fs/bcachefs/extents.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2010 Kent Overstreet <[email protected]>
   4  *
   5  * Code for managing the extent btree and dynamically updating the writeback
   6  * dirty sector count.
   7  */
   8
   9 #include "bcachefs.h"
  10 #include "bkey_methods.h"
  11 #include "btree_cache.h"
  12 #include "btree_gc.h"
  13 #include "btree_io.h"
  14 #include "btree_iter.h"
  15 #include "buckets.h"
  16 #include "checksum.h"
  17 #include "compress.h"
  18 #include "debug.h"
  19 #include "disk_groups.h"
  20 #include "error.h"
  21 #include "extents.h"
  22 #include "inode.h"
  23 #include "journal.h"
  24 #include "rebalance.h"
  25 #include "replicas.h"
  26 #include "super.h"
  27 #include "super-io.h"
  28 #include "trace.h"
  29 #include "util.h"
  30
  31 static unsigned bch2_crc_field_size_max[] = {
  32         [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
  33         [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
  34         [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
  35 };
  36
  37 static void bch2_extent_crc_pack(union bch_extent_crc *,
  38                                  struct bch_extent_crc_unpacked,
  39                                  enum bch_extent_entry_type);
  40
  41 struct bch_dev_io_failures *bch2_dev_io_failures(struct bch_io_failures *f,
  42                                                  unsigned dev)
  43 {
  44         struct bch_dev_io_failures *i;
  45
  46         for (i = f->devs; i < f->devs + f->nr; i++)
  47                 if (i->dev == dev)
  48                         return i;
  49
  50         return NULL;
  51 }
  52
  53 void bch2_mark_io_failure(struct bch_io_failures *failed,
  54                           struct extent_ptr_decoded *p)
  55 {
  56         struct bch_dev_io_failures *f = bch2_dev_io_failures(failed, p->ptr.dev);
  57
  58         if (!f) {
  59                 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
  60
  61                 f = &failed->devs[failed->nr++];
  62                 f->dev          = p->ptr.dev;
  63                 f->idx          = p->idx;
  64                 f->nr_failed    = 1;
  65                 f->nr_retries   = 0;
  66         } else if (p->idx != f->idx) {
  67                 f->idx          = p->idx;
  68                 f->nr_failed    = 1;
  69                 f->nr_retries   = 0;
  70         } else {
  71                 f->nr_failed++;
  72         }
  73 }
  74
  75 static inline u64 dev_latency(struct bch_fs *c, unsigned dev)
  76 {
  77         struct bch_dev *ca = bch2_dev_rcu(c, dev);
  78         return ca ? atomic64_read(&ca->cur_latency[READ]) : S64_MAX;
  79 }
  80
  81 /*
  82  * returns true if p1 is better than p2:
  83  */
  84 static inline bool ptr_better(struct bch_fs *c,
  85                               const struct extent_ptr_decoded p1,
  86                               const struct extent_ptr_decoded p2)
  87 {
  88         if (likely(!p1.idx && !p2.idx)) {
  89                 u64 l1 = dev_latency(c, p1.ptr.dev);
  90                 u64 l2 = dev_latency(c, p2.ptr.dev);
  91
  92                 /*
  93                  * Square the latencies, to bias more in favor of the faster
  94                  * device - we never want to stop issuing reads to the slower
  95                  * device altogether, so that we can update our latency numbers:
  96                  */
  97                 l1 *= l1;
  98                 l2 *= l2;
  99
 100                 /* Pick at random, biased in favor of the faster device: */
 101
 102                 return bch2_rand_range(l1 + l2) > l1;
 103         }
 104
 105         if (bch2_force_reconstruct_read)
 106                 return p1.idx > p2.idx;
 107
 108         return p1.idx < p2.idx;
 109 }
 110
 111 /*
 112  * This picks a non-stale pointer, preferably from a device other than @avoid.
 113  * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
 114  * other devices, it will still pick a pointer from avoid.
 115  */
 116 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 117                                struct bch_io_failures *failed,
 118                                struct extent_ptr_decoded *pick)
 119 {
 120         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 121         const union bch_extent_entry *entry;
 122         struct extent_ptr_decoded p;
 123         struct bch_dev_io_failures *f;
 124         int ret = 0;
 125
 126         if (k.k->type == KEY_TYPE_error)
 127                 return -BCH_ERR_key_type_error;
 128
 129         rcu_read_lock();
 130         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 131                 /*
 132                  * Unwritten extent: no need to actually read, treat it as a
 133                  * hole and return 0s:
 134                  */
 135                 if (p.ptr.unwritten) {
 136                         ret = 0;
 137                         break;
 138                 }
 139
 140                 /*
 141                  * If there are any dirty pointers it's an error if we can't
 142                  * read:
 143                  */
 144                 if (!ret && !p.ptr.cached)
 145                         ret = -BCH_ERR_no_device_to_read_from;
 146
 147                 struct bch_dev *ca = bch2_dev_rcu(c, p.ptr.dev);
 148
 149                 if (p.ptr.cached && (!ca || dev_ptr_stale_rcu(ca, &p.ptr)))
 150                         continue;
 151
 152                 f = failed ? bch2_dev_io_failures(failed, p.ptr.dev) : NULL;
 153                 if (f)
 154                         p.idx = f->nr_failed < f->nr_retries
 155                                 ? f->idx
 156                                 : f->idx + 1;
 157
 158                 if (!p.idx && (!ca || !bch2_dev_is_readable(ca)))
 159                         p.idx++;
 160
 161                 if (!p.idx && p.has_ec && bch2_force_reconstruct_read)
 162                         p.idx++;
 163
 164                 if (p.idx > (unsigned) p.has_ec)
 165                         continue;
 166
 167                 if (ret > 0 && !ptr_better(c, p, *pick))
 168                         continue;
 169
 170                 *pick = p;
 171                 ret = 1;
 172         }
 173         rcu_read_unlock();
 174
 175         return ret;
 176 }
 177
 178 /* KEY_TYPE_btree_ptr: */
 179
 180 int bch2_btree_ptr_validate(struct bch_fs *c, struct bkey_s_c k,
 181                             struct bkey_validate_context from)
 182 {
 183         int ret = 0;
 184
 185         bkey_fsck_err_on(bkey_val_u64s(k.k) > BCH_REPLICAS_MAX,
 186                          c, btree_ptr_val_too_big,
 187                          "value too big (%zu > %u)", bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 188
 189         ret = bch2_bkey_ptrs_validate(c, k, from);
 190 fsck_err:
 191         return ret;
 192 }
 193
 194 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 195                             struct bkey_s_c k)
 196 {
 197         bch2_bkey_ptrs_to_text(out, c, k);
 198 }
 199
 200 int bch2_btree_ptr_v2_validate(struct bch_fs *c, struct bkey_s_c k,
 201                                struct bkey_validate_context from)
 202 {
 203         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 204         int ret = 0;
 205
 206         bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX,
 207                          c, btree_ptr_v2_val_too_big,
 208                          "value too big (%zu > %zu)",
 209                          bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 210
 211         bkey_fsck_err_on(bpos_ge(bp.v->min_key, bp.k->p),
 212                          c, btree_ptr_v2_min_key_bad,
 213                          "min_key > key");
 214
 215         if ((from.flags & BCH_VALIDATE_write) &&
 216             c->sb.version_min >= bcachefs_metadata_version_btree_ptr_sectors_written)
 217                 bkey_fsck_err_on(!bp.v->sectors_written,
 218                                  c, btree_ptr_v2_written_0,
 219                                  "sectors_written == 0");
 220
 221         ret = bch2_bkey_ptrs_validate(c, k, from);
 222 fsck_err:
 223         return ret;
 224 }
 225
 226 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 227                                struct bkey_s_c k)
 228 {
 229         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 230
 231         prt_printf(out, "seq %llx written %u min_key %s",
 232                le64_to_cpu(bp.v->seq),
 233                le16_to_cpu(bp.v->sectors_written),
 234                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 235
 236         bch2_bpos_to_text(out, bp.v->min_key);
 237         prt_printf(out, " ");
 238         bch2_bkey_ptrs_to_text(out, c, k);
 239 }
 240
 241 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 242                               unsigned big_endian, int write,
 243                               struct bkey_s k)
 244 {
 245         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 246
 247         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 248
 249         if (version < bcachefs_metadata_version_inode_btree_change &&
 250             btree_id_is_extents(btree_id) &&
 251             !bkey_eq(bp.v->min_key, POS_MIN))
 252                 bp.v->min_key = write
 253                         ? bpos_nosnap_predecessor(bp.v->min_key)
 254                         : bpos_nosnap_successor(bp.v->min_key);
 255 }
 256
 257 /* KEY_TYPE_extent: */
 258
 259 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 260 {
 261         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 262         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 263         union bch_extent_entry *en_l;
 264         const union bch_extent_entry *en_r;
 265         struct extent_ptr_decoded lp, rp;
 266         bool use_right_ptr;
 267
 268         en_l = l_ptrs.start;
 269         en_r = r_ptrs.start;
 270         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 271                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 272                         return false;
 273
 274                 en_l = extent_entry_next(en_l);
 275                 en_r = extent_entry_next(en_r);
 276         }
 277
 278         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 279                 return false;
 280
 281         en_l = l_ptrs.start;
 282         en_r = r_ptrs.start;
 283         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 284         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 285
 286         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 287                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 288                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 289                     rp.ptr.offset + rp.crc.offset ||
 290                     lp.ptr.dev                  != rp.ptr.dev ||
 291                     lp.ptr.gen                  != rp.ptr.gen ||
 292                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 293                     lp.has_ec                   != rp.has_ec)
 294                         return false;
 295
 296                 /* Extents may not straddle buckets: */
 297                 rcu_read_lock();
 298                 struct bch_dev *ca = bch2_dev_rcu(c, lp.ptr.dev);
 299                 bool same_bucket = ca && PTR_BUCKET_NR(ca, &lp.ptr) == PTR_BUCKET_NR(ca, &rp.ptr);
 300                 rcu_read_unlock();
 301
 302                 if (!same_bucket)
 303                         return false;
 304
 305                 if (lp.has_ec                   != rp.has_ec ||
 306                     (lp.has_ec &&
 307                      (lp.ec.block               != rp.ec.block ||
 308                       lp.ec.redundancy          != rp.ec.redundancy ||
 309                       lp.ec.idx                 != rp.ec.idx)))
 310                         return false;
 311
 312                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 313                     lp.crc.nonce                != rp.crc.nonce)
 314                         return false;
 315
 316                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 317                     lp.crc.uncompressed_size) {
 318                         /* can use left extent's crc entry */
 319                 } else if (lp.crc.live_size <= rp.crc.offset) {
 320                         /* can use right extent's crc entry */
 321                 } else {
 322                         /* check if checksums can be merged: */
 323                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 324                             lp.crc.nonce                != rp.crc.nonce ||
 325                             crc_is_compressed(lp.crc) ||
 326                             !bch2_checksum_mergeable(lp.crc.csum_type))
 327                                 return false;
 328
 329                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 330                             rp.crc.offset)
 331                                 return false;
 332
 333                         if (lp.crc.csum_type &&
 334                             lp.crc.uncompressed_size +
 335                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 336                                 return false;
 337                 }
 338
 339                 en_l = extent_entry_next(en_l);
 340                 en_r = extent_entry_next(en_r);
 341         }
 342
 343         en_l = l_ptrs.start;
 344         en_r = r_ptrs.start;
 345         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 346                 if (extent_entry_is_crc(en_l)) {
 347                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 348                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 349
 350                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 351                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 352                                 return false;
 353                 }
 354
 355                 en_l = extent_entry_next(en_l);
 356                 en_r = extent_entry_next(en_r);
 357         }
 358
 359         use_right_ptr = false;
 360         en_l = l_ptrs.start;
 361         en_r = r_ptrs.start;
 362         while (en_l < l_ptrs.end) {
 363                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 364                     use_right_ptr)
 365                         en_l->ptr = en_r->ptr;
 366
 367                 if (extent_entry_is_crc(en_l)) {
 368                         struct bch_extent_crc_unpacked crc_l =
 369                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 370                         struct bch_extent_crc_unpacked crc_r =
 371                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 372
 373                         use_right_ptr = false;
 374
 375                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 376                             crc_l.uncompressed_size) {
 377                                 /* can use left extent's crc entry */
 378                         } else if (crc_l.live_size <= crc_r.offset) {
 379                                 /* can use right extent's crc entry */
 380                                 crc_r.offset -= crc_l.live_size;
 381                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 382                                                      extent_entry_type(en_l));
 383                                 use_right_ptr = true;
 384                         } else {
 385                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 386                                                                  crc_l.csum,
 387                                                                  crc_r.csum,
 388                                                                  crc_r.uncompressed_size << 9);
 389
 390                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 391                                 crc_l.compressed_size   += crc_r.compressed_size;
 392                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 393                                                      extent_entry_type(en_l));
 394                         }
 395                 }
 396
 397                 en_l = extent_entry_next(en_l);
 398                 en_r = extent_entry_next(en_r);
 399         }
 400
 401         bch2_key_resize(l.k, l.k->size + r.k->size);
 402         return true;
 403 }
 404
 405 /* KEY_TYPE_reservation: */
 406
 407 int bch2_reservation_validate(struct bch_fs *c, struct bkey_s_c k,
 408                               struct bkey_validate_context from)
 409 {
 410         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 411         int ret = 0;
 412
 413         bkey_fsck_err_on(!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX,
 414                          c, reservation_key_nr_replicas_invalid,
 415                          "invalid nr_replicas (%u)", r.v->nr_replicas);
 416 fsck_err:
 417         return ret;
 418 }
 419
 420 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 421                               struct bkey_s_c k)
 422 {
 423         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 424
 425         prt_printf(out, "generation %u replicas %u",
 426                le32_to_cpu(r.v->generation),
 427                r.v->nr_replicas);
 428 }
 429
 430 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 431 {
 432         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 433         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 434
 435         if (l.v->generation != r.v->generation ||
 436             l.v->nr_replicas != r.v->nr_replicas)
 437                 return false;
 438
 439         bch2_key_resize(l.k, l.k->size + r.k->size);
 440         return true;
 441 }
 442
 443 /* Extent checksum entries: */
 444
 445 /* returns true if not equal */
 446 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 447                                          struct bch_extent_crc_unpacked r)
 448 {
 449         return (l.csum_type             != r.csum_type ||
 450                 l.compression_type      != r.compression_type ||
 451                 l.compressed_size       != r.compressed_size ||
 452                 l.uncompressed_size     != r.uncompressed_size ||
 453                 l.offset                != r.offset ||
 454                 l.live_size             != r.live_size ||
 455                 l.nonce                 != r.nonce ||
 456                 bch2_crc_cmp(l.csum, r.csum));
 457 }
 458
 459 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 460                                   struct bch_extent_crc_unpacked n)
 461 {
 462         return !crc_is_compressed(u) &&
 463                 u.csum_type &&
 464                 u.uncompressed_size > u.live_size &&
 465                 bch2_csum_type_is_encryption(u.csum_type) ==
 466                 bch2_csum_type_is_encryption(n.csum_type);
 467 }
 468
 469 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 470                                  struct bch_extent_crc_unpacked n)
 471 {
 472         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 473         struct bch_extent_crc_unpacked crc;
 474         const union bch_extent_entry *i;
 475
 476         if (!n.csum_type)
 477                 return false;
 478
 479         bkey_for_each_crc(k.k, ptrs, crc, i)
 480                 if (can_narrow_crc(crc, n))
 481                         return true;
 482
 483         return false;
 484 }
 485
 486 /*
 487  * We're writing another replica for this extent, so while we've got the data in
 488  * memory we'll be computing a new checksum for the currently live data.
 489  *
 490  * If there are other replicas we aren't moving, and they are checksummed but
 491  * not compressed, we can modify them to point to only the data that is
 492  * currently live (so that readers won't have to bounce) while we've got the
 493  * checksum we need:
 494  */
 495 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 496 {
 497         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 498         struct bch_extent_crc_unpacked u;
 499         struct extent_ptr_decoded p;
 500         union bch_extent_entry *i;
 501         bool ret = false;
 502
 503         /* Find a checksum entry that covers only live data: */
 504         if (!n.csum_type) {
 505                 bkey_for_each_crc(&k->k, ptrs, u, i)
 506                         if (!crc_is_compressed(u) &&
 507                             u.csum_type &&
 508                             u.live_size == u.uncompressed_size) {
 509                                 n = u;
 510                                 goto found;
 511                         }
 512                 return false;
 513         }
 514 found:
 515         BUG_ON(crc_is_compressed(n));
 516         BUG_ON(n.offset);
 517         BUG_ON(n.live_size != k->k.size);
 518
 519 restart_narrow_pointers:
 520         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 521
 522         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 523                 if (can_narrow_crc(p.crc, n)) {
 524                         bch2_bkey_drop_ptr_noerror(bkey_i_to_s(k), &i->ptr);
 525                         p.ptr.offset += p.crc.offset;
 526                         p.crc = n;
 527                         bch2_extent_ptr_decoded_append(k, &p);
 528                         ret = true;
 529                         goto restart_narrow_pointers;
 530                 }
 531
 532         return ret;
 533 }
 534
 535 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 536                                  struct bch_extent_crc_unpacked src,
 537                                  enum bch_extent_entry_type type)
 538 {
 539 #define set_common_fields(_dst, _src)                                   \
 540                 _dst.type               = 1 << type;                    \
 541                 _dst.csum_type          = _src.csum_type,               \
 542                 _dst.compression_type   = _src.compression_type,        \
 543                 _dst._compressed_size   = _src.compressed_size - 1,     \
 544                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 545                 _dst.offset             = _src.offset
 546
 547         switch (type) {
 548         case BCH_EXTENT_ENTRY_crc32:
 549                 set_common_fields(dst->crc32, src);
 550                 dst->crc32.csum         = (u32 __force) *((__le32 *) &src.csum.lo);
 551                 break;
 552         case BCH_EXTENT_ENTRY_crc64:
 553                 set_common_fields(dst->crc64, src);
 554                 dst->crc64.nonce        = src.nonce;
 555                 dst->crc64.csum_lo      = (u64 __force) src.csum.lo;
 556                 dst->crc64.csum_hi      = (u64 __force) *((__le16 *) &src.csum.hi);
 557                 break;
 558         case BCH_EXTENT_ENTRY_crc128:
 559                 set_common_fields(dst->crc128, src);
 560                 dst->crc128.nonce       = src.nonce;
 561                 dst->crc128.csum        = src.csum;
 562                 break;
 563         default:
 564                 BUG();
 565         }
 566 #undef set_common_fields
 567 }
 568
 569 void bch2_extent_crc_append(struct bkey_i *k,
 570                             struct bch_extent_crc_unpacked new)
 571 {
 572         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 573         union bch_extent_crc *crc = (void *) ptrs.end;
 574         enum bch_extent_entry_type type;
 575
 576         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 577             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 578             new.nonce                           <= CRC32_NONCE_MAX)
 579                 type = BCH_EXTENT_ENTRY_crc32;
 580         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 581                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 582                    new.nonce                    <= CRC64_NONCE_MAX)
 583                 type = BCH_EXTENT_ENTRY_crc64;
 584         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 585                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 586                    new.nonce                    <= CRC128_NONCE_MAX)
 587                 type = BCH_EXTENT_ENTRY_crc128;
 588         else
 589                 BUG();
 590
 591         bch2_extent_crc_pack(crc, new, type);
 592
 593         k->k.u64s += extent_entry_u64s(ptrs.end);
 594
 595         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 596 }
 597
 598 /* Generic code for keys with pointers: */
 599
 600 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 601 {
 602         return bch2_bkey_devs(k).nr;
 603 }
 604
 605 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 606 {
 607         return k.k->type == KEY_TYPE_reservation
 608                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 609                 : bch2_bkey_dirty_devs(k).nr;
 610 }
 611
 612 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 613 {
 614         unsigned ret = 0;
 615
 616         if (k.k->type == KEY_TYPE_reservation) {
 617                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 618         } else {
 619                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 620                 const union bch_extent_entry *entry;
 621                 struct extent_ptr_decoded p;
 622
 623                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 624                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 625         }
 626
 627         return ret;
 628 }
 629
 630 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 631 {
 632         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 633         const union bch_extent_entry *entry;
 634         struct extent_ptr_decoded p;
 635         unsigned ret = 0;
 636
 637         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 638                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 639                         ret += p.crc.compressed_size;
 640
 641         return ret;
 642 }
 643
 644 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
 645 {
 646         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 647         const union bch_extent_entry *entry;
 648         struct bch_extent_crc_unpacked crc;
 649
 650         bkey_for_each_crc(k.k, ptrs, crc, entry)
 651                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 652                         return true;
 653         return false;
 654 }
 655
 656 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 657 {
 658         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 659         const union bch_extent_entry *entry;
 660         struct extent_ptr_decoded p = { 0 };
 661         unsigned replicas = 0;
 662
 663         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 664                 if (p.ptr.cached)
 665                         continue;
 666
 667                 if (p.has_ec)
 668                         replicas += p.ec.redundancy;
 669
 670                 replicas++;
 671
 672         }
 673
 674         return replicas;
 675 }
 676
 677 static inline unsigned __extent_ptr_durability(struct bch_dev *ca, struct extent_ptr_decoded *p)
 678 {
 679         if (p->ptr.cached)
 680                 return 0;
 681
 682         return p->has_ec
 683                 ? p->ec.redundancy + 1
 684                 : ca->mi.durability;
 685 }
 686
 687 unsigned bch2_extent_ptr_desired_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 688 {
 689         struct bch_dev *ca = bch2_dev_rcu(c, p->ptr.dev);
 690
 691         return ca ? __extent_ptr_durability(ca, p) : 0;
 692 }
 693
 694 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 695 {
 696         struct bch_dev *ca = bch2_dev_rcu(c, p->ptr.dev);
 697
 698         if (!ca || ca->mi.state == BCH_MEMBER_STATE_failed)
 699                 return 0;
 700
 701         return __extent_ptr_durability(ca, p);
 702 }
 703
 704 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 705 {
 706         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 707         const union bch_extent_entry *entry;
 708         struct extent_ptr_decoded p;
 709         unsigned durability = 0;
 710
 711         rcu_read_lock();
 712         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 713                 durability += bch2_extent_ptr_durability(c, &p);
 714         rcu_read_unlock();
 715
 716         return durability;
 717 }
 718
 719 static unsigned bch2_bkey_durability_safe(struct bch_fs *c, struct bkey_s_c k)
 720 {
 721         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 722         const union bch_extent_entry *entry;
 723         struct extent_ptr_decoded p;
 724         unsigned durability = 0;
 725
 726         rcu_read_lock();
 727         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 728                 if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 729                         durability += bch2_extent_ptr_durability(c, &p);
 730         rcu_read_unlock();
 731
 732         return durability;
 733 }
 734
 735 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
 736 {
 737         union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
 738         union bch_extent_entry *next = extent_entry_next(entry);
 739
 740         memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
 741         k->k.u64s -= extent_entry_u64s(entry);
 742 }
 743
 744 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
 745                                     struct extent_ptr_decoded *p)
 746 {
 747         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 748         struct bch_extent_crc_unpacked crc =
 749                 bch2_extent_crc_unpack(&k->k, NULL);
 750         union bch_extent_entry *pos;
 751
 752         if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 753                 pos = ptrs.start;
 754                 goto found;
 755         }
 756
 757         bkey_for_each_crc(&k->k, ptrs, crc, pos)
 758                 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 759                         pos = extent_entry_next(pos);
 760                         goto found;
 761                 }
 762
 763         bch2_extent_crc_append(k, p->crc);
 764         pos = bkey_val_end(bkey_i_to_s(k));
 765 found:
 766         p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
 767         __extent_entry_insert(k, pos, to_entry(&p->ptr));
 768
 769         if (p->has_ec) {
 770                 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
 771                 __extent_entry_insert(k, pos, to_entry(&p->ec));
 772         }
 773 }
 774
 775 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 776                                           union bch_extent_entry *entry)
 777 {
 778         union bch_extent_entry *i = ptrs.start;
 779
 780         if (i == entry)
 781                 return NULL;
 782
 783         while (extent_entry_next(i) != entry)
 784                 i = extent_entry_next(i);
 785         return i;
 786 }
 787
 788 /*
 789  * Returns pointer to the next entry after the one being dropped:
 790  */
 791 void bch2_bkey_drop_ptr_noerror(struct bkey_s k, struct bch_extent_ptr *ptr)
 792 {
 793         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 794         union bch_extent_entry *entry = to_entry(ptr), *next;
 795         bool drop_crc = true;
 796
 797         if (k.k->type == KEY_TYPE_stripe) {
 798                 ptr->dev = BCH_SB_MEMBER_INVALID;
 799                 return;
 800         }
 801
 802         EBUG_ON(ptr < &ptrs.start->ptr ||
 803                 ptr >= &ptrs.end->ptr);
 804         EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
 805
 806         for (next = extent_entry_next(entry);
 807              next != ptrs.end;
 808              next = extent_entry_next(next)) {
 809                 if (extent_entry_is_crc(next)) {
 810                         break;
 811                 } else if (extent_entry_is_ptr(next)) {
 812                         drop_crc = false;
 813                         break;
 814                 }
 815         }
 816
 817         extent_entry_drop(k, entry);
 818
 819         while ((entry = extent_entry_prev(ptrs, entry))) {
 820                 if (extent_entry_is_ptr(entry))
 821                         break;
 822
 823                 if ((extent_entry_is_crc(entry) && drop_crc) ||
 824                     extent_entry_is_stripe_ptr(entry))
 825                         extent_entry_drop(k, entry);
 826         }
 827 }
 828
 829 void bch2_bkey_drop_ptr(struct bkey_s k, struct bch_extent_ptr *ptr)
 830 {
 831         if (k.k->type != KEY_TYPE_stripe) {
 832                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k.s_c);
 833                 const union bch_extent_entry *entry;
 834                 struct extent_ptr_decoded p;
 835
 836                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 837                         if (p.ptr.dev == ptr->dev && p.has_ec) {
 838                                 ptr->dev = BCH_SB_MEMBER_INVALID;
 839                                 return;
 840                         }
 841         }
 842
 843         bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
 844
 845         bch2_bkey_drop_ptr_noerror(k, ptr);
 846
 847         /*
 848          * If we deleted all the dirty pointers and there's still cached
 849          * pointers, we could set the cached pointers to dirty if they're not
 850          * stale - but to do that correctly we'd need to grab an open_bucket
 851          * reference so that we don't race with bucket reuse:
 852          */
 853         if (have_dirty &&
 854             !bch2_bkey_dirty_devs(k.s_c).nr) {
 855                 k.k->type = KEY_TYPE_error;
 856                 set_bkey_val_u64s(k.k, 0);
 857         } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 858                 k.k->type = KEY_TYPE_deleted;
 859                 set_bkey_val_u64s(k.k, 0);
 860         }
 861 }
 862
 863 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 864 {
 865         bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 866 }
 867
 868 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 869 {
 870         bch2_bkey_drop_ptrs_noerror(k, ptr, ptr->dev == dev);
 871 }
 872
 873 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
 874 {
 875         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 876
 877         bkey_for_each_ptr(ptrs, ptr)
 878                 if (ptr->dev == dev)
 879                         return ptr;
 880
 881         return NULL;
 882 }
 883
 884 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 885 {
 886         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 887         struct bch_dev *ca;
 888         bool ret = false;
 889
 890         rcu_read_lock();
 891         bkey_for_each_ptr(ptrs, ptr)
 892                 if (bch2_dev_in_target(c, ptr->dev, target) &&
 893                     (ca = bch2_dev_rcu(c, ptr->dev)) &&
 894                     (!ptr->cached ||
 895                      !dev_ptr_stale_rcu(ca, ptr))) {
 896                         ret = true;
 897                         break;
 898                 }
 899         rcu_read_unlock();
 900
 901         return ret;
 902 }
 903
 904 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 905                            struct bch_extent_ptr m, u64 offset)
 906 {
 907         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 908         const union bch_extent_entry *entry;
 909         struct extent_ptr_decoded p;
 910
 911         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 912                 if (p.ptr.dev   == m.dev &&
 913                     p.ptr.gen   == m.gen &&
 914                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 915                     (s64) m.offset  - offset)
 916                         return true;
 917
 918         return false;
 919 }
 920
 921 /*
 922  * Returns true if two extents refer to the same data:
 923  */
 924 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 925 {
 926         if (k1.k->type != k2.k->type)
 927                 return false;
 928
 929         if (bkey_extent_is_direct_data(k1.k)) {
 930                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 931                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 932                 const union bch_extent_entry *entry1, *entry2;
 933                 struct extent_ptr_decoded p1, p2;
 934
 935                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 936                         return false;
 937
 938                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 939                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 940                                 if (p1.ptr.dev          == p2.ptr.dev &&
 941                                     p1.ptr.gen          == p2.ptr.gen &&
 942
 943                                     /*
 944                                      * This checks that the two pointers point
 945                                      * to the same region on disk - adjusting
 946                                      * for the difference in where the extents
 947                                      * start, since one may have been trimmed:
 948                                      */
 949                                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 950                                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k) &&
 951
 952                                     /*
 953                                      * This additionally checks that the
 954                                      * extents overlap on disk, since the
 955                                      * previous check may trigger spuriously
 956                                      * when one extent is immediately partially
 957                                      * overwritten with another extent (so that
 958                                      * on disk they are adjacent) and
 959                                      * compression is in use:
 960                                      */
 961                                     ((p1.ptr.offset >= p2.ptr.offset &&
 962                                       p1.ptr.offset  < p2.ptr.offset + p2.crc.compressed_size) ||
 963                                      (p2.ptr.offset >= p1.ptr.offset &&
 964                                       p2.ptr.offset  < p1.ptr.offset + p1.crc.compressed_size)))
 965                                         return true;
 966
 967                 return false;
 968         } else {
 969                 /* KEY_TYPE_deleted, etc. */
 970                 return true;
 971         }
 972 }
 973
 974 struct bch_extent_ptr *
 975 bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bkey_s k2)
 976 {
 977         struct bkey_ptrs ptrs2 = bch2_bkey_ptrs(k2);
 978         union bch_extent_entry *entry2;
 979         struct extent_ptr_decoded p2;
 980
 981         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 982                 if (p1.ptr.dev          == p2.ptr.dev &&
 983                     p1.ptr.gen          == p2.ptr.gen &&
 984                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 985                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 986                         return &entry2->ptr;
 987
 988         return NULL;
 989 }
 990
 991 static bool want_cached_ptr(struct bch_fs *c, struct bch_io_opts *opts,
 992                             struct bch_extent_ptr *ptr)
 993 {
 994         if (!opts->promote_target ||
 995             !bch2_dev_in_target(c, ptr->dev, opts->promote_target))
 996                 return false;
 997
 998         struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
 999
1000         return ca && bch2_dev_is_readable(ca) && !dev_ptr_stale_rcu(ca, ptr);
1001 }
1002
1003 void bch2_extent_ptr_set_cached(struct bch_fs *c,
1004                                 struct bch_io_opts *opts,
1005                                 struct bkey_s k,
1006                                 struct bch_extent_ptr *ptr)
1007 {
1008         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1009         union bch_extent_entry *entry;
1010         struct extent_ptr_decoded p;
1011
1012         rcu_read_lock();
1013         if (!want_cached_ptr(c, opts, ptr)) {
1014                 bch2_bkey_drop_ptr_noerror(k, ptr);
1015                 goto out;
1016         }
1017
1018         /*
1019          * Stripes can't contain cached data, for - reasons.
1020          *
1021          * Possibly something we can fix in the future?
1022          */
1023         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
1024                 if (&entry->ptr == ptr) {
1025                         if (p.has_ec)
1026                                 bch2_bkey_drop_ptr_noerror(k, ptr);
1027                         else
1028                                 ptr->cached = true;
1029                         goto out;
1030                 }
1031
1032         BUG();
1033 out:
1034         rcu_read_unlock();
1035 }
1036
1037 /*
1038  * bch2_extent_normalize - clean up an extent, dropping stale pointers etc.
1039  *
1040  * Returns true if @k should be dropped entirely
1041  *
1042  * For existing keys, only called when btree nodes are being rewritten, not when
1043  * they're merely being compacted/resorted in memory.
1044  */
1045 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
1046 {
1047         struct bch_dev *ca;
1048
1049         rcu_read_lock();
1050         bch2_bkey_drop_ptrs(k, ptr,
1051                 ptr->cached &&
1052                 (!(ca = bch2_dev_rcu(c, ptr->dev)) ||
1053                  dev_ptr_stale_rcu(ca, ptr) > 0));
1054         rcu_read_unlock();
1055
1056         return bkey_deleted(k.k);
1057 }
1058
1059 /*
1060  * bch2_extent_normalize_by_opts - clean up an extent, dropping stale pointers etc.
1061  *
1062  * Like bch2_extent_normalize(), but also only keeps a single cached pointer on
1063  * the promote target.
1064  */
1065 bool bch2_extent_normalize_by_opts(struct bch_fs *c,
1066                                    struct bch_io_opts *opts,
1067                                    struct bkey_s k)
1068 {
1069         struct bkey_ptrs ptrs;
1070         bool have_cached_ptr;
1071
1072         rcu_read_lock();
1073 restart_drop_ptrs:
1074         ptrs = bch2_bkey_ptrs(k);
1075         have_cached_ptr = false;
1076
1077         bkey_for_each_ptr(ptrs, ptr)
1078                 if (ptr->cached) {
1079                         if (have_cached_ptr || !want_cached_ptr(c, opts, ptr)) {
1080                                 bch2_bkey_drop_ptr(k, ptr);
1081                                 goto restart_drop_ptrs;
1082                         }
1083                         have_cached_ptr = true;
1084                 }
1085         rcu_read_unlock();
1086
1087         return bkey_deleted(k.k);
1088 }
1089
1090 void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *c, const struct bch_extent_ptr *ptr)
1091 {
1092         out->atomic++;
1093         rcu_read_lock();
1094         struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
1095         if (!ca) {
1096                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1097                            (u64) ptr->offset, ptr->gen,
1098                            ptr->cached ? " cached" : "");
1099         } else {
1100                 u32 offset;
1101                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1102
1103                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1104                            ptr->dev, b, offset, ptr->gen);
1105                 if (ca->mi.durability != 1)
1106                         prt_printf(out, " d=%u", ca->mi.durability);
1107                 if (ptr->cached)
1108                         prt_str(out, " cached");
1109                 if (ptr->unwritten)
1110                         prt_str(out, " unwritten");
1111                 int stale = dev_ptr_stale_rcu(ca, ptr);
1112                 if (stale > 0)
1113                         prt_printf(out, " stale");
1114                 else if (stale)
1115                         prt_printf(out, " invalid");
1116         }
1117         rcu_read_unlock();
1118         --out->atomic;
1119 }
1120
1121 void bch2_extent_crc_unpacked_to_text(struct printbuf *out, struct bch_extent_crc_unpacked *crc)
1122 {
1123         prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum ",
1124                    crc->compressed_size,
1125                    crc->uncompressed_size,
1126                    crc->offset, crc->nonce);
1127         bch2_prt_csum_type(out, crc->csum_type);
1128         prt_printf(out, " %0llx:%0llx ", crc->csum.hi, crc->csum.lo);
1129         prt_str(out, " compress ");
1130         bch2_prt_compression_type(out, crc->compression_type);
1131 }
1132
1133 static void bch2_extent_rebalance_to_text(struct printbuf *out, struct bch_fs *c,
1134                                           const struct bch_extent_rebalance *r)
1135 {
1136         prt_str(out, "rebalance:");
1137
1138         prt_printf(out, " replicas=%u", r->data_replicas);
1139         if (r->data_replicas_from_inode)
1140                 prt_str(out, " (inode)");
1141
1142         prt_str(out, " checksum=");
1143         bch2_prt_csum_opt(out, r->data_checksum);
1144         if (r->data_checksum_from_inode)
1145                 prt_str(out, " (inode)");
1146
1147         if (r->background_compression || r->background_compression_from_inode) {
1148                 prt_str(out, " background_compression=");
1149                 bch2_compression_opt_to_text(out, r->background_compression);
1150
1151                 if (r->background_compression_from_inode)
1152                         prt_str(out, " (inode)");
1153         }
1154
1155         if (r->background_target || r->background_target_from_inode) {
1156                 prt_str(out, " background_target=");
1157                 if (c)
1158                         bch2_target_to_text(out, c, r->background_target);
1159                 else
1160                         prt_printf(out, "%u", r->background_target);
1161
1162                 if (r->background_target_from_inode)
1163                         prt_str(out, " (inode)");
1164         }
1165
1166         if (r->promote_target || r->promote_target_from_inode) {
1167                 prt_str(out, " promote_target=");
1168                 if (c)
1169                         bch2_target_to_text(out, c, r->promote_target);
1170                 else
1171                         prt_printf(out, "%u", r->promote_target);
1172
1173                 if (r->promote_target_from_inode)
1174                         prt_str(out, " (inode)");
1175         }
1176
1177         if (r->erasure_code || r->erasure_code_from_inode) {
1178                 prt_printf(out, " ec=%u", r->erasure_code);
1179                 if (r->erasure_code_from_inode)
1180                         prt_str(out, " (inode)");
1181         }
1182 }
1183
1184 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
1185                             struct bkey_s_c k)
1186 {
1187         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1188         const union bch_extent_entry *entry;
1189         bool first = true;
1190
1191         if (c)
1192                 prt_printf(out, "durability: %u ", bch2_bkey_durability_safe(c, k));
1193
1194         bkey_extent_entry_for_each(ptrs, entry) {
1195                 if (!first)
1196                         prt_printf(out, " ");
1197
1198                 switch (__extent_entry_type(entry)) {
1199                 case BCH_EXTENT_ENTRY_ptr:
1200                         bch2_extent_ptr_to_text(out, c, entry_to_ptr(entry));
1201                         break;
1202
1203                 case BCH_EXTENT_ENTRY_crc32:
1204                 case BCH_EXTENT_ENTRY_crc64:
1205                 case BCH_EXTENT_ENTRY_crc128: {
1206                         struct bch_extent_crc_unpacked crc =
1207                                 bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1208
1209                         bch2_extent_crc_unpacked_to_text(out, &crc);
1210                         break;
1211                 }
1212                 case BCH_EXTENT_ENTRY_stripe_ptr: {
1213                         const struct bch_extent_stripe_ptr *ec = &entry->stripe_ptr;
1214
1215                         prt_printf(out, "ec: idx %llu block %u",
1216                                (u64) ec->idx, ec->block);
1217                         break;
1218                 }
1219                 case BCH_EXTENT_ENTRY_rebalance:
1220                         bch2_extent_rebalance_to_text(out, c, &entry->rebalance);
1221                         break;
1222
1223                 default:
1224                         prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1225                         return;
1226                 }
1227
1228                 first = false;
1229         }
1230 }
1231
1232 static int extent_ptr_validate(struct bch_fs *c,
1233                                struct bkey_s_c k,
1234                                struct bkey_validate_context from,
1235                                const struct bch_extent_ptr *ptr,
1236                                unsigned size_ondisk,
1237                                bool metadata)
1238 {
1239         int ret = 0;
1240
1241         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1242         bkey_for_each_ptr(ptrs, ptr2)
1243                 bkey_fsck_err_on(ptr != ptr2 && ptr->dev == ptr2->dev,
1244                                  c, ptr_to_duplicate_device,
1245                                  "multiple pointers to same device (%u)", ptr->dev);
1246
1247         /* bad pointers are repaired by check_fix_ptrs(): */
1248         rcu_read_lock();
1249         struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
1250         if (!ca) {
1251                 rcu_read_unlock();
1252                 return 0;
1253         }
1254         u32 bucket_offset;
1255         u64 bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1256         unsigned first_bucket   = ca->mi.first_bucket;
1257         u64 nbuckets            = ca->mi.nbuckets;
1258         unsigned bucket_size    = ca->mi.bucket_size;
1259         rcu_read_unlock();
1260
1261         bkey_fsck_err_on(bucket >= nbuckets,
1262                          c, ptr_after_last_bucket,
1263                          "pointer past last bucket (%llu > %llu)", bucket, nbuckets);
1264         bkey_fsck_err_on(bucket < first_bucket,
1265                          c, ptr_before_first_bucket,
1266                          "pointer before first bucket (%llu < %u)", bucket, first_bucket);
1267         bkey_fsck_err_on(bucket_offset + size_ondisk > bucket_size,
1268                          c, ptr_spans_multiple_buckets,
1269                          "pointer spans multiple buckets (%u + %u > %u)",
1270                        bucket_offset, size_ondisk, bucket_size);
1271 fsck_err:
1272         return ret;
1273 }
1274
1275 int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
1276                             struct bkey_validate_context from)
1277 {
1278         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1279         const union bch_extent_entry *entry;
1280         struct bch_extent_crc_unpacked crc;
1281         unsigned size_ondisk = k.k->size;
1282         unsigned nonce = UINT_MAX;
1283         unsigned nr_ptrs = 0;
1284         bool have_written = false, have_unwritten = false, have_ec = false, crc_since_last_ptr = false;
1285         int ret = 0;
1286
1287         if (bkey_is_btree_ptr(k.k))
1288                 size_ondisk = btree_sectors(c);
1289
1290         bkey_extent_entry_for_each(ptrs, entry) {
1291                 bkey_fsck_err_on(__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX,
1292                                  c, extent_ptrs_invalid_entry,
1293                                  "invalid extent entry type (got %u, max %u)",
1294                                  __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1295
1296                 bkey_fsck_err_on(bkey_is_btree_ptr(k.k) &&
1297                                  !extent_entry_is_ptr(entry),
1298                                  c, btree_ptr_has_non_ptr,
1299                                  "has non ptr field");
1300
1301                 switch (extent_entry_type(entry)) {
1302                 case BCH_EXTENT_ENTRY_ptr:
1303                         ret = extent_ptr_validate(c, k, from, &entry->ptr, size_ondisk, false);
1304                         if (ret)
1305                                 return ret;
1306
1307                         bkey_fsck_err_on(entry->ptr.cached && have_ec,
1308                                          c, ptr_cached_and_erasure_coded,
1309                                          "cached, erasure coded ptr");
1310
1311                         if (!entry->ptr.unwritten)
1312                                 have_written = true;
1313                         else
1314                                 have_unwritten = true;
1315
1316                         have_ec = false;
1317                         crc_since_last_ptr = false;
1318                         nr_ptrs++;
1319                         break;
1320                 case BCH_EXTENT_ENTRY_crc32:
1321                 case BCH_EXTENT_ENTRY_crc64:
1322                 case BCH_EXTENT_ENTRY_crc128:
1323                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1324
1325                         bkey_fsck_err_on(!bch2_checksum_type_valid(c, crc.csum_type),
1326                                          c, ptr_crc_csum_type_unknown,
1327                                          "invalid checksum type");
1328                         bkey_fsck_err_on(crc.compression_type >= BCH_COMPRESSION_TYPE_NR,
1329                                          c, ptr_crc_compression_type_unknown,
1330                                          "invalid compression type");
1331
1332                         bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size,
1333                                          c, ptr_crc_uncompressed_size_too_small,
1334                                          "checksum offset + key size > uncompressed size");
1335                         bkey_fsck_err_on(crc_is_encoded(crc) &&
1336                                          (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
1337                                          (from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit)),
1338                                          c, ptr_crc_uncompressed_size_too_big,
1339                                          "too large encoded extent");
1340                         bkey_fsck_err_on(!crc_is_compressed(crc) &&
1341                                          crc.compressed_size != crc.uncompressed_size,
1342                                          c, ptr_crc_uncompressed_size_mismatch,
1343                                          "not compressed but compressed != uncompressed size");
1344
1345                         if (bch2_csum_type_is_encryption(crc.csum_type)) {
1346                                 if (nonce == UINT_MAX)
1347                                         nonce = crc.offset + crc.nonce;
1348                                 else if (nonce != crc.offset + crc.nonce)
1349                                         bkey_fsck_err(c, ptr_crc_nonce_mismatch,
1350                                                       "incorrect nonce");
1351                         }
1352
1353                         bkey_fsck_err_on(crc_since_last_ptr,
1354                                          c, ptr_crc_redundant,
1355                                          "redundant crc entry");
1356                         crc_since_last_ptr = true;
1357
1358                         size_ondisk = crc.compressed_size;
1359                         break;
1360                 case BCH_EXTENT_ENTRY_stripe_ptr:
1361                         bkey_fsck_err_on(have_ec,
1362                                          c, ptr_stripe_redundant,
1363                                          "redundant stripe entry");
1364                         have_ec = true;
1365                         break;
1366                 case BCH_EXTENT_ENTRY_rebalance: {
1367                         /*
1368                          * this shouldn't be a fsck error, for forward
1369                          * compatibility; the rebalance code should just refetch
1370                          * the compression opt if it's unknown
1371                          */
1372 #if 0
1373                         const struct bch_extent_rebalance *r = &entry->rebalance;
1374
1375                         if (!bch2_compression_opt_valid(r->compression)) {
1376                                 struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
1377                                 prt_printf(err, "invalid compression opt %u:%u",
1378                                            opt.type, opt.level);
1379                                 return -BCH_ERR_invalid_bkey;
1380                         }
1381 #endif
1382                         break;
1383                 }
1384                 }
1385         }
1386
1387         bkey_fsck_err_on(!nr_ptrs,
1388                          c, extent_ptrs_no_ptrs,
1389                          "no ptrs");
1390         bkey_fsck_err_on(nr_ptrs > BCH_BKEY_PTRS_MAX,
1391                          c, extent_ptrs_too_many_ptrs,
1392                          "too many ptrs: %u > %u", nr_ptrs, BCH_BKEY_PTRS_MAX);
1393         bkey_fsck_err_on(have_written && have_unwritten,
1394                          c, extent_ptrs_written_and_unwritten,
1395                          "extent with unwritten and written ptrs");
1396         bkey_fsck_err_on(k.k->type != KEY_TYPE_extent && have_unwritten,
1397                          c, extent_ptrs_unwritten,
1398                          "has unwritten ptrs");
1399         bkey_fsck_err_on(crc_since_last_ptr,
1400                          c, extent_ptrs_redundant_crc,
1401                          "redundant crc entry");
1402         bkey_fsck_err_on(have_ec,
1403                          c, extent_ptrs_redundant_stripe,
1404                          "redundant stripe entry");
1405 fsck_err:
1406         return ret;
1407 }
1408
1409 void bch2_ptr_swab(struct bkey_s k)
1410 {
1411         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1412         union bch_extent_entry *entry;
1413         u64 *d;
1414
1415         for (d =  (u64 *) ptrs.start;
1416              d != (u64 *) ptrs.end;
1417              d++)
1418                 *d = swab64(*d);
1419
1420         for (entry = ptrs.start;
1421              entry < ptrs.end;
1422              entry = extent_entry_next(entry)) {
1423                 switch (__extent_entry_type(entry)) {
1424                 case BCH_EXTENT_ENTRY_ptr:
1425                         break;
1426                 case BCH_EXTENT_ENTRY_crc32:
1427                         entry->crc32.csum = swab32(entry->crc32.csum);
1428                         break;
1429                 case BCH_EXTENT_ENTRY_crc64:
1430                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1431                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1432                         break;
1433                 case BCH_EXTENT_ENTRY_crc128:
1434                         entry->crc128.csum.hi = (__force __le64)
1435                                 swab64((__force u64) entry->crc128.csum.hi);
1436                         entry->crc128.csum.lo = (__force __le64)
1437                                 swab64((__force u64) entry->crc128.csum.lo);
1438                         break;
1439                 case BCH_EXTENT_ENTRY_stripe_ptr:
1440                         break;
1441                 case BCH_EXTENT_ENTRY_rebalance:
1442                         break;
1443                 default:
1444                         /* Bad entry type: will be caught by validate() */
1445                         return;
1446                 }
1447         }
1448 }
1449
1450 /* Generic extent code: */
1451
1452 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1453 {
1454         unsigned new_val_u64s = bkey_val_u64s(k.k);
1455         int val_u64s_delta;
1456         u64 sub;
1457
1458         if (bkey_le(where, bkey_start_pos(k.k)))
1459                 return 0;
1460
1461         EBUG_ON(bkey_gt(where, k.k->p));
1462
1463         sub = where.offset - bkey_start_offset(k.k);
1464
1465         k.k->size -= sub;
1466
1467         if (!k.k->size) {
1468                 k.k->type = KEY_TYPE_deleted;
1469                 new_val_u64s = 0;
1470         }
1471
1472         switch (k.k->type) {
1473         case KEY_TYPE_extent:
1474         case KEY_TYPE_reflink_v: {
1475                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1476                 union bch_extent_entry *entry;
1477                 bool seen_crc = false;
1478
1479                 bkey_extent_entry_for_each(ptrs, entry) {
1480                         switch (extent_entry_type(entry)) {
1481                         case BCH_EXTENT_ENTRY_ptr:
1482                                 if (!seen_crc)
1483                                         entry->ptr.offset += sub;
1484                                 break;
1485                         case BCH_EXTENT_ENTRY_crc32:
1486                                 entry->crc32.offset += sub;
1487                                 break;
1488                         case BCH_EXTENT_ENTRY_crc64:
1489                                 entry->crc64.offset += sub;
1490                                 break;
1491                         case BCH_EXTENT_ENTRY_crc128:
1492                                 entry->crc128.offset += sub;
1493                                 break;
1494                         case BCH_EXTENT_ENTRY_stripe_ptr:
1495                                 break;
1496                         case BCH_EXTENT_ENTRY_rebalance:
1497                                 break;
1498                         }
1499
1500                         if (extent_entry_is_crc(entry))
1501                                 seen_crc = true;
1502                 }
1503
1504                 break;
1505         }
1506         case KEY_TYPE_reflink_p: {
1507                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1508
1509                 SET_REFLINK_P_IDX(p.v, REFLINK_P_IDX(p.v) + sub);
1510                 break;
1511         }
1512         case KEY_TYPE_inline_data:
1513         case KEY_TYPE_indirect_inline_data: {
1514                 void *p = bkey_inline_data_p(k);
1515                 unsigned bytes = bkey_inline_data_bytes(k.k);
1516
1517                 sub = min_t(u64, sub << 9, bytes);
1518
1519                 memmove(p, p + sub, bytes - sub);
1520
1521                 new_val_u64s -= sub >> 3;
1522                 break;
1523         }
1524         }
1525
1526         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1527         BUG_ON(val_u64s_delta < 0);
1528
1529         set_bkey_val_u64s(k.k, new_val_u64s);
1530         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1531         return -val_u64s_delta;
1532 }
1533
1534 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1535 {
1536         unsigned new_val_u64s = bkey_val_u64s(k.k);
1537         int val_u64s_delta;
1538         u64 len = 0;
1539
1540         if (bkey_ge(where, k.k->p))
1541                 return 0;
1542
1543         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1544
1545         len = where.offset - bkey_start_offset(k.k);
1546
1547         k.k->p.offset = where.offset;
1548         k.k->size = len;
1549
1550         if (!len) {
1551                 k.k->type = KEY_TYPE_deleted;
1552                 new_val_u64s = 0;
1553         }
1554
1555         switch (k.k->type) {
1556         case KEY_TYPE_inline_data:
1557         case KEY_TYPE_indirect_inline_data:
1558                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1559                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1560                 break;
1561         }
1562
1563         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1564         BUG_ON(val_u64s_delta < 0);
1565
1566         set_bkey_val_u64s(k.k, new_val_u64s);
1567         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1568         return -val_u64s_delta;
1569 }