fs-io.c 73.9 KB
Newer Older
1 2 3 4
// SPDX-License-Identifier: GPL-2.0
#ifndef NO_BCACHEFS_FS

#include "bcachefs.h"
5
#include "alloc_foreground.h"
Kent Overstreet's avatar
Kent Overstreet committed
6
#include "bkey_on_stack.h"
7 8 9 10
#include "btree_update.h"
#include "buckets.h"
#include "clock.h"
#include "error.h"
11
#include "extents.h"
12 13 14 15 16 17 18 19
#include "fs.h"
#include "fs-io.h"
#include "fsck.h"
#include "inode.h"
#include "journal.h"
#include "io.h"
#include "keylist.h"
#include "quota.h"
Kent Overstreet's avatar
Kent Overstreet committed
20
#include "reflink.h"
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
#include "trace.h"

#include <linux/aio.h>
#include <linux/backing-dev.h>
#include <linux/falloc.h>
#include <linux/migrate.h>
#include <linux/mmu_context.h>
#include <linux/pagevec.h>
#include <linux/sched/signal.h>
#include <linux/task_io_accounting_ops.h>
#include <linux/uio.h>
#include <linux/writeback.h>

#include <trace/events/writeback.h>

36 37 38 39 40 41 42 43 44
static inline bool bio_full(struct bio *bio, unsigned len)
{
	if (bio->bi_vcnt >= bio->bi_max_vecs)
		return true;
	if (bio->bi_iter.bi_size > UINT_MAX - len)
		return true;
	return false;
}

45 46 47 48 49 50
struct quota_res {
	u64				sectors;
};

struct bch_writepage_io {
	struct closure			cl;
51
	struct bch_inode_info		*inode;
52 53

	/* must be last: */
54
	struct bch_write_op		op;
55 56 57
};

struct dio_write {
58
	struct completion		done;
59
	struct kiocb			*req;
Kent Overstreet's avatar
Kent Overstreet committed
60
	struct mm_struct		*mm;
61 62 63 64 65 66 67 68 69
	unsigned			loop:1,
					sync:1,
					free_iov:1;
	struct quota_res		quota_res;

	struct iov_iter			iter;
	struct iovec			inline_vecs[2];

	/* must be last: */
70
	struct bch_write_op		op;
71 72 73 74 75 76 77 78 79 80
};

struct dio_read {
	struct closure			cl;
	struct kiocb			*req;
	long				ret;
	struct bch_read_bio		rbio;
};

/* pagecache_block must be held */
81
static noinline int write_invalidate_inode_pages_range(struct address_space *mapping,
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
					      loff_t start, loff_t end)
{
	int ret;

	/*
	 * XXX: the way this is currently implemented, we can spin if a process
	 * is continually redirtying a specific page
	 */
	do {
		if (!mapping->nrpages)
			return 0;

		ret = filemap_write_and_wait_range(mapping, start, end);
		if (ret)
			break;

		if (!mapping->nrpages)
			return 0;

		ret = invalidate_inode_pages2_range(mapping,
				start >> PAGE_SHIFT,
				end >> PAGE_SHIFT);
	} while (ret == -EBUSY);

	return ret;
}

/* quotas */

#ifdef CONFIG_BCACHEFS_QUOTA

static void bch2_quota_reservation_put(struct bch_fs *c,
				       struct bch_inode_info *inode,
				       struct quota_res *res)
{
	if (!res->sectors)
		return;

	mutex_lock(&inode->ei_quota_lock);
	BUG_ON(res->sectors > inode->ei_quota_reserved);

	bch2_quota_acct(c, inode->ei_qid, Q_SPC,
124
			-((s64) res->sectors), KEY_TYPE_QUOTA_PREALLOC);
125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
	inode->ei_quota_reserved -= res->sectors;
	mutex_unlock(&inode->ei_quota_lock);

	res->sectors = 0;
}

static int bch2_quota_reservation_add(struct bch_fs *c,
				      struct bch_inode_info *inode,
				      struct quota_res *res,
				      unsigned sectors,
				      bool check_enospc)
{
	int ret;

	mutex_lock(&inode->ei_quota_lock);
	ret = bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors,
141
			      check_enospc ? KEY_TYPE_QUOTA_PREALLOC : KEY_TYPE_QUOTA_NOCHECK);
142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
	if (likely(!ret)) {
		inode->ei_quota_reserved += sectors;
		res->sectors += sectors;
	}
	mutex_unlock(&inode->ei_quota_lock);

	return ret;
}

#else

static void bch2_quota_reservation_put(struct bch_fs *c,
				       struct bch_inode_info *inode,
				       struct quota_res *res)
{
}

static int bch2_quota_reservation_add(struct bch_fs *c,
				      struct bch_inode_info *inode,
				      struct quota_res *res,
				      unsigned sectors,
				      bool check_enospc)
{
	return 0;
}

#endif

/* i_size updates: */

172 173 174 175 176 177
struct inode_new_size {
	loff_t		new_size;
	u64		now;
	unsigned	fields;
};

178 179 180 181
static int inode_set_size(struct bch_inode_info *inode,
			  struct bch_inode_unpacked *bi,
			  void *p)
{
182
	struct inode_new_size *s = p;
183

184 185 186 187 188 189 190
	bi->bi_size = s->new_size;
	if (s->fields & ATTR_ATIME)
		bi->bi_atime = s->now;
	if (s->fields & ATTR_MTIME)
		bi->bi_mtime = s->now;
	if (s->fields & ATTR_CTIME)
		bi->bi_ctime = s->now;
191 192 193 194

	return 0;
}

Kent Overstreet's avatar
Kent Overstreet committed
195 196 197
int __must_check bch2_write_inode_size(struct bch_fs *c,
				       struct bch_inode_info *inode,
				       loff_t new_size, unsigned fields)
198
{
199 200 201 202 203 204 205
	struct inode_new_size s = {
		.new_size	= new_size,
		.now		= bch2_current_time(c),
		.fields		= fields,
	};

	return bch2_write_inode(c, inode, inode_set_size, &s, fields);
206 207 208
}

static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
209
			   struct quota_res *quota_res, s64 sectors)
210
{
211 212 213
	if (!sectors)
		return;

214 215 216 217 218 219 220 221 222
	mutex_lock(&inode->ei_quota_lock);
#ifdef CONFIG_BCACHEFS_QUOTA
	if (quota_res && sectors > 0) {
		BUG_ON(sectors > quota_res->sectors);
		BUG_ON(sectors > inode->ei_quota_reserved);

		quota_res->sectors -= sectors;
		inode->ei_quota_reserved -= sectors;
	} else {
223
		bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
224 225 226 227 228 229 230 231 232 233
	}
#endif
	inode->v.i_blocks += sectors;
	mutex_unlock(&inode->ei_quota_lock);
}

/* page state: */

/* stored in page->private: */

234
struct bch_page_sector {
235
	/* Uncompressed, fully allocated replicas: */
236
	unsigned		nr_replicas:3;
237

238
	/* Owns PAGE_SECTORS * replicas_reserved sized reservation: */
239 240 241 242 243
	unsigned		replicas_reserved:3;

	/* i_sectors: */
	enum {
		SECTOR_UNALLOCATED,
244
		SECTOR_RESERVED,
245 246 247 248
		SECTOR_DIRTY,
		SECTOR_ALLOCATED,
	}			state:2;
};
249

250
struct bch_page_state {
251
	spinlock_t		lock;
252
	atomic_t		write_count;
253
	struct bch_page_sector	s[PAGE_SECTORS];
254 255
};

256
static inline struct bch_page_state *__bch2_page_state(struct page *page)
257
{
258 259 260 261
	return page_has_private(page)
		? (struct bch_page_state *) page_private(page)
		: NULL;
}
262

263 264
static inline struct bch_page_state *bch2_page_state(struct page *page)
{
265
	EBUG_ON(!PageLocked(page));
266

267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
	return __bch2_page_state(page);
}

/* for newly allocated pages: */
static void __bch2_page_state_release(struct page *page)
{
	struct bch_page_state *s = __bch2_page_state(page);

	if (!s)
		return;

	ClearPagePrivate(page);
	set_page_private(page, 0);
	put_page(page);
	kfree(s);
}

static void bch2_page_state_release(struct page *page)
{
	struct bch_page_state *s = bch2_page_state(page);

	if (!s)
		return;

	ClearPagePrivate(page);
	set_page_private(page, 0);
	put_page(page);
	kfree(s);
}

/* for newly allocated pages: */
static struct bch_page_state *__bch2_page_state_create(struct page *page,
						       gfp_t gfp)
{
	struct bch_page_state *s;

	s = kzalloc(sizeof(*s), GFP_NOFS|gfp);
	if (!s)
		return NULL;
306

307
	spin_lock_init(&s->lock);
308 309 310 311 312 313 314
	/*
	 * migrate_page_move_mapping() assumes that pages with private data
	 * have their count elevated by 1.
	 */
	get_page(page);
	set_page_private(page, (unsigned long) s);
	SetPagePrivate(page);
315 316 317
	return s;
}

318 319 320 321 322 323
static struct bch_page_state *bch2_page_state_create(struct page *page,
						     gfp_t gfp)
{
	return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp);
}

324 325 326 327 328 329 330 331
static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode)
{
	/* XXX: this should not be open coded */
	return inode->ei_inode.bi_data_replicas
		? inode->ei_inode.bi_data_replicas - 1
		: c->opts.data_replicas;
}

332 333
static inline unsigned sectors_to_reserve(struct bch_page_sector *s,
						  unsigned nr_replicas)
334
{
335 336 337 338 339 340 341 342 343 344
	return max(0, (int) nr_replicas -
		   s->nr_replicas -
		   s->replicas_reserved);
}

static int bch2_get_page_disk_reservation(struct bch_fs *c,
				struct bch_inode_info *inode,
				struct page *page, bool check_enospc)
{
	struct bch_page_state *s = bch2_page_state_create(page, 0);
345
	unsigned nr_replicas = inode_nr_replicas(c, inode);
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373
	struct disk_reservation disk_res = { 0 };
	unsigned i, disk_res_sectors = 0;
	int ret;

	if (!s)
		return -ENOMEM;

	for (i = 0; i < ARRAY_SIZE(s->s); i++)
		disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas);

	if (!disk_res_sectors)
		return 0;

	ret = bch2_disk_reservation_get(c, &disk_res,
					disk_res_sectors, 1,
					!check_enospc
					? BCH_DISK_RESERVATION_NOFAIL
					: 0);
	if (unlikely(ret))
		return ret;

	for (i = 0; i < ARRAY_SIZE(s->s); i++)
		s->s[i].replicas_reserved +=
			sectors_to_reserve(&s->s[i], nr_replicas);

	return 0;
}

374 375 376 377 378 379
struct bch2_page_reservation {
	struct disk_reservation	disk;
	struct quota_res	quota;
};

static void bch2_page_reservation_init(struct bch_fs *c,
380
			struct bch_inode_info *inode,
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
			struct bch2_page_reservation *res)
{
	memset(res, 0, sizeof(*res));

	res->disk.nr_replicas = inode_nr_replicas(c, inode);
}

static void bch2_page_reservation_put(struct bch_fs *c,
			struct bch_inode_info *inode,
			struct bch2_page_reservation *res)
{
	bch2_disk_reservation_put(c, &res->disk);
	bch2_quota_reservation_put(c, inode, &res->quota);
}

static int bch2_page_reservation_get(struct bch_fs *c,
			struct bch_inode_info *inode, struct page *page,
			struct bch2_page_reservation *res,
			unsigned offset, unsigned len, bool check_enospc)
400 401
{
	struct bch_page_state *s = bch2_page_state_create(page, 0);
402
	unsigned i, disk_sectors = 0, quota_sectors = 0;
403
	int ret;
404

405 406
	if (!s)
		return -ENOMEM;
407

408 409
	for (i = round_down(offset, block_bytes(c)) >> 9;
	     i < round_up(offset + len, block_bytes(c)) >> 9;
410 411 412 413 414
	     i++) {
		disk_sectors += sectors_to_reserve(&s->s[i],
						res->disk.nr_replicas);
		quota_sectors += s->s[i].state == SECTOR_UNALLOCATED;
	}
415

416 417 418 419 420 421 422 423 424
	if (disk_sectors) {
		ret = bch2_disk_reservation_add(c, &res->disk,
						disk_sectors,
						!check_enospc
						? BCH_DISK_RESERVATION_NOFAIL
						: 0);
		if (unlikely(ret))
			return ret;
	}
425

426 427 428 429 430 431 432 433 434 435 436 437 438 439
	if (quota_sectors) {
		ret = bch2_quota_reservation_add(c, inode, &res->quota,
						 quota_sectors,
						 check_enospc);
		if (unlikely(ret)) {
			struct disk_reservation tmp = {
				.sectors = disk_sectors
			};

			bch2_disk_reservation_put(c, &tmp);
			res->disk.sectors -= disk_sectors;
			return ret;
		}
	}
440

441
	return 0;
442 443 444 445 446 447
}

static void bch2_clear_page_bits(struct page *page)
{
	struct bch_inode_info *inode = to_bch_ei(page->mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
448
	struct bch_page_state *s = bch2_page_state(page);
449
	struct disk_reservation disk_res = { 0 };
450
	int i, dirty_sectors = 0;
451

452
	if (!s)
453 454
		return;

455 456 457
	EBUG_ON(!PageLocked(page));
	EBUG_ON(PageWriteback(page));

458
	for (i = 0; i < ARRAY_SIZE(s->s); i++) {
459 460 461
		disk_res.sectors += s->s[i].replicas_reserved;
		s->s[i].replicas_reserved = 0;

462 463 464 465 466
		if (s->s[i].state == SECTOR_DIRTY) {
			dirty_sectors++;
			s->s[i].state = SECTOR_UNALLOCATED;
		}
	}
467

468 469
	bch2_disk_reservation_put(c, &disk_res);

470 471
	if (dirty_sectors)
		i_sectors_acct(c, inode, NULL, -dirty_sectors);
472

473
	bch2_page_state_release(page);
474 475
}

476 477 478 479
static void bch2_set_page_dirty(struct bch_fs *c,
			struct bch_inode_info *inode, struct page *page,
			struct bch2_page_reservation *res,
			unsigned offset, unsigned len)
480
{
481
	struct bch_page_state *s = bch2_page_state(page);
482
	unsigned i, dirty_sectors = 0;
483

484 485
	WARN_ON((u64) page_offset(page) + offset + len >
		round_up((u64) i_size_read(&inode->v), block_bytes(c)));
486

487 488
	spin_lock(&s->lock);

489 490
	for (i = round_down(offset, block_bytes(c)) >> 9;
	     i < round_up(offset + len, block_bytes(c)) >> 9;
491 492 493
	     i++) {
		unsigned sectors = sectors_to_reserve(&s->s[i],
						res->disk.nr_replicas);
494

495 496 497 498 499 500
		/*
		 * This can happen if we race with the error path in
		 * bch2_writepage_io_done():
		 */
		sectors = min_t(unsigned, sectors, res->disk.sectors);

501 502
		s->s[i].replicas_reserved += sectors;
		res->disk.sectors -= sectors;
503

504
		if (s->s[i].state == SECTOR_UNALLOCATED)
505
			dirty_sectors++;
506 507

		s->s[i].state = max_t(unsigned, s->s[i].state, SECTOR_DIRTY);
508 509
	}

510 511
	spin_unlock(&s->lock);

512
	if (dirty_sectors)
513
		i_sectors_acct(c, inode, &res->quota, dirty_sectors);
514

515 516
	if (!PageDirty(page))
		filemap_dirty_folio(inode->v.i_mapping, page_folio(page));
517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
}

vm_fault_t bch2_page_fault(struct vm_fault *vmf)
{
	struct file *file = vmf->vma->vm_file;
	struct bch_inode_info *inode = file_bch_inode(file);
	int ret;

	bch2_pagecache_add_get(&inode->ei_pagecache_lock);
	ret = filemap_fault(vmf);
	bch2_pagecache_add_put(&inode->ei_pagecache_lock);

	return ret;
}

vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf)
{
	struct page *page = vmf->page;
	struct file *file = vmf->vma->vm_file;
	struct bch_inode_info *inode = file_bch_inode(file);
	struct address_space *mapping = file->f_mapping;
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
539
	struct bch2_page_reservation res;
540 541
	unsigned len;
	loff_t isize;
542 543
	int ret = VM_FAULT_LOCKED;

544 545
	bch2_page_reservation_init(c, inode, &res);

546 547 548 549 550 551 552 553 554 555 556 557
	sb_start_pagefault(inode->v.i_sb);
	file_update_time(file);

	/*
	 * Not strictly necessary, but helps avoid dio writes livelocking in
	 * write_invalidate_inode_pages_range() - can drop this if/when we get
	 * a write_invalidate_inode_pages_range() that works without dropping
	 * page lock before invalidating page
	 */
	bch2_pagecache_add_get(&inode->ei_pagecache_lock);

	lock_page(page);
558 559 560
	isize = i_size_read(&inode->v);

	if (page->mapping != mapping || page_offset(page) >= isize) {
561 562 563 564 565
		unlock_page(page);
		ret = VM_FAULT_NOPAGE;
		goto out;
	}

566
	len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page));
567 568

	if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) {
569 570 571 572 573
		unlock_page(page);
		ret = VM_FAULT_SIGBUS;
		goto out;
	}

574
	bch2_set_page_dirty(c, inode, page, &res, 0, len);
575 576
	bch2_page_reservation_put(c, inode, &res);

577 578 579 580
	wait_for_stable_page(page);
out:
	bch2_pagecache_add_put(&inode->ei_pagecache_lock);
	sb_end_pagefault(inode->v.i_sb);
581

582 583 584 585 586 587 588 589 590 591 592 593 594
	return ret;
}

void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length)
{
	if (offset || length < folio_size(folio))
		return;

	bch2_clear_page_bits(&folio->page);
}

bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask)
{
595
	if (folio_test_dirty(folio) || folio_test_writeback(folio))
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
		return false;

	bch2_clear_page_bits(&folio->page);
	return true;
}

/* readpage(s): */

static void bch2_readpages_end_io(struct bio *bio)
{
	struct bvec_iter_all iter;
	struct bio_vec *bv;

	bio_for_each_segment_all(bv, bio, iter) {
		struct page *page = bv->bv_page;

		if (!bio->bi_status) {
			SetPageUptodate(page);
		} else {
			ClearPageUptodate(page);
			SetPageError(page);
		}
		unlock_page(page);
	}

	bio_put(bio);
}

static inline void page_state_init_for_read(struct page *page)
{
626 627
	SetPagePrivate(page);
	page->private = 0;
628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
}

struct readpages_iter {
	struct address_space	*mapping;
	struct page		**pages;
	unsigned		nr_pages;
	unsigned		idx;
	pgoff_t			offset;
};

static int readpages_iter_init(struct readpages_iter *iter,
			       struct readahead_control *ractl)
{
	unsigned i, nr_pages = readahead_count(ractl);

	memset(iter, 0, sizeof(*iter));

	iter->mapping	= ractl->mapping;
	iter->offset	= readahead_index(ractl);
	iter->nr_pages	= nr_pages;

	iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS);
	if (!iter->pages)
		return -ENOMEM;

	__readahead_batch(ractl, iter->pages, nr_pages);
	for (i = 0; i < nr_pages; i++) {
655
		__bch2_page_state_create(iter->pages[i], __GFP_NOFAIL);
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675
		put_page(iter->pages[i]);
	}

	return 0;
}

static inline struct page *readpage_iter_next(struct readpages_iter *iter)
{
	if (iter->idx >= iter->nr_pages)
		return NULL;

	EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx);

	return iter->pages[iter->idx];
}

static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k)
{
	struct bvec_iter iter;
	struct bio_vec bv;
Kent Overstreet's avatar
Kent Overstreet committed
676 677
	unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v
		? 0 : bch2_bkey_nr_ptrs_allocated(k);
678 679 680
	unsigned state = k.k->type == KEY_TYPE_reservation
		? SECTOR_RESERVED
		: SECTOR_ALLOCATED;
681

682 683 684 685 686 687 688 689
	bio_for_each_segment(bv, bio, iter) {
		struct bch_page_state *s = bch2_page_state(bv.bv_page);
		unsigned i;

		for (i = bv.bv_offset >> 9;
		     i < (bv.bv_offset + bv.bv_len) >> 9;
		     i++) {
			s->s[i].nr_replicas = nr_ptrs;
690
			s->s[i].state = state;
691
		}
692 693 694
	}
}

Kent Overstreet's avatar
Kent Overstreet committed
695 696 697 698 699 700 701 702 703 704 705 706
static bool extent_partial_reads_expensive(struct bkey_s_c k)
{
	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
	struct bch_extent_crc_unpacked crc;
	const union bch_extent_entry *i;

	bkey_for_each_crc(k.k, ptrs, crc, i)
		if (crc.csum_type || crc.compression_type)
			return true;
	return false;
}

707
static void readpage_bio_extend(struct readpages_iter *iter,
Kent Overstreet's avatar
Kent Overstreet committed
708 709
				struct bio *bio,
				unsigned sectors_this_extent,
710 711
				bool get_more)
{
Kent Overstreet's avatar
Kent Overstreet committed
712
	while (bio_sectors(bio) < sectors_this_extent &&
713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734
	       bio->bi_vcnt < bio->bi_max_vecs) {
		pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT;
		struct page *page = readpage_iter_next(iter);
		int ret;

		if (page) {
			if (iter->offset + iter->idx != page_offset)
				break;

			iter->idx++;
		} else {
			if (!get_more)
				break;

			page = xa_load(&iter->mapping->i_pages, page_offset);
			if (page && !xa_is_value(page))
				break;

			page = __page_cache_alloc(readahead_gfp_mask(iter->mapping));
			if (!page)
				break;

735 736 737 738
			if (!__bch2_page_state_create(page, 0)) {
				put_page(page);
				break;
			}
739 740 741 742

			ret = add_to_page_cache_lru(page, iter->mapping,
						    page_offset, GFP_NOFS);
			if (ret) {
743
				__bch2_page_state_release(page);
744 745 746 747 748 749 750
				put_page(page);
				break;
			}

			put_page(page);
		}

751
		BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0));
752 753 754
	}
}

755
static void bchfs_read(struct btree_trans *trans, struct btree_iter *iter,
756 757 758
		       struct bch_read_bio *rbio, u64 inum,
		       struct readpages_iter *readpages_iter)
{
759
	struct bch_fs *c = trans->c;
Kent Overstreet's avatar
Kent Overstreet committed
760
	struct bkey_on_stack sk;
761 762
	int flags = BCH_READ_RETRY_IF_STALE|
		BCH_READ_MAY_PROMOTE;
Kent Overstreet's avatar
Kent Overstreet committed
763
	int ret = 0;
764 765 766

	rbio->c = c;
	rbio->start_time = local_clock();
Kent Overstreet's avatar
Kent Overstreet committed
767 768

	bkey_on_stack_init(&sk);
Kent Overstreet's avatar
Kent Overstreet committed
769
retry:
770 771
	while (1) {
		struct bkey_s_c k;
Kent Overstreet's avatar
Kent Overstreet committed
772
		unsigned bytes, sectors, offset_into_extent;
773

774 775
		bch2_btree_iter_set_pos(iter,
				POS(inum, rbio->bio.bi_iter.bi_sector));
776 777

		k = bch2_btree_iter_peek_slot(iter);
Kent Overstreet's avatar
Kent Overstreet committed
778 779 780
		ret = bkey_err(k);
		if (ret)
			break;
781

Kent Overstreet's avatar
Kent Overstreet committed
782 783 784
		bkey_on_stack_realloc(&sk, c, k.k->u64s);
		bkey_reassemble(sk.k, k);
		k = bkey_i_to_s_c(sk.k);
785

786 787
		offset_into_extent = iter->pos.offset -
			bkey_start_offset(k.k);
Kent Overstreet's avatar
Kent Overstreet committed
788 789
		sectors = k.k->size - offset_into_extent;

790
		ret = bch2_read_indirect_extent(trans,
Kent Overstreet's avatar
Kent Overstreet committed
791
					&offset_into_extent, sk.k);
Kent Overstreet's avatar
Kent Overstreet committed
792 793 794 795 796 797
		if (ret)
			break;

		sectors = min(sectors, k.k->size - offset_into_extent);

		bch2_trans_unlock(trans);
798

Kent Overstreet's avatar
Kent Overstreet committed
799 800 801
		if (readpages_iter)
			readpage_bio_extend(readpages_iter, &rbio->bio, sectors,
					    extent_partial_reads_expensive(k));
802

Kent Overstreet's avatar
Kent Overstreet committed
803
		bytes = min(sectors, bio_sectors(&rbio->bio)) << 9;
804
		swap(rbio->bio.bi_iter.bi_size, bytes);
805

806
		if (rbio->bio.bi_iter.bi_size == bytes)
807 808 809
			flags |= BCH_READ_LAST_FRAGMENT;

		if (bkey_extent_is_allocation(k.k))
810
			bch2_add_page_sectors(&rbio->bio, k);
811

812
		bch2_read_extent(c, rbio, k, offset_into_extent, flags);
813 814

		if (flags & BCH_READ_LAST_FRAGMENT)
Kent Overstreet's avatar
Kent Overstreet committed
815
			break;
816

817 818
		swap(rbio->bio.bi_iter.bi_size, bytes);
		bio_advance(&rbio->bio, bytes);
819
	}
Kent Overstreet's avatar
Kent Overstreet committed
820 821 822 823

	if (ret == -EINTR)
		goto retry;

Kent Overstreet's avatar
Kent Overstreet committed
824 825 826 827 828 829
	if (ret) {
		bcache_io_error(c, &rbio->bio, "btree IO error %i", ret);
		bio_endio(&rbio->bio);
	}

	bkey_on_stack_exit(&sk, c);
830 831 832 833 834 835
}

void bch2_readahead(struct readahead_control *ractl)
{
	struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
836
	struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
837 838
	struct btree_trans trans;
	struct btree_iter *iter;
839 840 841 842 843 844 845
	struct page *page;
	struct readpages_iter readpages_iter;
	int ret;

	ret = readpages_iter_init(&readpages_iter, ractl);
	BUG_ON(ret);

846
	bch2_trans_init(&trans, c, 0, 0);
847 848 849

	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN,
				   BTREE_ITER_SLOTS);
850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867

	bch2_pagecache_add_get(&inode->ei_pagecache_lock);

	while ((page = readpage_iter_next(&readpages_iter))) {
		pgoff_t index = readpages_iter.offset + readpages_iter.idx;
		unsigned n = min_t(unsigned,
				   readpages_iter.nr_pages -
				   readpages_iter.idx,
				   BIO_MAX_VECS);
		struct bch_read_bio *rbio =
			rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ,
						   GFP_NOFS, &c->bio_read),
				  opts);

		readpages_iter.idx++;

		rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTOR_SHIFT;
		rbio->bio.bi_end_io = bch2_readpages_end_io;
868
		BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0));
869

870 871
		bchfs_read(&trans, iter, rbio, inode->v.i_ino,
			   &readpages_iter);
872 873 874
	}

	bch2_pagecache_add_put(&inode->ei_pagecache_lock);
875 876

	bch2_trans_exit(&trans);
877 878 879 880 881 882
	kfree(readpages_iter.pages);
}

static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio,
			     u64 inum, struct page *page)
{
883 884
	struct btree_trans trans;
	struct btree_iter *iter;
885

886
	bch2_page_state_create(page, __GFP_NOFAIL);
887 888

	rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC;
889 890 891
	rbio->bio.bi_iter.bi_sector =
		(sector_t) page->index << PAGE_SECTOR_SHIFT;
	BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0));
892

893
	bch2_trans_init(&trans, c, 0, 0);
894 895 896
	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN,
				   BTREE_ITER_SLOTS);

897
	bchfs_read(&trans, iter, rbio, inum, NULL);
898 899

	bch2_trans_exit(&trans);
900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916
}

static void bch2_read_single_page_end_io(struct bio *bio)
{
	complete(bio->bi_private);
}

static int bch2_read_single_page(struct page *page,
				 struct address_space *mapping)
{
	struct bch_inode_info *inode = to_bch_ei(mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct bch_read_bio *rbio;
	int ret;
	DECLARE_COMPLETION_ONSTACK(done);

	rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read),
917
			 io_opts(c, &inode->ei_inode));
918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
	rbio->bio.bi_private = &done;
	rbio->bio.bi_end_io = bch2_read_single_page_end_io;

	__bchfs_readpage(c, rbio, inode->v.i_ino, page);
	wait_for_completion(&done);

	ret = blk_status_to_errno(rbio->bio.bi_status);
	bio_put(&rbio->bio);

	if (ret < 0)
		return ret;

	SetPageUptodate(page);
	return 0;
}

int bch2_read_folio(struct file *file, struct folio *folio)
{
	struct page *page = &folio->page;
	int ret;

	ret = bch2_read_single_page(page, page->mapping);
	folio_unlock(folio);
	return ret;
}

/* writepages: */

struct bch_writepage_state {
	struct bch_writepage_io	*io;
	struct bch_io_opts	opts;
};

static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c,
								  struct bch_inode_info *inode)
{
954 955 956
	return (struct bch_writepage_state) {
		.opts = io_opts(c, &inode->ei_inode)
	};
957 958 959 960 961 962 963
}

static void bch2_writepage_io_free(struct closure *cl)
{
	struct bch_writepage_io *io = container_of(cl,
					struct bch_writepage_io, cl);

964
	bio_put(&io->op.wbio.bio);
965 966 967 968 969 970
}

static void bch2_writepage_io_done(struct closure *cl)
{
	struct bch_writepage_io *io = container_of(cl,
					struct bch_writepage_io, cl);
971 972
	struct bch_fs *c = io->op.c;
	struct bio *bio = &io->op.wbio.bio;
973 974
	struct bvec_iter_all iter;
	struct bio_vec *bvec;
975
	unsigned i;
976

977
	if (io->op.error) {
978
		bio_for_each_segment_all(bvec, bio, iter) {
979 980
			struct bch_page_state *s;

981
			SetPageError(bvec->bv_page);
982
			mapping_set_error(bvec->bv_page->mapping, -EIO);
983

984 985
			s = __bch2_page_state(bvec->bv_page);
			spin_lock(&s->lock);
986 987
			for (i = 0; i < PAGE_SECTORS; i++)
				s->s[i].nr_replicas = 0;
988
			spin_unlock(&s->lock);
989
		}
990 991 992 993 994 995
	}

	/*
	 * racing with fallocate can cause us to add fewer sectors than
	 * expected - but we shouldn't add more sectors than expected:
	 */
996
	BUG_ON(io->op.i_sectors_delta > 0);
997 998 999 1000 1001

	/*
	 * (error (due to going RO) halfway through a page can screw that up
	 * slightly)
	 * XXX wtf?
1002
	   BUG_ON(io->op.op.i_sectors_delta >= PAGE_SECTORS);
1003 1004 1005 1006 1007 1008
	 */

	/*
	 * PageWriteback is effectively our ref on the inode - fixup i_blocks
	 * before calling end_page_writeback:
	 */
1009
	i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta);
1010

1011 1012 1013 1014 1015 1016
	bio_for_each_segment_all(bvec, bio, iter) {
		struct bch_page_state *s = __bch2_page_state(bvec->bv_page);

		if (atomic_dec_and_test(&s->write_count))
			end_page_writeback(bvec->bv_page);
	}
1017 1018 1019 1020 1021 1022 1023 1024 1025

	closure_return_with_destructor(&io->cl, bch2_writepage_io_free);
}

static void bch2_writepage_do_io(struct bch_writepage_state *w)
{
	struct bch_writepage_io *io = w->io;

	w->io = NULL;
1026
	closure_call(&io->op.cl, bch2_write, NULL, &io->cl);
1027 1028 1029 1030 1031 1032 1033 1034
	continue_at(&io->cl, bch2_writepage_io_done, NULL);
}

/*
 * Get a bch_writepage_io and add @page to it - appending to an existing one if
 * possible, else allocating a new one:
 */
static void bch2_writepage_io_alloc(struct bch_fs *c,
1035
				    struct writeback_control *wbc,
1036 1037
				    struct bch_writepage_state *w,
				    struct bch_inode_info *inode,
1038
				    u64 sector,
1039 1040 1041 1042 1043 1044 1045 1046
				    unsigned nr_replicas)
{
	struct bch_write_op *op;

	w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS,
					      REQ_OP_WRITE,
					      GFP_NOFS,
					      &c->writepage_bioset),
1047
			     struct bch_writepage_io, op.wbio.bio);
1048 1049

	closure_init(&w->io->cl, NULL);
1050 1051 1052 1053 1054 1055
	w->io->inode		= inode;

	op			= &w->io->op;
	bch2_write_op_init(op, c, w->opts);
	op->target		= w->opts.foreground_target;
	op_journal_seq_set(op, &inode->ei_journal_seq);
1056 1057 1058
	op->nr_replicas		= nr_replicas;
	op->res.nr_replicas	= nr_replicas;
	op->write_point		= writepoint_hashed(inode->ei_last_dirtied);
1059 1060
	op->pos			= POS(inode->v.i_ino, sector);
	op->wbio.bio.bi_iter.bi_sector = sector;
1061
	op->wbio.bio.bi_opf	= wbc_to_write_flags(wbc);
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
}

static int __bch2_writepage(struct folio *folio,
			    struct writeback_control *wbc,
			    void *data)
{
	struct page *page = &folio->page;
	struct bch_inode_info *inode = to_bch_ei(page->mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct bch_writepage_state *w = data;
1072 1073
	struct bch_page_state *s, orig;
	unsigned i, offset, nr_replicas_this_write = U32_MAX;
1074 1075
	loff_t i_size = i_size_read(&inode->v);
	pgoff_t end_index = i_size >> PAGE_SHIFT;
1076
	int ret;
1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099

	EBUG_ON(!PageUptodate(page));

	/* Is the page fully inside i_size? */
	if (page->index < end_index)
		goto do_io;

	/* Is the page fully outside i_size? (truncate in progress) */
	offset = i_size & (PAGE_SIZE - 1);
	if (page->index > end_index || !offset) {
		unlock_page(page);
		return 0;
	}

	/*
	 * The page straddles i_size.  It must be zeroed out on each and every
	 * writepage invocation because it may be mmapped.  "A file is mapped
	 * in multiples of the page size.  For a file that is not a multiple of
	 * the  page size, the remaining memory is zeroed when mapped, and
	 * writes to that region are not written out to the file."
	 */
	zero_user_segment(page, offset, PAGE_SIZE);
do_io:
1100
	s = bch2_page_state_create(page, __GFP_NOFAIL);
1101

1102
	ret = bch2_get_page_disk_reservation(c, inode, page, true);
1103 1104 1105 1106 1107 1108
	if (ret) {
		SetPageError(page);
		mapping_set_error(page->mapping, ret);
		unlock_page(page);
		return 0;
	}
1109

1110 1111 1112 1113
	/* Before unlocking the page, get copy of reservations: */
	orig = *s;

	for (i = 0; i < PAGE_SECTORS; i++) {
1114
		if (s->s[i].state < SECTOR_DIRTY)
1115 1116
			continue;

1117 1118 1119 1120
		nr_replicas_this_write =
			min_t(unsigned, nr_replicas_this_write,
			      s->s[i].nr_replicas +
			      s->s[i].replicas_reserved);
1121
	}
1122

1123
	for (i = 0; i < PAGE_SECTORS; i++) {
1124
		if (s->s[i].state < SECTOR_DIRTY)
1125 1126
			continue;

1127 1128
		s->s[i].nr_replicas = w->opts.compression
			? 0 : nr_replicas_this_write;
1129

1130 1131 1132
		s->s[i].replicas_reserved = 0;
		s->s[i].state = SECTOR_ALLOCATED;
	}
1133

1134 1135 1136
	BUG_ON(atomic_read(&s->write_count));
	atomic_set(&s->write_count, 1);

1137 1138
	BUG_ON(PageWriteback(page));
	set_page_writeback(page);
1139

1140 1141
	unlock_page(page);

1142 1143 1144 1145 1146 1147
	offset = 0;
	while (1) {
		unsigned sectors = 1, dirty_sectors = 0, reserved_sectors = 0;
		u64 sector;

		while (offset < PAGE_SECTORS &&
1148
		       orig.s[offset].state < SECTOR_DIRTY)
1149
			offset++;
1150

1151 1152 1153 1154 1155 1156
		if (offset == PAGE_SECTORS)
			break;

		sector = ((u64) page->index << PAGE_SECTOR_SHIFT) + offset;

		while (offset + sectors < PAGE_SECTORS &&
1157
		       orig.s[offset + sectors].state >= SECTOR_DIRTY)
1158 1159 1160 1161 1162 1163 1164 1165
			sectors++;

		for (i = offset; i < offset + sectors; i++) {
			reserved_sectors += orig.s[i].replicas_reserved;
			dirty_sectors += orig.s[i].state == SECTOR_DIRTY;
		}

		if (w->io &&
1166 1167
		    (w->io->op.res.nr_replicas != nr_replicas_this_write ||
		     bio_full(&w->io->op.wbio.bio, PAGE_SIZE) ||
1168
		     w->io->op.wbio.bio.bi_iter.bi_size >= (256U << 20) ||
1169
		     bio_end_sector(&w->io->op.wbio.bio) != sector))
1170
			bch2_writepage_do_io(w);
1171

1172
		if (!w->io)
1173
			bch2_writepage_io_alloc(c, wbc, w, inode, sector,
1174
						nr_replicas_this_write);
1175

1176 1177
		atomic_inc(&s->write_count);

1178 1179
		BUG_ON(inode != w->io->inode);
		BUG_ON(!bio_add_page(&w->io->op.wbio.bio, page,
1180 1181
				     sectors << 9, offset << 9));

1182
		/* Check for writing past i_size: */
1183
		WARN_ON((bio_end_sector(&w->io->op.wbio.bio) << 9) >
1184
			round_up(i_size, block_bytes(c)));
1185

1186 1187
		w->io->op.res.sectors += reserved_sectors;
		w->io->op.i_sectors_delta -= dirty_sectors;
1188 1189 1190 1191
		w->io->op.new_i_size = i_size;

		offset += sectors;
	}
1192

1193 1194
	if (atomic_dec_and_test(&s->write_count))
		end_page_writeback(page);
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236

	return 0;
}

int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct bch_fs *c = mapping->host->i_sb->s_fs_info;
	struct bch_writepage_state w =
		bch_writepage_state_init(c, to_bch_ei(mapping->host));
	struct blk_plug plug;
	int ret;

	blk_start_plug(&plug);
	ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w);
	if (w.io)
		bch2_writepage_do_io(&w);
	blk_finish_plug(&plug);
	return ret;
}

int bch2_writepage(struct page *page, struct writeback_control *wbc)
{
	struct bch_fs *c = page->mapping->host->i_sb->s_fs_info;
	struct bch_writepage_state w =
		bch_writepage_state_init(c, to_bch_ei(page->mapping->host));
	int ret;

	ret = __bch2_writepage(page_folio(page), wbc, &w);
	if (w.io)
		bch2_writepage_do_io(&w);

	return ret;
}

/* buffered writes: */

int bch2_write_begin(struct file *file, struct address_space *mapping,
		     loff_t pos, unsigned len,
		     struct page **pagep, void **fsdata)
{
	struct bch_inode_info *inode = to_bch_ei(mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1237
	struct bch2_page_reservation *res;
1238 1239 1240 1241 1242
	pgoff_t index = pos >> PAGE_SHIFT;
	unsigned offset = pos & (PAGE_SIZE - 1);
	struct page *page;
	int ret = -ENOMEM;

1243 1244 1245 1246 1247 1248
	res = kmalloc(sizeof(*res), GFP_KERNEL);
	if (!res)
		return -ENOMEM;

	bch2_page_reservation_init(c, inode, res);
	*fsdata = res;
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278

	bch2_pagecache_add_get(&inode->ei_pagecache_lock);

	page = grab_cache_page_write_begin(mapping, index);
	if (!page)
		goto err_unlock;

	if (PageUptodate(page))
		goto out;

	/* If we're writing entire page, don't need to read it in first: */
	if (len == PAGE_SIZE)
		goto out;

	if (!offset && pos + len >= inode->v.i_size) {
		zero_user_segment(page, len, PAGE_SIZE);
		flush_dcache_page(page);
		goto out;
	}

	if (index > inode->v.i_size >> PAGE_SHIFT) {
		zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE);
		flush_dcache_page(page);
		goto out;
	}
readpage:
	ret = bch2_read_single_page(page, mapping);
	if (ret)
		goto err;
out:
1279 1280
	ret = bch2_page_reservation_get(c, inode, page, res,
					offset, len, true);
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
	if (ret) {
		if (!PageUptodate(page)) {
			/*
			 * If the page hasn't been read in, we won't know if we
			 * actually need a reservation - we don't actually need
			 * to read here, we just need to check if the page is
			 * fully backed by uncompressed data:
			 */
			goto readpage;
		}

		goto err;
	}

	*pagep = page;
	return 0;
err:
	unlock_page(page);
	put_page(page);
	*pagep = NULL;
err_unlock:
	bch2_pagecache_add_put(&inode->ei_pagecache_lock);
1303 1304
	kfree(res);
	*fsdata = NULL;
1305 1306 1307 1308 1309 1310 1311 1312 1313
	return ret;
}

int bch2_write_end(struct file *file, struct address_space *mapping,
		   loff_t pos, unsigned len, unsigned copied,
		   struct page *page, void *fsdata)
{
	struct bch_inode_info *inode = to_bch_ei(mapping->host);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1314 1315
	struct bch2_page_reservation *res = fsdata;
	unsigned offset = pos & (PAGE_SIZE - 1);
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337

	lockdep_assert_held(&inode->v.i_rwsem);

	if (unlikely(copied < len && !PageUptodate(page))) {
		/*
		 * The page needs to be read in, but that would destroy
		 * our partial write - simplest thing is to just force
		 * userspace to redo the write:
		 */
		zero_user(page, 0, PAGE_SIZE);
		flush_dcache_page(page);
		copied = 0;
	}

	spin_lock(&inode->v.i_lock);
	if (pos + copied > inode->v.i_size)
		i_size_write(&inode->v, pos + copied);
	spin_unlock(&inode->v.i_lock);

	if (copied) {
		if (!PageUptodate(page))
			SetPageUptodate(page);
1338 1339

		bch2_set_page_dirty(c, inode, page, res, offset, copied);
1340 1341 1342 1343 1344 1345 1346 1347

		inode->ei_last_dirtied = (unsigned long) current;
	}

	unlock_page(page);
	put_page(page);
	bch2_pagecache_add_put(&inode->ei_pagecache_lock);

1348 1349 1350
	bch2_page_reservation_put(c, inode, res);
	kfree(res);

1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362
	return copied;
}

#define WRITE_BATCH_PAGES	32

static int __bch2_buffered_write(struct bch_inode_info *inode,
				 struct address_space *mapping,
				 struct iov_iter *iter,
				 loff_t pos, unsigned len)
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct page *pages[WRITE_BATCH_PAGES];
1363
	struct bch2_page_reservation res;
1364 1365 1366
	unsigned long index = pos >> PAGE_SHIFT;
	unsigned offset = pos & (PAGE_SIZE - 1);
	unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1367 1368
	unsigned i, reserved = 0, set_dirty = 0;
	unsigned copied = 0, nr_pages_copied = 0;
1369 1370 1371 1372 1373
	int ret = 0;

	BUG_ON(!len);
	BUG_ON(nr_pages > ARRAY_SIZE(pages));

1374 1375
	bch2_page_reservation_init(c, inode, &res);

1376 1377 1378 1379
	for (i = 0; i < nr_pages; i++) {
		pages[i] = grab_cache_page_write_begin(mapping, index + i);
		if (!pages[i]) {
			nr_pages = i;
1380 1381 1382 1383 1384 1385 1386
			if (!i) {
				ret = -ENOMEM;
				goto out;
			}
			len = min_t(unsigned, len,
				    nr_pages * PAGE_SIZE - offset);
			break;
1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406
		}
	}

	if (offset && !PageUptodate(pages[0])) {
		ret = bch2_read_single_page(pages[0], mapping);
		if (ret)
			goto out;
	}

	if ((pos + len) & (PAGE_SIZE - 1) &&
	    !PageUptodate(pages[nr_pages - 1])) {
		if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) {
			zero_user(pages[nr_pages - 1], 0, PAGE_SIZE);
		} else {
			ret = bch2_read_single_page(pages[nr_pages - 1], mapping);
			if (ret)
				goto out;
		}
	}

1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419
	while (reserved < len) {
		struct page *page = pages[(offset + reserved) >> PAGE_SHIFT];
		unsigned pg_offset = (offset + reserved) & (PAGE_SIZE - 1);
		unsigned pg_len = min_t(unsigned, len - reserved,
					PAGE_SIZE - pg_offset);
retry_reservation:
		ret = bch2_page_reservation_get(c, inode, page, &res,
						pg_offset, pg_len, true);

		if (ret && !PageUptodate(page)) {
			ret = bch2_read_single_page(page, mapping);
			if (!ret)
				goto retry_reservation;
1420 1421 1422 1423
		}

		if (ret)
			goto out;
1424 1425

		reserved += pg_len;
1426 1427 1428 1429 1430 1431 1432 1433 1434
	}

	if (mapping_writably_mapped(mapping))
		for (i = 0; i < nr_pages; i++)
			flush_dcache_page(pages[i]);

	while (copied < len) {
		struct page *page = pages[(offset + copied) >> PAGE_SHIFT];
		unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1);
1435 1436
		unsigned pg_len = min_t(unsigned, len - copied,
					PAGE_SIZE - pg_offset);
1437
		unsigned pg_copied = copy_page_from_iter_atomic(page,
1438 1439 1440 1441
						pg_offset, pg_len, iter);

		if (!pg_copied)
			break;
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459

		flush_dcache_page(page);
		copied += pg_copied;
	}

	if (!copied)
		goto out;

	if (copied < len &&
	    ((offset + copied) & (PAGE_SIZE - 1))) {
		struct page *page = pages[(offset + copied) >> PAGE_SHIFT];

		if (!PageUptodate(page)) {
			zero_user(page, 0, PAGE_SIZE);
			copied -= (offset + copied) & (PAGE_SIZE - 1);
		}
	}

1460 1461 1462 1463 1464
	spin_lock(&inode->v.i_lock);
	if (pos + copied > inode->v.i_size)
		i_size_write(&inode->v, pos + copied);
	spin_unlock(&inode->v.i_lock);

1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479
	while (set_dirty < copied) {
		struct page *page = pages[(offset + set_dirty) >> PAGE_SHIFT];
		unsigned pg_offset = (offset + set_dirty) & (PAGE_SIZE - 1);
		unsigned pg_len = min_t(unsigned, copied - set_dirty,
					PAGE_SIZE - pg_offset);

		if (!PageUptodate(page))
			SetPageUptodate(page);

		bch2_set_page_dirty(c, inode, page, &res, pg_offset, pg_len);
		unlock_page(page);
		put_page(page);

		set_dirty += pg_len;
	}
1480 1481 1482

	nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE);
	inode->ei_last_dirtied = (unsigned long) current;
1483
out:
1484 1485 1486 1487 1488
	for (i = nr_pages_copied; i < nr_pages; i++) {
		unlock_page(pages[i]);
		put_page(pages[i]);
	}

1489 1490
	bch2_page_reservation_put(c, inode, &res);

1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595
	return copied ?: ret;
}

static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter)
{
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
	struct bch_inode_info *inode = file_bch_inode(file);
	loff_t pos = iocb->ki_pos;
	ssize_t written = 0;
	int ret = 0;

	bch2_pagecache_add_get(&inode->ei_pagecache_lock);

	do {
		unsigned offset = pos & (PAGE_SIZE - 1);
		unsigned bytes = min_t(unsigned long, iov_iter_count(iter),
			      PAGE_SIZE * WRITE_BATCH_PAGES - offset);
again:
		/*
		 * Bring in the user page that we will copy from _first_.
		 * Otherwise there's a nasty deadlock on copying from the
		 * same page as we're writing to, without it being marked
		 * up-to-date.
		 *
		 * Not only is this an optimisation, but it is also required
		 * to check that the address is actually valid, when atomic
		 * usercopies are used, below.
		 */
		if (unlikely(fault_in_iov_iter_readable(iter, bytes))) {
			bytes = min_t(unsigned long, iov_iter_count(iter),
				      PAGE_SIZE - offset);

			if (unlikely(fault_in_iov_iter_readable(iter, bytes))) {
				ret = -EFAULT;
				break;
			}
		}

		if (unlikely(fatal_signal_pending(current))) {
			ret = -EINTR;
			break;
		}

		ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes);
		if (unlikely(ret < 0))
			break;

		cond_resched();

		if (unlikely(ret == 0)) {
			/*
			 * If we were unable to copy any data at all, we must
			 * fall back to a single segment length write.
			 *
			 * If we didn't fallback here, we could livelock
			 * because not all segments in the iov can be copied at
			 * once without a pagefault.
			 */
			bytes = min_t(unsigned long, PAGE_SIZE - offset,
				      iov_iter_single_seg_count(iter));
			goto again;
		}
		pos += ret;
		written += ret;

		balance_dirty_pages_ratelimited(mapping);
	} while (iov_iter_count(iter));

	bch2_pagecache_add_put(&inode->ei_pagecache_lock);

	return written ? written : ret;
}

/* O_DIRECT reads */

static void bch2_dio_read_complete(struct closure *cl)
{
	struct dio_read *dio = container_of(cl, struct dio_read, cl);

	dio->req->ki_complete(dio->req, dio->ret);
	bio_check_pages_dirty(&dio->rbio.bio);	/* transfers ownership */
}

static void bch2_direct_IO_read_endio(struct bio *bio)
{
	struct dio_read *dio = bio->bi_private;

	if (bio->bi_status)
		dio->ret = blk_status_to_errno(bio->bi_status);

	closure_put(&dio->cl);
}

static void bch2_direct_IO_read_split_endio(struct bio *bio)
{
	bch2_direct_IO_read_endio(bio);
	bio_check_pages_dirty(bio);	/* transfers ownership */
}

static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
{
	struct file *file = req->ki_filp;
	struct bch_inode_info *inode = file_bch_inode(file);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1596
	struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701
	struct dio_read *dio;
	struct bio *bio;
	loff_t offset = req->ki_pos;
	bool sync = is_sync_kiocb(req);
	size_t shorten;
	ssize_t ret;

	if ((offset|iter->count) & (block_bytes(c) - 1))
		return -EINVAL;

	ret = min_t(loff_t, iter->count,
		    max_t(loff_t, 0, i_size_read(&inode->v) - offset));

	if (!ret)
		return ret;

	shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c));
	iter->count -= shorten;

	bio = bio_alloc_bioset(NULL,
			       iov_iter_npages(iter, BIO_MAX_VECS),
			       REQ_OP_READ,
			       GFP_KERNEL,
			       &c->dio_read_bioset);

	bio->bi_end_io = bch2_direct_IO_read_endio;

	dio = container_of(bio, struct dio_read, rbio.bio);
	closure_init(&dio->cl, NULL);

	/*
	 * this is a _really_ horrible hack just to avoid an atomic sub at the
	 * end:
	 */
	if (!sync) {
		set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL);
		atomic_set(&dio->cl.remaining,
			   CLOSURE_REMAINING_INITIALIZER -
			   CLOSURE_RUNNING +
			   CLOSURE_DESTRUCTOR);
	} else {
		atomic_set(&dio->cl.remaining,
			   CLOSURE_REMAINING_INITIALIZER + 1);
	}

	dio->req	= req;
	dio->ret	= ret;

	goto start;
	while (iter->count) {
		bio = bio_alloc_bioset(NULL,
				       iov_iter_npages(iter, BIO_MAX_VECS),
				       REQ_OP_READ,
				       GFP_KERNEL,
				       &c->bio_read);
		bio->bi_end_io		= bch2_direct_IO_read_split_endio;
start:
		bio->bi_opf		= REQ_OP_READ|REQ_SYNC;
		bio->bi_iter.bi_sector	= offset >> 9;
		bio->bi_private		= dio;

		ret = bio_iov_iter_get_pages(bio, iter);
		if (ret < 0) {
			/* XXX: fault inject this path */
			bio->bi_status = BLK_STS_RESOURCE;
			bio_endio(bio);
			break;
		}

		offset += bio->bi_iter.bi_size;
		bio_set_pages_dirty(bio);

		if (iter->count)
			closure_get(&dio->cl);

		bch2_read(c, rbio_init(bio, opts), inode->v.i_ino);
	}

	iter->count += shorten;

	if (sync) {
		closure_sync(&dio->cl);
		closure_debug_destroy(&dio->cl);
		ret = dio->ret;
		bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */
		return ret;
	} else {
		return -EIOCBQUEUED;
	}
}

ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter)
{
	struct file *file = iocb->ki_filp;
	struct bch_inode_info *inode = file_bch_inode(file);
	struct address_space *mapping = file->f_mapping;
	size_t count = iov_iter_count(iter);
	ssize_t ret;

	if (!count)
		return 0; /* skip atime */

	if (iocb->ki_flags & IOCB_DIRECT) {
		struct blk_plug plug;

1702 1703 1704 1705 1706 1707 1708
		if (unlikely(mapping->nrpages)) {
			ret = filemap_write_and_wait_range(mapping,
						iocb->ki_pos,
						iocb->ki_pos + count - 1);
			if (ret < 0)
				return ret;
		}
1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767

		file_accessed(file);

		blk_start_plug(&plug);
		ret = bch2_direct_IO_read(iocb, iter);
		blk_finish_plug(&plug);

		if (ret >= 0)
			iocb->ki_pos += ret;
	} else {
		bch2_pagecache_add_get(&inode->ei_pagecache_lock);
		ret = generic_file_read_iter(iocb, iter);
		bch2_pagecache_add_put(&inode->ei_pagecache_lock);
	}

	return ret;
}

/* O_DIRECT writes */

/*
 * We're going to return -EIOCBQUEUED, but we haven't finished consuming the
 * iov_iter yet, so we need to stash a copy of the iovec: it might be on the
 * caller's stack, we're not guaranteed that it will live for the duration of
 * the IO:
 */
static noinline int bch2_dio_write_copy_iov(struct dio_write *dio)
{
	struct iovec *iov = dio->inline_vecs;

	/*
	 * iov_iter has a single embedded iovec - nothing to do:
	 */
	if (iter_is_ubuf(&dio->iter))
		return 0;

	/*
	 * We don't currently handle non-iovec iov_iters here - return an error,
	 * and we'll fall back to doing the IO synchronously:
	 */
	if (!iter_is_iovec(&dio->iter))
		return -1;

	if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) {
		iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov),
				    GFP_KERNEL);
		if (unlikely(!iov))
			return -ENOMEM;

		dio->free_iov = true;
	}

	memcpy(iov, dio->iter.__iov, dio->iter.nr_segs * sizeof(*iov));
	dio->iter.__iov = iov;
	return 0;
}

static long bch2_dio_write_loop(struct dio_write *dio)
{
Kent Overstreet's avatar
Kent Overstreet committed
1768
	bool kthread = (current->flags & PF_KTHREAD) != 0;
1769
	struct bch_fs *c = dio->op.c;
1770 1771
	struct kiocb *req = dio->req;
	struct address_space *mapping = req->ki_filp->f_mapping;
1772 1773
	struct bch_inode_info *inode = file_bch_inode(req->ki_filp);
	struct bio *bio = &dio->op.wbio.bio;
1774 1775
	struct bvec_iter_all iter;
	struct bio_vec *bv;
1776
	unsigned unaligned;
1777
	u64 new_i_size;
1778 1779 1780 1781 1782 1783 1784
	bool sync;
	long ret;

	if (dio->loop)
		goto loop;

	while (1) {
Kent Overstreet's avatar
Kent Overstreet committed
1785 1786
		if (kthread)
			kthread_use_mm(dio->mm);
1787 1788 1789 1790 1791 1792
		BUG_ON(current->faults_disabled_mapping);
		current->faults_disabled_mapping = mapping;

		ret = bio_iov_iter_get_pages(bio, &dio->iter);

		current->faults_disabled_mapping = NULL;
Kent Overstreet's avatar
Kent Overstreet committed
1793 1794
		if (kthread)
			kthread_unuse_mm(dio->mm);
1795 1796 1797 1798

		if (unlikely(ret < 0))
			goto err;

1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813
		unaligned = bio->bi_iter.bi_size & (block_bytes(c) - 1);
		bio->bi_iter.bi_size -= unaligned;
		iov_iter_revert(&dio->iter, unaligned);

		if (!bio->bi_iter.bi_size) {
			/*
			 * bio_iov_iter_get_pages was only able to get <
			 * blocksize worth of pages:
			 */
			bio_for_each_segment_all(bv, bio, iter)
				put_page(bv->bv_page);
			ret = -EFAULT;
			goto err;
		}

1814 1815
		dio->op.pos = POS(inode->v.i_ino,
				  (req->ki_pos >> 9) + dio->op.written);
1816 1817 1818 1819 1820

		task_io_account_write(bio->bi_iter.bi_size);

		if (!dio->sync && !dio->loop && dio->iter.count) {
			if (bch2_dio_write_copy_iov(dio)) {
1821 1822
				dio->sync = true;
				goto do_io;
1823 1824
			}
		}
1825
do_io:
1826
		dio->loop = true;
1827
		closure_call(&dio->op.cl, bch2_write, NULL, NULL);
1828

1829 1830 1831
		if (dio->sync)
			wait_for_completion(&dio->done);
		else
1832 1833
			return -EIOCBQUEUED;
loop:
1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844
		i_sectors_acct(c, inode, &dio->quota_res,
			       dio->op.i_sectors_delta);
		dio->op.i_sectors_delta = 0;

		new_i_size = req->ki_pos + ((u64) dio->op.written << 9);

		spin_lock(&inode->v.i_lock);
		if (new_i_size > inode->v.i_size)
			i_size_write(&inode->v, new_i_size);
		spin_unlock(&inode->v.i_lock);

1845 1846
		bio_for_each_segment_all(bv, bio, iter)
			put_page(bv->bv_page);
1847
		if (!dio->iter.count || dio->op.error)
1848
			break;
1849

1850
		bio_reset(bio, NULL, REQ_OP_WRITE);
1851
		reinit_completion(&dio->done);
1852 1853
	}

1854
	ret = dio->op.error ?: ((long) dio->op.written << 9);
1855 1856
err:
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
1857
	bch2_disk_reservation_put(c, &dio->op.res);
1858
	bch2_quota_reservation_put(c, inode, &dio->quota_res);
1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875

	if (dio->free_iov)
		kfree(dio->iter.__iov);

	sync = dio->sync;
	bio_put(bio);

	/* inode->i_dio_count is our ref on inode and thus bch_fs */
	inode_dio_end(&inode->v);

	if (!sync) {
		req->ki_complete(req, ret);
		ret = -EIOCBQUEUED;
	}
	return ret;
}

1876
static void bch2_dio_write_loop_async(struct bch_write_op *op)
1877
{
1878
	struct dio_write *dio = container_of(op, struct dio_write, op);
1879

1880 1881 1882 1883
	if (dio->sync)
		complete(&dio->done);
	else
		bch2_dio_write_loop(dio);
1884 1885 1886 1887 1888 1889
}

static noinline
ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
{
	struct file *file = req->ki_filp;
1890
	struct address_space *mapping = file->f_mapping;
1891 1892
	struct bch_inode_info *inode = file_bch_inode(file);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
1893
	struct bch_io_opts opts = io_opts(c, &inode->ei_inode);
1894 1895
	struct dio_write *dio;
	struct bio *bio;
1896
	bool locked = true, extending;
1897 1898
	ssize_t ret;

1899 1900 1901 1902
	prefetch(&c->opts);
	prefetch((void *) &c->opts + 64);
	prefetch(&inode->ei_inode);
	prefetch((void *) &inode->ei_inode + 64);
1903

1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916
	inode_lock(&inode->v);

	ret = generic_write_checks(req, iter);
	if (unlikely(ret <= 0))
		goto err;

	ret = file_remove_privs(file);
	if (unlikely(ret))
		goto err;

	ret = file_update_time(file);
	if (unlikely(ret))
		goto err;
1917

1918
	if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1)))
1919 1920 1921 1922 1923 1924 1925 1926 1927 1928
		goto err;

	inode_dio_begin(&inode->v);
	bch2_pagecache_block_get(&inode->ei_pagecache_lock);

	extending = req->ki_pos + iter->count > inode->v.i_size;
	if (!extending) {
		inode_unlock(&inode->v);
		locked = false;
	}
1929 1930 1931 1932 1933 1934

	bio = bio_alloc_bioset(NULL,
			       iov_iter_npages(iter, BIO_MAX_VECS),
			       REQ_OP_WRITE,
			       GFP_KERNEL,
			       &c->dio_write_bioset);
1935
	dio = container_of(bio, struct dio_write, op.wbio.bio);
1936
	init_completion(&dio->done);
1937
	dio->req		= req;
Kent Overstreet's avatar
Kent Overstreet committed
1938
	dio->mm			= current->mm;
1939
	dio->loop		= false;
1940
	dio->sync		= is_sync_kiocb(req) || extending;
1941 1942 1943
	dio->free_iov		= false;
	dio->quota_res.sectors	= 0;
	dio->iter		= *iter;
1944 1945

	bch2_write_op_init(&dio->op, c, opts);
1946
	dio->op.end_io		= bch2_dio_write_loop_async;
1947 1948 1949 1950
	dio->op.target		= opts.foreground_target;
	op_journal_seq_set(&dio->op, &inode->ei_journal_seq);
	dio->op.write_point	= writepoint_hashed((unsigned long) current);
	dio->op.flags |= BCH_WRITE_NOPUT_RESERVATION;
1951 1952 1953

	if ((req->ki_flags & IOCB_DSYNC) &&
	    !c->opts.journal_flush_disabled)
1954
		dio->op.flags |= BCH_WRITE_FLUSH;
1955 1956 1957 1958

	ret = bch2_quota_reservation_add(c, inode, &dio->quota_res,
					 iter->count >> 9, true);
	if (unlikely(ret))
1959
		goto err_put_bio;
1960

1961
	dio->op.nr_replicas	= dio->op.opts.data_replicas;
1962

1963 1964 1965 1966 1967 1968 1969
	ret = bch2_disk_reservation_get(c, &dio->op.res, iter->count >> 9,
					dio->op.opts.data_replicas, 0);
	if (unlikely(ret) &&
	    !bch2_check_range_allocated(c, POS(inode->v.i_ino,
					       req->ki_pos >> 9),
					iter->count >> 9,
					dio->op.opts.data_replicas))
1970
		goto err_put_bio;
1971

1972 1973 1974 1975 1976 1977 1978
	if (unlikely(mapping->nrpages)) {
		ret = write_invalidate_inode_pages_range(mapping,
						req->ki_pos,
						req->ki_pos + iter->count - 1);
		if (unlikely(ret))
			goto err_put_bio;
	}
1979

1980
	ret = bch2_dio_write_loop(dio);
1981
err:
1982 1983 1984 1985 1986 1987 1988
	if (locked)
		inode_unlock(&inode->v);
	if (ret > 0)
		req->ki_pos += ret;
	return ret;
err_put_bio:
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
1989
	bch2_disk_reservation_put(c, &dio->op.res);
1990 1991
	bch2_quota_reservation_put(c, inode, &dio->quota_res);
	bio_put(bio);
1992 1993
	inode_dio_end(&inode->v);
	goto err;
1994 1995
}

1996
ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from)
1997 1998
{
	struct file *file = iocb->ki_filp;
1999
	struct bch_inode_info *inode = file_bch_inode(file);
2000 2001 2002 2003 2004
	ssize_t	ret;

	if (iocb->ki_flags & IOCB_DIRECT)
		return bch2_direct_write(iocb, from);

2005 2006 2007 2008 2009 2010
	inode_lock(&inode->v);

	ret = generic_write_checks(iocb, from);
	if (ret <= 0)
		goto unlock;

2011 2012
	ret = file_remove_privs(file);
	if (ret)
2013
		goto unlock;
2014 2015 2016

	ret = file_update_time(file);
	if (ret)
2017
		goto unlock;
2018

2019
	ret = bch2_buffered_write(iocb, from);
2020 2021
	if (likely(ret > 0))
		iocb->ki_pos += ret;
2022
unlock:
2023 2024
	inode_unlock(&inode->v);

2025
	if (ret > 0)
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036
		ret = generic_write_sync(iocb, ret);

	return ret;
}

/* fsync: */

int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
{
	struct bch_inode_info *inode = file_bch_inode(file);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
2037
	int ret, ret2;
2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049

	ret = file_write_and_wait_range(file, start, end);
	if (ret)
		return ret;

	if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC))
		goto out;

	ret = sync_inode_metadata(&inode->v, 1);
	if (ret)
		return ret;
out:
2050 2051 2052
	if (!c->opts.journal_flush_disabled)
		ret = bch2_journal_flush_seq(&c->journal,
					     inode->ei_journal_seq);
2053 2054 2055
	ret2 = file_check_and_advance_wb_err(file);

	return ret ?: ret2;
2056 2057 2058 2059 2060 2061 2062 2063
}

/* truncate: */

static inline int range_has_data(struct bch_fs *c,
				  struct bpos start,
				  struct bpos end)
{
2064 2065
	struct btree_trans trans;
	struct btree_iter *iter;
2066 2067 2068
	struct bkey_s_c k;
	int ret = 0;

2069
	bch2_trans_init(&trans, c, 0, 0);
2070

2071
	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, start, 0, k, ret) {
2072 2073 2074 2075 2076 2077 2078 2079 2080
		if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
			break;

		if (bkey_extent_is_data(k.k)) {
			ret = 1;
			break;
		}
	}

2081
	return bch2_trans_exit(&trans) ?: ret;
2082 2083 2084 2085 2086 2087 2088
}

static int __bch2_truncate_page(struct bch_inode_info *inode,
				pgoff_t index, loff_t start, loff_t end)
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct address_space *mapping = inode->v.i_mapping;
2089
	struct bch_page_state *s;
2090 2091
	unsigned start_offset = start & (PAGE_SIZE - 1);
	unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1;
2092
	unsigned i;
2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123
	struct page *page;
	int ret = 0;

	/* Page boundary? Nothing to do */
	if (!((index == start >> PAGE_SHIFT && start_offset) ||
	      (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE)))
		return 0;

	/* Above i_size? */
	if (index << PAGE_SHIFT >= inode->v.i_size)
		return 0;

	page = find_lock_page(mapping, index);
	if (!page) {
		/*
		 * XXX: we're doing two index lookups when we end up reading the
		 * page
		 */
		ret = range_has_data(c,
				POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT),
				POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT));
		if (ret <= 0)
			return ret;

		page = find_or_create_page(mapping, index, GFP_KERNEL);
		if (unlikely(!page)) {
			ret = -ENOMEM;
			goto out;
		}
	}

2124 2125 2126 2127 2128 2129
	s = bch2_page_state_create(page, 0);
	if (!s) {
		ret = -ENOMEM;
		goto unlock;
	}

2130 2131 2132 2133 2134 2135
	if (!PageUptodate(page)) {
		ret = bch2_read_single_page(page, mapping);
		if (ret)
			goto unlock;
	}

2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149
	if (index != start >> PAGE_SHIFT)
		start_offset = 0;
	if (index != end >> PAGE_SHIFT)
		end_offset = PAGE_SIZE;

	for (i = round_up(start_offset, block_bytes(c)) >> 9;
	     i < round_down(end_offset, block_bytes(c)) >> 9;
	     i++) {
		s->s[i].nr_replicas	= 0;
		s->s[i].state		= SECTOR_UNALLOCATED;
	}

	zero_user_segment(page, start_offset, end_offset);

2150 2151 2152 2153 2154 2155
	/*
	 * Bit of a hack - we don't want truncate to fail due to -ENOSPC.
	 *
	 * XXX: because we aren't currently tracking whether the page has actual
	 * data in it (vs. just 0s, or only partially written) this wrong. ick.
	 */
2156
	ret = bch2_get_page_disk_reservation(c, inode, page, false);
2157 2158
	BUG_ON(ret);

2159
	filemap_dirty_folio(mapping, page_folio(page));
2160 2161 2162 2163 2164 2165 2166 2167 2168 2169
unlock:
	unlock_page(page);
	put_page(page);
out:
	return ret;
}

static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from)
{
	return __bch2_truncate_page(inode, from >> PAGE_SHIFT,
2170
				    from, round_up(from, PAGE_SIZE));
2171 2172
}

2173 2174 2175
static int bch2_extend(struct bch_inode_info *inode,
		       struct bch_inode_unpacked *inode_u,
		       struct iattr *iattr)
2176 2177 2178 2179 2180
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct address_space *mapping = inode->v.i_mapping;
	int ret;

2181 2182
	/*
	 * sync appends:
2183 2184
	 *
	 * this has to be done _before_ extending i_size:
2185 2186
	 */
	ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
2187 2188 2189 2190 2191 2192 2193 2194
	if (ret)
		return ret;

	truncate_setsize(&inode->v, iattr->ia_size);
	/* ATTR_MODE will never be set here, ns argument isn't needed: */
	setattr_copy(NULL, &inode->v, iattr);

	mutex_lock(&inode->ei_update_lock);
2195 2196
	ret = bch2_write_inode_size(c, inode, inode->v.i_size,
				    ATTR_MTIME|ATTR_CTIME);
2197 2198 2199 2200 2201
	mutex_unlock(&inode->ei_update_lock);

	return ret;
}

2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222
static int bch2_truncate_finish_fn(struct bch_inode_info *inode,
				   struct bch_inode_unpacked *bi,
				   void *p)
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;

	bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
	bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
	return 0;
}

static int bch2_truncate_start_fn(struct bch_inode_info *inode,
				  struct bch_inode_unpacked *bi, void *p)
{
	u64 *new_i_size = p;

	bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY;
	bi->bi_size = *new_i_size;
	return 0;
}

2223 2224 2225 2226
int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr)
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct address_space *mapping = inode->v.i_mapping;
2227 2228 2229
	struct bch_inode_unpacked inode_u;
	struct btree_trans trans;
	struct btree_iter *iter;
2230
	u64 new_i_size = iattr->ia_size;
2231
	s64 i_sectors_delta = 0;
2232 2233 2234 2235 2236
	int ret = 0;

	inode_dio_wait(&inode->v);
	bch2_pagecache_block_get(&inode->ei_pagecache_lock);

2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247
	/*
	 * fetch current on disk i_size: inode is locked, i_size can only
	 * increase underneath us:
	 */
	bch2_trans_init(&trans, c, 0, 0);
	iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino, 0);
	ret = PTR_ERR_OR_ZERO(iter);
	bch2_trans_exit(&trans);

	if (ret)
		goto err;
2248

2249
	BUG_ON(inode->v.i_size < inode_u.bi_size);
2250

2251 2252
	if (iattr->ia_size > inode->v.i_size) {
		ret = bch2_extend(inode, &inode_u, iattr);
2253
		goto err;
2254 2255 2256 2257
	}

	ret = bch2_truncate_page(inode, iattr->ia_size);
	if (unlikely(ret))
2258
		goto err;
2259

2260 2261 2262 2263 2264 2265 2266 2267 2268 2269
	/*
	 * When extending, we're going to write the new i_size to disk
	 * immediately so we need to flush anything above the current on disk
	 * i_size first:
	 *
	 * Also, when extending we need to flush the page that i_size currently
	 * straddles - if it's mapped to userspace, we need to ensure that
	 * userspace has to redirty it and call .mkwrite -> set_page_dirty
	 * again to allocate the part of the page that was extended.
	 */
2270
	if (iattr->ia_size > inode_u.bi_size)
2271
		ret = filemap_write_and_wait_range(mapping,
2272
				inode_u.bi_size,
2273 2274 2275 2276 2277 2278
				iattr->ia_size - 1);
	else if (iattr->ia_size & (PAGE_SIZE - 1))
		ret = filemap_write_and_wait_range(mapping,
				round_down(iattr->ia_size, PAGE_SIZE),
				iattr->ia_size - 1);
	if (ret)
2279
		goto err;
2280

2281 2282 2283 2284
	mutex_lock(&inode->ei_update_lock);
	ret = bch2_write_inode(c, inode, bch2_truncate_start_fn,
			       &new_i_size, 0);
	mutex_unlock(&inode->ei_update_lock);
2285 2286

	if (unlikely(ret))
2287
		goto err;
2288 2289 2290

	truncate_setsize(&inode->v, iattr->ia_size);

2291
	ret = bch2_fpunch(c, inode->v.i_ino,
2292
			round_up(iattr->ia_size, block_bytes(c)) >> 9,
2293 2294 2295
			U64_MAX, &inode->ei_journal_seq, &i_sectors_delta);
	i_sectors_acct(c, inode, NULL, i_sectors_delta);

2296
	if (unlikely(ret))
2297
		goto err;
2298 2299 2300

	/* ATTR_MODE will never be set here, ns argument isn't needed: */
	setattr_copy(NULL, &inode->v, iattr);
2301 2302 2303 2304 2305 2306

	mutex_lock(&inode->ei_update_lock);
	ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL,
			       ATTR_MTIME|ATTR_CTIME);
	mutex_unlock(&inode->ei_update_lock);
err:
2307 2308 2309 2310 2311 2312
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
	return ret;
}

/* fallocate: */

2313
static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
2314 2315
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
2316 2317
	u64 discard_start = round_up(offset, block_bytes(c)) >> 9;
	u64 discard_end = round_down(offset + len, block_bytes(c)) >> 9;
2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
	int ret = 0;

	inode_lock(&inode->v);
	inode_dio_wait(&inode->v);
	bch2_pagecache_block_get(&inode->ei_pagecache_lock);

	ret = __bch2_truncate_page(inode,
				   offset >> PAGE_SHIFT,
				   offset, offset + len);
	if (unlikely(ret))
		goto err;

	if (offset >> PAGE_SHIFT !=
	    (offset + len) >> PAGE_SHIFT) {
		ret = __bch2_truncate_page(inode,
					   (offset + len) >> PAGE_SHIFT,
					   offset, offset + len);
		if (unlikely(ret))
			goto err;
	}

	truncate_pagecache_range(&inode->v, offset, offset + len - 1);

2341 2342 2343 2344 2345 2346 2347 2348 2349
	if (discard_start < discard_end) {
		s64 i_sectors_delta = 0;

		ret = bch2_fpunch(c, inode->v.i_ino,
				  discard_start, discard_end,
				  &inode->ei_journal_seq,
				  &i_sectors_delta);
		i_sectors_acct(c, inode, NULL, i_sectors_delta);
	}
2350 2351 2352 2353 2354 2355 2356
err:
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
	inode_unlock(&inode->v);

	return ret;
}

2357
static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
2358 2359
				   loff_t offset, loff_t len,
				   bool insert)
2360 2361 2362
{
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	struct address_space *mapping = inode->v.i_mapping;
Kent Overstreet's avatar
Kent Overstreet committed
2363
	struct bkey_on_stack copy;
2364
	struct btree_trans trans;
2365
	struct btree_iter *src, *dst, *del = NULL;
2366 2367
	loff_t shift, new_size;
	u64 src_start;
2368 2369 2370 2371 2372
	int ret;

	if ((offset | len) & (block_bytes(c) - 1))
		return -EINVAL;

Kent Overstreet's avatar
Kent Overstreet committed
2373
	bkey_on_stack_init(&copy);
2374
	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 256);
2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385

	/*
	 * We need i_mutex to keep the page cache consistent with the extents
	 * btree, and the btree consistent with i_size - we don't need outside
	 * locking for the extents btree itself, because we're using linked
	 * iterators
	 */
	inode_lock(&inode->v);
	inode_dio_wait(&inode->v);
	bch2_pagecache_block_get(&inode->ei_pagecache_lock);

2386 2387 2388 2389
	if (insert) {
		ret = -EFBIG;
		if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len)
			goto err;
2390

2391 2392 2393
		ret = -EINVAL;
		if (offset >= inode->v.i_size)
			goto err;
2394

2395 2396 2397 2398 2399 2400
		src_start	= U64_MAX;
		shift		= len;
	} else {
		ret = -EINVAL;
		if (offset + len >= inode->v.i_size)
			goto err;
2401

2402 2403 2404 2405 2406
		src_start	= offset + len;
		shift		= -len;
	}

	new_size = inode->v.i_size + shift;
2407

2408
	ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
2409 2410 2411
	if (ret)
		goto err;

2412 2413 2414 2415 2416 2417 2418
	if (insert) {
		i_size_write(&inode->v, new_size);
		mutex_lock(&inode->ei_update_lock);
		ret = bch2_write_inode_size(c, inode, new_size,
					    ATTR_MTIME|ATTR_CTIME);
		mutex_unlock(&inode->ei_update_lock);
	} else {
2419 2420 2421 2422 2423 2424 2425 2426
		s64 i_sectors_delta = 0;

		ret = bch2_fpunch(c, inode->v.i_ino,
				  offset >> 9, (offset + len) >> 9,
				  &inode->ei_journal_seq,
				  &i_sectors_delta);
		i_sectors_acct(c, inode, NULL, i_sectors_delta);

2427 2428 2429
		if (ret)
			goto err;
	}
2430 2431

	src = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
2432
			POS(inode->v.i_ino, src_start >> 9),
2433
			BTREE_ITER_INTENT);
2434
	BUG_ON(IS_ERR_OR_NULL(src));
2435

2436 2437 2438
	dst = bch2_trans_copy_iter(&trans, src);
	BUG_ON(IS_ERR_OR_NULL(dst));

2439 2440 2441 2442 2443 2444
	while (1) {
		struct disk_reservation disk_res =
			bch2_disk_reservation_init(c, 0);
		struct bkey_i delete;
		struct bkey_s_c k;
		struct bpos next_pos;
2445 2446
		struct bpos move_pos = POS(inode->v.i_ino, offset >> 9);
		struct bpos atomic_end;
2447 2448 2449
		unsigned commit_flags = BTREE_INSERT_NOFAIL|
			BTREE_INSERT_ATOMIC|
			BTREE_INSERT_USE_RESERVE;
2450

2451 2452 2453
		k = insert
			? bch2_btree_iter_peek_prev(src)
			: bch2_btree_iter_peek(src);
2454
		if ((ret = bkey_err(k)))
2455
			goto bkey_err;
2456

2457 2458
		if (!k.k || k.k->p.inode != inode->v.i_ino)
			break;
2459

2460 2461 2462 2463 2464 2465
		BUG_ON(bkey_cmp(src->pos, bkey_start_pos(k.k)));

		if (insert &&
		    bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0)
			break;
reassemble:
Kent Overstreet's avatar
Kent Overstreet committed
2466 2467
		bkey_on_stack_realloc(&copy, c, k.k->u64s);
		bkey_reassemble(copy.k, k);
2468 2469 2470

		if (insert &&
		    bkey_cmp(bkey_start_pos(k.k), move_pos) < 0) {
Kent Overstreet's avatar
Kent Overstreet committed
2471 2472
			bch2_cut_front(move_pos, copy.k);
			bch2_btree_iter_set_pos(src, bkey_start_pos(&copy.k->k));
2473
		}
2474

Kent Overstreet's avatar
Kent Overstreet committed
2475 2476
		copy.k->k.p.offset += shift >> 9;
		bch2_btree_iter_set_pos(dst, bkey_start_pos(&copy.k->k));
2477

Kent Overstreet's avatar
Kent Overstreet committed
2478
		ret = bch2_extent_atomic_end(dst, copy.k, &atomic_end);
2479 2480
		if (ret)
			goto bkey_err;
2481

Kent Overstreet's avatar
Kent Overstreet committed
2482
		if (bkey_cmp(atomic_end, copy.k->k.p)) {
2483 2484 2485 2486 2487
			if (insert) {
				move_pos = atomic_end;
				move_pos.offset -= shift >> 9;
				goto reassemble;
			} else {
2488
				bch2_cut_back(atomic_end, copy.k);
2489 2490 2491
			}
		}

2492 2493
		bkey_init(&delete.k);
		delete.k.p = src->pos;
Kent Overstreet's avatar
Kent Overstreet committed
2494
		bch2_key_resize(&delete.k, copy.k->k.size);
2495

2496
		next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p;
2497

2498 2499 2500 2501 2502 2503 2504 2505
		/*
		 * If the new and old keys overlap (because we're moving an
		 * extent that's bigger than the amount we're collapsing by),
		 * we need to trim the delete key here so they don't overlap
		 * because overlaps on insertions aren't handled before
		 * triggers are run, so the overwrite will get double counted
		 * by the triggers machinery:
		 */
2506
		if (insert &&
Kent Overstreet's avatar
Kent Overstreet committed
2507
		    bkey_cmp(bkey_start_pos(&copy.k->k), delete.k.p) < 0) {
2508
			bch2_cut_back(bkey_start_pos(&copy.k->k), &delete);
2509
		} else if (!insert &&
Kent Overstreet's avatar
Kent Overstreet committed
2510
			   bkey_cmp(copy.k->k.p,
2511
				    bkey_start_pos(&delete.k)) > 0) {
Kent Overstreet's avatar
Kent Overstreet committed
2512
			bch2_cut_front(copy.k->k.p, &delete);
2513

2514 2515 2516 2517 2518 2519 2520
			del = bch2_trans_copy_iter(&trans, src);
			BUG_ON(IS_ERR_OR_NULL(del));

			bch2_btree_iter_set_pos(del,
				bkey_start_pos(&delete.k));
		}

Kent Overstreet's avatar
Kent Overstreet committed
2521
		bch2_trans_update(&trans, dst, copy.k);
2522
		bch2_trans_update(&trans, del ?: src, &delete);
2523

Kent Overstreet's avatar
Kent Overstreet committed
2524
		if (copy.k->k.size == k.k->size) {
2525 2526 2527 2528 2529 2530 2531 2532
			/*
			 * If we're moving the entire extent, we can skip
			 * running triggers:
			 */
			commit_flags |= BTREE_INSERT_NOMARK;
		} else {
			/* We might end up splitting compressed extents: */
			unsigned nr_ptrs =
Kent Overstreet's avatar
Kent Overstreet committed
2533
				bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(copy.k));
2534 2535

			ret = bch2_disk_reservation_get(c, &disk_res,
Kent Overstreet's avatar
Kent Overstreet committed
2536
					copy.k->k.size, nr_ptrs,
2537 2538 2539 2540 2541 2542 2543
					BCH_DISK_RESERVATION_NOFAIL);
			BUG_ON(ret);
		}

		ret = bch2_trans_commit(&trans, &disk_res,
					&inode->ei_journal_seq,
					commit_flags);
2544
		bch2_disk_reservation_put(c, &disk_res);
2545
bkey_err:
2546
		if (del)
2547
			bch2_trans_iter_put(&trans, del);
2548 2549 2550 2551 2552
		del = NULL;

		if (!ret)
			bch2_btree_iter_set_pos(src, next_pos);

2553 2554
		if (ret == -EINTR)
			ret = 0;
2555 2556
		if (ret)
			goto err;
2557

2558
		bch2_trans_cond_resched(&trans);
2559
	}
2560
	bch2_trans_unlock(&trans);
2561

2562 2563 2564 2565 2566 2567 2568
	if (!insert) {
		i_size_write(&inode->v, new_size);
		mutex_lock(&inode->ei_update_lock);
		ret = bch2_write_inode_size(c, inode, new_size,
					    ATTR_MTIME|ATTR_CTIME);
		mutex_unlock(&inode->ei_update_lock);
	}
2569
err:
2570
	bch2_trans_exit(&trans);
Kent Overstreet's avatar
Kent Overstreet committed
2571
	bkey_on_stack_exit(&copy, c);
2572 2573 2574 2575 2576
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
	inode_unlock(&inode->v);
	return ret;
}

2577 2578
static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
			    loff_t offset, loff_t len)
2579 2580 2581
{
	struct address_space *mapping = inode->v.i_mapping;
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
2582 2583
	struct btree_trans trans;
	struct btree_iter *iter;
2584
	struct bpos end_pos;
2585 2586 2587
	loff_t end		= offset + len;
	loff_t block_start	= round_down(offset,	block_bytes(c));
	loff_t block_end	= round_up(end,		block_bytes(c));
2588
	unsigned sectors;
2589
	unsigned replicas = io_opts(c, &inode->ei_inode).data_replicas;
2590 2591
	int ret;

2592
	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620

	inode_lock(&inode->v);
	inode_dio_wait(&inode->v);
	bch2_pagecache_block_get(&inode->ei_pagecache_lock);

	if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
		ret = inode_newsize_ok(&inode->v, end);
		if (ret)
			goto err;
	}

	if (mode & FALLOC_FL_ZERO_RANGE) {
		ret = __bch2_truncate_page(inode,
					   offset >> PAGE_SHIFT,
					   offset, end);

		if (!ret &&
		    offset >> PAGE_SHIFT != end >> PAGE_SHIFT)
			ret = __bch2_truncate_page(inode,
						   end >> PAGE_SHIFT,
						   offset, end);

		if (unlikely(ret))
			goto err;

		truncate_pagecache_range(&inode->v, offset, end - 1);
	}

2621 2622 2623
	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
			POS(inode->v.i_ino, block_start >> 9),
			BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
2624 2625
	end_pos = POS(inode->v.i_ino, block_end >> 9);

2626
	while (bkey_cmp(iter->pos, end_pos) < 0) {
2627
		s64 i_sectors_delta = 0;
2628
		struct disk_reservation disk_res = { 0 };
2629
		struct quota_res quota_res = { 0 };
2630 2631 2632
		struct bkey_i_reservation reservation;
		struct bkey_s_c k;

2633
		k = bch2_btree_iter_peek_slot(iter);
2634 2635
		if ((ret = bkey_err(k)))
			goto bkey_err;
2636 2637

		/* already reserved */
2638
		if (k.k->type == KEY_TYPE_reservation &&
2639
		    bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) {
2640
			bch2_btree_iter_next_slot(iter);
2641 2642 2643
			continue;
		}

2644 2645 2646 2647
		if (bkey_extent_is_data(k.k) &&
		    !(mode & FALLOC_FL_ZERO_RANGE)) {
			bch2_btree_iter_next_slot(iter);
			continue;
2648 2649 2650
		}

		bkey_reservation_init(&reservation.k_i);
2651
		reservation.k.type	= KEY_TYPE_reservation;
2652 2653 2654
		reservation.k.p		= k.k->p;
		reservation.k.size	= k.k->size;

2655 2656
		bch2_cut_front(iter->pos,	&reservation.k_i);
		bch2_cut_back(end_pos,		&reservation.k_i);
2657 2658

		sectors = reservation.k.size;
2659
		reservation.v.nr_replicas = bch2_bkey_nr_dirty_ptrs(k);
2660 2661 2662

		if (!bkey_extent_is_allocation(k.k)) {
			ret = bch2_quota_reservation_add(c, inode,
2663
					&quota_res,
2664 2665
					sectors, true);
			if (unlikely(ret))
2666
				goto bkey_err;
2667 2668 2669 2670 2671 2672 2673
		}

		if (reservation.v.nr_replicas < replicas ||
		    bch2_extent_is_compressed(k)) {
			ret = bch2_disk_reservation_get(c, &disk_res, sectors,
							replicas, 0);
			if (unlikely(ret))
2674
				goto bkey_err;
2675 2676 2677 2678

			reservation.v.nr_replicas = disk_res.nr_replicas;
		}

2679 2680
		bch2_trans_begin_updates(&trans);

2681 2682 2683 2684
		ret = bch2_extent_update(&trans, iter, &reservation.k_i,
				&disk_res, &inode->ei_journal_seq,
				0, &i_sectors_delta);
		i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
2685
bkey_err:
2686
		bch2_quota_reservation_put(c, inode, &quota_res);
2687 2688 2689
		bch2_disk_reservation_put(c, &disk_res);
		if (ret == -EINTR)
			ret = 0;
2690 2691
		if (ret)
			goto err;
2692 2693
	}

2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712
	/*
	 * Do we need to extend the file?
	 *
	 * If we zeroed up to the end of the file, we dropped whatever writes
	 * were going to write out the current i_size, so we have to extend
	 * manually even if FL_KEEP_SIZE was set:
	 */
	if (end >= inode->v.i_size &&
	    (!(mode & FALLOC_FL_KEEP_SIZE) ||
	     (mode & FALLOC_FL_ZERO_RANGE))) {
		struct btree_iter *inode_iter;
		struct bch_inode_unpacked inode_u;

		do {
			bch2_trans_begin(&trans);
			inode_iter = bch2_inode_peek(&trans, &inode_u,
						     inode->v.i_ino, 0);
			ret = PTR_ERR_OR_ZERO(inode_iter);
		} while (ret == -EINTR);
2713

2714 2715 2716 2717
		bch2_trans_unlock(&trans);

		if (ret)
			goto err;
2718

2719 2720 2721 2722
		/*
		 * Sync existing appends before extending i_size,
		 * as in bch2_extend():
		 */
2723
		ret = filemap_write_and_wait_range(mapping,
2724
					inode_u.bi_size, S64_MAX);
2725 2726 2727
		if (ret)
			goto err;

2728 2729 2730 2731 2732 2733 2734 2735
		if (mode & FALLOC_FL_KEEP_SIZE)
			end = inode->v.i_size;
		else
			i_size_write(&inode->v, end);

		mutex_lock(&inode->ei_update_lock);
		ret = bch2_write_inode_size(c, inode, end, 0);
		mutex_unlock(&inode->ei_update_lock);
2736 2737
	}
err:
2738
	bch2_trans_exit(&trans);
2739 2740 2741 2742 2743 2744 2745 2746 2747
	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
	inode_unlock(&inode->v);
	return ret;
}

long bch2_fallocate_dispatch(struct file *file, int mode,
			     loff_t offset, loff_t len)
{
	struct bch_inode_info *inode = file_bch_inode(file);
2748 2749
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
	long ret;
2750

2751 2752
	if (!percpu_ref_tryget(&c->writes))
		return -EROFS;
2753

2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765
	if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
		ret = bchfs_fallocate(inode, mode, offset, len);
	else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
		ret = bchfs_fpunch(inode, offset, len);
	else if (mode == FALLOC_FL_INSERT_RANGE)
		ret = bchfs_fcollapse_finsert(inode, offset, len, true);
	else if (mode == FALLOC_FL_COLLAPSE_RANGE)
		ret = bchfs_fcollapse_finsert(inode, offset, len, false);
	else
		ret = -EOPNOTSUPP;

	percpu_ref_put(&c->writes);
2766

2767
	return ret;
2768 2769
}

Kent Overstreet's avatar
Kent Overstreet committed
2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788
static void mark_range_unallocated(struct bch_inode_info *inode,
				   loff_t start, loff_t end)
{
	pgoff_t index = start >> PAGE_SHIFT;
	pgoff_t end_index = (end - 1) >> PAGE_SHIFT;
	struct folio_batch fbatch;
	unsigned i, j;

	folio_batch_init(&fbatch);

	while (filemap_get_folios(inode->v.i_mapping,
				  &index, end_index, &fbatch)) {
		for (i = 0; i < folio_batch_count(&fbatch); i++) {
			struct folio *folio = fbatch.folios[i];
			struct bch_page_state *s;

			folio_lock(folio);
			s = bch2_page_state(&folio->page);

2789 2790
			if (s) {
				spin_lock(&s->lock);
Kent Overstreet's avatar
Kent Overstreet committed
2791 2792
				for (j = 0; j < PAGE_SECTORS; j++)
					s->s[j].nr_replicas = 0;
2793 2794
				spin_unlock(&s->lock);
			}
Kent Overstreet's avatar
Kent Overstreet committed
2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809

			folio_unlock(folio);
		}
		folio_batch_release(&fbatch);
		cond_resched();
	}
}

loff_t 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)
{
	struct bch_inode_info *src = file_bch_inode(file_src);
	struct bch_inode_info *dst = file_bch_inode(file_dst);
	struct bch_fs *c = src->v.i_sb->s_fs_info;
2810
	s64 i_sectors_delta = 0;
2811
	u64 aligned_len;
Kent Overstreet's avatar
Kent Overstreet committed
2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829
	loff_t ret = 0;

	if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
		return -EINVAL;

	if (remap_flags & REMAP_FILE_DEDUP)
		return -EOPNOTSUPP;

	if ((pos_src & (block_bytes(c) - 1)) ||
	    (pos_dst & (block_bytes(c) - 1)))
		return -EINVAL;

	if (src == dst &&
	    abs(pos_src - pos_dst) < len)
		return -EINVAL;

	bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);

2830 2831
	file_update_time(file_dst);

Kent Overstreet's avatar
Kent Overstreet committed
2832 2833 2834 2835 2836 2837 2838
	inode_dio_wait(&src->v);
	inode_dio_wait(&dst->v);

	ret = generic_remap_file_range_prep(file_src, pos_src,
					    file_dst, pos_dst,
					    &len, remap_flags);
	if (ret < 0 || len == 0)
2839
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
2840

2841
	aligned_len = round_up((u64) len, block_bytes(c));
Kent Overstreet's avatar
Kent Overstreet committed
2842 2843

	ret = write_invalidate_inode_pages_range(dst->v.i_mapping,
2844
				pos_dst, pos_dst + len - 1);
Kent Overstreet's avatar
Kent Overstreet committed
2845
	if (ret)
2846
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
2847 2848 2849

	mark_range_unallocated(src, pos_src, pos_src + aligned_len);

2850
	ret = bch2_remap_range(c,
Kent Overstreet's avatar
Kent Overstreet committed
2851 2852 2853
			       POS(dst->v.i_ino, pos_dst >> 9),
			       POS(src->v.i_ino, pos_src >> 9),
			       aligned_len >> 9,
2854 2855 2856 2857
			       &dst->ei_journal_seq,
			       pos_dst + len, &i_sectors_delta);
	if (ret < 0)
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
2858

2859 2860 2861
	/*
	 * due to alignment, we might have remapped slightly more than requsted
	 */
2862
	ret = min((u64) ret << 9, (u64) len);
2863 2864 2865 2866 2867

	/* XXX get a quota reservation */
	i_sectors_acct(c, dst, NULL, i_sectors_delta);

	spin_lock(&dst->v.i_lock);
2868 2869
	if (pos_dst + ret > dst->v.i_size)
		i_size_write(&dst->v, pos_dst + ret);
2870 2871
	spin_unlock(&dst->v.i_lock);
err:
Kent Overstreet's avatar
Kent Overstreet committed
2872 2873 2874 2875 2876
	bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);

	return ret;
}

2877 2878
/* fseek: */

2879
static int folio_data_offset(struct folio *folio, unsigned offset)
2880
{
2881 2882 2883
	struct bch_page_state *s = bch2_page_state(&folio->page);
	unsigned i;

2884 2885 2886 2887
	if (s)
		for (i = offset >> 9; i < PAGE_SECTORS; i++)
			if (s->s[i].state >= SECTOR_DIRTY)
				return i << 9;
2888

2889
	return -1;
2890 2891
}

2892
static loff_t bch2_seek_pagecache_data(struct inode *vinode,
2893 2894 2895 2896 2897 2898 2899 2900
				       loff_t start_offset,
				       loff_t end_offset)
{
	struct folio_batch fbatch;
	pgoff_t start_index	= start_offset >> PAGE_SHIFT;
	pgoff_t end_index	= end_offset >> PAGE_SHIFT;
	pgoff_t index		= start_index;
	unsigned i;
2901 2902
	loff_t ret;
	int offset;
2903 2904 2905 2906 2907 2908 2909 2910 2911

	folio_batch_init(&fbatch);

	while (filemap_get_folios(vinode->i_mapping,
				  &index, end_index, &fbatch)) {
		for (i = 0; i < folio_batch_count(&fbatch); i++) {
			struct folio *folio = fbatch.folios[i];

			folio_lock(folio);
2912 2913 2914 2915 2916 2917 2918 2919
			offset = folio_data_offset(folio,
					folio->index == start_index
					? start_offset & (PAGE_SIZE - 1)
					: 0);
			if (offset >= 0) {
				ret = clamp(((loff_t) folio->index << PAGE_SHIFT) +
					    offset,
					    start_offset, end_offset);
2920 2921
				folio_unlock(folio);
				folio_batch_release(&fbatch);
2922
				return ret;
2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936
			}
			folio_unlock(folio);
		}
		folio_batch_release(&fbatch);
		cond_resched();
	}

	return end_offset;
}

static loff_t bch2_seek_data(struct file *file, u64 offset)
{
	struct bch_inode_info *inode = file_bch_inode(file);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
2937 2938
	struct btree_trans trans;
	struct btree_iter *iter;
2939 2940 2941 2942 2943 2944 2945 2946
	struct bkey_s_c k;
	u64 isize, next_data = MAX_LFS_FILESIZE;
	int ret;

	isize = i_size_read(&inode->v);
	if (offset >= isize)
		return -ENXIO;

2947
	bch2_trans_init(&trans, c, 0, 0);
2948 2949

	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
2950
			   POS(inode->v.i_ino, offset >> 9), 0, k, ret) {
2951 2952 2953 2954 2955 2956 2957 2958 2959
		if (k.k->p.inode != inode->v.i_ino) {
			break;
		} else if (bkey_extent_is_data(k.k)) {
			next_data = max(offset, bkey_start_offset(k.k) << 9);
			break;
		} else if (k.k->p.offset >> 9 > isize)
			break;
	}

2960
	ret = bch2_trans_exit(&trans) ?: ret;
2961 2962 2963 2964
	if (ret)
		return ret;

	if (next_data > offset)
2965
		next_data = bch2_seek_pagecache_data(&inode->v,
2966 2967
						     offset, next_data);

2968
	if (next_data >= isize)
2969 2970 2971 2972 2973
		return -ENXIO;

	return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
}

2974
static int __page_hole_offset(struct page *page, unsigned offset)
2975
{
2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
	struct bch_page_state *s = bch2_page_state(page);
	unsigned i;

	if (!s)
		return 0;

	for (i = offset >> 9; i < PAGE_SECTORS; i++)
		if (s->s[i].state < SECTOR_DIRTY)
			return i << 9;

	return -1;
}

static loff_t page_hole_offset(struct address_space *mapping, loff_t offset)
{
	pgoff_t index = offset >> PAGE_SHIFT;
2992
	struct page *page;
2993 2994
	int pg_offset;
	loff_t ret = -1;
2995 2996 2997

	page = find_lock_page(mapping, index);
	if (!page)
2998 2999 3000 3001 3002
		return offset;

	pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1));
	if (pg_offset >= 0)
		ret = ((loff_t) index << PAGE_SHIFT) + pg_offset;
3003 3004 3005 3006 3007 3008

	unlock_page(page);

	return ret;
}

3009
static loff_t bch2_seek_pagecache_hole(struct inode *vinode,
3010 3011 3012 3013
				       loff_t start_offset,
				       loff_t end_offset)
{
	struct address_space *mapping = vinode->i_mapping;
3014
	loff_t offset = start_offset, hole;
3015

3016 3017 3018 3019 3020 3021 3022 3023
	while (offset < end_offset) {
		hole = page_hole_offset(mapping, offset);
		if (hole >= 0 && hole <= end_offset)
			return max(start_offset, hole);

		offset += PAGE_SIZE;
		offset &= PAGE_MASK;
	}
3024 3025 3026 3027 3028 3029 3030 3031

	return end_offset;
}

static loff_t bch2_seek_hole(struct file *file, u64 offset)
{
	struct bch_inode_info *inode = file_bch_inode(file);
	struct bch_fs *c = inode->v.i_sb->s_fs_info;
3032 3033
	struct btree_trans trans;
	struct btree_iter *iter;
3034 3035 3036 3037 3038 3039 3040 3041
	struct bkey_s_c k;
	u64 isize, next_hole = MAX_LFS_FILESIZE;
	int ret;

	isize = i_size_read(&inode->v);
	if (offset >= isize)
		return -ENXIO;

3042
	bch2_trans_init(&trans, c, 0, 0);
3043 3044

	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
3045
			   POS(inode->v.i_ino, offset >> 9),
3046
			   BTREE_ITER_SLOTS, k, ret) {
3047
		if (k.k->p.inode != inode->v.i_ino) {
3048
			next_hole = bch2_seek_pagecache_hole(&inode->v,
3049 3050 3051
					offset, MAX_LFS_FILESIZE);
			break;
		} else if (!bkey_extent_is_data(k.k)) {
3052
			next_hole = bch2_seek_pagecache_hole(&inode->v,
3053 3054 3055 3056 3057 3058 3059 3060 3061 3062
					max(offset, bkey_start_offset(k.k) << 9),
					k.k->p.offset << 9);

			if (next_hole < k.k->p.offset << 9)
				break;
		} else {
			offset = max(offset, bkey_start_offset(k.k) << 9);
		}
	}

3063
	ret = bch2_trans_exit(&trans) ?: ret;
3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102
	if (ret)
		return ret;

	if (next_hole > isize)
		next_hole = isize;

	return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
}

loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
{
	switch (whence) {
	case SEEK_SET:
	case SEEK_CUR:
	case SEEK_END:
		return generic_file_llseek(file, offset, whence);
	case SEEK_DATA:
		return bch2_seek_data(file, offset);
	case SEEK_HOLE:
		return bch2_seek_hole(file, offset);
	}

	return -EINVAL;
}

void bch2_fs_fsio_exit(struct bch_fs *c)
{
	bioset_exit(&c->dio_write_bioset);
	bioset_exit(&c->dio_read_bioset);
	bioset_exit(&c->writepage_bioset);
}

int bch2_fs_fsio_init(struct bch_fs *c)
{
	int ret = 0;

	pr_verbose_init(c->opts, "");

	if (bioset_init(&c->writepage_bioset,
3103
			4, offsetof(struct bch_writepage_io, op.wbio.bio),
3104 3105 3106 3107 3108
			BIOSET_NEED_BVECS) ||
	    bioset_init(&c->dio_read_bioset,
			4, offsetof(struct dio_read, rbio.bio),
			BIOSET_NEED_BVECS) ||
	    bioset_init(&c->dio_write_bioset,
3109
			4, offsetof(struct dio_write, op.wbio.bio),
3110 3111 3112 3113 3114 3115 3116 3117
			BIOSET_NEED_BVECS))
		ret = -ENOMEM;

	pr_verbose_init(c->opts, "ret %i", ret);
	return ret;
}

#endif /* NO_BCACHEFS_FS */