_size;
262 _enter("{%x:%u},{%lx}",
263 vnode->fid.vid, vnode->fid.vnode, page->index);
265 maybe_i_size = pos + copied;
267 i_size = i_size_read(&vnode->vfs_inode);
268 if (maybe_i_size > i_size) {
269 spin_lock(&vnode->writeback_lock);
270 i_size = i_size_read(&vnode->vfs_inode);
271 if (maybe_i_size > i_size)
272 i_size_write(&vnode->vfs_inode, maybe_i_size);
273 spin_unlock(&vnode->writeback_lock);
276 if (!PageUptodate(page)) {
278 /* Try and load any missing data from the server. The
279 * unmarshalling routine will take care of clearing any
280 * bits that are beyond the EOF.
282 ret = afs_fill_page(vnode, key, pos + copied,
287 SetPageUptodate(page);
290 set_page_dirty(page);
300 * kill all the pages in the given range
302 static void afs_kill_pages(struct afs_vnode *vnode, bool error,
303 pgoff_t first, pgoff_t last)
306 unsigned count, loop;
308 _enter("{%x:%u},%lx-%lx",
309 vnode->fid.vid, vnode->fid.vnode, first, last);
314 _debug("kill %lx-%lx", first, last);
316 count = last - first + 1;
317 if (count > PAGEVEC_SIZE)
318 count = PAGEVEC_SIZE;
319 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
320 first, count, pv.pages);
321 ASSERTCMP(pv.nr, ==, count);
323 for (loop = 0; loop < count; loop++) {
324 struct page *page = pv.pages[loop];
325 ClearPageUptodate(page);
328 if (PageWriteback(page))
329 end_page_writeback(page);
330 if (page->index >= first)
331 first = page->index + 1;
334 __pagevec_release(&pv);
335 } while (first < last);
341 * synchronously write back the locked page and any subsequent non-locked dirty
342 * pages also covered by the same writeback record
344 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
345 struct page *primary_page)
347 struct page *pages[8], *page;
349 unsigned n, offset, to;
350 pgoff_t start, first, last;
353 _enter(",%lx", primary_page->index);
356 if (test_set_page_writeback(primary_page))
359 /* find all consecutive lockable dirty pages, stopping when we find a
360 * page that is not immediately lockable, is not dirty or is missing,
361 * or we reach the end of the range */
362 start = primary_page->index;
363 if (start >= wb->last)
367 _debug("more %lx [%lx]", start, count);
368 n = wb->last - start + 1;
369 if (n > ARRAY_SIZE(pages))
370 n = ARRAY_SIZE(pages);
371 n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping,
373 _debug("fgpc %u", n);
376 if (pages[0]->index != start) {
378 put_page(pages[--n]);
383 for (loop = 0; loop < n; loop++) {
385 if (page->index > wb->last)
387 if (!trylock_page(page))
389 if (!PageDirty(page) ||
390 page_private(page) != (unsigned long) wb) {
394 if (!clear_page_dirty_for_io(page))
396 if (test_set_page_writeback(page))
403 for (; loop < n; loop++)
404 put_page(pages[loop]);
409 } while (start <= wb->last && count < 65536);
412 /* we now have a contiguous set of dirty pages, each with writeback set
413 * and the dirty mark cleared; the first page is locked and must remain
414 * so, all the rest are unlocked */
415 first = primary_page->index;
416 last = first + count - 1;
418 offset = (first == wb->first) ? wb->offset_first : 0;
419 to = (last == wb->last) ? wb->to_last : PAGE_SIZE;
421 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
423 ret = afs_vnode_store_data(wb, first, last, offset, to);
428 mapping_set_error(wb->vnode->vfs_inode.i_mapping, -ENOSPC);
437 afs_kill_pages(wb->vnode, true, first, last);
438 mapping_set_error(wb->vnode->vfs_inode.i_mapping, -EIO);
446 afs_kill_pages(wb->vnode, false, first, last);
455 _leave(" = %d", ret);
460 * write a page back to the server
461 * - the caller locked the page for us
463 int afs_writepage(struct page *page, struct writeback_control *wbc)
465 struct afs_writeback *wb;
468 _enter("{%lx},", page->index);
470 wb = (struct afs_writeback *) page_private(page);
473 ret = afs_write_back_from_locked_page(wb, page);
476 _leave(" = %d", ret);
480 wbc->nr_to_write -= ret;
487 * write a region of pages back to the server
489 static int afs_writepages_region(struct address_space *mapping,
490 struct writeback_control *wbc,
491 pgoff_t index, pgoff_t end, pgoff_t *_next)
493 struct afs_writeback *wb;
497 _enter(",,%lx,%lx,", index, end);
500 n = find_get_pages_range_tag(mapping, &index, end,
501 PAGECACHE_TAG_DIRTY, 1, &page);
505 _debug("wback %lx", page->index);
507 /* at this point we hold neither mapping->tree_lock nor lock on
508 * the page itself: the page may be truncated or invalidated
509 * (changing page->mapping to NULL), or even swizzled back from
510 * swapper_space to tmpfs file mapping
514 if (page->mapping != mapping || !PageDirty(page)) {
520 if (PageWriteback(page)) {
522 if (wbc->sync_mode != WB_SYNC_NONE)
523 wait_on_page_writeback(page);
528 wb = (struct afs_writeback *) page_private(page);
531 spin_lock(&wb->vnode->writeback_lock);
532 wb->state = AFS_WBACK_WRITING;
533 spin_unlock(&wb->vnode->writeback_lock);
535 if (!clear_page_dirty_for_io(page))
537 ret = afs_write_back_from_locked_page(wb, page);
541 _leave(" = %d", ret);
545 wbc->nr_to_write -= ret;
548 } while (index < end && wbc->nr_to_write > 0);
551 _leave(" = 0 [%lx]", *_next);
556 * write some of the pending data back to the server
558 int afs_writepages(struct address_space *mapping,
559 struct writeback_control *wbc)
561 pgoff_t start, end, next;
566 if (wbc->range_cyclic) {
567 start = mapping->writeback_index;
569 ret = afs_writepages_region(mapping, wbc, start, end, &next);
570 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
571 ret = afs_writepages_region(mapping, wbc, 0, start,
573 mapping->writeback_index = next;
574 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
575 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
576 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
577 if (wbc->nr_to_write > 0)
578 mapping->writeback_index = next;
580 start = wbc->range_start >> PAGE_SHIFT;
581 end = wbc->range_end >> PAGE_SHIFT;
582 ret = afs_writepages_region(mapping, wbc, start, end, &next);
585 _leave(" = %d", ret);
590 * completion of write to server
592 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
594 struct afs_writeback *wb = call->wb;
596 unsigned count, loop;
597 pgoff_t first = call->first, last = call->last;
600 _enter("{%x:%u},{%lx-%lx}",
601 vnode->fid.vid, vnode->fid.vnode, first, last);
608 _debug("done %lx-%lx", first, last);
610 count = last - first + 1;
611 if (count > PAGEVEC_SIZE)
612 count = PAGEVEC_SIZE;
613 pv.nr = find_get_pages_contig(call->mapping, first, count,
615 ASSERTCMP(pv.nr, ==, count);
617 spin_lock(&vnode->writeback_lock);
618 for (loop = 0; loop < count; loop++) {
619 struct page *page = pv.pages[loop];
620 end_page_writeback(page);
621 if (page_private(page) == (unsigned long) wb) {
622 set_page_private(page, 0);
623 ClearPagePrivate(page);
628 if (wb->usage == 0) {
629 afs_unlink_writeback(wb);
632 spin_unlock(&vnode->writeback_lock);
635 afs_free_writeback(wb);
639 __pagevec_release(&pv);
640 } while (first <= last);
646 * write to an AFS file
648 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
650 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
652 size_t count = iov_iter_count(from);
654 _enter("{%x.%u},{%zu},",
655 vnode->fid.vid, vnode->fid.vnode, count);
657 if (IS_SWAPFILE(&vnode->vfs_inode)) {
659 "AFS: Attempt to write to active swap file!\n");
666 result = generic_file_write_iter(iocb, from);
668 _leave(" = %zd", result);
673 * flush the vnode to the fileserver
675 int afs_writeback_all(struct afs_vnode *vnode)
677 struct address_space *mapping = vnode->vfs_inode.i_mapping;
678 struct writeback_control wbc = {
679 .sync_mode = WB_SYNC_ALL,
680 .nr_to_write = LONG_MAX,
687 ret = mapping->a_ops->writepages(mapping, &wbc);
688 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
690 _leave(" = %d", ret);
695 * flush any dirty pages for this process, and check for write errors.
696 * - the return status from this call provides a reliable indication of
697 * whether any write errors occurred for this process.
699 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
701 struct inode *inode = file_inode(file);
702 struct afs_writeback *wb, *xwb;
703 struct afs_vnode *vnode = AFS_FS_I(inode);
706 _enter("{%x:%u},{n=%pD},%d",
707 vnode->fid.vid, vnode->fid.vnode, file,
710 ret = file_write_and_wait_range(file, start, end);
715 /* use a writeback record as a marker in the queue - when this reaches
716 * the front of the queue, all the outstanding writes are either
717 * completed or rejected */
718 wb = kzalloc(sizeof(*wb), GFP_KERNEL);
726 wb->offset_first = 0;
727 wb->to_last = PAGE_SIZE;
729 wb->state = AFS_WBACK_SYNCING;
730 init_waitqueue_head(&wb->waitq);
732 spin_lock(&vnode->writeback_lock);
733 list_for_each_entry(xwb, &vnode->writebacks, link) {
734 if (xwb->state == AFS_WBACK_PENDING)
735 xwb->state = AFS_WBACK_CONFLICTING;
737 list_add_tail(&wb->link, &vnode->writebacks);
738 spin_unlock(&vnode->writeback_lock);
740 /* push all the outstanding writebacks to the server */
741 ret = afs_writeback_all(vnode);
743 afs_put_writeback(wb);
744 _leave(" = %d [wb]", ret);
748 /* wait for the preceding writes to actually complete */
749 ret = wait_event_interruptible(wb->waitq,
750 wb->state == AFS_WBACK_COMPLETE ||
751 vnode->writebacks.next == &wb->link);
752 afs_put_writeback(wb);
753 _leave(" = %d", ret);
760 * Flush out all outstanding writes on a file opened for writing when it is
763 int afs_flush(struct file *file, fl_owner_t id)
767 if ((file->f_mode & FMODE_WRITE) == 0)
770 return vfs_fsync(file, 0);
774 * notification that a previously read-only page is about to become writable
775 * - if it returns an error, the caller will deliver a bus error signal
777 int afs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
779 struct afs_vnode *vnode = AFS_FS_I(vma->vm_file->f_mapping->host);
781 _enter("{{%x:%u}},{%lx}",
782 vnode->fid.vid, vnode->fid.vnode, page->index);
784 /* wait for the page to be written to the cache before we allow it to
786 #ifdef CONFIG_AFS_FSCACHE
787 fscache_wait_on_page_write(vnode->cache, page);