git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_gc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
   4  * Copyright (C) 2014 Datera Inc.
   5  */
   6
   7 #include "bcachefs.h"
   8 #include "alloc_background.h"
   9 #include "alloc_foreground.h"
  10 #include "bkey_methods.h"
  11 #include "bkey_buf.h"
  12 #include "btree_locking.h"
  13 #include "btree_update_interior.h"
  14 #include "btree_io.h"
  15 #include "btree_gc.h"
  16 #include "buckets.h"
  17 #include "clock.h"
  18 #include "debug.h"
  19 #include "ec.h"
  20 #include "error.h"
  21 #include "extents.h"
  22 #include "journal.h"
  23 #include "keylist.h"
  24 #include "move.h"
  25 #include "recovery.h"
  26 #include "replicas.h"
  27 #include "super-io.h"
  28
  29 #include <linux/slab.h>
  30 #include <linux/bitops.h>
  31 #include <linux/freezer.h>
  32 #include <linux/kthread.h>
  33 #include <linux/preempt.h>
  34 #include <linux/rcupdate.h>
  35 #include <linux/sched/task.h>
  36 #include <trace/events/bcachefs.h>
  37
  38 static inline void __gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
  39 {
  40         preempt_disable();
  41         write_seqcount_begin(&c->gc_pos_lock);
  42         c->gc_pos = new_pos;
  43         write_seqcount_end(&c->gc_pos_lock);
  44         preempt_enable();
  45 }
  46
  47 static inline void gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
  48 {
  49         BUG_ON(gc_pos_cmp(new_pos, c->gc_pos) <= 0);
  50         __gc_pos_set(c, new_pos);
  51 }
  52
  53 static int bch2_gc_check_topology(struct bch_fs *c,
  54                                   struct bkey_s_c k,
  55                                   struct bpos *expected_start,
  56                                   struct bpos expected_end,
  57                                   bool is_last)
  58 {
  59         int ret = 0;
  60
  61         if (k.k->type == KEY_TYPE_btree_ptr_v2) {
  62                 struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
  63
  64                 if (fsck_err_on(bkey_cmp(*expected_start, bp.v->min_key), c,
  65                                 "btree node with incorrect min_key: got %llu:%llu, should be %llu:%llu",
  66                                 bp.v->min_key.inode,
  67                                 bp.v->min_key.offset,
  68                                 expected_start->inode,
  69                                 expected_start->offset)) {
  70                         BUG();
  71                 }
  72         }
  73
  74         *expected_start = bkey_cmp(k.k->p, POS_MAX)
  75                 ? bkey_successor(k.k->p)
  76                 : k.k->p;
  77
  78         if (fsck_err_on(is_last &&
  79                         bkey_cmp(k.k->p, expected_end), c,
  80                         "btree node with incorrect max_key: got %llu:%llu, should be %llu:%llu",
  81                         k.k->p.inode,
  82                         k.k->p.offset,
  83                         expected_end.inode,
  84                         expected_end.offset)) {
  85                 BUG();
  86         }
  87 fsck_err:
  88         return ret;
  89 }
  90
  91 /* marking of btree keys/nodes: */
  92
  93 static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
  94                             u8 *max_stale, bool initial)
  95 {
  96         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  97         const struct bch_extent_ptr *ptr;
  98         unsigned flags =
  99                 BTREE_TRIGGER_GC|
 100                 (initial ? BTREE_TRIGGER_NOATOMIC : 0);
 101         int ret = 0;
 102
 103         if (initial) {
 104                 BUG_ON(bch2_journal_seq_verify &&
 105                        k.k->version.lo > journal_cur_seq(&c->journal));
 106
 107                 /* XXX change to fsck check */
 108                 if (fsck_err_on(k.k->version.lo > atomic64_read(&c->key_version), c,
 109                                 "key version number higher than recorded: %llu > %llu",
 110                                 k.k->version.lo,
 111                                 atomic64_read(&c->key_version)))
 112                         atomic64_set(&c->key_version, k.k->version.lo);
 113
 114                 if (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
 115                     fsck_err_on(!bch2_bkey_replicas_marked(c, k), c,
 116                                 "superblock not marked as containing replicas (type %u)",
 117                                 k.k->type)) {
 118                         ret = bch2_mark_bkey_replicas(c, k);
 119                         if (ret)
 120                                 return ret;
 121                 }
 122
 123                 bkey_for_each_ptr(ptrs, ptr) {
 124                         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 125                         struct bucket *g = PTR_BUCKET(ca, ptr, true);
 126                         struct bucket *g2 = PTR_BUCKET(ca, ptr, false);
 127
 128                         if (mustfix_fsck_err_on(!g->gen_valid, c,
 129                                         "bucket %u:%zu data type %s ptr gen %u missing in alloc btree",
 130                                         ptr->dev, PTR_BUCKET_NR(ca, ptr),
 131                                         bch2_data_types[ptr_data_type(k.k, ptr)],
 132                                         ptr->gen)) {
 133                                 g2->_mark.gen   = g->_mark.gen          = ptr->gen;
 134                                 g2->gen_valid   = g->gen_valid          = true;
 135                                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);
 136                         }
 137
 138                         if (mustfix_fsck_err_on(gen_cmp(ptr->gen, g->mark.gen) > 0, c,
 139                                         "bucket %u:%zu data type %s ptr gen in the future: %u > %u",
 140                                         ptr->dev, PTR_BUCKET_NR(ca, ptr),
 141                                         bch2_data_types[ptr_data_type(k.k, ptr)],
 142                                         ptr->gen, g->mark.gen)) {
 143                                 g2->_mark.gen   = g->_mark.gen          = ptr->gen;
 144                                 g2->gen_valid   = g->gen_valid          = true;
 145                                 g2->_mark.data_type             = 0;
 146                                 g2->_mark.dirty_sectors         = 0;
 147                                 g2->_mark.cached_sectors        = 0;
 148                                 set_bit(BCH_FS_FIXED_GENS, &c->flags);
 149                                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);
 150                         }
 151                 }
 152         }
 153
 154         bkey_for_each_ptr(ptrs, ptr) {
 155                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 156                 struct bucket *g = PTR_BUCKET(ca, ptr, true);
 157
 158                 if (gen_after(g->oldest_gen, ptr->gen))
 159                         g->oldest_gen = ptr->gen;
 160
 161                 *max_stale = max(*max_stale, ptr_stale(ca, ptr));
 162         }
 163
 164         bch2_mark_key(c, k, 0, k.k->size, NULL, 0, flags);
 165 fsck_err:
 166         return ret;
 167 }
 168
 169 static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale,
 170                               bool initial)
 171 {
 172         struct bpos next_node_start = b->data->min_key;
 173         struct btree_node_iter iter;
 174         struct bkey unpacked;
 175         struct bkey_s_c k;
 176         int ret = 0;
 177
 178         *max_stale = 0;
 179
 180         if (!btree_node_type_needs_gc(btree_node_type(b)))
 181                 return 0;
 182
 183         bch2_btree_node_iter_init_from_start(&iter, b);
 184
 185         while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
 186                 bch2_bkey_debugcheck(c, b, k);
 187
 188                 ret = bch2_gc_mark_key(c, k, max_stale, initial);
 189                 if (ret)
 190                         break;
 191
 192                 bch2_btree_node_iter_advance(&iter, b);
 193
 194                 if (b->c.level) {
 195                         ret = bch2_gc_check_topology(c, k,
 196                                         &next_node_start,
 197                                         b->data->max_key,
 198                                         bch2_btree_node_iter_end(&iter));
 199                         if (ret)
 200                                 break;
 201                 }
 202         }
 203
 204         return ret;
 205 }
 206
 207 static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
 208                          bool initial, bool metadata_only)
 209 {
 210         struct btree_trans trans;
 211         struct btree_iter *iter;
 212         struct btree *b;
 213         unsigned depth = metadata_only                  ? 1
 214                 : bch2_expensive_debug_checks           ? 0
 215                 : !btree_node_type_needs_gc(btree_id)   ? 1
 216                 : 0;
 217         u8 max_stale = 0;
 218         int ret = 0;
 219
 220         bch2_trans_init(&trans, c, 0, 0);
 221
 222         gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
 223
 224         __for_each_btree_node(&trans, iter, btree_id, POS_MIN,
 225                               0, depth, BTREE_ITER_PREFETCH, b) {
 226                 bch2_verify_btree_nr_keys(b);
 227
 228                 gc_pos_set(c, gc_pos_btree_node(b));
 229
 230                 ret = btree_gc_mark_node(c, b, &max_stale, initial);
 231                 if (ret)
 232                         break;
 233
 234                 if (!initial) {
 235                         if (max_stale > 64)
 236                                 bch2_btree_node_rewrite(c, iter,
 237                                                 b->data->keys.seq,
 238                                                 BTREE_INSERT_NOWAIT|
 239                                                 BTREE_INSERT_GC_LOCK_HELD);
 240                         else if (!bch2_btree_gc_rewrite_disabled &&
 241                                  (bch2_btree_gc_always_rewrite || max_stale > 16))
 242                                 bch2_btree_node_rewrite(c, iter,
 243                                                 b->data->keys.seq,
 244                                                 BTREE_INSERT_NOWAIT|
 245                                                 BTREE_INSERT_GC_LOCK_HELD);
 246                 }
 247
 248                 bch2_trans_cond_resched(&trans);
 249         }
 250         ret = bch2_trans_exit(&trans) ?: ret;
 251         if (ret)
 252                 return ret;
 253
 254         mutex_lock(&c->btree_root_lock);
 255         b = c->btree_roots[btree_id].b;
 256         if (!btree_node_fake(b))
 257                 ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
 258                                        &max_stale, initial);
 259         gc_pos_set(c, gc_pos_btree_root(b->c.btree_id));
 260         mutex_unlock(&c->btree_root_lock);
 261
 262         return ret;
 263 }
 264
 265 static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b,
 266                                       struct journal_keys *journal_keys,
 267                                       unsigned target_depth)
 268 {
 269         struct btree_and_journal_iter iter;
 270         struct bkey_s_c k;
 271         struct bpos next_node_start = b->data->min_key;
 272         struct bkey_buf tmp;
 273         u8 max_stale = 0;
 274         int ret = 0;
 275
 276         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 277         bch2_bkey_buf_init(&tmp);
 278
 279         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 280                 bch2_bkey_debugcheck(c, b, k);
 281
 282                 BUG_ON(bkey_cmp(k.k->p, b->data->min_key) < 0);
 283                 BUG_ON(bkey_cmp(k.k->p, b->data->max_key) > 0);
 284
 285                 ret = bch2_gc_mark_key(c, k, &max_stale, true);
 286                 if (ret)
 287                         break;
 288
 289                 if (b->c.level) {
 290                         struct btree *child;
 291
 292                         bch2_bkey_buf_reassemble(&tmp, c, k);
 293                         k = bkey_i_to_s_c(tmp.k);
 294
 295                         bch2_btree_and_journal_iter_advance(&iter);
 296
 297                         ret = bch2_gc_check_topology(c, k,
 298                                         &next_node_start,
 299                                         b->data->max_key,
 300                                         !bch2_btree_and_journal_iter_peek(&iter).k);
 301                         if (ret)
 302                                 break;
 303
 304                         if (b->c.level > target_depth) {
 305                                 child = bch2_btree_node_get_noiter(c, tmp.k,
 306                                                         b->c.btree_id, b->c.level - 1);
 307                                 ret = PTR_ERR_OR_ZERO(child);
 308                                 if (ret)
 309                                         break;
 310
 311                                 ret = bch2_gc_btree_init_recurse(c, child,
 312                                                 journal_keys, target_depth);
 313                                 six_unlock_read(&child->c.lock);
 314
 315                                 if (ret)
 316                                         break;
 317                         }
 318                 } else {
 319                         bch2_btree_and_journal_iter_advance(&iter);
 320                 }
 321         }
 322
 323         bch2_bkey_buf_exit(&tmp, c);
 324         return ret;
 325 }
 326
 327 static int bch2_gc_btree_init(struct bch_fs *c,
 328                               struct journal_keys *journal_keys,
 329                               enum btree_id btree_id,
 330                               bool metadata_only)
 331 {
 332         struct btree *b;
 333         unsigned target_depth = metadata_only           ? 1
 334                 : bch2_expensive_debug_checks           ? 0
 335                 : !btree_node_type_needs_gc(btree_id)   ? 1
 336                 : 0;
 337         u8 max_stale = 0;
 338         int ret = 0;
 339
 340         b = c->btree_roots[btree_id].b;
 341
 342         if (btree_node_fake(b))
 343                 return 0;
 344
 345         six_lock_read(&b->c.lock, NULL, NULL);
 346         if (fsck_err_on(bkey_cmp(b->data->min_key, POS_MIN), c,
 347                         "btree root with incorrect min_key: %llu:%llu",
 348                         b->data->min_key.inode,
 349                         b->data->min_key.offset)) {
 350                 BUG();
 351         }
 352
 353         if (fsck_err_on(bkey_cmp(b->data->max_key, POS_MAX), c,
 354                         "btree root with incorrect min_key: %llu:%llu",
 355                         b->data->max_key.inode,
 356                         b->data->max_key.offset)) {
 357                 BUG();
 358         }
 359
 360         if (b->c.level >= target_depth)
 361                 ret = bch2_gc_btree_init_recurse(c, b,
 362                                         journal_keys, target_depth);
 363
 364         if (!ret)
 365                 ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
 366                                        &max_stale, true);
 367 fsck_err:
 368         six_unlock_read(&b->c.lock);
 369
 370         return ret;
 371 }
 372
 373 static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 374 {
 375         return  (int) btree_id_to_gc_phase(l) -
 376                 (int) btree_id_to_gc_phase(r);
 377 }
 378
 379 static int bch2_gc_btrees(struct bch_fs *c, struct journal_keys *journal_keys,
 380                           bool initial, bool metadata_only)
 381 {
 382         enum btree_id ids[BTREE_ID_NR];
 383         unsigned i;
 384
 385         for (i = 0; i < BTREE_ID_NR; i++)
 386                 ids[i] = i;
 387         bubble_sort(ids, BTREE_ID_NR, btree_id_gc_phase_cmp);
 388
 389         for (i = 0; i < BTREE_ID_NR; i++) {
 390                 enum btree_id id = ids[i];
 391                 int ret = initial
 392                         ? bch2_gc_btree_init(c, journal_keys,
 393                                              id, metadata_only)
 394                         : bch2_gc_btree(c, id, initial, metadata_only);
 395                 if (ret)
 396                         return ret;
 397         }
 398
 399         return 0;
 400 }
 401
 402 static void mark_metadata_sectors(struct bch_fs *c, struct bch_dev *ca,
 403                                   u64 start, u64 end,
 404                                   enum bch_data_type type,
 405                                   unsigned flags)
 406 {
 407         u64 b = sector_to_bucket(ca, start);
 408
 409         do {
 410                 unsigned sectors =
 411                         min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
 412
 413                 bch2_mark_metadata_bucket(c, ca, b, type, sectors,
 414                                           gc_phase(GC_PHASE_SB), flags);
 415                 b++;
 416                 start += sectors;
 417         } while (start < end);
 418 }
 419
 420 void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca,
 421                               unsigned flags)
 422 {
 423         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
 424         unsigned i;
 425         u64 b;
 426
 427         /*
 428          * This conditional is kind of gross, but we may be called from the
 429          * device add path, before the new device has actually been added to the
 430          * running filesystem:
 431          */
 432         if (c) {
 433                 lockdep_assert_held(&c->sb_lock);
 434                 percpu_down_read(&c->mark_lock);
 435         }
 436
 437         for (i = 0; i < layout->nr_superblocks; i++) {
 438                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
 439
 440                 if (offset == BCH_SB_SECTOR)
 441                         mark_metadata_sectors(c, ca, 0, BCH_SB_SECTOR,
 442                                               BCH_DATA_sb, flags);
 443
 444                 mark_metadata_sectors(c, ca, offset,
 445                                       offset + (1 << layout->sb_max_size_bits),
 446                                       BCH_DATA_sb, flags);
 447         }
 448
 449         for (i = 0; i < ca->journal.nr; i++) {
 450                 b = ca->journal.buckets[i];
 451                 bch2_mark_metadata_bucket(c, ca, b, BCH_DATA_journal,
 452                                           ca->mi.bucket_size,
 453                                           gc_phase(GC_PHASE_SB), flags);
 454         }
 455
 456         if (c)
 457                 percpu_up_read(&c->mark_lock);
 458 }
 459
 460 static void bch2_mark_superblocks(struct bch_fs *c)
 461 {
 462         struct bch_dev *ca;
 463         unsigned i;
 464
 465         mutex_lock(&c->sb_lock);
 466         gc_pos_set(c, gc_phase(GC_PHASE_SB));
 467
 468         for_each_online_member(ca, c, i)
 469                 bch2_mark_dev_superblock(c, ca, BTREE_TRIGGER_GC);
 470         mutex_unlock(&c->sb_lock);
 471 }
 472
 473 #if 0
 474 /* Also see bch2_pending_btree_node_free_insert_done() */
 475 static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
 476 {
 477         struct btree_update *as;
 478         struct pending_btree_node_free *d;
 479
 480         mutex_lock(&c->btree_interior_update_lock);
 481         gc_pos_set(c, gc_phase(GC_PHASE_PENDING_DELETE));
 482
 483         for_each_pending_btree_node_free(c, as, d)
 484                 if (d->index_update_done)
 485                         bch2_mark_key(c, bkey_i_to_s_c(&d->key),
 486                                       0, 0, NULL, 0,
 487                                       BTREE_TRIGGER_GC);
 488
 489         mutex_unlock(&c->btree_interior_update_lock);
 490 }
 491 #endif
 492
 493 static void bch2_mark_allocator_buckets(struct bch_fs *c)
 494 {
 495         struct bch_dev *ca;
 496         struct open_bucket *ob;
 497         size_t i, j, iter;
 498         unsigned ci;
 499
 500         percpu_down_read(&c->mark_lock);
 501
 502         spin_lock(&c->freelist_lock);
 503         gc_pos_set(c, gc_pos_alloc(c, NULL));
 504
 505         for_each_member_device(ca, c, ci) {
 506                 fifo_for_each_entry(i, &ca->free_inc, iter)
 507                         bch2_mark_alloc_bucket(c, ca, i, true,
 508                                                gc_pos_alloc(c, NULL),
 509                                                BTREE_TRIGGER_GC);
 510
 511
 512
 513                 for (j = 0; j < RESERVE_NR; j++)
 514                         fifo_for_each_entry(i, &ca->free[j], iter)
 515                                 bch2_mark_alloc_bucket(c, ca, i, true,
 516                                                        gc_pos_alloc(c, NULL),
 517                                                        BTREE_TRIGGER_GC);
 518         }
 519
 520         spin_unlock(&c->freelist_lock);
 521
 522         for (ob = c->open_buckets;
 523              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
 524              ob++) {
 525                 spin_lock(&ob->lock);
 526                 if (ob->valid) {
 527                         gc_pos_set(c, gc_pos_alloc(c, ob));
 528                         ca = bch_dev_bkey_exists(c, ob->ptr.dev);
 529                         bch2_mark_alloc_bucket(c, ca, PTR_BUCKET_NR(ca, &ob->ptr), true,
 530                                                gc_pos_alloc(c, ob),
 531                                                BTREE_TRIGGER_GC);
 532                 }
 533                 spin_unlock(&ob->lock);
 534         }
 535
 536         percpu_up_read(&c->mark_lock);
 537 }
 538
 539 static void bch2_gc_free(struct bch_fs *c)
 540 {
 541         struct bch_dev *ca;
 542         unsigned i;
 543
 544         genradix_free(&c->stripes[1]);
 545
 546         for_each_member_device(ca, c, i) {
 547                 kvpfree(rcu_dereference_protected(ca->buckets[1], 1),
 548                         sizeof(struct bucket_array) +
 549                         ca->mi.nbuckets * sizeof(struct bucket));
 550                 ca->buckets[1] = NULL;
 551
 552                 free_percpu(ca->usage[1]);
 553                 ca->usage[1] = NULL;
 554         }
 555
 556         free_percpu(c->usage_gc);
 557         c->usage_gc = NULL;
 558 }
 559
 560 static int bch2_gc_done(struct bch_fs *c,
 561                         bool initial, bool metadata_only)
 562 {
 563         struct bch_dev *ca;
 564         bool verify = !metadata_only &&
 565                 (!initial ||
 566                  (c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)));
 567         unsigned i;
 568         int ret = 0;
 569
 570 #define copy_field(_f, _msg, ...)                                       \
 571         if (dst->_f != src->_f) {                                       \
 572                 if (verify)                                             \
 573                         fsck_err(c, _msg ": got %llu, should be %llu"   \
 574                                 , ##__VA_ARGS__, dst->_f, src->_f);     \
 575                 dst->_f = src->_f;                                      \
 576                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
 577         }
 578 #define copy_stripe_field(_f, _msg, ...)                                \
 579         if (dst->_f != src->_f) {                                       \
 580                 if (verify)                                             \
 581                         fsck_err(c, "stripe %zu has wrong "_msg         \
 582                                 ": got %u, should be %u",               \
 583                                 dst_iter.pos, ##__VA_ARGS__,            \
 584                                 dst->_f, src->_f);                      \
 585                 dst->_f = src->_f;                                      \
 586                 dst->dirty = true;                                      \
 587                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
 588         }
 589 #define copy_bucket_field(_f)                                           \
 590         if (dst->b[b].mark._f != src->b[b].mark._f) {                   \
 591                 if (verify)                                             \
 592                         fsck_err(c, "bucket %u:%zu gen %u data type %s has wrong " #_f  \
 593                                 ": got %u, should be %u", i, b,         \
 594                                 dst->b[b].mark.gen,                     \
 595                                 bch2_data_types[dst->b[b].mark.data_type],\
 596                                 dst->b[b].mark._f, src->b[b].mark._f);  \
 597                 dst->b[b]._mark._f = src->b[b].mark._f;                 \
 598                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
 599         }
 600 #define copy_dev_field(_f, _msg, ...)                                   \
 601         copy_field(_f, "dev %u has wrong " _msg, i, ##__VA_ARGS__)
 602 #define copy_fs_field(_f, _msg, ...)                                    \
 603         copy_field(_f, "fs has wrong " _msg, ##__VA_ARGS__)
 604
 605         if (!metadata_only) {
 606                 struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
 607                 struct genradix_iter src_iter = genradix_iter_init(&c->stripes[1], 0);
 608                 struct stripe *dst, *src;
 609
 610                 c->ec_stripes_heap.used = 0;
 611
 612                 while ((dst = genradix_iter_peek(&dst_iter, &c->stripes[0])) &&
 613                        (src = genradix_iter_peek(&src_iter, &c->stripes[1]))) {
 614                         BUG_ON(src_iter.pos != dst_iter.pos);
 615
 616                         copy_stripe_field(alive,        "alive");
 617                         copy_stripe_field(sectors,      "sectors");
 618                         copy_stripe_field(algorithm,    "algorithm");
 619                         copy_stripe_field(nr_blocks,    "nr_blocks");
 620                         copy_stripe_field(nr_redundant, "nr_redundant");
 621                         copy_stripe_field(blocks_nonempty,
 622                                           "blocks_nonempty");
 623
 624                         for (i = 0; i < ARRAY_SIZE(dst->block_sectors); i++)
 625                                 copy_stripe_field(block_sectors[i],
 626                                                   "block_sectors[%u]", i);
 627
 628                         if (dst->alive) {
 629                                 spin_lock(&c->ec_stripes_heap_lock);
 630                                 bch2_stripes_heap_insert(c, dst, dst_iter.pos);
 631                                 spin_unlock(&c->ec_stripes_heap_lock);
 632                         }
 633
 634                         genradix_iter_advance(&dst_iter, &c->stripes[0]);
 635                         genradix_iter_advance(&src_iter, &c->stripes[1]);
 636                 }
 637         }
 638
 639         for_each_member_device(ca, c, i) {
 640                 struct bucket_array *dst = __bucket_array(ca, 0);
 641                 struct bucket_array *src = __bucket_array(ca, 1);
 642                 size_t b;
 643
 644                 for (b = 0; b < src->nbuckets; b++) {
 645                         copy_bucket_field(gen);
 646                         copy_bucket_field(data_type);
 647                         copy_bucket_field(owned_by_allocator);
 648                         copy_bucket_field(stripe);
 649                         copy_bucket_field(dirty_sectors);
 650                         copy_bucket_field(cached_sectors);
 651
 652                         dst->b[b].oldest_gen = src->b[b].oldest_gen;
 653                 }
 654         };
 655
 656         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 657                 bch2_fs_usage_acc_to_base(c, i);
 658
 659         bch2_dev_usage_from_buckets(c);
 660
 661         {
 662                 unsigned nr = fs_usage_u64s(c);
 663                 struct bch_fs_usage *dst = c->usage_base;
 664                 struct bch_fs_usage *src = (void *)
 665                         bch2_acc_percpu_u64s((void *) c->usage_gc, nr);
 666
 667                 copy_fs_field(hidden,           "hidden");
 668                 copy_fs_field(btree,            "btree");
 669
 670                 if (!metadata_only) {
 671                         copy_fs_field(data,     "data");
 672                         copy_fs_field(cached,   "cached");
 673                         copy_fs_field(reserved, "reserved");
 674                         copy_fs_field(nr_inodes,"nr_inodes");
 675
 676                         for (i = 0; i < BCH_REPLICAS_MAX; i++)
 677                                 copy_fs_field(persistent_reserved[i],
 678                                               "persistent_reserved[%i]", i);
 679                 }
 680
 681                 for (i = 0; i < c->replicas.nr; i++) {
 682                         struct bch_replicas_entry *e =
 683                                 cpu_replicas_entry(&c->replicas, i);
 684                         char buf[80];
 685
 686                         if (metadata_only &&
 687                             (e->data_type == BCH_DATA_user ||
 688                              e->data_type == BCH_DATA_cached))
 689                                 continue;
 690
 691                         bch2_replicas_entry_to_text(&PBUF(buf), e);
 692
 693                         copy_fs_field(replicas[i], "%s", buf);
 694                 }
 695         }
 696
 697 #undef copy_fs_field
 698 #undef copy_dev_field
 699 #undef copy_bucket_field
 700 #undef copy_stripe_field
 701 #undef copy_field
 702 fsck_err:
 703         return ret;
 704 }
 705
 706 static int bch2_gc_start(struct bch_fs *c,
 707                          bool metadata_only)
 708 {
 709         struct bch_dev *ca;
 710         unsigned i;
 711         int ret;
 712
 713         BUG_ON(c->usage_gc);
 714
 715         c->usage_gc = __alloc_percpu_gfp(fs_usage_u64s(c) * sizeof(u64),
 716                                          sizeof(u64), GFP_KERNEL);
 717         if (!c->usage_gc) {
 718                 bch_err(c, "error allocating c->usage_gc");
 719                 return -ENOMEM;
 720         }
 721
 722         for_each_member_device(ca, c, i) {
 723                 BUG_ON(ca->buckets[1]);
 724                 BUG_ON(ca->usage[1]);
 725
 726                 ca->buckets[1] = kvpmalloc(sizeof(struct bucket_array) +
 727                                 ca->mi.nbuckets * sizeof(struct bucket),
 728                                 GFP_KERNEL|__GFP_ZERO);
 729                 if (!ca->buckets[1]) {
 730                         percpu_ref_put(&ca->ref);
 731                         bch_err(c, "error allocating ca->buckets[gc]");
 732                         return -ENOMEM;
 733                 }
 734
 735                 ca->usage[1] = alloc_percpu(struct bch_dev_usage);
 736                 if (!ca->usage[1]) {
 737                         bch_err(c, "error allocating ca->usage[gc]");
 738                         percpu_ref_put(&ca->ref);
 739                         return -ENOMEM;
 740                 }
 741         }
 742
 743         ret = bch2_ec_mem_alloc(c, true);
 744         if (ret) {
 745                 bch_err(c, "error allocating ec gc mem");
 746                 return ret;
 747         }
 748
 749         percpu_down_write(&c->mark_lock);
 750
 751         /*
 752          * indicate to stripe code that we need to allocate for the gc stripes
 753          * radix tree, too
 754          */
 755         gc_pos_set(c, gc_phase(GC_PHASE_START));
 756
 757         for_each_member_device(ca, c, i) {
 758                 struct bucket_array *dst = __bucket_array(ca, 1);
 759                 struct bucket_array *src = __bucket_array(ca, 0);
 760                 size_t b;
 761
 762                 dst->first_bucket       = src->first_bucket;
 763                 dst->nbuckets           = src->nbuckets;
 764
 765                 for (b = 0; b < src->nbuckets; b++) {
 766                         struct bucket *d = &dst->b[b];
 767                         struct bucket *s = &src->b[b];
 768
 769                         d->_mark.gen = dst->b[b].oldest_gen = s->mark.gen;
 770                         d->gen_valid = s->gen_valid;
 771
 772                         if (metadata_only &&
 773                             (s->mark.data_type == BCH_DATA_user ||
 774                              s->mark.data_type == BCH_DATA_cached)) {
 775                                 d->_mark = s->mark;
 776                                 d->_mark.owned_by_allocator = 0;
 777                         }
 778                 }
 779         };
 780
 781         percpu_up_write(&c->mark_lock);
 782
 783         return 0;
 784 }
 785
 786 /**
 787  * bch2_gc - walk _all_ references to buckets, and recompute them:
 788  *
 789  * Order matters here:
 790  *  - Concurrent GC relies on the fact that we have a total ordering for
 791  *    everything that GC walks - see  gc_will_visit_node(),
 792  *    gc_will_visit_root()
 793  *
 794  *  - also, references move around in the course of index updates and
 795  *    various other crap: everything needs to agree on the ordering
 796  *    references are allowed to move around in - e.g., we're allowed to
 797  *    start with a reference owned by an open_bucket (the allocator) and
 798  *    move it to the btree, but not the reverse.
 799  *
 800  *    This is necessary to ensure that gc doesn't miss references that
 801  *    move around - if references move backwards in the ordering GC
 802  *    uses, GC could skip past them
 803  */
 804 int bch2_gc(struct bch_fs *c, struct journal_keys *journal_keys,
 805             bool initial, bool metadata_only)
 806 {
 807         struct bch_dev *ca;
 808         u64 start_time = local_clock();
 809         unsigned i, iter = 0;
 810         int ret;
 811
 812         lockdep_assert_held(&c->state_lock);
 813         trace_gc_start(c);
 814
 815         down_write(&c->gc_lock);
 816
 817         /* flush interior btree updates: */
 818         closure_wait_event(&c->btree_interior_update_wait,
 819                            !bch2_btree_interior_updates_nr_pending(c));
 820 again:
 821         ret = bch2_gc_start(c, metadata_only);
 822         if (ret)
 823                 goto out;
 824
 825         bch2_mark_superblocks(c);
 826
 827         ret = bch2_gc_btrees(c, journal_keys, initial, metadata_only);
 828         if (ret)
 829                 goto out;
 830
 831 #if 0
 832         bch2_mark_pending_btree_node_frees(c);
 833 #endif
 834         bch2_mark_allocator_buckets(c);
 835
 836         c->gc_count++;
 837 out:
 838         if (!ret &&
 839             (test_bit(BCH_FS_FIXED_GENS, &c->flags) ||
 840              (!iter && bch2_test_restart_gc))) {
 841                 /*
 842                  * XXX: make sure gens we fixed got saved
 843                  */
 844                 if (iter++ <= 2) {
 845                         bch_info(c, "Fixed gens, restarting mark and sweep:");
 846                         clear_bit(BCH_FS_FIXED_GENS, &c->flags);
 847                         __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
 848
 849                         percpu_down_write(&c->mark_lock);
 850                         bch2_gc_free(c);
 851                         percpu_up_write(&c->mark_lock);
 852                         /* flush fsck errors, reset counters */
 853                         bch2_flush_fsck_errs(c);
 854
 855                         goto again;
 856                 }
 857
 858                 bch_info(c, "Unable to fix bucket gens, looping");
 859                 ret = -EINVAL;
 860         }
 861
 862         if (!ret) {
 863                 bch2_journal_block(&c->journal);
 864
 865                 percpu_down_write(&c->mark_lock);
 866                 ret = bch2_gc_done(c, initial, metadata_only);
 867
 868                 bch2_journal_unblock(&c->journal);
 869         } else {
 870                 percpu_down_write(&c->mark_lock);
 871         }
 872
 873         /* Indicates that gc is no longer in progress: */
 874         __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
 875
 876         bch2_gc_free(c);
 877         percpu_up_write(&c->mark_lock);
 878
 879         up_write(&c->gc_lock);
 880
 881         trace_gc_end(c);
 882         bch2_time_stats_update(&c->times[BCH_TIME_btree_gc], start_time);
 883
 884         /*
 885          * Wake up allocator in case it was waiting for buckets
 886          * because of not being able to inc gens
 887          */
 888         for_each_member_device(ca, c, i)
 889                 bch2_wake_allocator(ca);
 890
 891         /*
 892          * At startup, allocations can happen directly instead of via the
 893          * allocator thread - issue wakeup in case they blocked on gc_lock:
 894          */
 895         closure_wake_up(&c->freelist_wait);
 896         return ret;
 897 }
 898
 899 static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k)
 900 {
 901         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 902         const struct bch_extent_ptr *ptr;
 903
 904         percpu_down_read(&c->mark_lock);
 905         bkey_for_each_ptr(ptrs, ptr) {
 906                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 907                 struct bucket *g = PTR_BUCKET(ca, ptr, false);
 908
 909                 if (gen_after(g->mark.gen, ptr->gen) > 16) {
 910                         percpu_up_read(&c->mark_lock);
 911                         return true;
 912                 }
 913         }
 914
 915         bkey_for_each_ptr(ptrs, ptr) {
 916                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 917                 struct bucket *g = PTR_BUCKET(ca, ptr, false);
 918
 919                 if (gen_after(g->gc_gen, ptr->gen))
 920                         g->gc_gen = ptr->gen;
 921         }
 922         percpu_up_read(&c->mark_lock);
 923
 924         return false;
 925 }
 926
 927 /*
 928  * For recalculating oldest gen, we only need to walk keys in leaf nodes; btree
 929  * node pointers currently never have cached pointers that can become stale:
 930  */
 931 static int bch2_gc_btree_gens(struct bch_fs *c, enum btree_id btree_id)
 932 {
 933         struct btree_trans trans;
 934         struct btree_iter *iter;
 935         struct bkey_s_c k;
 936         struct bkey_buf sk;
 937         int ret = 0;
 938
 939         bch2_bkey_buf_init(&sk);
 940         bch2_trans_init(&trans, c, 0, 0);
 941
 942         iter = bch2_trans_get_iter(&trans, btree_id, POS_MIN,
 943                                    BTREE_ITER_PREFETCH);
 944
 945         while ((k = bch2_btree_iter_peek(iter)).k &&
 946                !(ret = bkey_err(k))) {
 947                 if (gc_btree_gens_key(c, k)) {
 948                         bch2_bkey_buf_reassemble(&sk, c, k);
 949                         bch2_extent_normalize(c, bkey_i_to_s(sk.k));
 950
 951                         bch2_btree_iter_set_pos(iter, bkey_start_pos(&sk.k->k));
 952
 953                         bch2_trans_update(&trans, iter, sk.k, 0);
 954
 955                         ret = bch2_trans_commit(&trans, NULL, NULL,
 956                                                 BTREE_INSERT_NOFAIL);
 957                         if (ret == -EINTR)
 958                                 continue;
 959                         if (ret) {
 960                                 break;
 961                         }
 962                 }
 963
 964                 bch2_btree_iter_next(iter);
 965         }
 966
 967         bch2_trans_exit(&trans);
 968         bch2_bkey_buf_exit(&sk, c);
 969
 970         return ret;
 971 }
 972
 973 int bch2_gc_gens(struct bch_fs *c)
 974 {
 975         struct bch_dev *ca;
 976         struct bucket_array *buckets;
 977         struct bucket *g;
 978         unsigned i;
 979         int ret;
 980
 981         /*
 982          * Ideally we would be using state_lock and not gc_lock here, but that
 983          * introduces a deadlock in the RO path - we currently take the state
 984          * lock at the start of going RO, thus the gc thread may get stuck:
 985          */
 986         down_read(&c->gc_lock);
 987
 988         for_each_member_device(ca, c, i) {
 989                 down_read(&ca->bucket_lock);
 990                 buckets = bucket_array(ca);
 991
 992                 for_each_bucket(g, buckets)
 993                         g->gc_gen = g->mark.gen;
 994                 up_read(&ca->bucket_lock);
 995         }
 996
 997         for (i = 0; i < BTREE_ID_NR; i++)
 998                 if (btree_node_type_needs_gc(i)) {
 999                         ret = bch2_gc_btree_gens(c, i);
1000                         if (ret) {
1001                                 bch_err(c, "error recalculating oldest_gen: %i", ret);
1002                                 goto err;
1003                         }
1004                 }
1005
1006         for_each_member_device(ca, c, i) {
1007                 down_read(&ca->bucket_lock);
1008                 buckets = bucket_array(ca);
1009
1010                 for_each_bucket(g, buckets)
1011                         g->oldest_gen = g->gc_gen;
1012                 up_read(&ca->bucket_lock);
1013         }
1014
1015         c->gc_count++;
1016 err:
1017         up_read(&c->gc_lock);
1018         return ret;
1019 }
1020
1021 /* Btree coalescing */
1022
1023 static void recalc_packed_keys(struct btree *b)
1024 {
1025         struct bset *i = btree_bset_first(b);
1026         struct bkey_packed *k;
1027
1028         memset(&b->nr, 0, sizeof(b->nr));
1029
1030         BUG_ON(b->nsets != 1);
1031
1032         vstruct_for_each(i, k)
1033                 btree_keys_account_key_add(&b->nr, 0, k);
1034 }
1035
1036 static void bch2_coalesce_nodes(struct bch_fs *c, struct btree_iter *iter,
1037                                 struct btree *old_nodes[GC_MERGE_NODES])
1038 {
1039         struct btree *parent = btree_node_parent(iter, old_nodes[0]);
1040         unsigned i, nr_old_nodes, nr_new_nodes, u64s = 0;
1041         unsigned blocks = btree_blocks(c) * 2 / 3;
1042         struct btree *new_nodes[GC_MERGE_NODES];
1043         struct btree_update *as;
1044         struct keylist keylist;
1045         struct bkey_format_state format_state;
1046         struct bkey_format new_format;
1047
1048         memset(new_nodes, 0, sizeof(new_nodes));
1049         bch2_keylist_init(&keylist, NULL);
1050
1051         /* Count keys that are not deleted */
1052         for (i = 0; i < GC_MERGE_NODES && old_nodes[i]; i++)
1053                 u64s += old_nodes[i]->nr.live_u64s;
1054
1055         nr_old_nodes = nr_new_nodes = i;
1056
1057         /* Check if all keys in @old_nodes could fit in one fewer node */
1058         if (nr_old_nodes <= 1 ||
1059             __vstruct_blocks(struct btree_node, c->block_bits,
1060                              DIV_ROUND_UP(u64s, nr_old_nodes - 1)) > blocks)
1061                 return;
1062
1063         /* Find a format that all keys in @old_nodes can pack into */
1064         bch2_bkey_format_init(&format_state);
1065
1066         for (i = 0; i < nr_old_nodes; i++)
1067                 __bch2_btree_calc_format(&format_state, old_nodes[i]);
1068
1069         new_format = bch2_bkey_format_done(&format_state);
1070
1071         /* Check if repacking would make any nodes too big to fit */
1072         for (i = 0; i < nr_old_nodes; i++)
1073                 if (!bch2_btree_node_format_fits(c, old_nodes[i], &new_format)) {
1074                         trace_btree_gc_coalesce_fail(c,
1075                                         BTREE_GC_COALESCE_FAIL_FORMAT_FITS);
1076                         return;
1077                 }
1078
1079         if (bch2_keylist_realloc(&keylist, NULL, 0,
1080                         BKEY_BTREE_PTR_U64s_MAX * nr_old_nodes)) {
1081                 trace_btree_gc_coalesce_fail(c,
1082                                 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC);
1083                 return;
1084         }
1085
1086         as = bch2_btree_update_start(iter->trans, iter->btree_id,
1087                         btree_update_reserve_required(c, parent) + nr_old_nodes,
1088                         BTREE_INSERT_NOFAIL|
1089                         BTREE_INSERT_USE_RESERVE,
1090                         NULL);
1091         if (IS_ERR(as)) {
1092                 trace_btree_gc_coalesce_fail(c,
1093                                 BTREE_GC_COALESCE_FAIL_RESERVE_GET);
1094                 bch2_keylist_free(&keylist, NULL);
1095                 return;
1096         }
1097
1098         trace_btree_gc_coalesce(c, old_nodes[0]);
1099
1100         for (i = 0; i < nr_old_nodes; i++)
1101                 bch2_btree_interior_update_will_free_node(as, old_nodes[i]);
1102
1103         /* Repack everything with @new_format and sort down to one bset */
1104         for (i = 0; i < nr_old_nodes; i++)
1105                 new_nodes[i] =
1106                         __bch2_btree_node_alloc_replacement(as, old_nodes[i],
1107                                                             new_format);
1108
1109         /*
1110          * Conceptually we concatenate the nodes together and slice them
1111          * up at different boundaries.
1112          */
1113         for (i = nr_new_nodes - 1; i > 0; --i) {
1114                 struct btree *n1 = new_nodes[i];
1115                 struct btree *n2 = new_nodes[i - 1];
1116
1117                 struct bset *s1 = btree_bset_first(n1);
1118                 struct bset *s2 = btree_bset_first(n2);
1119                 struct bkey_packed *k, *last = NULL;
1120
1121                 /* Calculate how many keys from @n2 we could fit inside @n1 */
1122                 u64s = 0;
1123
1124                 for (k = s2->start;
1125                      k < vstruct_last(s2) &&
1126                      vstruct_blocks_plus(n1->data, c->block_bits,
1127                                          u64s + k->u64s) <= blocks;
1128                      k = bkey_next_skip_noops(k, vstruct_last(s2))) {
1129                         last = k;
1130                         u64s += k->u64s;
1131                 }
1132
1133                 if (u64s == le16_to_cpu(s2->u64s)) {
1134                         /* n2 fits entirely in n1 */
1135                         n1->key.k.p = n1->data->max_key = n2->data->max_key;
1136
1137                         memcpy_u64s(vstruct_last(s1),
1138                                     s2->start,
1139                                     le16_to_cpu(s2->u64s));
1140                         le16_add_cpu(&s1->u64s, le16_to_cpu(s2->u64s));
1141
1142                         set_btree_bset_end(n1, n1->set);
1143
1144                         six_unlock_write(&n2->c.lock);
1145                         bch2_btree_node_free_never_inserted(c, n2);
1146                         six_unlock_intent(&n2->c.lock);
1147
1148                         memmove(new_nodes + i - 1,
1149                                 new_nodes + i,
1150                                 sizeof(new_nodes[0]) * (nr_new_nodes - i));
1151                         new_nodes[--nr_new_nodes] = NULL;
1152                 } else if (u64s) {
1153                         /* move part of n2 into n1 */
1154                         n1->key.k.p = n1->data->max_key =
1155                                 bkey_unpack_pos(n1, last);
1156
1157                         n2->data->min_key = bkey_successor(n1->data->max_key);
1158
1159                         memcpy_u64s(vstruct_last(s1),
1160                                     s2->start, u64s);
1161                         le16_add_cpu(&s1->u64s, u64s);
1162
1163                         memmove(s2->start,
1164                                 vstruct_idx(s2, u64s),
1165                                 (le16_to_cpu(s2->u64s) - u64s) * sizeof(u64));
1166                         s2->u64s = cpu_to_le16(le16_to_cpu(s2->u64s) - u64s);
1167
1168                         set_btree_bset_end(n1, n1->set);
1169                         set_btree_bset_end(n2, n2->set);
1170                 }
1171         }
1172
1173         for (i = 0; i < nr_new_nodes; i++) {
1174                 struct btree *n = new_nodes[i];
1175
1176                 recalc_packed_keys(n);
1177                 btree_node_reset_sib_u64s(n);
1178
1179                 bch2_btree_build_aux_trees(n);
1180
1181                 bch2_btree_update_add_new_node(as, n);
1182                 six_unlock_write(&n->c.lock);
1183
1184                 bch2_btree_node_write(c, n, SIX_LOCK_intent);
1185         }
1186
1187         /*
1188          * The keys for the old nodes get deleted. We don't want to insert keys
1189          * that compare equal to the keys for the new nodes we'll also be
1190          * inserting - we can't because keys on a keylist must be strictly
1191          * greater than the previous keys, and we also don't need to since the
1192          * key for the new node will serve the same purpose (overwriting the key
1193          * for the old node).
1194          */
1195         for (i = 0; i < nr_old_nodes; i++) {
1196                 struct bkey_i delete;
1197                 unsigned j;
1198
1199                 for (j = 0; j < nr_new_nodes; j++)
1200                         if (!bkey_cmp(old_nodes[i]->key.k.p,
1201                                       new_nodes[j]->key.k.p))
1202                                 goto next;
1203
1204                 bkey_init(&delete.k);
1205                 delete.k.p = old_nodes[i]->key.k.p;
1206                 bch2_keylist_add_in_order(&keylist, &delete);
1207 next:
1208                 i = i;
1209         }
1210
1211         /*
1212          * Keys for the new nodes get inserted: bch2_btree_insert_keys() only
1213          * does the lookup once and thus expects the keys to be in sorted order
1214          * so we have to make sure the new keys are correctly ordered with
1215          * respect to the deleted keys added in the previous loop
1216          */
1217         for (i = 0; i < nr_new_nodes; i++)
1218                 bch2_keylist_add_in_order(&keylist, &new_nodes[i]->key);
1219
1220         /* Insert the newly coalesced nodes */
1221         bch2_btree_insert_node(as, parent, iter, &keylist, 0);
1222
1223         BUG_ON(!bch2_keylist_empty(&keylist));
1224
1225         BUG_ON(iter->l[old_nodes[0]->c.level].b != old_nodes[0]);
1226
1227         bch2_btree_iter_node_replace(iter, new_nodes[0]);
1228
1229         for (i = 0; i < nr_new_nodes; i++)
1230                 bch2_btree_update_get_open_buckets(as, new_nodes[i]);
1231
1232         /* Free the old nodes and update our sliding window */
1233         for (i = 0; i < nr_old_nodes; i++) {
1234                 bch2_btree_node_free_inmem(c, old_nodes[i], iter);
1235
1236                 /*
1237                  * the index update might have triggered a split, in which case
1238                  * the nodes we coalesced - the new nodes we just created -
1239                  * might not be sibling nodes anymore - don't add them to the
1240                  * sliding window (except the first):
1241                  */
1242                 if (!i) {
1243                         old_nodes[i] = new_nodes[i];
1244                 } else {
1245                         old_nodes[i] = NULL;
1246                 }
1247         }
1248
1249         for (i = 0; i < nr_new_nodes; i++)
1250                 six_unlock_intent(&new_nodes[i]->c.lock);
1251
1252         bch2_btree_update_done(as);
1253         bch2_keylist_free(&keylist, NULL);
1254 }
1255
1256 static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id)
1257 {
1258         struct btree_trans trans;
1259         struct btree_iter *iter;
1260         struct btree *b;
1261         bool kthread = (current->flags & PF_KTHREAD) != 0;
1262         unsigned i;
1263
1264         /* Sliding window of adjacent btree nodes */
1265         struct btree *merge[GC_MERGE_NODES];
1266         u32 lock_seq[GC_MERGE_NODES];
1267
1268         bch2_trans_init(&trans, c, 0, 0);
1269
1270         /*
1271          * XXX: We don't have a good way of positively matching on sibling nodes
1272          * that have the same parent - this code works by handling the cases
1273          * where they might not have the same parent, and is thus fragile. Ugh.
1274          *
1275          * Perhaps redo this to use multiple linked iterators?
1276          */
1277         memset(merge, 0, sizeof(merge));
1278
1279         __for_each_btree_node(&trans, iter, btree_id, POS_MIN,
1280                               BTREE_MAX_DEPTH, 0,
1281                               BTREE_ITER_PREFETCH, b) {
1282                 memmove(merge + 1, merge,
1283                         sizeof(merge) - sizeof(merge[0]));
1284                 memmove(lock_seq + 1, lock_seq,
1285                         sizeof(lock_seq) - sizeof(lock_seq[0]));
1286
1287                 merge[0] = b;
1288
1289                 for (i = 1; i < GC_MERGE_NODES; i++) {
1290                         if (!merge[i] ||
1291                             !six_relock_intent(&merge[i]->c.lock, lock_seq[i]))
1292                                 break;
1293
1294                         if (merge[i]->c.level != merge[0]->c.level) {
1295                                 six_unlock_intent(&merge[i]->c.lock);
1296                                 break;
1297                         }
1298                 }
1299                 memset(merge + i, 0, (GC_MERGE_NODES - i) * sizeof(merge[0]));
1300
1301                 bch2_coalesce_nodes(c, iter, merge);
1302
1303                 for (i = 1; i < GC_MERGE_NODES && merge[i]; i++) {
1304                         lock_seq[i] = merge[i]->c.lock.state.seq;
1305                         six_unlock_intent(&merge[i]->c.lock);
1306                 }
1307
1308                 lock_seq[0] = merge[0]->c.lock.state.seq;
1309
1310                 if (kthread && kthread_should_stop()) {
1311                         bch2_trans_exit(&trans);
1312                         return -ESHUTDOWN;
1313                 }
1314
1315                 bch2_trans_cond_resched(&trans);
1316
1317                 /*
1318                  * If the parent node wasn't relocked, it might have been split
1319                  * and the nodes in our sliding window might not have the same
1320                  * parent anymore - blow away the sliding window:
1321                  */
1322                 if (btree_iter_node(iter, iter->level + 1) &&
1323                     !btree_node_intent_locked(iter, iter->level + 1))
1324                         memset(merge + 1, 0,
1325                                (GC_MERGE_NODES - 1) * sizeof(merge[0]));
1326         }
1327         return bch2_trans_exit(&trans);
1328 }
1329
1330 /**
1331  * bch_coalesce - coalesce adjacent nodes with low occupancy
1332  */
1333 void bch2_coalesce(struct bch_fs *c)
1334 {
1335         enum btree_id id;
1336
1337         down_read(&c->gc_lock);
1338         trace_gc_coalesce_start(c);
1339
1340         for (id = 0; id < BTREE_ID_NR; id++) {
1341                 int ret = c->btree_roots[id].b
1342                         ? bch2_coalesce_btree(c, id)
1343                         : 0;
1344
1345                 if (ret) {
1346                         if (ret != -ESHUTDOWN)
1347                                 bch_err(c, "btree coalescing failed: %d", ret);
1348                         return;
1349                 }
1350         }
1351
1352         trace_gc_coalesce_end(c);
1353         up_read(&c->gc_lock);
1354 }
1355
1356 static int bch2_gc_thread(void *arg)
1357 {
1358         struct bch_fs *c = arg;
1359         struct io_clock *clock = &c->io_clock[WRITE];
1360         unsigned long last = atomic_long_read(&clock->now);
1361         unsigned last_kick = atomic_read(&c->kick_gc);
1362         int ret;
1363
1364         set_freezable();
1365
1366         while (1) {
1367                 while (1) {
1368                         set_current_state(TASK_INTERRUPTIBLE);
1369
1370                         if (kthread_should_stop()) {
1371                                 __set_current_state(TASK_RUNNING);
1372                                 return 0;
1373                         }
1374
1375                         if (atomic_read(&c->kick_gc) != last_kick)
1376                                 break;
1377
1378                         if (c->btree_gc_periodic) {
1379                                 unsigned long next = last + c->capacity / 16;
1380
1381                                 if (atomic_long_read(&clock->now) >= next)
1382                                         break;
1383
1384                                 bch2_io_clock_schedule_timeout(clock, next);
1385                         } else {
1386                                 schedule();
1387                         }
1388
1389                         try_to_freeze();
1390                 }
1391                 __set_current_state(TASK_RUNNING);
1392
1393                 last = atomic_long_read(&clock->now);
1394                 last_kick = atomic_read(&c->kick_gc);
1395
1396                 /*
1397                  * Full gc is currently incompatible with btree key cache:
1398                  */
1399 #if 0
1400                 ret = bch2_gc(c, NULL, false, false);
1401 #else
1402                 ret = bch2_gc_gens(c);
1403 #endif
1404                 if (ret < 0)
1405                         bch_err(c, "btree gc failed: %i", ret);
1406
1407                 debug_check_no_locks_held();
1408         }
1409
1410         return 0;
1411 }
1412
1413 void bch2_gc_thread_stop(struct bch_fs *c)
1414 {
1415         struct task_struct *p;
1416
1417         p = c->gc_thread;
1418         c->gc_thread = NULL;
1419
1420         if (p) {
1421                 kthread_stop(p);
1422                 put_task_struct(p);
1423         }
1424 }
1425
1426 int bch2_gc_thread_start(struct bch_fs *c)
1427 {
1428         struct task_struct *p;
1429
1430         BUG_ON(c->gc_thread);
1431
1432         p = kthread_create(bch2_gc_thread, c, "bch-gc/%s", c->name);
1433         if (IS_ERR(p))
1434                 return PTR_ERR(p);
1435
1436         get_task_struct(p);
1437         c->gc_thread = p;
1438         wake_up_process(p);
1439         return 0;
1440 }