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.
12 #include <linux/backing-dev.h>
13 #include <linux/slab.h>
15 #include <linux/pagemap.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
21 * mark a page as having been made dirty and thus needing writeback
23 int afs_set_page_dirty(struct page *page)
26 return __set_page_dirty_nobuffers(page);
30 * partly or wholly fill a page that's under preparation for writing
32 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
33 loff_t pos, unsigned int len, struct page *page)
38 _enter(",,%llu", (unsigned long long)pos);
40 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
45 refcount_set(&req->usage, 1);
49 req->pages = req->array;
53 ret = afs_fetch_data(vnode, key, req);
57 _debug("got NOENT from server"
58 " - marking file deleted and stale");
59 set_bit(AFS_VNODE_DELETED, &vnode->flags);
69 * prepare to perform part of a write to a page
71 int afs_write_begin(struct file *file, struct address_space *mapping,
72 loff_t pos, unsigned len, unsigned flags,
73 struct page **pagep, void **fsdata)
75 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
77 struct key *key = afs_file_key(file);
79 unsigned f, from = pos & (PAGE_SIZE - 1);
80 unsigned t, to = from + len;
81 pgoff_t index = pos >> PAGE_SHIFT;
84 _enter("{%x:%u},{%lx},%u,%u",
85 vnode->fid.vid, vnode->fid.vnode, index, from, to);
87 /* We want to store information about how much of a page is altered in
90 BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
92 page = grab_cache_page_write_begin(mapping, index, flags);
96 if (!PageUptodate(page) && len != PAGE_SIZE) {
97 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
101 _leave(" = %d [prep]", ret);
104 SetPageUptodate(page);
107 /* page won't leak in error case: it eventually gets cleaned off LRU */
111 /* See if this page is already partially written in a way that we can
112 * merge the new write with.
115 if (PagePrivate(page)) {
116 priv = page_private(page);
117 f = priv & AFS_PRIV_MAX;
118 t = priv >> AFS_PRIV_SHIFT;
123 if (PageWriteback(page)) {
124 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
126 goto flush_conflicting_write;
128 /* If the file is being filled locally, allow inter-write
129 * spaces to be merged into writes. If it's not, only write
130 * back what the user gives us.
132 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
133 (to < f || from > t))
134 goto flush_conflicting_write;
144 priv = (unsigned long)t << AFS_PRIV_SHIFT;
146 trace_afs_page_dirty(vnode, tracepoint_string("begin"),
148 SetPagePrivate(page);
149 set_page_private(page, priv);
153 /* The previous write and this write aren't adjacent or overlapping, so
154 * flush the page out.
156 flush_conflicting_write:
157 _debug("flush conflict");
158 ret = write_one_page(page);
160 _leave(" = %d", ret);
164 ret = lock_page_killable(page);
166 _leave(" = %d", ret);
173 * finalise part of a write to a page
175 int afs_write_end(struct file *file, struct address_space *mapping,
176 loff_t pos, unsigned len, unsigned copied,
177 struct page *page, void *fsdata)
179 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
180 struct key *key = afs_file_key(file);
181 loff_t i_size, maybe_i_size;
184 _enter("{%x:%u},{%lx}",
185 vnode->fid.vid, vnode->fid.vnode, page->index);
187 maybe_i_size = pos + copied;
189 i_size = i_size_read(&vnode->vfs_inode);
190 if (maybe_i_size > i_size) {
191 spin_lock(&vnode->wb_lock);
192 i_size = i_size_read(&vnode->vfs_inode);
193 if (maybe_i_size > i_size)
194 i_size_write(&vnode->vfs_inode, maybe_i_size);
195 spin_unlock(&vnode->wb_lock);
198 if (!PageUptodate(page)) {
200 /* Try and load any missing data from the server. The
201 * unmarshalling routine will take care of clearing any
202 * bits that are beyond the EOF.
204 ret = afs_fill_page(vnode, key, pos + copied,
209 SetPageUptodate(page);
212 set_page_dirty(page);
224 * kill all the pages in the given range
226 static void afs_kill_pages(struct address_space *mapping,
227 pgoff_t first, pgoff_t last)
229 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
231 unsigned count, loop;
233 _enter("{%x:%u},%lx-%lx",
234 vnode->fid.vid, vnode->fid.vnode, first, last);
239 _debug("kill %lx-%lx", first, last);
241 count = last - first + 1;
242 if (count > PAGEVEC_SIZE)
243 count = PAGEVEC_SIZE;
244 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
245 ASSERTCMP(pv.nr, ==, count);
247 for (loop = 0; loop < count; loop++) {
248 struct page *page = pv.pages[loop];
249 ClearPageUptodate(page);
251 end_page_writeback(page);
252 if (page->index >= first)
253 first = page->index + 1;
255 generic_error_remove_page(mapping, page);
258 __pagevec_release(&pv);
259 } while (first <= last);
265 * Redirty all the pages in a given range.
267 static void afs_redirty_pages(struct writeback_control *wbc,
268 struct address_space *mapping,
269 pgoff_t first, pgoff_t last)
271 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
273 unsigned count, loop;
275 _enter("{%x:%u},%lx-%lx",
276 vnode->fid.vid, vnode->fid.vnode, first, last);
281 _debug("redirty %lx-%lx", first, last);
283 count = last - first + 1;
284 if (count > PAGEVEC_SIZE)
285 count = PAGEVEC_SIZE;
286 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
287 ASSERTCMP(pv.nr, ==, count);
289 for (loop = 0; loop < count; loop++) {
290 struct page *page = pv.pages[loop];
292 redirty_page_for_writepage(wbc, page);
293 end_page_writeback(page);
294 if (page->index >= first)
295 first = page->index + 1;
298 __pagevec_release(&pv);
299 } while (first <= last);
307 static int afs_store_data(struct address_space *mapping,
308 pgoff_t first, pgoff_t last,
309 unsigned offset, unsigned to)
311 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
312 struct afs_fs_cursor fc;
313 struct afs_wb_key *wbk = NULL;
315 int ret = -ENOKEY, ret2;
317 _enter("%s{%x:%u.%u},%lx,%lx,%x,%x",
322 first, last, offset, to);
324 spin_lock(&vnode->wb_lock);
325 p = vnode->wb_keys.next;
327 /* Iterate through the list looking for a valid key to use. */
329 while (p != &vnode->wb_keys) {
330 wbk = list_entry(p, struct afs_wb_key, vnode_link);
331 _debug("wbk %u", key_serial(wbk->key));
332 ret2 = key_validate(wbk->key);
340 spin_unlock(&vnode->wb_lock);
342 _leave(" = %d [no keys]", ret);
346 refcount_inc(&wbk->usage);
347 spin_unlock(&vnode->wb_lock);
349 _debug("USE WB KEY %u", key_serial(wbk->key));
352 if (afs_begin_vnode_operation(&fc, vnode, wbk->key)) {
353 while (afs_select_fileserver(&fc)) {
354 fc.cb_break = vnode->cb_break + vnode->cb_s_break;
355 afs_fs_store_data(&fc, mapping, first, last, offset, to);
358 afs_check_for_remote_deletion(&fc, fc.vnode);
359 afs_vnode_commit_status(&fc, vnode, fc.cb_break);
360 ret = afs_end_vnode_operation(&fc);
365 afs_stat_v(vnode, n_stores);
366 atomic_long_add((last * PAGE_SIZE + to) -
367 (first * PAGE_SIZE + offset),
368 &afs_v2net(vnode)->n_store_bytes);
377 spin_lock(&vnode->wb_lock);
378 p = wbk->vnode_link.next;
384 _leave(" = %d", ret);
389 * Synchronously write back the locked page and any subsequent non-locked dirty
392 static int afs_write_back_from_locked_page(struct address_space *mapping,
393 struct writeback_control *wbc,
394 struct page *primary_page,
397 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
398 struct page *pages[8], *page;
399 unsigned long count, priv;
400 unsigned n, offset, to, f, t;
401 pgoff_t start, first, last;
404 _enter(",%lx", primary_page->index);
407 if (test_set_page_writeback(primary_page))
410 /* Find all consecutive lockable dirty pages that have contiguous
411 * written regions, stopping when we find a page that is not
412 * immediately lockable, is not dirty or is missing, or we reach the
415 start = primary_page->index;
416 priv = page_private(primary_page);
417 offset = priv & AFS_PRIV_MAX;
418 to = priv >> AFS_PRIV_SHIFT;
419 trace_afs_page_dirty(vnode, tracepoint_string("store"),
420 primary_page->index, priv);
422 WARN_ON(offset == to);
424 trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
425 primary_page->index, priv);
427 if (start >= final_page ||
428 (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
433 _debug("more %lx [%lx]", start, count);
434 n = final_page - start + 1;
435 if (n > ARRAY_SIZE(pages))
436 n = ARRAY_SIZE(pages);
437 n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
438 _debug("fgpc %u", n);
441 if (pages[0]->index != start) {
443 put_page(pages[--n]);
448 for (loop = 0; loop < n; loop++) {
450 if (to != PAGE_SIZE &&
451 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
453 if (page->index > final_page)
455 if (!trylock_page(page))
457 if (!PageDirty(page) || PageWriteback(page)) {
462 priv = page_private(page);
463 f = priv & AFS_PRIV_MAX;
464 t = priv >> AFS_PRIV_SHIFT;
466 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
472 trace_afs_page_dirty(vnode, tracepoint_string("store+"),
475 if (!clear_page_dirty_for_io(page))
477 if (test_set_page_writeback(page))
484 for (; loop < n; loop++)
485 put_page(pages[loop]);
490 } while (start <= final_page && count < 65536);
493 /* We now have a contiguous set of dirty pages, each with writeback
494 * set; the first page is still locked at this point, but all the rest
495 * have been unlocked.
497 unlock_page(primary_page);
499 first = primary_page->index;
500 last = first + count - 1;
502 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
504 ret = afs_store_data(mapping, first, last, offset, to);
511 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
519 afs_redirty_pages(wbc, mapping, first, last);
520 mapping_set_error(mapping, ret);
525 afs_redirty_pages(wbc, mapping, first, last);
526 mapping_set_error(mapping, -ENOSPC);
536 afs_kill_pages(mapping, first, last);
537 mapping_set_error(mapping, ret);
541 _leave(" = %d", ret);
546 * write a page back to the server
547 * - the caller locked the page for us
549 int afs_writepage(struct page *page, struct writeback_control *wbc)
553 _enter("{%lx},", page->index);
555 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
556 wbc->range_end >> PAGE_SHIFT);
558 _leave(" = %d", ret);
562 wbc->nr_to_write -= ret;
569 * write a region of pages back to the server
571 static int afs_writepages_region(struct address_space *mapping,
572 struct writeback_control *wbc,
573 pgoff_t index, pgoff_t end, pgoff_t *_next)
578 _enter(",,%lx,%lx,", index, end);
581 n = find_get_pages_range_tag(mapping, &index, end,
582 PAGECACHE_TAG_DIRTY, 1, &page);
586 _debug("wback %lx", page->index);
589 * at this point we hold neither the i_pages lock nor the
590 * page lock: the page may be truncated or invalidated
591 * (changing page->mapping to NULL), or even swizzled
592 * back from swapper_space to tmpfs file mapping
594 ret = lock_page_killable(page);
597 _leave(" = %d", ret);
601 if (page->mapping != mapping || !PageDirty(page)) {
607 if (PageWriteback(page)) {
609 if (wbc->sync_mode != WB_SYNC_NONE)
610 wait_on_page_writeback(page);
615 if (!clear_page_dirty_for_io(page))
617 ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
620 _leave(" = %d", ret);
624 wbc->nr_to_write -= ret;
627 } while (index < end && wbc->nr_to_write > 0);
630 _leave(" = 0 [%lx]", *_next);
635 * write some of the pending data back to the server
637 int afs_writepages(struct address_space *mapping,
638 struct writeback_control *wbc)
640 pgoff_t start, end, next;
645 if (wbc->range_cyclic) {
646 start = mapping->writeback_index;
648 ret = afs_writepages_region(mapping, wbc, start, end, &next);
649 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
650 ret = afs_writepages_region(mapping, wbc, 0, start,
652 mapping->writeback_index = next;
653 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
654 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
655 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
656 if (wbc->nr_to_write > 0)
657 mapping->writeback_index = next;
659 start = wbc->range_start >> PAGE_SHIFT;
660 end = wbc->range_end >> PAGE_SHIFT;
661 ret = afs_writepages_region(mapping, wbc, start, end, &next);
664 _leave(" = %d", ret);
669 * completion of write to server
671 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
675 unsigned count, loop;
676 pgoff_t first = call->first, last = call->last;
678 _enter("{%x:%u},{%lx-%lx}",
679 vnode->fid.vid, vnode->fid.vnode, first, last);
684 _debug("done %lx-%lx", first, last);
686 count = last - first + 1;
687 if (count > PAGEVEC_SIZE)
688 count = PAGEVEC_SIZE;
689 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
690 first, count, pv.pages);
691 ASSERTCMP(pv.nr, ==, count);
693 for (loop = 0; loop < count; loop++) {
694 priv = page_private(pv.pages[loop]);
695 trace_afs_page_dirty(vnode, tracepoint_string("clear"),
696 pv.pages[loop]->index, priv);
697 set_page_private(pv.pages[loop], 0);
698 end_page_writeback(pv.pages[loop]);
701 __pagevec_release(&pv);
702 } while (first <= last);
704 afs_prune_wb_keys(vnode);
709 * write to an AFS file
711 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
713 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
715 size_t count = iov_iter_count(from);
717 _enter("{%x.%u},{%zu},",
718 vnode->fid.vid, vnode->fid.vnode, count);
720 if (IS_SWAPFILE(&vnode->vfs_inode)) {
722 "AFS: Attempt to write to active swap file!\n");
729 result = generic_file_write_iter(iocb, from);
731 _leave(" = %zd", result);
736 * flush any dirty pages for this process, and check for write errors.
737 * - the return status from this call provides a reliable indication of
738 * whether any write errors occurred for this process.
740 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
742 struct inode *inode = file_inode(file);
743 struct afs_vnode *vnode = AFS_FS_I(inode);
745 _enter("{%x:%u},{n=%pD},%d",
746 vnode->fid.vid, vnode->fid.vnode, file,
749 return file_write_and_wait_range(file, start, end);
753 * notification that a previously read-only page is about to become writable
754 * - if it returns an error, the caller will deliver a bus error signal
756 int afs_page_mkwrite(struct vm_fault *vmf)
758 struct file *file = vmf->vma->vm_file;
759 struct inode *inode = file_inode(file);
760 struct afs_vnode *vnode = AFS_FS_I(inode);
763 _enter("{{%x:%u}},{%lx}",
764 vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
766 sb_start_pagefault(inode->i_sb);
768 /* Wait for the page to be written to the cache before we allow it to
769 * be modified. We then assume the entire page will need writing back.
771 #ifdef CONFIG_AFS_FSCACHE
772 fscache_wait_on_page_write(vnode->cache, vmf->page);
775 if (PageWriteback(vmf->page) &&
776 wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
777 return VM_FAULT_RETRY;
779 if (lock_page_killable(vmf->page) < 0)
780 return VM_FAULT_RETRY;
782 /* We mustn't change page->private until writeback is complete as that
783 * details the portion of the page we need to write back and we might
784 * need to redirty the page if there's a problem.
786 wait_on_page_writeback(vmf->page);
788 priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
789 priv |= 0; /* From */
790 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
791 vmf->page->index, priv);
792 SetPagePrivate(vmf->page);
793 set_page_private(vmf->page, priv);
795 sb_end_pagefault(inode->i_sb);
796 return VM_FAULT_LOCKED;
800 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
802 void afs_prune_wb_keys(struct afs_vnode *vnode)
804 LIST_HEAD(graveyard);
805 struct afs_wb_key *wbk, *tmp;
807 /* Discard unused keys */
808 spin_lock(&vnode->wb_lock);
810 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
811 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
812 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
813 if (refcount_read(&wbk->usage) == 1)
814 list_move(&wbk->vnode_link, &graveyard);
818 spin_unlock(&vnode->wb_lock);
820 while (!list_empty(&graveyard)) {
821 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
822 list_del(&wbk->vnode_link);
828 * Clean up a page during invalidation.
830 int afs_launder_page(struct page *page)
832 struct address_space *mapping = page->mapping;
833 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
838 _enter("{%lx}", page->index);
840 priv = page_private(page);
841 if (clear_page_dirty_for_io(page)) {
844 if (PagePrivate(page)) {
845 f = priv & AFS_PRIV_MAX;
846 t = priv >> AFS_PRIV_SHIFT;
849 trace_afs_page_dirty(vnode, tracepoint_string("launder"),
851 ret = afs_store_data(mapping, page->index, page->index, t, f);
854 trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
856 set_page_private(page, 0);
857 ClearPagePrivate(page);
859 #ifdef CONFIG_AFS_FSCACHE
860 if (PageFsCache(page)) {
861 fscache_wait_on_page_write(vnode->cache, page);
862 fscache_uncache_page(vnode->cache, page);