io.c 48.7 KB
Newer Older
1 2 3 4 5 6 7 8 9
// SPDX-License-Identifier: GPL-2.0
/*
 * Some low level IO code, and hacks for various block layer limitations
 *
 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
 * Copyright 2012 Google, Inc.
 */

#include "bcachefs.h"
10
#include "alloc_foreground.h"
11 12 13 14 15 16 17 18
#include "bset.h"
#include "btree_update.h"
#include "buckets.h"
#include "checksum.h"
#include "compress.h"
#include "clock.h"
#include "debug.h"
#include "disk_groups.h"
19
#include "ec.h"
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215
#include "error.h"
#include "extents.h"
#include "io.h"
#include "journal.h"
#include "keylist.h"
#include "move.h"
#include "rebalance.h"
#include "super.h"
#include "super-io.h"
#include "trace.h"

#include <linux/blkdev.h>
#include <linux/random.h>

#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT

static bool bch2_target_congested(struct bch_fs *c, u16 target)
{
	const struct bch_devs_mask *devs;
	unsigned d, nr = 0, total = 0;
	u64 now = local_clock(), last;
	s64 congested;
	struct bch_dev *ca;

	if (!target)
		return false;

	rcu_read_lock();
	devs = bch2_target_to_mask(c, target);
	for_each_set_bit(d, devs->d, BCH_SB_MEMBERS_MAX) {
		ca = rcu_dereference(c->devs[d]);
		if (!ca)
			continue;

		congested = atomic_read(&ca->congested);
		last = READ_ONCE(ca->congested_last);
		if (time_after64(now, last))
			congested -= (now - last) >> 12;

		total += max(congested, 0LL);
		nr++;
	}
	rcu_read_unlock();

	return bch2_rand_range(nr * CONGESTED_MAX) < total;
}

static inline void bch2_congested_acct(struct bch_dev *ca, u64 io_latency,
				       u64 now, int rw)
{
	u64 latency_capable =
		ca->io_latency[rw].quantiles.entries[QUANTILE_IDX(1)].m;
	/* ideally we'd be taking into account the device's variance here: */
	u64 latency_threshold = latency_capable << (rw == READ ? 2 : 3);
	s64 latency_over = io_latency - latency_threshold;

	if (latency_threshold && latency_over > 0) {
		/*
		 * bump up congested by approximately latency_over * 4 /
		 * latency_threshold - we don't need much accuracy here so don't
		 * bother with the divide:
		 */
		if (atomic_read(&ca->congested) < CONGESTED_MAX)
			atomic_add(latency_over >>
				   max_t(int, ilog2(latency_threshold) - 2, 0),
				   &ca->congested);

		ca->congested_last = now;
	} else if (atomic_read(&ca->congested) > 0) {
		atomic_dec(&ca->congested);
	}
}

void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw)
{
	atomic64_t *latency = &ca->cur_latency[rw];
	u64 now = local_clock();
	u64 io_latency = time_after64(now, submit_time)
		? now - submit_time
		: 0;
	u64 old, new, v = atomic64_read(latency);

	do {
		old = v;

		/*
		 * If the io latency was reasonably close to the current
		 * latency, skip doing the update and atomic operation - most of
		 * the time:
		 */
		if (abs((int) (old - io_latency)) < (old >> 1) &&
		    now & ~(~0 << 5))
			break;

		new = ewma_add(old, io_latency, 5);
	} while ((v = atomic64_cmpxchg(latency, old, new)) != old);

	bch2_congested_acct(ca, io_latency, now, rw);

	__bch2_time_stats_update(&ca->io_latency[rw], submit_time, now);
}

#else

static bool bch2_target_congested(struct bch_fs *c, u16 target)
{
	return false;
}

#endif

/* Allocate, free from mempool: */

void bch2_bio_free_pages_pool(struct bch_fs *c, struct bio *bio)
{
	struct bvec_iter_all iter;
	struct bio_vec *bv;

	bio_for_each_segment_all(bv, bio, iter)
		if (bv->bv_page != ZERO_PAGE(0))
			mempool_free(bv->bv_page, &c->bio_bounce_pages);
	bio->bi_vcnt = 0;
}

static void bch2_bio_alloc_page_pool(struct bch_fs *c, struct bio *bio,
				    bool *using_mempool)
{
	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt++];

	if (likely(!*using_mempool)) {
		bv->bv_page = alloc_page(GFP_NOIO);
		if (unlikely(!bv->bv_page)) {
			mutex_lock(&c->bio_bounce_pages_lock);
			*using_mempool = true;
			goto pool_alloc;

		}
	} else {
pool_alloc:
		bv->bv_page = mempool_alloc(&c->bio_bounce_pages, GFP_NOIO);
	}

	bv->bv_len = PAGE_SIZE;
	bv->bv_offset = 0;
}

void bch2_bio_alloc_pages_pool(struct bch_fs *c, struct bio *bio,
			       size_t bytes)
{
	bool using_mempool = false;

	BUG_ON(DIV_ROUND_UP(bytes, PAGE_SIZE) > bio->bi_max_vecs);

	bio->bi_iter.bi_size = bytes;

	while (bio->bi_vcnt < DIV_ROUND_UP(bytes, PAGE_SIZE))
		bch2_bio_alloc_page_pool(c, bio, &using_mempool);

	if (using_mempool)
		mutex_unlock(&c->bio_bounce_pages_lock);
}

void bch2_bio_alloc_more_pages_pool(struct bch_fs *c, struct bio *bio,
				    size_t bytes)
{
	while (bio->bi_vcnt < DIV_ROUND_UP(bytes, PAGE_SIZE)) {
		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];

		BUG_ON(bio->bi_vcnt >= bio->bi_max_vecs);

		bv->bv_page = alloc_page(GFP_NOIO);
		if (!bv->bv_page) {
			/*
			 * We already allocated from mempool, we can't allocate from it again
			 * without freeing the pages we already allocated or else we could
			 * deadlock:
			 */
			bch2_bio_free_pages_pool(c, bio);
			bch2_bio_alloc_pages_pool(c, bio, bytes);
			return;
		}

		bv->bv_len = PAGE_SIZE;
		bv->bv_offset = 0;
		bio->bi_vcnt++;
	}

	bio->bi_iter.bi_size = bytes;
}

/* Writes */

void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c,
			       enum bch_data_type type,
			       const struct bkey_i *k)
{
216
	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(k));
217 218 219 220 221 222
	const struct bch_extent_ptr *ptr;
	struct bch_write_bio *n;
	struct bch_dev *ca;

	BUG_ON(c->opts.nochanges);

223
	bkey_for_each_ptr(ptrs, ptr) {
224 225 226 227 228
		BUG_ON(ptr->dev >= BCH_SB_MEMBERS_MAX ||
		       !c->devs[ptr->dev]);

		ca = bch_dev_bkey_exists(c, ptr->dev);

229
		if (to_entry(ptr + 1) < ptrs.end) {
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295
			n = to_wbio(bio_alloc_clone(NULL, &wbio->bio,
						GFP_NOIO, &ca->replica_set));

			n->bio.bi_end_io	= wbio->bio.bi_end_io;
			n->bio.bi_private	= wbio->bio.bi_private;
			n->parent		= wbio;
			n->split		= true;
			n->bounce		= false;
			n->put_bio		= true;
			n->bio.bi_opf		= wbio->bio.bi_opf;
			bio_inc_remaining(&wbio->bio);
		} else {
			n = wbio;
			n->split		= false;
		}

		n->c			= c;
		n->dev			= ptr->dev;
		n->have_ioref		= bch2_dev_get_ioref(ca, WRITE);
		n->submit_time		= local_clock();
		n->bio.bi_iter.bi_sector = ptr->offset;

		if (!journal_flushes_device(ca))
			n->bio.bi_opf |= REQ_FUA;

		if (likely(n->have_ioref)) {
			this_cpu_add(ca->io_done->sectors[WRITE][type],
				     bio_sectors(&n->bio));

			bio_set_dev(&n->bio, ca->disk_sb.bdev);

			if (type != BCH_DATA_BTREE && unlikely(c->opts.no_data_io)) {
				bio_endio(&n->bio);
				continue;
			}

			submit_bio(&n->bio);
		} else {
			n->bio.bi_status	= BLK_STS_REMOVED;
			bio_endio(&n->bio);
		}
	}
}

static void __bch2_write(struct closure *);

static void bch2_write_done(struct closure *cl)
{
	struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
	struct bch_fs *c = op->c;

	if (!op->error && (op->flags & BCH_WRITE_FLUSH))
		op->error = bch2_journal_error(&c->journal);

	if (!(op->flags & BCH_WRITE_NOPUT_RESERVATION))
		bch2_disk_reservation_put(c, &op->res);
	percpu_ref_put(&c->writes);
	bch2_keylist_free(&op->insert_keys, op->inline_keys);

	bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time);

	closure_return(cl);
}

int bch2_write_index_default(struct bch_write_op *op)
{
296
	struct bch_fs *c = op->c;
297 298
	struct btree_trans trans;
	struct btree_iter *iter;
299 300 301
	struct keylist *keys = &op->insert_keys;
	int ret;

302 303 304 305 306 307 308 309
	BUG_ON(bch2_keylist_empty(keys));
	bch2_verify_keylist_sorted(keys);

	bch2_trans_init(&trans, c);

	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
				   bkey_start_pos(&bch2_keylist_front(keys)->k),
				   BTREE_ITER_INTENT);
310

311 312 313 314 315
	do {
		BKEY_PADDED(k) split;

		bkey_copy(&split.k, bch2_keylist_front(keys));

316
		bch2_extent_trim_atomic(&split.k, iter);
317

318 319 320 321 322 323
		bch2_trans_update(&trans,
				  BTREE_INSERT_ENTRY(iter, &split.k));

		ret = bch2_trans_commit(&trans, &op->res, op_journal_seq(op),
					BTREE_INSERT_NOFAIL|
					BTREE_INSERT_USE_RESERVE);
324 325 326
		if (ret)
			break;

327 328
		if (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) < 0)
			bch2_cut_front(iter->pos, bch2_keylist_front(keys));
329 330 331 332
		else
			bch2_keylist_pop_front(keys);
	} while (!bch2_keylist_empty(keys));

333
	bch2_trans_exit(&trans);
334 335 336 337 338 339 340 341 342 343 344 345 346

	return ret;
}

/**
 * bch_write_index - after a write, update index to point to new data
 */
static void __bch2_write_index(struct bch_write_op *op)
{
	struct bch_fs *c = op->c;
	struct keylist *keys = &op->insert_keys;
	struct bch_extent_ptr *ptr;
	struct bkey_i *src, *dst = keys->keys, *n, *k;
347
	unsigned dev;
348 349 350 351 352 353
	int ret;

	for (src = keys->keys; src != keys->top; src = n) {
		n = bkey_next(src);
		bkey_copy(dst, src);

354
		bch2_bkey_drop_ptrs(bkey_i_to_s(dst), ptr,
355
			test_bit(ptr->dev, op->failed.d));
356

357
		if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(dst))) {
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
			ret = -EIO;
			goto err;
		}

		dst = bkey_next(dst);
	}

	keys->top = dst;

	/*
	 * probably not the ideal place to hook this in, but I don't
	 * particularly want to plumb io_opts all the way through the btree
	 * update stack right now
	 */
	for_each_keylist_key(keys, k)
		bch2_rebalance_add_key(c, bkey_i_to_s_c(k), &op->opts);

	if (!bch2_keylist_empty(keys)) {
		u64 sectors_start = keylist_sectors(keys);
		int ret = op->index_update_fn(op);

		BUG_ON(keylist_sectors(keys) && !ret);

		op->written += sectors_start - keylist_sectors(keys);

		if (ret) {
			__bcache_io_error(c, "btree IO error %i", ret);
			op->error = ret;
		}
	}
out:
389 390 391 392
	/* If some a bucket wasn't written, we can't erasure code it: */
	for_each_set_bit(dev, op->failed.d, BCH_SB_MEMBERS_MAX)
		bch2_open_bucket_write_error(c, &op->open_buckets, dev);

393
	bch2_open_buckets_put(c, &op->open_buckets);
394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452
	return;
err:
	keys->top = keys->keys;
	op->error = ret;
	goto out;
}

static void bch2_write_index(struct closure *cl)
{
	struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
	struct bch_fs *c = op->c;

	__bch2_write_index(op);

	if (!op->error && (op->flags & BCH_WRITE_FLUSH)) {
		bch2_journal_flush_seq_async(&c->journal,
					     *op_journal_seq(op),
					     cl);
		continue_at(cl, bch2_write_done, index_update_wq(op));
	} else {
		continue_at_nobarrier(cl, bch2_write_done, NULL);
	}
}

static void bch2_write_endio(struct bio *bio)
{
	struct closure *cl		= bio->bi_private;
	struct bch_write_op *op		= container_of(cl, struct bch_write_op, cl);
	struct bch_write_bio *wbio	= to_wbio(bio);
	struct bch_write_bio *parent	= wbio->split ? wbio->parent : NULL;
	struct bch_fs *c		= wbio->c;
	struct bch_dev *ca		= bch_dev_bkey_exists(c, wbio->dev);

	if (bch2_dev_io_err_on(bio->bi_status, ca, "data write"))
		set_bit(wbio->dev, op->failed.d);

	if (wbio->have_ioref) {
		bch2_latency_acct(ca, wbio->submit_time, WRITE);
		percpu_ref_put(&ca->io_ref);
	}

	if (wbio->bounce)
		bch2_bio_free_pages_pool(c, bio);

	if (wbio->put_bio)
		bio_put(bio);

	if (parent)
		bio_endio(&parent->bio);
	else
		closure_put(cl);
}

static void init_append_extent(struct bch_write_op *op,
			       struct write_point *wp,
			       struct bversion version,
			       struct bch_extent_crc_unpacked crc)
{
	struct bkey_i_extent *e = bkey_extent_init(op->insert_keys.top);
453
	struct bch_extent_ptr *ptr;
454 455

	op->pos.offset += crc.uncompressed_size;
456 457 458
	e->k.p		= op->pos;
	e->k.size	= crc.uncompressed_size;
	e->k.version	= version;
459

460 461 462 463 464
	if (crc.csum_type ||
	    crc.compression_type ||
	    crc.nonce)
		bch2_extent_crc_append(e, crc);

465
	bch2_alloc_sectors_append_ptrs(op->c, wp, &e->k_i, crc.compressed_size);
466

467 468 469 470
	if (op->flags & BCH_WRITE_CACHED)
		extent_for_each_ptr(extent_i_to_s(e), ptr)
			ptr->cached = true;

471 472 473 474 475 476
	bch2_keylist_push(&op->insert_keys);
}

static struct bio *bch2_write_bio_alloc(struct bch_fs *c,
					struct write_point *wp,
					struct bio *src,
477 478
					bool *page_alloc_failed,
					void *buf)
479 480 481 482 483 484 485 486 487 488 489 490 491 492
{
	struct bch_write_bio *wbio;
	struct bio *bio;
	unsigned output_available =
		min(wp->sectors_free << 9, src->bi_iter.bi_size);
	unsigned pages = DIV_ROUND_UP(output_available, PAGE_SIZE);

	bio = bio_alloc_bioset(NULL, pages, 0,
			       GFP_NOIO, &c->bio_write);
	wbio			= wbio_init(bio);
	wbio->put_bio		= true;
	/* copy WRITE_SYNC flag */
	wbio->bio.bi_opf	= src->bi_opf;

493 494 495 496 497 498 499 500
	if (buf) {
		bio->bi_iter.bi_size = output_available;
		bch2_bio_map(bio, buf);
		return bio;
	}

	wbio->bounce		= true;

501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664
	/*
	 * We can't use mempool for more than c->sb.encoded_extent_max
	 * worth of pages, but we'd like to allocate more if we can:
	 */
	while (bio->bi_iter.bi_size < output_available) {
		unsigned len = min_t(unsigned, PAGE_SIZE,
				     output_available - bio->bi_iter.bi_size);
		struct page *p;

		p = alloc_page(GFP_NOIO);
		if (!p) {
			unsigned pool_max =
				min_t(unsigned, output_available,
				      c->sb.encoded_extent_max << 9);

			if (bio_sectors(bio) < pool_max)
				bch2_bio_alloc_pages_pool(c, bio, pool_max);
			break;
		}

		bio->bi_io_vec[bio->bi_vcnt++] = (struct bio_vec) {
			.bv_page	= p,
			.bv_len		= len,
			.bv_offset	= 0,
		};
		bio->bi_iter.bi_size += len;
	}

	*page_alloc_failed = bio->bi_vcnt < pages;
	return bio;
}

static int bch2_write_rechecksum(struct bch_fs *c,
				 struct bch_write_op *op,
				 unsigned new_csum_type)
{
	struct bio *bio = &op->wbio.bio;
	struct bch_extent_crc_unpacked new_crc;
	int ret;

	/* bch2_rechecksum_bio() can't encrypt or decrypt data: */

	if (bch2_csum_type_is_encryption(op->crc.csum_type) !=
	    bch2_csum_type_is_encryption(new_csum_type))
		new_csum_type = op->crc.csum_type;

	ret = bch2_rechecksum_bio(c, bio, op->version, op->crc,
				  NULL, &new_crc,
				  op->crc.offset, op->crc.live_size,
				  new_csum_type);
	if (ret)
		return ret;

	bio_advance(bio, op->crc.offset << 9);
	bio->bi_iter.bi_size = op->crc.live_size << 9;
	op->crc = new_crc;
	return 0;
}

static int bch2_write_decrypt(struct bch_write_op *op)
{
	struct bch_fs *c = op->c;
	struct nonce nonce = extent_nonce(op->version, op->crc);
	struct bch_csum csum;

	if (!bch2_csum_type_is_encryption(op->crc.csum_type))
		return 0;

	/*
	 * If we need to decrypt data in the write path, we'll no longer be able
	 * to verify the existing checksum (poly1305 mac, in this case) after
	 * it's decrypted - this is the last point we'll be able to reverify the
	 * checksum:
	 */
	csum = bch2_checksum_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
	if (bch2_crc_cmp(op->crc.csum, csum))
		return -EIO;

	bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
	op->crc.csum_type = 0;
	op->crc.csum = (struct bch_csum) { 0, 0 };
	return 0;
}

static enum prep_encoded_ret {
	PREP_ENCODED_OK,
	PREP_ENCODED_ERR,
	PREP_ENCODED_CHECKSUM_ERR,
	PREP_ENCODED_DO_WRITE,
} bch2_write_prep_encoded_data(struct bch_write_op *op, struct write_point *wp)
{
	struct bch_fs *c = op->c;
	struct bio *bio = &op->wbio.bio;

	if (!(op->flags & BCH_WRITE_DATA_ENCODED))
		return PREP_ENCODED_OK;

	BUG_ON(bio_sectors(bio) != op->crc.compressed_size);

	/* Can we just write the entire extent as is? */
	if (op->crc.uncompressed_size == op->crc.live_size &&
	    op->crc.compressed_size <= wp->sectors_free &&
	    op->crc.compression_type == op->compression_type) {
		if (!op->crc.compression_type &&
		    op->csum_type != op->crc.csum_type &&
		    bch2_write_rechecksum(c, op, op->csum_type))
			return PREP_ENCODED_CHECKSUM_ERR;

		return PREP_ENCODED_DO_WRITE;
	}

	/*
	 * If the data is compressed and we couldn't write the entire extent as
	 * is, we have to decompress it:
	 */
	if (op->crc.compression_type) {
		struct bch_csum csum;

		if (bch2_write_decrypt(op))
			return PREP_ENCODED_CHECKSUM_ERR;

		/* Last point we can still verify checksum: */
		csum = bch2_checksum_bio(c, op->crc.csum_type,
					 extent_nonce(op->version, op->crc),
					 bio);
		if (bch2_crc_cmp(op->crc.csum, csum))
			return PREP_ENCODED_CHECKSUM_ERR;

		if (bch2_bio_uncompress_inplace(c, bio, &op->crc))
			return PREP_ENCODED_ERR;
	}

	/*
	 * No longer have compressed data after this point - data might be
	 * encrypted:
	 */

	/*
	 * If the data is checksummed and we're only writing a subset,
	 * rechecksum and adjust bio to point to currently live data:
	 */
	if ((op->crc.live_size != op->crc.uncompressed_size ||
	     op->crc.csum_type != op->csum_type) &&
	    bch2_write_rechecksum(c, op, op->csum_type))
		return PREP_ENCODED_CHECKSUM_ERR;

	/*
	 * If we want to compress the data, it has to be decrypted:
	 */
	if ((op->compression_type ||
	     bch2_csum_type_is_encryption(op->crc.csum_type) !=
	     bch2_csum_type_is_encryption(op->csum_type)) &&
	    bch2_write_decrypt(op))
		return PREP_ENCODED_CHECKSUM_ERR;

	return PREP_ENCODED_OK;
}

static int bch2_write_extent(struct bch_write_op *op, struct write_point *wp)
{
	struct bch_fs *c = op->c;
	struct bio *src = &op->wbio.bio, *dst = src;
	struct bvec_iter saved_iter;
	struct bkey_i *key_to_write;
665
	void *ec_buf;
666 667
	unsigned key_to_write_offset = op->insert_keys.top_p -
		op->insert_keys.keys_p;
668 669 670
	unsigned total_output = 0, total_input = 0;
	bool bounce = false;
	bool page_alloc_failed = false;
671 672 673 674
	int ret, more = 0;

	BUG_ON(!bio_sectors(src));

675 676
	ec_buf = bch2_writepoint_ec_buf(c, wp);

677 678 679 680 681 682 683 684 685
	switch (bch2_write_prep_encoded_data(op, wp)) {
	case PREP_ENCODED_OK:
		break;
	case PREP_ENCODED_ERR:
		ret = -EIO;
		goto err;
	case PREP_ENCODED_CHECKSUM_ERR:
		goto csum_err;
	case PREP_ENCODED_DO_WRITE:
686 687 688 689 690 691 692
		if (ec_buf) {
			dst = bch2_write_bio_alloc(c, wp, src,
						   &page_alloc_failed,
						   ec_buf);
			bio_copy_data(dst, src);
			bounce = true;
		}
693 694 695 696
		init_append_extent(op, wp, op->version, op->crc);
		goto do_write;
	}

697 698
	if (ec_buf ||
	    op->compression_type ||
699 700 701 702
	    (op->csum_type &&
	     !(op->flags & BCH_WRITE_PAGES_STABLE)) ||
	    (bch2_csum_type_is_encryption(op->csum_type) &&
	     !(op->flags & BCH_WRITE_PAGES_OWNED))) {
703 704 705
		dst = bch2_write_bio_alloc(c, wp, src,
					   &page_alloc_failed,
					   ec_buf);
706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
		bounce = true;
	}

	saved_iter = dst->bi_iter;

	do {
		struct bch_extent_crc_unpacked crc =
			(struct bch_extent_crc_unpacked) { 0 };
		struct bversion version = op->version;
		size_t dst_len, src_len;

		if (page_alloc_failed &&
		    bio_sectors(dst) < wp->sectors_free &&
		    bio_sectors(dst) < c->sb.encoded_extent_max)
			break;

		BUG_ON(op->compression_type &&
		       (op->flags & BCH_WRITE_DATA_ENCODED) &&
		       bch2_csum_type_is_encryption(op->crc.csum_type));
		BUG_ON(op->compression_type && !bounce);

		crc.compression_type = op->compression_type
			?  bch2_bio_compress(c, dst, &dst_len, src, &src_len,
					     op->compression_type)
			: 0;
		if (!crc.compression_type) {
			dst_len = min(dst->bi_iter.bi_size, src->bi_iter.bi_size);
			dst_len = min_t(unsigned, dst_len, wp->sectors_free << 9);

			if (op->csum_type)
				dst_len = min_t(unsigned, dst_len,
						c->sb.encoded_extent_max << 9);

			if (bounce) {
				swap(dst->bi_iter.bi_size, dst_len);
				bio_copy_data(dst, src);
				swap(dst->bi_iter.bi_size, dst_len);
			}

			src_len = dst_len;
		}

		BUG_ON(!src_len || !dst_len);

		if (bch2_csum_type_is_encryption(op->csum_type)) {
			if (bversion_zero(version)) {
				version.lo = atomic64_inc_return(&c->key_version) + 1;
			} else {
				crc.nonce = op->nonce;
				op->nonce += src_len >> 9;
			}
		}

		if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
		    !crc.compression_type &&
		    bch2_csum_type_is_encryption(op->crc.csum_type) ==
		    bch2_csum_type_is_encryption(op->csum_type)) {
			/*
			 * Note: when we're using rechecksum(), we need to be
			 * checksumming @src because it has all the data our
			 * existing checksum covers - if we bounced (because we
			 * were trying to compress), @dst will only have the
			 * part of the data the new checksum will cover.
			 *
			 * But normally we want to be checksumming post bounce,
			 * because part of the reason for bouncing is so the
			 * data can't be modified (by userspace) while it's in
			 * flight.
			 */
			if (bch2_rechecksum_bio(c, src, version, op->crc,
					&crc, &op->crc,
					src_len >> 9,
					bio_sectors(src) - (src_len >> 9),
					op->csum_type))
				goto csum_err;
		} else {
			if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
			    bch2_rechecksum_bio(c, src, version, op->crc,
					NULL, &op->crc,
					src_len >> 9,
					bio_sectors(src) - (src_len >> 9),
					op->crc.csum_type))
				goto csum_err;

			crc.compressed_size	= dst_len >> 9;
			crc.uncompressed_size	= src_len >> 9;
			crc.live_size		= src_len >> 9;

			swap(dst->bi_iter.bi_size, dst_len);
			bch2_encrypt_bio(c, op->csum_type,
					 extent_nonce(version, crc), dst);
			crc.csum = bch2_checksum_bio(c, op->csum_type,
					 extent_nonce(version, crc), dst);
			crc.csum_type = op->csum_type;
			swap(dst->bi_iter.bi_size, dst_len);
		}

		init_append_extent(op, wp, version, crc);

		if (dst != src)
			bio_advance(dst, dst_len);
		bio_advance(src, src_len);
808 809
		total_output	+= dst_len;
		total_input	+= src_len;
810 811 812 813 814 815 816 817 818 819 820 821
	} while (dst->bi_iter.bi_size &&
		 src->bi_iter.bi_size &&
		 wp->sectors_free &&
		 !bch2_keylist_realloc(&op->insert_keys,
				      op->inline_keys,
				      ARRAY_SIZE(op->inline_keys),
				      BKEY_EXTENT_U64s_MAX));

	more = src->bi_iter.bi_size != 0;

	dst->bi_iter = saved_iter;

822 823 824 825
	if (dst == src && more) {
		BUG_ON(total_output != total_input);

		dst = bio_split(src, total_input >> 9,
826
				GFP_NOIO, &c->bio_write);
827 828 829
		wbio_init(dst)->put_bio	= true;
		/* copy WRITE_SYNC flag */
		dst->bi_opf		= src->bi_opf;
830 831 832 833 834
	}

	dst->bi_iter.bi_size = total_output;

	/* Free unneeded pages after compressing: */
835
	if (to_wbio(dst)->bounce)
836 837 838 839 840 841 842 843
		while (dst->bi_vcnt > DIV_ROUND_UP(dst->bi_iter.bi_size, PAGE_SIZE))
			mempool_free(dst->bi_io_vec[--dst->bi_vcnt].bv_page,
				     &c->bio_bounce_pages);
do_write:
	/* might have done a realloc... */

	key_to_write = (void *) (op->insert_keys.keys_p + key_to_write_offset);

844 845 846 847
	bch2_ec_add_backpointer(c, wp,
				bkey_start_pos(&key_to_write->k),
				total_input >> 9);

848 849 850 851 852 853 854 855 856 857 858 859 860 861
	dst->bi_end_io	= bch2_write_endio;
	dst->bi_private	= &op->cl;
	dst->bi_opf	= REQ_OP_WRITE;

	closure_get(dst->bi_private);

	bch2_submit_wbio_replicas(to_wbio(dst), c, BCH_DATA_USER,
				  key_to_write);
	return more;
csum_err:
	bch_err(c, "error verifying existing checksum while "
		"rewriting existing data (memory corruption?)");
	ret = -EIO;
err:
862
	if (to_wbio(dst)->bounce)
863
		bch2_bio_free_pages_pool(c, dst);
864
	if (to_wbio(dst)->put_bio)
865 866 867 868 869 870 871 872 873 874 875 876
		bio_put(dst);

	return ret;
}

static void __bch2_write(struct closure *cl)
{
	struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
	struct bch_fs *c = op->c;
	struct write_point *wp;
	int ret;
again:
877 878
	memset(&op->failed, 0, sizeof(op->failed));

879 880
	do {
		/* +1 for possible cache device: */
881 882
		if (op->open_buckets.nr + op->nr_replicas + 1 >
		    ARRAY_SIZE(op->open_buckets.v))
883 884 885 886 887 888 889 890 891 892
			goto flush_io;

		if (bch2_keylist_realloc(&op->insert_keys,
					op->inline_keys,
					ARRAY_SIZE(op->inline_keys),
					BKEY_EXTENT_U64s_MAX))
			goto flush_io;

		wp = bch2_alloc_sectors_start(c,
			op->target,
893
			op->opts.erasure_code,
894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913
			op->write_point,
			&op->devs_have,
			op->nr_replicas,
			op->nr_replicas_required,
			op->alloc_reserve,
			op->flags,
			(op->flags & BCH_WRITE_ALLOC_NOWAIT) ? NULL : cl);
		EBUG_ON(!wp);

		if (unlikely(IS_ERR(wp))) {
			if (unlikely(PTR_ERR(wp) != -EAGAIN)) {
				ret = PTR_ERR(wp);
				goto err;
			}

			goto flush_io;
		}

		ret = bch2_write_extent(op, wp);

914
		bch2_open_bucket_get(c, wp, &op->open_buckets);
915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994
		bch2_alloc_sectors_done(c, wp);

		if (ret < 0)
			goto err;
	} while (ret);

	continue_at(cl, bch2_write_index, index_update_wq(op));
	return;
err:
	op->error = ret;

	continue_at(cl, !bch2_keylist_empty(&op->insert_keys)
		    ? bch2_write_index
		    : bch2_write_done, index_update_wq(op));
	return;
flush_io:
	closure_sync(cl);

	if (!bch2_keylist_empty(&op->insert_keys)) {
		__bch2_write_index(op);

		if (op->error) {
			continue_at_nobarrier(cl, bch2_write_done, NULL);
			return;
		}
	}

	goto again;
}

/**
 * bch_write - handle a write to a cache device or flash only volume
 *
 * This is the starting point for any data to end up in a cache device; it could
 * be from a normal write, or a writeback write, or a write to a flash only
 * volume - it's also used by the moving garbage collector to compact data in
 * mostly empty buckets.
 *
 * It first writes the data to the cache, creating a list of keys to be inserted
 * (if the data won't fit in a single open bucket, there will be multiple keys);
 * after the data is written it calls bch_journal, and after the keys have been
 * added to the next journal write they're inserted into the btree.
 *
 * If op->discard is true, instead of inserting the data it invalidates the
 * region of the cache represented by op->bio and op->inode.
 */
void bch2_write(struct closure *cl)
{
	struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
	struct bch_fs *c = op->c;

	BUG_ON(!op->nr_replicas);
	BUG_ON(!op->write_point.v);
	BUG_ON(!bkey_cmp(op->pos, POS_MAX));
	BUG_ON(bio_sectors(&op->wbio.bio) > U16_MAX);

	op->start_time = local_clock();

	bch2_keylist_init(&op->insert_keys, op->inline_keys);
	wbio_init(&op->wbio.bio)->put_bio = false;

	if (c->opts.nochanges ||
	    !percpu_ref_tryget(&c->writes)) {
		__bcache_io_error(c, "read only");
		op->error = -EROFS;
		if (!(op->flags & BCH_WRITE_NOPUT_RESERVATION))
			bch2_disk_reservation_put(c, &op->res);
		closure_return(cl);
		return;
	}

	bch2_increment_clock(c, bio_sectors(&op->wbio.bio), WRITE);

	continue_at_nobarrier(cl, __bch2_write, NULL);
}

/* Cache promotion on read */

struct promote_op {
	struct closure		cl;
995
	struct rcu_head		rcu;
996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
	u64			start_time;

	struct rhash_head	hash;
	struct bpos		pos;

	struct migrate_write	write;
	struct bio_vec		bi_inline_vecs[0]; /* must be last */
};

static const struct rhashtable_params bch_promote_params = {
	.head_offset	= offsetof(struct promote_op, hash),
	.key_offset	= offsetof(struct promote_op, pos),
	.key_len	= sizeof(struct bpos),
};

static inline bool should_promote(struct bch_fs *c, struct bkey_s_c k,
				  struct bpos pos,
				  struct bch_io_opts opts,
				  unsigned flags)
{
	if (!opts.promote_target)
		return false;

	if (!(flags & BCH_READ_MAY_PROMOTE))
		return false;

	if (percpu_ref_is_dying(&c->writes))
		return false;

	if (!bkey_extent_is_data(k.k))
		return false;

	if (bch2_extent_has_target(c, bkey_s_c_to_extent(k), opts.promote_target))
		return false;

	if (bch2_target_congested(c, opts.promote_target))
		return false;

	if (rhashtable_lookup_fast(&c->promote_table, &pos,
				   bch_promote_params))
		return false;

	return true;
}

static void promote_free(struct bch_fs *c, struct promote_op *op)
{
	int ret;

	ret = rhashtable_remove_fast(&c->promote_table, &op->hash,
				     bch_promote_params);
	BUG_ON(ret);
	percpu_ref_put(&c->writes);
1049
	kfree_rcu(op, rcu);
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090
}

static void promote_done(struct closure *cl)
{
	struct promote_op *op =
		container_of(cl, struct promote_op, cl);
	struct bch_fs *c = op->write.op.c;

	bch2_time_stats_update(&c->times[BCH_TIME_data_promote],
			       op->start_time);

	bch2_bio_free_pages_pool(c, &op->write.op.wbio.bio);
	promote_free(c, op);
}

static void promote_start(struct promote_op *op, struct bch_read_bio *rbio)
{
	struct bch_fs *c = rbio->c;
	struct closure *cl = &op->cl;
	struct bio *bio = &op->write.op.wbio.bio;

	trace_promote(&rbio->bio);

	/* we now own pages: */
	BUG_ON(!rbio->bounce);
	BUG_ON(rbio->bio.bi_vcnt > bio->bi_max_vecs);

	memcpy(bio->bi_io_vec, rbio->bio.bi_io_vec,
	       sizeof(struct bio_vec) * rbio->bio.bi_vcnt);
	swap(bio->bi_vcnt, rbio->bio.bi_vcnt);

	bch2_migrate_read_done(&op->write, rbio);

	closure_init(cl, NULL);
	closure_call(&op->write.op.cl, bch2_write, c->wq, cl);
	closure_return_with_destructor(cl, promote_done);
}

noinline
static struct promote_op *__promote_alloc(struct bch_fs *c,
					  struct bpos pos,
1091
					  struct extent_ptr_decoded *pick,
1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168
					  struct bch_io_opts opts,
					  unsigned rbio_sectors,
					  struct bch_read_bio **rbio)
{
	struct promote_op *op = NULL;
	struct bio *bio;
	unsigned rbio_pages = DIV_ROUND_UP(rbio_sectors, PAGE_SECTORS);
	/* data might have to be decompressed in the write path: */
	unsigned wbio_pages = DIV_ROUND_UP(pick->crc.uncompressed_size,
					   PAGE_SECTORS);
	int ret;

	if (!percpu_ref_tryget(&c->writes))
		return NULL;

	op = kzalloc(sizeof(*op) + sizeof(struct bio_vec) * wbio_pages,
		     GFP_NOIO);
	if (!op)
		goto err;

	op->start_time = local_clock();
	op->pos = pos;

	/*
	 * promotes require bouncing, but if the extent isn't
	 * checksummed/compressed it might be too big for the mempool:
	 */
	if (rbio_sectors > c->sb.encoded_extent_max) {
		*rbio = kzalloc(sizeof(struct bch_read_bio) +
				sizeof(struct bio_vec) * rbio_pages,
				GFP_NOIO);
		if (!*rbio)
			goto err;

		rbio_init(&(*rbio)->bio, opts);
		bio_init(&(*rbio)->bio, NULL, (*rbio)->bio.bi_inline_vecs, rbio_pages, 0);

		if (bch2_bio_alloc_pages(&(*rbio)->bio, rbio_sectors << 9,
					 GFP_NOIO))
			goto err;

		(*rbio)->bounce		= true;
		(*rbio)->split		= true;
		(*rbio)->kmalloc	= true;
	}

	if (rhashtable_lookup_insert_fast(&c->promote_table, &op->hash,
					  bch_promote_params))
		goto err;

	bio = &op->write.op.wbio.bio;
	bio_init(bio, NULL, bio->bi_inline_vecs, wbio_pages, 0);

	ret = bch2_migrate_write_init(c, &op->write,
			writepoint_hashed((unsigned long) current),
			opts,
			DATA_PROMOTE,
			(struct data_opts) {
				.target = opts.promote_target
			},
			bkey_s_c_null);
	BUG_ON(ret);

	return op;
err:
	if (*rbio)
		bio_free_pages(&(*rbio)->bio);
	kfree(*rbio);
	*rbio = NULL;
	kfree(op);
	percpu_ref_put(&c->writes);
	return NULL;
}

static inline struct promote_op *promote_alloc(struct bch_fs *c,
					       struct bvec_iter iter,
					       struct bkey_s_c k,
1169
					       struct extent_ptr_decoded *pick,
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 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 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
					       struct bch_io_opts opts,
					       unsigned flags,
					       struct bch_read_bio **rbio,
					       bool *bounce,
					       bool *read_full)
{
	bool promote_full = *read_full || READ_ONCE(c->promote_whole_extents);
	unsigned sectors = promote_full
		? pick->crc.compressed_size
		: bvec_iter_sectors(iter);
	struct bpos pos = promote_full
		? bkey_start_pos(k.k)
		: POS(k.k->p.inode, iter.bi_sector);
	struct promote_op *promote;

	if (!should_promote(c, k, pos, opts, flags))
		return NULL;

	promote = __promote_alloc(c, pos, pick, opts, sectors, rbio);
	if (!promote)
		return NULL;

	*bounce		= true;
	*read_full	= promote_full;
	return promote;
}

/* Read */

#define READ_RETRY_AVOID	1
#define READ_RETRY		2
#define READ_ERR		3

enum rbio_context {
	RBIO_CONTEXT_NULL,
	RBIO_CONTEXT_HIGHPRI,
	RBIO_CONTEXT_UNBOUND,
};

static inline struct bch_read_bio *
bch2_rbio_parent(struct bch_read_bio *rbio)
{
	return rbio->split ? rbio->parent : rbio;
}

__always_inline
static void bch2_rbio_punt(struct bch_read_bio *rbio, work_func_t fn,
			   enum rbio_context context,
			   struct workqueue_struct *wq)
{
	if (context <= rbio->context) {
		fn(&rbio->work);
	} else {
		rbio->work.func		= fn;
		rbio->context		= context;
		queue_work(wq, &rbio->work);
	}
}

static inline struct bch_read_bio *bch2_rbio_free(struct bch_read_bio *rbio)
{
	BUG_ON(rbio->bounce && !rbio->split);

	if (rbio->promote)
		promote_free(rbio->c, rbio->promote);
	rbio->promote = NULL;

	if (rbio->bounce)
		bch2_bio_free_pages_pool(rbio->c, &rbio->bio);

	if (rbio->split) {
		struct bch_read_bio *parent = rbio->parent;

		if (rbio->kmalloc)
			kfree(rbio);
		else
			bio_put(&rbio->bio);

		rbio = parent;
	}

	return rbio;
}

static void bch2_rbio_done(struct bch_read_bio *rbio)
{
	bch2_time_stats_update(&rbio->c->times[BCH_TIME_data_read],
			       rbio->start_time);
	bio_endio(&rbio->bio);
}

static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio,
				     struct bvec_iter bvec_iter, u64 inode,
1263 1264
				     struct bch_io_failures *failed,
				     unsigned flags)
1265
{
1266 1267
	struct btree_trans trans;
	struct btree_iter *iter;
1268 1269 1270 1271 1272 1273
	BKEY_PADDED(k) tmp;
	struct bkey_s_c k;
	int ret;

	flags &= ~BCH_READ_LAST_FRAGMENT;

1274 1275 1276 1277
	bch2_trans_init(&trans, c);

	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
				   rbio->pos, BTREE_ITER_SLOTS);
1278 1279 1280
retry:
	rbio->bio.bi_status = 0;

1281
	k = bch2_btree_iter_peek_slot(iter);
1282
	if (bkey_err(k))
1283 1284 1285 1286
		goto err;

	bkey_reassemble(&tmp.k, k);
	k = bkey_i_to_s_c(&tmp.k);
1287
	bch2_trans_unlock(&trans);
1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298

	if (!bkey_extent_is_data(k.k) ||
	    !bch2_extent_matches_ptr(c, bkey_i_to_s_c_extent(&tmp.k),
				     rbio->pick.ptr,
				     rbio->pos.offset -
				     rbio->pick.crc.offset)) {
		/* extent we wanted to read no longer exists: */
		rbio->hole = true;
		goto out;
	}

1299
	ret = __bch2_read_extent(c, rbio, bvec_iter, k, failed, flags);
1300 1301 1302 1303 1304 1305
	if (ret == READ_RETRY)
		goto retry;
	if (ret)
		goto err;
out:
	bch2_rbio_done(rbio);
1306 1307 1308 1309 1310
	bch2_trans_exit(&trans);
	return;
err:
	rbio->bio.bi_status = BLK_STS_IOERR;
	goto out;
1311 1312 1313 1314
}

static void bch2_read_retry(struct bch_fs *c, struct bch_read_bio *rbio,
			    struct bvec_iter bvec_iter, u64 inode,
1315
			    struct bch_io_failures *failed, unsigned flags)
1316
{
1317 1318
	struct btree_trans trans;
	struct btree_iter *iter;
1319 1320 1321
	struct bkey_s_c k;
	int ret;

1322 1323
	bch2_trans_init(&trans, c);

1324 1325 1326
	flags &= ~BCH_READ_LAST_FRAGMENT;
	flags |= BCH_READ_MUST_CLONE;
retry:
1327
	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
1328
			   POS(inode, bvec_iter.bi_sector),
1329
			   BTREE_ITER_SLOTS, k, ret) {
1330 1331 1332 1333 1334
		BKEY_PADDED(k) tmp;
		unsigned bytes;

		bkey_reassemble(&tmp.k, k);
		k = bkey_i_to_s_c(&tmp.k);
1335
		bch2_btree_trans_unlock(&trans);
1336 1337 1338 1339 1340

		bytes = min_t(unsigned, bvec_iter.bi_size,
			      (k.k->p.offset - bvec_iter.bi_sector) << 9);
		swap(bvec_iter.bi_size, bytes);

1341
		ret = __bch2_read_extent(c, rbio, bvec_iter, k, failed, flags);
1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359
		switch (ret) {
		case READ_RETRY:
			goto retry;
		case READ_ERR:
			goto err;
		};

		if (bytes == bvec_iter.bi_size)
			goto out;

		swap(bvec_iter.bi_size, bytes);
		bio_advance_iter(&rbio->bio, &bvec_iter, bytes);
	}

	/*
	 * If we get here, it better have been because there was an error
	 * reading a btree node
	 */
1360 1361
	BUG_ON(!ret);
	__bcache_io_error(c, "btree IO error: %i", ret);
1362 1363 1364
err:
	rbio->bio.bi_status = BLK_STS_IOERR;
out:
1365
	bch2_trans_exit(&trans);
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376
	bch2_rbio_done(rbio);
}

static void bch2_rbio_retry(struct work_struct *work)
{
	struct bch_read_bio *rbio =
		container_of(work, struct bch_read_bio, work);
	struct bch_fs *c	= rbio->c;
	struct bvec_iter iter	= rbio->bvec_iter;
	unsigned flags		= rbio->flags;
	u64 inode		= rbio->pos.inode;
1377
	struct bch_io_failures failed = { .nr = 0 };
1378 1379 1380 1381

	trace_read_retry(&rbio->bio);

	if (rbio->retry == READ_RETRY_AVOID)
1382
		bch2_mark_io_failure(&failed, &rbio->pick);
1383 1384 1385 1386 1387 1388 1389 1390 1391

	rbio->bio.bi_status = 0;

	rbio = bch2_rbio_free(rbio);

	flags |= BCH_READ_IN_RETRY;
	flags &= ~BCH_READ_MAY_PROMOTE;

	if (flags & BCH_READ_NODECODE)
1392
		bch2_read_retry_nodecode(c, rbio, iter, inode, &failed, flags);
1393
	else
1394
		bch2_read_retry(c, rbio, iter, inode, &failed, flags);
1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418
}

static void bch2_rbio_error(struct bch_read_bio *rbio, int retry,
			    blk_status_t error)
{
	rbio->retry = retry;

	if (rbio->flags & BCH_READ_IN_RETRY)
		return;

	if (retry == READ_ERR) {
		rbio = bch2_rbio_free(rbio);

		rbio->bio.bi_status = error;
		bch2_rbio_done(rbio);
	} else {
		bch2_rbio_punt(rbio, bch2_rbio_retry,
			       RBIO_CONTEXT_UNBOUND, system_unbound_wq);
	}
}

static void bch2_rbio_narrow_crcs(struct bch_read_bio *rbio)
{
	struct bch_fs *c = rbio->c;
1419 1420
	struct btree_trans trans;
	struct btree_iter *iter;
1421 1422 1423 1424 1425 1426 1427 1428 1429 1430
	struct bkey_s_c k;
	struct bkey_i_extent *e;
	BKEY_PADDED(k) new;
	struct bch_extent_crc_unpacked new_crc;
	unsigned offset;
	int ret;

	if (rbio->pick.crc.compression_type)
		return;

1431
	bch2_trans_init(&trans, c);
1432
retry:
1433 1434 1435 1436 1437
	bch2_trans_begin(&trans);

	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, rbio->pos,
				   BTREE_ITER_INTENT);
	k = bch2_btree_iter_peek(iter);
1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472
	if (IS_ERR_OR_NULL(k.k))
		goto out;

	if (!bkey_extent_is_data(k.k))
		goto out;

	bkey_reassemble(&new.k, k);
	e = bkey_i_to_extent(&new.k);

	if (!bch2_extent_matches_ptr(c, extent_i_to_s_c(e),
				     rbio->pick.ptr,
				     rbio->pos.offset -
				     rbio->pick.crc.offset) ||
	    bversion_cmp(e->k.version, rbio->version))
		goto out;

	/* Extent was merged? */
	if (bkey_start_offset(&e->k) < rbio->pos.offset ||
	    e->k.p.offset > rbio->pos.offset + rbio->pick.crc.uncompressed_size)
		goto out;

	/* The extent might have been partially overwritten since we read it: */
	offset = rbio->pick.crc.offset + (bkey_start_offset(&e->k) - rbio->pos.offset);

	if (bch2_rechecksum_bio(c, &rbio->bio, rbio->version,
				rbio->pick.crc, NULL, &new_crc,
				offset, e->k.size,
				rbio->pick.crc.csum_type)) {
		bch_err(c, "error verifying existing checksum while narrowing checksum (memory corruption?)");
		goto out;
	}

	if (!bch2_extent_narrow_crcs(e, new_crc))
		goto out;

1473 1474 1475 1476 1477
	bch2_trans_update(&trans, BTREE_INSERT_ENTRY(iter, &e->k_i));
	ret = bch2_trans_commit(&trans, NULL, NULL,
				BTREE_INSERT_ATOMIC|
				BTREE_INSERT_NOFAIL|
				BTREE_INSERT_NOWAIT);
1478 1479 1480
	if (ret == -EINTR)
		goto retry;
out:
1481
	bch2_trans_exit(&trans);
1482 1483 1484
}

static bool should_narrow_crcs(struct bkey_s_c k,
1485
			       struct extent_ptr_decoded *pick,
1486 1487 1488 1489 1490 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 1596 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
			       unsigned flags)
{
	return !(flags & BCH_READ_IN_RETRY) &&
		bkey_extent_is_data(k.k) &&
		bch2_can_narrow_extent_crcs(bkey_s_c_to_extent(k), pick->crc);
}

/* Inner part that may run in process context */
static void __bch2_read_endio(struct work_struct *work)
{
	struct bch_read_bio *rbio =
		container_of(work, struct bch_read_bio, work);
	struct bch_fs *c	= rbio->c;
	struct bch_dev *ca	= bch_dev_bkey_exists(c, rbio->pick.ptr.dev);
	struct bio *src		= &rbio->bio;
	struct bio *dst		= &bch2_rbio_parent(rbio)->bio;
	struct bvec_iter dst_iter = rbio->bvec_iter;
	struct bch_extent_crc_unpacked crc = rbio->pick.crc;
	struct nonce nonce = extent_nonce(rbio->version, crc);
	struct bch_csum csum;

	/* Reset iterator for checksumming and copying bounced data: */
	if (rbio->bounce) {
		src->bi_iter.bi_size		= crc.compressed_size << 9;
		src->bi_iter.bi_idx		= 0;
		src->bi_iter.bi_bvec_done	= 0;
	} else {
		src->bi_iter			= rbio->bvec_iter;
	}

	csum = bch2_checksum_bio(c, crc.csum_type, nonce, src);
	if (bch2_crc_cmp(csum, rbio->pick.crc.csum) && !c->opts.no_data_io)
		goto csum_err;

	if (unlikely(rbio->narrow_crcs))
		bch2_rbio_narrow_crcs(rbio);

	if (rbio->flags & BCH_READ_NODECODE)
		goto nodecode;

	/* Adjust crc to point to subset of data we want: */
	crc.offset     += rbio->bvec_iter.bi_sector - rbio->pos.offset;
	crc.live_size	= bvec_iter_sectors(rbio->bvec_iter);

	if (crc.compression_type != BCH_COMPRESSION_NONE) {
		bch2_encrypt_bio(c, crc.csum_type, nonce, src);
		if (bch2_bio_uncompress(c, src, dst, dst_iter, crc))
			goto decompression_err;
	} else {
		/* don't need to decrypt the entire bio: */
		nonce = nonce_add(nonce, crc.offset << 9);
		bio_advance(src, crc.offset << 9);

		BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size);
		src->bi_iter.bi_size = dst_iter.bi_size;

		bch2_encrypt_bio(c, crc.csum_type, nonce, src);

		if (rbio->bounce) {
			struct bvec_iter src_iter = src->bi_iter;
			bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
		}
	}

	if (rbio->promote) {
		/*
		 * Re encrypt data we decrypted, so it's consistent with
		 * rbio->crc:
		 */
		bch2_encrypt_bio(c, crc.csum_type, nonce, src);
		promote_start(rbio->promote, rbio);
		rbio->promote = NULL;
	}
nodecode:
	if (likely(!(rbio->flags & BCH_READ_IN_RETRY))) {
		rbio = bch2_rbio_free(rbio);
		bch2_rbio_done(rbio);
	}
	return;
csum_err:
	/*
	 * Checksum error: if the bio wasn't bounced, we may have been
	 * reading into buffers owned by userspace (that userspace can
	 * scribble over) - retry the read, bouncing it this time:
	 */
	if (!rbio->bounce && (rbio->flags & BCH_READ_USER_MAPPED)) {
		rbio->flags |= BCH_READ_MUST_BOUNCE;
		bch2_rbio_error(rbio, READ_RETRY, BLK_STS_IOERR);
		return;
	}

	bch2_dev_io_error(ca,
		"data checksum error, inode %llu offset %llu: expected %0llx:%0llx got %0llx:%0llx (type %u)",
		rbio->pos.inode, (u64) rbio->bvec_iter.bi_sector,
		rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo,
		csum.hi, csum.lo, crc.csum_type);
	bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
	return;
decompression_err:
	__bcache_io_error(c, "decompression error, inode %llu offset %llu",
			  rbio->pos.inode,
			  (u64) rbio->bvec_iter.bi_sector);
	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
	return;
}

static void bch2_read_endio(struct bio *bio)
{
	struct bch_read_bio *rbio =
		container_of(bio, struct bch_read_bio, bio);
	struct bch_fs *c	= rbio->c;
	struct bch_dev *ca	= bch_dev_bkey_exists(c, rbio->pick.ptr.dev);
	struct workqueue_struct *wq = NULL;
	enum rbio_context context = RBIO_CONTEXT_NULL;

	if (rbio->have_ioref) {
		bch2_latency_acct(ca, rbio->submit_time, READ);
		percpu_ref_put(&ca->io_ref);
	}

	if (!rbio->split)
		rbio->bio.bi_end_io = rbio->end_io;

	if (bch2_dev_io_err_on(bio->bi_status, ca, "data read")) {
		bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
		return;
	}

	if (rbio->pick.ptr.cached &&
	    (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) ||
	     ptr_stale(ca, &rbio->pick.ptr))) {
		atomic_long_inc(&c->read_realloc_races);

		if (rbio->flags & BCH_READ_RETRY_IF_STALE)
			bch2_rbio_error(rbio, READ_RETRY, BLK_STS_AGAIN);
		else
			bch2_rbio_error(rbio, READ_ERR, BLK_STS_AGAIN);
		return;
	}

	if (rbio->narrow_crcs ||
	    rbio->pick.crc.compression_type ||
	    bch2_csum_type_is_encryption(rbio->pick.crc.csum_type))
		context = RBIO_CONTEXT_UNBOUND,	wq = system_unbound_wq;
	else if (rbio->pick.crc.csum_type)
		context = RBIO_CONTEXT_HIGHPRI,	wq = system_highpri_wq;

	bch2_rbio_punt(rbio, __bch2_read_endio, context, wq);
}

int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
		       struct bvec_iter iter, struct bkey_s_c k,
1638
		       struct bch_io_failures *failed, unsigned flags)
1639
{
1640
	struct extent_ptr_decoded pick;
1641 1642 1643 1644 1645 1646 1647
	struct bch_read_bio *rbio = NULL;
	struct bch_dev *ca;
	struct promote_op *promote = NULL;
	bool bounce = false, read_full = false, narrow_crcs = false;
	struct bpos pos = bkey_start_pos(k.k);
	int pick_ret;

1648
	pick_ret = bch2_bkey_pick_read_device(c, k, failed, &pick);
1649 1650 1651 1652 1653

	/* hole or reservation - just zero fill: */
	if (!pick_ret)
		goto hole;

1654 1655 1656 1657
	if (pick_ret < 0) {
		__bcache_io_error(c, "no device to read from");
		goto err;
	}
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 1702 1703 1704 1705 1706 1707 1708 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 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783

	if (pick_ret > 0)
		ca = bch_dev_bkey_exists(c, pick.ptr.dev);

	if (flags & BCH_READ_NODECODE) {
		/*
		 * can happen if we retry, and the extent we were going to read
		 * has been merged in the meantime:
		 */
		if (pick.crc.compressed_size > orig->bio.bi_vcnt * PAGE_SECTORS)
			goto hole;

		iter.bi_sector	= pos.offset;
		iter.bi_size	= pick.crc.compressed_size << 9;
		goto noclone;
	}

	if (!(flags & BCH_READ_LAST_FRAGMENT) ||
	    bio_flagged(&orig->bio, BIO_CHAIN))
		flags |= BCH_READ_MUST_CLONE;

	narrow_crcs = should_narrow_crcs(k, &pick, flags);

	if (narrow_crcs && (flags & BCH_READ_USER_MAPPED))
		flags |= BCH_READ_MUST_BOUNCE;

	EBUG_ON(bkey_start_offset(k.k) > iter.bi_sector ||
		k.k->p.offset < bvec_iter_end_sector(iter));

	if (pick.crc.compression_type != BCH_COMPRESSION_NONE ||
	    (pick.crc.csum_type != BCH_CSUM_NONE &&
	     (bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
	      (bch2_csum_type_is_encryption(pick.crc.csum_type) &&
	       (flags & BCH_READ_USER_MAPPED)) ||
	      (flags & BCH_READ_MUST_BOUNCE)))) {
		read_full = true;
		bounce = true;
	}

	promote = promote_alloc(c, iter, k, &pick, orig->opts, flags,
				&rbio, &bounce, &read_full);

	if (!read_full) {
		EBUG_ON(pick.crc.compression_type);
		EBUG_ON(pick.crc.csum_type &&
			(bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
			 bvec_iter_sectors(iter) != pick.crc.live_size ||
			 pick.crc.offset ||
			 iter.bi_sector != pos.offset));

		pick.ptr.offset += pick.crc.offset +
			(iter.bi_sector - pos.offset);
		pick.crc.compressed_size	= bvec_iter_sectors(iter);
		pick.crc.uncompressed_size	= bvec_iter_sectors(iter);
		pick.crc.offset			= 0;
		pick.crc.live_size		= bvec_iter_sectors(iter);
		pos.offset			= iter.bi_sector;
	}

	if (rbio) {
		/* promote already allocated bounce rbio */
	} else if (bounce) {
		unsigned sectors = pick.crc.compressed_size;

		rbio = rbio_init(bio_alloc_bioset(NULL,
						  DIV_ROUND_UP(sectors, PAGE_SECTORS),
						  0,
						  GFP_NOIO,
						  &c->bio_read_split),
				 orig->opts);

		bch2_bio_alloc_pages_pool(c, &rbio->bio, sectors << 9);
		rbio->bounce	= true;
		rbio->split	= true;
	} else if (flags & BCH_READ_MUST_CLONE) {
		/*
		 * Have to clone if there were any splits, due to error
		 * reporting issues (if a split errored, and retrying didn't
		 * work, when it reports the error to its parent (us) we don't
		 * know if the error was from our bio, and we should retry, or
		 * from the whole bio, in which case we don't want to retry and
		 * lose the error)
		 */
		rbio = rbio_init(bio_alloc_clone(NULL, &orig->bio, GFP_NOIO,
						 &c->bio_read_split),
				 orig->opts);
		rbio->bio.bi_iter = iter;
		rbio->split	= true;
	} else {
noclone:
		rbio = orig;
		rbio->bio.bi_iter = iter;
		BUG_ON(bio_flagged(&rbio->bio, BIO_CHAIN));
	}

	BUG_ON(bio_sectors(&rbio->bio) != pick.crc.compressed_size);

	rbio->c			= c;
	rbio->submit_time	= local_clock();
	if (rbio->split)
		rbio->parent	= orig;
	else
		rbio->end_io	= orig->bio.bi_end_io;
	rbio->bvec_iter		= iter;
	rbio->flags		= flags;
	rbio->have_ioref	= pick_ret > 0 && bch2_dev_get_ioref(ca, READ);
	rbio->narrow_crcs	= narrow_crcs;
	rbio->hole		= 0;
	rbio->retry		= 0;
	rbio->context		= 0;
	rbio->devs_have		= bch2_bkey_devs(k);
	rbio->pick		= pick;
	rbio->pos		= pos;
	rbio->version		= k.k->version;
	rbio->promote		= promote;
	INIT_WORK(&rbio->work, NULL);

	rbio->bio.bi_opf	= orig->bio.bi_opf;
	rbio->bio.bi_iter.bi_sector = pick.ptr.offset;
	rbio->bio.bi_end_io	= bch2_read_endio;

	if (rbio->bounce)
		trace_read_bounce(&rbio->bio);

	bch2_increment_clock(c, bio_sectors(&rbio->bio), READ);

1784
	percpu_down_read(&c->mark_lock);
1785
	bucket_io_clock_reset(c, ca, PTR_BUCKET_NR(ca, &pick.ptr), READ);
1786
	percpu_up_read(&c->mark_lock);
1787

1788 1789 1790 1791
	if (likely(!(flags & (BCH_READ_IN_RETRY|BCH_READ_LAST_FRAGMENT)))) {
		bio_inc_remaining(&orig->bio);
		trace_read_split(&orig->bio);
	}
1792

1793 1794 1795 1796 1797
	if (!rbio->pick.idx) {
		if (!rbio->have_ioref) {
			__bcache_io_error(c, "no device to read from");
			bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
			goto out;
1798 1799
		}

1800 1801 1802 1803
		this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_USER],
			     bio_sectors(&rbio->bio));
		bio_set_dev(&rbio->bio, ca->disk_sb.bdev);

1804
		if (unlikely(c->opts.no_data_io)) {
1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817
			if (likely(!(flags & BCH_READ_IN_RETRY)))
				bio_endio(&rbio->bio);
		} else {
			if (likely(!(flags & BCH_READ_IN_RETRY)))
				submit_bio(&rbio->bio);
			else
				submit_bio_wait(&rbio->bio);
		}
	} else {
		/* Attempting reconstruct read: */
		if (bch2_ec_read_extent(c, rbio)) {
			bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
			goto out;
1818 1819
		}

1820 1821 1822 1823 1824
		if (likely(!(flags & BCH_READ_IN_RETRY)))
			bio_endio(&rbio->bio);
	}
out:
	if (likely(!(flags & BCH_READ_IN_RETRY))) {
1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835
		return 0;
	} else {
		int ret;

		rbio->context = RBIO_CONTEXT_UNBOUND;
		bch2_read_endio(&rbio->bio);

		ret = rbio->retry;
		rbio = bch2_rbio_free(rbio);

		if (ret == READ_RETRY_AVOID) {
1836
			bch2_mark_io_failure(failed, &pick);
1837 1838 1839 1840 1841 1842
			ret = READ_RETRY;
		}

		return ret;
	}

1843 1844
err:
	if (flags & BCH_READ_IN_RETRY)
1845
		return READ_ERR;
1846 1847 1848

	orig->bio.bi_status = BLK_STS_IOERR;
	goto out_read_done;
1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859

hole:
	/*
	 * won't normally happen in the BCH_READ_NODECODE
	 * (bch2_move_extent()) path, but if we retry and the extent we wanted
	 * to read no longer exists we have to signal that:
	 */
	if (flags & BCH_READ_NODECODE)
		orig->hole = true;

	zero_fill_bio_iter(&orig->bio, iter);
1860
out_read_done:
1861 1862 1863 1864 1865 1866 1867
	if (flags & BCH_READ_LAST_FRAGMENT)
		bch2_rbio_done(orig);
	return 0;
}

void bch2_read(struct bch_fs *c, struct bch_read_bio *rbio, u64 inode)
{
1868 1869
	struct btree_trans trans;
	struct btree_iter *iter;
1870 1871 1872 1873
	struct bkey_s_c k;
	unsigned flags = BCH_READ_RETRY_IF_STALE|
		BCH_READ_MAY_PROMOTE|
		BCH_READ_USER_MAPPED;
1874
	int ret;
1875 1876

	bch2_trans_init(&trans, c);
1877 1878 1879 1880 1881 1882 1883 1884

	BUG_ON(rbio->_state);
	BUG_ON(flags & BCH_READ_NODECODE);
	BUG_ON(flags & BCH_READ_IN_RETRY);

	rbio->c = c;
	rbio->start_time = local_clock();

1885
	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
1886
			   POS(inode, rbio->bio.bi_iter.bi_sector),
1887
			   BTREE_ITER_SLOTS, k, ret) {
1888 1889 1890 1891 1892 1893 1894 1895 1896
		BKEY_PADDED(k) tmp;
		unsigned bytes;

		/*
		 * Unlock the iterator while the btree node's lock is still in
		 * cache, before doing the IO:
		 */
		bkey_reassemble(&tmp.k, k);
		k = bkey_i_to_s_c(&tmp.k);
1897
		bch2_btree_trans_unlock(&trans);
1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918

		bytes = min_t(unsigned, rbio->bio.bi_iter.bi_size,
			      (k.k->p.offset - rbio->bio.bi_iter.bi_sector) << 9);
		swap(rbio->bio.bi_iter.bi_size, bytes);

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

		bch2_read_extent(c, rbio, k, flags);

		if (flags & BCH_READ_LAST_FRAGMENT)
			return;

		swap(rbio->bio.bi_iter.bi_size, bytes);
		bio_advance(&rbio->bio, bytes);
	}

	/*
	 * If we get here, it better have been because there was an error
	 * reading a btree node
	 */
1919 1920
	BUG_ON(!ret);
	bcache_io_error(c, &rbio->bio, "btree IO error: %i", ret);
1921 1922

	bch2_trans_exit(&trans);
1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953
	bch2_rbio_done(rbio);
}

void bch2_fs_io_exit(struct bch_fs *c)
{
	if (c->promote_table.tbl)
		rhashtable_destroy(&c->promote_table);
	mempool_exit(&c->bio_bounce_pages);
	bioset_exit(&c->bio_write);
	bioset_exit(&c->bio_read_split);
	bioset_exit(&c->bio_read);
}

int bch2_fs_io_init(struct bch_fs *c)
{
	if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
			BIOSET_NEED_BVECS) ||
	    bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
			BIOSET_NEED_BVECS) ||
	    bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio),
			BIOSET_NEED_BVECS) ||
	    mempool_init_page_pool(&c->bio_bounce_pages,
				   max_t(unsigned,
					 c->opts.btree_node_size,
					 c->sb.encoded_extent_max) /
				   PAGE_SECTORS, 0) ||
	    rhashtable_init(&c->promote_table, &bch_promote_params))
		return -ENOMEM;

	return 0;
}