2017-03-17 06:18:50 +00:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
|
|
|
|
#include "bcachefs.h"
|
2022-03-18 04:42:09 +00:00
|
|
|
#include "alloc_background.h"
|
2018-10-06 04:46:55 +00:00
|
|
|
#include "alloc_foreground.h"
|
2022-03-18 04:42:09 +00:00
|
|
|
#include "backpointers.h"
|
2020-12-17 20:08:58 +00:00
|
|
|
#include "bkey_buf.h"
|
2017-03-17 06:18:50 +00:00
|
|
|
#include "btree_gc.h"
|
|
|
|
#include "btree_update.h"
|
2019-01-21 20:32:13 +00:00
|
|
|
#include "btree_update_interior.h"
|
2022-03-18 04:42:09 +00:00
|
|
|
#include "btree_write_buffer.h"
|
2018-11-05 04:10:09 +00:00
|
|
|
#include "disk_groups.h"
|
2021-10-29 20:29:13 +00:00
|
|
|
#include "ec.h"
|
2022-07-18 23:42:58 +00:00
|
|
|
#include "errcode.h"
|
2022-03-18 04:42:09 +00:00
|
|
|
#include "error.h"
|
2017-03-17 06:18:50 +00:00
|
|
|
#include "inode.h"
|
|
|
|
#include "io.h"
|
|
|
|
#include "journal_reclaim.h"
|
|
|
|
#include "keylist.h"
|
|
|
|
#include "move.h"
|
|
|
|
#include "replicas.h"
|
|
|
|
#include "super-io.h"
|
|
|
|
#include "trace.h"
|
|
|
|
|
|
|
|
#include <linux/ioprio.h>
|
|
|
|
#include <linux/kthread.h>
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
|
|
|
|
{
|
|
|
|
mutex_lock(&c->data_progress_lock);
|
|
|
|
list_add(&stats->list, &c->data_progress_list);
|
|
|
|
mutex_unlock(&c->data_progress_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
|
|
|
|
{
|
|
|
|
mutex_lock(&c->data_progress_lock);
|
|
|
|
list_del(&stats->list);
|
|
|
|
mutex_unlock(&c->data_progress_lock);
|
|
|
|
}
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
struct moving_io {
|
|
|
|
struct list_head list;
|
|
|
|
struct closure cl;
|
|
|
|
bool read_completed;
|
|
|
|
|
|
|
|
unsigned read_sectors;
|
|
|
|
unsigned write_sectors;
|
|
|
|
|
|
|
|
struct bch_read_bio rbio;
|
|
|
|
|
2022-06-13 23:07:19 +00:00
|
|
|
struct data_update write;
|
2017-03-17 06:18:50 +00:00
|
|
|
/* Must be last since it is variable size */
|
|
|
|
struct bio_vec bi_inline_vecs[0];
|
|
|
|
};
|
|
|
|
|
2022-10-29 06:47:33 +00:00
|
|
|
static void move_free(struct moving_io *io)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
|
|
|
struct moving_context *ctxt = io->write.ctxt;
|
2022-06-20 23:43:35 +00:00
|
|
|
struct bch_fs *c = ctxt->c;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
bch2_data_update_exit(&io->write);
|
2017-03-17 06:18:50 +00:00
|
|
|
wake_up(&ctxt->wait);
|
2023-02-09 17:21:45 +00:00
|
|
|
bch2_write_ref_put(c, BCH_WRITE_REF_move);
|
2017-03-17 06:18:50 +00:00
|
|
|
kfree(io);
|
|
|
|
}
|
|
|
|
|
2022-10-29 06:47:33 +00:00
|
|
|
static void move_write_done(struct bch_write_op *op)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
2022-10-29 06:47:33 +00:00
|
|
|
struct moving_io *io = container_of(op, struct moving_io, write.op);
|
|
|
|
struct moving_context *ctxt = io->write.ctxt;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-03-18 04:42:09 +00:00
|
|
|
if (io->write.op.error)
|
|
|
|
ctxt->write_error = true;
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
|
2022-10-29 06:47:33 +00:00
|
|
|
move_free(io);
|
|
|
|
closure_put(&ctxt->cl);
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
|
|
|
|
2022-10-29 06:47:33 +00:00
|
|
|
static void move_write(struct moving_io *io)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
|
|
|
if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
|
2022-10-29 06:47:33 +00:00
|
|
|
move_free(io);
|
2017-03-17 06:18:50 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2022-10-29 06:47:33 +00:00
|
|
|
closure_get(&io->write.ctxt->cl);
|
2017-03-17 06:18:50 +00:00
|
|
|
atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
|
2022-10-29 06:47:33 +00:00
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
bch2_data_update_read_done(&io->write, io->rbio.pick.crc);
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
|
|
|
|
{
|
|
|
|
struct moving_io *io =
|
|
|
|
list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
|
|
|
|
|
|
|
|
return io && io->read_completed ? io : NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void move_read_endio(struct bio *bio)
|
|
|
|
{
|
|
|
|
struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
|
|
|
|
struct moving_context *ctxt = io->write.ctxt;
|
|
|
|
|
|
|
|
atomic_sub(io->read_sectors, &ctxt->read_sectors);
|
|
|
|
io->read_completed = true;
|
|
|
|
|
2022-02-21 18:22:11 +00:00
|
|
|
wake_up(&ctxt->wait);
|
2017-03-17 06:18:50 +00:00
|
|
|
closure_put(&ctxt->cl);
|
|
|
|
}
|
|
|
|
|
2022-02-16 04:40:30 +00:00
|
|
|
static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
|
|
|
struct moving_io *io;
|
|
|
|
|
2022-02-16 04:40:30 +00:00
|
|
|
if (trans)
|
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
while ((io = next_pending_write(ctxt))) {
|
|
|
|
list_del(&io->list);
|
2022-10-29 06:47:33 +00:00
|
|
|
move_write(io);
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-02-16 04:40:30 +00:00
|
|
|
#define move_ctxt_wait_event(_ctxt, _trans, _cond) \
|
2017-03-17 06:18:50 +00:00
|
|
|
do { \
|
2022-02-16 04:40:30 +00:00
|
|
|
do_pending_writes(_ctxt, _trans); \
|
2017-03-17 06:18:50 +00:00
|
|
|
\
|
|
|
|
if (_cond) \
|
|
|
|
break; \
|
|
|
|
__wait_event((_ctxt)->wait, \
|
|
|
|
next_pending_write(_ctxt) || (_cond)); \
|
|
|
|
} while (1)
|
|
|
|
|
2022-02-16 04:40:30 +00:00
|
|
|
static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
|
|
|
|
struct btree_trans *trans)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
|
|
|
unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
|
|
|
|
|
2022-02-16 04:40:30 +00:00
|
|
|
move_ctxt_wait_event(ctxt, trans,
|
2017-03-17 06:18:50 +00:00
|
|
|
!atomic_read(&ctxt->write_sectors) ||
|
|
|
|
atomic_read(&ctxt->write_sectors) != sectors_pending);
|
|
|
|
}
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
void bch2_moving_ctxt_exit(struct moving_context *ctxt)
|
|
|
|
{
|
|
|
|
move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
|
|
|
|
closure_sync(&ctxt->cl);
|
|
|
|
EBUG_ON(atomic_read(&ctxt->write_sectors));
|
|
|
|
|
2022-06-29 21:14:06 +00:00
|
|
|
if (ctxt->stats) {
|
|
|
|
progress_list_del(ctxt->c, ctxt->stats);
|
|
|
|
|
|
|
|
trace_move_data(ctxt->c,
|
|
|
|
atomic64_read(&ctxt->stats->sectors_moved),
|
|
|
|
atomic64_read(&ctxt->stats->keys_moved));
|
|
|
|
}
|
2022-06-20 19:40:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void bch2_moving_ctxt_init(struct moving_context *ctxt,
|
|
|
|
struct bch_fs *c,
|
|
|
|
struct bch_ratelimit *rate,
|
|
|
|
struct bch_move_stats *stats,
|
|
|
|
struct write_point_specifier wp,
|
|
|
|
bool wait_on_copygc)
|
|
|
|
{
|
|
|
|
memset(ctxt, 0, sizeof(*ctxt));
|
|
|
|
|
|
|
|
ctxt->c = c;
|
|
|
|
ctxt->rate = rate;
|
|
|
|
ctxt->stats = stats;
|
|
|
|
ctxt->wp = wp;
|
|
|
|
ctxt->wait_on_copygc = wait_on_copygc;
|
|
|
|
|
|
|
|
closure_init_stack(&ctxt->cl);
|
|
|
|
INIT_LIST_HEAD(&ctxt->reads);
|
|
|
|
init_waitqueue_head(&ctxt->wait);
|
|
|
|
|
2022-06-29 21:14:06 +00:00
|
|
|
if (stats) {
|
|
|
|
progress_list_add(c, stats);
|
2022-06-20 19:40:26 +00:00
|
|
|
stats->data_type = BCH_DATA_user;
|
2022-06-29 21:14:06 +00:00
|
|
|
}
|
2022-06-20 19:40:26 +00:00
|
|
|
}
|
|
|
|
|
2022-11-14 01:01:42 +00:00
|
|
|
void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
|
2022-06-20 19:40:26 +00:00
|
|
|
{
|
|
|
|
memset(stats, 0, sizeof(*stats));
|
|
|
|
scnprintf(stats->name, sizeof(stats->name), "%s", name);
|
|
|
|
}
|
|
|
|
|
2022-10-09 07:32:17 +00:00
|
|
|
static int bch2_extent_drop_ptrs(struct btree_trans *trans,
|
|
|
|
struct btree_iter *iter,
|
|
|
|
struct bkey_s_c k,
|
|
|
|
struct data_update_opts data_opts)
|
|
|
|
{
|
|
|
|
struct bch_fs *c = trans->c;
|
|
|
|
struct bkey_i *n;
|
|
|
|
int ret;
|
|
|
|
|
2022-11-24 03:13:19 +00:00
|
|
|
n = bch2_bkey_make_mut(trans, k);
|
2022-10-09 07:32:17 +00:00
|
|
|
ret = PTR_ERR_OR_ZERO(n);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
while (data_opts.kill_ptrs) {
|
|
|
|
unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
|
|
|
|
struct bch_extent_ptr *ptr;
|
|
|
|
|
|
|
|
bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
|
|
|
|
data_opts.kill_ptrs ^= 1U << drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the new extent no longer has any pointers, bch2_extent_normalize()
|
|
|
|
* will do the appropriate thing with it (turning it into a
|
|
|
|
* KEY_TYPE_error key, or just a discard if it was a cached extent)
|
|
|
|
*/
|
|
|
|
bch2_extent_normalize(c, bkey_i_to_s(n));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Since we're not inserting through an extent iterator
|
|
|
|
* (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
|
|
|
|
* we aren't using the extent overwrite path to delete, we're
|
|
|
|
* just using the normal key deletion path:
|
|
|
|
*/
|
|
|
|
if (bkey_deleted(&n->k))
|
|
|
|
n->k.size = 0;
|
|
|
|
|
|
|
|
return bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
|
|
|
|
bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
|
|
|
|
}
|
|
|
|
|
2020-10-17 01:39:16 +00:00
|
|
|
static int bch2_move_extent(struct btree_trans *trans,
|
2022-10-09 07:32:17 +00:00
|
|
|
struct btree_iter *iter,
|
2017-03-17 06:18:50 +00:00
|
|
|
struct moving_context *ctxt,
|
|
|
|
struct bch_io_opts io_opts,
|
2019-08-16 13:59:56 +00:00
|
|
|
enum btree_id btree_id,
|
2019-07-25 17:52:14 +00:00
|
|
|
struct bkey_s_c k,
|
2022-06-13 23:17:45 +00:00
|
|
|
struct data_update_opts data_opts)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
2020-10-17 01:39:16 +00:00
|
|
|
struct bch_fs *c = trans->c;
|
2019-07-25 17:52:14 +00:00
|
|
|
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
|
2017-03-17 06:18:50 +00:00
|
|
|
struct moving_io *io;
|
2018-09-28 01:08:39 +00:00
|
|
|
const union bch_extent_entry *entry;
|
|
|
|
struct extent_ptr_decoded p;
|
2019-07-25 17:52:14 +00:00
|
|
|
unsigned sectors = k.k->size, pages;
|
2017-03-17 06:18:50 +00:00
|
|
|
int ret = -ENOMEM;
|
|
|
|
|
2022-10-09 07:32:17 +00:00
|
|
|
bch2_data_update_opts_normalize(k, &data_opts);
|
|
|
|
|
|
|
|
if (!data_opts.rewrite_ptrs &&
|
|
|
|
!data_opts.extra_replicas) {
|
|
|
|
if (data_opts.kill_ptrs)
|
|
|
|
return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2023-02-09 17:21:45 +00:00
|
|
|
if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_move))
|
2022-12-12 01:37:11 +00:00
|
|
|
return -BCH_ERR_erofs_no_writes;
|
2022-06-20 23:43:35 +00:00
|
|
|
|
bcachefs: Nocow support
This adds support for nocow mode, where we do writes in-place when
possible. Patch components:
- New boolean filesystem and inode option, nocow: note that when nocow
is enabled, data checksumming and compression are implicitly disabled
- To prevent in-place writes from racing with data moves
(data_update.c) or bucket reuse (i.e. a bucket being reused and
re-allocated while a nocow write is in flight, we have a new locking
mechanism.
Buckets can be locked for either data update or data move, using a
fixed size hash table of two_state_shared locks. We don't have any
chaining, meaning updates and moves to different buckets that hash to
the same lock will wait unnecessarily - we'll want to watch for this
becoming an issue.
- The allocator path also needs to check for in-place writes in flight
to a given bucket before giving it out: thus we add another counter
to bucket_alloc_state so we can track this.
- Fsync now may need to issue cache flushes to block devices instead of
flushing the journal. We add a device bitmask to bch_inode_info,
ei_devs_need_flush, which tracks devices that need to have flushes
issued - note that this will lead to unnecessary flushes when other
codepaths have already issued flushes, we may want to replace this with
a sequence number.
- New nocow write path: look up extents, and if they're writable write
to them - otherwise fall back to the normal COW write path.
XXX: switch to sequence numbers instead of bitmask for devs needing
journal flush
XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to
run in process context - see if we can improve this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
|
|
|
/*
|
|
|
|
* Before memory allocations & taking nocow locks in
|
|
|
|
* bch2_data_update_init():
|
|
|
|
*/
|
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
/* write path might have to decompress data: */
|
2019-07-25 17:52:14 +00:00
|
|
|
bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
|
2018-09-28 01:08:39 +00:00
|
|
|
sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
|
|
|
|
io = kzalloc(sizeof(struct moving_io) +
|
|
|
|
sizeof(struct bio_vec) * pages, GFP_KERNEL);
|
|
|
|
if (!io)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
io->write.ctxt = ctxt;
|
2019-07-25 17:52:14 +00:00
|
|
|
io->read_sectors = k.k->size;
|
|
|
|
io->write_sectors = k.k->size;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
|
|
|
|
bio_set_prio(&io->write.op.wbio.bio,
|
|
|
|
IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
|
|
|
|
|
|
|
|
if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
|
|
|
|
GFP_KERNEL))
|
|
|
|
goto err_free;
|
|
|
|
|
2019-09-07 17:16:41 +00:00
|
|
|
io->rbio.c = c;
|
|
|
|
io->rbio.opts = io_opts;
|
2017-03-17 06:18:50 +00:00
|
|
|
bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
|
|
|
|
io->rbio.bio.bi_vcnt = pages;
|
|
|
|
bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
|
|
|
|
io->rbio.bio.bi_iter.bi_size = sectors << 9;
|
|
|
|
|
|
|
|
io->rbio.bio.bi_opf = REQ_OP_READ;
|
2019-07-25 17:52:14 +00:00
|
|
|
io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
|
2017-03-17 06:18:50 +00:00
|
|
|
io->rbio.bio.bi_end_io = move_read_endio;
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
|
2022-06-13 23:17:45 +00:00
|
|
|
data_opts, btree_id, k);
|
2022-11-14 06:31:10 +00:00
|
|
|
if (ret && ret != -BCH_ERR_unwritten_extent_update)
|
2017-03-17 06:18:50 +00:00
|
|
|
goto err_free_pages;
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
io->write.ctxt = ctxt;
|
2022-10-29 06:47:33 +00:00
|
|
|
io->write.op.end_io = move_write_done;
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
atomic64_inc(&ctxt->stats->keys_moved);
|
2019-07-25 17:52:14 +00:00
|
|
|
atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
|
2022-11-14 06:31:10 +00:00
|
|
|
|
|
|
|
if (ret == -BCH_ERR_unwritten_extent_update) {
|
|
|
|
bch2_update_unwritten_extent(trans, &io->write);
|
|
|
|
move_free(io);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
BUG_ON(ret);
|
|
|
|
|
2022-03-15 08:36:33 +00:00
|
|
|
this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
|
2022-08-27 16:48:36 +00:00
|
|
|
this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
|
|
|
|
trace_move_extent_read(k.k);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
atomic_add(io->read_sectors, &ctxt->read_sectors);
|
|
|
|
list_add_tail(&io->list, &ctxt->reads);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* dropped by move_read_endio() - guards against use after free of
|
|
|
|
* ctxt when doing wakeup
|
|
|
|
*/
|
|
|
|
closure_get(&ctxt->cl);
|
2021-03-15 01:30:08 +00:00
|
|
|
bch2_read_extent(trans, &io->rbio,
|
|
|
|
bkey_start_pos(k.k),
|
|
|
|
btree_id, k, 0,
|
2017-03-17 06:18:50 +00:00
|
|
|
BCH_READ_NODECODE|
|
|
|
|
BCH_READ_LAST_FRAGMENT);
|
|
|
|
return 0;
|
|
|
|
err_free_pages:
|
|
|
|
bio_free_pages(&io->write.op.wbio.bio);
|
|
|
|
err_free:
|
|
|
|
kfree(io);
|
|
|
|
err:
|
2023-02-09 17:21:45 +00:00
|
|
|
bch2_write_ref_put(c, BCH_WRITE_REF_move);
|
2022-08-27 16:48:36 +00:00
|
|
|
trace_and_count(c, move_extent_alloc_mem_fail, k.k);
|
2017-03-17 06:18:50 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2021-03-16 22:08:10 +00:00
|
|
|
static int lookup_inode(struct btree_trans *trans, struct bpos pos,
|
|
|
|
struct bch_inode_unpacked *inode)
|
|
|
|
{
|
2021-08-30 19:18:31 +00:00
|
|
|
struct btree_iter iter;
|
2021-03-16 22:08:10 +00:00
|
|
|
struct bkey_s_c k;
|
|
|
|
int ret;
|
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
|
|
|
|
BTREE_ITER_ALL_SNAPSHOTS);
|
|
|
|
k = bch2_btree_iter_peek(&iter);
|
2021-03-16 22:08:10 +00:00
|
|
|
ret = bkey_err(k);
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
2022-11-24 08:12:22 +00:00
|
|
|
if (!k.k || !bkey_eq(k.k->p, pos)) {
|
2021-05-23 22:42:51 +00:00
|
|
|
ret = -ENOENT;
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
2021-10-30 01:14:23 +00:00
|
|
|
ret = bkey_is_inode(k.k) ? 0 : -EIO;
|
2021-03-16 22:08:10 +00:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
2021-10-30 01:14:23 +00:00
|
|
|
ret = bch2_inode_unpack(k, inode);
|
2021-03-16 22:08:10 +00:00
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
err:
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_trans_iter_exit(trans, &iter);
|
2021-03-16 22:08:10 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2022-06-15 14:06:43 +00:00
|
|
|
static int move_ratelimit(struct btree_trans *trans,
|
2022-06-20 19:40:26 +00:00
|
|
|
struct moving_context *ctxt)
|
2022-06-15 14:06:43 +00:00
|
|
|
{
|
|
|
|
struct bch_fs *c = trans->c;
|
|
|
|
u64 delay;
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
if (ctxt->wait_on_copygc) {
|
2022-06-15 14:06:43 +00:00
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
wait_event_killable(c->copygc_running_wq,
|
|
|
|
!c->copygc_running ||
|
|
|
|
kthread_should_stop());
|
|
|
|
}
|
|
|
|
|
|
|
|
do {
|
2022-06-20 19:40:26 +00:00
|
|
|
delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
|
2022-06-15 14:06:43 +00:00
|
|
|
|
|
|
|
if (delay) {
|
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
|
|
|
|
__set_current_state(TASK_RUNNING);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (delay)
|
|
|
|
schedule_timeout(delay);
|
|
|
|
|
|
|
|
if (unlikely(freezing(current))) {
|
|
|
|
move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
|
|
|
|
try_to_freeze();
|
|
|
|
}
|
|
|
|
} while (delay);
|
|
|
|
|
|
|
|
move_ctxt_wait_event(ctxt, trans,
|
|
|
|
atomic_read(&ctxt->write_sectors) <
|
|
|
|
c->opts.move_bytes_in_flight >> 9);
|
|
|
|
|
|
|
|
move_ctxt_wait_event(ctxt, trans,
|
|
|
|
atomic_read(&ctxt->read_sectors) <
|
|
|
|
c->opts.move_bytes_in_flight >> 9);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2022-03-18 04:42:09 +00:00
|
|
|
static int move_get_io_opts(struct btree_trans *trans,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct bkey_s_c k, u64 *cur_inum)
|
|
|
|
{
|
|
|
|
struct bch_inode_unpacked inode;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (*cur_inum == k.k->p.inode)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
ret = lookup_inode(trans,
|
|
|
|
SPOS(0, k.k->p.inode, k.k->p.snapshot),
|
|
|
|
&inode);
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (!ret)
|
|
|
|
bch2_inode_opts_get(io_opts, trans->c, &inode);
|
|
|
|
else
|
|
|
|
*io_opts = bch2_opts_to_inode_opts(trans->c->opts);
|
|
|
|
*cur_inum = k.k->p.inode;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
static int __bch2_move_data(struct moving_context *ctxt,
|
2022-06-15 14:06:43 +00:00
|
|
|
struct bpos start,
|
|
|
|
struct bpos end,
|
|
|
|
move_pred_fn pred, void *arg,
|
2022-06-20 19:40:26 +00:00
|
|
|
enum btree_id btree_id)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
2022-06-20 19:40:26 +00:00
|
|
|
struct bch_fs *c = ctxt->c;
|
2017-03-17 06:18:50 +00:00
|
|
|
struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
|
2020-12-17 20:08:58 +00:00
|
|
|
struct bkey_buf sk;
|
2019-03-25 19:10:15 +00:00
|
|
|
struct btree_trans trans;
|
2021-08-30 19:18:31 +00:00
|
|
|
struct btree_iter iter;
|
2017-03-17 06:18:50 +00:00
|
|
|
struct bkey_s_c k;
|
2022-06-13 23:17:45 +00:00
|
|
|
struct data_update_opts data_opts;
|
2022-06-15 14:06:43 +00:00
|
|
|
u64 cur_inum = U64_MAX;
|
2017-03-17 06:18:50 +00:00
|
|
|
int ret = 0, ret2;
|
|
|
|
|
2020-12-17 20:08:58 +00:00
|
|
|
bch2_bkey_buf_init(&sk);
|
2019-05-15 14:54:43 +00:00
|
|
|
bch2_trans_init(&trans, c, 0, 0);
|
2019-03-25 19:10:15 +00:00
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
ctxt->stats->data_type = BCH_DATA_user;
|
|
|
|
ctxt->stats->btree_id = btree_id;
|
|
|
|
ctxt->stats->pos = start;
|
2019-03-25 19:10:15 +00:00
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_trans_iter_init(&trans, &iter, btree_id, start,
|
2021-03-16 04:28:17 +00:00
|
|
|
BTREE_ITER_PREFETCH|
|
|
|
|
BTREE_ITER_ALL_SNAPSHOTS);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
if (ctxt->rate)
|
|
|
|
bch2_ratelimit_reset(ctxt->rate);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
while (!move_ratelimit(&trans, ctxt)) {
|
2021-07-25 00:24:10 +00:00
|
|
|
bch2_trans_begin(&trans);
|
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
k = bch2_btree_iter_peek(&iter);
|
2017-03-17 06:18:50 +00:00
|
|
|
if (!k.k)
|
|
|
|
break;
|
2022-01-10 01:52:10 +00:00
|
|
|
|
2019-03-28 02:03:30 +00:00
|
|
|
ret = bkey_err(k);
|
2022-07-18 03:06:38 +00:00
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
2022-01-10 01:52:10 +00:00
|
|
|
continue;
|
2017-03-17 06:18:50 +00:00
|
|
|
if (ret)
|
|
|
|
break;
|
2022-01-10 01:52:10 +00:00
|
|
|
|
2022-11-24 08:12:22 +00:00
|
|
|
if (bkey_ge(bkey_start_pos(k.k), end))
|
2017-03-17 06:18:50 +00:00
|
|
|
break;
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
ctxt->stats->pos = iter.pos;
|
2022-01-10 01:52:10 +00:00
|
|
|
|
2019-09-07 20:13:20 +00:00
|
|
|
if (!bkey_extent_is_direct_data(k.k))
|
2017-03-17 06:18:50 +00:00
|
|
|
goto next_nondata;
|
|
|
|
|
2022-03-18 04:42:09 +00:00
|
|
|
ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
|
|
|
|
if (ret)
|
|
|
|
continue;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
memset(&data_opts, 0, sizeof(data_opts));
|
|
|
|
if (!pred(c, arg, k, &io_opts, &data_opts))
|
2017-03-17 06:18:50 +00:00
|
|
|
goto next;
|
|
|
|
|
2022-02-15 05:06:59 +00:00
|
|
|
/*
|
|
|
|
* The iterator gets unlocked by __bch2_read_extent - need to
|
|
|
|
* save a copy of @k elsewhere:
|
2022-10-19 22:31:33 +00:00
|
|
|
*/
|
2020-12-17 20:08:58 +00:00
|
|
|
bch2_bkey_buf_reassemble(&sk, c, k);
|
2019-11-09 21:01:15 +00:00
|
|
|
k = bkey_i_to_s_c(sk.k);
|
bcachefs: Nocow support
This adds support for nocow mode, where we do writes in-place when
possible. Patch components:
- New boolean filesystem and inode option, nocow: note that when nocow
is enabled, data checksumming and compression are implicitly disabled
- To prevent in-place writes from racing with data moves
(data_update.c) or bucket reuse (i.e. a bucket being reused and
re-allocated while a nocow write is in flight, we have a new locking
mechanism.
Buckets can be locked for either data update or data move, using a
fixed size hash table of two_state_shared locks. We don't have any
chaining, meaning updates and moves to different buckets that hash to
the same lock will wait unnecessarily - we'll want to watch for this
becoming an issue.
- The allocator path also needs to check for in-place writes in flight
to a given bucket before giving it out: thus we add another counter
to bucket_alloc_state so we can track this.
- Fsync now may need to issue cache flushes to block devices instead of
flushing the journal. We add a device bitmask to bch_inode_info,
ei_devs_need_flush, which tracks devices that need to have flushes
issued - note that this will lead to unnecessary flushes when other
codepaths have already issued flushes, we may want to replace this with
a sequence number.
- New nocow write path: look up extents, and if they're writable write
to them - otherwise fall back to the normal COW write path.
XXX: switch to sequence numbers instead of bitmask for devs needing
journal flush
XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to
run in process context - see if we can improve this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
|
|
|
bch2_trans_unlock(&trans);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-10-09 07:32:17 +00:00
|
|
|
ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
|
2022-06-13 23:17:45 +00:00
|
|
|
btree_id, k, data_opts);
|
2017-03-17 06:18:50 +00:00
|
|
|
if (ret2) {
|
2022-07-18 03:06:38 +00:00
|
|
|
if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
|
2020-12-21 02:42:19 +00:00
|
|
|
continue;
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
if (ret2 == -ENOMEM) {
|
|
|
|
/* memory allocation failure, wait for some IO to finish */
|
2022-02-16 04:40:30 +00:00
|
|
|
bch2_move_ctxt_wait_for_io(ctxt, &trans);
|
2017-03-17 06:18:50 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* XXX signal failure */
|
|
|
|
goto next;
|
|
|
|
}
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
if (ctxt->rate)
|
|
|
|
bch2_ratelimit_increment(ctxt->rate, k.k->size);
|
2017-03-17 06:18:50 +00:00
|
|
|
next:
|
2022-06-20 19:40:26 +00:00
|
|
|
atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
|
2017-03-17 06:18:50 +00:00
|
|
|
next_nondata:
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_btree_iter_advance(&iter);
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
2021-03-20 00:29:11 +00:00
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
2021-10-19 19:08:00 +00:00
|
|
|
bch2_trans_exit(&trans);
|
2020-12-17 20:08:58 +00:00
|
|
|
bch2_bkey_buf_exit(&sk, c);
|
2019-08-16 13:59:56 +00:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int bch2_move_data(struct bch_fs *c,
|
2021-03-14 23:01:14 +00:00
|
|
|
enum btree_id start_btree_id, struct bpos start_pos,
|
|
|
|
enum btree_id end_btree_id, struct bpos end_pos,
|
2019-08-16 13:59:56 +00:00
|
|
|
struct bch_ratelimit *rate,
|
2022-06-15 14:06:43 +00:00
|
|
|
struct bch_move_stats *stats,
|
2022-06-20 19:40:26 +00:00
|
|
|
struct write_point_specifier wp,
|
|
|
|
bool wait_on_copygc,
|
|
|
|
move_pred_fn pred, void *arg)
|
2019-08-16 13:59:56 +00:00
|
|
|
{
|
2022-06-20 19:40:26 +00:00
|
|
|
struct moving_context ctxt;
|
2021-03-14 23:01:14 +00:00
|
|
|
enum btree_id id;
|
2019-08-16 13:59:56 +00:00
|
|
|
int ret;
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
|
2019-08-16 13:59:56 +00:00
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
for (id = start_btree_id;
|
|
|
|
id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
|
|
|
|
id++) {
|
|
|
|
stats->btree_id = id;
|
|
|
|
|
2021-02-21 00:27:37 +00:00
|
|
|
if (id != BTREE_ID_extents &&
|
|
|
|
id != BTREE_ID_reflink)
|
2021-03-14 23:01:14 +00:00
|
|
|
continue;
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
ret = __bch2_move_data(&ctxt,
|
2021-03-14 23:01:14 +00:00
|
|
|
id == start_btree_id ? start_pos : POS_MIN,
|
|
|
|
id == end_btree_id ? end_pos : POS_MAX,
|
2022-06-20 19:40:26 +00:00
|
|
|
pred, arg, id);
|
2021-03-14 23:01:14 +00:00
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2022-06-20 19:40:26 +00:00
|
|
|
bch2_moving_ctxt_exit(&ctxt);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2022-03-18 04:42:09 +00:00
|
|
|
static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
|
|
|
|
{
|
|
|
|
struct bch_fs *c = trans->c;
|
|
|
|
struct btree_iter iter;
|
|
|
|
struct bkey_s_c k;
|
|
|
|
struct printbuf buf = PRINTBUF;
|
|
|
|
struct bch_backpointer bp;
|
|
|
|
u64 bp_offset = 0;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
|
|
|
|
bucket, BTREE_ITER_CACHED);
|
|
|
|
again:
|
|
|
|
k = bch2_btree_iter_peek_slot(&iter);
|
|
|
|
ret = bkey_err(k);
|
|
|
|
|
|
|
|
if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
|
|
|
|
struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
|
|
|
|
|
|
|
|
if (a.v->gen == gen &&
|
|
|
|
a.v->dirty_sectors) {
|
|
|
|
if (a.v->data_type == BCH_DATA_btree) {
|
|
|
|
bch2_trans_unlock(trans);
|
|
|
|
if (bch2_btree_interior_updates_flush(c))
|
|
|
|
goto again;
|
|
|
|
goto failed_to_evacuate;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_trans_iter_exit(trans, &iter);
|
|
|
|
return ret;
|
|
|
|
failed_to_evacuate:
|
|
|
|
bch2_trans_iter_exit(trans, &iter);
|
|
|
|
|
|
|
|
prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
|
|
|
|
bch2_bkey_val_to_text(&buf, c, k);
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
bch2_trans_begin(trans);
|
|
|
|
|
|
|
|
ret = bch2_get_next_backpointer(trans, bucket, gen,
|
2022-10-14 11:02:36 +00:00
|
|
|
&bp_offset, &bp,
|
|
|
|
BTREE_ITER_CACHED);
|
2022-03-18 04:42:09 +00:00
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
if (bp_offset == U64_MAX)
|
|
|
|
break;
|
|
|
|
|
|
|
|
k = bch2_backpointer_get_key(trans, &iter,
|
|
|
|
bucket, bp_offset, bp);
|
|
|
|
ret = bkey_err(k);
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
if (!k.k)
|
|
|
|
continue;
|
|
|
|
prt_newline(&buf);
|
|
|
|
bch2_bkey_val_to_text(&buf, c, k);
|
|
|
|
bch2_trans_iter_exit(trans, &iter);
|
|
|
|
}
|
|
|
|
|
|
|
|
bch2_print_string_as_lines(KERN_ERR, buf.buf);
|
|
|
|
printbuf_exit(&buf);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int __bch2_evacuate_bucket(struct moving_context *ctxt,
|
|
|
|
struct bpos bucket, int gen,
|
|
|
|
struct data_update_opts _data_opts)
|
|
|
|
{
|
|
|
|
struct bch_fs *c = ctxt->c;
|
|
|
|
struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
|
|
|
|
struct btree_trans trans;
|
|
|
|
struct btree_iter iter;
|
|
|
|
struct bkey_buf sk;
|
|
|
|
struct bch_backpointer bp;
|
|
|
|
struct bch_alloc_v4 a_convert;
|
|
|
|
const struct bch_alloc_v4 *a;
|
|
|
|
struct bkey_s_c k;
|
|
|
|
struct data_update_opts data_opts;
|
|
|
|
unsigned dirty_sectors, bucket_size;
|
|
|
|
u64 bp_offset = 0, cur_inum = U64_MAX;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
bch2_bkey_buf_init(&sk);
|
|
|
|
bch2_trans_init(&trans, c, 0, 0);
|
|
|
|
|
|
|
|
bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc,
|
|
|
|
bucket, BTREE_ITER_CACHED);
|
|
|
|
ret = lockrestart_do(&trans,
|
|
|
|
bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
|
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
|
|
|
|
|
|
|
if (ret) {
|
|
|
|
bch_err(c, "%s: error looking up alloc key: %s", __func__, bch2_err_str(ret));
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
a = bch2_alloc_to_v4(k, &a_convert);
|
|
|
|
dirty_sectors = a->dirty_sectors;
|
|
|
|
bucket_size = bch_dev_bkey_exists(c, bucket.inode)->mi.bucket_size;
|
|
|
|
|
|
|
|
ret = bch2_btree_write_buffer_flush(&trans);
|
|
|
|
if (ret) {
|
|
|
|
bch_err(c, "%s: error flushing btree write buffer: %s", __func__, bch2_err_str(ret));
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
while (!(ret = move_ratelimit(&trans, ctxt))) {
|
|
|
|
bch2_trans_begin(&trans);
|
|
|
|
|
|
|
|
ret = bch2_get_next_backpointer(&trans, bucket, gen,
|
2022-10-14 11:02:36 +00:00
|
|
|
&bp_offset, &bp,
|
|
|
|
BTREE_ITER_CACHED);
|
2022-03-18 04:42:09 +00:00
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
if (bp_offset == U64_MAX)
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (!bp.level) {
|
|
|
|
const struct bch_extent_ptr *ptr;
|
|
|
|
struct bkey_s_c k;
|
|
|
|
unsigned i = 0;
|
|
|
|
|
|
|
|
k = bch2_backpointer_get_key(&trans, &iter,
|
|
|
|
bucket, bp_offset, bp);
|
|
|
|
ret = bkey_err(k);
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
if (!k.k)
|
|
|
|
goto next;
|
|
|
|
|
|
|
|
bch2_bkey_buf_reassemble(&sk, c, k);
|
|
|
|
k = bkey_i_to_s_c(sk.k);
|
|
|
|
|
|
|
|
ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
|
|
|
|
if (ret) {
|
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
data_opts = _data_opts;
|
|
|
|
data_opts.target = io_opts.background_target;
|
|
|
|
data_opts.rewrite_ptrs = 0;
|
|
|
|
|
|
|
|
bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
|
|
|
|
if (ptr->dev == bucket.inode)
|
|
|
|
data_opts.rewrite_ptrs |= 1U << i;
|
|
|
|
i++;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = bch2_move_extent(&trans, &iter, ctxt, io_opts,
|
|
|
|
bp.btree_id, k, data_opts);
|
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
|
|
|
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret == -ENOMEM) {
|
|
|
|
/* memory allocation failure, wait for some IO to finish */
|
|
|
|
bch2_move_ctxt_wait_for_io(ctxt, &trans);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
if (ctxt->rate)
|
|
|
|
bch2_ratelimit_increment(ctxt->rate, k.k->size);
|
|
|
|
atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
|
|
|
|
} else {
|
|
|
|
struct btree *b;
|
|
|
|
|
|
|
|
b = bch2_backpointer_get_node(&trans, &iter,
|
|
|
|
bucket, bp_offset, bp);
|
|
|
|
ret = PTR_ERR_OR_ZERO(b);
|
|
|
|
if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
|
|
|
|
continue;
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
if (!b)
|
|
|
|
goto next;
|
|
|
|
|
|
|
|
ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
|
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
|
|
|
|
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
if (ctxt->rate)
|
|
|
|
bch2_ratelimit_increment(ctxt->rate,
|
|
|
|
c->opts.btree_node_size >> 9);
|
|
|
|
atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
|
|
|
|
atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
|
|
|
|
}
|
|
|
|
next:
|
|
|
|
bp_offset++;
|
|
|
|
}
|
|
|
|
|
|
|
|
trace_evacuate_bucket(c, &bucket, dirty_sectors, bucket_size, ret);
|
|
|
|
|
|
|
|
if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
|
|
|
|
bch2_trans_unlock(&trans);
|
|
|
|
move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
|
|
|
|
closure_sync(&ctxt->cl);
|
|
|
|
if (!ctxt->write_error)
|
|
|
|
lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
|
|
|
|
}
|
|
|
|
err:
|
|
|
|
bch2_trans_exit(&trans);
|
|
|
|
bch2_bkey_buf_exit(&sk, c);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int bch2_evacuate_bucket(struct bch_fs *c,
|
|
|
|
struct bpos bucket, int gen,
|
|
|
|
struct data_update_opts data_opts,
|
|
|
|
struct bch_ratelimit *rate,
|
|
|
|
struct bch_move_stats *stats,
|
|
|
|
struct write_point_specifier wp,
|
|
|
|
bool wait_on_copygc)
|
|
|
|
{
|
|
|
|
struct moving_context ctxt;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
|
|
|
|
ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
|
|
|
|
bch2_moving_ctxt_exit(&ctxt);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
typedef bool (*move_btree_pred)(struct bch_fs *, void *,
|
|
|
|
struct btree *, struct bch_io_opts *,
|
|
|
|
struct data_update_opts *);
|
2021-03-14 23:01:14 +00:00
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
static int bch2_move_btree(struct bch_fs *c,
|
2021-03-14 23:01:14 +00:00
|
|
|
enum btree_id start_btree_id, struct bpos start_pos,
|
|
|
|
enum btree_id end_btree_id, struct bpos end_pos,
|
|
|
|
move_btree_pred pred, void *arg,
|
2017-03-17 06:18:50 +00:00
|
|
|
struct bch_move_stats *stats)
|
|
|
|
{
|
2021-03-14 23:01:14 +00:00
|
|
|
bool kthread = (current->flags & PF_KTHREAD) != 0;
|
2017-03-17 06:18:50 +00:00
|
|
|
struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
|
2019-03-25 19:10:15 +00:00
|
|
|
struct btree_trans trans;
|
2021-08-30 19:18:31 +00:00
|
|
|
struct btree_iter iter;
|
2017-03-17 06:18:50 +00:00
|
|
|
struct btree *b;
|
2021-03-14 23:01:14 +00:00
|
|
|
enum btree_id id;
|
2022-06-13 23:17:45 +00:00
|
|
|
struct data_update_opts data_opts;
|
2017-03-17 06:18:50 +00:00
|
|
|
int ret = 0;
|
|
|
|
|
2019-05-15 14:54:43 +00:00
|
|
|
bch2_trans_init(&trans, c, 0, 0);
|
2021-07-23 19:57:19 +00:00
|
|
|
progress_list_add(c, stats);
|
2019-03-25 19:10:15 +00:00
|
|
|
|
2020-07-09 22:28:11 +00:00
|
|
|
stats->data_type = BCH_DATA_btree;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
for (id = start_btree_id;
|
|
|
|
id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
|
|
|
|
id++) {
|
2019-03-25 19:10:15 +00:00
|
|
|
stats->btree_id = id;
|
|
|
|
|
2021-10-07 22:08:01 +00:00
|
|
|
bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
|
|
|
|
BTREE_ITER_PREFETCH);
|
2021-10-19 18:20:50 +00:00
|
|
|
retry:
|
2021-10-19 19:11:45 +00:00
|
|
|
ret = 0;
|
2021-10-07 22:08:01 +00:00
|
|
|
while (bch2_trans_begin(&trans),
|
2021-10-19 18:20:50 +00:00
|
|
|
(b = bch2_btree_iter_peek_node(&iter)) &&
|
|
|
|
!(ret = PTR_ERR_OR_ZERO(b))) {
|
2021-03-14 23:01:14 +00:00
|
|
|
if (kthread && kthread_should_stop())
|
2021-04-21 00:21:39 +00:00
|
|
|
break;
|
2021-03-14 23:01:14 +00:00
|
|
|
|
|
|
|
if ((cmp_int(id, end_btree_id) ?:
|
2021-07-06 02:02:07 +00:00
|
|
|
bpos_cmp(b->key.k.p, end_pos)) > 0)
|
2021-03-14 23:01:14 +00:00
|
|
|
break;
|
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
stats->pos = iter.pos;
|
2019-03-25 19:10:15 +00:00
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
if (!pred(c, arg, b, &io_opts, &data_opts))
|
2017-03-17 06:18:50 +00:00
|
|
|
goto next;
|
|
|
|
|
2021-10-24 20:59:33 +00:00
|
|
|
ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
|
2022-07-18 03:06:38 +00:00
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
2021-10-24 20:59:33 +00:00
|
|
|
continue;
|
|
|
|
if (ret)
|
|
|
|
break;
|
2017-03-17 06:18:50 +00:00
|
|
|
next:
|
2021-10-07 22:08:01 +00:00
|
|
|
bch2_btree_iter_next_node(&iter);
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
2022-07-18 03:06:38 +00:00
|
|
|
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
|
2021-10-19 18:20:50 +00:00
|
|
|
goto retry;
|
|
|
|
|
2021-08-30 19:18:31 +00:00
|
|
|
bch2_trans_iter_exit(&trans, &iter);
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2021-04-21 00:21:39 +00:00
|
|
|
if (kthread && kthread_should_stop())
|
|
|
|
break;
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
2021-04-21 00:21:39 +00:00
|
|
|
|
2019-03-25 19:10:15 +00:00
|
|
|
bch2_trans_exit(&trans);
|
|
|
|
|
2021-04-05 01:57:35 +00:00
|
|
|
if (ret)
|
2022-07-18 23:42:58 +00:00
|
|
|
bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
|
2021-04-05 01:57:35 +00:00
|
|
|
|
2022-04-17 21:30:49 +00:00
|
|
|
bch2_btree_interior_updates_flush(c);
|
2021-10-24 21:00:33 +00:00
|
|
|
|
2021-07-23 19:57:19 +00:00
|
|
|
progress_list_del(c, stats);
|
2017-03-17 06:18:50 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
static bool rereplicate_pred(struct bch_fs *c, void *arg,
|
|
|
|
struct bkey_s_c k,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct data_update_opts *data_opts)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
2018-11-01 19:10:01 +00:00
|
|
|
unsigned nr_good = bch2_bkey_durability(c, k);
|
2021-10-12 18:15:45 +00:00
|
|
|
unsigned replicas = bkey_is_btree_ptr(k.k)
|
|
|
|
? c->opts.metadata_replicas
|
|
|
|
: io_opts->data_replicas;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
if (!nr_good || nr_good >= replicas)
|
2022-06-13 23:17:45 +00:00
|
|
|
return false;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
|
|
|
data_opts->target = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
data_opts->extra_replicas = replicas - nr_good;
|
2018-11-01 19:10:01 +00:00
|
|
|
data_opts->btree_insert_flags = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
return true;
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
static bool migrate_pred(struct bch_fs *c, void *arg,
|
|
|
|
struct bkey_s_c k,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct data_update_opts *data_opts)
|
2017-03-17 06:18:50 +00:00
|
|
|
{
|
2022-06-13 23:17:45 +00:00
|
|
|
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
|
|
|
|
const struct bch_extent_ptr *ptr;
|
2017-03-17 06:18:50 +00:00
|
|
|
struct bch_ioctl_data *op = arg;
|
2022-06-13 23:17:45 +00:00
|
|
|
unsigned i = 0;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
data_opts->rewrite_ptrs = 0;
|
2017-03-17 06:18:50 +00:00
|
|
|
data_opts->target = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
data_opts->extra_replicas = 0;
|
2017-03-17 06:18:50 +00:00
|
|
|
data_opts->btree_insert_flags = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
|
|
|
|
bkey_for_each_ptr(ptrs, ptr) {
|
|
|
|
if (ptr->dev == op->migrate.dev)
|
|
|
|
data_opts->rewrite_ptrs |= 1U << i;
|
|
|
|
i++;
|
|
|
|
}
|
|
|
|
|
2022-10-19 22:31:33 +00:00
|
|
|
return data_opts->rewrite_ptrs != 0;
|
2017-03-17 06:18:50 +00:00
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
|
|
|
|
struct btree *b,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct data_update_opts *data_opts)
|
2021-03-14 23:01:14 +00:00
|
|
|
{
|
|
|
|
return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
|
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
static bool migrate_btree_pred(struct bch_fs *c, void *arg,
|
|
|
|
struct btree *b,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct data_update_opts *data_opts)
|
2021-03-14 23:01:14 +00:00
|
|
|
{
|
|
|
|
return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
|
|
|
|
}
|
|
|
|
|
2021-03-21 03:55:36 +00:00
|
|
|
static bool bformat_needs_redo(struct bkey_format *f)
|
|
|
|
{
|
|
|
|
unsigned i;
|
|
|
|
|
|
|
|
for (i = 0; i < f->nr_fields; i++) {
|
|
|
|
unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
|
|
|
|
u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
|
|
|
|
u64 field_offset = le64_to_cpu(f->field_offset[i]);
|
|
|
|
|
|
|
|
if (f->bits_per_field[i] > unpacked_bits)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
|
|
|
|
unpacked_mask) <
|
|
|
|
field_offset)
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
|
|
|
|
struct btree *b,
|
|
|
|
struct bch_io_opts *io_opts,
|
|
|
|
struct data_update_opts *data_opts)
|
2021-03-14 23:01:14 +00:00
|
|
|
{
|
|
|
|
if (b->version_ondisk != c->sb.version ||
|
2021-03-21 03:55:36 +00:00
|
|
|
btree_node_need_rewrite(b) ||
|
|
|
|
bformat_needs_redo(&b->format)) {
|
2021-03-14 23:01:14 +00:00
|
|
|
data_opts->target = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
data_opts->extra_replicas = 0;
|
2021-03-14 23:01:14 +00:00
|
|
|
data_opts->btree_insert_flags = 0;
|
2022-06-13 23:17:45 +00:00
|
|
|
return true;
|
2021-03-14 23:01:14 +00:00
|
|
|
}
|
|
|
|
|
2022-06-13 23:17:45 +00:00
|
|
|
return false;
|
2021-03-14 23:01:14 +00:00
|
|
|
}
|
|
|
|
|
2021-03-22 22:39:16 +00:00
|
|
|
int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = bch2_move_btree(c,
|
|
|
|
0, POS_MIN,
|
2021-07-06 02:02:07 +00:00
|
|
|
BTREE_ID_NR, SPOS_MAX,
|
2021-03-22 22:39:16 +00:00
|
|
|
rewrite_old_nodes_pred, c, stats);
|
|
|
|
if (!ret) {
|
|
|
|
mutex_lock(&c->sb_lock);
|
2021-05-23 21:04:13 +00:00
|
|
|
c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
|
|
|
|
c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
|
2021-03-22 22:39:16 +00:00
|
|
|
c->disk_sb.sb->version_min = c->disk_sb.sb->version;
|
|
|
|
bch2_write_super(c);
|
|
|
|
mutex_unlock(&c->sb_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-03-17 06:18:50 +00:00
|
|
|
int bch2_data_job(struct bch_fs *c,
|
|
|
|
struct bch_move_stats *stats,
|
|
|
|
struct bch_ioctl_data op)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
switch (op.op) {
|
|
|
|
case BCH_DATA_OP_REREPLICATE:
|
2022-11-14 01:01:42 +00:00
|
|
|
bch2_move_stats_init(stats, "rereplicate");
|
2020-07-09 22:28:11 +00:00
|
|
|
stats->data_type = BCH_DATA_journal;
|
2017-03-17 06:18:50 +00:00
|
|
|
ret = bch2_journal_flush_device_pins(&c->journal, -1);
|
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
ret = bch2_move_btree(c,
|
|
|
|
op.start_btree, op.start_pos,
|
|
|
|
op.end_btree, op.end_pos,
|
|
|
|
rereplicate_btree_pred, c, stats) ?: ret;
|
2019-04-30 21:15:39 +00:00
|
|
|
ret = bch2_replicas_gc2(c) ?: ret;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
ret = bch2_move_data(c,
|
|
|
|
op.start_btree, op.start_pos,
|
|
|
|
op.end_btree, op.end_pos,
|
2022-06-20 19:40:26 +00:00
|
|
|
NULL,
|
|
|
|
stats,
|
|
|
|
writepoint_hashed((unsigned long) current),
|
|
|
|
true,
|
|
|
|
rereplicate_pred, c) ?: ret;
|
2019-04-30 21:15:39 +00:00
|
|
|
ret = bch2_replicas_gc2(c) ?: ret;
|
2017-03-17 06:18:50 +00:00
|
|
|
break;
|
|
|
|
case BCH_DATA_OP_MIGRATE:
|
|
|
|
if (op.migrate.dev >= c->sb.nr_devices)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2022-11-14 01:01:42 +00:00
|
|
|
bch2_move_stats_init(stats, "migrate");
|
2020-07-09 22:28:11 +00:00
|
|
|
stats->data_type = BCH_DATA_journal;
|
2017-03-17 06:18:50 +00:00
|
|
|
ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
|
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
ret = bch2_move_btree(c,
|
|
|
|
op.start_btree, op.start_pos,
|
|
|
|
op.end_btree, op.end_pos,
|
|
|
|
migrate_btree_pred, &op, stats) ?: ret;
|
2019-04-30 21:15:39 +00:00
|
|
|
ret = bch2_replicas_gc2(c) ?: ret;
|
2017-03-17 06:18:50 +00:00
|
|
|
|
2021-03-14 23:01:14 +00:00
|
|
|
ret = bch2_move_data(c,
|
|
|
|
op.start_btree, op.start_pos,
|
|
|
|
op.end_btree, op.end_pos,
|
2022-06-20 19:40:26 +00:00
|
|
|
NULL,
|
|
|
|
stats,
|
|
|
|
writepoint_hashed((unsigned long) current),
|
|
|
|
true,
|
|
|
|
migrate_pred, &op) ?: ret;
|
2019-04-30 21:15:39 +00:00
|
|
|
ret = bch2_replicas_gc2(c) ?: ret;
|
2017-03-17 06:18:50 +00:00
|
|
|
break;
|
2021-03-14 23:01:14 +00:00
|
|
|
case BCH_DATA_OP_REWRITE_OLD_NODES:
|
2022-11-14 01:01:42 +00:00
|
|
|
bch2_move_stats_init(stats, "rewrite_old_nodes");
|
2021-03-22 22:39:16 +00:00
|
|
|
ret = bch2_scan_old_btree_nodes(c, stats);
|
2021-03-14 23:01:14 +00:00
|
|
|
break;
|
2017-03-17 06:18:50 +00:00
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|