2019-08-16 21:59:56 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
#include "bcachefs.h"
|
2020-12-18 04:08:58 +08:00
|
|
|
#include "bkey_buf.h"
|
2019-08-16 21:59:56 +08:00
|
|
|
#include "btree_update.h"
|
2021-05-17 12:08:06 +08:00
|
|
|
#include "buckets.h"
|
2019-08-16 21:59:56 +08:00
|
|
|
#include "extents.h"
|
2019-10-11 00:47:22 +08:00
|
|
|
#include "inode.h"
|
2023-09-11 06:05:17 +08:00
|
|
|
#include "io_misc.h"
|
|
|
|
#include "io_write.h"
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
#include "rebalance.h"
|
2019-08-16 21:59:56 +08:00
|
|
|
#include "reflink.h"
|
2021-03-16 12:28:17 +08:00
|
|
|
#include "subvolume.h"
|
2023-09-11 06:05:17 +08:00
|
|
|
#include "super-io.h"
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
#include <linux/sched/signal.h>
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-08-16 21:59:56 +08:00
|
|
|
/* reflink pointers */
|
|
|
|
|
2023-10-25 08:44:36 +08:00
|
|
|
int bch2_reflink_p_invalid(struct bch_fs *c, struct bkey_s_c k,
|
2023-07-07 09:16:10 +08:00
|
|
|
enum bkey_invalid_flags flags,
|
|
|
|
struct printbuf *err)
|
2019-08-16 21:59:56 +08:00
|
|
|
{
|
|
|
|
struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
|
|
|
|
|
2021-10-14 21:54:47 +08:00
|
|
|
if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix &&
|
2022-04-04 05:50:01 +08:00
|
|
|
le64_to_cpu(p.v->idx) < le32_to_cpu(p.v->front_pad)) {
|
2023-02-04 10:01:40 +08:00
|
|
|
prt_printf(err, "idx < front_pad (%llu < %u)",
|
2022-04-04 05:50:01 +08:00
|
|
|
le64_to_cpu(p.v->idx), le32_to_cpu(p.v->front_pad));
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2021-10-14 21:54:47 +08:00
|
|
|
|
2022-04-04 05:50:01 +08:00
|
|
|
return 0;
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
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);
|
|
|
|
|
2023-02-04 10:01:40 +08:00
|
|
|
prt_printf(out, "idx %llu front_pad %u back_pad %u",
|
2021-11-14 06:44:13 +08:00
|
|
|
le64_to_cpu(p.v->idx),
|
|
|
|
le32_to_cpu(p.v->front_pad),
|
|
|
|
le32_to_cpu(p.v->back_pad));
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
2021-04-29 11:49:30 +08:00
|
|
|
bool bch2_reflink_p_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
|
2019-08-16 21:59:56 +08:00
|
|
|
{
|
|
|
|
struct bkey_s_reflink_p l = bkey_s_to_reflink_p(_l);
|
2021-04-29 11:49:30 +08:00
|
|
|
struct bkey_s_c_reflink_p r = bkey_s_c_to_reflink_p(_r);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-04-29 11:49:30 +08:00
|
|
|
/*
|
|
|
|
* Disabled for now, the triggers code needs to be reworked for merging
|
|
|
|
* of reflink pointers to work:
|
|
|
|
*/
|
|
|
|
return false;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-04-29 11:49:30 +08:00
|
|
|
if (le64_to_cpu(l.v->idx) + l.k->size != le64_to_cpu(r.v->idx))
|
|
|
|
return false;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
bch2_key_resize(l.k, l.k->size + r.k->size);
|
2021-04-29 11:49:30 +08:00
|
|
|
return true;
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* indirect extents */
|
|
|
|
|
2023-10-25 08:44:36 +08:00
|
|
|
int bch2_reflink_v_invalid(struct bch_fs *c, struct bkey_s_c k,
|
2023-07-07 09:16:10 +08:00
|
|
|
enum bkey_invalid_flags flags,
|
|
|
|
struct printbuf *err)
|
2019-08-16 21:59:56 +08:00
|
|
|
{
|
2022-12-21 08:58:16 +08:00
|
|
|
return bch2_bkey_ptrs_invalid(c, k, flags, err);
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
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);
|
|
|
|
|
2023-02-04 10:01:40 +08:00
|
|
|
prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
bch2_bkey_ptrs_to_text(out, c, k);
|
|
|
|
}
|
|
|
|
|
2023-09-13 06:41:22 +08:00
|
|
|
#if 0
|
|
|
|
Currently disabled, needs to be debugged:
|
|
|
|
|
2021-05-16 03:04:08 +08:00
|
|
|
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);
|
|
|
|
}
|
2023-09-13 06:41:22 +08:00
|
|
|
#endif
|
2021-05-16 03:04:08 +08:00
|
|
|
|
2023-06-23 08:18:12 +08:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-03-31 12:03:37 +08:00
|
|
|
int bch2_trans_mark_reflink_v(struct btree_trans *trans,
|
2022-04-01 09:44:55 +08:00
|
|
|
enum btree_id btree_id, unsigned level,
|
2022-03-31 12:03:37 +08:00
|
|
|
struct bkey_s_c old, struct bkey_i *new,
|
|
|
|
unsigned flags)
|
|
|
|
{
|
2023-06-23 08:18:12 +08:00
|
|
|
check_indirect_extent_deleting(new, &flags);
|
2022-03-31 12:03:37 +08:00
|
|
|
|
2023-12-04 02:05:21 +08:00
|
|
|
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;
|
|
|
|
|
2022-04-01 09:44:55 +08:00
|
|
|
return bch2_trans_mark_extent(trans, btree_id, level, old, new, flags);
|
2022-03-31 12:03:37 +08:00
|
|
|
}
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
/* indirect inline data */
|
|
|
|
|
2023-10-25 08:44:36 +08:00
|
|
|
int bch2_indirect_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k,
|
2023-07-07 09:16:10 +08:00
|
|
|
enum bkey_invalid_flags flags,
|
|
|
|
struct printbuf *err)
|
2020-10-25 07:51:34 +08:00
|
|
|
{
|
2022-04-04 05:50:01 +08:00
|
|
|
return 0;
|
2020-10-25 07:51:34 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void bch2_indirect_inline_data_to_text(struct printbuf *out,
|
2023-06-23 08:18:12 +08:00
|
|
|
struct bch_fs *c, struct bkey_s_c k)
|
2020-10-25 07:51:34 +08:00
|
|
|
{
|
|
|
|
struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k);
|
|
|
|
unsigned datalen = bkey_inline_data_bytes(k.k);
|
|
|
|
|
2023-02-04 10:01:40 +08:00
|
|
|
prt_printf(out, "refcount %llu datalen %u: %*phN",
|
2020-10-25 07:51:34 +08:00
|
|
|
le64_to_cpu(d.v->refcount), datalen,
|
|
|
|
min(datalen, 32U), d.v->data);
|
|
|
|
}
|
|
|
|
|
2022-03-31 12:03:37 +08:00
|
|
|
int bch2_trans_mark_indirect_inline_data(struct btree_trans *trans,
|
2022-04-01 09:44:55 +08:00
|
|
|
enum btree_id btree_id, unsigned level,
|
2022-03-31 12:03:37 +08:00
|
|
|
struct bkey_s_c old, struct bkey_i *new,
|
|
|
|
unsigned flags)
|
|
|
|
{
|
2023-06-23 08:18:12 +08:00
|
|
|
check_indirect_extent_deleting(new, &flags);
|
2022-03-31 12:03:37 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-08-16 21:59:56 +08:00
|
|
|
static int bch2_make_extent_indirect(struct btree_trans *trans,
|
|
|
|
struct btree_iter *extent_iter,
|
2020-10-25 07:51:34 +08:00
|
|
|
struct bkey_i *orig)
|
2019-08-16 21:59:56 +08:00
|
|
|
{
|
|
|
|
struct bch_fs *c = trans->c;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_iter reflink_iter = { NULL };
|
2019-08-16 21:59:56 +08:00
|
|
|
struct bkey_s_c k;
|
2020-10-25 07:51:34 +08:00
|
|
|
struct bkey_i *r_v;
|
2019-08-16 21:59:56 +08:00
|
|
|
struct bkey_i_reflink_p *r_p;
|
2020-10-25 07:51:34 +08:00
|
|
|
__le64 *refcount;
|
2019-08-16 21:59:56 +08:00
|
|
|
int ret;
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
if (orig->k.type == KEY_TYPE_inline_data)
|
|
|
|
bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data);
|
|
|
|
|
2023-05-26 11:37:06 +08:00
|
|
|
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);
|
2019-08-16 21:59:56 +08:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
2021-06-11 01:21:39 +08:00
|
|
|
r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k));
|
2019-08-16 21:59:56 +08:00
|
|
|
ret = PTR_ERR_OR_ZERO(r_v);
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
bkey_init(&r_v->k);
|
|
|
|
r_v->k.type = bkey_type_to_indirect(&orig->k);
|
2021-08-31 03:18:31 +08:00
|
|
|
r_v->k.p = reflink_iter.pos;
|
2020-10-25 07:51:34 +08:00
|
|
|
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));
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-05-23 14:31:33 +08:00
|
|
|
refcount = bkey_refcount(r_v);
|
2020-10-25 07:51:34 +08:00
|
|
|
*refcount = 0;
|
|
|
|
memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
ret = bch2_trans_update(trans, &reflink_iter, r_v, 0);
|
2021-05-19 11:17:03 +08:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-10-18 23:32:06 +08:00
|
|
|
/*
|
|
|
|
* orig is in a bkey_buf which statically allocates 5 64s for the val,
|
|
|
|
* so we know it will be big enough:
|
|
|
|
*/
|
2020-10-25 07:51:34 +08:00
|
|
|
orig->k.type = KEY_TYPE_reflink_p;
|
|
|
|
r_p = bkey_i_to_reflink_p(orig);
|
2019-08-16 21:59:56 +08:00
|
|
|
set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
|
2023-03-05 12:05:55 +08:00
|
|
|
|
|
|
|
/* 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
|
2021-10-18 23:32:06 +08:00
|
|
|
memset(&r_p->v, 0, sizeof(r_p->v));
|
2023-03-05 12:05:55 +08:00
|
|
|
#endif
|
2021-10-18 23:32:06 +08:00
|
|
|
|
2019-08-16 21:59:56 +08:00
|
|
|
r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
|
|
|
|
|
2021-11-30 05:36:50 +08:00
|
|
|
ret = bch2_trans_update(trans, extent_iter, &r_p->k_i,
|
|
|
|
BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
|
2019-08-16 21:59:56 +08:00
|
|
|
err:
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_trans_iter_exit(trans, &reflink_iter);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
|
|
|
|
{
|
2021-03-21 10:14:10 +08:00
|
|
|
struct bkey_s_c k;
|
2019-10-05 08:40:47 +08:00
|
|
|
int ret;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2022-11-14 07:59:01 +08:00
|
|
|
for_each_btree_key_upto_continue_norestart(*iter, end, 0, k, ret) {
|
|
|
|
if (bkey_extent_is_unwritten(k))
|
|
|
|
continue;
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
if (bkey_extent_is_data(k.k))
|
2021-03-21 10:14:10 +08:00
|
|
|
return k;
|
2022-11-14 07:59:01 +08:00
|
|
|
}
|
2019-10-05 08:40:47 +08:00
|
|
|
|
2022-11-24 16:12:22 +08:00
|
|
|
if (bkey_ge(iter->pos, end))
|
2021-07-25 07:50:40 +08:00
|
|
|
bch2_btree_iter_set_pos(iter, end);
|
|
|
|
return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
s64 bch2_remap_range(struct bch_fs *c,
|
2021-03-16 12:28:17 +08:00
|
|
|
subvol_inum dst_inum, u64 dst_offset,
|
|
|
|
subvol_inum src_inum, u64 src_offset,
|
2021-11-06 03:17:13 +08:00
|
|
|
u64 remap_sectors,
|
2019-10-11 00:47:22 +08:00
|
|
|
u64 new_i_size, s64 *i_sectors_delta)
|
2019-08-16 21:59:56 +08:00
|
|
|
{
|
2023-09-13 05:16:02 +08:00
|
|
|
struct btree_trans *trans;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_iter dst_iter, src_iter;
|
2019-08-16 21:59:56 +08:00
|
|
|
struct bkey_s_c src_k;
|
2020-12-18 04:08:58 +08:00
|
|
|
struct bkey_buf new_dst, new_src;
|
2021-03-16 12:28:17 +08:00
|
|
|
struct bpos dst_start = POS(dst_inum.inum, dst_offset);
|
|
|
|
struct bpos src_start = POS(src_inum.inum, src_offset);
|
2019-08-16 21:59:56 +08:00
|
|
|
struct bpos dst_end = dst_start, src_end = src_start;
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
struct bch_io_opts opts;
|
2021-03-21 10:14:10 +08:00
|
|
|
struct bpos src_want;
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
u64 dst_done = 0;
|
2021-03-13 09:30:39 +08:00
|
|
|
u32 dst_snapshot, src_snapshot;
|
2019-10-11 00:47:22 +08:00
|
|
|
int ret = 0, ret2 = 0;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2023-02-10 01:21:45 +08:00
|
|
|
if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_reflink))
|
2022-12-12 09:37:11 +08:00
|
|
|
return -BCH_ERR_erofs_no_writes;
|
2019-10-20 07:03:23 +08:00
|
|
|
|
2019-12-29 09:17:06 +08:00
|
|
|
bch2_check_set_feature(c, BCH_FEATURE_reflink);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
dst_end.offset += remap_sectors;
|
|
|
|
src_end.offset += remap_sectors;
|
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_init(&new_dst);
|
|
|
|
bch2_bkey_buf_init(&new_src);
|
2023-09-13 05:16:02 +08:00
|
|
|
trans = bch2_trans_get(c);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
ret = bch2_inum_opts_get(trans, src_inum, &opts);
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_iter_init(trans, &src_iter, BTREE_ID_extents, src_start,
|
2021-08-31 03:18:31 +08:00
|
|
|
BTREE_ITER_INTENT);
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_iter_init(trans, &dst_iter, BTREE_ID_extents, dst_start,
|
2021-08-31 03:18:31 +08:00
|
|
|
BTREE_ITER_INTENT);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2022-07-18 11:06:38 +08:00
|
|
|
while ((ret == 0 ||
|
|
|
|
bch2_err_matches(ret, BCH_ERR_transaction_restart)) &&
|
2022-11-24 16:12:22 +08:00
|
|
|
bkey_lt(dst_iter.pos, dst_end)) {
|
2021-05-17 12:08:06 +08:00
|
|
|
struct disk_reservation disk_res = { 0 };
|
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_begin(trans);
|
2019-12-21 05:35:24 +08:00
|
|
|
|
2019-08-16 21:59:56 +08:00
|
|
|
if (fatal_signal_pending(current)) {
|
|
|
|
ret = -EINTR;
|
2021-03-20 08:29:11 +08:00
|
|
|
break;
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
ret = bch2_subvolume_get_snapshot(trans, src_inum.subvol,
|
2021-03-13 09:30:39 +08:00
|
|
|
&src_snapshot);
|
2021-03-16 12:28:17 +08:00
|
|
|
if (ret)
|
|
|
|
continue;
|
|
|
|
|
2021-03-13 09:30:39 +08:00
|
|
|
bch2_btree_iter_set_snapshot(&src_iter, src_snapshot);
|
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
ret = bch2_subvolume_get_snapshot(trans, dst_inum.subvol,
|
2021-03-13 09:30:39 +08:00
|
|
|
&dst_snapshot);
|
2021-03-16 12:28:17 +08:00
|
|
|
if (ret)
|
|
|
|
continue;
|
|
|
|
|
2021-03-13 09:30:39 +08:00
|
|
|
bch2_btree_iter_set_snapshot(&dst_iter, dst_snapshot);
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
dst_done = dst_iter.pos.offset - dst_start.offset;
|
2021-03-21 10:14:10 +08:00
|
|
|
src_want = POS(src_start.inode, src_start.offset + dst_done);
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_iter_set_pos(&src_iter, src_want);
|
2021-03-21 10:14:10 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
src_k = get_next_src(&src_iter, src_end);
|
2019-08-16 21:59:56 +08:00
|
|
|
ret = bkey_err(src_k);
|
|
|
|
if (ret)
|
2021-03-20 08:29:11 +08:00
|
|
|
continue;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2022-11-24 16:12:22 +08:00
|
|
|
if (bkey_lt(src_want, src_iter.pos)) {
|
2023-09-13 05:16:02 +08:00
|
|
|
ret = bch2_fpunch_at(trans, &dst_iter, dst_inum,
|
2021-03-13 09:30:39 +08:00
|
|
|
min(dst_end.offset,
|
|
|
|
dst_iter.pos.offset +
|
|
|
|
src_iter.pos.offset - src_want.offset),
|
2021-11-06 03:17:13 +08:00
|
|
|
i_sectors_delta);
|
2019-08-16 21:59:56 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2020-10-25 07:51:34 +08:00
|
|
|
if (src_k.k->type != KEY_TYPE_reflink_p) {
|
2021-08-31 03:18:31 +08:00
|
|
|
bch2_btree_iter_set_pos_to_extent_start(&src_iter);
|
2021-06-15 06:16:10 +08:00
|
|
|
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_reassemble(&new_src, c, src_k);
|
2019-11-10 05:01:15 +08:00
|
|
|
src_k = bkey_i_to_s_c(new_src.k);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
ret = bch2_make_extent_indirect(trans, &src_iter,
|
2020-10-25 07:51:34 +08:00
|
|
|
new_src.k);
|
2019-08-16 21:59:56 +08:00
|
|
|
if (ret)
|
2021-03-20 08:29:11 +08:00
|
|
|
continue;
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
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 =
|
2020-12-18 04:08:58 +08:00
|
|
|
bkey_reflink_p_init(new_dst.k);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
|
|
|
u64 offset = le64_to_cpu(src_p.v->idx) +
|
2021-03-21 10:14:10 +08:00
|
|
|
(src_want.offset -
|
2019-08-16 21:59:56 +08:00
|
|
|
bkey_start_offset(src_k.k));
|
|
|
|
|
|
|
|
dst_p->v.idx = cpu_to_le64(offset);
|
|
|
|
} else {
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
new_dst.k->k.p = dst_iter.pos;
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_key_resize(&new_dst.k->k,
|
2021-03-21 10:14:10 +08:00
|
|
|
min(src_k.k->p.offset - src_want.offset,
|
2021-08-31 03:18:31 +08:00
|
|
|
dst_end.offset - dst_iter.pos.offset));
|
2021-03-13 09:30:39 +08:00
|
|
|
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
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);
|
2021-05-17 12:08:06 +08:00
|
|
|
bch2_disk_reservation_put(c, &disk_res);
|
2019-08-16 21:59:56 +08:00
|
|
|
}
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_iter_exit(trans, &dst_iter);
|
|
|
|
bch2_trans_iter_exit(trans, &src_iter);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2022-11-24 16:12:22 +08:00
|
|
|
BUG_ON(!ret && !bkey_eq(dst_iter.pos, dst_end));
|
|
|
|
BUG_ON(bkey_gt(dst_iter.pos, dst_end));
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2021-08-31 03:18:31 +08:00
|
|
|
dst_done = dst_iter.pos.offset - dst_start.offset;
|
|
|
|
new_i_size = min(dst_iter.pos.offset << 9, new_i_size);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2019-10-11 00:47:22 +08:00
|
|
|
do {
|
|
|
|
struct bch_inode_unpacked inode_u;
|
2021-08-31 03:18:31 +08:00
|
|
|
struct btree_iter inode_iter = { NULL };
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_begin(trans);
|
2021-07-25 08:24:10 +08:00
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
ret2 = bch2_inode_peek(trans, &inode_iter, &inode_u,
|
2021-03-16 12:28:17 +08:00
|
|
|
dst_inum, BTREE_ITER_INTENT);
|
2019-08-16 21:59:56 +08:00
|
|
|
|
2019-10-11 00:47:22 +08:00
|
|
|
if (!ret2 &&
|
2019-11-05 11:22:13 +08:00
|
|
|
inode_u.bi_size < new_i_size) {
|
|
|
|
inode_u.bi_size = new_i_size;
|
2023-09-13 05:16:02 +08:00
|
|
|
ret2 = bch2_inode_write(trans, &inode_iter, &inode_u) ?:
|
|
|
|
bch2_trans_commit(trans, NULL, NULL,
|
2021-11-10 06:20:06 +08:00
|
|
|
BTREE_INSERT_NOFAIL);
|
2019-11-05 11:22:13 +08:00
|
|
|
}
|
2021-03-20 08:29:11 +08:00
|
|
|
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_iter_exit(trans, &inode_iter);
|
2022-07-18 11:06:38 +08:00
|
|
|
} while (bch2_err_matches(ret2, BCH_ERR_transaction_restart));
|
bcachefs: rebalance_work
This adds a new btree, rebalance_work, to eliminate scanning required
for finding extents that need work done on them in the background - i.e.
for the background_target and background_compression options.
rebalance_work is a bitset btree, where a KEY_TYPE_set corresponds to an
extent in the extents or reflink btree at the same pos.
A new extent field is added, bch_extent_rebalance, which indicates that
this extent has work that needs to be done in the background - and which
options to use. This allows per-inode options to be propagated to
indirect extents - at least in some circumstances. In this patch,
changing IO options on a file will not propagate the new options to
indirect extents pointed to by that file.
Updating (setting/clearing) the rebalance_work btree is done by the
extent trigger, which looks at the bch_extent_rebalance field.
Scanning is still requrired after changing IO path options - either just
for a given inode, or for the whole filesystem. We indicate that
scanning is required by adding a KEY_TYPE_cookie key to the
rebalance_work btree: the cookie counter is so that we can detect that
scanning is still required when an option has been flipped mid-way
through an existing scan.
Future possible work:
- Propagate options to indirect extents when being changed
- Add other IO path options - nr_replicas, ec, to rebalance_work so
they can be applied in the background when they change
- Add a counter, for bcachefs fs usage output, showing the pending
amount of rebalance work: we'll probably want to do this after the
disk space accounting rewrite (moving it to a new btree)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-10-21 01:33:14 +08:00
|
|
|
err:
|
2023-09-13 05:16:02 +08:00
|
|
|
bch2_trans_put(trans);
|
2020-12-18 04:08:58 +08:00
|
|
|
bch2_bkey_buf_exit(&new_src, c);
|
|
|
|
bch2_bkey_buf_exit(&new_dst, c);
|
2019-10-11 00:47:22 +08:00
|
|
|
|
2023-02-10 01:21:45 +08:00
|
|
|
bch2_write_ref_put(c, BCH_WRITE_REF_reflink);
|
2019-10-20 07:03:23 +08:00
|
|
|
|
2019-10-11 00:47:22 +08:00
|
|
|
return dst_done ?: ret ?: ret2;
|
|
|
|
}
|