X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Freflink.c;h=6e1bfe9feb59e4abe96e1dc74b30196fa5766f48;hb=0a08ddf78c9cf4b6671ba64b049c37da64233f4f;hp=930547de33091754aae2a19c62ce2e3fd4ded961;hpb=f39f0bde7875aceb3e82a330f3a86223c6dd8af4;p=bcachefs-tools-debian

diff --git a/libbcachefs/reflink.c b/libbcachefs/reflink.c
index 930547d..6e1bfe9 100644
--- a/libbcachefs/reflink.c
+++ b/libbcachefs/reflink.c
@@ -2,10 +2,15 @@
 #include "bcachefs.h"
 #include "bkey_buf.h"
 #include "btree_update.h"
+#include "buckets.h"
 #include "extents.h"
 #include "inode.h"
-#include "io.h"
+#include "io_misc.h"
+#include "io_write.h"
+#include "rebalance.h"
 #include "reflink.h"
+#include "subvolume.h"
+#include "super-io.h"
 
 #include <linux/sched/signal.h>
 
@@ -23,14 +28,20 @@ static inline unsigned bkey_type_to_indirect(const struct bkey *k)
 
 /* reflink pointers */
 
-const char *bch2_reflink_p_invalid(const struct bch_fs *c, struct bkey_s_c k)
+int bch2_reflink_p_invalid(struct bch_fs *c, struct bkey_s_c k,
+			   enum bkey_invalid_flags flags,
+			   struct printbuf *err)
 {
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 
-	if (bkey_val_bytes(p.k) != sizeof(*p.v))
-		return "incorrect value size";
+	if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix &&
+	    le64_to_cpu(p.v->idx) < le32_to_cpu(p.v->front_pad)) {
+		prt_printf(err, "idx < front_pad (%llu < %u)",
+		       le64_to_cpu(p.v->idx), le32_to_cpu(p.v->front_pad));
+		return -EINVAL;
+	}
 
-	return NULL;
+	return 0;
 }
 
 void bch2_reflink_p_to_text(struct printbuf *out, struct bch_fs *c,
@@ -38,39 +49,37 @@ void bch2_reflink_p_to_text(struct printbuf *out, struct bch_fs *c,
 {
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 
-	pr_buf(out, "idx %llu", le64_to_cpu(p.v->idx));
+	prt_printf(out, "idx %llu front_pad %u back_pad %u",
+	       le64_to_cpu(p.v->idx),
+	       le32_to_cpu(p.v->front_pad),
+	       le32_to_cpu(p.v->back_pad));
 }
 
-enum merge_result bch2_reflink_p_merge(struct bch_fs *c,
-				       struct bkey_s _l, struct bkey_s _r)
+bool bch2_reflink_p_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 {
 	struct bkey_s_reflink_p l = bkey_s_to_reflink_p(_l);
-	struct bkey_s_reflink_p r = bkey_s_to_reflink_p(_r);
+	struct bkey_s_c_reflink_p r = bkey_s_c_to_reflink_p(_r);
 
-	if (le64_to_cpu(l.v->idx) + l.k->size != le64_to_cpu(r.v->idx))
-		return BCH_MERGE_NOMERGE;
+	/*
+	 * Disabled for now, the triggers code needs to be reworked for merging
+	 * of reflink pointers to work:
+	 */
+	return false;
 
-	if ((u64) l.k->size + r.k->size > KEY_SIZE_MAX) {
-		bch2_key_resize(l.k, KEY_SIZE_MAX);
-		bch2_cut_front_s(l.k->p, _r);
-		return BCH_MERGE_PARTIAL;
-	}
+	if (le64_to_cpu(l.v->idx) + l.k->size != le64_to_cpu(r.v->idx))
+		return false;
 
 	bch2_key_resize(l.k, l.k->size + r.k->size);
-
-	return BCH_MERGE_MERGE;
+	return true;
 }
 
 /* indirect extents */
 
-const char *bch2_reflink_v_invalid(const struct bch_fs *c, struct bkey_s_c k)
+int bch2_reflink_v_invalid(struct bch_fs *c, struct bkey_s_c k,
+			   enum bkey_invalid_flags flags,
+			   struct printbuf *err)
 {
-	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);
-
-	if (bkey_val_bytes(r.k) < sizeof(*r.v))
-		return "incorrect value size";
-
-	return bch2_bkey_ptrs_invalid(c, k);
+	return bch2_bkey_ptrs_invalid(c, k, flags, err);
 }
 
 void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
@@ -78,38 +87,79 @@ void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
 {
 	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);
 
-	pr_buf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
+	prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
 
 	bch2_bkey_ptrs_to_text(out, c, k);
 }
 
+#if 0
+Currently disabled, needs to be debugged:
+
+bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
+{
+	struct bkey_s_reflink_v   l = bkey_s_to_reflink_v(_l);
+	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(_r);
+
+	return l.v->refcount == r.v->refcount && bch2_extent_merge(c, _l, _r);
+}
+#endif
+
+static inline void check_indirect_extent_deleting(struct bkey_i *new, unsigned *flags)
+{
+	if ((*flags & BTREE_TRIGGER_INSERT) && !*bkey_refcount(new)) {
+		new->k.type = KEY_TYPE_deleted;
+		new->k.size = 0;
+		set_bkey_val_u64s(&new->k, 0);;
+		*flags &= ~BTREE_TRIGGER_INSERT;
+	}
+}
+
+int bch2_trans_mark_reflink_v(struct btree_trans *trans,
+			      enum btree_id btree_id, unsigned level,
+			      struct bkey_s_c old, struct bkey_i *new,
+			      unsigned flags)
+{
+	check_indirect_extent_deleting(new, &flags);
+
+	return bch2_trans_mark_extent(trans, btree_id, level, old, new, flags);
+}
+
 /* indirect inline data */
 
-const char *bch2_indirect_inline_data_invalid(const struct bch_fs *c,
-					      struct bkey_s_c k)
+int bch2_indirect_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k,
+				      enum bkey_invalid_flags flags,
+				      struct printbuf *err)
 {
-	if (bkey_val_bytes(k.k) < sizeof(struct bch_indirect_inline_data))
-		return "incorrect value size";
-	return NULL;
+	return 0;
 }
 
 void bch2_indirect_inline_data_to_text(struct printbuf *out,
-					struct bch_fs *c, struct bkey_s_c k)
+				       struct bch_fs *c, struct bkey_s_c k)
 {
 	struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k);
 	unsigned datalen = bkey_inline_data_bytes(k.k);
 
-	pr_buf(out, "refcount %llu datalen %u: %*phN",
+	prt_printf(out, "refcount %llu datalen %u: %*phN",
 	       le64_to_cpu(d.v->refcount), datalen,
 	       min(datalen, 32U), d.v->data);
 }
 
+int bch2_trans_mark_indirect_inline_data(struct btree_trans *trans,
+			      enum btree_id btree_id, unsigned level,
+			      struct bkey_s_c old, struct bkey_i *new,
+			      unsigned flags)
+{
+	check_indirect_extent_deleting(new, &flags);
+
+	return 0;
+}
+
 static int bch2_make_extent_indirect(struct btree_trans *trans,
 				     struct btree_iter *extent_iter,
 				     struct bkey_i *orig)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *reflink_iter;
+	struct btree_iter reflink_iter = { NULL };
 	struct bkey_s_c k;
 	struct bkey_i *r_v;
 	struct bkey_i_reflink_p *r_p;
@@ -119,96 +169,98 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	if (orig->k.type == KEY_TYPE_inline_data)
 		bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data);
 
-	for_each_btree_key(trans, reflink_iter, BTREE_ID_REFLINK,
-			   POS(0, c->reflink_hint),
-			   BTREE_ITER_INTENT|BTREE_ITER_SLOTS, k, ret) {
-		if (reflink_iter->pos.inode) {
-			bch2_btree_iter_set_pos(reflink_iter, POS_MIN);
-			continue;
-		}
-
-		if (bkey_deleted(k.k) && orig->k.size <= k.k->size)
-			break;
-	}
-
+	bch2_trans_iter_init(trans, &reflink_iter, BTREE_ID_reflink, POS_MAX,
+			     BTREE_ITER_INTENT);
+	k = bch2_btree_iter_peek_prev(&reflink_iter);
+	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	/* rewind iter to start of hole, if necessary: */
-	bch2_btree_iter_set_pos(reflink_iter, bkey_start_pos(k.k));
-
-	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_val_bytes(&orig->k));
+	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k));
 	ret = PTR_ERR_OR_ZERO(r_v);
 	if (ret)
 		goto err;
 
 	bkey_init(&r_v->k);
 	r_v->k.type	= bkey_type_to_indirect(&orig->k);
-	r_v->k.p	= reflink_iter->pos;
+	r_v->k.p	= reflink_iter.pos;
 	bch2_key_resize(&r_v->k, orig->k.size);
 	r_v->k.version	= orig->k.version;
 
 	set_bkey_val_bytes(&r_v->k, sizeof(__le64) + bkey_val_bytes(&orig->k));
 
-	refcount	= (void *) &r_v->v;
+	refcount	= bkey_refcount(r_v);
 	*refcount	= 0;
 	memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));
 
-	bch2_trans_update(trans, reflink_iter, r_v, 0);
-
-	r_p = bch2_trans_kmalloc(trans, sizeof(*r_p));
-	if (IS_ERR(r_p))
-		return PTR_ERR(r_p);
+	ret = bch2_trans_update(trans, &reflink_iter, r_v, 0);
+	if (ret)
+		goto err;
 
+	/*
+	 * orig is in a bkey_buf which statically allocates 5 64s for the val,
+	 * so we know it will be big enough:
+	 */
 	orig->k.type = KEY_TYPE_reflink_p;
 	r_p = bkey_i_to_reflink_p(orig);
 	set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
+
+	/* FORTIFY_SOURCE is broken here, and doesn't provide unsafe_memset() */
+#if !defined(__NO_FORTIFY) && defined(__OPTIMIZE__) && defined(CONFIG_FORTIFY_SOURCE)
+	__underlying_memset(&r_p->v, 0, sizeof(r_p->v));
+#else
+	memset(&r_p->v, 0, sizeof(r_p->v));
+#endif
+
 	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
 
-	bch2_trans_update(trans, extent_iter, &r_p->k_i, 0);
+	ret = bch2_trans_update(trans, extent_iter, &r_p->k_i,
+				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 err:
-	if (!IS_ERR(reflink_iter))
-		c->reflink_hint = reflink_iter->pos.offset;
-	bch2_trans_iter_put(trans, reflink_iter);
+	bch2_trans_iter_exit(trans, &reflink_iter);
 
 	return ret;
 }
 
 static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
 {
-	struct bkey_s_c k = bch2_btree_iter_peek(iter);
+	struct bkey_s_c k;
 	int ret;
 
-	for_each_btree_key_continue(iter, 0, k, ret) {
-		if (bkey_cmp(iter->pos, end) >= 0)
-			return bkey_s_c_null;
+	for_each_btree_key_upto_continue_norestart(*iter, end, 0, k, ret) {
+		if (bkey_extent_is_unwritten(k))
+			continue;
 
 		if (bkey_extent_is_data(k.k))
-			break;
+			return k;
 	}
 
-	return k;
+	if (bkey_ge(iter->pos, end))
+		bch2_btree_iter_set_pos(iter, end);
+	return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
 }
 
 s64 bch2_remap_range(struct bch_fs *c,
-		     struct bpos dst_start, struct bpos src_start,
-		     u64 remap_sectors, u64 *journal_seq,
+		     subvol_inum dst_inum, u64 dst_offset,
+		     subvol_inum src_inum, u64 src_offset,
+		     u64 remap_sectors,
 		     u64 new_i_size, s64 *i_sectors_delta)
 {
-	struct btree_trans trans;
-	struct btree_iter *dst_iter, *src_iter;
+	struct btree_trans *trans;
+	struct btree_iter dst_iter, src_iter;
 	struct bkey_s_c src_k;
 	struct bkey_buf new_dst, new_src;
+	struct bpos dst_start = POS(dst_inum.inum, dst_offset);
+	struct bpos src_start = POS(src_inum.inum, src_offset);
 	struct bpos dst_end = dst_start, src_end = src_start;
-	struct bpos dst_want, src_want;
-	u64 src_done, dst_done;
+	struct bch_io_opts opts;
+	struct bpos src_want;
+	u64 dst_done = 0;
+	u32 dst_snapshot, src_snapshot;
 	int ret = 0, ret2 = 0;
 
-	if (!c->opts.reflink)
-		return -EOPNOTSUPP;
-
-	if (!percpu_ref_tryget(&c->writes))
-		return -EROFS;
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_reflink))
+		return -BCH_ERR_erofs_no_writes;
 
 	bch2_check_set_feature(c, BCH_FEATURE_reflink);
 
@@ -217,56 +269,71 @@ s64 bch2_remap_range(struct bch_fs *c,
 
 	bch2_bkey_buf_init(&new_dst);
 	bch2_bkey_buf_init(&new_src);
-	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 4096);
+	trans = bch2_trans_get(c);
+
+	ret = bch2_inum_opts_get(trans, src_inum, &opts);
+	if (ret)
+		goto err;
 
-	src_iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, src_start,
-				       BTREE_ITER_INTENT);
-	dst_iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, dst_start,
-				       BTREE_ITER_INTENT);
+	bch2_trans_iter_init(trans, &src_iter, BTREE_ID_extents, src_start,
+			     BTREE_ITER_INTENT);
+	bch2_trans_iter_init(trans, &dst_iter, BTREE_ID_extents, dst_start,
+			     BTREE_ITER_INTENT);
 
-	while (1) {
-		bch2_trans_begin(&trans);
+	while ((ret == 0 ||
+		bch2_err_matches(ret, BCH_ERR_transaction_restart)) &&
+	       bkey_lt(dst_iter.pos, dst_end)) {
+		struct disk_reservation disk_res = { 0 };
 
-		trans.mem_top = 0;
+		bch2_trans_begin(trans);
 
 		if (fatal_signal_pending(current)) {
 			ret = -EINTR;
-			goto err;
+			break;
 		}
 
-		src_k = get_next_src(src_iter, src_end);
-		ret = bkey_err(src_k);
+		ret = bch2_subvolume_get_snapshot(trans, src_inum.subvol,
+						  &src_snapshot);
 		if (ret)
-			goto btree_err;
+			continue;
 
-		src_done = bpos_min(src_iter->pos, src_end).offset -
-			src_start.offset;
-		dst_want = POS(dst_start.inode, dst_start.offset + src_done);
+		bch2_btree_iter_set_snapshot(&src_iter, src_snapshot);
 
-		if (bkey_cmp(dst_iter->pos, dst_want) < 0) {
-			ret = bch2_fpunch_at(&trans, dst_iter, dst_want,
-					     journal_seq, i_sectors_delta);
-			if (ret)
-				goto btree_err;
+		ret = bch2_subvolume_get_snapshot(trans, dst_inum.subvol,
+						  &dst_snapshot);
+		if (ret)
 			continue;
-		}
 
-		BUG_ON(bkey_cmp(dst_iter->pos, dst_want));
+		bch2_btree_iter_set_snapshot(&dst_iter, dst_snapshot);
 
-		if (!bkey_cmp(dst_iter->pos, dst_end))
-			break;
+		dst_done = dst_iter.pos.offset - dst_start.offset;
+		src_want = POS(src_start.inode, src_start.offset + dst_done);
+		bch2_btree_iter_set_pos(&src_iter, src_want);
+
+		src_k = get_next_src(&src_iter, src_end);
+		ret = bkey_err(src_k);
+		if (ret)
+			continue;
+
+		if (bkey_lt(src_want, src_iter.pos)) {
+			ret = bch2_fpunch_at(trans, &dst_iter, dst_inum,
+					min(dst_end.offset,
+					    dst_iter.pos.offset +
+					    src_iter.pos.offset - src_want.offset),
+					i_sectors_delta);
+			continue;
+		}
 
 		if (src_k.k->type != KEY_TYPE_reflink_p) {
+			bch2_btree_iter_set_pos_to_extent_start(&src_iter);
+
 			bch2_bkey_buf_reassemble(&new_src, c, src_k);
 			src_k = bkey_i_to_s_c(new_src.k);
 
-			bch2_cut_front(src_iter->pos,	new_src.k);
-			bch2_cut_back(src_end,		new_src.k);
-
-			ret = bch2_make_extent_indirect(&trans, src_iter,
+			ret = bch2_make_extent_indirect(trans, &src_iter,
 						new_src.k);
 			if (ret)
-				goto btree_err;
+				continue;
 
 			BUG_ON(src_k.k->type != KEY_TYPE_reflink_p);
 		}
@@ -278,7 +345,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 				bkey_reflink_p_init(new_dst.k);
 
 			u64 offset = le64_to_cpu(src_p.v->idx) +
-				(src_iter->pos.offset -
+				(src_want.offset -
 				 bkey_start_offset(src_k.k));
 
 			dst_p->v.idx = cpu_to_le64(offset);
@@ -286,57 +353,54 @@ s64 bch2_remap_range(struct bch_fs *c,
 			BUG();
 		}
 
-		new_dst.k->k.p = dst_iter->pos;
+		new_dst.k->k.p = dst_iter.pos;
 		bch2_key_resize(&new_dst.k->k,
-				min(src_k.k->p.offset - src_iter->pos.offset,
-				    dst_end.offset - dst_iter->pos.offset));
-
-		ret = bch2_extent_update(&trans, dst_iter, new_dst.k,
-					 NULL, journal_seq,
-					 new_i_size, i_sectors_delta);
-		if (ret)
-			goto btree_err;
-
-		dst_done = dst_iter->pos.offset - dst_start.offset;
-		src_want = POS(src_start.inode, src_start.offset + dst_done);
-		bch2_btree_iter_set_pos(src_iter, src_want);
-btree_err:
-		if (ret == -EINTR)
-			ret = 0;
-		if (ret)
-			goto err;
+				min(src_k.k->p.offset - src_want.offset,
+				    dst_end.offset - dst_iter.pos.offset));
+
+		ret =   bch2_bkey_set_needs_rebalance(c, new_dst.k,
+					opts.background_target,
+					opts.background_compression) ?:
+			bch2_extent_update(trans, dst_inum, &dst_iter,
+					new_dst.k, &disk_res,
+					new_i_size, i_sectors_delta,
+					true);
+		bch2_disk_reservation_put(c, &disk_res);
 	}
+	bch2_trans_iter_exit(trans, &dst_iter);
+	bch2_trans_iter_exit(trans, &src_iter);
 
-	BUG_ON(bkey_cmp(dst_iter->pos, dst_end));
-err:
-	BUG_ON(bkey_cmp(dst_iter->pos, dst_end) > 0);
-
-	dst_done = dst_iter->pos.offset - dst_start.offset;
-	new_i_size = min(dst_iter->pos.offset << 9, new_i_size);
+	BUG_ON(!ret && !bkey_eq(dst_iter.pos, dst_end));
+	BUG_ON(bkey_gt(dst_iter.pos, dst_end));
 
-	bch2_trans_begin(&trans);
+	dst_done = dst_iter.pos.offset - dst_start.offset;
+	new_i_size = min(dst_iter.pos.offset << 9, new_i_size);
 
 	do {
 		struct bch_inode_unpacked inode_u;
-		struct btree_iter *inode_iter;
+		struct btree_iter inode_iter = { NULL };
 
-		inode_iter = bch2_inode_peek(&trans, &inode_u,
-				dst_start.inode, BTREE_ITER_INTENT);
-		ret2 = PTR_ERR_OR_ZERO(inode_iter);
+		bch2_trans_begin(trans);
+
+		ret2 = bch2_inode_peek(trans, &inode_iter, &inode_u,
+				       dst_inum, BTREE_ITER_INTENT);
 
 		if (!ret2 &&
 		    inode_u.bi_size < new_i_size) {
 			inode_u.bi_size = new_i_size;
-			ret2  = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
-				bch2_trans_commit(&trans, NULL, journal_seq, 0);
+			ret2  = bch2_inode_write(trans, &inode_iter, &inode_u) ?:
+				bch2_trans_commit(trans, NULL, NULL,
+						  BTREE_INSERT_NOFAIL);
 		}
-	} while (ret2 == -EINTR);
 
-	ret = bch2_trans_exit(&trans) ?: ret;
+		bch2_trans_iter_exit(trans, &inode_iter);
+	} while (bch2_err_matches(ret2, BCH_ERR_transaction_restart));
+err:
+	bch2_trans_put(trans);
 	bch2_bkey_buf_exit(&new_src, c);
 	bch2_bkey_buf_exit(&new_dst, c);
 
-	percpu_ref_put(&c->writes);
+	bch2_write_ref_put(c, BCH_WRITE_REF_reflink);
 
 	return dst_done ?: ret ?: ret2;
 }