1 files changed, 607 insertions, 382 deletions

diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 61395b113df9..036e4ad95987 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -21,12 +21,20 @@
 #include "extents.h"
 #include "inode.h"
 #include "journal.h"
+#include "rebalance.h"
 #include "replicas.h"
 #include "super.h"
 #include "super-io.h"
 #include "trace.h"
 #include "util.h"
 
+static const char * const bch2_extent_flags_strs[] = {
+#define x(n, v)	[BCH_EXTENT_FLAG_##n] = #n,
+	BCH_EXTENT_FLAGS()
+#undef x
+	NULL,
+};
+
 static unsigned bch2_crc_field_size_max[] = {
 	[BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
 	[BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
@@ -37,8 +45,51 @@ static void bch2_extent_crc_pack(union bch_extent_crc *,
 				 struct bch_extent_crc_unpacked,
 				 enum bch_extent_entry_type);
 
-static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
-						   unsigned dev)
+void bch2_io_failures_to_text(struct printbuf *out,
+			      struct bch_fs *c,
+			      struct bch_io_failures *failed)
+{
+	static const char * const error_types[] = {
+		"io", "checksum", "ec reconstruct", NULL
+	};
+
+	for (struct bch_dev_io_failures *f = failed->devs;
+	     f < failed->devs + failed->nr;
+	     f++) {
+		unsigned errflags =
+			((!!f->failed_io)	<< 0) |
+			((!!f->failed_csum_nr)	<< 1) |
+			((!!f->failed_ec)	<< 2);
+
+		if (!errflags)
+			continue;
+
+		bch2_printbuf_make_room(out, 1024);
+		out->atomic++;
+		scoped_guard(rcu) {
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, f->dev);
+			if (ca)
+				prt_str(out, ca->name);
+			else
+				prt_printf(out, "(invalid device %u)", f->dev);
+		}
+		--out->atomic;
+
+		prt_char(out, ' ');
+
+		if (is_power_of_2(errflags)) {
+			prt_bitflags(out, error_types, errflags);
+			prt_str(out, " error");
+		} else {
+			prt_str(out, "errors: ");
+			prt_bitflags(out, error_types, errflags);
+		}
+		prt_newline(out);
+	}
+}
+
+struct bch_dev_io_failures *bch2_dev_io_failures(struct bch_io_failures *f,
+						 unsigned dev)
 {
 	struct bch_dev_io_failures *i;
 
@@ -50,25 +101,51 @@ static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
 }
 
 void bch2_mark_io_failure(struct bch_io_failures *failed,
-			  struct extent_ptr_decoded *p)
+			  struct extent_ptr_decoded *p,
+			  bool csum_error)
 {
-	struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
+	struct bch_dev_io_failures *f = bch2_dev_io_failures(failed, p->ptr.dev);
 
 	if (!f) {
 		BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
 
 		f = &failed->devs[failed->nr++];
-		f->dev		= p->ptr.dev;
-		f->idx		= p->idx;
-		f->nr_failed	= 1;
-		f->nr_retries	= 0;
-	} else if (p->idx != f->idx) {
-		f->idx		= p->idx;
-		f->nr_failed	= 1;
-		f->nr_retries	= 0;
-	} else {
-		f->nr_failed++;
+		memset(f, 0, sizeof(*f));
+		f->dev = p->ptr.dev;
 	}
+
+	if (p->do_ec_reconstruct)
+		f->failed_ec = true;
+	else if (!csum_error)
+		f->failed_io = true;
+	else
+		f->failed_csum_nr++;
+}
+
+void bch2_mark_btree_validate_failure(struct bch_io_failures *failed,
+				      unsigned dev)
+{
+	struct bch_dev_io_failures *f = bch2_dev_io_failures(failed, dev);
+
+	if (!f) {
+		BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
+
+		f = &failed->devs[failed->nr++];
+		memset(f, 0, sizeof(*f));
+		f->dev = dev;
+	}
+
+	f->failed_btree_validate = true;
+}
+
+static inline u64 dev_latency(struct bch_dev *ca)
+{
+	return ca ? atomic64_read(&ca->cur_latency[READ]) : S64_MAX;
+}
+
+static inline int dev_failed(struct bch_dev *ca)
+{
+	return !ca || ca->mi.state == BCH_MEMBER_STATE_failed;
 }
 
 /*
@@ -76,24 +153,30 @@ void bch2_mark_io_failure(struct bch_io_failures *failed,
  */
 static inline bool ptr_better(struct bch_fs *c,
 			      const struct extent_ptr_decoded p1,
-			      const struct extent_ptr_decoded p2)
+			      u64 p1_latency,
+			      struct bch_dev *ca1,
+			      const struct extent_ptr_decoded p2,
+			      u64 p2_latency)
 {
-	if (likely(!p1.idx && !p2.idx)) {
-		struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
-		struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
+	struct bch_dev *ca2 = bch2_dev_rcu(c, p2.ptr.dev);
 
-		u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
-		u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
+	int failed_delta = dev_failed(ca1) - dev_failed(ca2);
+	if (unlikely(failed_delta))
+		return failed_delta < 0;
 
-		/* Pick at random, biased in favor of the faster device: */
+	if (static_branch_unlikely(&bch2_force_reconstruct_read))
+		return p1.do_ec_reconstruct > p2.do_ec_reconstruct;
 
-		return bch2_rand_range(l1 + l2) > l1;
-	}
+	if (unlikely(p1.do_ec_reconstruct || p2.do_ec_reconstruct))
+		return p1.do_ec_reconstruct < p2.do_ec_reconstruct;
 
-	if (bch2_force_reconstruct_read)
-		return p1.idx > p2.idx;
+	int crc_retry_delta = (int) p1.crc_retry_nr - (int) p2.crc_retry_nr;
+	if (unlikely(crc_retry_delta))
+		return crc_retry_delta < 0;
 
-	return p1.idx < p2.idx;
+	/* Pick at random, biased in favor of the faster device: */
+
+	return bch2_get_random_u64_below(p1_latency + p2_latency) > p1_latency;
 }
 
 /*
@@ -103,78 +186,131 @@ static inline bool ptr_better(struct bch_fs *c,
  */
 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 			       struct bch_io_failures *failed,
-			       struct extent_ptr_decoded *pick)
+			       struct extent_ptr_decoded *pick,
+			       int dev)
 {
+	bool have_csum_errors = false, have_io_errors = false, have_missing_devs = false;
+	bool have_dirty_ptrs = false, have_pick = false;
+
+	if (k.k->type == KEY_TYPE_error)
+		return bch_err_throw(c, key_type_error);
+
+	rcu_read_lock();
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
 	struct extent_ptr_decoded p;
-	struct bch_dev_io_failures *f;
-	struct bch_dev *ca;
-	int ret = 0;
-
-	if (k.k->type == KEY_TYPE_error)
-		return -EIO;
+	u64 pick_latency;
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+		have_dirty_ptrs |= !p.ptr.cached;
+
 		/*
 		 * Unwritten extent: no need to actually read, treat it as a
 		 * hole and return 0s:
 		 */
-		if (p.ptr.unwritten)
+		if (p.ptr.unwritten) {
+			rcu_read_unlock();
 			return 0;
+		}
+
+		/* Are we being asked to read from a specific device? */
+		if (dev >= 0 && p.ptr.dev != dev)
+			continue;
 
-		ca = bch_dev_bkey_exists(c, p.ptr.dev);
+		struct bch_dev *ca = bch2_dev_rcu_noerror(c, p.ptr.dev);
 
-		/*
-		 * If there are any dirty pointers it's an error if we can't
-		 * read:
-		 */
-		if (!ret && !p.ptr.cached)
-			ret = -EIO;
+		if (unlikely(!ca && p.ptr.dev != BCH_SB_MEMBER_INVALID)) {
+			rcu_read_unlock();
+			int ret = bch2_dev_missing_bkey(c, k, p.ptr.dev);
+			if (ret)
+				return ret;
+			rcu_read_lock();
+		}
 
-		if (p.ptr.cached && ptr_stale(ca, &p.ptr))
+		if (p.ptr.cached && (!ca || dev_ptr_stale_rcu(ca, &p.ptr)))
 			continue;
 
-		f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
-		if (f)
-			p.idx = f->nr_failed < f->nr_retries
-				? f->idx
-				: f->idx + 1;
+		struct bch_dev_io_failures *f =
+			unlikely(failed) ? bch2_dev_io_failures(failed, p.ptr.dev) : NULL;
+		if (unlikely(f)) {
+			p.crc_retry_nr	   = f->failed_csum_nr;
+			p.has_ec	  &= ~f->failed_ec;
 
-		if (!p.idx &&
-		    !bch2_dev_is_readable(ca))
-			p.idx++;
+			if (ca && ca->mi.state != BCH_MEMBER_STATE_failed) {
+				have_io_errors	|= f->failed_io;
+				have_io_errors	|= f->failed_btree_validate;
+				have_io_errors	|= f->failed_ec;
+			}
+			have_csum_errors	|= !!f->failed_csum_nr;
+
+			if (p.has_ec && (f->failed_io || f->failed_csum_nr))
+				p.do_ec_reconstruct = true;
+			else if (f->failed_io ||
+				 f->failed_btree_validate ||
+				 f->failed_csum_nr > c->opts.checksum_err_retry_nr)
+				continue;
+		}
 
-		if (bch2_force_reconstruct_read &&
-		    !p.idx && p.has_ec)
-			p.idx++;
+		have_missing_devs |= ca && !bch2_dev_is_online(ca);
 
-		if (p.idx >= (unsigned) p.has_ec + 1)
-			continue;
+		if (!ca || !bch2_dev_is_online(ca)) {
+			if (!p.has_ec)
+				continue;
+			p.do_ec_reconstruct = true;
+		}
 
-		if (ret > 0 && !ptr_better(c, p, *pick))
-			continue;
+		if (static_branch_unlikely(&bch2_force_reconstruct_read) && p.has_ec)
+			p.do_ec_reconstruct = true;
 
-		*pick = p;
-		ret = 1;
+		u64 p_latency = dev_latency(ca);
+		/*
+		 * Square the latencies, to bias more in favor of the faster
+		 * device - we never want to stop issuing reads to the slower
+		 * device altogether, so that we can update our latency numbers:
+		 */
+		p_latency *= p_latency;
+
+		if (!have_pick ||
+		    ptr_better(c,
+			       p, p_latency, ca,
+			       *pick, pick_latency)) {
+			*pick = p;
+			pick_latency = p_latency;
+			have_pick = true;
+		}
 	}
+	rcu_read_unlock();
 
-	return ret;
+	if (have_pick)
+		return 1;
+	if (!have_dirty_ptrs)
+		return 0;
+	if (have_missing_devs)
+		return bch_err_throw(c, no_device_to_read_from);
+	if (have_csum_errors)
+		return bch_err_throw(c, data_read_csum_err);
+	if (have_io_errors)
+		return bch_err_throw(c, data_read_io_err);
+
+	/*
+	 * If we get here, we have pointers (bkey_ptrs_validate() ensures that),
+	 * but they don't point to valid devices:
+	 */
+	return bch_err_throw(c, no_devices_valid);
 }
 
 /* KEY_TYPE_btree_ptr: */
 
-int bch2_btree_ptr_invalid(struct bch_fs *c, struct bkey_s_c k,
-			   enum bkey_invalid_flags flags,
-			   struct printbuf *err)
+int bch2_btree_ptr_validate(struct bch_fs *c, struct bkey_s_c k,
+			    struct bkey_validate_context from)
 {
 	int ret = 0;
 
-	bkey_fsck_err_on(bkey_val_u64s(k.k) > BCH_REPLICAS_MAX, c, err,
-			 btree_ptr_val_too_big,
+	bkey_fsck_err_on(bkey_val_u64s(k.k) > BCH_REPLICAS_MAX,
+			 c, btree_ptr_val_too_big,
 			 "value too big (%zu > %u)", bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 
-	ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
+	ret = bch2_bkey_ptrs_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -185,18 +321,28 @@ void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 	bch2_bkey_ptrs_to_text(out, c, k);
 }
 
-int bch2_btree_ptr_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
-			      enum bkey_invalid_flags flags,
-			      struct printbuf *err)
+int bch2_btree_ptr_v2_validate(struct bch_fs *c, struct bkey_s_c k,
+			       struct bkey_validate_context from)
 {
+	struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX, c, err,
-			 btree_ptr_v2_val_too_big,
+	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX,
+			 c, btree_ptr_v2_val_too_big,
 			 "value too big (%zu > %zu)",
 			 bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 
-	ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
+	bkey_fsck_err_on(bpos_ge(bp.v->min_key, bp.k->p),
+			 c, btree_ptr_v2_min_key_bad,
+			 "min_key > key");
+
+	if ((from.flags & BCH_VALIDATE_write) &&
+	    c->sb.version_min >= bcachefs_metadata_version_btree_ptr_sectors_written)
+		bkey_fsck_err_on(!bp.v->sectors_written,
+				 c, btree_ptr_v2_written_0,
+				 "sectors_written == 0");
+
+	ret = bch2_bkey_ptrs_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -242,7 +388,6 @@ bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 	const union bch_extent_entry *en_r;
 	struct extent_ptr_decoded lp, rp;
 	bool use_right_ptr;
-	struct bch_dev *ca;
 
 	en_l = l_ptrs.start;
 	en_r = r_ptrs.start;
@@ -262,6 +407,8 @@ bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 	lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 	rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 
+	guard(rcu)();
+
 	while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 	       __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 		if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
@@ -273,8 +420,10 @@ bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 			return false;
 
 		/* Extents may not straddle buckets: */
-		ca = bch_dev_bkey_exists(c, lp.ptr.dev);
-		if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
+		struct bch_dev *ca = bch2_dev_rcu(c, lp.ptr.dev);
+		bool same_bucket = ca && PTR_BUCKET_NR(ca, &lp.ptr) == PTR_BUCKET_NR(ca, &rp.ptr);
+
+		if (!same_bucket)
 			return false;
 
 		if (lp.has_ec			!= rp.has_ec ||
@@ -379,15 +528,14 @@ bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 
 /* KEY_TYPE_reservation: */
 
-int bch2_reservation_invalid(struct bch_fs *c, struct bkey_s_c k,
-			     enum bkey_invalid_flags flags,
-			     struct printbuf *err)
+int bch2_reservation_validate(struct bch_fs *c, struct bkey_s_c k,
+			      struct bkey_validate_context from)
 {
 	struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX, c, err,
-			 reservation_key_nr_replicas_invalid,
+	bkey_fsck_err_on(!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX,
+			 c, reservation_key_nr_replicas_invalid,
 			 "invalid nr_replicas (%u)", r.v->nr_replicas);
 fsck_err:
 	return ret;
@@ -512,29 +660,35 @@ static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 				 struct bch_extent_crc_unpacked src,
 				 enum bch_extent_entry_type type)
 {
-#define set_common_fields(_dst, _src)					\
-		_dst.type		= 1 << type;			\
-		_dst.csum_type		= _src.csum_type,		\
-		_dst.compression_type	= _src.compression_type,	\
-		_dst._compressed_size	= _src.compressed_size - 1,	\
-		_dst._uncompressed_size	= _src.uncompressed_size - 1,	\
-		_dst.offset		= _src.offset
+#define common_fields(_src)						\
+		.type			= BIT(type),			\
+		.csum_type		= _src.csum_type,		\
+		.compression_type	= _src.compression_type,	\
+		._compressed_size	= _src.compressed_size - 1,	\
+		._uncompressed_size	= _src.uncompressed_size - 1,	\
+		.offset			= _src.offset
 
 	switch (type) {
 	case BCH_EXTENT_ENTRY_crc32:
-		set_common_fields(dst->crc32, src);
-		dst->crc32.csum		= (u32 __force) *((__le32 *) &src.csum.lo);
+		dst->crc32		= (struct bch_extent_crc32) {
+			common_fields(src),
+			.csum		= (u32 __force) *((__le32 *) &src.csum.lo),
+		};
 		break;
 	case BCH_EXTENT_ENTRY_crc64:
-		set_common_fields(dst->crc64, src);
-		dst->crc64.nonce	= src.nonce;
-		dst->crc64.csum_lo	= (u64 __force) src.csum.lo;
-		dst->crc64.csum_hi	= (u64 __force) *((__le16 *) &src.csum.hi);
+		dst->crc64		= (struct bch_extent_crc64) {
+			common_fields(src),
+			.nonce		= src.nonce,
+			.csum_lo	= (u64 __force) src.csum.lo,
+			.csum_hi	= (u64 __force) *((__le16 *) &src.csum.hi),
+		};
 		break;
 	case BCH_EXTENT_ENTRY_crc128:
-		set_common_fields(dst->crc128, src);
-		dst->crc128.nonce	= src.nonce;
-		dst->crc128.csum	= src.csum;
+		dst->crc128		= (struct bch_extent_crc128) {
+			common_fields(src),
+			.nonce		= src.nonce,
+			.csum		= src.csum,
+		};
 		break;
 	default:
 		BUG();
@@ -662,16 +816,16 @@ static inline unsigned __extent_ptr_durability(struct bch_dev *ca, struct extent
 
 unsigned bch2_extent_ptr_desired_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 {
-	struct bch_dev *ca = bch_dev_bkey_exists(c, p->ptr.dev);
+	struct bch_dev *ca = bch2_dev_rcu(c, p->ptr.dev);
 
-	return __extent_ptr_durability(ca, p);
+	return ca ? __extent_ptr_durability(ca, p) : 0;
 }
 
 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 {
-	struct bch_dev *ca = bch_dev_bkey_exists(c, p->ptr.dev);
+	struct bch_dev *ca = bch2_dev_rcu(c, p->ptr.dev);
 
-	if (ca->mi.state == BCH_MEMBER_STATE_failed)
+	if (!ca || ca->mi.state == BCH_MEMBER_STATE_failed)
 		return 0;
 
 	return __extent_ptr_durability(ca, p);
@@ -684,9 +838,9 @@ unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 	struct extent_ptr_decoded p;
 	unsigned durability = 0;
 
+	guard(rcu)();
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		durability += bch2_extent_ptr_durability(c, &p);
-
 	return durability;
 }
 
@@ -697,10 +851,10 @@ static unsigned bch2_bkey_durability_safe(struct bch_fs *c, struct bkey_s_c k)
 	struct extent_ptr_decoded p;
 	unsigned durability = 0;
 
+	guard(rcu)();
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 			durability += bch2_extent_ptr_durability(c, &p);
-
 	return durability;
 }
 
@@ -760,14 +914,17 @@ static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 /*
  * Returns pointer to the next entry after the one being dropped:
  */
-union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
-						   struct bch_extent_ptr *ptr)
+void bch2_bkey_drop_ptr_noerror(struct bkey_s k, struct bch_extent_ptr *ptr)
 {
 	struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 	union bch_extent_entry *entry = to_entry(ptr), *next;
-	union bch_extent_entry *ret = entry;
 	bool drop_crc = true;
 
+	if (k.k->type == KEY_TYPE_stripe) {
+		ptr->dev = BCH_SB_MEMBER_INVALID;
+		return;
+	}
+
 	EBUG_ON(ptr < &ptrs.start->ptr ||
 		ptr >= &ptrs.end->ptr);
 	EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
@@ -790,21 +947,28 @@ union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
 			break;
 
 		if ((extent_entry_is_crc(entry) && drop_crc) ||
-		    extent_entry_is_stripe_ptr(entry)) {
-			ret = (void *) ret - extent_entry_bytes(entry);
+		    extent_entry_is_stripe_ptr(entry))
 			extent_entry_drop(k, entry);
-		}
 	}
-
-	return ret;
 }
 
-union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
-					   struct bch_extent_ptr *ptr)
+void bch2_bkey_drop_ptr(struct bkey_s k, struct bch_extent_ptr *ptr)
 {
+	if (k.k->type != KEY_TYPE_stripe) {
+		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k.s_c);
+		const union bch_extent_entry *entry;
+		struct extent_ptr_decoded p;
+
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+			if (p.ptr.dev == ptr->dev && p.has_ec) {
+				ptr->dev = BCH_SB_MEMBER_INVALID;
+				return;
+			}
+	}
+
 	bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
-	union bch_extent_entry *ret =
-		bch2_bkey_drop_ptr_noerror(k, ptr);
+
+	bch2_bkey_drop_ptr_noerror(k, ptr);
 
 	/*
 	 * If we deleted all the dirty pointers and there's still cached
@@ -816,29 +980,20 @@ union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 	    !bch2_bkey_dirty_devs(k.s_c).nr) {
 		k.k->type = KEY_TYPE_error;
 		set_bkey_val_u64s(k.k, 0);
-		ret = NULL;
 	} else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 		k.k->type = KEY_TYPE_deleted;
 		set_bkey_val_u64s(k.k, 0);
-		ret = NULL;
 	}
-
-	return ret;
 }
 
 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 {
-	struct bch_extent_ptr *ptr;
-
 	bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 }
 
 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 {
-	struct bch_extent_ptr *ptr = bch2_bkey_has_device(k, dev);
-
-	if (ptr)
-		bch2_bkey_drop_ptr_noerror(k, ptr);
+	bch2_bkey_drop_ptrs_noerror(k, ptr, ptr->dev == dev);
 }
 
 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
@@ -855,11 +1010,14 @@ const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned
 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	struct bch_dev *ca;
 
+	guard(rcu)();
 	bkey_for_each_ptr(ptrs, ptr)
 		if (bch2_dev_in_target(c, ptr->dev, target) &&
+		    (ca = bch2_dev_rcu(c, ptr->dev)) &&
 		    (!ptr->cached ||
-		     !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
+		     !dev_ptr_stale_rcu(ca, ptr)))
 			return true;
 
 	return false;
@@ -903,8 +1061,29 @@ bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 			bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 				if (p1.ptr.dev		== p2.ptr.dev &&
 				    p1.ptr.gen		== p2.ptr.gen &&
+
+				    /*
+				     * This checks that the two pointers point
+				     * to the same region on disk - adjusting
+				     * for the difference in where the extents
+				     * start, since one may have been trimmed:
+				     */
 				    (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
-				    (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
+				    (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k) &&
+
+				    /*
+				     * This additionally checks that the
+				     * extents overlap on disk, since the
+				     * previous check may trigger spuriously
+				     * when one extent is immediately partially
+				     * overwritten with another extent (so that
+				     * on disk they are adjacent) and
+				     * compression is in use:
+				     */
+				    ((p1.ptr.offset >= p2.ptr.offset &&
+				      p1.ptr.offset  < p2.ptr.offset + p2.crc.compressed_size) ||
+				     (p2.ptr.offset >= p1.ptr.offset &&
+				      p2.ptr.offset  < p1.ptr.offset + p1.crc.compressed_size)))
 					return true;
 
 		return false;
@@ -931,31 +1110,70 @@ bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bke
 	return NULL;
 }
 
-void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
+static bool want_cached_ptr(struct bch_fs *c, struct bch_io_opts *opts,
+			    struct bch_extent_ptr *ptr)
 {
-	struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
+	unsigned target = opts->promote_target ?: opts->foreground_target;
+
+	if (target && !bch2_dev_in_target(c, ptr->dev, target))
+		return false;
+
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
+
+	return ca && bch2_dev_is_healthy(ca) && !dev_ptr_stale_rcu(ca, ptr);
+}
+
+void bch2_extent_ptr_set_cached(struct bch_fs *c,
+				struct bch_io_opts *opts,
+				struct bkey_s k,
+				struct bch_extent_ptr *ptr)
+{
+	struct bkey_ptrs ptrs;
 	union bch_extent_entry *entry;
-	union bch_extent_entry *ec = NULL;
+	struct extent_ptr_decoded p;
+	bool have_cached_ptr;
+	unsigned drop_dev = ptr->dev;
 
-	bkey_extent_entry_for_each(ptrs, entry) {
-		if (&entry->ptr == ptr) {
-			ptr->cached = true;
-			if (ec)
-				extent_entry_drop(k, ec);
-			return;
-		}
+	guard(rcu)();
+restart_drop_ptrs:
+	ptrs = bch2_bkey_ptrs(k);
+	have_cached_ptr = false;
 
-		if (extent_entry_is_stripe_ptr(entry))
-			ec = entry;
-		else if (extent_entry_is_ptr(entry))
-			ec = NULL;
+	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+		/*
+		 * Check if it's erasure coded - stripes can't contain cached
+		 * data. Possibly something we can fix in the future?
+		 */
+		if (&entry->ptr == ptr && p.has_ec)
+			goto drop;
+
+		if (p.ptr.cached) {
+			if (have_cached_ptr || !want_cached_ptr(c, opts, &p.ptr)) {
+				bch2_bkey_drop_ptr_noerror(k, &entry->ptr);
+				ptr = NULL;
+				goto restart_drop_ptrs;
+			}
+
+			have_cached_ptr = true;
+		}
 	}
 
-	BUG();
+	if (!ptr)
+		bkey_for_each_ptr(ptrs, ptr2)
+			if (ptr2->dev == drop_dev)
+				ptr = ptr2;
+
+	if (have_cached_ptr || !want_cached_ptr(c, opts, ptr))
+		goto drop;
+
+	ptr->cached = true;
+	return;
+drop:
+	bch2_bkey_drop_ptr_noerror(k, ptr);
 }
 
 /*
- * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
+ * bch2_extent_normalize - clean up an extent, dropping stale pointers etc.
  *
  * Returns true if @k should be dropped entirely
  *
@@ -964,15 +1182,140 @@ void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
  */
 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 {
-	struct bch_extent_ptr *ptr;
+	struct bch_dev *ca;
 
+	guard(rcu)();
 	bch2_bkey_drop_ptrs(k, ptr,
 		ptr->cached &&
-		ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
+		(!(ca = bch2_dev_rcu(c, ptr->dev)) ||
+		 dev_ptr_stale_rcu(ca, ptr) > 0));
 
 	return bkey_deleted(k.k);
 }
 
+/*
+ * bch2_extent_normalize_by_opts - clean up an extent, dropping stale pointers etc.
+ *
+ * Like bch2_extent_normalize(), but also only keeps a single cached pointer on
+ * the promote target.
+ */
+bool bch2_extent_normalize_by_opts(struct bch_fs *c,
+				   struct bch_io_opts *opts,
+				   struct bkey_s k)
+{
+	struct bkey_ptrs ptrs;
+	bool have_cached_ptr;
+
+	guard(rcu)();
+restart_drop_ptrs:
+	ptrs = bch2_bkey_ptrs(k);
+	have_cached_ptr = false;
+
+	bkey_for_each_ptr(ptrs, ptr)
+		if (ptr->cached) {
+			if (have_cached_ptr || !want_cached_ptr(c, opts, ptr)) {
+				bch2_bkey_drop_ptr(k, ptr);
+				goto restart_drop_ptrs;
+			}
+			have_cached_ptr = true;
+		}
+
+	return bkey_deleted(k.k);
+}
+
+void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *c, const struct bch_extent_ptr *ptr)
+{
+	out->atomic++;
+	guard(rcu)();
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
+	if (!ca) {
+		prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
+			   (u64) ptr->offset, ptr->gen,
+			   ptr->cached ? " cached" : "");
+	} else {
+		u32 offset;
+		u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
+
+		prt_printf(out, "ptr: %u:%llu:%u gen %u",
+			   ptr->dev, b, offset, ptr->gen);
+		if (ca->mi.durability != 1)
+			prt_printf(out, " d=%u", ca->mi.durability);
+		if (ptr->cached)
+			prt_str(out, " cached");
+		if (ptr->unwritten)
+			prt_str(out, " unwritten");
+		int stale = dev_ptr_stale_rcu(ca, ptr);
+		if (stale > 0)
+			prt_printf(out, " stale");
+		else if (stale)
+			prt_printf(out, " invalid");
+	}
+	--out->atomic;
+}
+
+void bch2_extent_crc_unpacked_to_text(struct printbuf *out, struct bch_extent_crc_unpacked *crc)
+{
+	prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum ",
+		   crc->compressed_size,
+		   crc->uncompressed_size,
+		   crc->offset, crc->nonce);
+	bch2_prt_csum_type(out, crc->csum_type);
+	prt_printf(out, " %0llx:%0llx ", crc->csum.hi, crc->csum.lo);
+	prt_str(out, " compress ");
+	bch2_prt_compression_type(out, crc->compression_type);
+}
+
+static void bch2_extent_rebalance_to_text(struct printbuf *out, struct bch_fs *c,
+					  const struct bch_extent_rebalance *r)
+{
+	prt_str(out, "rebalance:");
+
+	prt_printf(out, " replicas=%u", r->data_replicas);
+	if (r->data_replicas_from_inode)
+		prt_str(out, " (inode)");
+
+	prt_str(out, " checksum=");
+	bch2_prt_csum_opt(out, r->data_checksum);
+	if (r->data_checksum_from_inode)
+		prt_str(out, " (inode)");
+
+	if (r->background_compression || r->background_compression_from_inode) {
+		prt_str(out, " background_compression=");
+		bch2_compression_opt_to_text(out, r->background_compression);
+
+		if (r->background_compression_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->background_target || r->background_target_from_inode) {
+		prt_str(out, " background_target=");
+		if (c)
+			bch2_target_to_text(out, c, r->background_target);
+		else
+			prt_printf(out, "%u", r->background_target);
+
+		if (r->background_target_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->promote_target || r->promote_target_from_inode) {
+		prt_str(out, " promote_target=");
+		if (c)
+			bch2_target_to_text(out, c, r->promote_target);
+		else
+			prt_printf(out, "%u", r->promote_target);
+
+		if (r->promote_target_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->erasure_code || r->erasure_code_from_inode) {
+		prt_printf(out, " ec=%u", r->erasure_code);
+		if (r->erasure_code_from_inode)
+			prt_str(out, " (inode)");
+	}
+}
+
 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			    struct bkey_s_c k)
 {
@@ -988,43 +1331,17 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			prt_printf(out, " ");
 
 		switch (__extent_entry_type(entry)) {
-		case BCH_EXTENT_ENTRY_ptr: {
-			const struct bch_extent_ptr *ptr = entry_to_ptr(entry);
-			struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
-				? bch_dev_bkey_exists(c, ptr->dev)
-				: NULL;
-
-			if (!ca) {
-				prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
-				       (u64) ptr->offset, ptr->gen,
-				       ptr->cached ? " cached" : "");
-			} else {
-				u32 offset;
-				u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
-
-				prt_printf(out, "ptr: %u:%llu:%u gen %u",
-					   ptr->dev, b, offset, ptr->gen);
-				if (ptr->cached)
-					prt_str(out, " cached");
-				if (ptr->unwritten)
-					prt_str(out, " unwritten");
-				if (ca && ptr_stale(ca, ptr))
-					prt_printf(out, " stale");
-			}
+		case BCH_EXTENT_ENTRY_ptr:
+			bch2_extent_ptr_to_text(out, c, entry_to_ptr(entry));
 			break;
-		}
+
 		case BCH_EXTENT_ENTRY_crc32:
 		case BCH_EXTENT_ENTRY_crc64:
 		case BCH_EXTENT_ENTRY_crc128: {
 			struct bch_extent_crc_unpacked crc =
 				bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
 
-			prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress ",
-			       crc.compressed_size,
-			       crc.uncompressed_size,
-			       crc.offset, crc.nonce,
-			       bch2_csum_types[crc.csum_type]);
-			bch2_prt_compression_type(out, crc.compression_type);
+			bch2_extent_crc_unpacked_to_text(out, &crc);
 			break;
 		}
 		case BCH_EXTENT_ENTRY_stripe_ptr: {
@@ -1034,18 +1351,14 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			       (u64) ec->idx, ec->block);
 			break;
 		}
-		case BCH_EXTENT_ENTRY_rebalance: {
-			const struct bch_extent_rebalance *r = &entry->rebalance;
+		case BCH_EXTENT_ENTRY_rebalance:
+			bch2_extent_rebalance_to_text(out, c, &entry->rebalance);
+			break;
 
-			prt_str(out, "rebalance: target ");
-			if (c)
-				bch2_target_to_text(out, c, r->target);
-			else
-				prt_printf(out, "%u", r->target);
-			prt_str(out, " compression ");
-			bch2_compression_opt_to_text(out, r->compression);
+		case BCH_EXTENT_ENTRY_flags:
+			prt_bitflags(out, bch2_extent_flags_strs, entry->flags.flags);
 			break;
-		}
+
 		default:
 			prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
 			return;
@@ -1055,58 +1368,51 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 	}
 }
 
-static int extent_ptr_invalid(struct bch_fs *c,
-			      struct bkey_s_c k,
-			      enum bkey_invalid_flags flags,
-			      const struct bch_extent_ptr *ptr,
-			      unsigned size_ondisk,
-			      bool metadata,
-			      struct printbuf *err)
+static int extent_ptr_validate(struct bch_fs *c,
+			       struct bkey_s_c k,
+			       struct bkey_validate_context from,
+			       const struct bch_extent_ptr *ptr,
+			       unsigned size_ondisk,
+			       bool metadata)
 {
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	u64 bucket;
-	u32 bucket_offset;
-	struct bch_dev *ca;
 	int ret = 0;
 
-	if (!bch2_dev_exists2(c, ptr->dev)) {
-		/*
-		 * If we're in the write path this key might have already been
-		 * overwritten, and we could be seeing a device that doesn't
-		 * exist anymore due to racing with device removal:
-		 */
-		if (flags & BKEY_INVALID_WRITE)
-			return 0;
-
-		bkey_fsck_err(c, err, ptr_to_invalid_device,
-			   "pointer to invalid device (%u)", ptr->dev);
-	}
-
-	ca = bch_dev_bkey_exists(c, ptr->dev);
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	bkey_for_each_ptr(ptrs, ptr2)
-		bkey_fsck_err_on(ptr != ptr2 && ptr->dev == ptr2->dev, c, err,
-				 ptr_to_duplicate_device,
+		bkey_fsck_err_on(ptr != ptr2 && ptr->dev == ptr2->dev,
+				 c, ptr_to_duplicate_device,
 				 "multiple pointers to same device (%u)", ptr->dev);
 
-	bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
-
-	bkey_fsck_err_on(bucket >= ca->mi.nbuckets, c, err,
-			 ptr_after_last_bucket,
-			 "pointer past last bucket (%llu > %llu)", bucket, ca->mi.nbuckets);
-	bkey_fsck_err_on(ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket), c, err,
-			 ptr_before_first_bucket,
-			 "pointer before first bucket (%llu < %u)", bucket, ca->mi.first_bucket);
-	bkey_fsck_err_on(bucket_offset + size_ondisk > ca->mi.bucket_size, c, err,
-			 ptr_spans_multiple_buckets,
+	/* bad pointers are repaired by check_fix_ptrs(): */
+	rcu_read_lock();
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
+	if (!ca) {
+		rcu_read_unlock();
+		return 0;
+	}
+	u32 bucket_offset;
+	u64 bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
+	unsigned first_bucket	= ca->mi.first_bucket;
+	u64 nbuckets		= ca->mi.nbuckets;
+	unsigned bucket_size	= ca->mi.bucket_size;
+	rcu_read_unlock();
+
+	bkey_fsck_err_on(bucket >= nbuckets,
+			 c, ptr_after_last_bucket,
+			 "pointer past last bucket (%llu > %llu)", bucket, nbuckets);
+	bkey_fsck_err_on(bucket < first_bucket,
+			 c, ptr_before_first_bucket,
+			 "pointer before first bucket (%llu < %u)", bucket, first_bucket);
+	bkey_fsck_err_on(bucket_offset + size_ondisk > bucket_size,
+			 c, ptr_spans_multiple_buckets,
 			 "pointer spans multiple buckets (%u + %u > %u)",
-		       bucket_offset, size_ondisk, ca->mi.bucket_size);
+		       bucket_offset, size_ondisk, bucket_size);
 fsck_err:
 	return ret;
 }
 
-int bch2_bkey_ptrs_invalid(struct bch_fs *c, struct bkey_s_c k,
-			   enum bkey_invalid_flags flags,
-			   struct printbuf *err)
+int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
+			    struct bkey_validate_context from)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
@@ -1121,25 +1427,24 @@ int bch2_bkey_ptrs_invalid(struct bch_fs *c, struct bkey_s_c k,
 		size_ondisk = btree_sectors(c);
 
 	bkey_extent_entry_for_each(ptrs, entry) {
-		bkey_fsck_err_on(__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX, c, err,
-			extent_ptrs_invalid_entry,
-			"invalid extent entry type (got %u, max %u)",
-			__extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
+		bkey_fsck_err_on(__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX,
+				 c, extent_ptrs_invalid_entry,
+				 "invalid extent entry type (got %u, max %u)",
+				 __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
 
 		bkey_fsck_err_on(bkey_is_btree_ptr(k.k) &&
-				 !extent_entry_is_ptr(entry), c, err,
-				 btree_ptr_has_non_ptr,
+				 !extent_entry_is_ptr(entry),
+				 c, btree_ptr_has_non_ptr,
 				 "has non ptr field");
 
 		switch (extent_entry_type(entry)) {
 		case BCH_EXTENT_ENTRY_ptr:
-			ret = extent_ptr_invalid(c, k, flags, &entry->ptr,
-						 size_ondisk, false, err);
+			ret = extent_ptr_validate(c, k, from, &entry->ptr, size_ondisk, false);
 			if (ret)
 				return ret;
 
-			bkey_fsck_err_on(entry->ptr.cached && have_ec, c, err,
-					 ptr_cached_and_erasure_coded,
+			bkey_fsck_err_on(entry->ptr.cached && have_ec,
+					 c, ptr_cached_and_erasure_coded,
 					 "cached, erasure coded ptr");
 
 			if (!entry->ptr.unwritten)
@@ -1156,74 +1461,90 @@ int bch2_bkey_ptrs_invalid(struct bch_fs *c, struct bkey_s_c k,
 		case BCH_EXTENT_ENTRY_crc128:
 			crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
 
-			bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size, c, err,
-					 ptr_crc_uncompressed_size_too_small,
-					 "checksum offset + key size > uncompressed size");
-			bkey_fsck_err_on(!bch2_checksum_type_valid(c, crc.csum_type), c, err,
-					 ptr_crc_csum_type_unknown,
+			bkey_fsck_err_on(!bch2_checksum_type_valid(c, crc.csum_type),
+					 c, ptr_crc_csum_type_unknown,
 					 "invalid checksum type");
-			bkey_fsck_err_on(crc.compression_type >= BCH_COMPRESSION_TYPE_NR, c, err,
-					 ptr_crc_compression_type_unknown,
+			bkey_fsck_err_on(crc.compression_type >= BCH_COMPRESSION_TYPE_NR,
+					 c, ptr_crc_compression_type_unknown,
 					 "invalid compression type");
 
+			bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size,
+					 c, ptr_crc_uncompressed_size_too_small,
+					 "checksum offset + key size > uncompressed size");
+			bkey_fsck_err_on(crc_is_encoded(crc) &&
+					 (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
+					 (from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit)),
+					 c, ptr_crc_uncompressed_size_too_big,
+					 "too large encoded extent");
+			bkey_fsck_err_on(!crc_is_compressed(crc) &&
+					 crc.compressed_size != crc.uncompressed_size,
+					 c, ptr_crc_uncompressed_size_mismatch,
+					 "not compressed but compressed != uncompressed size");
+
 			if (bch2_csum_type_is_encryption(crc.csum_type)) {
 				if (nonce == UINT_MAX)
 					nonce = crc.offset + crc.nonce;
 				else if (nonce != crc.offset + crc.nonce)
-					bkey_fsck_err(c, err, ptr_crc_nonce_mismatch,
+					bkey_fsck_err(c, ptr_crc_nonce_mismatch,
 						      "incorrect nonce");
 			}
 
-			bkey_fsck_err_on(crc_since_last_ptr, c, err,
-					 ptr_crc_redundant,
+			bkey_fsck_err_on(crc_since_last_ptr,
+					 c, ptr_crc_redundant,
 					 "redundant crc entry");
 			crc_since_last_ptr = true;
 
-			bkey_fsck_err_on(crc_is_encoded(crc) &&
-					 (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
-					 (flags & (BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT)), c, err,
-					 ptr_crc_uncompressed_size_too_big,
-					 "too large encoded extent");
-
 			size_ondisk = crc.compressed_size;
 			break;
 		case BCH_EXTENT_ENTRY_stripe_ptr:
-			bkey_fsck_err_on(have_ec, c, err,
-					 ptr_stripe_redundant,
+			bkey_fsck_err_on(have_ec,
+					 c, ptr_stripe_redundant,
 					 "redundant stripe entry");
 			have_ec = true;
 			break;
 		case BCH_EXTENT_ENTRY_rebalance: {
+			/*
+			 * this shouldn't be a fsck error, for forward
+			 * compatibility; the rebalance code should just refetch
+			 * the compression opt if it's unknown
+			 */
+#if 0
 			const struct bch_extent_rebalance *r = &entry->rebalance;
 
 			if (!bch2_compression_opt_valid(r->compression)) {
 				struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
 				prt_printf(err, "invalid compression opt %u:%u",
 					   opt.type, opt.level);
-				return -BCH_ERR_invalid_bkey;
+				return bch_err_throw(c, invalid_bkey);
 			}
+#endif
 			break;
 		}
+		case BCH_EXTENT_ENTRY_flags:
+			bkey_fsck_err_on(entry != ptrs.start,
+					 c, extent_flags_not_at_start,
+					 "extent flags entry not at start");
+			break;
 		}
 	}
 
-	bkey_fsck_err_on(!nr_ptrs, c, err,
-			 extent_ptrs_no_ptrs,
+	bkey_fsck_err_on(!nr_ptrs,
+			 c, extent_ptrs_no_ptrs,
 			 "no ptrs");
-	bkey_fsck_err_on(nr_ptrs > BCH_BKEY_PTRS_MAX, c, err,
-			 extent_ptrs_too_many_ptrs,
+	bkey_fsck_err_on(nr_ptrs > BCH_BKEY_PTRS_MAX,
+			 c, extent_ptrs_too_many_ptrs,
 			 "too many ptrs: %u > %u", nr_ptrs, BCH_BKEY_PTRS_MAX);
-	bkey_fsck_err_on(have_written && have_unwritten, c, err,
-			 extent_ptrs_written_and_unwritten,
+	bkey_fsck_err_on(have_written && have_unwritten,
+			 c, extent_ptrs_written_and_unwritten,
 			 "extent with unwritten and written ptrs");
-	bkey_fsck_err_on(k.k->type != KEY_TYPE_extent && have_unwritten, c, err,
-			 extent_ptrs_unwritten,
+	bkey_fsck_err_on(k.k->type != KEY_TYPE_extent && have_unwritten,
+			 c, extent_ptrs_unwritten,
 			 "has unwritten ptrs");
-	bkey_fsck_err_on(crc_since_last_ptr, c, err,
-			 extent_ptrs_redundant_crc,
+	bkey_fsck_err_on(crc_since_last_ptr,
+			 c, extent_ptrs_redundant_crc,
 			 "redundant crc entry");
-	bkey_fsck_err_on(have_ec, c, err,
-			 extent_ptrs_redundant_stripe,
+	bkey_fsck_err_on(have_ec,
+			 c, extent_ptrs_redundant_stripe,
 			 "redundant stripe entry");
 fsck_err:
 	return ret;
@@ -1243,7 +1564,7 @@ void bch2_ptr_swab(struct bkey_s k)
 	for (entry = ptrs.start;
 	     entry < ptrs.end;
 	     entry = extent_entry_next(entry)) {
-		switch (extent_entry_type(entry)) {
+		switch (__extent_entry_type(entry)) {
 		case BCH_EXTENT_ENTRY_ptr:
 			break;
 		case BCH_EXTENT_ENTRY_crc32:
@@ -1263,126 +1584,30 @@ void bch2_ptr_swab(struct bkey_s k)
 			break;
 		case BCH_EXTENT_ENTRY_rebalance:
 			break;
+		default:
+			/* Bad entry type: will be caught by validate() */
+			return;
 		}
 	}
 }
 
-const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c k)
-{
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	const union bch_extent_entry *entry;
-
-	bkey_extent_entry_for_each(ptrs, entry)
-		if (__extent_entry_type(entry) == BCH_EXTENT_ENTRY_rebalance)
-			return &entry->rebalance;
-
-	return NULL;
-}
-
-unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *c, struct bkey_s_c k,
-				       unsigned target, unsigned compression)
-{
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	unsigned rewrite_ptrs = 0;
-
-	if (compression) {
-		unsigned compression_type = bch2_compression_opt_to_type(compression);
-		const union bch_extent_entry *entry;
-		struct extent_ptr_decoded p;
-		unsigned i = 0;
-
-		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-			if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible ||
-			    p.ptr.unwritten) {
-				rewrite_ptrs = 0;
-				goto incompressible;
-			}
-
-			if (!p.ptr.cached && p.crc.compression_type != compression_type)
-				rewrite_ptrs |= 1U << i;
-			i++;
-		}
-	}
-incompressible:
-	if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
-		unsigned i = 0;
-
-		bkey_for_each_ptr(ptrs, ptr) {
-			if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, target))
-				rewrite_ptrs |= 1U << i;
-			i++;
-		}
-	}
-
-	return rewrite_ptrs;
-}
-
-bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k)
-{
-	const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
-
-	/*
-	 * If it's an indirect extent, we don't delete the rebalance entry when
-	 * done so that we know what options were applied - check if it still
-	 * needs work done:
-	 */
-	if (r &&
-	    k.k->type == KEY_TYPE_reflink_v &&
-	    !bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression))
-		r = NULL;
-
-	return r != NULL;
-}
-
-int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
-				  struct bch_io_opts *opts)
+int bch2_bkey_extent_flags_set(struct bch_fs *c, struct bkey_i *k, u64 flags)
 {
-	struct bkey_s k = bkey_i_to_s(_k);
-	struct bch_extent_rebalance *r;
-	unsigned target = opts->background_target;
-	unsigned compression = background_compression(*opts);
-	bool needs_rebalance;
+	int ret = bch2_request_incompat_feature(c, bcachefs_metadata_version_extent_flags);
+	if (ret)
+		return ret;
 
-	if (!bkey_extent_is_direct_data(k.k))
-		return 0;
-
-	/* get existing rebalance entry: */
-	r = (struct bch_extent_rebalance *) bch2_bkey_rebalance_opts(k.s_c);
-	if (r) {
-		if (k.k->type == KEY_TYPE_reflink_v) {
-			/*
-			 * indirect extents: existing options take precedence,
-			 * so that we don't move extents back and forth if
-			 * they're referenced by different inodes with different
-			 * options:
-			 */
-			if (r->target)
-				target = r->target;
-			if (r->compression)
-				compression = r->compression;
-		}
-
-		r->target	= target;
-		r->compression	= compression;
-	}
-
-	needs_rebalance = bch2_bkey_ptrs_need_rebalance(c, k.s_c, target, compression);
-
-	if (needs_rebalance && !r) {
-		union bch_extent_entry *new = bkey_val_end(k);
+	struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 
-		new->rebalance.type		= 1U << BCH_EXTENT_ENTRY_rebalance;
-		new->rebalance.compression	= compression;
-		new->rebalance.target		= target;
-		new->rebalance.unused		= 0;
-		k.k->u64s += extent_entry_u64s(new);
-	} else if (!needs_rebalance && r && k.k->type != KEY_TYPE_reflink_v) {
-		/*
-		 * For indirect extents, don't delete the rebalance entry when
-		 * we're finished so that we know we specifically moved it or
-		 * compressed it to its current location/compression type
-		 */
-		extent_entry_drop(k, (union bch_extent_entry *) r);
+	if (ptrs.start != ptrs.end &&
+	    extent_entry_type(ptrs.start) == BCH_EXTENT_ENTRY_flags) {
+		ptrs.start->flags.flags = flags;
+	} else {
+		struct bch_extent_flags f = {
+			.type	= BIT(BCH_EXTENT_ENTRY_flags),
+			.flags	= flags,
+		};
+		__extent_entry_insert(k, ptrs.start, (union bch_extent_entry *) &f);
 	}
 
 	return 0;
@@ -1433,8 +1658,8 @@ int bch2_cut_front_s(struct bpos where, struct bkey_s k)
 				entry->crc128.offset += sub;
 				break;
 			case BCH_EXTENT_ENTRY_stripe_ptr:
-				break;
 			case BCH_EXTENT_ENTRY_rebalance:
+			case BCH_EXTENT_ENTRY_flags:
 				break;
 			}
 
@@ -1447,7 +1672,7 @@ int bch2_cut_front_s(struct bpos where, struct bkey_s k)
 	case KEY_TYPE_reflink_p: {
 		struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
 
-		le64_add_cpu(&p.v->idx, sub);
+		SET_REFLINK_P_IDX(p.v, REFLINK_P_IDX(p.v) + sub);
 		break;
 	}
 	case KEY_TYPE_inline_data: