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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
   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_gc.h"
  12 #include "btree_io.h"
  13 #include "btree_iter.h"
  14 #include "buckets.h"
  15 #include "checksum.h"
  16 #include "debug.h"
  17 #include "disk_groups.h"
  18 #include "error.h"
  19 #include "extents.h"
  20 #include "inode.h"
  21 #include "journal.h"
  22 #include "replicas.h"
  23 #include "super.h"
  24 #include "super-io.h"
  25 #include "trace.h"
  26 #include "util.h"
  27
  28 static unsigned bch2_crc_field_size_max[] = {
  29         [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
  30         [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
  31         [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
  32 };
  33
  34 static void bch2_extent_crc_pack(union bch_extent_crc *,
  35                                  struct bch_extent_crc_unpacked,
  36                                  enum bch_extent_entry_type);
  37
  38 static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
  39                                                    unsigned dev)
  40 {
  41         struct bch_dev_io_failures *i;
  42
  43         for (i = f->devs; i < f->devs + f->nr; i++)
  44                 if (i->dev == dev)
  45                         return i;
  46
  47         return NULL;
  48 }
  49
  50 void bch2_mark_io_failure(struct bch_io_failures *failed,
  51                           struct extent_ptr_decoded *p)
  52 {
  53         struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
  54
  55         if (!f) {
  56                 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
  57
  58                 f = &failed->devs[failed->nr++];
  59                 f->dev          = p->ptr.dev;
  60                 f->idx          = p->idx;
  61                 f->nr_failed    = 1;
  62                 f->nr_retries   = 0;
  63         } else if (p->idx != f->idx) {
  64                 f->idx          = p->idx;
  65                 f->nr_failed    = 1;
  66                 f->nr_retries   = 0;
  67         } else {
  68                 f->nr_failed++;
  69         }
  70 }
  71
  72 /*
  73  * returns true if p1 is better than p2:
  74  */
  75 static inline bool ptr_better(struct bch_fs *c,
  76                               const struct extent_ptr_decoded p1,
  77                               const struct extent_ptr_decoded p2)
  78 {
  79         if (likely(!p1.idx && !p2.idx)) {
  80                 struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
  81                 struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
  82
  83                 u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
  84                 u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
  85
  86                 /* Pick at random, biased in favor of the faster device: */
  87
  88                 return bch2_rand_range(l1 + l2) > l1;
  89         }
  90
  91         if (bch2_force_reconstruct_read)
  92                 return p1.idx > p2.idx;
  93
  94         return p1.idx < p2.idx;
  95 }
  96
  97 /*
  98  * This picks a non-stale pointer, preferably from a device other than @avoid.
  99  * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
 100  * other devices, it will still pick a pointer from avoid.
 101  */
 102 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 103                                struct bch_io_failures *failed,
 104                                struct extent_ptr_decoded *pick)
 105 {
 106         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 107         const union bch_extent_entry *entry;
 108         struct extent_ptr_decoded p;
 109         struct bch_dev_io_failures *f;
 110         struct bch_dev *ca;
 111         int ret = 0;
 112
 113         if (k.k->type == KEY_TYPE_error)
 114                 return -EIO;
 115
 116         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 117                 /*
 118                  * Unwritten extent: no need to actually read, treat it as a
 119                  * hole and return 0s:
 120                  */
 121                 if (p.ptr.unwritten)
 122                         return 0;
 123
 124                 ca = bch_dev_bkey_exists(c, p.ptr.dev);
 125
 126                 /*
 127                  * If there are any dirty pointers it's an error if we can't
 128                  * read:
 129                  */
 130                 if (!ret && !p.ptr.cached)
 131                         ret = -EIO;
 132
 133                 if (p.ptr.cached && ptr_stale(ca, &p.ptr))
 134                         continue;
 135
 136                 f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
 137                 if (f)
 138                         p.idx = f->nr_failed < f->nr_retries
 139                                 ? f->idx
 140                                 : f->idx + 1;
 141
 142                 if (!p.idx &&
 143                     !bch2_dev_is_readable(ca))
 144                         p.idx++;
 145
 146                 if (bch2_force_reconstruct_read &&
 147                     !p.idx && p.has_ec)
 148                         p.idx++;
 149
 150                 if (p.idx >= (unsigned) p.has_ec + 1)
 151                         continue;
 152
 153                 if (ret > 0 && !ptr_better(c, p, *pick))
 154                         continue;
 155
 156                 *pick = p;
 157                 ret = 1;
 158         }
 159
 160         return ret;
 161 }
 162
 163 /* KEY_TYPE_btree_ptr: */
 164
 165 int bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k,
 166                            unsigned flags, struct printbuf *err)
 167 {
 168         if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX) {
 169                 prt_printf(err, "value too big (%zu > %u)",
 170                        bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 171                 return -BCH_ERR_invalid_bkey;
 172         }
 173
 174         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 175 }
 176
 177 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 178                             struct bkey_s_c k)
 179 {
 180         bch2_bkey_ptrs_to_text(out, c, k);
 181 }
 182
 183 int bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 184                               unsigned flags, struct printbuf *err)
 185 {
 186         if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) {
 187                 prt_printf(err, "value too big (%zu > %zu)",
 188                        bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 189                 return -BCH_ERR_invalid_bkey;
 190         }
 191
 192         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 193 }
 194
 195 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 196                                struct bkey_s_c k)
 197 {
 198         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 199
 200         prt_printf(out, "seq %llx written %u min_key %s",
 201                le64_to_cpu(bp.v->seq),
 202                le16_to_cpu(bp.v->sectors_written),
 203                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 204
 205         bch2_bpos_to_text(out, bp.v->min_key);
 206         prt_printf(out, " ");
 207         bch2_bkey_ptrs_to_text(out, c, k);
 208 }
 209
 210 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 211                               unsigned big_endian, int write,
 212                               struct bkey_s k)
 213 {
 214         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 215
 216         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 217
 218         if (version < bcachefs_metadata_version_inode_btree_change &&
 219             btree_node_type_is_extents(btree_id) &&
 220             !bkey_eq(bp.v->min_key, POS_MIN))
 221                 bp.v->min_key = write
 222                         ? bpos_nosnap_predecessor(bp.v->min_key)
 223                         : bpos_nosnap_successor(bp.v->min_key);
 224 }
 225
 226 /* KEY_TYPE_extent: */
 227
 228 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 229 {
 230         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 231         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 232         union bch_extent_entry *en_l;
 233         const union bch_extent_entry *en_r;
 234         struct extent_ptr_decoded lp, rp;
 235         bool use_right_ptr;
 236         struct bch_dev *ca;
 237
 238         en_l = l_ptrs.start;
 239         en_r = r_ptrs.start;
 240         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 241                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 242                         return false;
 243
 244                 en_l = extent_entry_next(en_l);
 245                 en_r = extent_entry_next(en_r);
 246         }
 247
 248         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 249                 return false;
 250
 251         en_l = l_ptrs.start;
 252         en_r = r_ptrs.start;
 253         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 254         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 255
 256         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 257                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 258                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 259                     rp.ptr.offset + rp.crc.offset ||
 260                     lp.ptr.dev                  != rp.ptr.dev ||
 261                     lp.ptr.gen                  != rp.ptr.gen ||
 262                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 263                     lp.has_ec                   != rp.has_ec)
 264                         return false;
 265
 266                 /* Extents may not straddle buckets: */
 267                 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
 268                 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
 269                         return false;
 270
 271                 if (lp.has_ec                   != rp.has_ec ||
 272                     (lp.has_ec &&
 273                      (lp.ec.block               != rp.ec.block ||
 274                       lp.ec.redundancy          != rp.ec.redundancy ||
 275                       lp.ec.idx                 != rp.ec.idx)))
 276                         return false;
 277
 278                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 279                     lp.crc.nonce                != rp.crc.nonce)
 280                         return false;
 281
 282                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 283                     lp.crc.uncompressed_size) {
 284                         /* can use left extent's crc entry */
 285                 } else if (lp.crc.live_size <= rp.crc.offset) {
 286                         /* can use right extent's crc entry */
 287                 } else {
 288                         /* check if checksums can be merged: */
 289                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 290                             lp.crc.nonce                != rp.crc.nonce ||
 291                             crc_is_compressed(lp.crc) ||
 292                             !bch2_checksum_mergeable(lp.crc.csum_type))
 293                                 return false;
 294
 295                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 296                             rp.crc.offset)
 297                                 return false;
 298
 299                         if (lp.crc.csum_type &&
 300                             lp.crc.uncompressed_size +
 301                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 302                                 return false;
 303                 }
 304
 305                 en_l = extent_entry_next(en_l);
 306                 en_r = extent_entry_next(en_r);
 307         }
 308
 309         en_l = l_ptrs.start;
 310         en_r = r_ptrs.start;
 311         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 312                 if (extent_entry_is_crc(en_l)) {
 313                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 314                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 315
 316                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 317                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 318                                 return false;
 319                 }
 320
 321                 en_l = extent_entry_next(en_l);
 322                 en_r = extent_entry_next(en_r);
 323         }
 324
 325         use_right_ptr = false;
 326         en_l = l_ptrs.start;
 327         en_r = r_ptrs.start;
 328         while (en_l < l_ptrs.end) {
 329                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 330                     use_right_ptr)
 331                         en_l->ptr = en_r->ptr;
 332
 333                 if (extent_entry_is_crc(en_l)) {
 334                         struct bch_extent_crc_unpacked crc_l =
 335                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 336                         struct bch_extent_crc_unpacked crc_r =
 337                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 338
 339                         use_right_ptr = false;
 340
 341                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 342                             crc_l.uncompressed_size) {
 343                                 /* can use left extent's crc entry */
 344                         } else if (crc_l.live_size <= crc_r.offset) {
 345                                 /* can use right extent's crc entry */
 346                                 crc_r.offset -= crc_l.live_size;
 347                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 348                                                      extent_entry_type(en_l));
 349                                 use_right_ptr = true;
 350                         } else {
 351                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 352                                                                  crc_l.csum,
 353                                                                  crc_r.csum,
 354                                                                  crc_r.uncompressed_size << 9);
 355
 356                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 357                                 crc_l.compressed_size   += crc_r.compressed_size;
 358                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 359                                                      extent_entry_type(en_l));
 360                         }
 361                 }
 362
 363                 en_l = extent_entry_next(en_l);
 364                 en_r = extent_entry_next(en_r);
 365         }
 366
 367         bch2_key_resize(l.k, l.k->size + r.k->size);
 368         return true;
 369 }
 370
 371 /* KEY_TYPE_reservation: */
 372
 373 int bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k,
 374                              unsigned flags, struct printbuf *err)
 375 {
 376         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 377
 378         if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX) {
 379                 prt_printf(err, "invalid nr_replicas (%u)",
 380                        r.v->nr_replicas);
 381                 return -BCH_ERR_invalid_bkey;
 382         }
 383
 384         return 0;
 385 }
 386
 387 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 388                               struct bkey_s_c k)
 389 {
 390         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 391
 392         prt_printf(out, "generation %u replicas %u",
 393                le32_to_cpu(r.v->generation),
 394                r.v->nr_replicas);
 395 }
 396
 397 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 398 {
 399         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 400         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 401
 402         if (l.v->generation != r.v->generation ||
 403             l.v->nr_replicas != r.v->nr_replicas)
 404                 return false;
 405
 406         bch2_key_resize(l.k, l.k->size + r.k->size);
 407         return true;
 408 }
 409
 410 /* Extent checksum entries: */
 411
 412 /* returns true if not equal */
 413 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 414                                          struct bch_extent_crc_unpacked r)
 415 {
 416         return (l.csum_type             != r.csum_type ||
 417                 l.compression_type      != r.compression_type ||
 418                 l.compressed_size       != r.compressed_size ||
 419                 l.uncompressed_size     != r.uncompressed_size ||
 420                 l.offset                != r.offset ||
 421                 l.live_size             != r.live_size ||
 422                 l.nonce                 != r.nonce ||
 423                 bch2_crc_cmp(l.csum, r.csum));
 424 }
 425
 426 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 427                                   struct bch_extent_crc_unpacked n)
 428 {
 429         return !crc_is_compressed(u) &&
 430                 u.csum_type &&
 431                 u.uncompressed_size > u.live_size &&
 432                 bch2_csum_type_is_encryption(u.csum_type) ==
 433                 bch2_csum_type_is_encryption(n.csum_type);
 434 }
 435
 436 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 437                                  struct bch_extent_crc_unpacked n)
 438 {
 439         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 440         struct bch_extent_crc_unpacked crc;
 441         const union bch_extent_entry *i;
 442
 443         if (!n.csum_type)
 444                 return false;
 445
 446         bkey_for_each_crc(k.k, ptrs, crc, i)
 447                 if (can_narrow_crc(crc, n))
 448                         return true;
 449
 450         return false;
 451 }
 452
 453 /*
 454  * We're writing another replica for this extent, so while we've got the data in
 455  * memory we'll be computing a new checksum for the currently live data.
 456  *
 457  * If there are other replicas we aren't moving, and they are checksummed but
 458  * not compressed, we can modify them to point to only the data that is
 459  * currently live (so that readers won't have to bounce) while we've got the
 460  * checksum we need:
 461  */
 462 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 463 {
 464         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 465         struct bch_extent_crc_unpacked u;
 466         struct extent_ptr_decoded p;
 467         union bch_extent_entry *i;
 468         bool ret = false;
 469
 470         /* Find a checksum entry that covers only live data: */
 471         if (!n.csum_type) {
 472                 bkey_for_each_crc(&k->k, ptrs, u, i)
 473                         if (!crc_is_compressed(u) &&
 474                             u.csum_type &&
 475                             u.live_size == u.uncompressed_size) {
 476                                 n = u;
 477                                 goto found;
 478                         }
 479                 return false;
 480         }
 481 found:
 482         BUG_ON(crc_is_compressed(n));
 483         BUG_ON(n.offset);
 484         BUG_ON(n.live_size != k->k.size);
 485
 486 restart_narrow_pointers:
 487         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 488
 489         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 490                 if (can_narrow_crc(p.crc, n)) {
 491                         bch2_bkey_drop_ptr_noerror(bkey_i_to_s(k), &i->ptr);
 492                         p.ptr.offset += p.crc.offset;
 493                         p.crc = n;
 494                         bch2_extent_ptr_decoded_append(k, &p);
 495                         ret = true;
 496                         goto restart_narrow_pointers;
 497                 }
 498
 499         return ret;
 500 }
 501
 502 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 503                                  struct bch_extent_crc_unpacked src,
 504                                  enum bch_extent_entry_type type)
 505 {
 506 #define set_common_fields(_dst, _src)                                   \
 507                 _dst.type               = 1 << type;                    \
 508                 _dst.csum_type          = _src.csum_type,               \
 509                 _dst.compression_type   = _src.compression_type,        \
 510                 _dst._compressed_size   = _src.compressed_size - 1,     \
 511                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 512                 _dst.offset             = _src.offset
 513
 514         switch (type) {
 515         case BCH_EXTENT_ENTRY_crc32:
 516                 set_common_fields(dst->crc32, src);
 517                 dst->crc32.csum  = *((__le32 *) &src.csum.lo);
 518                 break;
 519         case BCH_EXTENT_ENTRY_crc64:
 520                 set_common_fields(dst->crc64, src);
 521                 dst->crc64.nonce        = src.nonce;
 522                 dst->crc64.csum_lo      = src.csum.lo;
 523                 dst->crc64.csum_hi      = *((__le16 *) &src.csum.hi);
 524                 break;
 525         case BCH_EXTENT_ENTRY_crc128:
 526                 set_common_fields(dst->crc128, src);
 527                 dst->crc128.nonce       = src.nonce;
 528                 dst->crc128.csum        = src.csum;
 529                 break;
 530         default:
 531                 BUG();
 532         }
 533 #undef set_common_fields
 534 }
 535
 536 void bch2_extent_crc_append(struct bkey_i *k,
 537                             struct bch_extent_crc_unpacked new)
 538 {
 539         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 540         union bch_extent_crc *crc = (void *) ptrs.end;
 541         enum bch_extent_entry_type type;
 542
 543         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 544             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 545             new.nonce                           <= CRC32_NONCE_MAX)
 546                 type = BCH_EXTENT_ENTRY_crc32;
 547         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 548                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 549                    new.nonce                    <= CRC64_NONCE_MAX)
 550                 type = BCH_EXTENT_ENTRY_crc64;
 551         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 552                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 553                    new.nonce                    <= CRC128_NONCE_MAX)
 554                 type = BCH_EXTENT_ENTRY_crc128;
 555         else
 556                 BUG();
 557
 558         bch2_extent_crc_pack(crc, new, type);
 559
 560         k->k.u64s += extent_entry_u64s(ptrs.end);
 561
 562         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 563 }
 564
 565 /* Generic code for keys with pointers: */
 566
 567 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 568 {
 569         return bch2_bkey_devs(k).nr;
 570 }
 571
 572 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 573 {
 574         return k.k->type == KEY_TYPE_reservation
 575                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 576                 : bch2_bkey_dirty_devs(k).nr;
 577 }
 578
 579 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 580 {
 581         unsigned ret = 0;
 582
 583         if (k.k->type == KEY_TYPE_reservation) {
 584                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 585         } else {
 586                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 587                 const union bch_extent_entry *entry;
 588                 struct extent_ptr_decoded p;
 589
 590                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 591                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 592         }
 593
 594         return ret;
 595 }
 596
 597 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 598 {
 599         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 600         const union bch_extent_entry *entry;
 601         struct extent_ptr_decoded p;
 602         unsigned ret = 0;
 603
 604         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 605                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 606                         ret += p.crc.compressed_size;
 607
 608         return ret;
 609 }
 610
 611 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
 612 {
 613         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 614         const union bch_extent_entry *entry;
 615         struct bch_extent_crc_unpacked crc;
 616
 617         bkey_for_each_crc(k.k, ptrs, crc, entry)
 618                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 619                         return true;
 620         return false;
 621 }
 622
 623 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 624 {
 625         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 626         const union bch_extent_entry *entry;
 627         struct extent_ptr_decoded p = { 0 };
 628         unsigned replicas = 0;
 629
 630         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 631                 if (p.ptr.cached)
 632                         continue;
 633
 634                 if (p.has_ec)
 635                         replicas += p.ec.redundancy;
 636
 637                 replicas++;
 638
 639         }
 640
 641         return replicas;
 642 }
 643
 644 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 645 {
 646         unsigned durability = 0;
 647         struct bch_dev *ca;
 648
 649         if (p->ptr.cached)
 650                 return 0;
 651
 652         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 653
 654         if (ca->mi.state != BCH_MEMBER_STATE_failed)
 655                 durability = max_t(unsigned, durability, ca->mi.durability);
 656
 657         if (p->has_ec)
 658                 durability += p->ec.redundancy;
 659
 660         return durability;
 661 }
 662
 663 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 664 {
 665         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 666         const union bch_extent_entry *entry;
 667         struct extent_ptr_decoded p;
 668         unsigned durability = 0;
 669
 670         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 671                 durability += bch2_extent_ptr_durability(c, &p);
 672
 673         return durability;
 674 }
 675
 676 static unsigned bch2_bkey_durability_safe(struct bch_fs *c, struct bkey_s_c k)
 677 {
 678         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 679         const union bch_extent_entry *entry;
 680         struct extent_ptr_decoded p;
 681         unsigned durability = 0;
 682
 683         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 684                 if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 685                         durability += bch2_extent_ptr_durability(c, &p);
 686
 687         return durability;
 688 }
 689
 690 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
 691 {
 692         union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
 693         union bch_extent_entry *next = extent_entry_next(entry);
 694
 695         memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
 696         k->k.u64s -= extent_entry_u64s(entry);
 697 }
 698
 699 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
 700                                     struct extent_ptr_decoded *p)
 701 {
 702         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 703         struct bch_extent_crc_unpacked crc =
 704                 bch2_extent_crc_unpack(&k->k, NULL);
 705         union bch_extent_entry *pos;
 706
 707         if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 708                 pos = ptrs.start;
 709                 goto found;
 710         }
 711
 712         bkey_for_each_crc(&k->k, ptrs, crc, pos)
 713                 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 714                         pos = extent_entry_next(pos);
 715                         goto found;
 716                 }
 717
 718         bch2_extent_crc_append(k, p->crc);
 719         pos = bkey_val_end(bkey_i_to_s(k));
 720 found:
 721         p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
 722         __extent_entry_insert(k, pos, to_entry(&p->ptr));
 723
 724         if (p->has_ec) {
 725                 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
 726                 __extent_entry_insert(k, pos, to_entry(&p->ec));
 727         }
 728 }
 729
 730 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 731                                           union bch_extent_entry *entry)
 732 {
 733         union bch_extent_entry *i = ptrs.start;
 734
 735         if (i == entry)
 736                 return NULL;
 737
 738         while (extent_entry_next(i) != entry)
 739                 i = extent_entry_next(i);
 740         return i;
 741 }
 742
 743 static void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
 744 {
 745         union bch_extent_entry *next = extent_entry_next(entry);
 746
 747         /* stripes have ptrs, but their layout doesn't work with this code */
 748         BUG_ON(k.k->type == KEY_TYPE_stripe);
 749
 750         memmove_u64s_down(entry, next,
 751                           (u64 *) bkey_val_end(k) - (u64 *) next);
 752         k.k->u64s -= (u64 *) next - (u64 *) entry;
 753 }
 754
 755 /*
 756  * Returns pointer to the next entry after the one being dropped:
 757  */
 758 union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
 759                                                    struct bch_extent_ptr *ptr)
 760 {
 761         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 762         union bch_extent_entry *entry = to_entry(ptr), *next;
 763         union bch_extent_entry *ret = entry;
 764         bool drop_crc = true;
 765
 766         EBUG_ON(ptr < &ptrs.start->ptr ||
 767                 ptr >= &ptrs.end->ptr);
 768         EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
 769
 770         for (next = extent_entry_next(entry);
 771              next != ptrs.end;
 772              next = extent_entry_next(next)) {
 773                 if (extent_entry_is_crc(next)) {
 774                         break;
 775                 } else if (extent_entry_is_ptr(next)) {
 776                         drop_crc = false;
 777                         break;
 778                 }
 779         }
 780
 781         extent_entry_drop(k, entry);
 782
 783         while ((entry = extent_entry_prev(ptrs, entry))) {
 784                 if (extent_entry_is_ptr(entry))
 785                         break;
 786
 787                 if ((extent_entry_is_crc(entry) && drop_crc) ||
 788                     extent_entry_is_stripe_ptr(entry)) {
 789                         ret = (void *) ret - extent_entry_bytes(entry);
 790                         extent_entry_drop(k, entry);
 791                 }
 792         }
 793
 794         return ret;
 795 }
 796
 797 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 798                                            struct bch_extent_ptr *ptr)
 799 {
 800         bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
 801         union bch_extent_entry *ret =
 802                 bch2_bkey_drop_ptr_noerror(k, ptr);
 803
 804         /*
 805          * If we deleted all the dirty pointers and there's still cached
 806          * pointers, we could set the cached pointers to dirty if they're not
 807          * stale - but to do that correctly we'd need to grab an open_bucket
 808          * reference so that we don't race with bucket reuse:
 809          */
 810         if (have_dirty &&
 811             !bch2_bkey_dirty_devs(k.s_c).nr) {
 812                 k.k->type = KEY_TYPE_error;
 813                 set_bkey_val_u64s(k.k, 0);
 814                 ret = NULL;
 815         } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 816                 k.k->type = KEY_TYPE_deleted;
 817                 set_bkey_val_u64s(k.k, 0);
 818                 ret = NULL;
 819         }
 820
 821         return ret;
 822 }
 823
 824 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 825 {
 826         struct bch_extent_ptr *ptr;
 827
 828         bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 829 }
 830
 831 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 832 {
 833         struct bch_extent_ptr *ptr = bch2_bkey_has_device(k, dev);
 834
 835         if (ptr)
 836                 bch2_bkey_drop_ptr_noerror(k, ptr);
 837 }
 838
 839 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
 840 {
 841         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 842         const struct bch_extent_ptr *ptr;
 843
 844         bkey_for_each_ptr(ptrs, ptr)
 845                 if (ptr->dev == dev)
 846                         return ptr;
 847
 848         return NULL;
 849 }
 850
 851 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 852 {
 853         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 854         const struct bch_extent_ptr *ptr;
 855
 856         bkey_for_each_ptr(ptrs, ptr)
 857                 if (bch2_dev_in_target(c, ptr->dev, target) &&
 858                     (!ptr->cached ||
 859                      !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
 860                         return true;
 861
 862         return false;
 863 }
 864
 865 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 866                            struct bch_extent_ptr m, u64 offset)
 867 {
 868         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 869         const union bch_extent_entry *entry;
 870         struct extent_ptr_decoded p;
 871
 872         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 873                 if (p.ptr.dev   == m.dev &&
 874                     p.ptr.gen   == m.gen &&
 875                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 876                     (s64) m.offset  - offset)
 877                         return true;
 878
 879         return false;
 880 }
 881
 882 /*
 883  * Returns true if two extents refer to the same data:
 884  */
 885 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 886 {
 887         if (k1.k->type != k2.k->type)
 888                 return false;
 889
 890         if (bkey_extent_is_direct_data(k1.k)) {
 891                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 892                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 893                 const union bch_extent_entry *entry1, *entry2;
 894                 struct extent_ptr_decoded p1, p2;
 895
 896                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 897                         return false;
 898
 899                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 900                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 901                         if (p1.ptr.dev          == p2.ptr.dev &&
 902                             p1.ptr.gen          == p2.ptr.gen &&
 903                             (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 904                             (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 905                                 return true;
 906
 907                 return false;
 908         } else {
 909                 /* KEY_TYPE_deleted, etc. */
 910                 return true;
 911         }
 912 }
 913
 914 struct bch_extent_ptr *
 915 bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bkey_s k2)
 916 {
 917         struct bkey_ptrs ptrs2 = bch2_bkey_ptrs(k2);
 918         union bch_extent_entry *entry2;
 919         struct extent_ptr_decoded p2;
 920
 921         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 922                 if (p1.ptr.dev          == p2.ptr.dev &&
 923                     p1.ptr.gen          == p2.ptr.gen &&
 924                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 925                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 926                         return &entry2->ptr;
 927
 928         return NULL;
 929 }
 930
 931 void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
 932 {
 933         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 934         union bch_extent_entry *entry;
 935         union bch_extent_entry *ec = NULL;
 936
 937         bkey_extent_entry_for_each(ptrs, entry) {
 938                 if (&entry->ptr == ptr) {
 939                         ptr->cached = true;
 940                         if (ec)
 941                                 extent_entry_drop(k, ec);
 942                         return;
 943                 }
 944
 945                 if (extent_entry_is_stripe_ptr(entry))
 946                         ec = entry;
 947                 else if (extent_entry_is_ptr(entry))
 948                         ec = NULL;
 949         }
 950
 951         BUG();
 952 }
 953
 954 /*
 955  * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 956  *
 957  * Returns true if @k should be dropped entirely
 958  *
 959  * For existing keys, only called when btree nodes are being rewritten, not when
 960  * they're merely being compacted/resorted in memory.
 961  */
 962 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 963 {
 964         struct bch_extent_ptr *ptr;
 965
 966         bch2_bkey_drop_ptrs(k, ptr,
 967                 ptr->cached &&
 968                 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
 969
 970         return bkey_deleted(k.k);
 971 }
 972
 973 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 974                             struct bkey_s_c k)
 975 {
 976         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 977         const union bch_extent_entry *entry;
 978         struct bch_extent_crc_unpacked crc;
 979         const struct bch_extent_ptr *ptr;
 980         const struct bch_extent_stripe_ptr *ec;
 981         struct bch_dev *ca;
 982         bool first = true;
 983
 984         if (c)
 985                 prt_printf(out, "durability: %u ", bch2_bkey_durability_safe(c, k));
 986
 987         bkey_extent_entry_for_each(ptrs, entry) {
 988                 if (!first)
 989                         prt_printf(out, " ");
 990
 991                 switch (__extent_entry_type(entry)) {
 992                 case BCH_EXTENT_ENTRY_ptr:
 993                         ptr = entry_to_ptr(entry);
 994                         ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
 995                                 ? bch_dev_bkey_exists(c, ptr->dev)
 996                                 : NULL;
 997
 998                         if (!ca) {
 999                                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1000                                        (u64) ptr->offset, ptr->gen,
1001                                        ptr->cached ? " cached" : "");
1002                         } else {
1003                                 u32 offset;
1004                                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1005
1006                                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1007                                            ptr->dev, b, offset, ptr->gen);
1008                                 if (ptr->cached)
1009                                         prt_str(out, " cached");
1010                                 if (ptr->unwritten)
1011                                         prt_str(out, " unwritten");
1012                                 if (ca && ptr_stale(ca, ptr))
1013                                         prt_printf(out, " stale");
1014                         }
1015                         break;
1016                 case BCH_EXTENT_ENTRY_crc32:
1017                 case BCH_EXTENT_ENTRY_crc64:
1018                 case BCH_EXTENT_ENTRY_crc128:
1019                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1020
1021                         prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
1022                                crc.compressed_size,
1023                                crc.uncompressed_size,
1024                                crc.offset, crc.nonce,
1025                                bch2_csum_types[crc.csum_type],
1026                                bch2_compression_types[crc.compression_type]);
1027                         break;
1028                 case BCH_EXTENT_ENTRY_stripe_ptr:
1029                         ec = &entry->stripe_ptr;
1030
1031                         prt_printf(out, "ec: idx %llu block %u",
1032                                (u64) ec->idx, ec->block);
1033                         break;
1034                 default:
1035                         prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1036                         return;
1037                 }
1038
1039                 first = false;
1040         }
1041 }
1042
1043 static int extent_ptr_invalid(const struct bch_fs *c,
1044                               struct bkey_s_c k,
1045                               const struct bch_extent_ptr *ptr,
1046                               unsigned size_ondisk,
1047                               bool metadata,
1048                               struct printbuf *err)
1049 {
1050         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1051         const struct bch_extent_ptr *ptr2;
1052         u64 bucket;
1053         u32 bucket_offset;
1054         struct bch_dev *ca;
1055
1056         if (!bch2_dev_exists2(c, ptr->dev)) {
1057                 prt_printf(err, "pointer to invalid device (%u)", ptr->dev);
1058                 return -BCH_ERR_invalid_bkey;
1059         }
1060
1061         ca = bch_dev_bkey_exists(c, ptr->dev);
1062         bkey_for_each_ptr(ptrs, ptr2)
1063                 if (ptr != ptr2 && ptr->dev == ptr2->dev) {
1064                         prt_printf(err, "multiple pointers to same device (%u)", ptr->dev);
1065                         return -BCH_ERR_invalid_bkey;
1066                 }
1067
1068         bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1069
1070         if (bucket >= ca->mi.nbuckets) {
1071                 prt_printf(err, "pointer past last bucket (%llu > %llu)",
1072                        bucket, ca->mi.nbuckets);
1073                 return -BCH_ERR_invalid_bkey;
1074         }
1075
1076         if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket)) {
1077                 prt_printf(err, "pointer before first bucket (%llu < %u)",
1078                        bucket, ca->mi.first_bucket);
1079                 return -BCH_ERR_invalid_bkey;
1080         }
1081
1082         if (bucket_offset + size_ondisk > ca->mi.bucket_size) {
1083                 prt_printf(err, "pointer spans multiple buckets (%u + %u > %u)",
1084                        bucket_offset, size_ondisk, ca->mi.bucket_size);
1085                 return -BCH_ERR_invalid_bkey;
1086         }
1087
1088         return 0;
1089 }
1090
1091 int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
1092                            unsigned flags, struct printbuf *err)
1093 {
1094         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1095         const union bch_extent_entry *entry;
1096         struct bch_extent_crc_unpacked crc;
1097         unsigned size_ondisk = k.k->size;
1098         unsigned nonce = UINT_MAX;
1099         unsigned nr_ptrs = 0;
1100         bool unwritten = false, have_ec = false, crc_since_last_ptr = false;
1101         int ret;
1102
1103         if (bkey_is_btree_ptr(k.k))
1104                 size_ondisk = btree_sectors(c);
1105
1106         bkey_extent_entry_for_each(ptrs, entry) {
1107                 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX) {
1108                         prt_printf(err, "invalid extent entry type (got %u, max %u)",
1109                                __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1110                         return -BCH_ERR_invalid_bkey;
1111                 }
1112
1113                 if (bkey_is_btree_ptr(k.k) &&
1114                     !extent_entry_is_ptr(entry)) {
1115                         prt_printf(err, "has non ptr field");
1116                         return -BCH_ERR_invalid_bkey;
1117                 }
1118
1119                 switch (extent_entry_type(entry)) {
1120                 case BCH_EXTENT_ENTRY_ptr:
1121                         ret = extent_ptr_invalid(c, k, &entry->ptr, size_ondisk,
1122                                                  false, err);
1123                         if (ret)
1124                                 return ret;
1125
1126                         if (nr_ptrs && unwritten != entry->ptr.unwritten) {
1127                                 prt_printf(err, "extent with unwritten and written ptrs");
1128                                 return -BCH_ERR_invalid_bkey;
1129                         }
1130
1131                         if (k.k->type != KEY_TYPE_extent && entry->ptr.unwritten) {
1132                                 prt_printf(err, "has unwritten ptrs");
1133                                 return -BCH_ERR_invalid_bkey;
1134                         }
1135
1136                         if (entry->ptr.cached && have_ec) {
1137                                 prt_printf(err, "cached, erasure coded ptr");
1138                                 return -BCH_ERR_invalid_bkey;
1139                         }
1140
1141                         unwritten = entry->ptr.unwritten;
1142                         have_ec = false;
1143                         crc_since_last_ptr = false;
1144                         nr_ptrs++;
1145                         break;
1146                 case BCH_EXTENT_ENTRY_crc32:
1147                 case BCH_EXTENT_ENTRY_crc64:
1148                 case BCH_EXTENT_ENTRY_crc128:
1149                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1150
1151                         if (crc.offset + crc.live_size >
1152                             crc.uncompressed_size) {
1153                                 prt_printf(err, "checksum offset + key size > uncompressed size");
1154                                 return -BCH_ERR_invalid_bkey;
1155                         }
1156
1157                         size_ondisk = crc.compressed_size;
1158
1159                         if (!bch2_checksum_type_valid(c, crc.csum_type)) {
1160                                 prt_printf(err, "invalid checksum type");
1161                                 return -BCH_ERR_invalid_bkey;
1162                         }
1163
1164                         if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR) {
1165                                 prt_printf(err, "invalid compression type");
1166                                 return -BCH_ERR_invalid_bkey;
1167                         }
1168
1169                         if (bch2_csum_type_is_encryption(crc.csum_type)) {
1170                                 if (nonce == UINT_MAX)
1171                                         nonce = crc.offset + crc.nonce;
1172                                 else if (nonce != crc.offset + crc.nonce) {
1173                                         prt_printf(err, "incorrect nonce");
1174                                         return -BCH_ERR_invalid_bkey;
1175                                 }
1176                         }
1177
1178                         if (crc_since_last_ptr) {
1179                                 prt_printf(err, "redundant crc entry");
1180                                 return -BCH_ERR_invalid_bkey;
1181                         }
1182                         crc_since_last_ptr = true;
1183                         break;
1184                 case BCH_EXTENT_ENTRY_stripe_ptr:
1185                         if (have_ec) {
1186                                 prt_printf(err, "redundant stripe entry");
1187                                 return -BCH_ERR_invalid_bkey;
1188                         }
1189                         have_ec = true;
1190                         break;
1191                 }
1192         }
1193
1194         if (!nr_ptrs) {
1195                 prt_str(err, "no ptrs");
1196                 return -BCH_ERR_invalid_bkey;
1197         }
1198
1199         if (nr_ptrs >= BCH_BKEY_PTRS_MAX) {
1200                 prt_str(err, "too many ptrs");
1201                 return -BCH_ERR_invalid_bkey;
1202         }
1203
1204         if (crc_since_last_ptr) {
1205                 prt_printf(err, "redundant crc entry");
1206                 return -BCH_ERR_invalid_bkey;
1207         }
1208
1209         if (have_ec) {
1210                 prt_printf(err, "redundant stripe entry");
1211                 return -BCH_ERR_invalid_bkey;
1212         }
1213
1214         return 0;
1215 }
1216
1217 void bch2_ptr_swab(struct bkey_s k)
1218 {
1219         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1220         union bch_extent_entry *entry;
1221         u64 *d;
1222
1223         for (d =  (u64 *) ptrs.start;
1224              d != (u64 *) ptrs.end;
1225              d++)
1226                 *d = swab64(*d);
1227
1228         for (entry = ptrs.start;
1229              entry < ptrs.end;
1230              entry = extent_entry_next(entry)) {
1231                 switch (extent_entry_type(entry)) {
1232                 case BCH_EXTENT_ENTRY_ptr:
1233                         break;
1234                 case BCH_EXTENT_ENTRY_crc32:
1235                         entry->crc32.csum = swab32(entry->crc32.csum);
1236                         break;
1237                 case BCH_EXTENT_ENTRY_crc64:
1238                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1239                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1240                         break;
1241                 case BCH_EXTENT_ENTRY_crc128:
1242                         entry->crc128.csum.hi = (__force __le64)
1243                                 swab64((__force u64) entry->crc128.csum.hi);
1244                         entry->crc128.csum.lo = (__force __le64)
1245                                 swab64((__force u64) entry->crc128.csum.lo);
1246                         break;
1247                 case BCH_EXTENT_ENTRY_stripe_ptr:
1248                         break;
1249                 }
1250         }
1251 }
1252
1253 /* Generic extent code: */
1254
1255 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1256 {
1257         unsigned new_val_u64s = bkey_val_u64s(k.k);
1258         int val_u64s_delta;
1259         u64 sub;
1260
1261         if (bkey_le(where, bkey_start_pos(k.k)))
1262                 return 0;
1263
1264         EBUG_ON(bkey_gt(where, k.k->p));
1265
1266         sub = where.offset - bkey_start_offset(k.k);
1267
1268         k.k->size -= sub;
1269
1270         if (!k.k->size) {
1271                 k.k->type = KEY_TYPE_deleted;
1272                 new_val_u64s = 0;
1273         }
1274
1275         switch (k.k->type) {
1276         case KEY_TYPE_extent:
1277         case KEY_TYPE_reflink_v: {
1278                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1279                 union bch_extent_entry *entry;
1280                 bool seen_crc = false;
1281
1282                 bkey_extent_entry_for_each(ptrs, entry) {
1283                         switch (extent_entry_type(entry)) {
1284                         case BCH_EXTENT_ENTRY_ptr:
1285                                 if (!seen_crc)
1286                                         entry->ptr.offset += sub;
1287                                 break;
1288                         case BCH_EXTENT_ENTRY_crc32:
1289                                 entry->crc32.offset += sub;
1290                                 break;
1291                         case BCH_EXTENT_ENTRY_crc64:
1292                                 entry->crc64.offset += sub;
1293                                 break;
1294                         case BCH_EXTENT_ENTRY_crc128:
1295                                 entry->crc128.offset += sub;
1296                                 break;
1297                         case BCH_EXTENT_ENTRY_stripe_ptr:
1298                                 break;
1299                         }
1300
1301                         if (extent_entry_is_crc(entry))
1302                                 seen_crc = true;
1303                 }
1304
1305                 break;
1306         }
1307         case KEY_TYPE_reflink_p: {
1308                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1309
1310                 le64_add_cpu(&p.v->idx, sub);
1311                 break;
1312         }
1313         case KEY_TYPE_inline_data:
1314         case KEY_TYPE_indirect_inline_data: {
1315                 void *p = bkey_inline_data_p(k);
1316                 unsigned bytes = bkey_inline_data_bytes(k.k);
1317
1318                 sub = min_t(u64, sub << 9, bytes);
1319
1320                 memmove(p, p + sub, bytes - sub);
1321
1322                 new_val_u64s -= sub >> 3;
1323                 break;
1324         }
1325         }
1326
1327         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1328         BUG_ON(val_u64s_delta < 0);
1329
1330         set_bkey_val_u64s(k.k, new_val_u64s);
1331         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1332         return -val_u64s_delta;
1333 }
1334
1335 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1336 {
1337         unsigned new_val_u64s = bkey_val_u64s(k.k);
1338         int val_u64s_delta;
1339         u64 len = 0;
1340
1341         if (bkey_ge(where, k.k->p))
1342                 return 0;
1343
1344         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1345
1346         len = where.offset - bkey_start_offset(k.k);
1347
1348         k.k->p.offset = where.offset;
1349         k.k->size = len;
1350
1351         if (!len) {
1352                 k.k->type = KEY_TYPE_deleted;
1353                 new_val_u64s = 0;
1354         }
1355
1356         switch (k.k->type) {
1357         case KEY_TYPE_inline_data:
1358         case KEY_TYPE_indirect_inline_data:
1359                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1360                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1361                 break;
1362         }
1363
1364         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1365         BUG_ON(val_u64s_delta < 0);
1366
1367         set_bkey_val_u64s(k.k, new_val_u64s);
1368         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1369         return -val_u64s_delta;
1370 }