xref: /linux/fs/bcachefs/fs-io.c (revision dd83757f6e686a2188997cb58b5975f744bb7786)
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 
nocow_flush_endio(struct bio * _bio)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 
bch2_inode_flush_nocow_writes_async(struct bch_fs * c,struct bch_inode_info * inode,struct closure * cl)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 
bch2_inode_flush_nocow_writes(struct bch_fs * c,struct bch_inode_info * inode)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 
inode_set_size(struct btree_trans * trans,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,void * p)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 
bch2_write_inode_size(struct bch_fs * c,struct bch_inode_info * inode,loff_t new_size,unsigned fields)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 
__bch2_i_sectors_acct(struct bch_fs * c,struct bch_inode_info * inode,struct quota_res * quota_res,s64 sectors)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 
bch2_get_inode_journal_seq_trans(struct btree_trans * trans,subvol_inum inum,u64 * seq)170 static int bch2_get_inode_journal_seq_trans(struct btree_trans *trans, subvol_inum inum,
171 					    u64 *seq)
172 {
173 	struct printbuf buf = PRINTBUF;
174 	struct bch_inode_unpacked u;
175 	struct btree_iter iter;
176 	int ret = bch2_inode_peek(trans, &iter, &u, inum, 0);
177 	if (ret)
178 		return ret;
179 
180 	u64 cur_seq = journal_cur_seq(&trans->c->journal);
181 	*seq = min(cur_seq, u.bi_journal_seq);
182 
183 	if (fsck_err_on(u.bi_journal_seq > cur_seq,
184 			trans, inode_journal_seq_in_future,
185 			"inode journal seq in future (currently at %llu)\n%s",
186 			cur_seq,
187 			(bch2_inode_unpacked_to_text(&buf, &u),
188 			buf.buf))) {
189 		u.bi_journal_seq = cur_seq;
190 		ret = bch2_inode_write(trans, &iter, &u);
191 	}
192 fsck_err:
193 	bch2_trans_iter_exit(trans, &iter);
194 	printbuf_exit(&buf);
195 	return ret;
196 }
197 
198 /*
199  * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
200  * insert trigger: look up the btree inode instead
201  */
bch2_flush_inode(struct bch_fs * c,struct bch_inode_info * inode)202 static int bch2_flush_inode(struct bch_fs *c,
203 			    struct bch_inode_info *inode)
204 {
205 	if (c->opts.journal_flush_disabled)
206 		return 0;
207 
208 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
209 		return -EROFS;
210 
211 	u64 seq;
212 	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
213 			bch2_get_inode_journal_seq_trans(trans, inode_inum(inode), &seq)) ?:
214 		  bch2_journal_flush_seq(&c->journal, seq, TASK_INTERRUPTIBLE) ?:
215 		  bch2_inode_flush_nocow_writes(c, inode);
216 	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
217 	return ret;
218 }
219 
bch2_fsync(struct file * file,loff_t start,loff_t end,int datasync)220 int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
221 {
222 	struct bch_inode_info *inode = file_bch_inode(file);
223 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
224 	int ret, err;
225 
226 	trace_bch2_fsync(file, datasync);
227 
228 	ret = file_write_and_wait_range(file, start, end);
229 	if (ret)
230 		goto out;
231 	ret = sync_inode_metadata(&inode->v, 1);
232 	if (ret)
233 		goto out;
234 	ret = bch2_flush_inode(c, inode);
235 out:
236 	ret = bch2_err_class(ret);
237 	if (ret == -EROFS)
238 		ret = -EIO;
239 
240 	err = file_check_and_advance_wb_err(file);
241 	if (!ret)
242 		ret = err;
243 
244 	return ret;
245 }
246 
247 /* truncate: */
248 
range_has_data(struct bch_fs * c,u32 subvol,struct bpos start,struct bpos end)249 static inline int range_has_data(struct bch_fs *c, u32 subvol,
250 				 struct bpos start,
251 				 struct bpos end)
252 {
253 	return bch2_trans_run(c,
254 		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents, start, end,
255 						    subvol, 0, k, ({
256 			bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k);
257 		})));
258 }
259 
__bch2_truncate_folio(struct bch_inode_info * inode,pgoff_t index,loff_t start,loff_t end)260 static int __bch2_truncate_folio(struct bch_inode_info *inode,
261 				 pgoff_t index, loff_t start, loff_t end)
262 {
263 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
264 	struct address_space *mapping = inode->v.i_mapping;
265 	struct bch_folio *s;
266 	unsigned start_offset;
267 	unsigned end_offset;
268 	unsigned i;
269 	struct folio *folio;
270 	s64 i_sectors_delta = 0;
271 	int ret = 0;
272 	u64 end_pos;
273 
274 	folio = filemap_lock_folio(mapping, index);
275 	if (IS_ERR_OR_NULL(folio)) {
276 		/*
277 		 * XXX: we're doing two index lookups when we end up reading the
278 		 * folio
279 		 */
280 		ret = range_has_data(c, inode->ei_inum.subvol,
281 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
282 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
283 		if (ret <= 0)
284 			return ret;
285 
286 		folio = __filemap_get_folio(mapping, index,
287 					    FGP_LOCK|FGP_CREAT, GFP_KERNEL);
288 		if (IS_ERR(folio)) {
289 			ret = -ENOMEM;
290 			goto out;
291 		}
292 	}
293 
294 	BUG_ON(start	>= folio_end_pos(folio));
295 	BUG_ON(end	<= folio_pos(folio));
296 
297 	start_offset	= max(start, folio_pos(folio)) - folio_pos(folio);
298 	end_offset	= min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
299 
300 	/* Folio boundary? Nothing to do */
301 	if (start_offset == 0 &&
302 	    end_offset == folio_size(folio)) {
303 		ret = 0;
304 		goto unlock;
305 	}
306 
307 	s = bch2_folio_create(folio, 0);
308 	if (!s) {
309 		ret = -ENOMEM;
310 		goto unlock;
311 	}
312 
313 	if (!folio_test_uptodate(folio)) {
314 		ret = bch2_read_single_folio(folio, mapping);
315 		if (ret)
316 			goto unlock;
317 	}
318 
319 	ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
320 	if (ret)
321 		goto unlock;
322 
323 	for (i = round_up(start_offset, block_bytes(c)) >> 9;
324 	     i < round_down(end_offset, block_bytes(c)) >> 9;
325 	     i++) {
326 		s->s[i].nr_replicas	= 0;
327 
328 		i_sectors_delta -= s->s[i].state == SECTOR_dirty;
329 		bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
330 	}
331 
332 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
333 
334 	/*
335 	 * Caller needs to know whether this folio will be written out by
336 	 * writeback - doing an i_size update if necessary - or whether it will
337 	 * be responsible for the i_size update.
338 	 *
339 	 * Note that we shouldn't ever see a folio beyond EOF, but check and
340 	 * warn if so. This has been observed by failure to clean up folios
341 	 * after a short write and there's still a chance reclaim will fix
342 	 * things up.
343 	 */
344 	WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
345 	end_pos = folio_end_pos(folio);
346 	if (inode->v.i_size > folio_pos(folio))
347 		end_pos = min_t(u64, inode->v.i_size, end_pos);
348 	ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
349 
350 	folio_zero_segment(folio, start_offset, end_offset);
351 
352 	/*
353 	 * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
354 	 *
355 	 * XXX: because we aren't currently tracking whether the folio has actual
356 	 * data in it (vs. just 0s, or only partially written) this wrong. ick.
357 	 */
358 	BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
359 
360 	/*
361 	 * This removes any writeable userspace mappings; we need to force
362 	 * .page_mkwrite to be called again before any mmapped writes, to
363 	 * redirty the full page:
364 	 */
365 	folio_mkclean(folio);
366 	filemap_dirty_folio(mapping, folio);
367 unlock:
368 	folio_unlock(folio);
369 	folio_put(folio);
370 out:
371 	return ret;
372 }
373 
bch2_truncate_folio(struct bch_inode_info * inode,loff_t from)374 static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
375 {
376 	return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
377 				     from, ANYSINT_MAX(loff_t));
378 }
379 
bch2_truncate_folios(struct bch_inode_info * inode,loff_t start,loff_t end)380 static int bch2_truncate_folios(struct bch_inode_info *inode,
381 				loff_t start, loff_t end)
382 {
383 	int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
384 					start, end);
385 
386 	if (ret >= 0 &&
387 	    start >> PAGE_SHIFT != end >> PAGE_SHIFT)
388 		ret = __bch2_truncate_folio(inode,
389 					(end - 1) >> PAGE_SHIFT,
390 					start, end);
391 	return ret;
392 }
393 
bch2_extend(struct mnt_idmap * idmap,struct bch_inode_info * inode,struct bch_inode_unpacked * inode_u,struct iattr * iattr)394 static int bch2_extend(struct mnt_idmap *idmap,
395 		       struct bch_inode_info *inode,
396 		       struct bch_inode_unpacked *inode_u,
397 		       struct iattr *iattr)
398 {
399 	struct address_space *mapping = inode->v.i_mapping;
400 	int ret;
401 
402 	/*
403 	 * sync appends:
404 	 *
405 	 * this has to be done _before_ extending i_size:
406 	 */
407 	ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
408 	if (ret)
409 		return ret;
410 
411 	truncate_setsize(&inode->v, iattr->ia_size);
412 
413 	return bch2_setattr_nonsize(idmap, inode, iattr);
414 }
415 
bchfs_truncate(struct mnt_idmap * idmap,struct bch_inode_info * inode,struct iattr * iattr)416 int bchfs_truncate(struct mnt_idmap *idmap,
417 		  struct bch_inode_info *inode, struct iattr *iattr)
418 {
419 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
420 	struct address_space *mapping = inode->v.i_mapping;
421 	struct bch_inode_unpacked inode_u;
422 	s64 i_sectors_delta = 0;
423 	int ret = 0;
424 
425 	/*
426 	 * If the truncate call with change the size of the file, the
427 	 * cmtimes should be updated. If the size will not change, we
428 	 * do not need to update the cmtimes.
429 	 */
430 	if (iattr->ia_size != inode->v.i_size) {
431 		if (!(iattr->ia_valid & ATTR_MTIME))
432 			ktime_get_coarse_real_ts64(&iattr->ia_mtime);
433 		if (!(iattr->ia_valid & ATTR_CTIME))
434 			ktime_get_coarse_real_ts64(&iattr->ia_ctime);
435 		iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
436 	}
437 
438 	inode_dio_wait(&inode->v);
439 	bch2_pagecache_block_get(inode);
440 
441 	ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
442 	if (ret)
443 		goto err;
444 
445 	/*
446 	 * check this before next assertion; on filesystem error our normal
447 	 * invariants are a bit broken (truncate has to truncate the page cache
448 	 * before the inode).
449 	 */
450 	ret = bch2_journal_error(&c->journal);
451 	if (ret)
452 		goto err;
453 
454 	WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
455 		  inode->v.i_size < inode_u.bi_size,
456 		  "truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
457 		  (u64) inode->v.i_size, inode_u.bi_size);
458 
459 	if (iattr->ia_size > inode->v.i_size) {
460 		ret = bch2_extend(idmap, inode, &inode_u, iattr);
461 		goto err;
462 	}
463 
464 	iattr->ia_valid &= ~ATTR_SIZE;
465 
466 	ret = bch2_truncate_folio(inode, iattr->ia_size);
467 	if (unlikely(ret < 0))
468 		goto err;
469 	ret = 0;
470 
471 	truncate_setsize(&inode->v, iattr->ia_size);
472 
473 	/*
474 	 * When extending, we're going to write the new i_size to disk
475 	 * immediately so we need to flush anything above the current on disk
476 	 * i_size first:
477 	 *
478 	 * Also, when extending we need to flush the page that i_size currently
479 	 * straddles - if it's mapped to userspace, we need to ensure that
480 	 * userspace has to redirty it and call .mkwrite -> set_page_dirty
481 	 * again to allocate the part of the page that was extended.
482 	 */
483 	if (iattr->ia_size > inode_u.bi_size)
484 		ret = filemap_write_and_wait_range(mapping,
485 				inode_u.bi_size,
486 				iattr->ia_size - 1);
487 	else if (iattr->ia_size & (PAGE_SIZE - 1))
488 		ret = filemap_write_and_wait_range(mapping,
489 				round_down(iattr->ia_size, PAGE_SIZE),
490 				iattr->ia_size - 1);
491 	if (ret)
492 		goto err;
493 
494 	ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta);
495 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
496 
497 	if (unlikely(ret)) {
498 		/*
499 		 * If we error here, VFS caches are now inconsistent with btree
500 		 */
501 		set_bit(EI_INODE_ERROR, &inode->ei_flags);
502 		goto err;
503 	}
504 
505 	bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
506 				!bch2_journal_error(&c->journal), c,
507 				"inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
508 				inode->v.i_ino, (u64) inode->v.i_blocks,
509 				inode->ei_inode.bi_sectors);
510 
511 	ret = bch2_setattr_nonsize(idmap, inode, iattr);
512 err:
513 	bch2_pagecache_block_put(inode);
514 	return bch2_err_class(ret);
515 }
516 
517 /* fallocate: */
518 
inode_update_times_fn(struct btree_trans * trans,struct bch_inode_info * inode,struct bch_inode_unpacked * bi,void * p)519 static int inode_update_times_fn(struct btree_trans *trans,
520 				 struct bch_inode_info *inode,
521 				 struct bch_inode_unpacked *bi, void *p)
522 {
523 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
524 
525 	bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
526 	return 0;
527 }
528 
bchfs_fpunch(struct bch_inode_info * inode,loff_t offset,loff_t len)529 static noinline long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
530 {
531 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
532 	u64 end		= offset + len;
533 	u64 block_start	= round_up(offset, block_bytes(c));
534 	u64 block_end	= round_down(end, block_bytes(c));
535 	bool truncated_last_page;
536 	int ret = 0;
537 
538 	ret = bch2_truncate_folios(inode, offset, end);
539 	if (unlikely(ret < 0))
540 		goto err;
541 
542 	truncated_last_page = ret;
543 
544 	truncate_pagecache_range(&inode->v, offset, end - 1);
545 
546 	if (block_start < block_end) {
547 		s64 i_sectors_delta = 0;
548 
549 		ret = bch2_fpunch(c, inode_inum(inode),
550 				  block_start >> 9, block_end >> 9,
551 				  &i_sectors_delta);
552 		bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
553 	}
554 
555 	mutex_lock(&inode->ei_update_lock);
556 	if (end >= inode->v.i_size && !truncated_last_page) {
557 		ret = bch2_write_inode_size(c, inode, inode->v.i_size,
558 					    ATTR_MTIME|ATTR_CTIME);
559 	} else {
560 		ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
561 				       ATTR_MTIME|ATTR_CTIME);
562 	}
563 	mutex_unlock(&inode->ei_update_lock);
564 err:
565 	return ret;
566 }
567 
bchfs_fcollapse_finsert(struct bch_inode_info * inode,loff_t offset,loff_t len,bool insert)568 static noinline long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
569 				   loff_t offset, loff_t len,
570 				   bool insert)
571 {
572 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
573 	struct address_space *mapping = inode->v.i_mapping;
574 	s64 i_sectors_delta = 0;
575 	int ret = 0;
576 
577 	if ((offset | len) & (block_bytes(c) - 1))
578 		return -EINVAL;
579 
580 	if (insert) {
581 		if (offset >= inode->v.i_size)
582 			return -EINVAL;
583 	} else {
584 		if (offset + len >= inode->v.i_size)
585 			return -EINVAL;
586 	}
587 
588 	ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
589 	if (ret)
590 		return ret;
591 
592 	if (insert)
593 		i_size_write(&inode->v, inode->v.i_size + len);
594 
595 	ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9,
596 				     insert, &i_sectors_delta);
597 	if (!ret && !insert)
598 		i_size_write(&inode->v, inode->v.i_size - len);
599 	bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
600 
601 	return ret;
602 }
603 
__bchfs_fallocate(struct bch_inode_info * inode,int mode,u64 start_sector,u64 end_sector)604 static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
605 			     u64 start_sector, u64 end_sector)
606 {
607 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
608 	struct btree_trans *trans = bch2_trans_get(c);
609 	struct btree_iter iter;
610 	struct bpos end_pos = POS(inode->v.i_ino, end_sector);
611 	struct bch_io_opts opts;
612 	int ret = 0;
613 
614 	bch2_inode_opts_get(&opts, c, &inode->ei_inode);
615 
616 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
617 			POS(inode->v.i_ino, start_sector),
618 			BTREE_ITER_slots|BTREE_ITER_intent);
619 
620 	while (!ret) {
621 		s64 i_sectors_delta = 0;
622 		struct quota_res quota_res = { 0 };
623 		struct bkey_s_c k;
624 		unsigned sectors;
625 		bool is_allocation;
626 		u64 hole_start, hole_end;
627 		u32 snapshot;
628 
629 		bch2_trans_begin(trans);
630 
631 		if (bkey_ge(iter.pos, end_pos))
632 			break;
633 
634 		ret = bch2_subvolume_get_snapshot(trans,
635 					inode->ei_inum.subvol, &snapshot);
636 		if (ret)
637 			goto bkey_err;
638 
639 		bch2_btree_iter_set_snapshot(&iter, snapshot);
640 
641 		k = bch2_btree_iter_peek_slot(&iter);
642 		if ((ret = bkey_err(k)))
643 			goto bkey_err;
644 
645 		hole_start	= iter.pos.offset;
646 		hole_end	= bpos_min(k.k->p, end_pos).offset;
647 		is_allocation	= bkey_extent_is_allocation(k.k);
648 
649 		/* already reserved */
650 		if (bkey_extent_is_reservation(k) &&
651 		    bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
652 			bch2_btree_iter_advance(&iter);
653 			continue;
654 		}
655 
656 		if (bkey_extent_is_data(k.k) &&
657 		    !(mode & FALLOC_FL_ZERO_RANGE)) {
658 			bch2_btree_iter_advance(&iter);
659 			continue;
660 		}
661 
662 		if (!(mode & FALLOC_FL_ZERO_RANGE)) {
663 			/*
664 			 * Lock ordering - can't be holding btree locks while
665 			 * blocking on a folio lock:
666 			 */
667 			if (bch2_clamp_data_hole(&inode->v,
668 						 &hole_start,
669 						 &hole_end,
670 						 opts.data_replicas, true)) {
671 				ret = drop_locks_do(trans,
672 					(bch2_clamp_data_hole(&inode->v,
673 							      &hole_start,
674 							      &hole_end,
675 							      opts.data_replicas, false), 0));
676 				if (ret)
677 					goto bkey_err;
678 			}
679 			bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
680 
681 			if (ret)
682 				goto bkey_err;
683 
684 			if (hole_start == hole_end)
685 				continue;
686 		}
687 
688 		sectors	= hole_end - hole_start;
689 
690 		if (!is_allocation) {
691 			ret = bch2_quota_reservation_add(c, inode,
692 					&quota_res, sectors, true);
693 			if (unlikely(ret))
694 				goto bkey_err;
695 		}
696 
697 		ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter,
698 					    sectors, opts, &i_sectors_delta,
699 					    writepoint_hashed((unsigned long) current));
700 		if (ret)
701 			goto bkey_err;
702 
703 		bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
704 
705 		if (bch2_mark_pagecache_reserved(inode, &hole_start,
706 						 iter.pos.offset, true)) {
707 			ret = drop_locks_do(trans,
708 				bch2_mark_pagecache_reserved(inode, &hole_start,
709 							     iter.pos.offset, false));
710 			if (ret)
711 				goto bkey_err;
712 		}
713 bkey_err:
714 		bch2_quota_reservation_put(c, inode, &quota_res);
715 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
716 			ret = 0;
717 	}
718 
719 	if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
720 		struct quota_res quota_res = { 0 };
721 		s64 i_sectors_delta = 0;
722 
723 		bch2_fpunch_at(trans, &iter, inode_inum(inode),
724 			       end_sector, &i_sectors_delta);
725 		bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
726 		bch2_quota_reservation_put(c, inode, &quota_res);
727 	}
728 
729 	bch2_trans_iter_exit(trans, &iter);
730 	bch2_trans_put(trans);
731 	return ret;
732 }
733 
bchfs_fallocate(struct bch_inode_info * inode,int mode,loff_t offset,loff_t len)734 static noinline long bchfs_fallocate(struct bch_inode_info *inode, int mode,
735 			    loff_t offset, loff_t len)
736 {
737 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
738 	u64 end		= offset + len;
739 	u64 block_start	= round_down(offset,	block_bytes(c));
740 	u64 block_end	= round_up(end,		block_bytes(c));
741 	bool truncated_last_page = false;
742 	int ret, ret2 = 0;
743 
744 	if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
745 		ret = inode_newsize_ok(&inode->v, end);
746 		if (ret)
747 			return ret;
748 	}
749 
750 	if (mode & FALLOC_FL_ZERO_RANGE) {
751 		ret = bch2_truncate_folios(inode, offset, end);
752 		if (unlikely(ret < 0))
753 			return ret;
754 
755 		truncated_last_page = ret;
756 
757 		truncate_pagecache_range(&inode->v, offset, end - 1);
758 
759 		block_start	= round_up(offset,	block_bytes(c));
760 		block_end	= round_down(end,	block_bytes(c));
761 	}
762 
763 	ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
764 
765 	/*
766 	 * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
767 	 * so that the VFS cache i_size is consistent with the btree i_size:
768 	 */
769 	if (ret &&
770 	    !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
771 		return ret;
772 
773 	if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
774 		end = inode->v.i_size;
775 
776 	if (end >= inode->v.i_size &&
777 	    (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
778 	     !(mode & FALLOC_FL_KEEP_SIZE))) {
779 		spin_lock(&inode->v.i_lock);
780 		i_size_write(&inode->v, end);
781 		spin_unlock(&inode->v.i_lock);
782 
783 		mutex_lock(&inode->ei_update_lock);
784 		ret2 = bch2_write_inode_size(c, inode, end, 0);
785 		mutex_unlock(&inode->ei_update_lock);
786 	}
787 
788 	return ret ?: ret2;
789 }
790 
bch2_fallocate_dispatch(struct file * file,int mode,loff_t offset,loff_t len)791 long bch2_fallocate_dispatch(struct file *file, int mode,
792 			     loff_t offset, loff_t len)
793 {
794 	struct bch_inode_info *inode = file_bch_inode(file);
795 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
796 	long ret;
797 
798 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
799 		return -EROFS;
800 
801 	inode_lock(&inode->v);
802 	inode_dio_wait(&inode->v);
803 	bch2_pagecache_block_get(inode);
804 
805 	ret = file_modified(file);
806 	if (ret)
807 		goto err;
808 
809 	if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
810 		ret = bchfs_fallocate(inode, mode, offset, len);
811 	else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
812 		ret = bchfs_fpunch(inode, offset, len);
813 	else if (mode == FALLOC_FL_INSERT_RANGE)
814 		ret = bchfs_fcollapse_finsert(inode, offset, len, true);
815 	else if (mode == FALLOC_FL_COLLAPSE_RANGE)
816 		ret = bchfs_fcollapse_finsert(inode, offset, len, false);
817 	else
818 		ret = -EOPNOTSUPP;
819 err:
820 	bch2_pagecache_block_put(inode);
821 	inode_unlock(&inode->v);
822 	bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
823 
824 	return bch2_err_class(ret);
825 }
826 
827 /*
828  * Take a quota reservation for unallocated blocks in a given file range
829  * Does not check pagecache
830  */
quota_reserve_range(struct bch_inode_info * inode,struct quota_res * res,u64 start,u64 end)831 static int quota_reserve_range(struct bch_inode_info *inode,
832 			       struct quota_res *res,
833 			       u64 start, u64 end)
834 {
835 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
836 	u64 sectors = end - start;
837 
838 	int ret = bch2_trans_run(c,
839 		for_each_btree_key_in_subvolume_max(trans, iter,
840 				BTREE_ID_extents,
841 				POS(inode->v.i_ino, start),
842 				POS(inode->v.i_ino, end - 1),
843 				inode->ei_inum.subvol, 0, k, ({
844 			if (bkey_extent_is_allocation(k.k)) {
845 				u64 s = min(end, k.k->p.offset) -
846 					max(start, bkey_start_offset(k.k));
847 				BUG_ON(s > sectors);
848 				sectors -= s;
849 			}
850 
851 			0;
852 		})));
853 
854 	return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true);
855 }
856 
bch2_remap_file_range(struct file * file_src,loff_t pos_src,struct file * file_dst,loff_t pos_dst,loff_t len,unsigned remap_flags)857 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
858 			     struct file *file_dst, loff_t pos_dst,
859 			     loff_t len, unsigned remap_flags)
860 {
861 	struct bch_inode_info *src = file_bch_inode(file_src);
862 	struct bch_inode_info *dst = file_bch_inode(file_dst);
863 	struct bch_fs *c = src->v.i_sb->s_fs_info;
864 	struct quota_res quota_res = { 0 };
865 	s64 i_sectors_delta = 0;
866 	u64 aligned_len;
867 	loff_t ret = 0;
868 
869 	if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
870 		return -EINVAL;
871 
872 	if ((pos_src & (block_bytes(c) - 1)) ||
873 	    (pos_dst & (block_bytes(c) - 1)))
874 		return -EINVAL;
875 
876 	if (src == dst &&
877 	    abs(pos_src - pos_dst) < len)
878 		return -EINVAL;
879 
880 	lock_two_nondirectories(&src->v, &dst->v);
881 	bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
882 
883 	inode_dio_wait(&src->v);
884 	inode_dio_wait(&dst->v);
885 
886 	ret = generic_remap_file_range_prep(file_src, pos_src,
887 					    file_dst, pos_dst,
888 					    &len, remap_flags);
889 	if (ret < 0 || len == 0)
890 		goto err;
891 
892 	aligned_len = round_up((u64) len, block_bytes(c));
893 
894 	ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
895 				pos_dst, pos_dst + len - 1);
896 	if (ret)
897 		goto err;
898 
899 	ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
900 				  (pos_dst + aligned_len) >> 9);
901 	if (ret)
902 		goto err;
903 
904 	if (!(remap_flags & REMAP_FILE_DEDUP))
905 		file_update_time(file_dst);
906 
907 	bch2_mark_pagecache_unallocated(src, pos_src >> 9,
908 				   (pos_src + aligned_len) >> 9);
909 
910 	/*
911 	 * XXX: we'd like to be telling bch2_remap_range() if we have
912 	 * permission to write to the source file, and thus if io path option
913 	 * changes should be propagated through the copy, but we need mnt_idmap
914 	 * from the pathwalk, awkward
915 	 */
916 	ret = bch2_remap_range(c,
917 			       inode_inum(dst), pos_dst >> 9,
918 			       inode_inum(src), pos_src >> 9,
919 			       aligned_len >> 9,
920 			       pos_dst + len, &i_sectors_delta,
921 			       false);
922 	if (ret < 0)
923 		goto err;
924 
925 	/*
926 	 * due to alignment, we might have remapped slightly more than requsted
927 	 */
928 	ret = min((u64) ret << 9, (u64) len);
929 
930 	bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
931 
932 	spin_lock(&dst->v.i_lock);
933 	if (pos_dst + ret > dst->v.i_size)
934 		i_size_write(&dst->v, pos_dst + ret);
935 	spin_unlock(&dst->v.i_lock);
936 
937 	if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
938 	    IS_SYNC(file_inode(file_dst)))
939 		ret = bch2_flush_inode(c, dst);
940 err:
941 	bch2_quota_reservation_put(c, dst, &quota_res);
942 	bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
943 	unlock_two_nondirectories(&src->v, &dst->v);
944 
945 	return bch2_err_class(ret);
946 }
947 
948 /* fseek: */
949 
bch2_seek_data(struct file * file,u64 offset)950 static loff_t bch2_seek_data(struct file *file, u64 offset)
951 {
952 	struct bch_inode_info *inode = file_bch_inode(file);
953 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
954 	subvol_inum inum = inode_inum(inode);
955 	u64 isize, next_data = MAX_LFS_FILESIZE;
956 
957 	isize = i_size_read(&inode->v);
958 	if (offset >= isize)
959 		return -ENXIO;
960 
961 	int ret = bch2_trans_run(c,
962 		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
963 				   POS(inode->v.i_ino, offset >> 9),
964 				   POS(inode->v.i_ino, U64_MAX),
965 				   inum.subvol, 0, k, ({
966 			if (bkey_extent_is_data(k.k)) {
967 				next_data = max(offset, bkey_start_offset(k.k) << 9);
968 				break;
969 			} else if (k.k->p.offset >> 9 > isize)
970 				break;
971 			0;
972 		})));
973 	if (ret)
974 		return ret;
975 
976 	if (next_data > offset)
977 		next_data = bch2_seek_pagecache_data(&inode->v,
978 					offset, next_data, 0, false);
979 
980 	if (next_data >= isize)
981 		return -ENXIO;
982 
983 	return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
984 }
985 
bch2_seek_hole(struct file * file,u64 offset)986 static loff_t bch2_seek_hole(struct file *file, u64 offset)
987 {
988 	struct bch_inode_info *inode = file_bch_inode(file);
989 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
990 	subvol_inum inum = inode_inum(inode);
991 	u64 isize, next_hole = MAX_LFS_FILESIZE;
992 
993 	isize = i_size_read(&inode->v);
994 	if (offset >= isize)
995 		return -ENXIO;
996 
997 	int ret = bch2_trans_run(c,
998 		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
999 				   POS(inode->v.i_ino, offset >> 9),
1000 				   POS(inode->v.i_ino, U64_MAX),
1001 				   inum.subvol, BTREE_ITER_slots, k, ({
1002 			if (k.k->p.inode != inode->v.i_ino) {
1003 				next_hole = bch2_seek_pagecache_hole(&inode->v,
1004 						offset, MAX_LFS_FILESIZE, 0, false);
1005 				break;
1006 			} else if (!bkey_extent_is_data(k.k)) {
1007 				next_hole = bch2_seek_pagecache_hole(&inode->v,
1008 						max(offset, bkey_start_offset(k.k) << 9),
1009 						k.k->p.offset << 9, 0, false);
1010 
1011 				if (next_hole < k.k->p.offset << 9)
1012 					break;
1013 			} else {
1014 				offset = max(offset, bkey_start_offset(k.k) << 9);
1015 			}
1016 			0;
1017 		})));
1018 	if (ret)
1019 		return ret;
1020 
1021 	if (next_hole > isize)
1022 		next_hole = isize;
1023 
1024 	return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
1025 }
1026 
bch2_llseek(struct file * file,loff_t offset,int whence)1027 loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
1028 {
1029 	loff_t ret;
1030 
1031 	switch (whence) {
1032 	case SEEK_SET:
1033 	case SEEK_CUR:
1034 	case SEEK_END:
1035 		ret = generic_file_llseek(file, offset, whence);
1036 		break;
1037 	case SEEK_DATA:
1038 		ret = bch2_seek_data(file, offset);
1039 		break;
1040 	case SEEK_HOLE:
1041 		ret = bch2_seek_hole(file, offset);
1042 		break;
1043 	default:
1044 		ret = -EINVAL;
1045 		break;
1046 	}
1047 
1048 	return bch2_err_class(ret);
1049 }
1050 
bch2_fs_fsio_exit(struct bch_fs * c)1051 void bch2_fs_fsio_exit(struct bch_fs *c)
1052 {
1053 	bioset_exit(&c->nocow_flush_bioset);
1054 }
1055 
bch2_fs_fsio_init(struct bch_fs * c)1056 int bch2_fs_fsio_init(struct bch_fs *c)
1057 {
1058 	if (bioset_init(&c->nocow_flush_bioset,
1059 			1, offsetof(struct nocow_flush, bio), 0))
1060 		return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
1061 
1062 	return 0;
1063 }
1064 
1065 #endif /* NO_BCACHEFS_FS */
1066