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 "reflink.h"
  27 #include "replicas.h"
  28 #include "super-io.h"
  29
  30 #include <linux/slab.h>
  31 #include <linux/bitops.h>
  32 #include <linux/freezer.h>
  33 #include <linux/kthread.h>
  34 #include <linux/preempt.h>
  35 #include <linux/rcupdate.h>
  36 #include <linux/sched/task.h>
  37 #include <trace/events/bcachefs.h>
  38
  39 #define DROP_THIS_NODE          10
  40 #define DROP_PREV_NODE          11
  41
  42 static inline void __gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
  43 {
  44         preempt_disable();
  45         write_seqcount_begin(&c->gc_pos_lock);
  46         c->gc_pos = new_pos;
  47         write_seqcount_end(&c->gc_pos_lock);
  48         preempt_enable();
  49 }
  50
  51 static inline void gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
  52 {
  53         BUG_ON(gc_pos_cmp(new_pos, c->gc_pos) <= 0);
  54         __gc_pos_set(c, new_pos);
  55 }
  56
  57 /*
  58  * Missing: if an interior btree node is empty, we need to do something -
  59  * perhaps just kill it
  60  */
  61 static int bch2_gc_check_topology(struct bch_fs *c,
  62                                   struct btree *b,
  63                                   struct bkey_buf *prev,
  64                                   struct bkey_buf cur,
  65                                   bool is_last)
  66 {
  67         struct bpos node_start  = b->data->min_key;
  68         struct bpos node_end    = b->data->max_key;
  69         struct bpos expected_start = bkey_deleted(&prev->k->k)
  70                 ? node_start
  71                 : bpos_successor(prev->k->k.p);
  72         char buf1[200], buf2[200];
  73         int ret = 0;
  74
  75         if (cur.k->k.type == KEY_TYPE_btree_ptr_v2) {
  76                 struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(cur.k);
  77
  78                 if (bkey_deleted(&prev->k->k)) {
  79                         struct printbuf out = PBUF(buf1);
  80                         pr_buf(&out, "start of node: ");
  81                         bch2_bpos_to_text(&out, node_start);
  82                 } else {
  83                         bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev->k));
  84                 }
  85
  86                 if (bpos_cmp(expected_start, bp->v.min_key)) {
  87                         bch2_topology_error(c);
  88
  89                         if (fsck_err(c, "btree node with incorrect min_key at btree %s level %u:\n"
  90                                      "  prev %s\n"
  91                                      "  cur %s",
  92                                      bch2_btree_ids[b->c.btree_id], b->c.level,
  93                                      buf1,
  94                                      (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(cur.k)), buf2))) {
  95                                 bch_info(c, "Halting mark and sweep to start topology repair pass");
  96                                 return FSCK_ERR_START_TOPOLOGY_REPAIR;
  97                         } else {
  98                                 set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags);
  99                         }
 100                 }
 101         }
 102
 103         if (is_last && bpos_cmp(cur.k->k.p, node_end)) {
 104                 bch2_topology_error(c);
 105
 106                 if (fsck_err(c, "btree node with incorrect max_key at btree %s level %u:\n"
 107                              "  %s\n"
 108                              "  expected %s",
 109                              bch2_btree_ids[b->c.btree_id], b->c.level,
 110                              (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(cur.k)), buf1),
 111                              (bch2_bpos_to_text(&PBUF(buf2), node_end), buf2))) {
 112                         bch_info(c, "Halting mark and sweep to start topology repair pass");
 113                         return FSCK_ERR_START_TOPOLOGY_REPAIR;
 114                 } else {
 115                         set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags);
 116                 }
 117         }
 118
 119         bch2_bkey_buf_copy(prev, c, cur.k);
 120 fsck_err:
 121         return ret;
 122 }
 123
 124 static void btree_ptr_to_v2(struct btree *b, struct bkey_i_btree_ptr_v2 *dst)
 125 {
 126         switch (b->key.k.type) {
 127         case KEY_TYPE_btree_ptr: {
 128                 struct bkey_i_btree_ptr *src = bkey_i_to_btree_ptr(&b->key);
 129
 130                 dst->k.p                = src->k.p;
 131                 dst->v.mem_ptr          = 0;
 132                 dst->v.seq              = b->data->keys.seq;
 133                 dst->v.sectors_written  = 0;
 134                 dst->v.flags            = 0;
 135                 dst->v.min_key          = b->data->min_key;
 136                 set_bkey_val_bytes(&dst->k, sizeof(dst->v) + bkey_val_bytes(&src->k));
 137                 memcpy(dst->v.start, src->v.start, bkey_val_bytes(&src->k));
 138                 break;
 139         }
 140         case KEY_TYPE_btree_ptr_v2:
 141                 bkey_copy(&dst->k_i, &b->key);
 142                 break;
 143         default:
 144                 BUG();
 145         }
 146 }
 147
 148 static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min)
 149 {
 150         struct bkey_i_btree_ptr_v2 *new;
 151         int ret;
 152
 153         new = kmalloc(BKEY_BTREE_PTR_U64s_MAX * sizeof(u64), GFP_KERNEL);
 154         if (!new)
 155                 return -ENOMEM;
 156
 157         btree_ptr_to_v2(b, new);
 158         b->data->min_key        = new_min;
 159         new->v.min_key          = new_min;
 160         SET_BTREE_PTR_RANGE_UPDATED(&new->v, true);
 161
 162         ret = bch2_journal_key_insert(c, b->c.btree_id, b->c.level + 1, &new->k_i);
 163         if (ret) {
 164                 kfree(new);
 165                 return ret;
 166         }
 167
 168         bch2_btree_node_drop_keys_outside_node(b);
 169
 170         return 0;
 171 }
 172
 173 static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max)
 174 {
 175         struct bkey_i_btree_ptr_v2 *new;
 176         int ret;
 177
 178         ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level + 1, b->key.k.p);
 179         if (ret)
 180                 return ret;
 181
 182         new = kmalloc(BKEY_BTREE_PTR_U64s_MAX * sizeof(u64), GFP_KERNEL);
 183         if (!new)
 184                 return -ENOMEM;
 185
 186         btree_ptr_to_v2(b, new);
 187         b->data->max_key        = new_max;
 188         new->k.p                = new_max;
 189         SET_BTREE_PTR_RANGE_UPDATED(&new->v, true);
 190
 191         ret = bch2_journal_key_insert(c, b->c.btree_id, b->c.level + 1, &new->k_i);
 192         if (ret) {
 193                 kfree(new);
 194                 return ret;
 195         }
 196
 197         bch2_btree_node_drop_keys_outside_node(b);
 198
 199         mutex_lock(&c->btree_cache.lock);
 200         bch2_btree_node_hash_remove(&c->btree_cache, b);
 201
 202         bkey_copy(&b->key, &new->k_i);
 203         ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
 204         BUG_ON(ret);
 205         mutex_unlock(&c->btree_cache.lock);
 206         return 0;
 207 }
 208
 209 static int btree_repair_node_start(struct bch_fs *c, struct btree *b,
 210                                    struct btree *prev, struct btree *cur)
 211 {
 212         struct bpos expected_start = !prev
 213                 ? b->data->min_key
 214                 : bpos_successor(prev->key.k.p);
 215         char buf1[200], buf2[200];
 216         int ret = 0;
 217
 218         if (!prev) {
 219                 struct printbuf out = PBUF(buf1);
 220                 pr_buf(&out, "start of node: ");
 221                 bch2_bpos_to_text(&out, b->data->min_key);
 222         } else {
 223                 bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(&prev->key));
 224         }
 225
 226         if (mustfix_fsck_err_on(bpos_cmp(expected_start, cur->data->min_key), c,
 227                         "btree node with incorrect min_key at btree %s level %u:\n"
 228                         "  prev %s\n"
 229                         "  cur %s",
 230                         bch2_btree_ids[b->c.btree_id], b->c.level,
 231                         buf1,
 232                         (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(&cur->key)), buf2))) {
 233                 if (prev &&
 234                     bpos_cmp(expected_start, cur->data->min_key) > 0 &&
 235                     BTREE_NODE_SEQ(cur->data) > BTREE_NODE_SEQ(prev->data)) {
 236                         if (bkey_cmp(prev->data->min_key,
 237                                      cur->data->min_key) <= 0)
 238                                 return DROP_PREV_NODE;
 239
 240                         ret = set_node_max(c, prev,
 241                                            bpos_predecessor(cur->data->min_key));
 242                 } else {
 243                         if (bkey_cmp(expected_start, b->data->max_key) >= 0)
 244                                 return DROP_THIS_NODE;
 245
 246                         ret = set_node_min(c, cur, expected_start);
 247                 }
 248                 if (ret)
 249                         return ret;
 250         }
 251 fsck_err:
 252         return ret;
 253 }
 254
 255 static int btree_repair_node_end(struct bch_fs *c, struct btree *b,
 256                                  struct btree *child)
 257 {
 258         char buf1[200], buf2[200];
 259         int ret = 0;
 260
 261         if (mustfix_fsck_err_on(bpos_cmp(child->key.k.p, b->key.k.p), c,
 262                         "btree node with incorrect max_key at btree %s level %u:\n"
 263                         "  %s\n"
 264                         "  expected %s",
 265                         bch2_btree_ids[b->c.btree_id], b->c.level,
 266                         (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(&child->key)), buf1),
 267                         (bch2_bpos_to_text(&PBUF(buf2), b->key.k.p), buf2))) {
 268                 ret = set_node_max(c, child, b->key.k.p);
 269                 if (ret)
 270                         return ret;
 271         }
 272 fsck_err:
 273         return ret;
 274 }
 275
 276 static int bch2_btree_repair_topology_recurse(struct bch_fs *c, struct btree *b)
 277 {
 278         struct btree_and_journal_iter iter;
 279         struct bkey_s_c k;
 280         struct bkey_buf prev_k, cur_k;
 281         struct btree *prev = NULL, *cur = NULL;
 282         bool have_child, dropped_children = false;
 283         char buf[200];
 284         int ret = 0;
 285
 286         if (!b->c.level)
 287                 return 0;
 288 again:
 289         have_child = dropped_children = false;
 290         bch2_bkey_buf_init(&prev_k);
 291         bch2_bkey_buf_init(&cur_k);
 292         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 293
 294         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 295                 bch2_btree_and_journal_iter_advance(&iter);
 296                 bch2_bkey_buf_reassemble(&cur_k, c, k);
 297
 298                 cur = bch2_btree_node_get_noiter(c, cur_k.k,
 299                                         b->c.btree_id, b->c.level - 1,
 300                                         false);
 301                 ret = PTR_ERR_OR_ZERO(cur);
 302
 303                 if (mustfix_fsck_err_on(ret == -EIO, c,
 304                                 "Unreadable btree node at btree %s level %u:\n"
 305                                 "  %s",
 306                                 bch2_btree_ids[b->c.btree_id],
 307                                 b->c.level - 1,
 308                                 (bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(cur_k.k)), buf))) {
 309                         bch2_btree_node_evict(c, cur_k.k);
 310                         ret = bch2_journal_key_delete(c, b->c.btree_id,
 311                                                       b->c.level, cur_k.k->k.p);
 312                         if (ret)
 313                                 goto err;
 314                         continue;
 315                 }
 316
 317                 if (ret) {
 318                         bch_err(c, "%s: error %i getting btree node",
 319                                 __func__, ret);
 320                         break;
 321                 }
 322
 323                 ret = btree_repair_node_start(c, b, prev, cur);
 324
 325                 if (ret == DROP_THIS_NODE) {
 326                         six_unlock_read(&cur->c.lock);
 327                         bch2_btree_node_evict(c, cur_k.k);
 328                         ret = bch2_journal_key_delete(c, b->c.btree_id,
 329                                                       b->c.level, cur_k.k->k.p);
 330                         if (ret)
 331                                 goto err;
 332                         continue;
 333                 }
 334
 335                 if (prev)
 336                         six_unlock_read(&prev->c.lock);
 337
 338                 if (ret == DROP_PREV_NODE) {
 339                         bch2_btree_node_evict(c, prev_k.k);
 340                         ret = bch2_journal_key_delete(c, b->c.btree_id,
 341                                                       b->c.level, prev_k.k->k.p);
 342                         if (ret)
 343                                 goto err;
 344                         goto again;
 345                 } else if (ret)
 346                         break;
 347
 348                 prev = cur;
 349                 cur = NULL;
 350                 bch2_bkey_buf_copy(&prev_k, c, cur_k.k);
 351         }
 352
 353         if (!ret && !IS_ERR_OR_NULL(prev)) {
 354                 BUG_ON(cur);
 355                 ret = btree_repair_node_end(c, b, prev);
 356         }
 357
 358         if (!IS_ERR_OR_NULL(prev))
 359                 six_unlock_read(&prev->c.lock);
 360         prev = NULL;
 361         if (!IS_ERR_OR_NULL(cur))
 362                 six_unlock_read(&cur->c.lock);
 363         cur = NULL;
 364
 365         if (ret)
 366                 goto err;
 367
 368         bch2_btree_and_journal_iter_exit(&iter);
 369         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 370
 371         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 372                 bch2_bkey_buf_reassemble(&cur_k, c, k);
 373                 bch2_btree_and_journal_iter_advance(&iter);
 374
 375                 cur = bch2_btree_node_get_noiter(c, cur_k.k,
 376                                         b->c.btree_id, b->c.level - 1,
 377                                         false);
 378                 ret = PTR_ERR_OR_ZERO(cur);
 379
 380                 if (ret) {
 381                         bch_err(c, "%s: error %i getting btree node",
 382                                 __func__, ret);
 383                         goto err;
 384                 }
 385
 386                 ret = bch2_btree_repair_topology_recurse(c, cur);
 387                 six_unlock_read(&cur->c.lock);
 388                 cur = NULL;
 389
 390                 if (ret == DROP_THIS_NODE) {
 391                         bch2_btree_node_evict(c, cur_k.k);
 392                         ret = bch2_journal_key_delete(c, b->c.btree_id,
 393                                                       b->c.level, cur_k.k->k.p);
 394                         dropped_children = true;
 395                 }
 396
 397                 if (ret)
 398                         goto err;
 399
 400                 have_child = true;
 401         }
 402
 403         if (mustfix_fsck_err_on(!have_child, c,
 404                         "empty interior btree node at btree %s level %u\n"
 405                         "  %s",
 406                         bch2_btree_ids[b->c.btree_id],
 407                         b->c.level,
 408                         (bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(&b->key)), buf)))
 409                 ret = DROP_THIS_NODE;
 410 err:
 411 fsck_err:
 412         if (!IS_ERR_OR_NULL(prev))
 413                 six_unlock_read(&prev->c.lock);
 414         if (!IS_ERR_OR_NULL(cur))
 415                 six_unlock_read(&cur->c.lock);
 416
 417         bch2_btree_and_journal_iter_exit(&iter);
 418         bch2_bkey_buf_exit(&prev_k, c);
 419         bch2_bkey_buf_exit(&cur_k, c);
 420
 421         if (!ret && dropped_children)
 422                 goto again;
 423
 424         return ret;
 425 }
 426
 427 static int bch2_repair_topology(struct bch_fs *c)
 428 {
 429         struct btree *b;
 430         unsigned i;
 431         int ret = 0;
 432
 433         for (i = 0; i < BTREE_ID_NR && !ret; i++) {
 434                 b = c->btree_roots[i].b;
 435                 if (btree_node_fake(b))
 436                         continue;
 437
 438                 six_lock_read(&b->c.lock, NULL, NULL);
 439                 ret = bch2_btree_repair_topology_recurse(c, b);
 440                 six_unlock_read(&b->c.lock);
 441
 442                 if (ret == DROP_THIS_NODE) {
 443                         bch_err(c, "empty btree root - repair unimplemented");
 444                         ret = FSCK_ERR_EXIT;
 445                 }
 446         }
 447
 448         return ret;
 449 }
 450
 451 static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id,
 452                                unsigned level, bool is_root,
 453                                struct bkey_s_c *k)
 454 {
 455         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(*k);
 456         const union bch_extent_entry *entry;
 457         struct extent_ptr_decoded p = { 0 };
 458         bool do_update = false;
 459         char buf[200];
 460         int ret = 0;
 461
 462         bkey_for_each_ptr_decode(k->k, ptrs, p, entry) {
 463                 struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 464                 struct bucket *g = PTR_BUCKET(ca, &p.ptr, true);
 465                 struct bucket *g2 = PTR_BUCKET(ca, &p.ptr, false);
 466                 enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, &entry->ptr);
 467
 468                 if (fsck_err_on(g->mark.data_type &&
 469                                 g->mark.data_type != data_type, c,
 470                                 "bucket %u:%zu different types of data in same bucket: %s, %s\n"
 471                                 "while marking %s",
 472                                 p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr),
 473                                 bch2_data_types[g->mark.data_type],
 474                                 bch2_data_types[data_type],
 475                                 (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf))) {
 476                         if (data_type == BCH_DATA_btree) {
 477                                 g2->_mark.data_type = g->_mark.data_type = data_type;
 478                                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);
 479                         } else {
 480                                 do_update = true;
 481                         }
 482                 }
 483
 484                 if (fsck_err_on(!g->gen_valid, c,
 485                                 "bucket %u:%zu data type %s ptr gen %u missing in alloc btree\n"
 486                                 "while marking %s",
 487                                 p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr),
 488                                 bch2_data_types[ptr_data_type(k->k, &p.ptr)],
 489                                 p.ptr.gen,
 490                                 (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf))) {
 491                         if (!p.ptr.cached) {
 492                                 g2->_mark.gen   = g->_mark.gen          = p.ptr.gen;
 493                                 g2->gen_valid   = g->gen_valid          = true;
 494                                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);
 495                         } else {
 496                                 do_update = true;
 497                         }
 498                 }
 499
 500                 if (fsck_err_on(gen_cmp(p.ptr.gen, g->mark.gen) > 0, c,
 501                                 "bucket %u:%zu data type %s ptr gen in the future: %u > %u\n"
 502                                 "while marking %s",
 503                                 p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr),
 504                                 bch2_data_types[ptr_data_type(k->k, &p.ptr)],
 505                                 p.ptr.gen, g->mark.gen,
 506                                 (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf))) {
 507                         if (!p.ptr.cached) {
 508                                 g2->_mark.gen   = g->_mark.gen  = p.ptr.gen;
 509                                 g2->gen_valid   = g->gen_valid  = true;
 510                                 g2->_mark.data_type             = 0;
 511                                 g2->_mark.dirty_sectors         = 0;
 512                                 g2->_mark.cached_sectors        = 0;
 513                                 set_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags);
 514                                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);
 515                         } else {
 516                                 do_update = true;
 517                         }
 518                 }
 519
 520                 if (fsck_err_on(!p.ptr.cached &&
 521                                 gen_cmp(p.ptr.gen, g->mark.gen) < 0, c,
 522                                 "bucket %u:%zu data type %s stale dirty ptr: %u < %u\n"
 523                                 "while marking %s",
 524                                 p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr),
 525                                 bch2_data_types[ptr_data_type(k->k, &p.ptr)],
 526                                 p.ptr.gen, g->mark.gen,
 527                                 (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf)))
 528                         do_update = true;
 529
 530                 if (p.has_ec) {
 531                         struct stripe *m = genradix_ptr(&c->stripes[true], p.ec.idx);
 532
 533                         if (fsck_err_on(!m || !m->alive, c,
 534                                         "pointer to nonexistent stripe %llu\n"
 535                                         "while marking %s",
 536                                         (u64) p.ec.idx,
 537                                         (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf)))
 538                                 do_update = true;
 539
 540                         if (fsck_err_on(!bch2_ptr_matches_stripe_m(m, p), c,
 541                                         "pointer does not match stripe %llu\n"
 542                                         "while marking %s",
 543                                         (u64) p.ec.idx,
 544                                         (bch2_bkey_val_to_text(&PBUF(buf), c, *k), buf)))
 545                                 do_update = true;
 546                 }
 547         }
 548
 549         if (do_update) {
 550                 struct bkey_ptrs ptrs;
 551                 union bch_extent_entry *entry;
 552                 struct bch_extent_ptr *ptr;
 553                 struct bkey_i *new;
 554
 555                 if (is_root) {
 556                         bch_err(c, "cannot update btree roots yet");
 557                         return -EINVAL;
 558                 }
 559
 560                 new = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
 561                 if (!new) {
 562                         bch_err(c, "%s: error allocating new key", __func__);
 563                         return -ENOMEM;
 564                 }
 565
 566                 bkey_reassemble(new, *k);
 567
 568                 if (level) {
 569                         /*
 570                          * We don't want to drop btree node pointers - if the
 571                          * btree node isn't there anymore, the read path will
 572                          * sort it out:
 573                          */
 574                         ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 575                         bkey_for_each_ptr(ptrs, ptr) {
 576                                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 577                                 struct bucket *g = PTR_BUCKET(ca, ptr, true);
 578
 579                                 ptr->gen = g->mark.gen;
 580                         }
 581                 } else {
 582                         bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, ({
 583                                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 584                                 struct bucket *g = PTR_BUCKET(ca, ptr, true);
 585                                 enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, ptr);
 586
 587                                 (ptr->cached &&
 588                                  (!g->gen_valid || gen_cmp(ptr->gen, g->mark.gen) > 0)) ||
 589                                 (!ptr->cached &&
 590                                  gen_cmp(ptr->gen, g->mark.gen) < 0) ||
 591                                 (g->mark.data_type &&
 592                                  g->mark.data_type != data_type);
 593                         }));
 594 again:
 595                         ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 596                         bkey_extent_entry_for_each(ptrs, entry) {
 597                                 if (extent_entry_type(entry) == BCH_EXTENT_ENTRY_stripe_ptr) {
 598                                         struct stripe *m = genradix_ptr(&c->stripes[true],
 599                                                                         entry->stripe_ptr.idx);
 600                                         union bch_extent_entry *next_ptr;
 601
 602                                         bkey_extent_entry_for_each_from(ptrs, next_ptr, entry)
 603                                                 if (extent_entry_type(next_ptr) == BCH_EXTENT_ENTRY_ptr)
 604                                                         goto found;
 605                                         next_ptr = NULL;
 606 found:
 607                                         if (!next_ptr) {
 608                                                 bch_err(c, "aieee, found stripe ptr with no data ptr");
 609                                                 continue;
 610                                         }
 611
 612                                         if (!m || !m->alive ||
 613                                             !__bch2_ptr_matches_stripe(&m->ptrs[entry->stripe_ptr.block],
 614                                                                        &next_ptr->ptr,
 615                                                                        m->sectors)) {
 616                                                 bch2_bkey_extent_entry_drop(new, entry);
 617                                                 goto again;
 618                                         }
 619                                 }
 620                         }
 621                 }
 622
 623                 ret = bch2_journal_key_insert(c, btree_id, level, new);
 624                 if (ret)
 625                         kfree(new);
 626                 else
 627                         *k = bkey_i_to_s_c(new);
 628         }
 629 fsck_err:
 630         return ret;
 631 }
 632
 633 /* marking of btree keys/nodes: */
 634
 635 static int bch2_gc_mark_key(struct bch_fs *c, enum btree_id btree_id,
 636                             unsigned level, bool is_root,
 637                             struct bkey_s_c *k,
 638                             u8 *max_stale, bool initial)
 639 {
 640         struct bkey_ptrs_c ptrs;
 641         const struct bch_extent_ptr *ptr;
 642         unsigned flags =
 643                 BTREE_TRIGGER_GC|
 644                 (initial ? BTREE_TRIGGER_NOATOMIC : 0);
 645         int ret = 0;
 646
 647         if (initial) {
 648                 BUG_ON(bch2_journal_seq_verify &&
 649                        k->k->version.lo > journal_cur_seq(&c->journal));
 650
 651                 ret = bch2_check_fix_ptrs(c, btree_id, level, is_root, k);
 652                 if (ret)
 653                         goto err;
 654
 655                 if (fsck_err_on(k->k->version.lo > atomic64_read(&c->key_version), c,
 656                                 "key version number higher than recorded: %llu > %llu",
 657                                 k->k->version.lo,
 658                                 atomic64_read(&c->key_version)))
 659                         atomic64_set(&c->key_version, k->k->version.lo);
 660
 661                 if (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
 662                     fsck_err_on(!bch2_bkey_replicas_marked(c, *k), c,
 663                                 "superblock not marked as containing replicas (type %u)",
 664                                 k->k->type)) {
 665                         ret = bch2_mark_bkey_replicas(c, *k);
 666                         if (ret) {
 667                                 bch_err(c, "error marking bkey replicas: %i", ret);
 668                                 goto err;
 669                         }
 670                 }
 671         }
 672
 673         ptrs = bch2_bkey_ptrs_c(*k);
 674         bkey_for_each_ptr(ptrs, ptr) {
 675                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 676                 struct bucket *g = PTR_BUCKET(ca, ptr, true);
 677
 678                 if (gen_after(g->oldest_gen, ptr->gen))
 679                         g->oldest_gen = ptr->gen;
 680
 681                 *max_stale = max(*max_stale, ptr_stale(ca, ptr));
 682         }
 683
 684         bch2_mark_key(c, *k, 0, k->k->size, NULL, 0, flags);
 685 fsck_err:
 686 err:
 687         if (ret)
 688                 bch_err(c, "%s: ret %i", __func__, ret);
 689         return ret;
 690 }
 691
 692 static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale,
 693                               bool initial)
 694 {
 695         struct btree_node_iter iter;
 696         struct bkey unpacked;
 697         struct bkey_s_c k;
 698         struct bkey_buf prev, cur;
 699         int ret = 0;
 700
 701         *max_stale = 0;
 702
 703         if (!btree_node_type_needs_gc(btree_node_type(b)))
 704                 return 0;
 705
 706         bch2_btree_node_iter_init_from_start(&iter, b);
 707         bch2_bkey_buf_init(&prev);
 708         bch2_bkey_buf_init(&cur);
 709         bkey_init(&prev.k->k);
 710
 711         while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
 712                 ret = bch2_gc_mark_key(c, b->c.btree_id, b->c.level, false,
 713                                        &k, max_stale, initial);
 714                 if (ret)
 715                         break;
 716
 717                 bch2_btree_node_iter_advance(&iter, b);
 718
 719                 if (b->c.level) {
 720                         bch2_bkey_buf_reassemble(&cur, c, k);
 721
 722                         ret = bch2_gc_check_topology(c, b, &prev, cur,
 723                                         bch2_btree_node_iter_end(&iter));
 724                         if (ret)
 725                                 break;
 726                 }
 727         }
 728
 729         bch2_bkey_buf_exit(&cur, c);
 730         bch2_bkey_buf_exit(&prev, c);
 731         return ret;
 732 }
 733
 734 static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
 735                          bool initial, bool metadata_only)
 736 {
 737         struct btree_trans trans;
 738         struct btree_iter *iter;
 739         struct btree *b;
 740         unsigned depth = metadata_only                  ? 1
 741                 : bch2_expensive_debug_checks           ? 0
 742                 : !btree_node_type_needs_gc(btree_id)   ? 1
 743                 : 0;
 744         u8 max_stale = 0;
 745         int ret = 0;
 746
 747         bch2_trans_init(&trans, c, 0, 0);
 748
 749         gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
 750
 751         __for_each_btree_node(&trans, iter, btree_id, POS_MIN,
 752                               0, depth, BTREE_ITER_PREFETCH, b) {
 753                 bch2_verify_btree_nr_keys(b);
 754
 755                 gc_pos_set(c, gc_pos_btree_node(b));
 756
 757                 ret = btree_gc_mark_node(c, b, &max_stale, initial);
 758                 if (ret)
 759                         break;
 760
 761                 if (!initial) {
 762                         if (max_stale > 64)
 763                                 bch2_btree_node_rewrite(c, iter,
 764                                                 b->data->keys.seq,
 765                                                 BTREE_INSERT_NOWAIT|
 766                                                 BTREE_INSERT_GC_LOCK_HELD);
 767                         else if (!bch2_btree_gc_rewrite_disabled &&
 768                                  (bch2_btree_gc_always_rewrite || max_stale > 16))
 769                                 bch2_btree_node_rewrite(c, iter,
 770                                                 b->data->keys.seq,
 771                                                 BTREE_INSERT_NOWAIT|
 772                                                 BTREE_INSERT_GC_LOCK_HELD);
 773                 }
 774
 775                 bch2_trans_cond_resched(&trans);
 776         }
 777         bch2_trans_iter_put(&trans, iter);
 778
 779         ret = bch2_trans_exit(&trans) ?: ret;
 780         if (ret)
 781                 return ret;
 782
 783         mutex_lock(&c->btree_root_lock);
 784         b = c->btree_roots[btree_id].b;
 785         if (!btree_node_fake(b)) {
 786                 struct bkey_s_c k = bkey_i_to_s_c(&b->key);
 787
 788                 ret = bch2_gc_mark_key(c, b->c.btree_id, b->c.level, true,
 789                                        &k, &max_stale, initial);
 790         }
 791         gc_pos_set(c, gc_pos_btree_root(b->c.btree_id));
 792         mutex_unlock(&c->btree_root_lock);
 793
 794         return ret;
 795 }
 796
 797 static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b,
 798                                       unsigned target_depth)
 799 {
 800         struct btree_and_journal_iter iter;
 801         struct bkey_s_c k;
 802         struct bkey_buf cur, prev;
 803         u8 max_stale = 0;
 804         char buf[200];
 805         int ret = 0;
 806
 807         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 808         bch2_bkey_buf_init(&prev);
 809         bch2_bkey_buf_init(&cur);
 810         bkey_init(&prev.k->k);
 811
 812         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 813                 BUG_ON(bpos_cmp(k.k->p, b->data->min_key) < 0);
 814                 BUG_ON(bpos_cmp(k.k->p, b->data->max_key) > 0);
 815
 816                 ret = bch2_gc_mark_key(c, b->c.btree_id, b->c.level, false,
 817                                        &k, &max_stale, true);
 818                 if (ret) {
 819                         bch_err(c, "%s: error %i from bch2_gc_mark_key", __func__, ret);
 820                         goto fsck_err;
 821                 }
 822
 823                 if (b->c.level) {
 824                         bch2_bkey_buf_reassemble(&cur, c, k);
 825                         k = bkey_i_to_s_c(cur.k);
 826
 827                         bch2_btree_and_journal_iter_advance(&iter);
 828
 829                         ret = bch2_gc_check_topology(c, b,
 830                                         &prev, cur,
 831                                         !bch2_btree_and_journal_iter_peek(&iter).k);
 832                         if (ret)
 833                                 goto fsck_err;
 834                 } else {
 835                         bch2_btree_and_journal_iter_advance(&iter);
 836                 }
 837         }
 838
 839         if (b->c.level > target_depth) {
 840                 bch2_btree_and_journal_iter_exit(&iter);
 841                 bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 842
 843                 while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 844                         struct btree *child;
 845
 846                         bch2_bkey_buf_reassemble(&cur, c, k);
 847                         bch2_btree_and_journal_iter_advance(&iter);
 848
 849                         child = bch2_btree_node_get_noiter(c, cur.k,
 850                                                 b->c.btree_id, b->c.level - 1,
 851                                                 false);
 852                         ret = PTR_ERR_OR_ZERO(child);
 853
 854                         if (ret == -EIO) {
 855                                 bch2_topology_error(c);
 856
 857                                 if (fsck_err(c, "Unreadable btree node at btree %s level %u:\n"
 858                                         "  %s",
 859                                         bch2_btree_ids[b->c.btree_id],
 860                                         b->c.level - 1,
 861                                         (bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(cur.k)), buf))) {
 862                                         ret = FSCK_ERR_START_TOPOLOGY_REPAIR;
 863                                         bch_info(c, "Halting mark and sweep to start topology repair pass");
 864                                         goto fsck_err;
 865                                 } else {
 866                                         /* Continue marking when opted to not
 867                                          * fix the error: */
 868                                         ret = 0;
 869                                         set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags);
 870                                         continue;
 871                                 }
 872                         } else if (ret) {
 873                                 bch_err(c, "%s: error %i getting btree node",
 874                                         __func__, ret);
 875                                 break;
 876                         }
 877
 878                         ret = bch2_gc_btree_init_recurse(c, child,
 879                                                          target_depth);
 880                         six_unlock_read(&child->c.lock);
 881
 882                         if (ret)
 883                                 break;
 884                 }
 885         }
 886 fsck_err:
 887         bch2_bkey_buf_exit(&cur, c);
 888         bch2_bkey_buf_exit(&prev, c);
 889         bch2_btree_and_journal_iter_exit(&iter);
 890         return ret;
 891 }
 892
 893 static int bch2_gc_btree_init(struct bch_fs *c,
 894                               enum btree_id btree_id,
 895                               bool metadata_only)
 896 {
 897         struct btree *b;
 898         unsigned target_depth = metadata_only           ? 1
 899                 : bch2_expensive_debug_checks           ? 0
 900                 : !btree_node_type_needs_gc(btree_id)   ? 1
 901                 : 0;
 902         u8 max_stale = 0;
 903         char buf[100];
 904         int ret = 0;
 905
 906         b = c->btree_roots[btree_id].b;
 907
 908         if (btree_node_fake(b))
 909                 return 0;
 910
 911         six_lock_read(&b->c.lock, NULL, NULL);
 912         if (mustfix_fsck_err_on(bpos_cmp(b->data->min_key, POS_MIN), c,
 913                         "btree root with incorrect min_key: %s",
 914                         (bch2_bpos_to_text(&PBUF(buf), b->data->min_key), buf))) {
 915                 bch_err(c, "repair unimplemented");
 916                 ret = FSCK_ERR_EXIT;
 917                 goto fsck_err;
 918         }
 919
 920         if (mustfix_fsck_err_on(bpos_cmp(b->data->max_key, POS_MAX), c,
 921                         "btree root with incorrect max_key: %s",
 922                         (bch2_bpos_to_text(&PBUF(buf), b->data->max_key), buf))) {
 923                 bch_err(c, "repair unimplemented");
 924                 ret = FSCK_ERR_EXIT;
 925                 goto fsck_err;
 926         }
 927
 928         if (b->c.level >= target_depth)
 929                 ret = bch2_gc_btree_init_recurse(c, b, target_depth);
 930
 931         if (!ret) {
 932                 struct bkey_s_c k = bkey_i_to_s_c(&b->key);
 933
 934                 ret = bch2_gc_mark_key(c, b->c.btree_id, b->c.level, true,
 935                                        &k, &max_stale, true);
 936         }
 937 fsck_err:
 938         six_unlock_read(&b->c.lock);
 939
 940         if (ret < 0)
 941                 bch_err(c, "%s: ret %i", __func__, ret);
 942         return ret;
 943 }
 944
 945 static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 946 {
 947         return  (int) btree_id_to_gc_phase(l) -
 948                 (int) btree_id_to_gc_phase(r);
 949 }
 950
 951 static int bch2_gc_btrees(struct bch_fs *c, bool initial, bool metadata_only)
 952 {
 953         enum btree_id ids[BTREE_ID_NR];
 954         unsigned i;
 955         int ret = 0;
 956
 957         for (i = 0; i < BTREE_ID_NR; i++)
 958                 ids[i] = i;
 959         bubble_sort(ids, BTREE_ID_NR, btree_id_gc_phase_cmp);
 960
 961         for (i = 0; i < BTREE_ID_NR && !ret; i++)
 962                 ret = initial
 963                         ? bch2_gc_btree_init(c, ids[i], metadata_only)
 964                         : bch2_gc_btree(c, ids[i], initial, metadata_only);
 965
 966         if (ret < 0)
 967                 bch_err(c, "%s: ret %i", __func__, ret);
 968         return ret;
 969 }
 970
 971 static void mark_metadata_sectors(struct bch_fs *c, struct bch_dev *ca,
 972                                   u64 start, u64 end,
 973                                   enum bch_data_type type,
 974                                   unsigned flags)
 975 {
 976         u64 b = sector_to_bucket(ca, start);
 977
 978         do {
 979                 unsigned sectors =
 980                         min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
 981
 982                 bch2_mark_metadata_bucket(c, ca, b, type, sectors,
 983                                           gc_phase(GC_PHASE_SB), flags);
 984                 b++;
 985                 start += sectors;
 986         } while (start < end);
 987 }
 988
 989 void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca,
 990                               unsigned flags)
 991 {
 992         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
 993         unsigned i;
 994         u64 b;
 995
 996         /*
 997          * This conditional is kind of gross, but we may be called from the
 998          * device add path, before the new device has actually been added to the
 999          * running filesystem:
1000          */
1001         if (c) {
1002                 lockdep_assert_held(&c->sb_lock);
1003                 percpu_down_read(&c->mark_lock);
1004         }
1005
1006         for (i = 0; i < layout->nr_superblocks; i++) {
1007                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1008
1009                 if (offset == BCH_SB_SECTOR)
1010                         mark_metadata_sectors(c, ca, 0, BCH_SB_SECTOR,
1011                                               BCH_DATA_sb, flags);
1012
1013                 mark_metadata_sectors(c, ca, offset,
1014                                       offset + (1 << layout->sb_max_size_bits),
1015                                       BCH_DATA_sb, flags);
1016         }
1017
1018         for (i = 0; i < ca->journal.nr; i++) {
1019                 b = ca->journal.buckets[i];
1020                 bch2_mark_metadata_bucket(c, ca, b, BCH_DATA_journal,
1021                                           ca->mi.bucket_size,
1022                                           gc_phase(GC_PHASE_SB), flags);
1023         }
1024
1025         if (c)
1026                 percpu_up_read(&c->mark_lock);
1027 }
1028
1029 static void bch2_mark_superblocks(struct bch_fs *c)
1030 {
1031         struct bch_dev *ca;
1032         unsigned i;
1033
1034         mutex_lock(&c->sb_lock);
1035         gc_pos_set(c, gc_phase(GC_PHASE_SB));
1036
1037         for_each_online_member(ca, c, i)
1038                 bch2_mark_dev_superblock(c, ca, BTREE_TRIGGER_GC);
1039         mutex_unlock(&c->sb_lock);
1040 }
1041
1042 #if 0
1043 /* Also see bch2_pending_btree_node_free_insert_done() */
1044 static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
1045 {
1046         struct btree_update *as;
1047         struct pending_btree_node_free *d;
1048
1049         mutex_lock(&c->btree_interior_update_lock);
1050         gc_pos_set(c, gc_phase(GC_PHASE_PENDING_DELETE));
1051
1052         for_each_pending_btree_node_free(c, as, d)
1053                 if (d->index_update_done)
1054                         bch2_mark_key(c, bkey_i_to_s_c(&d->key),
1055                                       0, 0, NULL, 0,
1056                                       BTREE_TRIGGER_GC);
1057
1058         mutex_unlock(&c->btree_interior_update_lock);
1059 }
1060 #endif
1061
1062 static void bch2_gc_free(struct bch_fs *c)
1063 {
1064         struct bch_dev *ca;
1065         unsigned i;
1066
1067         genradix_free(&c->stripes[1]);
1068
1069         for_each_member_device(ca, c, i) {
1070                 kvpfree(rcu_dereference_protected(ca->buckets[1], 1),
1071                         sizeof(struct bucket_array) +
1072                         ca->mi.nbuckets * sizeof(struct bucket));
1073                 ca->buckets[1] = NULL;
1074
1075                 free_percpu(ca->usage_gc);
1076                 ca->usage_gc = NULL;
1077         }
1078
1079         free_percpu(c->usage_gc);
1080         c->usage_gc = NULL;
1081 }
1082
1083 static int bch2_gc_done(struct bch_fs *c,
1084                         bool initial, bool metadata_only)
1085 {
1086         struct bch_dev *ca = NULL;
1087         bool verify = !metadata_only && (!initial ||
1088                        (c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)));
1089         unsigned i, dev;
1090         int ret = 0;
1091
1092 #define copy_field(_f, _msg, ...)                                       \
1093         if (dst->_f != src->_f) {                                       \
1094                 if (verify)                                             \
1095                         fsck_err(c, _msg ": got %llu, should be %llu"   \
1096                                 , ##__VA_ARGS__, dst->_f, src->_f);     \
1097                 dst->_f = src->_f;                                      \
1098                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
1099         }
1100 #define copy_stripe_field(_f, _msg, ...)                                \
1101         if (dst->_f != src->_f) {                                       \
1102                 if (verify)                                             \
1103                         fsck_err(c, "stripe %zu has wrong "_msg         \
1104                                 ": got %u, should be %u",               \
1105                                 iter.pos, ##__VA_ARGS__,                \
1106                                 dst->_f, src->_f);                      \
1107                 dst->_f = src->_f;                                      \
1108                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
1109         }
1110 #define copy_bucket_field(_f)                                           \
1111         if (dst->b[b].mark._f != src->b[b].mark._f) {                   \
1112                 if (verify)                                             \
1113                         fsck_err(c, "bucket %u:%zu gen %u data type %s has wrong " #_f  \
1114                                 ": got %u, should be %u", dev, b,       \
1115                                 dst->b[b].mark.gen,                     \
1116                                 bch2_data_types[dst->b[b].mark.data_type],\
1117                                 dst->b[b].mark._f, src->b[b].mark._f);  \
1118                 dst->b[b]._mark._f = src->b[b].mark._f;                 \
1119                 set_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags);            \
1120         }
1121 #define copy_dev_field(_f, _msg, ...)                                   \
1122         copy_field(_f, "dev %u has wrong " _msg, dev, ##__VA_ARGS__)
1123 #define copy_fs_field(_f, _msg, ...)                                    \
1124         copy_field(_f, "fs has wrong " _msg, ##__VA_ARGS__)
1125
1126         if (!metadata_only) {
1127                 struct genradix_iter iter = genradix_iter_init(&c->stripes[1], 0);
1128                 struct stripe *dst, *src;
1129
1130                 while ((src = genradix_iter_peek(&iter, &c->stripes[1]))) {
1131                         dst = genradix_ptr_alloc(&c->stripes[0], iter.pos, GFP_KERNEL);
1132
1133                         if (dst->alive          != src->alive ||
1134                             dst->sectors        != src->sectors ||
1135                             dst->algorithm      != src->algorithm ||
1136                             dst->nr_blocks      != src->nr_blocks ||
1137                             dst->nr_redundant   != src->nr_redundant) {
1138                                 bch_err(c, "unexpected stripe inconsistency at bch2_gc_done, confused");
1139                                 ret = -EINVAL;
1140                                 goto fsck_err;
1141                         }
1142
1143                         for (i = 0; i < ARRAY_SIZE(dst->block_sectors); i++)
1144                                 copy_stripe_field(block_sectors[i],
1145                                                   "block_sectors[%u]", i);
1146
1147                         dst->blocks_nonempty = 0;
1148                         for (i = 0; i < dst->nr_blocks; i++)
1149                                 dst->blocks_nonempty += dst->block_sectors[i] != 0;
1150
1151                         genradix_iter_advance(&iter, &c->stripes[1]);
1152                 }
1153         }
1154
1155         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1156                 bch2_fs_usage_acc_to_base(c, i);
1157
1158         for_each_member_device(ca, c, dev) {
1159                 struct bucket_array *dst = __bucket_array(ca, 0);
1160                 struct bucket_array *src = __bucket_array(ca, 1);
1161                 size_t b;
1162
1163                 for (b = 0; b < src->nbuckets; b++) {
1164                         copy_bucket_field(gen);
1165                         copy_bucket_field(data_type);
1166                         copy_bucket_field(stripe);
1167                         copy_bucket_field(dirty_sectors);
1168                         copy_bucket_field(cached_sectors);
1169
1170                         dst->b[b].oldest_gen = src->b[b].oldest_gen;
1171                 }
1172
1173                 {
1174                         struct bch_dev_usage *dst = ca->usage_base;
1175                         struct bch_dev_usage *src = (void *)
1176                                 bch2_acc_percpu_u64s((void *) ca->usage_gc,
1177                                                      dev_usage_u64s());
1178
1179                         copy_dev_field(buckets_ec,              "buckets_ec");
1180                         copy_dev_field(buckets_unavailable,     "buckets_unavailable");
1181
1182                         for (i = 0; i < BCH_DATA_NR; i++) {
1183                                 copy_dev_field(d[i].buckets,    "%s buckets", bch2_data_types[i]);
1184                                 copy_dev_field(d[i].sectors,    "%s sectors", bch2_data_types[i]);
1185                                 copy_dev_field(d[i].fragmented, "%s fragmented", bch2_data_types[i]);
1186                         }
1187                 }
1188         };
1189
1190         {
1191                 unsigned nr = fs_usage_u64s(c);
1192                 struct bch_fs_usage *dst = c->usage_base;
1193                 struct bch_fs_usage *src = (void *)
1194                         bch2_acc_percpu_u64s((void *) c->usage_gc, nr);
1195
1196                 copy_fs_field(hidden,           "hidden");
1197                 copy_fs_field(btree,            "btree");
1198
1199                 if (!metadata_only) {
1200                         copy_fs_field(data,     "data");
1201                         copy_fs_field(cached,   "cached");
1202                         copy_fs_field(reserved, "reserved");
1203                         copy_fs_field(nr_inodes,"nr_inodes");
1204
1205                         for (i = 0; i < BCH_REPLICAS_MAX; i++)
1206                                 copy_fs_field(persistent_reserved[i],
1207                                               "persistent_reserved[%i]", i);
1208                 }
1209
1210                 for (i = 0; i < c->replicas.nr; i++) {
1211                         struct bch_replicas_entry *e =
1212                                 cpu_replicas_entry(&c->replicas, i);
1213                         char buf[80];
1214
1215                         if (metadata_only &&
1216                             (e->data_type == BCH_DATA_user ||
1217                              e->data_type == BCH_DATA_cached))
1218                                 continue;
1219
1220                         bch2_replicas_entry_to_text(&PBUF(buf), e);
1221
1222                         copy_fs_field(replicas[i], "%s", buf);
1223                 }
1224         }
1225
1226 #undef copy_fs_field
1227 #undef copy_dev_field
1228 #undef copy_bucket_field
1229 #undef copy_stripe_field
1230 #undef copy_field
1231 fsck_err:
1232         if (ca)
1233                 percpu_ref_put(&ca->ref);
1234         if (ret)
1235                 bch_err(c, "%s: ret %i", __func__, ret);
1236         return ret;
1237 }
1238
1239 static int bch2_gc_start(struct bch_fs *c,
1240                          bool metadata_only)
1241 {
1242         struct bch_dev *ca = NULL;
1243         unsigned i;
1244         int ret;
1245
1246         BUG_ON(c->usage_gc);
1247
1248         c->usage_gc = __alloc_percpu_gfp(fs_usage_u64s(c) * sizeof(u64),
1249                                          sizeof(u64), GFP_KERNEL);
1250         if (!c->usage_gc) {
1251                 bch_err(c, "error allocating c->usage_gc");
1252                 return -ENOMEM;
1253         }
1254
1255         for_each_member_device(ca, c, i) {
1256                 BUG_ON(ca->buckets[1]);
1257                 BUG_ON(ca->usage_gc);
1258
1259                 ca->buckets[1] = kvpmalloc(sizeof(struct bucket_array) +
1260                                 ca->mi.nbuckets * sizeof(struct bucket),
1261                                 GFP_KERNEL|__GFP_ZERO);
1262                 if (!ca->buckets[1]) {
1263                         percpu_ref_put(&ca->ref);
1264                         bch_err(c, "error allocating ca->buckets[gc]");
1265                         return -ENOMEM;
1266                 }
1267
1268                 ca->usage_gc = alloc_percpu(struct bch_dev_usage);
1269                 if (!ca->usage_gc) {
1270                         bch_err(c, "error allocating ca->usage_gc");
1271                         percpu_ref_put(&ca->ref);
1272                         return -ENOMEM;
1273                 }
1274         }
1275
1276         ret = bch2_ec_mem_alloc(c, true);
1277         if (ret) {
1278                 bch_err(c, "error allocating ec gc mem");
1279                 return ret;
1280         }
1281
1282         percpu_down_write(&c->mark_lock);
1283
1284         /*
1285          * indicate to stripe code that we need to allocate for the gc stripes
1286          * radix tree, too
1287          */
1288         gc_pos_set(c, gc_phase(GC_PHASE_START));
1289
1290         for_each_member_device(ca, c, i) {
1291                 struct bucket_array *dst = __bucket_array(ca, 1);
1292                 struct bucket_array *src = __bucket_array(ca, 0);
1293                 size_t b;
1294
1295                 dst->first_bucket       = src->first_bucket;
1296                 dst->nbuckets           = src->nbuckets;
1297
1298                 for (b = 0; b < src->nbuckets; b++) {
1299                         struct bucket *d = &dst->b[b];
1300                         struct bucket *s = &src->b[b];
1301
1302                         d->_mark.gen = dst->b[b].oldest_gen = s->mark.gen;
1303                         d->gen_valid = s->gen_valid;
1304
1305                         if (metadata_only &&
1306                             (s->mark.data_type == BCH_DATA_user ||
1307                              s->mark.data_type == BCH_DATA_cached))
1308                                 d->_mark = s->mark;
1309                 }
1310         };
1311
1312         percpu_up_write(&c->mark_lock);
1313
1314         return 0;
1315 }
1316
1317 static int bch2_gc_reflink_done_initial_fn(struct bch_fs *c, struct bkey_s_c k)
1318 {
1319         struct reflink_gc *r;
1320         const __le64 *refcount = bkey_refcount_c(k);
1321         char buf[200];
1322         int ret = 0;
1323
1324         if (!refcount)
1325                 return 0;
1326
1327         r = genradix_ptr(&c->reflink_gc_table, c->reflink_gc_idx++);
1328         if (!r)
1329                 return -ENOMEM;
1330
1331         if (!r ||
1332             r->offset != k.k->p.offset ||
1333             r->size != k.k->size) {
1334                 bch_err(c, "unexpected inconsistency walking reflink table at gc finish");
1335                 return -EINVAL;
1336         }
1337
1338         if (fsck_err_on(r->refcount != le64_to_cpu(*refcount), c,
1339                         "reflink key has wrong refcount:\n"
1340                         "  %s\n"
1341                         "  should be %u",
1342                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf),
1343                         r->refcount)) {
1344                 struct bkey_i *new;
1345
1346                 new = kmalloc(bkey_bytes(k.k), GFP_KERNEL);
1347                 if (!new) {
1348                         ret = -ENOMEM;
1349                         goto fsck_err;
1350                 }
1351
1352                 bkey_reassemble(new, k);
1353
1354                 if (!r->refcount) {
1355                         new->k.type = KEY_TYPE_deleted;
1356                         new->k.size = 0;
1357                 } else {
1358                         *bkey_refcount(new) = cpu_to_le64(r->refcount);
1359                 }
1360
1361                 ret = bch2_journal_key_insert(c, BTREE_ID_reflink, 0, new);
1362                 if (ret)
1363                         kfree(new);
1364         }
1365 fsck_err:
1366         return ret;
1367 }
1368
1369 static int bch2_gc_reflink_done(struct bch_fs *c, bool initial,
1370                                 bool metadata_only)
1371 {
1372         struct btree_trans trans;
1373         struct btree_iter *iter;
1374         struct bkey_s_c k;
1375         struct reflink_gc *r;
1376         size_t idx = 0;
1377         char buf[200];
1378         int ret = 0;
1379
1380         if (metadata_only)
1381                 return 0;
1382
1383         if (initial) {
1384                 c->reflink_gc_idx = 0;
1385
1386                 ret = bch2_btree_and_journal_walk(c, BTREE_ID_reflink,
1387                                 bch2_gc_reflink_done_initial_fn);
1388                 goto out;
1389         }
1390
1391         bch2_trans_init(&trans, c, 0, 0);
1392
1393         for_each_btree_key(&trans, iter, BTREE_ID_reflink, POS_MIN,
1394                            BTREE_ITER_PREFETCH, k, ret) {
1395                 const __le64 *refcount = bkey_refcount_c(k);
1396
1397                 if (!refcount)
1398                         continue;
1399
1400                 r = genradix_ptr(&c->reflink_gc_table, idx);
1401                 if (!r ||
1402                     r->offset != k.k->p.offset ||
1403                     r->size != k.k->size) {
1404                         bch_err(c, "unexpected inconsistency walking reflink table at gc finish");
1405                         ret = -EINVAL;
1406                         break;
1407                 }
1408
1409                 if (fsck_err_on(r->refcount != le64_to_cpu(*refcount), c,
1410                                 "reflink key has wrong refcount:\n"
1411                                 "  %s\n"
1412                                 "  should be %u",
1413                                 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf),
1414                                 r->refcount)) {
1415                         struct bkey_i *new;
1416
1417                         new = kmalloc(bkey_bytes(k.k), GFP_KERNEL);
1418                         if (!new) {
1419                                 ret = -ENOMEM;
1420                                 break;
1421                         }
1422
1423                         bkey_reassemble(new, k);
1424
1425                         if (!r->refcount)
1426                                 new->k.type = KEY_TYPE_deleted;
1427                         else
1428                                 *bkey_refcount(new) = cpu_to_le64(r->refcount);
1429
1430                         ret = __bch2_trans_do(&trans, NULL, NULL, 0,
1431                                         __bch2_btree_insert(&trans, BTREE_ID_reflink, new));
1432                         kfree(new);
1433
1434                         if (ret)
1435                                 break;
1436                 }
1437         }
1438 fsck_err:
1439         bch2_trans_iter_put(&trans, iter);
1440         bch2_trans_exit(&trans);
1441 out:
1442         genradix_free(&c->reflink_gc_table);
1443         c->reflink_gc_nr = 0;
1444         return ret;
1445 }
1446
1447 static int bch2_gc_reflink_start_initial_fn(struct bch_fs *c, struct bkey_s_c k)
1448 {
1449
1450         struct reflink_gc *r;
1451         const __le64 *refcount = bkey_refcount_c(k);
1452
1453         if (!refcount)
1454                 return 0;
1455
1456         r = genradix_ptr_alloc(&c->reflink_gc_table, c->reflink_gc_nr++,
1457                                GFP_KERNEL);
1458         if (!r)
1459                 return -ENOMEM;
1460
1461         r->offset       = k.k->p.offset;
1462         r->size         = k.k->size;
1463         r->refcount     = 0;
1464         return 0;
1465 }
1466
1467 static int bch2_gc_reflink_start(struct bch_fs *c, bool initial,
1468                                  bool metadata_only)
1469 {
1470         struct btree_trans trans;
1471         struct btree_iter *iter;
1472         struct bkey_s_c k;
1473         struct reflink_gc *r;
1474         int ret;
1475
1476         if (metadata_only)
1477                 return 0;
1478
1479         genradix_free(&c->reflink_gc_table);
1480         c->reflink_gc_nr = 0;
1481
1482         if (initial)
1483                 return bch2_btree_and_journal_walk(c, BTREE_ID_reflink,
1484                                 bch2_gc_reflink_start_initial_fn);
1485
1486         bch2_trans_init(&trans, c, 0, 0);
1487
1488         for_each_btree_key(&trans, iter, BTREE_ID_reflink, POS_MIN,
1489                            BTREE_ITER_PREFETCH, k, ret) {
1490                 const __le64 *refcount = bkey_refcount_c(k);
1491
1492                 if (!refcount)
1493                         continue;
1494
1495                 r = genradix_ptr_alloc(&c->reflink_gc_table, c->reflink_gc_nr++,
1496                                        GFP_KERNEL);
1497                 if (!r) {
1498                         ret = -ENOMEM;
1499                         break;
1500                 }
1501
1502                 r->offset       = k.k->p.offset;
1503                 r->size         = k.k->size;
1504                 r->refcount     = 0;
1505         }
1506         bch2_trans_iter_put(&trans, iter);
1507
1508         bch2_trans_exit(&trans);
1509         return 0;
1510 }
1511
1512 /**
1513  * bch2_gc - walk _all_ references to buckets, and recompute them:
1514  *
1515  * Order matters here:
1516  *  - Concurrent GC relies on the fact that we have a total ordering for
1517  *    everything that GC walks - see  gc_will_visit_node(),
1518  *    gc_will_visit_root()
1519  *
1520  *  - also, references move around in the course of index updates and
1521  *    various other crap: everything needs to agree on the ordering
1522  *    references are allowed to move around in - e.g., we're allowed to
1523  *    start with a reference owned by an open_bucket (the allocator) and
1524  *    move it to the btree, but not the reverse.
1525  *
1526  *    This is necessary to ensure that gc doesn't miss references that
1527  *    move around - if references move backwards in the ordering GC
1528  *    uses, GC could skip past them
1529  */
1530 int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
1531 {
1532         struct bch_dev *ca;
1533         u64 start_time = local_clock();
1534         unsigned i, iter = 0;
1535         int ret;
1536
1537         lockdep_assert_held(&c->state_lock);
1538         trace_gc_start(c);
1539
1540         down_write(&c->gc_lock);
1541
1542         /* flush interior btree updates: */
1543         closure_wait_event(&c->btree_interior_update_wait,
1544                            !bch2_btree_interior_updates_nr_pending(c));
1545 again:
1546         ret   = bch2_gc_start(c, metadata_only) ?:
1547                 bch2_gc_reflink_start(c, initial, metadata_only);
1548         if (ret)
1549                 goto out;
1550
1551         bch2_mark_superblocks(c);
1552
1553         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags) &&
1554             !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags) &&
1555             c->opts.fix_errors != FSCK_OPT_NO) {
1556                 bch_info(c, "starting topology repair pass");
1557                 ret = bch2_repair_topology(c);
1558                 if (ret)
1559                         goto out;
1560                 bch_info(c, "topology repair pass done");
1561         }
1562
1563         ret = bch2_gc_btrees(c, initial, metadata_only);
1564
1565         if (ret == FSCK_ERR_START_TOPOLOGY_REPAIR &&
1566             !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags)) {
1567                 set_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags);
1568                 ret = 0;
1569         }
1570
1571         if (ret == FSCK_ERR_START_TOPOLOGY_REPAIR)
1572                 ret = FSCK_ERR_EXIT;
1573
1574         if (ret)
1575                 goto out;
1576
1577 #if 0
1578         bch2_mark_pending_btree_node_frees(c);
1579 #endif
1580         c->gc_count++;
1581
1582         if (test_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags) ||
1583             (!iter && bch2_test_restart_gc)) {
1584                 /*
1585                  * XXX: make sure gens we fixed got saved
1586                  */
1587                 if (iter++ <= 2) {
1588                         bch_info(c, "Second GC pass needed, restarting:");
1589                         clear_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags);
1590                         __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
1591
1592                         percpu_down_write(&c->mark_lock);
1593                         bch2_gc_free(c);
1594                         percpu_up_write(&c->mark_lock);
1595                         /* flush fsck errors, reset counters */
1596                         bch2_flush_fsck_errs(c);
1597
1598                         goto again;
1599                 }
1600
1601                 bch_info(c, "Unable to fix bucket gens, looping");
1602                 ret = -EINVAL;
1603         }
1604 out:
1605         if (!ret) {
1606                 bch2_journal_block(&c->journal);
1607
1608                 percpu_down_write(&c->mark_lock);
1609                 ret   = bch2_gc_reflink_done(c, initial, metadata_only) ?:
1610                         bch2_gc_done(c, initial, metadata_only);
1611
1612                 bch2_journal_unblock(&c->journal);
1613         } else {
1614                 percpu_down_write(&c->mark_lock);
1615         }
1616
1617         /* Indicates that gc is no longer in progress: */
1618         __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
1619
1620         bch2_gc_free(c);
1621         percpu_up_write(&c->mark_lock);
1622
1623         up_write(&c->gc_lock);
1624
1625         trace_gc_end(c);
1626         bch2_time_stats_update(&c->times[BCH_TIME_btree_gc], start_time);
1627
1628         /*
1629          * Wake up allocator in case it was waiting for buckets
1630          * because of not being able to inc gens
1631          */
1632         for_each_member_device(ca, c, i)
1633                 bch2_wake_allocator(ca);
1634
1635         /*
1636          * At startup, allocations can happen directly instead of via the
1637          * allocator thread - issue wakeup in case they blocked on gc_lock:
1638          */
1639         closure_wake_up(&c->freelist_wait);
1640         return ret;
1641 }
1642
1643 static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k)
1644 {
1645         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1646         const struct bch_extent_ptr *ptr;
1647
1648         percpu_down_read(&c->mark_lock);
1649         bkey_for_each_ptr(ptrs, ptr) {
1650                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1651                 struct bucket *g = PTR_BUCKET(ca, ptr, false);
1652
1653                 if (gen_after(g->mark.gen, ptr->gen) > 16) {
1654                         percpu_up_read(&c->mark_lock);
1655                         return true;
1656                 }
1657         }
1658
1659         bkey_for_each_ptr(ptrs, ptr) {
1660                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1661                 struct bucket *g = PTR_BUCKET(ca, ptr, false);
1662
1663                 if (gen_after(g->gc_gen, ptr->gen))
1664                         g->gc_gen = ptr->gen;
1665         }
1666         percpu_up_read(&c->mark_lock);
1667
1668         return false;
1669 }
1670
1671 /*
1672  * For recalculating oldest gen, we only need to walk keys in leaf nodes; btree
1673  * node pointers currently never have cached pointers that can become stale:
1674  */
1675 static int bch2_gc_btree_gens(struct bch_fs *c, enum btree_id btree_id)
1676 {
1677         struct btree_trans trans;
1678         struct btree_iter *iter;
1679         struct bkey_s_c k;
1680         struct bkey_buf sk;
1681         int ret = 0, commit_err = 0;
1682
1683         bch2_bkey_buf_init(&sk);
1684         bch2_trans_init(&trans, c, 0, 0);
1685
1686         iter = bch2_trans_get_iter(&trans, btree_id, POS_MIN,
1687                                    BTREE_ITER_PREFETCH|
1688                                    BTREE_ITER_NOT_EXTENTS|
1689                                    BTREE_ITER_ALL_SNAPSHOTS);
1690
1691         while ((k = bch2_btree_iter_peek(iter)).k &&
1692                !(ret = bkey_err(k))) {
1693                 c->gc_gens_pos = iter->pos;
1694
1695                 if (gc_btree_gens_key(c, k) && !commit_err) {
1696                         bch2_bkey_buf_reassemble(&sk, c, k);
1697                         bch2_extent_normalize(c, bkey_i_to_s(sk.k));
1698
1699
1700                         commit_err =
1701                                 bch2_trans_update(&trans, iter, sk.k, 0) ?:
1702                                 bch2_trans_commit(&trans, NULL, NULL,
1703                                                        BTREE_INSERT_NOWAIT|
1704                                                        BTREE_INSERT_NOFAIL);
1705                         if (commit_err == -EINTR) {
1706                                 commit_err = 0;
1707                                 continue;
1708                         }
1709                 }
1710
1711                 bch2_btree_iter_advance(iter);
1712         }
1713         bch2_trans_iter_put(&trans, iter);
1714
1715         bch2_trans_exit(&trans);
1716         bch2_bkey_buf_exit(&sk, c);
1717
1718         return ret;
1719 }
1720
1721 int bch2_gc_gens(struct bch_fs *c)
1722 {
1723         struct bch_dev *ca;
1724         struct bucket_array *buckets;
1725         struct bucket *g;
1726         unsigned i;
1727         int ret;
1728
1729         /*
1730          * Ideally we would be using state_lock and not gc_lock here, but that
1731          * introduces a deadlock in the RO path - we currently take the state
1732          * lock at the start of going RO, thus the gc thread may get stuck:
1733          */
1734         down_read(&c->gc_lock);
1735
1736         for_each_member_device(ca, c, i) {
1737                 down_read(&ca->bucket_lock);
1738                 buckets = bucket_array(ca);
1739
1740                 for_each_bucket(g, buckets)
1741                         g->gc_gen = g->mark.gen;
1742                 up_read(&ca->bucket_lock);
1743         }
1744
1745         for (i = 0; i < BTREE_ID_NR; i++)
1746                 if ((1 << i) & BTREE_ID_HAS_PTRS) {
1747                         c->gc_gens_btree = i;
1748                         c->gc_gens_pos = POS_MIN;
1749                         ret = bch2_gc_btree_gens(c, i);
1750                         if (ret) {
1751                                 bch_err(c, "error recalculating oldest_gen: %i", ret);
1752                                 goto err;
1753                         }
1754                 }
1755
1756         for_each_member_device(ca, c, i) {
1757                 down_read(&ca->bucket_lock);
1758                 buckets = bucket_array(ca);
1759
1760                 for_each_bucket(g, buckets)
1761                         g->oldest_gen = g->gc_gen;
1762                 up_read(&ca->bucket_lock);
1763         }
1764
1765         c->gc_gens_btree        = 0;
1766         c->gc_gens_pos          = POS_MIN;
1767
1768         c->gc_count++;
1769 err:
1770         up_read(&c->gc_lock);
1771         return ret;
1772 }
1773
1774 static int bch2_gc_thread(void *arg)
1775 {
1776         struct bch_fs *c = arg;
1777         struct io_clock *clock = &c->io_clock[WRITE];
1778         unsigned long last = atomic64_read(&clock->now);
1779         unsigned last_kick = atomic_read(&c->kick_gc);
1780         int ret;
1781
1782         set_freezable();
1783
1784         while (1) {
1785                 while (1) {
1786                         set_current_state(TASK_INTERRUPTIBLE);
1787
1788                         if (kthread_should_stop()) {
1789                                 __set_current_state(TASK_RUNNING);
1790                                 return 0;
1791                         }
1792
1793                         if (atomic_read(&c->kick_gc) != last_kick)
1794                                 break;
1795
1796                         if (c->btree_gc_periodic) {
1797                                 unsigned long next = last + c->capacity / 16;
1798
1799                                 if (atomic64_read(&clock->now) >= next)
1800                                         break;
1801
1802                                 bch2_io_clock_schedule_timeout(clock, next);
1803                         } else {
1804                                 schedule();
1805                         }
1806
1807                         try_to_freeze();
1808                 }
1809                 __set_current_state(TASK_RUNNING);
1810
1811                 last = atomic64_read(&clock->now);
1812                 last_kick = atomic_read(&c->kick_gc);
1813
1814                 /*
1815                  * Full gc is currently incompatible with btree key cache:
1816                  */
1817 #if 0
1818                 ret = bch2_gc(c, false, false);
1819 #else
1820                 ret = bch2_gc_gens(c);
1821 #endif
1822                 if (ret < 0)
1823                         bch_err(c, "btree gc failed: %i", ret);
1824
1825                 debug_check_no_locks_held();
1826         }
1827
1828         return 0;
1829 }
1830
1831 void bch2_gc_thread_stop(struct bch_fs *c)
1832 {
1833         struct task_struct *p;
1834
1835         p = c->gc_thread;
1836         c->gc_thread = NULL;
1837
1838         if (p) {
1839                 kthread_stop(p);
1840                 put_task_struct(p);
1841         }
1842 }
1843
1844 int bch2_gc_thread_start(struct bch_fs *c)
1845 {
1846         struct task_struct *p;
1847
1848         if (c->gc_thread)
1849                 return 0;
1850
1851         p = kthread_create(bch2_gc_thread, c, "bch-gc/%s", c->name);
1852         if (IS_ERR(p)) {
1853                 bch_err(c, "error creating gc thread: %li", PTR_ERR(p));
1854                 return PTR_ERR(p);
1855         }
1856
1857         get_task_struct(p);
1858         c->gc_thread = p;
1859         wake_up_process(p);
1860         return 0;
1861 }