1 // SPDX-License-Identifier: GPL-2.0
5 #include "alloc_foreground.h"
6 #include "bkey_on_stack.h"
7 #include "btree_update.h"
12 #include "extent_update.h"
23 #include <linux/aio.h>
24 #include <linux/backing-dev.h>
25 #include <linux/falloc.h>
26 #include <linux/migrate.h>
27 #include <linux/mmu_context.h>
28 #include <linux/pagevec.h>
29 #include <linux/rmap.h>
30 #include <linux/sched/signal.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/uio.h>
33 #include <linux/writeback.h>
35 #include <trace/events/bcachefs.h>
36 #include <trace/events/writeback.h>
42 struct bch_writepage_io {
44 struct bch_inode_info *inode;
47 struct bch_write_op op;
51 struct completion done;
57 struct quota_res quota_res;
61 struct iovec inline_vecs[2];
64 struct bch_write_op op;
71 struct bch_read_bio rbio;
74 /* pagecache_block must be held */
75 static int write_invalidate_inode_pages_range(struct address_space *mapping,
76 loff_t start, loff_t end)
81 * XXX: the way this is currently implemented, we can spin if a process
82 * is continually redirtying a specific page
85 if (!mapping->nrpages &&
86 !mapping->nrexceptional)
89 ret = filemap_write_and_wait_range(mapping, start, end);
93 if (!mapping->nrpages)
96 ret = invalidate_inode_pages2_range(mapping,
99 } while (ret == -EBUSY);
106 #ifdef CONFIG_BCACHEFS_QUOTA
108 static void bch2_quota_reservation_put(struct bch_fs *c,
109 struct bch_inode_info *inode,
110 struct quota_res *res)
115 mutex_lock(&inode->ei_quota_lock);
116 BUG_ON(res->sectors > inode->ei_quota_reserved);
118 bch2_quota_acct(c, inode->ei_qid, Q_SPC,
119 -((s64) res->sectors), KEY_TYPE_QUOTA_PREALLOC);
120 inode->ei_quota_reserved -= res->sectors;
121 mutex_unlock(&inode->ei_quota_lock);
126 static int bch2_quota_reservation_add(struct bch_fs *c,
127 struct bch_inode_info *inode,
128 struct quota_res *res,
134 mutex_lock(&inode->ei_quota_lock);
135 ret = bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors,
136 check_enospc ? KEY_TYPE_QUOTA_PREALLOC : KEY_TYPE_QUOTA_NOCHECK);
138 inode->ei_quota_reserved += sectors;
139 res->sectors += sectors;
141 mutex_unlock(&inode->ei_quota_lock);
148 static void bch2_quota_reservation_put(struct bch_fs *c,
149 struct bch_inode_info *inode,
150 struct quota_res *res)
154 static int bch2_quota_reservation_add(struct bch_fs *c,
155 struct bch_inode_info *inode,
156 struct quota_res *res,
165 /* i_size updates: */
167 struct inode_new_size {
173 static int inode_set_size(struct bch_inode_info *inode,
174 struct bch_inode_unpacked *bi,
177 struct inode_new_size *s = p;
179 bi->bi_size = s->new_size;
180 if (s->fields & ATTR_ATIME)
181 bi->bi_atime = s->now;
182 if (s->fields & ATTR_MTIME)
183 bi->bi_mtime = s->now;
184 if (s->fields & ATTR_CTIME)
185 bi->bi_ctime = s->now;
190 int __must_check bch2_write_inode_size(struct bch_fs *c,
191 struct bch_inode_info *inode,
192 loff_t new_size, unsigned fields)
194 struct inode_new_size s = {
195 .new_size = new_size,
196 .now = bch2_current_time(c),
200 return bch2_write_inode(c, inode, inode_set_size, &s, fields);
203 static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
204 struct quota_res *quota_res, s64 sectors)
209 mutex_lock(&inode->ei_quota_lock);
210 #ifdef CONFIG_BCACHEFS_QUOTA
211 if (quota_res && sectors > 0) {
212 BUG_ON(sectors > quota_res->sectors);
213 BUG_ON(sectors > inode->ei_quota_reserved);
215 quota_res->sectors -= sectors;
216 inode->ei_quota_reserved -= sectors;
218 bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
221 inode->v.i_blocks += sectors;
222 mutex_unlock(&inode->ei_quota_lock);
227 /* stored in page->private: */
229 struct bch_page_sector {
230 /* Uncompressed, fully allocated replicas: */
231 unsigned nr_replicas:3;
233 /* Owns PAGE_SECTORS * replicas_reserved sized reservation: */
234 unsigned replicas_reserved:3;
245 struct bch_page_state {
247 atomic_t write_count;
248 struct bch_page_sector s[PAGE_SECTORS];
251 static inline struct bch_page_state *__bch2_page_state(struct page *page)
253 return page_has_private(page)
254 ? (struct bch_page_state *) page_private(page)
258 static inline struct bch_page_state *bch2_page_state(struct page *page)
260 EBUG_ON(!PageLocked(page));
262 return __bch2_page_state(page);
265 /* for newly allocated pages: */
266 static void __bch2_page_state_release(struct page *page)
268 struct bch_page_state *s = __bch2_page_state(page);
273 ClearPagePrivate(page);
274 set_page_private(page, 0);
279 static void bch2_page_state_release(struct page *page)
281 struct bch_page_state *s = bch2_page_state(page);
286 ClearPagePrivate(page);
287 set_page_private(page, 0);
292 /* for newly allocated pages: */
293 static struct bch_page_state *__bch2_page_state_create(struct page *page,
296 struct bch_page_state *s;
298 s = kzalloc(sizeof(*s), GFP_NOFS|gfp);
302 spin_lock_init(&s->lock);
304 * migrate_page_move_mapping() assumes that pages with private data
305 * have their count elevated by 1.
308 set_page_private(page, (unsigned long) s);
309 SetPagePrivate(page);
313 static struct bch_page_state *bch2_page_state_create(struct page *page,
316 return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp);
319 static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode)
321 /* XXX: this should not be open coded */
322 return inode->ei_inode.bi_data_replicas
323 ? inode->ei_inode.bi_data_replicas - 1
324 : c->opts.data_replicas;
327 static inline unsigned sectors_to_reserve(struct bch_page_sector *s,
328 unsigned nr_replicas)
330 return max(0, (int) nr_replicas -
332 s->replicas_reserved);
335 static int bch2_get_page_disk_reservation(struct bch_fs *c,
336 struct bch_inode_info *inode,
337 struct page *page, bool check_enospc)
339 struct bch_page_state *s = bch2_page_state_create(page, 0);
340 unsigned nr_replicas = inode_nr_replicas(c, inode);
341 struct disk_reservation disk_res = { 0 };
342 unsigned i, disk_res_sectors = 0;
348 for (i = 0; i < ARRAY_SIZE(s->s); i++)
349 disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas);
351 if (!disk_res_sectors)
354 ret = bch2_disk_reservation_get(c, &disk_res,
357 ? BCH_DISK_RESERVATION_NOFAIL
362 for (i = 0; i < ARRAY_SIZE(s->s); i++)
363 s->s[i].replicas_reserved +=
364 sectors_to_reserve(&s->s[i], nr_replicas);
369 struct bch2_page_reservation {
370 struct disk_reservation disk;
371 struct quota_res quota;
374 static void bch2_page_reservation_init(struct bch_fs *c,
375 struct bch_inode_info *inode,
376 struct bch2_page_reservation *res)
378 memset(res, 0, sizeof(*res));
380 res->disk.nr_replicas = inode_nr_replicas(c, inode);
383 static void bch2_page_reservation_put(struct bch_fs *c,
384 struct bch_inode_info *inode,
385 struct bch2_page_reservation *res)
387 bch2_disk_reservation_put(c, &res->disk);
388 bch2_quota_reservation_put(c, inode, &res->quota);
391 static int bch2_page_reservation_get(struct bch_fs *c,
392 struct bch_inode_info *inode, struct page *page,
393 struct bch2_page_reservation *res,
394 unsigned offset, unsigned len, bool check_enospc)
396 struct bch_page_state *s = bch2_page_state_create(page, 0);
397 unsigned i, disk_sectors = 0, quota_sectors = 0;
403 for (i = round_down(offset, block_bytes(c)) >> 9;
404 i < round_up(offset + len, block_bytes(c)) >> 9;
406 disk_sectors += sectors_to_reserve(&s->s[i],
407 res->disk.nr_replicas);
408 quota_sectors += s->s[i].state == SECTOR_UNALLOCATED;
412 ret = bch2_disk_reservation_add(c, &res->disk,
415 ? BCH_DISK_RESERVATION_NOFAIL
422 ret = bch2_quota_reservation_add(c, inode, &res->quota,
426 struct disk_reservation tmp = {
427 .sectors = disk_sectors
430 bch2_disk_reservation_put(c, &tmp);
431 res->disk.sectors -= disk_sectors;
439 static void bch2_clear_page_bits(struct page *page)
441 struct bch_inode_info *inode = to_bch_ei(page->mapping->host);
442 struct bch_fs *c = inode->v.i_sb->s_fs_info;
443 struct bch_page_state *s = bch2_page_state(page);
444 struct disk_reservation disk_res = { 0 };
445 int i, dirty_sectors = 0;
450 EBUG_ON(!PageLocked(page));
451 EBUG_ON(PageWriteback(page));
453 for (i = 0; i < ARRAY_SIZE(s->s); i++) {
454 disk_res.sectors += s->s[i].replicas_reserved;
455 s->s[i].replicas_reserved = 0;
457 if (s->s[i].state == SECTOR_DIRTY) {
459 s->s[i].state = SECTOR_UNALLOCATED;
463 bch2_disk_reservation_put(c, &disk_res);
466 i_sectors_acct(c, inode, NULL, -dirty_sectors);
468 bch2_page_state_release(page);
471 static void bch2_set_page_dirty(struct bch_fs *c,
472 struct bch_inode_info *inode, struct page *page,
473 struct bch2_page_reservation *res,
474 unsigned offset, unsigned len)
476 struct bch_page_state *s = bch2_page_state(page);
477 unsigned i, dirty_sectors = 0;
479 WARN_ON((u64) page_offset(page) + offset + len >
480 round_up((u64) i_size_read(&inode->v), block_bytes(c)));
484 for (i = round_down(offset, block_bytes(c)) >> 9;
485 i < round_up(offset + len, block_bytes(c)) >> 9;
487 unsigned sectors = sectors_to_reserve(&s->s[i],
488 res->disk.nr_replicas);
491 * This can happen if we race with the error path in
492 * bch2_writepage_io_done():
494 sectors = min_t(unsigned, sectors, res->disk.sectors);
496 s->s[i].replicas_reserved += sectors;
497 res->disk.sectors -= sectors;
499 if (s->s[i].state == SECTOR_UNALLOCATED)
502 s->s[i].state = max_t(unsigned, s->s[i].state, SECTOR_DIRTY);
505 spin_unlock(&s->lock);
508 i_sectors_acct(c, inode, &res->quota, dirty_sectors);
510 if (!PageDirty(page))
511 __set_page_dirty_nobuffers(page);
514 vm_fault_t bch2_page_fault(struct vm_fault *vmf)
516 struct file *file = vmf->vma->vm_file;
517 struct bch_inode_info *inode = file_bch_inode(file);
520 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
521 ret = filemap_fault(vmf);
522 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
527 vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf)
529 struct page *page = vmf->page;
530 struct file *file = vmf->vma->vm_file;
531 struct bch_inode_info *inode = file_bch_inode(file);
532 struct address_space *mapping = file->f_mapping;
533 struct bch_fs *c = inode->v.i_sb->s_fs_info;
534 struct bch2_page_reservation res;
537 int ret = VM_FAULT_LOCKED;
539 bch2_page_reservation_init(c, inode, &res);
541 sb_start_pagefault(inode->v.i_sb);
542 file_update_time(file);
545 * Not strictly necessary, but helps avoid dio writes livelocking in
546 * write_invalidate_inode_pages_range() - can drop this if/when we get
547 * a write_invalidate_inode_pages_range() that works without dropping
548 * page lock before invalidating page
550 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
553 isize = i_size_read(&inode->v);
555 if (page->mapping != mapping || page_offset(page) >= isize) {
557 ret = VM_FAULT_NOPAGE;
561 len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page));
563 if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) {
565 ret = VM_FAULT_SIGBUS;
569 bch2_set_page_dirty(c, inode, page, &res, 0, len);
570 bch2_page_reservation_put(c, inode, &res);
572 wait_for_stable_page(page);
574 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
575 sb_end_pagefault(inode->v.i_sb);
580 void bch2_invalidatepage(struct page *page, unsigned int offset,
583 if (offset || length < PAGE_SIZE)
586 bch2_clear_page_bits(page);
589 int bch2_releasepage(struct page *page, gfp_t gfp_mask)
594 bch2_clear_page_bits(page);
598 #ifdef CONFIG_MIGRATION
599 int bch2_migrate_page(struct address_space *mapping, struct page *newpage,
600 struct page *page, enum migrate_mode mode)
604 EBUG_ON(!PageLocked(page));
605 EBUG_ON(!PageLocked(newpage));
607 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
608 if (ret != MIGRATEPAGE_SUCCESS)
611 if (PagePrivate(page)) {
612 ClearPagePrivate(page);
614 set_page_private(newpage, page_private(page));
615 set_page_private(page, 0);
617 SetPagePrivate(newpage);
620 if (mode != MIGRATE_SYNC_NO_COPY)
621 migrate_page_copy(newpage, page);
623 migrate_page_states(newpage, page);
624 return MIGRATEPAGE_SUCCESS;
630 static void bch2_readpages_end_io(struct bio *bio)
632 struct bvec_iter_all iter;
635 bio_for_each_segment_all(bv, bio, iter) {
636 struct page *page = bv->bv_page;
638 if (!bio->bi_status) {
639 SetPageUptodate(page);
641 ClearPageUptodate(page);
650 static inline void page_state_init_for_read(struct page *page)
652 SetPagePrivate(page);
656 struct readpages_iter {
657 struct address_space *mapping;
665 static int readpages_iter_init(struct readpages_iter *iter,
666 struct address_space *mapping,
667 struct list_head *pages, unsigned nr_pages)
669 memset(iter, 0, sizeof(*iter));
671 iter->mapping = mapping;
672 iter->offset = list_last_entry(pages, struct page, lru)->index;
674 iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS);
678 while (!list_empty(pages)) {
679 struct page *page = list_last_entry(pages, struct page, lru);
681 __bch2_page_state_create(page, __GFP_NOFAIL);
683 iter->pages[iter->nr_pages++] = page;
684 list_del(&page->lru);
690 static inline struct page *readpage_iter_next(struct readpages_iter *iter)
696 BUG_ON(iter->idx > iter->nr_added);
697 BUG_ON(iter->nr_added > iter->nr_pages);
699 if (iter->idx < iter->nr_added)
703 if (iter->idx == iter->nr_pages)
706 ret = add_to_page_cache_lru_vec(iter->mapping,
707 iter->pages + iter->nr_added,
708 iter->nr_pages - iter->nr_added,
709 iter->offset + iter->nr_added,
714 page = iter->pages[iter->nr_added];
718 __bch2_page_state_release(page);
722 iter->nr_added += ret;
724 for (i = iter->idx; i < iter->nr_added; i++)
725 put_page(iter->pages[i]);
727 EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx);
729 return iter->pages[iter->idx];
732 static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k)
734 struct bvec_iter iter;
736 unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v
737 ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k);
738 unsigned state = k.k->type == KEY_TYPE_reservation
742 bio_for_each_segment(bv, bio, iter) {
743 struct bch_page_state *s = bch2_page_state(bv.bv_page);
746 for (i = bv.bv_offset >> 9;
747 i < (bv.bv_offset + bv.bv_len) >> 9;
749 s->s[i].nr_replicas = nr_ptrs;
750 s->s[i].state = state;
755 static bool extent_partial_reads_expensive(struct bkey_s_c k)
757 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
758 struct bch_extent_crc_unpacked crc;
759 const union bch_extent_entry *i;
761 bkey_for_each_crc(k.k, ptrs, crc, i)
762 if (crc.csum_type || crc.compression_type)
767 static void readpage_bio_extend(struct readpages_iter *iter,
769 unsigned sectors_this_extent,
772 while (bio_sectors(bio) < sectors_this_extent &&
773 bio->bi_vcnt < bio->bi_max_vecs) {
774 pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT;
775 struct page *page = readpage_iter_next(iter);
779 if (iter->offset + iter->idx != page_offset)
787 page = xa_load(&iter->mapping->i_pages, page_offset);
788 if (page && !xa_is_value(page))
791 page = __page_cache_alloc(readahead_gfp_mask(iter->mapping));
795 if (!__bch2_page_state_create(page, 0)) {
800 ret = add_to_page_cache_lru(page, iter->mapping,
801 page_offset, GFP_NOFS);
803 __bch2_page_state_release(page);
811 BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0));
815 static void bchfs_read(struct btree_trans *trans, struct btree_iter *iter,
816 struct bch_read_bio *rbio, u64 inum,
817 struct readpages_iter *readpages_iter)
819 struct bch_fs *c = trans->c;
820 struct bkey_on_stack sk;
821 int flags = BCH_READ_RETRY_IF_STALE|
822 BCH_READ_MAY_PROMOTE;
826 rbio->start_time = local_clock();
828 bkey_on_stack_init(&sk);
832 unsigned bytes, sectors, offset_into_extent;
834 bch2_btree_iter_set_pos(iter,
835 POS(inum, rbio->bio.bi_iter.bi_sector));
837 k = bch2_btree_iter_peek_slot(iter);
842 bkey_on_stack_reassemble(&sk, c, k);
843 k = bkey_i_to_s_c(sk.k);
845 offset_into_extent = iter->pos.offset -
846 bkey_start_offset(k.k);
847 sectors = k.k->size - offset_into_extent;
849 ret = bch2_read_indirect_extent(trans,
850 &offset_into_extent, &sk);
854 sectors = min(sectors, k.k->size - offset_into_extent);
856 bch2_trans_unlock(trans);
859 readpage_bio_extend(readpages_iter, &rbio->bio, sectors,
860 extent_partial_reads_expensive(k));
862 bytes = min(sectors, bio_sectors(&rbio->bio)) << 9;
863 swap(rbio->bio.bi_iter.bi_size, bytes);
865 if (rbio->bio.bi_iter.bi_size == bytes)
866 flags |= BCH_READ_LAST_FRAGMENT;
868 if (bkey_extent_is_allocation(k.k))
869 bch2_add_page_sectors(&rbio->bio, k);
871 bch2_read_extent(trans, rbio, k, offset_into_extent, flags);
873 if (flags & BCH_READ_LAST_FRAGMENT)
876 swap(rbio->bio.bi_iter.bi_size, bytes);
877 bio_advance(&rbio->bio, bytes);
884 bcache_io_error(c, &rbio->bio, "btree IO error %i", ret);
885 bio_endio(&rbio->bio);
888 bkey_on_stack_exit(&sk, c);
891 int bch2_readpages(struct file *file, struct address_space *mapping,
892 struct list_head *pages, unsigned nr_pages)
894 struct bch_inode_info *inode = to_bch_ei(mapping->host);
895 struct bch_fs *c = inode->v.i_sb->s_fs_info;
896 struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
897 struct btree_trans trans;
898 struct btree_iter *iter;
900 struct readpages_iter readpages_iter;
903 ret = readpages_iter_init(&readpages_iter, mapping, pages, nr_pages);
906 bch2_trans_init(&trans, c, 0, 0);
908 iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN,
911 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
913 while ((page = readpage_iter_next(&readpages_iter))) {
914 pgoff_t index = readpages_iter.offset + readpages_iter.idx;
915 unsigned n = min_t(unsigned,
916 readpages_iter.nr_pages -
919 struct bch_read_bio *rbio =
920 rbio_init(bio_alloc_bioset(GFP_NOFS, n, &c->bio_read),
923 readpages_iter.idx++;
925 bio_set_op_attrs(&rbio->bio, REQ_OP_READ, 0);
926 rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTOR_SHIFT;
927 rbio->bio.bi_end_io = bch2_readpages_end_io;
928 BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0));
930 bchfs_read(&trans, iter, rbio, inode->v.i_ino,
934 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
936 bch2_trans_exit(&trans);
937 kfree(readpages_iter.pages);
942 static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio,
943 u64 inum, struct page *page)
945 struct btree_trans trans;
946 struct btree_iter *iter;
948 bch2_page_state_create(page, __GFP_NOFAIL);
950 bio_set_op_attrs(&rbio->bio, REQ_OP_READ, REQ_SYNC);
951 rbio->bio.bi_iter.bi_sector =
952 (sector_t) page->index << PAGE_SECTOR_SHIFT;
953 BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0));
955 bch2_trans_init(&trans, c, 0, 0);
956 iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN,
959 bchfs_read(&trans, iter, rbio, inum, NULL);
961 bch2_trans_exit(&trans);
964 int bch2_readpage(struct file *file, struct page *page)
966 struct bch_inode_info *inode = to_bch_ei(page->mapping->host);
967 struct bch_fs *c = inode->v.i_sb->s_fs_info;
968 struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
969 struct bch_read_bio *rbio;
971 rbio = rbio_init(bio_alloc_bioset(GFP_NOFS, 1, &c->bio_read), opts);
972 rbio->bio.bi_end_io = bch2_readpages_end_io;
974 __bchfs_readpage(c, rbio, inode->v.i_ino, page);
978 static void bch2_read_single_page_end_io(struct bio *bio)
980 complete(bio->bi_private);
983 static int bch2_read_single_page(struct page *page,
984 struct address_space *mapping)
986 struct bch_inode_info *inode = to_bch_ei(mapping->host);
987 struct bch_fs *c = inode->v.i_sb->s_fs_info;
988 struct bch_read_bio *rbio;
990 DECLARE_COMPLETION_ONSTACK(done);
992 rbio = rbio_init(bio_alloc_bioset(GFP_NOFS, 1, &c->bio_read),
993 io_opts(c, &inode->ei_inode));
994 rbio->bio.bi_private = &done;
995 rbio->bio.bi_end_io = bch2_read_single_page_end_io;
997 __bchfs_readpage(c, rbio, inode->v.i_ino, page);
998 wait_for_completion(&done);
1000 ret = blk_status_to_errno(rbio->bio.bi_status);
1001 bio_put(&rbio->bio);
1006 SetPageUptodate(page);
1012 struct bch_writepage_state {
1013 struct bch_writepage_io *io;
1014 struct bch_io_opts opts;
1017 static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c,
1018 struct bch_inode_info *inode)
1020 return (struct bch_writepage_state) {
1021 .opts = io_opts(c, &inode->ei_inode)
1025 static void bch2_writepage_io_free(struct closure *cl)
1027 struct bch_writepage_io *io = container_of(cl,
1028 struct bch_writepage_io, cl);
1030 bio_put(&io->op.wbio.bio);
1033 static void bch2_writepage_io_done(struct closure *cl)
1035 struct bch_writepage_io *io = container_of(cl,
1036 struct bch_writepage_io, cl);
1037 struct bch_fs *c = io->op.c;
1038 struct bio *bio = &io->op.wbio.bio;
1039 struct bvec_iter_all iter;
1040 struct bio_vec *bvec;
1044 bio_for_each_segment_all(bvec, bio, iter) {
1045 struct bch_page_state *s;
1047 SetPageError(bvec->bv_page);
1048 mapping_set_error(bvec->bv_page->mapping, -EIO);
1050 s = __bch2_page_state(bvec->bv_page);
1051 spin_lock(&s->lock);
1052 for (i = 0; i < PAGE_SECTORS; i++)
1053 s->s[i].nr_replicas = 0;
1054 spin_unlock(&s->lock);
1058 if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) {
1059 bio_for_each_segment_all(bvec, bio, iter) {
1060 struct bch_page_state *s;
1062 s = __bch2_page_state(bvec->bv_page);
1063 spin_lock(&s->lock);
1064 for (i = 0; i < PAGE_SECTORS; i++)
1065 s->s[i].nr_replicas = 0;
1066 spin_unlock(&s->lock);
1071 * racing with fallocate can cause us to add fewer sectors than
1072 * expected - but we shouldn't add more sectors than expected:
1074 BUG_ON(io->op.i_sectors_delta > 0);
1077 * (error (due to going RO) halfway through a page can screw that up
1080 BUG_ON(io->op.op.i_sectors_delta >= PAGE_SECTORS);
1084 * PageWriteback is effectively our ref on the inode - fixup i_blocks
1085 * before calling end_page_writeback:
1087 i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta);
1089 bio_for_each_segment_all(bvec, bio, iter) {
1090 struct bch_page_state *s = __bch2_page_state(bvec->bv_page);
1092 if (atomic_dec_and_test(&s->write_count))
1093 end_page_writeback(bvec->bv_page);
1096 closure_return_with_destructor(&io->cl, bch2_writepage_io_free);
1099 static void bch2_writepage_do_io(struct bch_writepage_state *w)
1101 struct bch_writepage_io *io = w->io;
1104 closure_call(&io->op.cl, bch2_write, NULL, &io->cl);
1105 continue_at(&io->cl, bch2_writepage_io_done, NULL);
1109 * Get a bch_writepage_io and add @page to it - appending to an existing one if
1110 * possible, else allocating a new one:
1112 static void bch2_writepage_io_alloc(struct bch_fs *c,
1113 struct writeback_control *wbc,
1114 struct bch_writepage_state *w,
1115 struct bch_inode_info *inode,
1117 unsigned nr_replicas)
1119 struct bch_write_op *op;
1121 w->io = container_of(bio_alloc_bioset(GFP_NOFS,
1123 &c->writepage_bioset),
1124 struct bch_writepage_io, op.wbio.bio);
1126 closure_init(&w->io->cl, NULL);
1127 w->io->inode = inode;
1130 bch2_write_op_init(op, c, w->opts);
1131 op->target = w->opts.foreground_target;
1132 op_journal_seq_set(op, &inode->ei_journal_seq);
1133 op->nr_replicas = nr_replicas;
1134 op->res.nr_replicas = nr_replicas;
1135 op->write_point = writepoint_hashed(inode->ei_last_dirtied);
1136 op->pos = POS(inode->v.i_ino, sector);
1137 op->wbio.bio.bi_iter.bi_sector = sector;
1138 op->wbio.bio.bi_opf = wbc_to_write_flags(wbc);
1141 static int __bch2_writepage(struct page *page,
1142 struct writeback_control *wbc,
1145 struct bch_inode_info *inode = to_bch_ei(page->mapping->host);
1146 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1147 struct bch_writepage_state *w = data;
1148 struct bch_page_state *s, orig;
1149 unsigned i, offset, nr_replicas_this_write = U32_MAX;
1150 loff_t i_size = i_size_read(&inode->v);
1151 pgoff_t end_index = i_size >> PAGE_SHIFT;
1154 EBUG_ON(!PageUptodate(page));
1156 /* Is the page fully inside i_size? */
1157 if (page->index < end_index)
1160 /* Is the page fully outside i_size? (truncate in progress) */
1161 offset = i_size & (PAGE_SIZE - 1);
1162 if (page->index > end_index || !offset) {
1168 * The page straddles i_size. It must be zeroed out on each and every
1169 * writepage invocation because it may be mmapped. "A file is mapped
1170 * in multiples of the page size. For a file that is not a multiple of
1171 * the page size, the remaining memory is zeroed when mapped, and
1172 * writes to that region are not written out to the file."
1174 zero_user_segment(page, offset, PAGE_SIZE);
1176 s = bch2_page_state_create(page, __GFP_NOFAIL);
1178 ret = bch2_get_page_disk_reservation(c, inode, page, true);
1181 mapping_set_error(page->mapping, ret);
1186 /* Before unlocking the page, get copy of reservations: */
1189 for (i = 0; i < PAGE_SECTORS; i++) {
1190 if (s->s[i].state < SECTOR_DIRTY)
1193 nr_replicas_this_write =
1194 min_t(unsigned, nr_replicas_this_write,
1195 s->s[i].nr_replicas +
1196 s->s[i].replicas_reserved);
1199 for (i = 0; i < PAGE_SECTORS; i++) {
1200 if (s->s[i].state < SECTOR_DIRTY)
1203 s->s[i].nr_replicas = w->opts.compression
1204 ? 0 : nr_replicas_this_write;
1206 s->s[i].replicas_reserved = 0;
1207 s->s[i].state = SECTOR_ALLOCATED;
1210 BUG_ON(atomic_read(&s->write_count));
1211 atomic_set(&s->write_count, 1);
1213 BUG_ON(PageWriteback(page));
1214 set_page_writeback(page);
1220 unsigned sectors = 1, dirty_sectors = 0, reserved_sectors = 0;
1223 while (offset < PAGE_SECTORS &&
1224 orig.s[offset].state < SECTOR_DIRTY)
1227 if (offset == PAGE_SECTORS)
1230 sector = ((u64) page->index << PAGE_SECTOR_SHIFT) + offset;
1232 while (offset + sectors < PAGE_SECTORS &&
1233 orig.s[offset + sectors].state >= SECTOR_DIRTY)
1236 for (i = offset; i < offset + sectors; i++) {
1237 reserved_sectors += orig.s[i].replicas_reserved;
1238 dirty_sectors += orig.s[i].state == SECTOR_DIRTY;
1242 (w->io->op.res.nr_replicas != nr_replicas_this_write ||
1243 bio_full(&w->io->op.wbio.bio, PAGE_SIZE) ||
1244 w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >=
1245 (BIO_MAX_PAGES * PAGE_SIZE) ||
1246 bio_end_sector(&w->io->op.wbio.bio) != sector))
1247 bch2_writepage_do_io(w);
1250 bch2_writepage_io_alloc(c, wbc, w, inode, sector,
1251 nr_replicas_this_write);
1253 atomic_inc(&s->write_count);
1255 BUG_ON(inode != w->io->inode);
1256 BUG_ON(!bio_add_page(&w->io->op.wbio.bio, page,
1257 sectors << 9, offset << 9));
1259 /* Check for writing past i_size: */
1260 WARN_ON((bio_end_sector(&w->io->op.wbio.bio) << 9) >
1261 round_up(i_size, block_bytes(c)));
1263 w->io->op.res.sectors += reserved_sectors;
1264 w->io->op.i_sectors_delta -= dirty_sectors;
1265 w->io->op.new_i_size = i_size;
1270 if (atomic_dec_and_test(&s->write_count))
1271 end_page_writeback(page);
1276 int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc)
1278 struct bch_fs *c = mapping->host->i_sb->s_fs_info;
1279 struct bch_writepage_state w =
1280 bch_writepage_state_init(c, to_bch_ei(mapping->host));
1281 struct blk_plug plug;
1284 blk_start_plug(&plug);
1285 ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w);
1287 bch2_writepage_do_io(&w);
1288 blk_finish_plug(&plug);
1292 int bch2_writepage(struct page *page, struct writeback_control *wbc)
1294 struct bch_fs *c = page->mapping->host->i_sb->s_fs_info;
1295 struct bch_writepage_state w =
1296 bch_writepage_state_init(c, to_bch_ei(page->mapping->host));
1299 ret = __bch2_writepage(page, wbc, &w);
1301 bch2_writepage_do_io(&w);
1306 /* buffered writes: */
1308 int bch2_write_begin(struct file *file, struct address_space *mapping,
1309 loff_t pos, unsigned len, unsigned flags,
1310 struct page **pagep, void **fsdata)
1312 struct bch_inode_info *inode = to_bch_ei(mapping->host);
1313 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1314 struct bch2_page_reservation *res;
1315 pgoff_t index = pos >> PAGE_SHIFT;
1316 unsigned offset = pos & (PAGE_SIZE - 1);
1320 res = kmalloc(sizeof(*res), GFP_KERNEL);
1324 bch2_page_reservation_init(c, inode, res);
1327 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
1329 page = grab_cache_page_write_begin(mapping, index, flags);
1333 if (PageUptodate(page))
1336 /* If we're writing entire page, don't need to read it in first: */
1337 if (len == PAGE_SIZE)
1340 if (!offset && pos + len >= inode->v.i_size) {
1341 zero_user_segment(page, len, PAGE_SIZE);
1342 flush_dcache_page(page);
1346 if (index > inode->v.i_size >> PAGE_SHIFT) {
1347 zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE);
1348 flush_dcache_page(page);
1352 ret = bch2_read_single_page(page, mapping);
1356 ret = bch2_page_reservation_get(c, inode, page, res,
1359 if (!PageUptodate(page)) {
1361 * If the page hasn't been read in, we won't know if we
1362 * actually need a reservation - we don't actually need
1363 * to read here, we just need to check if the page is
1364 * fully backed by uncompressed data:
1379 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
1385 int bch2_write_end(struct file *file, struct address_space *mapping,
1386 loff_t pos, unsigned len, unsigned copied,
1387 struct page *page, void *fsdata)
1389 struct bch_inode_info *inode = to_bch_ei(mapping->host);
1390 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1391 struct bch2_page_reservation *res = fsdata;
1392 unsigned offset = pos & (PAGE_SIZE - 1);
1394 lockdep_assert_held(&inode->v.i_rwsem);
1396 if (unlikely(copied < len && !PageUptodate(page))) {
1398 * The page needs to be read in, but that would destroy
1399 * our partial write - simplest thing is to just force
1400 * userspace to redo the write:
1402 zero_user(page, 0, PAGE_SIZE);
1403 flush_dcache_page(page);
1407 spin_lock(&inode->v.i_lock);
1408 if (pos + copied > inode->v.i_size)
1409 i_size_write(&inode->v, pos + copied);
1410 spin_unlock(&inode->v.i_lock);
1413 if (!PageUptodate(page))
1414 SetPageUptodate(page);
1416 bch2_set_page_dirty(c, inode, page, res, offset, copied);
1418 inode->ei_last_dirtied = (unsigned long) current;
1423 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
1425 bch2_page_reservation_put(c, inode, res);
1431 #define WRITE_BATCH_PAGES 32
1433 static int __bch2_buffered_write(struct bch_inode_info *inode,
1434 struct address_space *mapping,
1435 struct iov_iter *iter,
1436 loff_t pos, unsigned len)
1438 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1439 struct page *pages[WRITE_BATCH_PAGES];
1440 struct bch2_page_reservation res;
1441 unsigned long index = pos >> PAGE_SHIFT;
1442 unsigned offset = pos & (PAGE_SIZE - 1);
1443 unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1444 unsigned i, reserved = 0, set_dirty = 0;
1445 unsigned copied = 0, nr_pages_copied = 0;
1449 BUG_ON(nr_pages > ARRAY_SIZE(pages));
1451 bch2_page_reservation_init(c, inode, &res);
1453 for (i = 0; i < nr_pages; i++) {
1454 pages[i] = grab_cache_page_write_begin(mapping, index + i, 0);
1461 len = min_t(unsigned, len,
1462 nr_pages * PAGE_SIZE - offset);
1467 if (offset && !PageUptodate(pages[0])) {
1468 ret = bch2_read_single_page(pages[0], mapping);
1473 if ((pos + len) & (PAGE_SIZE - 1) &&
1474 !PageUptodate(pages[nr_pages - 1])) {
1475 if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) {
1476 zero_user(pages[nr_pages - 1], 0, PAGE_SIZE);
1478 ret = bch2_read_single_page(pages[nr_pages - 1], mapping);
1484 while (reserved < len) {
1485 struct page *page = pages[(offset + reserved) >> PAGE_SHIFT];
1486 unsigned pg_offset = (offset + reserved) & (PAGE_SIZE - 1);
1487 unsigned pg_len = min_t(unsigned, len - reserved,
1488 PAGE_SIZE - pg_offset);
1490 ret = bch2_page_reservation_get(c, inode, page, &res,
1491 pg_offset, pg_len, true);
1493 if (ret && !PageUptodate(page)) {
1494 ret = bch2_read_single_page(page, mapping);
1496 goto retry_reservation;
1505 if (mapping_writably_mapped(mapping))
1506 for (i = 0; i < nr_pages; i++)
1507 flush_dcache_page(pages[i]);
1509 while (copied < len) {
1510 struct page *page = pages[(offset + copied) >> PAGE_SHIFT];
1511 unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1);
1512 unsigned pg_len = min_t(unsigned, len - copied,
1513 PAGE_SIZE - pg_offset);
1514 unsigned pg_copied = iov_iter_copy_from_user_atomic(page,
1515 iter, pg_offset, pg_len);
1520 if (!PageUptodate(page) &&
1521 pg_copied != PAGE_SIZE &&
1522 pos + copied + pg_copied < inode->v.i_size) {
1523 zero_user(page, 0, PAGE_SIZE);
1527 flush_dcache_page(page);
1528 iov_iter_advance(iter, pg_copied);
1529 copied += pg_copied;
1531 if (pg_copied != pg_len)
1538 spin_lock(&inode->v.i_lock);
1539 if (pos + copied > inode->v.i_size)
1540 i_size_write(&inode->v, pos + copied);
1541 spin_unlock(&inode->v.i_lock);
1543 while (set_dirty < copied) {
1544 struct page *page = pages[(offset + set_dirty) >> PAGE_SHIFT];
1545 unsigned pg_offset = (offset + set_dirty) & (PAGE_SIZE - 1);
1546 unsigned pg_len = min_t(unsigned, copied - set_dirty,
1547 PAGE_SIZE - pg_offset);
1549 if (!PageUptodate(page))
1550 SetPageUptodate(page);
1552 bch2_set_page_dirty(c, inode, page, &res, pg_offset, pg_len);
1556 set_dirty += pg_len;
1559 nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE);
1560 inode->ei_last_dirtied = (unsigned long) current;
1562 for (i = nr_pages_copied; i < nr_pages; i++) {
1563 unlock_page(pages[i]);
1567 bch2_page_reservation_put(c, inode, &res);
1569 return copied ?: ret;
1572 static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter)
1574 struct file *file = iocb->ki_filp;
1575 struct address_space *mapping = file->f_mapping;
1576 struct bch_inode_info *inode = file_bch_inode(file);
1577 loff_t pos = iocb->ki_pos;
1578 ssize_t written = 0;
1581 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
1584 unsigned offset = pos & (PAGE_SIZE - 1);
1585 unsigned bytes = min_t(unsigned long, iov_iter_count(iter),
1586 PAGE_SIZE * WRITE_BATCH_PAGES - offset);
1589 * Bring in the user page that we will copy from _first_.
1590 * Otherwise there's a nasty deadlock on copying from the
1591 * same page as we're writing to, without it being marked
1594 * Not only is this an optimisation, but it is also required
1595 * to check that the address is actually valid, when atomic
1596 * usercopies are used, below.
1598 if (unlikely(iov_iter_fault_in_readable(iter, bytes))) {
1599 bytes = min_t(unsigned long, iov_iter_count(iter),
1600 PAGE_SIZE - offset);
1602 if (unlikely(iov_iter_fault_in_readable(iter, bytes))) {
1608 if (unlikely(fatal_signal_pending(current))) {
1613 ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes);
1614 if (unlikely(ret < 0))
1619 if (unlikely(ret == 0)) {
1621 * If we were unable to copy any data at all, we must
1622 * fall back to a single segment length write.
1624 * If we didn't fallback here, we could livelock
1625 * because not all segments in the iov can be copied at
1626 * once without a pagefault.
1628 bytes = min_t(unsigned long, PAGE_SIZE - offset,
1629 iov_iter_single_seg_count(iter));
1636 balance_dirty_pages_ratelimited(mapping);
1637 } while (iov_iter_count(iter));
1639 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
1641 return written ? written : ret;
1644 /* O_DIRECT reads */
1646 static void bch2_dio_read_complete(struct closure *cl)
1648 struct dio_read *dio = container_of(cl, struct dio_read, cl);
1650 dio->req->ki_complete(dio->req, dio->ret, 0);
1651 bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */
1654 static void bch2_direct_IO_read_endio(struct bio *bio)
1656 struct dio_read *dio = bio->bi_private;
1659 dio->ret = blk_status_to_errno(bio->bi_status);
1661 closure_put(&dio->cl);
1664 static void bch2_direct_IO_read_split_endio(struct bio *bio)
1666 bch2_direct_IO_read_endio(bio);
1667 bio_check_pages_dirty(bio); /* transfers ownership */
1670 static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
1672 struct file *file = req->ki_filp;
1673 struct bch_inode_info *inode = file_bch_inode(file);
1674 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1675 struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
1676 struct dio_read *dio;
1678 loff_t offset = req->ki_pos;
1679 bool sync = is_sync_kiocb(req);
1683 if ((offset|iter->count) & (block_bytes(c) - 1))
1686 ret = min_t(loff_t, iter->count,
1687 max_t(loff_t, 0, i_size_read(&inode->v) - offset));
1692 shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c));
1693 iter->count -= shorten;
1695 bio = bio_alloc_bioset(GFP_KERNEL,
1696 iov_iter_npages(iter, BIO_MAX_PAGES),
1697 &c->dio_read_bioset);
1699 bio->bi_end_io = bch2_direct_IO_read_endio;
1701 dio = container_of(bio, struct dio_read, rbio.bio);
1702 closure_init(&dio->cl, NULL);
1705 * this is a _really_ horrible hack just to avoid an atomic sub at the
1709 set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL);
1710 atomic_set(&dio->cl.remaining,
1711 CLOSURE_REMAINING_INITIALIZER -
1713 CLOSURE_DESTRUCTOR);
1715 atomic_set(&dio->cl.remaining,
1716 CLOSURE_REMAINING_INITIALIZER + 1);
1723 while (iter->count) {
1724 bio = bio_alloc_bioset(GFP_KERNEL,
1725 iov_iter_npages(iter, BIO_MAX_PAGES),
1727 bio->bi_end_io = bch2_direct_IO_read_split_endio;
1729 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC);
1730 bio->bi_iter.bi_sector = offset >> 9;
1731 bio->bi_private = dio;
1733 ret = bio_iov_iter_get_pages(bio, iter);
1735 /* XXX: fault inject this path */
1736 bio->bi_status = BLK_STS_RESOURCE;
1741 offset += bio->bi_iter.bi_size;
1742 bio_set_pages_dirty(bio);
1745 closure_get(&dio->cl);
1747 bch2_read(c, rbio_init(bio, opts), inode->v.i_ino);
1750 iter->count += shorten;
1753 closure_sync(&dio->cl);
1754 closure_debug_destroy(&dio->cl);
1756 bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */
1759 return -EIOCBQUEUED;
1763 ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1765 struct file *file = iocb->ki_filp;
1766 struct bch_inode_info *inode = file_bch_inode(file);
1767 struct address_space *mapping = file->f_mapping;
1768 size_t count = iov_iter_count(iter);
1772 return 0; /* skip atime */
1774 if (iocb->ki_flags & IOCB_DIRECT) {
1775 struct blk_plug plug;
1777 ret = filemap_write_and_wait_range(mapping,
1779 iocb->ki_pos + count - 1);
1783 file_accessed(file);
1785 blk_start_plug(&plug);
1786 ret = bch2_direct_IO_read(iocb, iter);
1787 blk_finish_plug(&plug);
1790 iocb->ki_pos += ret;
1792 bch2_pagecache_add_get(&inode->ei_pagecache_lock);
1793 ret = generic_file_read_iter(iocb, iter);
1794 bch2_pagecache_add_put(&inode->ei_pagecache_lock);
1800 /* O_DIRECT writes */
1802 static void bch2_dio_write_loop_async(struct bch_write_op *);
1804 static long bch2_dio_write_loop(struct dio_write *dio)
1806 bool kthread = (current->flags & PF_KTHREAD) != 0;
1807 struct kiocb *req = dio->req;
1808 struct address_space *mapping = req->ki_filp->f_mapping;
1809 struct bch_inode_info *inode = file_bch_inode(req->ki_filp);
1810 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1811 struct bio *bio = &dio->op.wbio.bio;
1812 struct bvec_iter_all iter;
1815 bool sync = dio->sync;
1823 kthread_use_mm(dio->mm);
1824 BUG_ON(current->faults_disabled_mapping);
1825 current->faults_disabled_mapping = mapping;
1827 ret = bio_iov_iter_get_pages(bio, &dio->iter);
1829 current->faults_disabled_mapping = NULL;
1831 kthread_unuse_mm(dio->mm);
1833 if (unlikely(ret < 0))
1836 unaligned = bio->bi_iter.bi_size & (block_bytes(c) - 1);
1837 bio->bi_iter.bi_size -= unaligned;
1838 iov_iter_revert(&dio->iter, unaligned);
1840 if (!bio->bi_iter.bi_size) {
1842 * bio_iov_iter_get_pages was only able to get <
1843 * blocksize worth of pages:
1845 bio_for_each_segment_all(bv, bio, iter)
1846 put_page(bv->bv_page);
1851 bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode));
1852 dio->op.end_io = bch2_dio_write_loop_async;
1853 dio->op.target = dio->op.opts.foreground_target;
1854 op_journal_seq_set(&dio->op, &inode->ei_journal_seq);
1855 dio->op.write_point = writepoint_hashed((unsigned long) current);
1856 dio->op.nr_replicas = dio->op.opts.data_replicas;
1857 dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9);
1859 if ((req->ki_flags & IOCB_DSYNC) &&
1860 !c->opts.journal_flush_disabled)
1861 dio->op.flags |= BCH_WRITE_FLUSH;
1863 ret = bch2_disk_reservation_get(c, &dio->op.res, bio_sectors(bio),
1864 dio->op.opts.data_replicas, 0);
1865 if (unlikely(ret) &&
1866 !bch2_check_range_allocated(c, dio->op.pos,
1867 bio_sectors(bio), dio->op.opts.data_replicas))
1870 task_io_account_write(bio->bi_iter.bi_size);
1872 if (!dio->sync && !dio->loop && dio->iter.count) {
1873 struct iovec *iov = dio->inline_vecs;
1875 if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) {
1876 iov = kmalloc(dio->iter.nr_segs * sizeof(*iov),
1878 if (unlikely(!iov)) {
1879 dio->sync = sync = true;
1883 dio->free_iov = true;
1886 memcpy(iov, dio->iter.iov, dio->iter.nr_segs * sizeof(*iov));
1887 dio->iter.iov = iov;
1891 closure_call(&dio->op.cl, bch2_write, NULL, NULL);
1894 wait_for_completion(&dio->done);
1896 return -EIOCBQUEUED;
1898 i_sectors_acct(c, inode, &dio->quota_res,
1899 dio->op.i_sectors_delta);
1900 req->ki_pos += (u64) dio->op.written << 9;
1901 dio->written += dio->op.written;
1903 spin_lock(&inode->v.i_lock);
1904 if (req->ki_pos > inode->v.i_size)
1905 i_size_write(&inode->v, req->ki_pos);
1906 spin_unlock(&inode->v.i_lock);
1908 bio_for_each_segment_all(bv, bio, iter)
1909 put_page(bv->bv_page);
1910 if (!dio->iter.count || dio->op.error)
1914 reinit_completion(&dio->done);
1917 ret = dio->op.error ?: ((long) dio->written << 9);
1919 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
1920 bch2_quota_reservation_put(c, inode, &dio->quota_res);
1923 kfree(dio->iter.iov);
1927 /* inode->i_dio_count is our ref on inode and thus bch_fs */
1928 inode_dio_end(&inode->v);
1931 req->ki_complete(req, ret, 0);
1937 static void bch2_dio_write_loop_async(struct bch_write_op *op)
1939 struct dio_write *dio = container_of(op, struct dio_write, op);
1942 complete(&dio->done);
1944 bch2_dio_write_loop(dio);
1948 ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
1950 struct file *file = req->ki_filp;
1951 struct address_space *mapping = file->f_mapping;
1952 struct bch_inode_info *inode = file_bch_inode(file);
1953 struct bch_fs *c = inode->v.i_sb->s_fs_info;
1954 struct dio_write *dio;
1956 bool locked = true, extending;
1960 prefetch((void *) &c->opts + 64);
1961 prefetch(&inode->ei_inode);
1962 prefetch((void *) &inode->ei_inode + 64);
1964 inode_lock(&inode->v);
1966 ret = generic_write_checks(req, iter);
1967 if (unlikely(ret <= 0))
1970 ret = file_remove_privs(file);
1974 ret = file_update_time(file);
1978 if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1)))
1981 inode_dio_begin(&inode->v);
1982 bch2_pagecache_block_get(&inode->ei_pagecache_lock);
1984 extending = req->ki_pos + iter->count > inode->v.i_size;
1986 inode_unlock(&inode->v);
1990 bio = bio_alloc_bioset(GFP_KERNEL,
1991 iov_iter_npages(iter, BIO_MAX_PAGES),
1992 &c->dio_write_bioset);
1993 dio = container_of(bio, struct dio_write, op.wbio.bio);
1994 init_completion(&dio->done);
1996 dio->mm = current->mm;
1998 dio->sync = is_sync_kiocb(req) || extending;
1999 dio->free_iov = false;
2000 dio->quota_res.sectors = 0;
2004 ret = bch2_quota_reservation_add(c, inode, &dio->quota_res,
2005 iter->count >> 9, true);
2009 ret = write_invalidate_inode_pages_range(mapping,
2011 req->ki_pos + iter->count - 1);
2015 ret = bch2_dio_write_loop(dio);
2018 inode_unlock(&inode->v);
2021 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
2022 bch2_quota_reservation_put(c, inode, &dio->quota_res);
2024 inode_dio_end(&inode->v);
2028 ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from)
2030 struct file *file = iocb->ki_filp;
2031 struct bch_inode_info *inode = file_bch_inode(file);
2034 if (iocb->ki_flags & IOCB_DIRECT)
2035 return bch2_direct_write(iocb, from);
2037 /* We can write back this queue in page reclaim */
2038 current->backing_dev_info = inode_to_bdi(&inode->v);
2039 inode_lock(&inode->v);
2041 ret = generic_write_checks(iocb, from);
2045 ret = file_remove_privs(file);
2049 ret = file_update_time(file);
2053 ret = bch2_buffered_write(iocb, from);
2054 if (likely(ret > 0))
2055 iocb->ki_pos += ret;
2057 inode_unlock(&inode->v);
2058 current->backing_dev_info = NULL;
2061 ret = generic_write_sync(iocb, ret);
2068 int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2070 struct bch_inode_info *inode = file_bch_inode(file);
2071 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2074 ret = file_write_and_wait_range(file, start, end);
2078 if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC))
2081 ret = sync_inode_metadata(&inode->v, 1);
2085 if (!c->opts.journal_flush_disabled)
2086 ret = bch2_journal_flush_seq(&c->journal,
2087 inode->ei_journal_seq);
2088 ret2 = file_check_and_advance_wb_err(file);
2095 static inline int range_has_data(struct bch_fs *c,
2099 struct btree_trans trans;
2100 struct btree_iter *iter;
2104 bch2_trans_init(&trans, c, 0, 0);
2106 for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, start, 0, k, ret) {
2107 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
2110 if (bkey_extent_is_data(k.k)) {
2116 return bch2_trans_exit(&trans) ?: ret;
2119 static int __bch2_truncate_page(struct bch_inode_info *inode,
2120 pgoff_t index, loff_t start, loff_t end)
2122 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2123 struct address_space *mapping = inode->v.i_mapping;
2124 struct bch_page_state *s;
2125 unsigned start_offset = start & (PAGE_SIZE - 1);
2126 unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1;
2131 /* Page boundary? Nothing to do */
2132 if (!((index == start >> PAGE_SHIFT && start_offset) ||
2133 (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE)))
2137 if (index << PAGE_SHIFT >= inode->v.i_size)
2140 page = find_lock_page(mapping, index);
2143 * XXX: we're doing two index lookups when we end up reading the
2146 ret = range_has_data(c,
2147 POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT),
2148 POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT));
2152 page = find_or_create_page(mapping, index, GFP_KERNEL);
2153 if (unlikely(!page)) {
2159 s = bch2_page_state_create(page, 0);
2165 if (!PageUptodate(page)) {
2166 ret = bch2_read_single_page(page, mapping);
2171 if (index != start >> PAGE_SHIFT)
2173 if (index != end >> PAGE_SHIFT)
2174 end_offset = PAGE_SIZE;
2176 for (i = round_up(start_offset, block_bytes(c)) >> 9;
2177 i < round_down(end_offset, block_bytes(c)) >> 9;
2179 s->s[i].nr_replicas = 0;
2180 s->s[i].state = SECTOR_UNALLOCATED;
2183 zero_user_segment(page, start_offset, end_offset);
2186 * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
2188 * XXX: because we aren't currently tracking whether the page has actual
2189 * data in it (vs. just 0s, or only partially written) this wrong. ick.
2191 ret = bch2_get_page_disk_reservation(c, inode, page, false);
2195 * This removes any writeable userspace mappings; we need to force
2196 * .page_mkwrite to be called again before any mmapped writes, to
2197 * redirty the full page:
2200 __set_page_dirty_nobuffers(page);
2208 static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from)
2210 return __bch2_truncate_page(inode, from >> PAGE_SHIFT,
2211 from, round_up(from, PAGE_SIZE));
2214 static int bch2_extend(struct bch_inode_info *inode,
2215 struct bch_inode_unpacked *inode_u,
2216 struct iattr *iattr)
2218 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2219 struct address_space *mapping = inode->v.i_mapping;
2225 * this has to be done _before_ extending i_size:
2227 ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
2231 truncate_setsize(&inode->v, iattr->ia_size);
2232 setattr_copy(&inode->v, iattr);
2234 mutex_lock(&inode->ei_update_lock);
2235 ret = bch2_write_inode_size(c, inode, inode->v.i_size,
2236 ATTR_MTIME|ATTR_CTIME);
2237 mutex_unlock(&inode->ei_update_lock);
2242 static int bch2_truncate_finish_fn(struct bch_inode_info *inode,
2243 struct bch_inode_unpacked *bi,
2246 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2248 bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
2249 bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
2253 static int bch2_truncate_start_fn(struct bch_inode_info *inode,
2254 struct bch_inode_unpacked *bi, void *p)
2256 u64 *new_i_size = p;
2258 bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY;
2259 bi->bi_size = *new_i_size;
2263 int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr)
2265 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2266 struct address_space *mapping = inode->v.i_mapping;
2267 struct bch_inode_unpacked inode_u;
2268 struct btree_trans trans;
2269 struct btree_iter *iter;
2270 u64 new_i_size = iattr->ia_size;
2271 s64 i_sectors_delta = 0;
2274 inode_dio_wait(&inode->v);
2275 bch2_pagecache_block_get(&inode->ei_pagecache_lock);
2278 * fetch current on disk i_size: inode is locked, i_size can only
2279 * increase underneath us:
2281 bch2_trans_init(&trans, c, 0, 0);
2282 iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino, 0);
2283 ret = PTR_ERR_OR_ZERO(iter);
2284 bch2_trans_exit(&trans);
2290 * check this before next assertion; on filesystem error our normal
2291 * invariants are a bit broken (truncate has to truncate the page cache
2292 * before the inode).
2294 ret = bch2_journal_error(&c->journal);
2298 BUG_ON(inode->v.i_size < inode_u.bi_size);
2300 if (iattr->ia_size > inode->v.i_size) {
2301 ret = bch2_extend(inode, &inode_u, iattr);
2305 ret = bch2_truncate_page(inode, iattr->ia_size);
2310 * When extending, we're going to write the new i_size to disk
2311 * immediately so we need to flush anything above the current on disk
2314 * Also, when extending we need to flush the page that i_size currently
2315 * straddles - if it's mapped to userspace, we need to ensure that
2316 * userspace has to redirty it and call .mkwrite -> set_page_dirty
2317 * again to allocate the part of the page that was extended.
2319 if (iattr->ia_size > inode_u.bi_size)
2320 ret = filemap_write_and_wait_range(mapping,
2322 iattr->ia_size - 1);
2323 else if (iattr->ia_size & (PAGE_SIZE - 1))
2324 ret = filemap_write_and_wait_range(mapping,
2325 round_down(iattr->ia_size, PAGE_SIZE),
2326 iattr->ia_size - 1);
2330 mutex_lock(&inode->ei_update_lock);
2331 ret = bch2_write_inode(c, inode, bch2_truncate_start_fn,
2333 mutex_unlock(&inode->ei_update_lock);
2338 truncate_setsize(&inode->v, iattr->ia_size);
2340 ret = bch2_fpunch(c, inode->v.i_ino,
2341 round_up(iattr->ia_size, block_bytes(c)) >> 9,
2342 U64_MAX, &inode->ei_journal_seq, &i_sectors_delta);
2343 i_sectors_acct(c, inode, NULL, i_sectors_delta);
2348 setattr_copy(&inode->v, iattr);
2350 mutex_lock(&inode->ei_update_lock);
2351 ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL,
2352 ATTR_MTIME|ATTR_CTIME);
2353 mutex_unlock(&inode->ei_update_lock);
2355 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
2361 static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
2363 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2364 u64 discard_start = round_up(offset, block_bytes(c)) >> 9;
2365 u64 discard_end = round_down(offset + len, block_bytes(c)) >> 9;
2368 inode_lock(&inode->v);
2369 inode_dio_wait(&inode->v);
2370 bch2_pagecache_block_get(&inode->ei_pagecache_lock);
2372 ret = __bch2_truncate_page(inode,
2373 offset >> PAGE_SHIFT,
2374 offset, offset + len);
2378 if (offset >> PAGE_SHIFT !=
2379 (offset + len) >> PAGE_SHIFT) {
2380 ret = __bch2_truncate_page(inode,
2381 (offset + len) >> PAGE_SHIFT,
2382 offset, offset + len);
2387 truncate_pagecache_range(&inode->v, offset, offset + len - 1);
2389 if (discard_start < discard_end) {
2390 s64 i_sectors_delta = 0;
2392 ret = bch2_fpunch(c, inode->v.i_ino,
2393 discard_start, discard_end,
2394 &inode->ei_journal_seq,
2396 i_sectors_acct(c, inode, NULL, i_sectors_delta);
2399 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
2400 inode_unlock(&inode->v);
2405 static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
2406 loff_t offset, loff_t len,
2409 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2410 struct address_space *mapping = inode->v.i_mapping;
2411 struct bkey_on_stack copy;
2412 struct btree_trans trans;
2413 struct btree_iter *src, *dst;
2414 loff_t shift, new_size;
2418 if ((offset | len) & (block_bytes(c) - 1))
2421 bkey_on_stack_init(©);
2422 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 256);
2425 * We need i_mutex to keep the page cache consistent with the extents
2426 * btree, and the btree consistent with i_size - we don't need outside
2427 * locking for the extents btree itself, because we're using linked
2430 inode_lock(&inode->v);
2431 inode_dio_wait(&inode->v);
2432 bch2_pagecache_block_get(&inode->ei_pagecache_lock);
2436 if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len)
2440 if (offset >= inode->v.i_size)
2443 src_start = U64_MAX;
2447 if (offset + len >= inode->v.i_size)
2450 src_start = offset + len;
2454 new_size = inode->v.i_size + shift;
2456 ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
2461 i_size_write(&inode->v, new_size);
2462 mutex_lock(&inode->ei_update_lock);
2463 ret = bch2_write_inode_size(c, inode, new_size,
2464 ATTR_MTIME|ATTR_CTIME);
2465 mutex_unlock(&inode->ei_update_lock);
2467 s64 i_sectors_delta = 0;
2469 ret = bch2_fpunch(c, inode->v.i_ino,
2470 offset >> 9, (offset + len) >> 9,
2471 &inode->ei_journal_seq,
2473 i_sectors_acct(c, inode, NULL, i_sectors_delta);
2479 src = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
2480 POS(inode->v.i_ino, src_start >> 9),
2482 BUG_ON(IS_ERR_OR_NULL(src));
2484 dst = bch2_trans_copy_iter(&trans, src);
2485 BUG_ON(IS_ERR_OR_NULL(dst));
2488 struct disk_reservation disk_res =
2489 bch2_disk_reservation_init(c, 0);
2490 struct bkey_i delete;
2492 struct bpos next_pos;
2493 struct bpos move_pos = POS(inode->v.i_ino, offset >> 9);
2494 struct bpos atomic_end;
2495 unsigned trigger_flags = 0;
2498 ? bch2_btree_iter_peek_prev(src)
2499 : bch2_btree_iter_peek(src);
2500 if ((ret = bkey_err(k)))
2503 if (!k.k || k.k->p.inode != inode->v.i_ino)
2506 BUG_ON(bkey_cmp(src->pos, bkey_start_pos(k.k)));
2509 bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0)
2512 bkey_on_stack_reassemble(©, c, k);
2515 bkey_cmp(bkey_start_pos(k.k), move_pos) < 0)
2516 bch2_cut_front(move_pos, copy.k);
2518 copy.k->k.p.offset += shift >> 9;
2519 bch2_btree_iter_set_pos(dst, bkey_start_pos(©.k->k));
2521 ret = bch2_extent_atomic_end(dst, copy.k, &atomic_end);
2525 if (bkey_cmp(atomic_end, copy.k->k.p)) {
2527 move_pos = atomic_end;
2528 move_pos.offset -= shift >> 9;
2531 bch2_cut_back(atomic_end, copy.k);
2535 bkey_init(&delete.k);
2536 delete.k.p = copy.k->k.p;
2537 delete.k.size = copy.k->k.size;
2538 delete.k.p.offset -= shift >> 9;
2540 next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p;
2542 if (copy.k->k.size == k.k->size) {
2544 * If we're moving the entire extent, we can skip
2547 trigger_flags |= BTREE_TRIGGER_NORUN;
2549 /* We might end up splitting compressed extents: */
2551 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k));
2553 ret = bch2_disk_reservation_get(c, &disk_res,
2554 copy.k->k.size, nr_ptrs,
2555 BCH_DISK_RESERVATION_NOFAIL);
2559 bch2_btree_iter_set_pos(src, bkey_start_pos(&delete.k));
2561 ret = bch2_trans_update(&trans, src, &delete, trigger_flags) ?:
2562 bch2_trans_update(&trans, dst, copy.k, trigger_flags) ?:
2563 bch2_trans_commit(&trans, &disk_res,
2564 &inode->ei_journal_seq,
2565 BTREE_INSERT_NOFAIL);
2566 bch2_disk_reservation_put(c, &disk_res);
2569 bch2_btree_iter_set_pos(src, next_pos);
2576 bch2_trans_cond_resched(&trans);
2578 bch2_trans_unlock(&trans);
2581 i_size_write(&inode->v, new_size);
2582 mutex_lock(&inode->ei_update_lock);
2583 ret = bch2_write_inode_size(c, inode, new_size,
2584 ATTR_MTIME|ATTR_CTIME);
2585 mutex_unlock(&inode->ei_update_lock);
2588 bch2_trans_exit(&trans);
2589 bkey_on_stack_exit(©, c);
2590 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
2591 inode_unlock(&inode->v);
2595 static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
2596 loff_t offset, loff_t len)
2598 struct address_space *mapping = inode->v.i_mapping;
2599 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2600 struct btree_trans trans;
2601 struct btree_iter *iter;
2602 struct bpos end_pos;
2603 loff_t end = offset + len;
2604 loff_t block_start = round_down(offset, block_bytes(c));
2605 loff_t block_end = round_up(end, block_bytes(c));
2607 unsigned replicas = io_opts(c, &inode->ei_inode).data_replicas;
2610 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2612 inode_lock(&inode->v);
2613 inode_dio_wait(&inode->v);
2614 bch2_pagecache_block_get(&inode->ei_pagecache_lock);
2616 if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
2617 ret = inode_newsize_ok(&inode->v, end);
2622 if (mode & FALLOC_FL_ZERO_RANGE) {
2623 ret = __bch2_truncate_page(inode,
2624 offset >> PAGE_SHIFT,
2628 offset >> PAGE_SHIFT != end >> PAGE_SHIFT)
2629 ret = __bch2_truncate_page(inode,
2636 truncate_pagecache_range(&inode->v, offset, end - 1);
2639 iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
2640 POS(inode->v.i_ino, block_start >> 9),
2641 BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
2642 end_pos = POS(inode->v.i_ino, block_end >> 9);
2644 while (bkey_cmp(iter->pos, end_pos) < 0) {
2645 s64 i_sectors_delta = 0;
2646 struct disk_reservation disk_res = { 0 };
2647 struct quota_res quota_res = { 0 };
2648 struct bkey_i_reservation reservation;
2651 bch2_trans_begin(&trans);
2653 k = bch2_btree_iter_peek_slot(iter);
2654 if ((ret = bkey_err(k)))
2657 /* already reserved */
2658 if (k.k->type == KEY_TYPE_reservation &&
2659 bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) {
2660 bch2_btree_iter_next_slot(iter);
2664 if (bkey_extent_is_data(k.k) &&
2665 !(mode & FALLOC_FL_ZERO_RANGE)) {
2666 bch2_btree_iter_next_slot(iter);
2670 bkey_reservation_init(&reservation.k_i);
2671 reservation.k.type = KEY_TYPE_reservation;
2672 reservation.k.p = k.k->p;
2673 reservation.k.size = k.k->size;
2675 bch2_cut_front(iter->pos, &reservation.k_i);
2676 bch2_cut_back(end_pos, &reservation.k_i);
2678 sectors = reservation.k.size;
2679 reservation.v.nr_replicas = bch2_bkey_nr_ptrs_allocated(k);
2681 if (!bkey_extent_is_allocation(k.k)) {
2682 ret = bch2_quota_reservation_add(c, inode,
2689 if (reservation.v.nr_replicas < replicas ||
2690 bch2_bkey_sectors_compressed(k)) {
2691 ret = bch2_disk_reservation_get(c, &disk_res, sectors,
2696 reservation.v.nr_replicas = disk_res.nr_replicas;
2699 ret = bch2_extent_update(&trans, iter, &reservation.k_i,
2700 &disk_res, &inode->ei_journal_seq,
2701 0, &i_sectors_delta);
2702 i_sectors_acct(c, inode, "a_res, i_sectors_delta);
2704 bch2_quota_reservation_put(c, inode, "a_res);
2705 bch2_disk_reservation_put(c, &disk_res);
2713 * Do we need to extend the file?
2715 * If we zeroed up to the end of the file, we dropped whatever writes
2716 * were going to write out the current i_size, so we have to extend
2717 * manually even if FL_KEEP_SIZE was set:
2719 if (end >= inode->v.i_size &&
2720 (!(mode & FALLOC_FL_KEEP_SIZE) ||
2721 (mode & FALLOC_FL_ZERO_RANGE))) {
2722 struct btree_iter *inode_iter;
2723 struct bch_inode_unpacked inode_u;
2726 bch2_trans_begin(&trans);
2727 inode_iter = bch2_inode_peek(&trans, &inode_u,
2729 ret = PTR_ERR_OR_ZERO(inode_iter);
2730 } while (ret == -EINTR);
2732 bch2_trans_unlock(&trans);
2738 * Sync existing appends before extending i_size,
2739 * as in bch2_extend():
2741 ret = filemap_write_and_wait_range(mapping,
2742 inode_u.bi_size, S64_MAX);
2746 if (mode & FALLOC_FL_KEEP_SIZE)
2747 end = inode->v.i_size;
2749 i_size_write(&inode->v, end);
2751 mutex_lock(&inode->ei_update_lock);
2752 ret = bch2_write_inode_size(c, inode, end, 0);
2753 mutex_unlock(&inode->ei_update_lock);
2756 bch2_trans_exit(&trans);
2757 bch2_pagecache_block_put(&inode->ei_pagecache_lock);
2758 inode_unlock(&inode->v);
2762 long bch2_fallocate_dispatch(struct file *file, int mode,
2763 loff_t offset, loff_t len)
2765 struct bch_inode_info *inode = file_bch_inode(file);
2766 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2769 if (!percpu_ref_tryget(&c->writes))
2772 if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
2773 ret = bchfs_fallocate(inode, mode, offset, len);
2774 else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
2775 ret = bchfs_fpunch(inode, offset, len);
2776 else if (mode == FALLOC_FL_INSERT_RANGE)
2777 ret = bchfs_fcollapse_finsert(inode, offset, len, true);
2778 else if (mode == FALLOC_FL_COLLAPSE_RANGE)
2779 ret = bchfs_fcollapse_finsert(inode, offset, len, false);
2783 percpu_ref_put(&c->writes);
2788 static void mark_range_unallocated(struct bch_inode_info *inode,
2789 loff_t start, loff_t end)
2791 pgoff_t index = start >> PAGE_SHIFT;
2792 pgoff_t end_index = (end - 1) >> PAGE_SHIFT;
2793 struct pagevec pvec;
2795 pagevec_init(&pvec);
2798 unsigned nr_pages, i, j;
2800 nr_pages = pagevec_lookup_range(&pvec, inode->v.i_mapping,
2805 for (i = 0; i < nr_pages; i++) {
2806 struct page *page = pvec.pages[i];
2807 struct bch_page_state *s;
2810 s = bch2_page_state(page);
2813 spin_lock(&s->lock);
2814 for (j = 0; j < PAGE_SECTORS; j++)
2815 s->s[j].nr_replicas = 0;
2816 spin_unlock(&s->lock);
2821 pagevec_release(&pvec);
2822 } while (index <= end_index);
2825 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
2826 struct file *file_dst, loff_t pos_dst,
2827 loff_t len, unsigned remap_flags)
2829 struct bch_inode_info *src = file_bch_inode(file_src);
2830 struct bch_inode_info *dst = file_bch_inode(file_dst);
2831 struct bch_fs *c = src->v.i_sb->s_fs_info;
2832 s64 i_sectors_delta = 0;
2836 if (!c->opts.reflink)
2839 if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
2842 if (remap_flags & REMAP_FILE_DEDUP)
2845 if ((pos_src & (block_bytes(c) - 1)) ||
2846 (pos_dst & (block_bytes(c) - 1)))
2850 abs(pos_src - pos_dst) < len)
2853 bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);
2855 file_update_time(file_dst);
2857 inode_dio_wait(&src->v);
2858 inode_dio_wait(&dst->v);
2860 ret = generic_remap_file_range_prep(file_src, pos_src,
2863 if (ret < 0 || len == 0)
2866 aligned_len = round_up((u64) len, block_bytes(c));
2868 ret = write_invalidate_inode_pages_range(dst->v.i_mapping,
2869 pos_dst, pos_dst + len - 1);
2873 mark_range_unallocated(src, pos_src, pos_src + aligned_len);
2875 ret = bch2_remap_range(c,
2876 POS(dst->v.i_ino, pos_dst >> 9),
2877 POS(src->v.i_ino, pos_src >> 9),
2879 &dst->ei_journal_seq,
2880 pos_dst + len, &i_sectors_delta);
2885 * due to alignment, we might have remapped slightly more than requsted
2887 ret = min((u64) ret << 9, (u64) len);
2889 /* XXX get a quota reservation */
2890 i_sectors_acct(c, dst, NULL, i_sectors_delta);
2892 spin_lock(&dst->v.i_lock);
2893 if (pos_dst + ret > dst->v.i_size)
2894 i_size_write(&dst->v, pos_dst + ret);
2895 spin_unlock(&dst->v.i_lock);
2897 bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);
2904 static int page_data_offset(struct page *page, unsigned offset)
2906 struct bch_page_state *s = bch2_page_state(page);
2910 for (i = offset >> 9; i < PAGE_SECTORS; i++)
2911 if (s->s[i].state >= SECTOR_DIRTY)
2917 static loff_t bch2_seek_pagecache_data(struct inode *vinode,
2918 loff_t start_offset,
2921 struct address_space *mapping = vinode->i_mapping;
2923 pgoff_t start_index = start_offset >> PAGE_SHIFT;
2924 pgoff_t end_index = end_offset >> PAGE_SHIFT;
2925 pgoff_t index = start_index;
2929 while (index <= end_index) {
2930 if (find_get_pages_range(mapping, &index, end_index, 1, &page)) {
2933 offset = page_data_offset(page,
2934 page->index == start_index
2935 ? start_offset & (PAGE_SIZE - 1)
2938 ret = clamp(((loff_t) page->index << PAGE_SHIFT) +
2940 start_offset, end_offset);
2956 static loff_t bch2_seek_data(struct file *file, u64 offset)
2958 struct bch_inode_info *inode = file_bch_inode(file);
2959 struct bch_fs *c = inode->v.i_sb->s_fs_info;
2960 struct btree_trans trans;
2961 struct btree_iter *iter;
2963 u64 isize, next_data = MAX_LFS_FILESIZE;
2966 isize = i_size_read(&inode->v);
2967 if (offset >= isize)
2970 bch2_trans_init(&trans, c, 0, 0);
2972 for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
2973 POS(inode->v.i_ino, offset >> 9), 0, k, ret) {
2974 if (k.k->p.inode != inode->v.i_ino) {
2976 } else if (bkey_extent_is_data(k.k)) {
2977 next_data = max(offset, bkey_start_offset(k.k) << 9);
2979 } else if (k.k->p.offset >> 9 > isize)
2983 ret = bch2_trans_exit(&trans) ?: ret;
2987 if (next_data > offset)
2988 next_data = bch2_seek_pagecache_data(&inode->v,
2991 if (next_data >= isize)
2994 return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
2997 static int __page_hole_offset(struct page *page, unsigned offset)
2999 struct bch_page_state *s = bch2_page_state(page);
3005 for (i = offset >> 9; i < PAGE_SECTORS; i++)
3006 if (s->s[i].state < SECTOR_DIRTY)
3012 static loff_t page_hole_offset(struct address_space *mapping, loff_t offset)
3014 pgoff_t index = offset >> PAGE_SHIFT;
3019 page = find_lock_entry(mapping, index);
3020 if (!page || xa_is_value(page))
3023 pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1));
3025 ret = ((loff_t) index << PAGE_SHIFT) + pg_offset;
3032 static loff_t bch2_seek_pagecache_hole(struct inode *vinode,
3033 loff_t start_offset,
3036 struct address_space *mapping = vinode->i_mapping;
3037 loff_t offset = start_offset, hole;
3039 while (offset < end_offset) {
3040 hole = page_hole_offset(mapping, offset);
3041 if (hole >= 0 && hole <= end_offset)
3042 return max(start_offset, hole);
3044 offset += PAGE_SIZE;
3045 offset &= PAGE_MASK;
3051 static loff_t bch2_seek_hole(struct file *file, u64 offset)
3053 struct bch_inode_info *inode = file_bch_inode(file);
3054 struct bch_fs *c = inode->v.i_sb->s_fs_info;
3055 struct btree_trans trans;
3056 struct btree_iter *iter;
3058 u64 isize, next_hole = MAX_LFS_FILESIZE;
3061 isize = i_size_read(&inode->v);
3062 if (offset >= isize)
3065 bch2_trans_init(&trans, c, 0, 0);
3067 for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
3068 POS(inode->v.i_ino, offset >> 9),
3069 BTREE_ITER_SLOTS, k, ret) {
3070 if (k.k->p.inode != inode->v.i_ino) {
3071 next_hole = bch2_seek_pagecache_hole(&inode->v,
3072 offset, MAX_LFS_FILESIZE);
3074 } else if (!bkey_extent_is_data(k.k)) {
3075 next_hole = bch2_seek_pagecache_hole(&inode->v,
3076 max(offset, bkey_start_offset(k.k) << 9),
3077 k.k->p.offset << 9);
3079 if (next_hole < k.k->p.offset << 9)
3082 offset = max(offset, bkey_start_offset(k.k) << 9);
3086 ret = bch2_trans_exit(&trans) ?: ret;
3090 if (next_hole > isize)
3093 return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
3096 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
3102 return generic_file_llseek(file, offset, whence);
3104 return bch2_seek_data(file, offset);
3106 return bch2_seek_hole(file, offset);
3112 void bch2_fs_fsio_exit(struct bch_fs *c)
3114 bioset_exit(&c->dio_write_bioset);
3115 bioset_exit(&c->dio_read_bioset);
3116 bioset_exit(&c->writepage_bioset);
3119 int bch2_fs_fsio_init(struct bch_fs *c)
3123 pr_verbose_init(c->opts, "");
3125 if (bioset_init(&c->writepage_bioset,
3126 4, offsetof(struct bch_writepage_io, op.wbio.bio),
3127 BIOSET_NEED_BVECS) ||
3128 bioset_init(&c->dio_read_bioset,
3129 4, offsetof(struct dio_read, rbio.bio),
3130 BIOSET_NEED_BVECS) ||
3131 bioset_init(&c->dio_write_bioset,
3132 4, offsetof(struct dio_write, op.wbio.bio),
3136 pr_verbose_init(c->opts, "ret %i", ret);
3140 #endif /* NO_BCACHEFS_FS */