Merge branch 'x86-smep-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6/next.git] / fs / afs / write.c
blob789b3afb342328dcfe551d39227a81f09ffb3470
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>
13 #include <linux/fs.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/pagevec.h>
17 #include "internal.h"
19 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
20 struct page *page);
23 * mark a page as having been made dirty and thus needing writeback
25 int afs_set_page_dirty(struct page *page)
27 _enter("");
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
43 * up */
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)
59 _enter("");
60 key_put(wb->key);
61 kfree(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);
74 if (--wb->usage == 0)
75 afs_unlink_writeback(wb);
76 else
77 wb = NULL;
78 spin_unlock(&vnode->writeback_lock);
79 if (wb)
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 len, struct page *page)
89 loff_t i_size;
90 unsigned eof;
91 int ret;
93 _enter(",,%llu,%u", (unsigned long long)pos, len);
95 ASSERTCMP(len, <=, PAGE_CACHE_SIZE);
97 i_size = i_size_read(&vnode->vfs_inode);
98 if (pos + len > i_size)
99 eof = i_size;
100 else
101 eof = PAGE_CACHE_SIZE;
103 ret = afs_vnode_fetch_data(vnode, key, 0, eof, page);
104 if (ret < 0) {
105 if (ret == -ENOENT) {
106 _debug("got NOENT from server"
107 " - marking file deleted and stale");
108 set_bit(AFS_VNODE_DELETED, &vnode->flags);
109 ret = -ESTALE;
113 _leave(" = %d", ret);
114 return ret;
118 * prepare to perform part of a write to a page
120 int afs_write_begin(struct file *file, struct address_space *mapping,
121 loff_t pos, unsigned len, unsigned flags,
122 struct page **pagep, void **fsdata)
124 struct afs_writeback *candidate, *wb;
125 struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
126 struct page *page;
127 struct key *key = file->private_data;
128 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
129 unsigned to = from + len;
130 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
131 int ret;
133 _enter("{%x:%u},{%lx},%u,%u",
134 vnode->fid.vid, vnode->fid.vnode, index, from, to);
136 candidate = kzalloc(sizeof(*candidate), GFP_KERNEL);
137 if (!candidate)
138 return -ENOMEM;
139 candidate->vnode = vnode;
140 candidate->first = candidate->last = index;
141 candidate->offset_first = from;
142 candidate->to_last = to;
143 INIT_LIST_HEAD(&candidate->link);
144 candidate->usage = 1;
145 candidate->state = AFS_WBACK_PENDING;
146 init_waitqueue_head(&candidate->waitq);
148 page = grab_cache_page_write_begin(mapping, index, flags);
149 if (!page) {
150 kfree(candidate);
151 return -ENOMEM;
153 *pagep = page;
154 /* page won't leak in error case: it eventually gets cleaned off LRU */
156 if (!PageUptodate(page)) {
157 _debug("not up to date");
158 ret = afs_fill_page(vnode, key, pos, len, page);
159 if (ret < 0) {
160 kfree(candidate);
161 _leave(" = %d [prep]", ret);
162 return ret;
164 SetPageUptodate(page);
167 try_again:
168 spin_lock(&vnode->writeback_lock);
170 /* see if this page is already pending a writeback under a suitable key
171 * - if so we can just join onto that one */
172 wb = (struct afs_writeback *) page_private(page);
173 if (wb) {
174 if (wb->key == key && wb->state == AFS_WBACK_PENDING)
175 goto subsume_in_current_wb;
176 goto flush_conflicting_wb;
179 if (index > 0) {
180 /* see if we can find an already pending writeback that we can
181 * append this page to */
182 list_for_each_entry(wb, &vnode->writebacks, link) {
183 if (wb->last == index - 1 && wb->key == key &&
184 wb->state == AFS_WBACK_PENDING)
185 goto append_to_previous_wb;
189 list_add_tail(&candidate->link, &vnode->writebacks);
190 candidate->key = key_get(key);
191 spin_unlock(&vnode->writeback_lock);
192 SetPagePrivate(page);
193 set_page_private(page, (unsigned long) candidate);
194 _leave(" = 0 [new]");
195 return 0;
197 subsume_in_current_wb:
198 _debug("subsume");
199 ASSERTRANGE(wb->first, <=, index, <=, wb->last);
200 if (index == wb->first && from < wb->offset_first)
201 wb->offset_first = from;
202 if (index == wb->last && to > wb->to_last)
203 wb->to_last = to;
204 spin_unlock(&vnode->writeback_lock);
205 kfree(candidate);
206 _leave(" = 0 [sub]");
207 return 0;
209 append_to_previous_wb:
210 _debug("append into %lx-%lx", wb->first, wb->last);
211 wb->usage++;
212 wb->last++;
213 wb->to_last = to;
214 spin_unlock(&vnode->writeback_lock);
215 SetPagePrivate(page);
216 set_page_private(page, (unsigned long) wb);
217 kfree(candidate);
218 _leave(" = 0 [app]");
219 return 0;
221 /* the page is currently bound to another context, so if it's dirty we
222 * need to flush it before we can use the new context */
223 flush_conflicting_wb:
224 _debug("flush conflict");
225 if (wb->state == AFS_WBACK_PENDING)
226 wb->state = AFS_WBACK_CONFLICTING;
227 spin_unlock(&vnode->writeback_lock);
228 if (PageDirty(page)) {
229 ret = afs_write_back_from_locked_page(wb, page);
230 if (ret < 0) {
231 afs_put_writeback(candidate);
232 _leave(" = %d", ret);
233 return ret;
237 /* the page holds a ref on the writeback record */
238 afs_put_writeback(wb);
239 set_page_private(page, 0);
240 ClearPagePrivate(page);
241 goto try_again;
245 * finalise part of a write to a page
247 int afs_write_end(struct file *file, struct address_space *mapping,
248 loff_t pos, unsigned len, unsigned copied,
249 struct page *page, void *fsdata)
251 struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
252 loff_t i_size, maybe_i_size;
254 _enter("{%x:%u},{%lx}",
255 vnode->fid.vid, vnode->fid.vnode, page->index);
257 maybe_i_size = pos + copied;
259 i_size = i_size_read(&vnode->vfs_inode);
260 if (maybe_i_size > i_size) {
261 spin_lock(&vnode->writeback_lock);
262 i_size = i_size_read(&vnode->vfs_inode);
263 if (maybe_i_size > i_size)
264 i_size_write(&vnode->vfs_inode, maybe_i_size);
265 spin_unlock(&vnode->writeback_lock);
268 set_page_dirty(page);
269 if (PageDirty(page))
270 _debug("dirtied");
271 unlock_page(page);
272 page_cache_release(page);
274 return copied;
278 * kill all the pages in the given range
280 static void afs_kill_pages(struct afs_vnode *vnode, bool error,
281 pgoff_t first, pgoff_t last)
283 struct pagevec pv;
284 unsigned count, loop;
286 _enter("{%x:%u},%lx-%lx",
287 vnode->fid.vid, vnode->fid.vnode, first, last);
289 pagevec_init(&pv, 0);
291 do {
292 _debug("kill %lx-%lx", first, last);
294 count = last - first + 1;
295 if (count > PAGEVEC_SIZE)
296 count = PAGEVEC_SIZE;
297 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
298 first, count, pv.pages);
299 ASSERTCMP(pv.nr, ==, count);
301 for (loop = 0; loop < count; loop++) {
302 ClearPageUptodate(pv.pages[loop]);
303 if (error)
304 SetPageError(pv.pages[loop]);
305 end_page_writeback(pv.pages[loop]);
308 __pagevec_release(&pv);
309 } while (first < last);
311 _leave("");
315 * synchronously write back the locked page and any subsequent non-locked dirty
316 * pages also covered by the same writeback record
318 static int afs_write_back_from_locked_page(struct afs_writeback *wb,
319 struct page *primary_page)
321 struct page *pages[8], *page;
322 unsigned long count;
323 unsigned n, offset, to;
324 pgoff_t start, first, last;
325 int loop, ret;
327 _enter(",%lx", primary_page->index);
329 count = 1;
330 if (!clear_page_dirty_for_io(primary_page))
331 BUG();
332 if (test_set_page_writeback(primary_page))
333 BUG();
335 /* find all consecutive lockable dirty pages, stopping when we find a
336 * page that is not immediately lockable, is not dirty or is missing,
337 * or we reach the end of the range */
338 start = primary_page->index;
339 if (start >= wb->last)
340 goto no_more;
341 start++;
342 do {
343 _debug("more %lx [%lx]", start, count);
344 n = wb->last - start + 1;
345 if (n > ARRAY_SIZE(pages))
346 n = ARRAY_SIZE(pages);
347 n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping,
348 start, n, pages);
349 _debug("fgpc %u", n);
350 if (n == 0)
351 goto no_more;
352 if (pages[0]->index != start) {
353 do {
354 put_page(pages[--n]);
355 } while (n > 0);
356 goto no_more;
359 for (loop = 0; loop < n; loop++) {
360 page = pages[loop];
361 if (page->index > wb->last)
362 break;
363 if (!trylock_page(page))
364 break;
365 if (!PageDirty(page) ||
366 page_private(page) != (unsigned long) wb) {
367 unlock_page(page);
368 break;
370 if (!clear_page_dirty_for_io(page))
371 BUG();
372 if (test_set_page_writeback(page))
373 BUG();
374 unlock_page(page);
375 put_page(page);
377 count += loop;
378 if (loop < n) {
379 for (; loop < n; loop++)
380 put_page(pages[loop]);
381 goto no_more;
384 start += loop;
385 } while (start <= wb->last && count < 65536);
387 no_more:
388 /* we now have a contiguous set of dirty pages, each with writeback set
389 * and the dirty mark cleared; the first page is locked and must remain
390 * so, all the rest are unlocked */
391 first = primary_page->index;
392 last = first + count - 1;
394 offset = (first == wb->first) ? wb->offset_first : 0;
395 to = (last == wb->last) ? wb->to_last : PAGE_SIZE;
397 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
399 ret = afs_vnode_store_data(wb, first, last, offset, to);
400 if (ret < 0) {
401 switch (ret) {
402 case -EDQUOT:
403 case -ENOSPC:
404 set_bit(AS_ENOSPC,
405 &wb->vnode->vfs_inode.i_mapping->flags);
406 break;
407 case -EROFS:
408 case -EIO:
409 case -EREMOTEIO:
410 case -EFBIG:
411 case -ENOENT:
412 case -ENOMEDIUM:
413 case -ENXIO:
414 afs_kill_pages(wb->vnode, true, first, last);
415 set_bit(AS_EIO, &wb->vnode->vfs_inode.i_mapping->flags);
416 break;
417 case -EACCES:
418 case -EPERM:
419 case -ENOKEY:
420 case -EKEYEXPIRED:
421 case -EKEYREJECTED:
422 case -EKEYREVOKED:
423 afs_kill_pages(wb->vnode, false, first, last);
424 break;
425 default:
426 break;
428 } else {
429 ret = count;
432 _leave(" = %d", ret);
433 return ret;
437 * write a page back to the server
438 * - the caller locked the page for us
440 int afs_writepage(struct page *page, struct writeback_control *wbc)
442 struct afs_writeback *wb;
443 int ret;
445 _enter("{%lx},", page->index);
447 wb = (struct afs_writeback *) page_private(page);
448 ASSERT(wb != NULL);
450 ret = afs_write_back_from_locked_page(wb, page);
451 unlock_page(page);
452 if (ret < 0) {
453 _leave(" = %d", ret);
454 return 0;
457 wbc->nr_to_write -= ret;
459 _leave(" = 0");
460 return 0;
464 * write a region of pages back to the server
466 static int afs_writepages_region(struct address_space *mapping,
467 struct writeback_control *wbc,
468 pgoff_t index, pgoff_t end, pgoff_t *_next)
470 struct afs_writeback *wb;
471 struct page *page;
472 int ret, n;
474 _enter(",,%lx,%lx,", index, end);
476 do {
477 n = find_get_pages_tag(mapping, &index, PAGECACHE_TAG_DIRTY,
478 1, &page);
479 if (!n)
480 break;
482 _debug("wback %lx", page->index);
484 if (page->index > end) {
485 *_next = index;
486 page_cache_release(page);
487 _leave(" = 0 [%lx]", *_next);
488 return 0;
491 /* at this point we hold neither mapping->tree_lock nor lock on
492 * the page itself: the page may be truncated or invalidated
493 * (changing page->mapping to NULL), or even swizzled back from
494 * swapper_space to tmpfs file mapping
496 lock_page(page);
498 if (page->mapping != mapping) {
499 unlock_page(page);
500 page_cache_release(page);
501 continue;
504 if (wbc->sync_mode != WB_SYNC_NONE)
505 wait_on_page_writeback(page);
507 if (PageWriteback(page) || !PageDirty(page)) {
508 unlock_page(page);
509 continue;
512 wb = (struct afs_writeback *) page_private(page);
513 ASSERT(wb != NULL);
515 spin_lock(&wb->vnode->writeback_lock);
516 wb->state = AFS_WBACK_WRITING;
517 spin_unlock(&wb->vnode->writeback_lock);
519 ret = afs_write_back_from_locked_page(wb, page);
520 unlock_page(page);
521 page_cache_release(page);
522 if (ret < 0) {
523 _leave(" = %d", ret);
524 return ret;
527 wbc->nr_to_write -= ret;
529 cond_resched();
530 } while (index < end && wbc->nr_to_write > 0);
532 *_next = index;
533 _leave(" = 0 [%lx]", *_next);
534 return 0;
538 * write some of the pending data back to the server
540 int afs_writepages(struct address_space *mapping,
541 struct writeback_control *wbc)
543 pgoff_t start, end, next;
544 int ret;
546 _enter("");
548 if (wbc->range_cyclic) {
549 start = mapping->writeback_index;
550 end = -1;
551 ret = afs_writepages_region(mapping, wbc, start, end, &next);
552 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
553 ret = afs_writepages_region(mapping, wbc, 0, start,
554 &next);
555 mapping->writeback_index = next;
556 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
557 end = (pgoff_t)(LLONG_MAX >> PAGE_CACHE_SHIFT);
558 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
559 if (wbc->nr_to_write > 0)
560 mapping->writeback_index = next;
561 } else {
562 start = wbc->range_start >> PAGE_CACHE_SHIFT;
563 end = wbc->range_end >> PAGE_CACHE_SHIFT;
564 ret = afs_writepages_region(mapping, wbc, start, end, &next);
567 _leave(" = %d", ret);
568 return ret;
572 * completion of write to server
574 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
576 struct afs_writeback *wb = call->wb;
577 struct pagevec pv;
578 unsigned count, loop;
579 pgoff_t first = call->first, last = call->last;
580 bool free_wb;
582 _enter("{%x:%u},{%lx-%lx}",
583 vnode->fid.vid, vnode->fid.vnode, first, last);
585 ASSERT(wb != NULL);
587 pagevec_init(&pv, 0);
589 do {
590 _debug("done %lx-%lx", first, last);
592 count = last - first + 1;
593 if (count > PAGEVEC_SIZE)
594 count = PAGEVEC_SIZE;
595 pv.nr = find_get_pages_contig(call->mapping, first, count,
596 pv.pages);
597 ASSERTCMP(pv.nr, ==, count);
599 spin_lock(&vnode->writeback_lock);
600 for (loop = 0; loop < count; loop++) {
601 struct page *page = pv.pages[loop];
602 end_page_writeback(page);
603 if (page_private(page) == (unsigned long) wb) {
604 set_page_private(page, 0);
605 ClearPagePrivate(page);
606 wb->usage--;
609 free_wb = false;
610 if (wb->usage == 0) {
611 afs_unlink_writeback(wb);
612 free_wb = true;
614 spin_unlock(&vnode->writeback_lock);
615 first += count;
616 if (free_wb) {
617 afs_free_writeback(wb);
618 wb = NULL;
621 __pagevec_release(&pv);
622 } while (first <= last);
624 _leave("");
628 * write to an AFS file
630 ssize_t afs_file_write(struct kiocb *iocb, const struct iovec *iov,
631 unsigned long nr_segs, loff_t pos)
633 struct dentry *dentry = iocb->ki_filp->f_path.dentry;
634 struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
635 ssize_t result;
636 size_t count = iov_length(iov, nr_segs);
638 _enter("{%x.%u},{%zu},%lu,",
639 vnode->fid.vid, vnode->fid.vnode, count, nr_segs);
641 if (IS_SWAPFILE(&vnode->vfs_inode)) {
642 printk(KERN_INFO
643 "AFS: Attempt to write to active swap file!\n");
644 return -EBUSY;
647 if (!count)
648 return 0;
650 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
651 if (IS_ERR_VALUE(result)) {
652 _leave(" = %zd", result);
653 return result;
656 _leave(" = %zd", result);
657 return result;
661 * flush the vnode to the fileserver
663 int afs_writeback_all(struct afs_vnode *vnode)
665 struct address_space *mapping = vnode->vfs_inode.i_mapping;
666 struct writeback_control wbc = {
667 .sync_mode = WB_SYNC_ALL,
668 .nr_to_write = LONG_MAX,
669 .range_cyclic = 1,
671 int ret;
673 _enter("");
675 ret = mapping->a_ops->writepages(mapping, &wbc);
676 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
678 _leave(" = %d", ret);
679 return ret;
683 * flush any dirty pages for this process, and check for write errors.
684 * - the return status from this call provides a reliable indication of
685 * whether any write errors occurred for this process.
687 int afs_fsync(struct file *file, int datasync)
689 struct dentry *dentry = file->f_path.dentry;
690 struct afs_writeback *wb, *xwb;
691 struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode);
692 int ret;
694 _enter("{%x:%u},{n=%s},%d",
695 vnode->fid.vid, vnode->fid.vnode, dentry->d_name.name,
696 datasync);
698 /* use a writeback record as a marker in the queue - when this reaches
699 * the front of the queue, all the outstanding writes are either
700 * completed or rejected */
701 wb = kzalloc(sizeof(*wb), GFP_KERNEL);
702 if (!wb)
703 return -ENOMEM;
704 wb->vnode = vnode;
705 wb->first = 0;
706 wb->last = -1;
707 wb->offset_first = 0;
708 wb->to_last = PAGE_SIZE;
709 wb->usage = 1;
710 wb->state = AFS_WBACK_SYNCING;
711 init_waitqueue_head(&wb->waitq);
713 spin_lock(&vnode->writeback_lock);
714 list_for_each_entry(xwb, &vnode->writebacks, link) {
715 if (xwb->state == AFS_WBACK_PENDING)
716 xwb->state = AFS_WBACK_CONFLICTING;
718 list_add_tail(&wb->link, &vnode->writebacks);
719 spin_unlock(&vnode->writeback_lock);
721 /* push all the outstanding writebacks to the server */
722 ret = afs_writeback_all(vnode);
723 if (ret < 0) {
724 afs_put_writeback(wb);
725 _leave(" = %d [wb]", ret);
726 return ret;
729 /* wait for the preceding writes to actually complete */
730 ret = wait_event_interruptible(wb->waitq,
731 wb->state == AFS_WBACK_COMPLETE ||
732 vnode->writebacks.next == &wb->link);
733 afs_put_writeback(wb);
734 _leave(" = %d", ret);
735 return ret;
739 * notification that a previously read-only page is about to become writable
740 * - if it returns an error, the caller will deliver a bus error signal
742 int afs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
744 struct afs_vnode *vnode = AFS_FS_I(vma->vm_file->f_mapping->host);
746 _enter("{{%x:%u}},{%lx}",
747 vnode->fid.vid, vnode->fid.vnode, page->index);
749 /* wait for the page to be written to the cache before we allow it to
750 * be modified */
751 #ifdef CONFIG_AFS_FSCACHE
752 fscache_wait_on_page_write(vnode->cache, page);
753 #endif
755 _leave(" = 0");
756 return 0;