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(struct_size(req, array, 1), GFP_KERNEL);
51 refcount_set(&req->usage, 1);
55 req->pages = req->array;
59 ret = afs_fetch_data(vnode, key, req);
63 _debug("got NOENT from server"
64 " - marking file deleted and stale");
65 set_bit(AFS_VNODE_DELETED, &vnode->flags);
75 * prepare to perform part of a write to a page
77 int afs_write_begin(struct file *file, struct address_space *mapping,
78 loff_t pos, unsigned len, unsigned flags,
79 struct page **pagep, void **fsdata)
81 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
83 struct key *key = afs_file_key(file);
85 unsigned f, from = pos & (PAGE_SIZE - 1);
86 unsigned t, to = from + len;
87 pgoff_t index = pos >> PAGE_SHIFT;
90 _enter("{%llx:%llu},{%lx},%u,%u",
91 vnode->fid.vid, vnode->fid.vnode, index, from, to);
93 /* We want to store information about how much of a page is altered in
96 BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
98 page = grab_cache_page_write_begin(mapping, index, flags);
102 if (!PageUptodate(page) && len != PAGE_SIZE) {
103 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
107 _leave(" = %d [prep]", ret);
110 SetPageUptodate(page);
113 /* page won't leak in error case: it eventually gets cleaned off LRU */
117 /* See if this page is already partially written in a way that we can
118 * merge the new write with.
121 if (PagePrivate(page)) {
122 priv = page_private(page);
123 f = priv & AFS_PRIV_MAX;
124 t = priv >> AFS_PRIV_SHIFT;
129 if (PageWriteback(page)) {
130 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
132 goto flush_conflicting_write;
134 /* If the file is being filled locally, allow inter-write
135 * spaces to be merged into writes. If it's not, only write
136 * back what the user gives us.
138 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
139 (to < f || from > t))
140 goto flush_conflicting_write;
150 priv = (unsigned long)t << AFS_PRIV_SHIFT;
152 trace_afs_page_dirty(vnode, tracepoint_string("begin"),
154 SetPagePrivate(page);
155 set_page_private(page, priv);
159 /* The previous write and this write aren't adjacent or overlapping, so
160 * flush the page out.
162 flush_conflicting_write:
163 _debug("flush conflict");
164 ret = write_one_page(page);
166 _leave(" = %d", ret);
170 ret = lock_page_killable(page);
172 _leave(" = %d", ret);
179 * finalise part of a write to a page
181 int afs_write_end(struct file *file, struct address_space *mapping,
182 loff_t pos, unsigned len, unsigned copied,
183 struct page *page, void *fsdata)
185 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
186 struct key *key = afs_file_key(file);
187 loff_t i_size, maybe_i_size;
190 _enter("{%llx:%llu},{%lx}",
191 vnode->fid.vid, vnode->fid.vnode, page->index);
193 maybe_i_size = pos + copied;
195 i_size = i_size_read(&vnode->vfs_inode);
196 if (maybe_i_size > i_size) {
197 spin_lock(&vnode->wb_lock);
198 i_size = i_size_read(&vnode->vfs_inode);
199 if (maybe_i_size > i_size)
200 i_size_write(&vnode->vfs_inode, maybe_i_size);
201 spin_unlock(&vnode->wb_lock);
204 if (!PageUptodate(page)) {
206 /* Try and load any missing data from the server. The
207 * unmarshalling routine will take care of clearing any
208 * bits that are beyond the EOF.
210 ret = afs_fill_page(vnode, key, pos + copied,
215 SetPageUptodate(page);
218 set_page_dirty(page);
230 * kill all the pages in the given range
232 static void afs_kill_pages(struct address_space *mapping,
233 pgoff_t first, pgoff_t last)
235 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
237 unsigned count, loop;
239 _enter("{%llx:%llu},%lx-%lx",
240 vnode->fid.vid, vnode->fid.vnode, first, last);
245 _debug("kill %lx-%lx", first, last);
247 count = last - first + 1;
248 if (count > PAGEVEC_SIZE)
249 count = PAGEVEC_SIZE;
250 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
251 ASSERTCMP(pv.nr, ==, count);
253 for (loop = 0; loop < count; loop++) {
254 struct page *page = pv.pages[loop];
255 ClearPageUptodate(page);
257 end_page_writeback(page);
258 if (page->index >= first)
259 first = page->index + 1;
261 generic_error_remove_page(mapping, page);
265 __pagevec_release(&pv);
266 } while (first <= last);
272 * Redirty all the pages in a given range.
274 static void afs_redirty_pages(struct writeback_control *wbc,
275 struct address_space *mapping,
276 pgoff_t first, pgoff_t last)
278 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
280 unsigned count, loop;
282 _enter("{%llx:%llu},%lx-%lx",
283 vnode->fid.vid, vnode->fid.vnode, first, last);
288 _debug("redirty %lx-%lx", first, last);
290 count = last - first + 1;
291 if (count > PAGEVEC_SIZE)
292 count = PAGEVEC_SIZE;
293 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
294 ASSERTCMP(pv.nr, ==, count);
296 for (loop = 0; loop < count; loop++) {
297 struct page *page = pv.pages[loop];
299 redirty_page_for_writepage(wbc, page);
300 end_page_writeback(page);
301 if (page->index >= first)
302 first = page->index + 1;
305 __pagevec_release(&pv);
306 } while (first <= last);
312 * completion of write to server
314 static void afs_pages_written_back(struct afs_vnode *vnode,
315 pgoff_t first, pgoff_t last)
319 unsigned count, loop;
321 _enter("{%llx:%llu},{%lx-%lx}",
322 vnode->fid.vid, vnode->fid.vnode, first, last);
327 _debug("done %lx-%lx", first, last);
329 count = last - first + 1;
330 if (count > PAGEVEC_SIZE)
331 count = PAGEVEC_SIZE;
332 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
333 first, count, pv.pages);
334 ASSERTCMP(pv.nr, ==, count);
336 for (loop = 0; loop < count; loop++) {
337 priv = page_private(pv.pages[loop]);
338 trace_afs_page_dirty(vnode, tracepoint_string("clear"),
339 pv.pages[loop]->index, priv);
340 set_page_private(pv.pages[loop], 0);
341 end_page_writeback(pv.pages[loop]);
344 __pagevec_release(&pv);
345 } while (first <= last);
347 afs_prune_wb_keys(vnode);
352 * Find a key to use for the writeback. We cached the keys used to author the
353 * writes on the vnode. *_wbk will contain the last writeback key used or NULL
354 * and we need to start from there if it's set.
356 static int afs_get_writeback_key(struct afs_vnode *vnode,
357 struct afs_wb_key **_wbk)
359 struct afs_wb_key *wbk = NULL;
361 int ret = -ENOKEY, ret2;
363 spin_lock(&vnode->wb_lock);
365 p = (*_wbk)->vnode_link.next;
367 p = vnode->wb_keys.next;
369 while (p != &vnode->wb_keys) {
370 wbk = list_entry(p, struct afs_wb_key, vnode_link);
371 _debug("wbk %u", key_serial(wbk->key));
372 ret2 = key_validate(wbk->key);
374 refcount_inc(&wbk->usage);
375 _debug("USE WB KEY %u", key_serial(wbk->key));
385 spin_unlock(&vnode->wb_lock);
387 afs_put_wb_key(*_wbk);
392 static void afs_store_data_success(struct afs_operation *op)
394 struct afs_vnode *vnode = op->file[0].vnode;
396 afs_vnode_commit_status(op, &op->file[0]);
397 if (op->error == 0) {
398 afs_pages_written_back(vnode, op->store.first, op->store.last);
399 afs_stat_v(vnode, n_stores);
400 atomic_long_add((op->store.last * PAGE_SIZE + op->store.last_to) -
401 (op->store.first * PAGE_SIZE + op->store.first_offset),
402 &afs_v2net(vnode)->n_store_bytes);
406 static const struct afs_operation_ops afs_store_data_operation = {
407 .issue_afs_rpc = afs_fs_store_data,
408 .issue_yfs_rpc = yfs_fs_store_data,
409 .success = afs_store_data_success,
415 static int afs_store_data(struct address_space *mapping,
416 pgoff_t first, pgoff_t last,
417 unsigned offset, unsigned to)
419 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
420 struct afs_operation *op;
421 struct afs_wb_key *wbk = NULL;
424 _enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
429 first, last, offset, to);
431 ret = afs_get_writeback_key(vnode, &wbk);
433 _leave(" = %d [no keys]", ret);
437 op = afs_alloc_operation(wbk->key, vnode->volume);
443 afs_op_set_vnode(op, 0, vnode);
444 op->file[0].dv_delta = 1;
445 op->store.mapping = mapping;
446 op->store.first = first;
447 op->store.last = last;
448 op->store.first_offset = offset;
449 op->store.last_to = to;
450 op->ops = &afs_store_data_operation;
453 afs_begin_vnode_operation(op);
454 afs_wait_for_operation(op);
465 ret = afs_get_writeback_key(vnode, &wbk);
468 op->key = key_get(wbk->key);
475 _leave(" = %d", op->error);
476 return afs_put_operation(op);
480 * Synchronously write back the locked page and any subsequent non-locked dirty
483 static int afs_write_back_from_locked_page(struct address_space *mapping,
484 struct writeback_control *wbc,
485 struct page *primary_page,
488 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
489 struct page *pages[8], *page;
490 unsigned long count, priv;
491 unsigned n, offset, to, f, t;
492 pgoff_t start, first, last;
495 _enter(",%lx", primary_page->index);
498 if (test_set_page_writeback(primary_page))
501 /* Find all consecutive lockable dirty pages that have contiguous
502 * written regions, stopping when we find a page that is not
503 * immediately lockable, is not dirty or is missing, or we reach the
506 start = primary_page->index;
507 priv = page_private(primary_page);
508 offset = priv & AFS_PRIV_MAX;
509 to = priv >> AFS_PRIV_SHIFT;
510 trace_afs_page_dirty(vnode, tracepoint_string("store"),
511 primary_page->index, priv);
513 WARN_ON(offset == to);
515 trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
516 primary_page->index, priv);
518 if (start >= final_page ||
519 (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
524 _debug("more %lx [%lx]", start, count);
525 n = final_page - start + 1;
526 if (n > ARRAY_SIZE(pages))
527 n = ARRAY_SIZE(pages);
528 n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
529 _debug("fgpc %u", n);
532 if (pages[0]->index != start) {
534 put_page(pages[--n]);
539 for (loop = 0; loop < n; loop++) {
541 if (to != PAGE_SIZE &&
542 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
544 if (page->index > final_page)
546 if (!trylock_page(page))
548 if (!PageDirty(page) || PageWriteback(page)) {
553 priv = page_private(page);
554 f = priv & AFS_PRIV_MAX;
555 t = priv >> AFS_PRIV_SHIFT;
557 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
563 trace_afs_page_dirty(vnode, tracepoint_string("store+"),
566 if (!clear_page_dirty_for_io(page))
568 if (test_set_page_writeback(page))
575 for (; loop < n; loop++)
576 put_page(pages[loop]);
581 } while (start <= final_page && count < 65536);
584 /* We now have a contiguous set of dirty pages, each with writeback
585 * set; the first page is still locked at this point, but all the rest
586 * have been unlocked.
588 unlock_page(primary_page);
590 first = primary_page->index;
591 last = first + count - 1;
593 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
595 ret = afs_store_data(mapping, first, last, offset, to);
602 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
610 afs_redirty_pages(wbc, mapping, first, last);
611 mapping_set_error(mapping, ret);
616 afs_redirty_pages(wbc, mapping, first, last);
617 mapping_set_error(mapping, -ENOSPC);
627 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
628 afs_kill_pages(mapping, first, last);
629 mapping_set_error(mapping, ret);
633 _leave(" = %d", ret);
638 * write a page back to the server
639 * - the caller locked the page for us
641 int afs_writepage(struct page *page, struct writeback_control *wbc)
645 _enter("{%lx},", page->index);
647 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
648 wbc->range_end >> PAGE_SHIFT);
650 _leave(" = %d", ret);
654 wbc->nr_to_write -= ret;
661 * write a region of pages back to the server
663 static int afs_writepages_region(struct address_space *mapping,
664 struct writeback_control *wbc,
665 pgoff_t index, pgoff_t end, pgoff_t *_next)
670 _enter(",,%lx,%lx,", index, end);
673 n = find_get_pages_range_tag(mapping, &index, end,
674 PAGECACHE_TAG_DIRTY, 1, &page);
678 _debug("wback %lx", page->index);
681 * at this point we hold neither the i_pages lock nor the
682 * page lock: the page may be truncated or invalidated
683 * (changing page->mapping to NULL), or even swizzled
684 * back from swapper_space to tmpfs file mapping
686 ret = lock_page_killable(page);
689 _leave(" = %d", ret);
693 if (page->mapping != mapping || !PageDirty(page)) {
699 if (PageWriteback(page)) {
701 if (wbc->sync_mode != WB_SYNC_NONE)
702 wait_on_page_writeback(page);
707 if (!clear_page_dirty_for_io(page))
709 ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
712 _leave(" = %d", ret);
716 wbc->nr_to_write -= ret;
719 } while (index < end && wbc->nr_to_write > 0);
722 _leave(" = 0 [%lx]", *_next);
727 * write some of the pending data back to the server
729 int afs_writepages(struct address_space *mapping,
730 struct writeback_control *wbc)
732 pgoff_t start, end, next;
737 if (wbc->range_cyclic) {
738 start = mapping->writeback_index;
740 ret = afs_writepages_region(mapping, wbc, start, end, &next);
741 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
742 ret = afs_writepages_region(mapping, wbc, 0, start,
744 mapping->writeback_index = next;
745 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
746 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
747 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
748 if (wbc->nr_to_write > 0)
749 mapping->writeback_index = next;
751 start = wbc->range_start >> PAGE_SHIFT;
752 end = wbc->range_end >> PAGE_SHIFT;
753 ret = afs_writepages_region(mapping, wbc, start, end, &next);
756 _leave(" = %d", ret);
761 * write to an AFS file
763 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
765 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
767 size_t count = iov_iter_count(from);
769 _enter("{%llx:%llu},{%zu},",
770 vnode->fid.vid, vnode->fid.vnode, count);
772 if (IS_SWAPFILE(&vnode->vfs_inode)) {
774 "AFS: Attempt to write to active swap file!\n");
781 result = generic_file_write_iter(iocb, from);
783 _leave(" = %zd", result);
788 * flush any dirty pages for this process, and check for write errors.
789 * - the return status from this call provides a reliable indication of
790 * whether any write errors occurred for this process.
792 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
794 struct inode *inode = file_inode(file);
795 struct afs_vnode *vnode = AFS_FS_I(inode);
797 _enter("{%llx:%llu},{n=%pD},%d",
798 vnode->fid.vid, vnode->fid.vnode, file,
801 return file_write_and_wait_range(file, start, end);
805 * notification that a previously read-only page is about to become writable
806 * - if it returns an error, the caller will deliver a bus error signal
808 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
810 struct file *file = vmf->vma->vm_file;
811 struct inode *inode = file_inode(file);
812 struct afs_vnode *vnode = AFS_FS_I(inode);
815 _enter("{{%llx:%llu}},{%lx}",
816 vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
818 sb_start_pagefault(inode->i_sb);
820 /* Wait for the page to be written to the cache before we allow it to
821 * be modified. We then assume the entire page will need writing back.
823 #ifdef CONFIG_AFS_FSCACHE
824 fscache_wait_on_page_write(vnode->cache, vmf->page);
827 if (PageWriteback(vmf->page) &&
828 wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
829 return VM_FAULT_RETRY;
831 if (lock_page_killable(vmf->page) < 0)
832 return VM_FAULT_RETRY;
834 /* We mustn't change page->private until writeback is complete as that
835 * details the portion of the page we need to write back and we might
836 * need to redirty the page if there's a problem.
838 wait_on_page_writeback(vmf->page);
840 priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
841 priv |= 0; /* From */
842 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
843 vmf->page->index, priv);
844 SetPagePrivate(vmf->page);
845 set_page_private(vmf->page, priv);
847 sb_end_pagefault(inode->i_sb);
848 return VM_FAULT_LOCKED;
852 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
854 void afs_prune_wb_keys(struct afs_vnode *vnode)
856 LIST_HEAD(graveyard);
857 struct afs_wb_key *wbk, *tmp;
859 /* Discard unused keys */
860 spin_lock(&vnode->wb_lock);
862 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
863 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
864 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
865 if (refcount_read(&wbk->usage) == 1)
866 list_move(&wbk->vnode_link, &graveyard);
870 spin_unlock(&vnode->wb_lock);
872 while (!list_empty(&graveyard)) {
873 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
874 list_del(&wbk->vnode_link);
880 * Clean up a page during invalidation.
882 int afs_launder_page(struct page *page)
884 struct address_space *mapping = page->mapping;
885 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
890 _enter("{%lx}", page->index);
892 priv = page_private(page);
893 if (clear_page_dirty_for_io(page)) {
896 if (PagePrivate(page)) {
897 f = priv & AFS_PRIV_MAX;
898 t = priv >> AFS_PRIV_SHIFT;
901 trace_afs_page_dirty(vnode, tracepoint_string("launder"),
903 ret = afs_store_data(mapping, page->index, page->index, t, f);
906 trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
908 set_page_private(page, 0);
909 ClearPagePrivate(page);
911 #ifdef CONFIG_AFS_FSCACHE
912 if (PageFsCache(page)) {
913 fscache_wait_on_page_write(vnode->cache, page);
914 fscache_uncache_page(vnode->cache, page);