1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* handling of writes to regular files and writing back to the server
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/backing-dev.h>
9 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
17 * mark a page as having been made dirty and thus needing writeback
19 int afs_set_page_dirty(struct page *page)
22 return __set_page_dirty_nobuffers(page);
26 * partly or wholly fill a page that's under preparation for writing
28 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
29 loff_t pos, unsigned int len, struct page *page)
36 _enter(",,%llu", (unsigned long long)pos);
38 if (pos >= vnode->vfs_inode.i_size) {
40 ASSERTCMP(p + len, <=, PAGE_SIZE);
42 memset(data + p, 0, len);
47 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
52 refcount_set(&req->usage, 1);
56 req->pages = req->array;
60 ret = afs_fetch_data(vnode, key, req);
64 _debug("got NOENT from server"
65 " - marking file deleted and stale");
66 set_bit(AFS_VNODE_DELETED, &vnode->flags);
76 * prepare to perform part of a write to a page
78 int afs_write_begin(struct file *file, struct address_space *mapping,
79 loff_t pos, unsigned len, unsigned flags,
80 struct page **pagep, void **fsdata)
82 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
84 struct key *key = afs_file_key(file);
86 unsigned f, from = pos & (PAGE_SIZE - 1);
87 unsigned t, to = from + len;
88 pgoff_t index = pos >> PAGE_SHIFT;
91 _enter("{%llx:%llu},{%lx},%u,%u",
92 vnode->fid.vid, vnode->fid.vnode, index, from, to);
94 /* We want to store information about how much of a page is altered in
97 BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
99 page = grab_cache_page_write_begin(mapping, index, flags);
103 if (!PageUptodate(page) && len != PAGE_SIZE) {
104 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
108 _leave(" = %d [prep]", ret);
111 SetPageUptodate(page);
114 /* page won't leak in error case: it eventually gets cleaned off LRU */
118 /* See if this page is already partially written in a way that we can
119 * merge the new write with.
122 if (PagePrivate(page)) {
123 priv = page_private(page);
124 f = priv & AFS_PRIV_MAX;
125 t = priv >> AFS_PRIV_SHIFT;
130 if (PageWriteback(page)) {
131 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
133 goto flush_conflicting_write;
135 /* If the file is being filled locally, allow inter-write
136 * spaces to be merged into writes. If it's not, only write
137 * back what the user gives us.
139 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
140 (to < f || from > t))
141 goto flush_conflicting_write;
151 priv = (unsigned long)t << AFS_PRIV_SHIFT;
153 trace_afs_page_dirty(vnode, tracepoint_string("begin"),
155 SetPagePrivate(page);
156 set_page_private(page, priv);
160 /* The previous write and this write aren't adjacent or overlapping, so
161 * flush the page out.
163 flush_conflicting_write:
164 _debug("flush conflict");
165 ret = write_one_page(page);
167 _leave(" = %d", ret);
171 ret = lock_page_killable(page);
173 _leave(" = %d", ret);
180 * finalise part of a write to a page
182 int afs_write_end(struct file *file, struct address_space *mapping,
183 loff_t pos, unsigned len, unsigned copied,
184 struct page *page, void *fsdata)
186 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
187 struct key *key = afs_file_key(file);
188 loff_t i_size, maybe_i_size;
191 _enter("{%llx:%llu},{%lx}",
192 vnode->fid.vid, vnode->fid.vnode, page->index);
194 maybe_i_size = pos + copied;
196 i_size = i_size_read(&vnode->vfs_inode);
197 if (maybe_i_size > i_size) {
198 spin_lock(&vnode->wb_lock);
199 i_size = i_size_read(&vnode->vfs_inode);
200 if (maybe_i_size > i_size)
201 i_size_write(&vnode->vfs_inode, maybe_i_size);
202 spin_unlock(&vnode->wb_lock);
205 if (!PageUptodate(page)) {
207 /* Try and load any missing data from the server. The
208 * unmarshalling routine will take care of clearing any
209 * bits that are beyond the EOF.
211 ret = afs_fill_page(vnode, key, pos + copied,
216 SetPageUptodate(page);
219 set_page_dirty(page);
231 * kill all the pages in the given range
233 static void afs_kill_pages(struct address_space *mapping,
234 pgoff_t first, pgoff_t last)
236 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
238 unsigned count, loop;
240 _enter("{%llx:%llu},%lx-%lx",
241 vnode->fid.vid, vnode->fid.vnode, first, last);
246 _debug("kill %lx-%lx", first, last);
248 count = last - first + 1;
249 if (count > PAGEVEC_SIZE)
250 count = PAGEVEC_SIZE;
251 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
252 ASSERTCMP(pv.nr, ==, count);
254 for (loop = 0; loop < count; loop++) {
255 struct page *page = pv.pages[loop];
256 ClearPageUptodate(page);
258 end_page_writeback(page);
259 if (page->index >= first)
260 first = page->index + 1;
262 generic_error_remove_page(mapping, page);
266 __pagevec_release(&pv);
267 } while (first <= last);
273 * Redirty all the pages in a given range.
275 static void afs_redirty_pages(struct writeback_control *wbc,
276 struct address_space *mapping,
277 pgoff_t first, pgoff_t last)
279 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
281 unsigned count, loop;
283 _enter("{%llx:%llu},%lx-%lx",
284 vnode->fid.vid, vnode->fid.vnode, first, last);
289 _debug("redirty %lx-%lx", first, last);
291 count = last - first + 1;
292 if (count > PAGEVEC_SIZE)
293 count = PAGEVEC_SIZE;
294 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
295 ASSERTCMP(pv.nr, ==, count);
297 for (loop = 0; loop < count; loop++) {
298 struct page *page = pv.pages[loop];
300 redirty_page_for_writepage(wbc, page);
301 end_page_writeback(page);
302 if (page->index >= first)
303 first = page->index + 1;
306 __pagevec_release(&pv);
307 } while (first <= last);
313 * completion of write to server
315 static void afs_pages_written_back(struct afs_vnode *vnode,
316 pgoff_t first, pgoff_t last)
320 unsigned count, loop;
322 _enter("{%llx:%llu},{%lx-%lx}",
323 vnode->fid.vid, vnode->fid.vnode, first, last);
328 _debug("done %lx-%lx", first, last);
330 count = last - first + 1;
331 if (count > PAGEVEC_SIZE)
332 count = PAGEVEC_SIZE;
333 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
334 first, count, pv.pages);
335 ASSERTCMP(pv.nr, ==, count);
337 for (loop = 0; loop < count; loop++) {
338 priv = page_private(pv.pages[loop]);
339 trace_afs_page_dirty(vnode, tracepoint_string("clear"),
340 pv.pages[loop]->index, priv);
341 set_page_private(pv.pages[loop], 0);
342 end_page_writeback(pv.pages[loop]);
345 __pagevec_release(&pv);
346 } while (first <= last);
348 afs_prune_wb_keys(vnode);
355 static int afs_store_data(struct address_space *mapping,
356 pgoff_t first, pgoff_t last,
357 unsigned offset, unsigned to)
359 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
360 struct afs_fs_cursor fc;
361 struct afs_status_cb *scb;
362 struct afs_wb_key *wbk = NULL;
364 int ret = -ENOKEY, ret2;
366 _enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
371 first, last, offset, to);
373 scb = kzalloc(sizeof(struct afs_status_cb), GFP_NOFS);
377 spin_lock(&vnode->wb_lock);
378 p = vnode->wb_keys.next;
380 /* Iterate through the list looking for a valid key to use. */
382 while (p != &vnode->wb_keys) {
383 wbk = list_entry(p, struct afs_wb_key, vnode_link);
384 _debug("wbk %u", key_serial(wbk->key));
385 ret2 = key_validate(wbk->key);
393 spin_unlock(&vnode->wb_lock);
396 _leave(" = %d [no keys]", ret);
400 refcount_inc(&wbk->usage);
401 spin_unlock(&vnode->wb_lock);
403 _debug("USE WB KEY %u", key_serial(wbk->key));
406 if (afs_begin_vnode_operation(&fc, vnode, wbk->key, false)) {
407 afs_dataversion_t data_version = vnode->status.data_version + 1;
409 while (afs_select_fileserver(&fc)) {
410 fc.cb_break = afs_calc_vnode_cb_break(vnode);
411 afs_fs_store_data(&fc, mapping, first, last, offset, to, scb);
414 afs_check_for_remote_deletion(&fc, vnode);
415 afs_vnode_commit_status(&fc, vnode, fc.cb_break,
417 if (fc.ac.error == 0)
418 afs_pages_written_back(vnode, first, last);
419 ret = afs_end_vnode_operation(&fc);
424 afs_stat_v(vnode, n_stores);
425 atomic_long_add((last * PAGE_SIZE + to) -
426 (first * PAGE_SIZE + offset),
427 &afs_v2net(vnode)->n_store_bytes);
436 spin_lock(&vnode->wb_lock);
437 p = wbk->vnode_link.next;
444 _leave(" = %d", ret);
449 * Synchronously write back the locked page and any subsequent non-locked dirty
452 static int afs_write_back_from_locked_page(struct address_space *mapping,
453 struct writeback_control *wbc,
454 struct page *primary_page,
457 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
458 struct page *pages[8], *page;
459 unsigned long count, priv;
460 unsigned n, offset, to, f, t;
461 pgoff_t start, first, last;
464 _enter(",%lx", primary_page->index);
467 if (test_set_page_writeback(primary_page))
470 /* Find all consecutive lockable dirty pages that have contiguous
471 * written regions, stopping when we find a page that is not
472 * immediately lockable, is not dirty or is missing, or we reach the
475 start = primary_page->index;
476 priv = page_private(primary_page);
477 offset = priv & AFS_PRIV_MAX;
478 to = priv >> AFS_PRIV_SHIFT;
479 trace_afs_page_dirty(vnode, tracepoint_string("store"),
480 primary_page->index, priv);
482 WARN_ON(offset == to);
484 trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
485 primary_page->index, priv);
487 if (start >= final_page ||
488 (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
493 _debug("more %lx [%lx]", start, count);
494 n = final_page - start + 1;
495 if (n > ARRAY_SIZE(pages))
496 n = ARRAY_SIZE(pages);
497 n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
498 _debug("fgpc %u", n);
501 if (pages[0]->index != start) {
503 put_page(pages[--n]);
508 for (loop = 0; loop < n; loop++) {
510 if (to != PAGE_SIZE &&
511 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
513 if (page->index > final_page)
515 if (!trylock_page(page))
517 if (!PageDirty(page) || PageWriteback(page)) {
522 priv = page_private(page);
523 f = priv & AFS_PRIV_MAX;
524 t = priv >> AFS_PRIV_SHIFT;
526 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
532 trace_afs_page_dirty(vnode, tracepoint_string("store+"),
535 if (!clear_page_dirty_for_io(page))
537 if (test_set_page_writeback(page))
544 for (; loop < n; loop++)
545 put_page(pages[loop]);
550 } while (start <= final_page && count < 65536);
553 /* We now have a contiguous set of dirty pages, each with writeback
554 * set; the first page is still locked at this point, but all the rest
555 * have been unlocked.
557 unlock_page(primary_page);
559 first = primary_page->index;
560 last = first + count - 1;
562 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
564 ret = afs_store_data(mapping, first, last, offset, to);
571 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
579 afs_redirty_pages(wbc, mapping, first, last);
580 mapping_set_error(mapping, ret);
585 afs_redirty_pages(wbc, mapping, first, last);
586 mapping_set_error(mapping, -ENOSPC);
596 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
597 afs_kill_pages(mapping, first, last);
598 mapping_set_error(mapping, ret);
602 _leave(" = %d", ret);
607 * write a page back to the server
608 * - the caller locked the page for us
610 int afs_writepage(struct page *page, struct writeback_control *wbc)
614 _enter("{%lx},", page->index);
616 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
617 wbc->range_end >> PAGE_SHIFT);
619 _leave(" = %d", ret);
623 wbc->nr_to_write -= ret;
630 * write a region of pages back to the server
632 static int afs_writepages_region(struct address_space *mapping,
633 struct writeback_control *wbc,
634 pgoff_t index, pgoff_t end, pgoff_t *_next)
639 _enter(",,%lx,%lx,", index, end);
642 n = find_get_pages_range_tag(mapping, &index, end,
643 PAGECACHE_TAG_DIRTY, 1, &page);
647 _debug("wback %lx", page->index);
650 * at this point we hold neither the i_pages lock nor the
651 * page lock: the page may be truncated or invalidated
652 * (changing page->mapping to NULL), or even swizzled
653 * back from swapper_space to tmpfs file mapping
655 ret = lock_page_killable(page);
658 _leave(" = %d", ret);
662 if (page->mapping != mapping || !PageDirty(page)) {
668 if (PageWriteback(page)) {
670 if (wbc->sync_mode != WB_SYNC_NONE)
671 wait_on_page_writeback(page);
676 if (!clear_page_dirty_for_io(page))
678 ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
681 _leave(" = %d", ret);
685 wbc->nr_to_write -= ret;
688 } while (index < end && wbc->nr_to_write > 0);
691 _leave(" = 0 [%lx]", *_next);
696 * write some of the pending data back to the server
698 int afs_writepages(struct address_space *mapping,
699 struct writeback_control *wbc)
701 pgoff_t start, end, next;
706 if (wbc->range_cyclic) {
707 start = mapping->writeback_index;
709 ret = afs_writepages_region(mapping, wbc, start, end, &next);
710 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
711 ret = afs_writepages_region(mapping, wbc, 0, start,
713 mapping->writeback_index = next;
714 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
715 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
716 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
717 if (wbc->nr_to_write > 0)
718 mapping->writeback_index = next;
720 start = wbc->range_start >> PAGE_SHIFT;
721 end = wbc->range_end >> PAGE_SHIFT;
722 ret = afs_writepages_region(mapping, wbc, start, end, &next);
725 _leave(" = %d", ret);
730 * write to an AFS file
732 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
734 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
736 size_t count = iov_iter_count(from);
738 _enter("{%llx:%llu},{%zu},",
739 vnode->fid.vid, vnode->fid.vnode, count);
741 if (IS_SWAPFILE(&vnode->vfs_inode)) {
743 "AFS: Attempt to write to active swap file!\n");
750 result = generic_file_write_iter(iocb, from);
752 _leave(" = %zd", result);
757 * flush any dirty pages for this process, and check for write errors.
758 * - the return status from this call provides a reliable indication of
759 * whether any write errors occurred for this process.
761 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
763 struct inode *inode = file_inode(file);
764 struct afs_vnode *vnode = AFS_FS_I(inode);
766 _enter("{%llx:%llu},{n=%pD},%d",
767 vnode->fid.vid, vnode->fid.vnode, file,
770 return file_write_and_wait_range(file, start, end);
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 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
779 struct file *file = vmf->vma->vm_file;
780 struct inode *inode = file_inode(file);
781 struct afs_vnode *vnode = AFS_FS_I(inode);
784 _enter("{{%llx:%llu}},{%lx}",
785 vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
787 sb_start_pagefault(inode->i_sb);
789 /* Wait for the page to be written to the cache before we allow it to
790 * be modified. We then assume the entire page will need writing back.
792 #ifdef CONFIG_AFS_FSCACHE
793 fscache_wait_on_page_write(vnode->cache, vmf->page);
796 if (PageWriteback(vmf->page) &&
797 wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
798 return VM_FAULT_RETRY;
800 if (lock_page_killable(vmf->page) < 0)
801 return VM_FAULT_RETRY;
803 /* We mustn't change page->private until writeback is complete as that
804 * details the portion of the page we need to write back and we might
805 * need to redirty the page if there's a problem.
807 wait_on_page_writeback(vmf->page);
809 priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
810 priv |= 0; /* From */
811 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
812 vmf->page->index, priv);
813 SetPagePrivate(vmf->page);
814 set_page_private(vmf->page, priv);
816 sb_end_pagefault(inode->i_sb);
817 return VM_FAULT_LOCKED;
821 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
823 void afs_prune_wb_keys(struct afs_vnode *vnode)
825 LIST_HEAD(graveyard);
826 struct afs_wb_key *wbk, *tmp;
828 /* Discard unused keys */
829 spin_lock(&vnode->wb_lock);
831 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
832 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
833 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
834 if (refcount_read(&wbk->usage) == 1)
835 list_move(&wbk->vnode_link, &graveyard);
839 spin_unlock(&vnode->wb_lock);
841 while (!list_empty(&graveyard)) {
842 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
843 list_del(&wbk->vnode_link);
849 * Clean up a page during invalidation.
851 int afs_launder_page(struct page *page)
853 struct address_space *mapping = page->mapping;
854 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
859 _enter("{%lx}", page->index);
861 priv = page_private(page);
862 if (clear_page_dirty_for_io(page)) {
865 if (PagePrivate(page)) {
866 f = priv & AFS_PRIV_MAX;
867 t = priv >> AFS_PRIV_SHIFT;
870 trace_afs_page_dirty(vnode, tracepoint_string("launder"),
872 ret = afs_store_data(mapping, page->index, page->index, t, f);
875 trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
877 set_page_private(page, 0);
878 ClearPagePrivate(page);
880 #ifdef CONFIG_AFS_FSCACHE
881 if (PageFsCache(page)) {
882 fscache_wait_on_page_write(vnode->cache, page);
883 fscache_uncache_page(vnode->cache, page);