xref: /linux/fs/bcachefs/fs-io.c (revision 7a4ffec9fd54ea27395e24dff726dbf58e2fe06b)
1 // SPDX-License-Identifier: GPL-2.0
2 #ifndef NO_BCACHEFS_FS
3 
4 #include "bcachefs.h"
5 #include "alloc_foreground.h"
6 #include "bkey_buf.h"
7 #include "btree_update.h"
8 #include "buckets.h"
9 #include "clock.h"
10 #include "error.h"
11 #include "extents.h"
12 #include "extent_update.h"
13 #include "fs.h"
14 #include "fs-io.h"
15 #include "fs-io-buffered.h"
16 #include "fs-io-pagecache.h"
17 #include "fsck.h"
18 #include "inode.h"
19 #include "journal.h"
20 #include "io_misc.h"
21 #include "keylist.h"
22 #include "quota.h"
23 #include "reflink.h"
24 #include "trace.h"
25 
26 #include <linux/aio.h>
27 #include <linux/backing-dev.h>
28 #include <linux/falloc.h>
29 #include <linux/migrate.h>
30 #include <linux/mmu_context.h>
31 #include <linux/pagevec.h>
32 #include <linux/rmap.h>
33 #include <linux/sched/signal.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/uio.h>
36 
37 #include <trace/events/writeback.h>
38 
39 struct nocow_flush {
40 	struct closure	*cl;
41 	struct bch_dev	*ca;
42 	struct bio	bio;
43 };
44 
45 static void nocow_flush_endio(struct bio *_bio)
46 {
47 
48 	struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio);
49 
50 	closure_put(bio->cl);
51 	percpu_ref_put(&bio->ca->io_ref);
52 	bio_put(&bio->bio);
53 }
54 
55 void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
56 					 struct bch_inode_info *inode,
57 					 struct closure *cl)
58 {
59 	struct nocow_flush *bio;
60 	struct bch_dev *ca;
61 	struct bch_devs_mask devs;
62 	unsigned dev;
63 
64 	dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX);
65 	if (dev == BCH_SB_MEMBERS_MAX)
66 		return;
67 
68 	devs = inode->ei_devs_need_flush;
69 	memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
70 
71 	for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
72 		rcu_read_lock();
73 		ca = rcu_dereference(c->devs[dev]);
74 		if (ca && !percpu_ref_tryget(&ca->io_ref))
75 			ca = NULL;
76 		rcu_read_unlock();
77 
78 		if (!ca)
79 			continue;
80 
81 		bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0,
82 						    REQ_OP_WRITE|REQ_PREFLUSH,
83 						    GFP_KERNEL,
84 						    &c->nocow_flush_bioset),
85 				   struct nocow_flush, bio);
86 		bio->cl			= cl;
87 		bio->ca			= ca;
88 		bio->bio.bi_end_io	= nocow_flush_endio;
89 		closure_bio_submit(&bio->bio, cl);
90 	}
91 }
92 
93 static int bch2_inode_flush_nocow_writes(struct bch_fs *c,
94 					 struct bch_inode_info *inode)
95 {
96 	struct closure cl;
97 
98 	closure_init_stack(&cl);
99 	bch2_inode_flush_nocow_writes_async(c, inode, &cl);
100 	closure_sync(&cl);
101 
102 	return 0;
103 }
104 
105 /* i_size updates: */
106 
107 struct inode_new_size {
108 	loff_t		new_size;
109 	u64		now;
110 	unsigned	fields;
111 };
112 
113 static int inode_set_size(struct btree_trans *trans,
114 			  struct bch_inode_info *inode,
115 			  struct bch_inode_unpacked *bi,
116 			  void *p)
117 {
118 	struct inode_new_size *s = p;
119 
120 	bi->bi_size = s->new_size;
121 	if (s->fields & ATTR_ATIME)
122 		bi->bi_atime = s->now;
123 	if (s->fields & ATTR_MTIME)
124 		bi->bi_mtime = s->now;
125 	if (s->fields & ATTR_CTIME)
126 		bi->bi_ctime = s->now;
127 
128 	return 0;
129 }
130 
131 int __must_check bch2_write_inode_size(struct bch_fs *c,
132 				       struct bch_inode_info *inode,
133 				       loff_t new_size, unsigned fields)
134 {
135 	struct inode_new_size s = {
136 		.new_size	= new_size,
137 		.now		= bch2_current_time(c),
138 		.fields		= fields,
139 	};
140 
141 	return bch2_write_inode(c, inode, inode_set_size, &s, fields);
142 }
143 
144 void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
145 			   struct quota_res *quota_res, s64 sectors)
146 {
147 	bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c,
148 				"inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
149 				inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
150 				inode->ei_inode.bi_sectors);
151 	inode->v.i_blocks += sectors;
152 
153 #ifdef CONFIG_BCACHEFS_QUOTA
154 	if (quota_res &&
155 	    !test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) &&
156 	    sectors > 0) {
157 		BUG_ON(sectors > quota_res->sectors);
158 		BUG_ON(sectors > inode->ei_quota_reserved);
159 
160 		quota_res->sectors -= sectors;
161 		inode->ei_quota_reserved -= sectors;
162 	} else {
163 		bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
164 	}
165 #endif
166 }
167 
168 /* fsync: */
169 
170 /*
171  * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
172  * insert trigger: look up the btree inode instead
173  */
174 static int bch2_flush_inode(struct bch_fs *c,
175 			    struct bch_inode_info *inode)
176 {
177 	if (c->opts.journal_flush_disabled)
178 		return 0;
179 
180 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
181 		return -EROFS;
182 
183 	struct bch_inode_unpacked u;
184 	int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
185 		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
186 		  bch2_inode_flush_nocow_writes(c, inode);
187 	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
188 	return ret;
189 }
190 
191 int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
192 {
193 	struct bch_inode_info *inode = file_bch_inode(file);
194 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
195 	int ret, err;
196 
197 	trace_bch2_fsync(file, datasync);
198 
199 	ret = file_write_and_wait_range(file, start, end);
200 	if (ret)
201 		goto out;
202 	ret = sync_inode_metadata(&inode->v, 1);
203 	if (ret)
204 		goto out;
205 	ret = bch2_flush_inode(c, inode);
206 out:
207 	ret = bch2_err_class(ret);
208 	if (ret == -EROFS)
209 		ret = -EIO;
210 
211 	err = file_check_and_advance_wb_err(file);
212 	if (!ret)
213 		ret = err;
214 
215 	return ret;
216 }
217 
218 /* truncate: */
219 
220 static inline int range_has_data(struct bch_fs *c, u32 subvol,
221 				 struct bpos start,
222 				 struct bpos end)
223 {
224 	return bch2_trans_run(c,
225 		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents, start, end,
226 						    subvol, 0, k, ({
227 			bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k);
228 		})));
229 }
230 
231 static int __bch2_truncate_folio(struct bch_inode_info *inode,
232 				 pgoff_t index, loff_t start, loff_t end)
233 {
234 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
235 	struct address_space *mapping = inode->v.i_mapping;
236 	struct bch_folio *s;
237 	unsigned start_offset;
238 	unsigned end_offset;
239 	unsigned i;
240 	struct folio *folio;
241 	s64 i_sectors_delta = 0;
242 	int ret = 0;
243 	u64 end_pos;
244 
245 	folio = filemap_lock_folio(mapping, index);
246 	if (IS_ERR_OR_NULL(folio)) {
247 		/*
248 		 * XXX: we're doing two index lookups when we end up reading the
249 		 * folio
250 		 */
251 		ret = range_has_data(c, inode->ei_inum.subvol,
252 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
253 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
254 		if (ret <= 0)
255 			return ret;
256 
257 		folio = __filemap_get_folio(mapping, index,
258 					    FGP_LOCK|FGP_CREAT, GFP_KERNEL);
259 		if (IS_ERR_OR_NULL(folio)) {
260 			ret = -ENOMEM;
261 			goto out;
262 		}
263 	}
264 
265 	BUG_ON(start	>= folio_end_pos(folio));
266 	BUG_ON(end	<= folio_pos(folio));
267 
268 	start_offset	= max(start, folio_pos(folio)) - folio_pos(folio);
269 	end_offset	= min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
270 
271 	/* Folio boundary? Nothing to do */
272 	if (start_offset == 0 &&
273 	    end_offset == folio_size(folio)) {
274 		ret = 0;
275 		goto unlock;
276 	}
277 
278 	s = bch2_folio_create(folio, 0);
279 	if (!s) {
280 		ret = -ENOMEM;
281 		goto unlock;
282 	}
283 
284 	if (!folio_test_uptodate(folio)) {
285 		ret = bch2_read_single_folio(folio, mapping);
286 		if (ret)
287 			goto unlock;
288 	}
289 
290 	ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
291 	if (ret)
292 		goto unlock;
293 
294 	for (i = round_up(start_offset, block_bytes(c)) >> 9;
295 	     i < round_down(end_offset, block_bytes(c)) >> 9;
296 	     i++) {
297 		s->s[i].nr_replicas	= 0;
298 
299 		i_sectors_delta -= s->s[i].state == SECTOR_dirty;
300 		bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
301 	}
302 
303 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
304 
305 	/*
306 	 * Caller needs to know whether this folio will be written out by
307 	 * writeback - doing an i_size update if necessary - or whether it will
308 	 * be responsible for the i_size update.
309 	 *
310 	 * Note that we shouldn't ever see a folio beyond EOF, but check and
311 	 * warn if so. This has been observed by failure to clean up folios
312 	 * after a short write and there's still a chance reclaim will fix
313 	 * things up.
314 	 */
315 	WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
316 	end_pos = folio_end_pos(folio);
317 	if (inode->v.i_size > folio_pos(folio))
318 		end_pos = min_t(u64, inode->v.i_size, end_pos);
319 	ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
320 
321 	folio_zero_segment(folio, start_offset, end_offset);
322 
323 	/*
324 	 * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
325 	 *
326 	 * XXX: because we aren't currently tracking whether the folio has actual
327 	 * data in it (vs. just 0s, or only partially written) this wrong. ick.
328 	 */
329 	BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
330 
331 	/*
332 	 * This removes any writeable userspace mappings; we need to force
333 	 * .page_mkwrite to be called again before any mmapped writes, to
334 	 * redirty the full page:
335 	 */
336 	folio_mkclean(folio);
337 	filemap_dirty_folio(mapping, folio);
338 unlock:
339 	folio_unlock(folio);
340 	folio_put(folio);
341 out:
342 	return ret;
343 }
344 
345 static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
346 {
347 	return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
348 				     from, ANYSINT_MAX(loff_t));
349 }
350 
351 static int bch2_truncate_folios(struct bch_inode_info *inode,
352 				loff_t start, loff_t end)
353 {
354 	int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
355 					start, end);
356 
357 	if (ret >= 0 &&
358 	    start >> PAGE_SHIFT != end >> PAGE_SHIFT)
359 		ret = __bch2_truncate_folio(inode,
360 					(end - 1) >> PAGE_SHIFT,
361 					start, end);
362 	return ret;
363 }
364 
365 static int bch2_extend(struct mnt_idmap *idmap,
366 		       struct bch_inode_info *inode,
367 		       struct bch_inode_unpacked *inode_u,
368 		       struct iattr *iattr)
369 {
370 	struct address_space *mapping = inode->v.i_mapping;
371 	int ret;
372 
373 	/*
374 	 * sync appends:
375 	 *
376 	 * this has to be done _before_ extending i_size:
377 	 */
378 	ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
379 	if (ret)
380 		return ret;
381 
382 	truncate_setsize(&inode->v, iattr->ia_size);
383 
384 	return bch2_setattr_nonsize(idmap, inode, iattr);
385 }
386 
387 int bchfs_truncate(struct mnt_idmap *idmap,
388 		  struct bch_inode_info *inode, struct iattr *iattr)
389 {
390 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
391 	struct address_space *mapping = inode->v.i_mapping;
392 	struct bch_inode_unpacked inode_u;
393 	s64 i_sectors_delta = 0;
394 	int ret = 0;
395 
396 	/*
397 	 * If the truncate call with change the size of the file, the
398 	 * cmtimes should be updated. If the size will not change, we
399 	 * do not need to update the cmtimes.
400 	 */
401 	if (iattr->ia_size != inode->v.i_size) {
402 		if (!(iattr->ia_valid & ATTR_MTIME))
403 			ktime_get_coarse_real_ts64(&iattr->ia_mtime);
404 		if (!(iattr->ia_valid & ATTR_CTIME))
405 			ktime_get_coarse_real_ts64(&iattr->ia_ctime);
406 		iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
407 	}
408 
409 	inode_dio_wait(&inode->v);
410 	bch2_pagecache_block_get(inode);
411 
412 	ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
413 	if (ret)
414 		goto err;
415 
416 	/*
417 	 * check this before next assertion; on filesystem error our normal
418 	 * invariants are a bit broken (truncate has to truncate the page cache
419 	 * before the inode).
420 	 */
421 	ret = bch2_journal_error(&c->journal);
422 	if (ret)
423 		goto err;
424 
425 	WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
426 		  inode->v.i_size < inode_u.bi_size,
427 		  "truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
428 		  (u64) inode->v.i_size, inode_u.bi_size);
429 
430 	if (iattr->ia_size > inode->v.i_size) {
431 		ret = bch2_extend(idmap, inode, &inode_u, iattr);
432 		goto err;
433 	}
434 
435 	iattr->ia_valid &= ~ATTR_SIZE;
436 
437 	ret = bch2_truncate_folio(inode, iattr->ia_size);
438 	if (unlikely(ret < 0))
439 		goto err;
440 
441 	truncate_setsize(&inode->v, iattr->ia_size);
442 
443 	/*
444 	 * When extending, we're going to write the new i_size to disk
445 	 * immediately so we need to flush anything above the current on disk
446 	 * i_size first:
447 	 *
448 	 * Also, when extending we need to flush the page that i_size currently
449 	 * straddles - if it's mapped to userspace, we need to ensure that
450 	 * userspace has to redirty it and call .mkwrite -> set_page_dirty
451 	 * again to allocate the part of the page that was extended.
452 	 */
453 	if (iattr->ia_size > inode_u.bi_size)
454 		ret = filemap_write_and_wait_range(mapping,
455 				inode_u.bi_size,
456 				iattr->ia_size - 1);
457 	else if (iattr->ia_size & (PAGE_SIZE - 1))
458 		ret = filemap_write_and_wait_range(mapping,
459 				round_down(iattr->ia_size, PAGE_SIZE),
460 				iattr->ia_size - 1);
461 	if (ret)
462 		goto err;
463 
464 	ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta);
465 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
466 
467 	if (unlikely(ret)) {
468 		/*
469 		 * If we error here, VFS caches are now inconsistent with btree
470 		 */
471 		set_bit(EI_INODE_ERROR, &inode->ei_flags);
472 		goto err;
473 	}
474 
475 	bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
476 				!bch2_journal_error(&c->journal), c,
477 				"inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
478 				inode->v.i_ino, (u64) inode->v.i_blocks,
479 				inode->ei_inode.bi_sectors);
480 
481 	ret = bch2_setattr_nonsize(idmap, inode, iattr);
482 err:
483 	bch2_pagecache_block_put(inode);
484 	return bch2_err_class(ret);
485 }
486 
487 /* fallocate: */
488 
489 static int inode_update_times_fn(struct btree_trans *trans,
490 				 struct bch_inode_info *inode,
491 				 struct bch_inode_unpacked *bi, void *p)
492 {
493 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
494 
495 	bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
496 	return 0;
497 }
498 
499 static noinline long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
500 {
501 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
502 	u64 end		= offset + len;
503 	u64 block_start	= round_up(offset, block_bytes(c));
504 	u64 block_end	= round_down(end, block_bytes(c));
505 	bool truncated_last_page;
506 	int ret = 0;
507 
508 	ret = bch2_truncate_folios(inode, offset, end);
509 	if (unlikely(ret < 0))
510 		goto err;
511 
512 	truncated_last_page = ret;
513 
514 	truncate_pagecache_range(&inode->v, offset, end - 1);
515 
516 	if (block_start < block_end) {
517 		s64 i_sectors_delta = 0;
518 
519 		ret = bch2_fpunch(c, inode_inum(inode),
520 				  block_start >> 9, block_end >> 9,
521 				  &i_sectors_delta);
522 		bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
523 	}
524 
525 	mutex_lock(&inode->ei_update_lock);
526 	if (end >= inode->v.i_size && !truncated_last_page) {
527 		ret = bch2_write_inode_size(c, inode, inode->v.i_size,
528 					    ATTR_MTIME|ATTR_CTIME);
529 	} else {
530 		ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
531 				       ATTR_MTIME|ATTR_CTIME);
532 	}
533 	mutex_unlock(&inode->ei_update_lock);
534 err:
535 	return ret;
536 }
537 
538 static noinline long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
539 				   loff_t offset, loff_t len,
540 				   bool insert)
541 {
542 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
543 	struct address_space *mapping = inode->v.i_mapping;
544 	s64 i_sectors_delta = 0;
545 	int ret = 0;
546 
547 	if ((offset | len) & (block_bytes(c) - 1))
548 		return -EINVAL;
549 
550 	if (insert) {
551 		if (offset >= inode->v.i_size)
552 			return -EINVAL;
553 	} else {
554 		if (offset + len >= inode->v.i_size)
555 			return -EINVAL;
556 	}
557 
558 	ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
559 	if (ret)
560 		return ret;
561 
562 	if (insert)
563 		i_size_write(&inode->v, inode->v.i_size + len);
564 
565 	ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9,
566 				     insert, &i_sectors_delta);
567 	if (!ret && !insert)
568 		i_size_write(&inode->v, inode->v.i_size - len);
569 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
570 
571 	return ret;
572 }
573 
574 static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
575 			     u64 start_sector, u64 end_sector)
576 {
577 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
578 	struct btree_trans *trans = bch2_trans_get(c);
579 	struct btree_iter iter;
580 	struct bpos end_pos = POS(inode->v.i_ino, end_sector);
581 	struct bch_io_opts opts;
582 	int ret = 0;
583 
584 	bch2_inode_opts_get(&opts, c, &inode->ei_inode);
585 
586 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
587 			POS(inode->v.i_ino, start_sector),
588 			BTREE_ITER_slots|BTREE_ITER_intent);
589 
590 	while (!ret && bkey_lt(iter.pos, end_pos)) {
591 		s64 i_sectors_delta = 0;
592 		struct quota_res quota_res = { 0 };
593 		struct bkey_s_c k;
594 		unsigned sectors;
595 		bool is_allocation;
596 		u64 hole_start, hole_end;
597 		u32 snapshot;
598 
599 		bch2_trans_begin(trans);
600 
601 		ret = bch2_subvolume_get_snapshot(trans,
602 					inode->ei_inum.subvol, &snapshot);
603 		if (ret)
604 			goto bkey_err;
605 
606 		bch2_btree_iter_set_snapshot(&iter, snapshot);
607 
608 		k = bch2_btree_iter_peek_slot(&iter);
609 		if ((ret = bkey_err(k)))
610 			goto bkey_err;
611 
612 		hole_start	= iter.pos.offset;
613 		hole_end	= bpos_min(k.k->p, end_pos).offset;
614 		is_allocation	= bkey_extent_is_allocation(k.k);
615 
616 		/* already reserved */
617 		if (bkey_extent_is_reservation(k) &&
618 		    bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
619 			bch2_btree_iter_advance(&iter);
620 			continue;
621 		}
622 
623 		if (bkey_extent_is_data(k.k) &&
624 		    !(mode & FALLOC_FL_ZERO_RANGE)) {
625 			bch2_btree_iter_advance(&iter);
626 			continue;
627 		}
628 
629 		if (!(mode & FALLOC_FL_ZERO_RANGE)) {
630 			/*
631 			 * Lock ordering - can't be holding btree locks while
632 			 * blocking on a folio lock:
633 			 */
634 			if (bch2_clamp_data_hole(&inode->v,
635 						 &hole_start,
636 						 &hole_end,
637 						 opts.data_replicas, true))
638 				ret = drop_locks_do(trans,
639 					(bch2_clamp_data_hole(&inode->v,
640 							      &hole_start,
641 							      &hole_end,
642 							      opts.data_replicas, false), 0));
643 			bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
644 
645 			if (ret)
646 				goto bkey_err;
647 
648 			if (hole_start == hole_end)
649 				continue;
650 		}
651 
652 		sectors	= hole_end - hole_start;
653 
654 		if (!is_allocation) {
655 			ret = bch2_quota_reservation_add(c, inode,
656 					&quota_res, sectors, true);
657 			if (unlikely(ret))
658 				goto bkey_err;
659 		}
660 
661 		ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter,
662 					    sectors, opts, &i_sectors_delta,
663 					    writepoint_hashed((unsigned long) current));
664 		if (ret)
665 			goto bkey_err;
666 
667 		bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
668 
669 		if (bch2_mark_pagecache_reserved(inode, &hole_start,
670 						 iter.pos.offset, true))
671 			drop_locks_do(trans,
672 				bch2_mark_pagecache_reserved(inode, &hole_start,
673 							     iter.pos.offset, false));
674 bkey_err:
675 		bch2_quota_reservation_put(c, inode, &quota_res);
676 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
677 			ret = 0;
678 	}
679 
680 	if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
681 		struct quota_res quota_res = { 0 };
682 		s64 i_sectors_delta = 0;
683 
684 		bch2_fpunch_at(trans, &iter, inode_inum(inode),
685 			       end_sector, &i_sectors_delta);
686 		bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
687 		bch2_quota_reservation_put(c, inode, &quota_res);
688 	}
689 
690 	bch2_trans_iter_exit(trans, &iter);
691 	bch2_trans_put(trans);
692 	return ret;
693 }
694 
695 static noinline long bchfs_fallocate(struct bch_inode_info *inode, int mode,
696 			    loff_t offset, loff_t len)
697 {
698 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
699 	u64 end		= offset + len;
700 	u64 block_start	= round_down(offset,	block_bytes(c));
701 	u64 block_end	= round_up(end,		block_bytes(c));
702 	bool truncated_last_page = false;
703 	int ret, ret2 = 0;
704 
705 	if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
706 		ret = inode_newsize_ok(&inode->v, end);
707 		if (ret)
708 			return ret;
709 	}
710 
711 	if (mode & FALLOC_FL_ZERO_RANGE) {
712 		ret = bch2_truncate_folios(inode, offset, end);
713 		if (unlikely(ret < 0))
714 			return ret;
715 
716 		truncated_last_page = ret;
717 
718 		truncate_pagecache_range(&inode->v, offset, end - 1);
719 
720 		block_start	= round_up(offset,	block_bytes(c));
721 		block_end	= round_down(end,	block_bytes(c));
722 	}
723 
724 	ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
725 
726 	/*
727 	 * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
728 	 * so that the VFS cache i_size is consistent with the btree i_size:
729 	 */
730 	if (ret &&
731 	    !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
732 		return ret;
733 
734 	if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
735 		end = inode->v.i_size;
736 
737 	if (end >= inode->v.i_size &&
738 	    (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
739 	     !(mode & FALLOC_FL_KEEP_SIZE))) {
740 		spin_lock(&inode->v.i_lock);
741 		i_size_write(&inode->v, end);
742 		spin_unlock(&inode->v.i_lock);
743 
744 		mutex_lock(&inode->ei_update_lock);
745 		ret2 = bch2_write_inode_size(c, inode, end, 0);
746 		mutex_unlock(&inode->ei_update_lock);
747 	}
748 
749 	return ret ?: ret2;
750 }
751 
752 long bch2_fallocate_dispatch(struct file *file, int mode,
753 			     loff_t offset, loff_t len)
754 {
755 	struct bch_inode_info *inode = file_bch_inode(file);
756 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
757 	long ret;
758 
759 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
760 		return -EROFS;
761 
762 	inode_lock(&inode->v);
763 	inode_dio_wait(&inode->v);
764 	bch2_pagecache_block_get(inode);
765 
766 	ret = file_modified(file);
767 	if (ret)
768 		goto err;
769 
770 	if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
771 		ret = bchfs_fallocate(inode, mode, offset, len);
772 	else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
773 		ret = bchfs_fpunch(inode, offset, len);
774 	else if (mode == FALLOC_FL_INSERT_RANGE)
775 		ret = bchfs_fcollapse_finsert(inode, offset, len, true);
776 	else if (mode == FALLOC_FL_COLLAPSE_RANGE)
777 		ret = bchfs_fcollapse_finsert(inode, offset, len, false);
778 	else
779 		ret = -EOPNOTSUPP;
780 err:
781 	bch2_pagecache_block_put(inode);
782 	inode_unlock(&inode->v);
783 	bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
784 
785 	return bch2_err_class(ret);
786 }
787 
788 /*
789  * Take a quota reservation for unallocated blocks in a given file range
790  * Does not check pagecache
791  */
792 static int quota_reserve_range(struct bch_inode_info *inode,
793 			       struct quota_res *res,
794 			       u64 start, u64 end)
795 {
796 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
797 	u64 sectors = end - start;
798 
799 	int ret = bch2_trans_run(c,
800 		for_each_btree_key_in_subvolume_upto(trans, iter,
801 				BTREE_ID_extents,
802 				POS(inode->v.i_ino, start),
803 				POS(inode->v.i_ino, end - 1),
804 				inode->ei_inum.subvol, 0, k, ({
805 			if (bkey_extent_is_allocation(k.k)) {
806 				u64 s = min(end, k.k->p.offset) -
807 					max(start, bkey_start_offset(k.k));
808 				BUG_ON(s > sectors);
809 				sectors -= s;
810 			}
811 
812 			0;
813 		})));
814 
815 	return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true);
816 }
817 
818 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
819 			     struct file *file_dst, loff_t pos_dst,
820 			     loff_t len, unsigned remap_flags)
821 {
822 	struct bch_inode_info *src = file_bch_inode(file_src);
823 	struct bch_inode_info *dst = file_bch_inode(file_dst);
824 	struct bch_fs *c = src->v.i_sb->s_fs_info;
825 	struct quota_res quota_res = { 0 };
826 	s64 i_sectors_delta = 0;
827 	u64 aligned_len;
828 	loff_t ret = 0;
829 
830 	if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
831 		return -EINVAL;
832 
833 	if ((pos_src & (block_bytes(c) - 1)) ||
834 	    (pos_dst & (block_bytes(c) - 1)))
835 		return -EINVAL;
836 
837 	if (src == dst &&
838 	    abs(pos_src - pos_dst) < len)
839 		return -EINVAL;
840 
841 	lock_two_nondirectories(&src->v, &dst->v);
842 	bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
843 
844 	inode_dio_wait(&src->v);
845 	inode_dio_wait(&dst->v);
846 
847 	ret = generic_remap_file_range_prep(file_src, pos_src,
848 					    file_dst, pos_dst,
849 					    &len, remap_flags);
850 	if (ret < 0 || len == 0)
851 		goto err;
852 
853 	aligned_len = round_up((u64) len, block_bytes(c));
854 
855 	ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
856 				pos_dst, pos_dst + len - 1);
857 	if (ret)
858 		goto err;
859 
860 	ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
861 				  (pos_dst + aligned_len) >> 9);
862 	if (ret)
863 		goto err;
864 
865 	if (!(remap_flags & REMAP_FILE_DEDUP))
866 		file_update_time(file_dst);
867 
868 	bch2_mark_pagecache_unallocated(src, pos_src >> 9,
869 				   (pos_src + aligned_len) >> 9);
870 
871 	ret = bch2_remap_range(c,
872 			       inode_inum(dst), pos_dst >> 9,
873 			       inode_inum(src), pos_src >> 9,
874 			       aligned_len >> 9,
875 			       pos_dst + len, &i_sectors_delta);
876 	if (ret < 0)
877 		goto err;
878 
879 	/*
880 	 * due to alignment, we might have remapped slightly more than requsted
881 	 */
882 	ret = min((u64) ret << 9, (u64) len);
883 
884 	bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
885 
886 	spin_lock(&dst->v.i_lock);
887 	if (pos_dst + ret > dst->v.i_size)
888 		i_size_write(&dst->v, pos_dst + ret);
889 	spin_unlock(&dst->v.i_lock);
890 
891 	if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
892 	    IS_SYNC(file_inode(file_dst)))
893 		ret = bch2_flush_inode(c, dst);
894 err:
895 	bch2_quota_reservation_put(c, dst, &quota_res);
896 	bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
897 	unlock_two_nondirectories(&src->v, &dst->v);
898 
899 	return bch2_err_class(ret);
900 }
901 
902 /* fseek: */
903 
904 static loff_t bch2_seek_data(struct file *file, u64 offset)
905 {
906 	struct bch_inode_info *inode = file_bch_inode(file);
907 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
908 	subvol_inum inum = inode_inum(inode);
909 	u64 isize, next_data = MAX_LFS_FILESIZE;
910 
911 	isize = i_size_read(&inode->v);
912 	if (offset >= isize)
913 		return -ENXIO;
914 
915 	int ret = bch2_trans_run(c,
916 		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
917 				   POS(inode->v.i_ino, offset >> 9),
918 				   POS(inode->v.i_ino, U64_MAX),
919 				   inum.subvol, 0, k, ({
920 			if (bkey_extent_is_data(k.k)) {
921 				next_data = max(offset, bkey_start_offset(k.k) << 9);
922 				break;
923 			} else if (k.k->p.offset >> 9 > isize)
924 				break;
925 			0;
926 		})));
927 	if (ret)
928 		return ret;
929 
930 	if (next_data > offset)
931 		next_data = bch2_seek_pagecache_data(&inode->v,
932 					offset, next_data, 0, false);
933 
934 	if (next_data >= isize)
935 		return -ENXIO;
936 
937 	return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
938 }
939 
940 static loff_t bch2_seek_hole(struct file *file, u64 offset)
941 {
942 	struct bch_inode_info *inode = file_bch_inode(file);
943 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
944 	subvol_inum inum = inode_inum(inode);
945 	u64 isize, next_hole = MAX_LFS_FILESIZE;
946 
947 	isize = i_size_read(&inode->v);
948 	if (offset >= isize)
949 		return -ENXIO;
950 
951 	int ret = bch2_trans_run(c,
952 		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
953 				   POS(inode->v.i_ino, offset >> 9),
954 				   POS(inode->v.i_ino, U64_MAX),
955 				   inum.subvol, BTREE_ITER_slots, k, ({
956 			if (k.k->p.inode != inode->v.i_ino) {
957 				next_hole = bch2_seek_pagecache_hole(&inode->v,
958 						offset, MAX_LFS_FILESIZE, 0, false);
959 				break;
960 			} else if (!bkey_extent_is_data(k.k)) {
961 				next_hole = bch2_seek_pagecache_hole(&inode->v,
962 						max(offset, bkey_start_offset(k.k) << 9),
963 						k.k->p.offset << 9, 0, false);
964 
965 				if (next_hole < k.k->p.offset << 9)
966 					break;
967 			} else {
968 				offset = max(offset, bkey_start_offset(k.k) << 9);
969 			}
970 			0;
971 		})));
972 	if (ret)
973 		return ret;
974 
975 	if (next_hole > isize)
976 		next_hole = isize;
977 
978 	return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
979 }
980 
981 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
982 {
983 	loff_t ret;
984 
985 	switch (whence) {
986 	case SEEK_SET:
987 	case SEEK_CUR:
988 	case SEEK_END:
989 		ret = generic_file_llseek(file, offset, whence);
990 		break;
991 	case SEEK_DATA:
992 		ret = bch2_seek_data(file, offset);
993 		break;
994 	case SEEK_HOLE:
995 		ret = bch2_seek_hole(file, offset);
996 		break;
997 	default:
998 		ret = -EINVAL;
999 		break;
1000 	}
1001 
1002 	return bch2_err_class(ret);
1003 }
1004 
1005 void bch2_fs_fsio_exit(struct bch_fs *c)
1006 {
1007 	bioset_exit(&c->nocow_flush_bioset);
1008 }
1009 
1010 int bch2_fs_fsio_init(struct bch_fs *c)
1011 {
1012 	if (bioset_init(&c->nocow_flush_bioset,
1013 			1, offsetof(struct nocow_flush, bio), 0))
1014 		return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
1015 
1016 	return 0;
1017 }
1018 
1019 #endif /* NO_BCACHEFS_FS */
1020