[thirdparty/kernel/stable.git] / fs / bcachefs / reflink.c

// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
#include "bkey_buf.h"
#include "btree_update.h"
#include "buckets.h"
#include "extents.h"
#include "inode.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>

static inline unsigned bkey_type_to_indirect(const struct bkey *k)
{
	switch (k->type) {
	case KEY_TYPE_extent:
		return KEY_TYPE_reflink_v;
	case KEY_TYPE_inline_data:
		return KEY_TYPE_indirect_inline_data;
	default:
		return 0;
	}
}

/* reflink pointers */

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 (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 0;
}

void bch2_reflink_p_to_text(struct printbuf *out, struct bch_fs *c,
			    struct bkey_s_c k)
{
	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);

	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));
}

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_c_reflink_p r = bkey_s_c_to_reflink_p(_r);

	/*
	 * Disabled for now, the triggers code needs to be reworked for merging
	 * of reflink pointers to work:
	 */
	return false;

	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 true;
}

/* indirect extents */

int bch2_reflink_v_invalid(struct bch_fs *c, struct bkey_s_c k,
			   enum bkey_invalid_flags flags,
			   struct printbuf *err)
{
	return bch2_bkey_ptrs_invalid(c, k, flags, err);
}

void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
			    struct bkey_s_c k)
{
	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);

	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);

	if (old.k->type == KEY_TYPE_reflink_v &&
	    new->k.type == KEY_TYPE_reflink_v &&
	    old.k->u64s == new->k.u64s &&
	    !memcmp(bkey_s_c_to_reflink_v(old).v->start,
		    bkey_i_to_reflink_v(new)->v.start,
		    bkey_val_bytes(&new->k) - 8))
		return 0;

	return bch2_trans_mark_extent(trans, btree_id, level, old, new, flags);
}

/* indirect inline data */

int bch2_indirect_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k,
				      enum bkey_invalid_flags flags,
				      struct printbuf *err)
{
	return 0;
}

void bch2_indirect_inline_data_to_text(struct printbuf *out,
				       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);

	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 = { NULL };
	struct bkey_s_c k;
	struct bkey_i *r_v;
	struct bkey_i_reflink_p *r_p;
	__le64 *refcount;
	int ret;

	if (orig->k.type == KEY_TYPE_inline_data)
		bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data);

	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;

	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;
	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	= bkey_refcount(r_v);
	*refcount	= 0;
	memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));

	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));

	ret = bch2_trans_update(trans, extent_iter, &r_p->k_i,
				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
err:
	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;
	int ret;

	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))
			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,
		     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 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 bch_io_opts opts;
	struct bpos src_want;
	u64 dst_done = 0;
	u32 dst_snapshot, src_snapshot;
	int ret = 0, ret2 = 0;

	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_reflink))
		return -BCH_ERR_erofs_no_writes;

	bch2_check_set_feature(c, BCH_FEATURE_reflink);

	dst_end.offset += remap_sectors;
	src_end.offset += remap_sectors;

	bch2_bkey_buf_init(&new_dst);
	bch2_bkey_buf_init(&new_src);
	trans = bch2_trans_get(c);

	ret = bch2_inum_opts_get(trans, src_inum, &opts);
	if (ret)
		goto err;

	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 ((ret == 0 ||
		bch2_err_matches(ret, BCH_ERR_transaction_restart)) &&
	       bkey_lt(dst_iter.pos, dst_end)) {
		struct disk_reservation disk_res = { 0 };

		bch2_trans_begin(trans);

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

		ret = bch2_subvolume_get_snapshot(trans, src_inum.subvol,
						  &src_snapshot);
		if (ret)
			continue;

		bch2_btree_iter_set_snapshot(&src_iter, src_snapshot);

		ret = bch2_subvolume_get_snapshot(trans, dst_inum.subvol,
						  &dst_snapshot);
		if (ret)
			continue;

		bch2_btree_iter_set_snapshot(&dst_iter, dst_snapshot);

		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);

			ret = bch2_make_extent_indirect(trans, &src_iter,
						new_src.k);
			if (ret)
				continue;

			BUG_ON(src_k.k->type != KEY_TYPE_reflink_p);
		}

		if (src_k.k->type == KEY_TYPE_reflink_p) {
			struct bkey_s_c_reflink_p src_p =
				bkey_s_c_to_reflink_p(src_k);
			struct bkey_i_reflink_p *dst_p =
				bkey_reflink_p_init(new_dst.k);

			u64 offset = le64_to_cpu(src_p.v->idx) +
				(src_want.offset -
				 bkey_start_offset(src_k.k));

			dst_p->v.idx = cpu_to_le64(offset);
		} else {
			BUG();
		}

		new_dst.k->k.p = dst_iter.pos;
		bch2_key_resize(&new_dst.k->k,
				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(!ret && !bkey_eq(dst_iter.pos, dst_end));
	BUG_ON(bkey_gt(dst_iter.pos, dst_end));

	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 = { NULL };

		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, NULL,
						  BTREE_INSERT_NOFAIL);
		}

		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);

	bch2_write_ref_put(c, BCH_WRITE_REF_reflink);

	return dst_done ?: ret ?: ret2;
}
Commit	Line	Data
76426098 KO	1	// SPDX-License-Identifier: GPL-2.0
76426098 KO	2	#include "bcachefs.h"
07a1006a	3	#include "bkey_buf.h"
76426098	4	#include "btree_update.h"
2e8f9d23	5	#include "buckets.h"
76426098	6	#include "extents.h"
2e87eae1	7	#include "inode.h"
1809b8cb KO	8	#include "io_misc.h"
1809b8cb KO	9	#include "io_write.h"
fb3f57bb	10	#include "rebalance.h"
76426098	11	#include "reflink.h"
6fed42bb	12	#include "subvolume.h"
1809b8cb	13	#include "super-io.h"
76426098 KO	14
	15	#include <linux/sched/signal.h>
	16
801a3de6 KO	17	static inline unsigned bkey_type_to_indirect(const struct bkey *k)
	18	{
	19	switch (k->type) {
	20	case KEY_TYPE_extent:
	21	return KEY_TYPE_reflink_v;
	22	case KEY_TYPE_inline_data:
	23	return KEY_TYPE_indirect_inline_data;
	24	default:
	25	return 0;
	26	}
	27	}
	28
76426098 KO	29	/* reflink pointers */
76426098 KO	30
b65db750	31	int bch2_reflink_p_invalid(struct bch_fs *c, struct bkey_s_c k,
8726dc93 KO	32	enum bkey_invalid_flags flags,
8726dc93 KO	33	struct printbuf *err)
76426098 KO	34	{
	35	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
	36
6d76aefe	37	if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix &&
f0ac7df2	38	le64_to_cpu(p.v->idx) < le32_to_cpu(p.v->front_pad)) {
401ec4db	39	prt_printf(err, "idx < front_pad (%llu < %u)",
f0ac7df2 KO	40	le64_to_cpu(p.v->idx), le32_to_cpu(p.v->front_pad));
	41	return -EINVAL;
	42	}
6d76aefe	43
f0ac7df2	44	return 0;
76426098 KO	45	}
	46
	47	void bch2_reflink_p_to_text(struct printbuf out, struct bch_fs c,
	48	struct bkey_s_c k)
	49	{
	50	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
	51
401ec4db	52	prt_printf(out, "idx %llu front_pad %u back_pad %u",
61d876c2 KO	53	le64_to_cpu(p.v->idx),
	54	le32_to_cpu(p.v->front_pad),
	55	le32_to_cpu(p.v->back_pad));
76426098 KO	56	}
76426098 KO	57
59ba21d9	58	bool bch2_reflink_p_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
76426098 KO	59	{
76426098 KO	60	struct bkey_s_reflink_p l = bkey_s_to_reflink_p(_l);
59ba21d9	61	struct bkey_s_c_reflink_p r = bkey_s_c_to_reflink_p(_r);
76426098	62
59ba21d9 KO	63	/*
	64	* Disabled for now, the triggers code needs to be reworked for merging
	65	* of reflink pointers to work:
	66	*/
	67	return false;
76426098	68
59ba21d9 KO	69	if (le64_to_cpu(l.v->idx) + l.k->size != le64_to_cpu(r.v->idx))
59ba21d9 KO	70	return false;
76426098 KO	71
76426098 KO	72	bch2_key_resize(l.k, l.k->size + r.k->size);
59ba21d9	73	return true;
76426098 KO	74	}
	75
	76	/* indirect extents */
	77
b65db750	78	int bch2_reflink_v_invalid(struct bch_fs *c, struct bkey_s_c k,
8726dc93 KO	79	enum bkey_invalid_flags flags,
8726dc93 KO	80	struct printbuf *err)
76426098	81	{
facafdcb	82	return bch2_bkey_ptrs_invalid(c, k, flags, err);
76426098 KO	83	}
	84
	85	void bch2_reflink_v_to_text(struct printbuf out, struct bch_fs c,
	86	struct bkey_s_c k)
	87	{
	88	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);
	89
401ec4db	90	prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
76426098 KO	91
	92	bch2_bkey_ptrs_to_text(out, c, k);
	93	}
	94
96dea3d5 KO	95	#if 0
	96	Currently disabled, needs to be debugged:
	97
b058ac20 KO	98	bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
	99	{
	100	struct bkey_s_reflink_v l = bkey_s_to_reflink_v(_l);
	101	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(_r);
	102
	103	return l.v->refcount == r.v->refcount && bch2_extent_merge(c, _l, _r);
	104	}
96dea3d5	105	#endif
b058ac20	106
523f33ef KO	107	static inline void check_indirect_extent_deleting(struct bkey_i new, unsigned flags)
	108	{
	109	if ((flags & BTREE_TRIGGER_INSERT) && !bkey_refcount(new)) {
	110	new->k.type = KEY_TYPE_deleted;
	111	new->k.size = 0;
	112	set_bkey_val_u64s(&new->k, 0);;
	113	*flags &= ~BTREE_TRIGGER_INSERT;
	114	}
	115	}
	116
78668fe0	117	int bch2_trans_mark_reflink_v(struct btree_trans *trans,
e1b8f5f5	118	enum btree_id btree_id, unsigned level,
78668fe0 KO	119	struct bkey_s_c old, struct bkey_i *new,
	120	unsigned flags)
	121	{
523f33ef	122	check_indirect_extent_deleting(new, &flags);
78668fe0	123
f88d811a KO	124	if (old.k->type == KEY_TYPE_reflink_v &&
	125	new->k.type == KEY_TYPE_reflink_v &&
	126	old.k->u64s == new->k.u64s &&
	127	!memcmp(bkey_s_c_to_reflink_v(old).v->start,
	128	bkey_i_to_reflink_v(new)->v.start,
	129	bkey_val_bytes(&new->k) - 8))
	130	return 0;
	131
e1b8f5f5	132	return bch2_trans_mark_extent(trans, btree_id, level, old, new, flags);
78668fe0 KO	133	}
78668fe0 KO	134
801a3de6 KO	135	/* indirect inline data */
801a3de6 KO	136
b65db750	137	int bch2_indirect_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k,
8726dc93 KO	138	enum bkey_invalid_flags flags,
8726dc93 KO	139	struct printbuf *err)
801a3de6	140	{
f0ac7df2	141	return 0;
801a3de6 KO	142	}
	143
	144	void bch2_indirect_inline_data_to_text(struct printbuf *out,
523f33ef	145	struct bch_fs *c, struct bkey_s_c k)
801a3de6 KO	146	{
	147	struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k);
	148	unsigned datalen = bkey_inline_data_bytes(k.k);
	149
401ec4db	150	prt_printf(out, "refcount %llu datalen %u: %*phN",
801a3de6 KO	151	le64_to_cpu(d.v->refcount), datalen,
	152	min(datalen, 32U), d.v->data);
	153	}
	154
78668fe0	155	int bch2_trans_mark_indirect_inline_data(struct btree_trans *trans,
e1b8f5f5	156	enum btree_id btree_id, unsigned level,
78668fe0 KO	157	struct bkey_s_c old, struct bkey_i *new,
	158	unsigned flags)
	159	{
523f33ef	160	check_indirect_extent_deleting(new, &flags);
78668fe0 KO	161
	162	return 0;
	163	}
	164
76426098 KO	165	static int bch2_make_extent_indirect(struct btree_trans *trans,
76426098 KO	166	struct btree_iter *extent_iter,
801a3de6	167	struct bkey_i *orig)
76426098 KO	168	{
76426098 KO	169	struct bch_fs *c = trans->c;
67e0dd8f	170	struct btree_iter reflink_iter = { NULL };
76426098	171	struct bkey_s_c k;
801a3de6	172	struct bkey_i *r_v;
76426098	173	struct bkey_i_reflink_p *r_p;
801a3de6	174	__le64 *refcount;
76426098 KO	175	int ret;
76426098 KO	176
801a3de6 KO	177	if (orig->k.type == KEY_TYPE_inline_data)
	178	bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data);
	179
fc0ee376 KO	180	bch2_trans_iter_init(trans, &reflink_iter, BTREE_ID_reflink, POS_MAX,
	181	BTREE_ITER_INTENT);
	182	k = bch2_btree_iter_peek_prev(&reflink_iter);
	183	ret = bkey_err(k);
76426098 KO	184	if (ret)
	185	goto err;
	186
74cc1abd	187	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k));
76426098 KO	188	ret = PTR_ERR_OR_ZERO(r_v);
	189	if (ret)
	190	goto err;
	191
801a3de6 KO	192	bkey_init(&r_v->k);
801a3de6 KO	193	r_v->k.type = bkey_type_to_indirect(&orig->k);
67e0dd8f	194	r_v->k.p = reflink_iter.pos;
801a3de6 KO	195	bch2_key_resize(&r_v->k, orig->k.size);
	196	r_v->k.version = orig->k.version;
	197
	198	set_bkey_val_bytes(&r_v->k, sizeof(__le64) + bkey_val_bytes(&orig->k));
76426098	199
890b74f0	200	refcount = bkey_refcount(r_v);
801a3de6 KO	201	*refcount = 0;
801a3de6 KO	202	memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));
76426098	203
67e0dd8f	204	ret = bch2_trans_update(trans, &reflink_iter, r_v, 0);
d6462f49 KO	205	if (ret)
d6462f49 KO	206	goto err;
76426098	207
70d61a70 KO	208	/*
	209	* orig is in a bkey_buf which statically allocates 5 64s for the val,
	210	* so we know it will be big enough:
	211	*/
801a3de6 KO	212	orig->k.type = KEY_TYPE_reflink_p;
801a3de6 KO	213	r_p = bkey_i_to_reflink_p(orig);
76426098	214	set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
ac2ccddc KO	215
	216	/* FORTIFY_SOURCE is broken here, and doesn't provide unsafe_memset() */
	217	#if !defined(__NO_FORTIFY) && defined(__OPTIMIZE__) && defined(CONFIG_FORTIFY_SOURCE)
	218	__underlying_memset(&r_p->v, 0, sizeof(r_p->v));
	219	#else
70d61a70	220	memset(&r_p->v, 0, sizeof(r_p->v));
ac2ccddc	221	#endif
70d61a70	222
76426098 KO	223	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
76426098 KO	224
f449bedb KO	225	ret = bch2_trans_update(trans, extent_iter, &r_p->k_i,
f449bedb KO	226	BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
76426098	227	err:
67e0dd8f	228	bch2_trans_iter_exit(trans, &reflink_iter);
76426098 KO	229
	230	return ret;
	231	}
	232
	233	static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
	234	{
649d9a4d	235	struct bkey_s_c k;
63fbf458	236	int ret;
76426098	237
79203111 KO	238	for_each_btree_key_upto_continue_norestart(*iter, end, 0, k, ret) {
	239	if (bkey_extent_is_unwritten(k))
	240	continue;
	241
801a3de6	242	if (bkey_extent_is_data(k.k))
649d9a4d	243	return k;
79203111	244	}
63fbf458	245
e88a75eb	246	if (bkey_ge(iter->pos, end))
8b3e9bd6 KO	247	bch2_btree_iter_set_pos(iter, end);
8b3e9bd6 KO	248	return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
76426098 KO	249	}
	250
	251	s64 bch2_remap_range(struct bch_fs *c,
6fed42bb KO	252	subvol_inum dst_inum, u64 dst_offset,
6fed42bb KO	253	subvol_inum src_inum, u64 src_offset,
68a2054d	254	u64 remap_sectors,
2e87eae1	255	u64 new_i_size, s64 *i_sectors_delta)
76426098	256	{
6bd68ec2	257	struct btree_trans *trans;
67e0dd8f	258	struct btree_iter dst_iter, src_iter;
76426098	259	struct bkey_s_c src_k;
07a1006a	260	struct bkey_buf new_dst, new_src;
6fed42bb KO	261	struct bpos dst_start = POS(dst_inum.inum, dst_offset);
6fed42bb KO	262	struct bpos src_start = POS(src_inum.inum, src_offset);
76426098	263	struct bpos dst_end = dst_start, src_end = src_start;
fb3f57bb	264	struct bch_io_opts opts;
649d9a4d	265	struct bpos src_want;
fb3f57bb	266	u64 dst_done = 0;
8c6d298a	267	u32 dst_snapshot, src_snapshot;
2e87eae1	268	int ret = 0, ret2 = 0;
76426098	269
d94189ad	270	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_reflink))
858536c7	271	return -BCH_ERR_erofs_no_writes;
2a9101a9	272
1c3ff72c	273	bch2_check_set_feature(c, BCH_FEATURE_reflink);
76426098 KO	274
	275	dst_end.offset += remap_sectors;
	276	src_end.offset += remap_sectors;
	277
07a1006a KO	278	bch2_bkey_buf_init(&new_dst);
07a1006a KO	279	bch2_bkey_buf_init(&new_src);
6bd68ec2	280	trans = bch2_trans_get(c);
76426098	281
fb3f57bb KO	282	ret = bch2_inum_opts_get(trans, src_inum, &opts);
	283	if (ret)
	284	goto err;
	285
6bd68ec2	286	bch2_trans_iter_init(trans, &src_iter, BTREE_ID_extents, src_start,
67e0dd8f	287	BTREE_ITER_INTENT);
6bd68ec2	288	bch2_trans_iter_init(trans, &dst_iter, BTREE_ID_extents, dst_start,
67e0dd8f	289	BTREE_ITER_INTENT);
76426098	290
549d173c KO	291	while ((ret == 0 \|\|
549d173c KO	292	bch2_err_matches(ret, BCH_ERR_transaction_restart)) &&
e88a75eb	293	bkey_lt(dst_iter.pos, dst_end)) {
2e8f9d23 KO	294	struct disk_reservation disk_res = { 0 };
2e8f9d23 KO	295
6bd68ec2	296	bch2_trans_begin(trans);
a8abd3a7	297
76426098 KO	298	if (fatal_signal_pending(current)) {
76426098 KO	299	ret = -EINTR;
50dc0f69	300	break;
76426098 KO	301	}
76426098 KO	302
6bd68ec2	303	ret = bch2_subvolume_get_snapshot(trans, src_inum.subvol,
8c6d298a	304	&src_snapshot);
6fed42bb KO	305	if (ret)
	306	continue;
	307
8c6d298a KO	308	bch2_btree_iter_set_snapshot(&src_iter, src_snapshot);
8c6d298a KO	309
6bd68ec2	310	ret = bch2_subvolume_get_snapshot(trans, dst_inum.subvol,
8c6d298a	311	&dst_snapshot);
6fed42bb KO	312	if (ret)
	313	continue;
	314
8c6d298a KO	315	bch2_btree_iter_set_snapshot(&dst_iter, dst_snapshot);
8c6d298a KO	316
67e0dd8f	317	dst_done = dst_iter.pos.offset - dst_start.offset;
649d9a4d	318	src_want = POS(src_start.inode, src_start.offset + dst_done);
67e0dd8f	319	bch2_btree_iter_set_pos(&src_iter, src_want);
649d9a4d	320
67e0dd8f	321	src_k = get_next_src(&src_iter, src_end);
76426098 KO	322	ret = bkey_err(src_k);
76426098 KO	323	if (ret)
50dc0f69	324	continue;
76426098	325
e88a75eb	326	if (bkey_lt(src_want, src_iter.pos)) {
6bd68ec2	327	ret = bch2_fpunch_at(trans, &dst_iter, dst_inum,
8c6d298a KO	328	min(dst_end.offset,
	329	dst_iter.pos.offset +
	330	src_iter.pos.offset - src_want.offset),
68a2054d	331	i_sectors_delta);
76426098 KO	332	continue;
	333	}
	334
801a3de6	335	if (src_k.k->type != KEY_TYPE_reflink_p) {
67e0dd8f	336	bch2_btree_iter_set_pos_to_extent_start(&src_iter);
8c3f6da9	337
07a1006a	338	bch2_bkey_buf_reassemble(&new_src, c, src_k);
35189e09	339	src_k = bkey_i_to_s_c(new_src.k);
76426098	340
6bd68ec2	341	ret = bch2_make_extent_indirect(trans, &src_iter,
801a3de6	342	new_src.k);
76426098	343	if (ret)
50dc0f69	344	continue;
76426098 KO	345
	346	BUG_ON(src_k.k->type != KEY_TYPE_reflink_p);
	347	}
	348
	349	if (src_k.k->type == KEY_TYPE_reflink_p) {
	350	struct bkey_s_c_reflink_p src_p =
	351	bkey_s_c_to_reflink_p(src_k);
	352	struct bkey_i_reflink_p *dst_p =
07a1006a	353	bkey_reflink_p_init(new_dst.k);
76426098 KO	354
76426098 KO	355	u64 offset = le64_to_cpu(src_p.v->idx) +
649d9a4d	356	(src_want.offset -
76426098 KO	357	bkey_start_offset(src_k.k));
	358
	359	dst_p->v.idx = cpu_to_le64(offset);
	360	} else {
	361	BUG();
	362	}
	363
67e0dd8f	364	new_dst.k->k.p = dst_iter.pos;
07a1006a	365	bch2_key_resize(&new_dst.k->k,
649d9a4d	366	min(src_k.k->p.offset - src_want.offset,
67e0dd8f	367	dst_end.offset - dst_iter.pos.offset));
8c6d298a	368
fb3f57bb KO	369	ret = bch2_bkey_set_needs_rebalance(c, new_dst.k,
	370	opts.background_target,
	371	opts.background_compression) ?:
	372	bch2_extent_update(trans, dst_inum, &dst_iter,
	373	new_dst.k, &disk_res,
	374	new_i_size, i_sectors_delta,
	375	true);
2e8f9d23	376	bch2_disk_reservation_put(c, &disk_res);
76426098	377	}
6bd68ec2 KO	378	bch2_trans_iter_exit(trans, &dst_iter);
6bd68ec2 KO	379	bch2_trans_iter_exit(trans, &src_iter);
76426098	380
e88a75eb KO	381	BUG_ON(!ret && !bkey_eq(dst_iter.pos, dst_end));
e88a75eb KO	382	BUG_ON(bkey_gt(dst_iter.pos, dst_end));
76426098	383
67e0dd8f KO	384	dst_done = dst_iter.pos.offset - dst_start.offset;
67e0dd8f KO	385	new_i_size = min(dst_iter.pos.offset << 9, new_i_size);
76426098	386
2e87eae1 KO	387	do {
2e87eae1 KO	388	struct bch_inode_unpacked inode_u;
67e0dd8f	389	struct btree_iter inode_iter = { NULL };
76426098	390
6bd68ec2	391	bch2_trans_begin(trans);
700c25b3	392
6bd68ec2	393	ret2 = bch2_inode_peek(trans, &inode_iter, &inode_u,
6fed42bb	394	dst_inum, BTREE_ITER_INTENT);
76426098	395
2e87eae1	396	if (!ret2 &&
677fc056 KO	397	inode_u.bi_size < new_i_size) {
677fc056 KO	398	inode_u.bi_size = new_i_size;
6bd68ec2 KO	399	ret2 = bch2_inode_write(trans, &inode_iter, &inode_u) ?:
6bd68ec2 KO	400	bch2_trans_commit(trans, NULL, NULL,
770e8214	401	BTREE_INSERT_NOFAIL);
677fc056	402	}
50dc0f69	403
6bd68ec2	404	bch2_trans_iter_exit(trans, &inode_iter);
549d173c	405	} while (bch2_err_matches(ret2, BCH_ERR_transaction_restart));
fb3f57bb	406	err:
6bd68ec2	407	bch2_trans_put(trans);
07a1006a KO	408	bch2_bkey_buf_exit(&new_src, c);
07a1006a KO	409	bch2_bkey_buf_exit(&new_dst, c);
2e87eae1	410
d94189ad	411	bch2_write_ref_put(c, BCH_WRITE_REF_reflink);
2a9101a9	412
2e87eae1 KO	413	return dst_done ?: ret ?: ret2;
2e87eae1 KO	414	}