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 "util.h"
  26
  27 #include <trace/events/bcachefs.h>
  28
  29 static unsigned bch2_crc_field_size_max[] = {
  30         [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
  31         [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
  32         [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
  33 };
  34
  35 static void bch2_extent_crc_pack(union bch_extent_crc *,
  36                                  struct bch_extent_crc_unpacked,
  37                                  enum bch_extent_entry_type);
  38
  39 static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
  40                                                    unsigned dev)
  41 {
  42         struct bch_dev_io_failures *i;
  43
  44         for (i = f->devs; i < f->devs + f->nr; i++)
  45                 if (i->dev == dev)
  46                         return i;
  47
  48         return NULL;
  49 }
  50
  51 void bch2_mark_io_failure(struct bch_io_failures *failed,
  52                           struct extent_ptr_decoded *p)
  53 {
  54         struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
  55
  56         if (!f) {
  57                 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
  58
  59                 f = &failed->devs[failed->nr++];
  60                 f->dev          = p->ptr.dev;
  61                 f->idx          = p->idx;
  62                 f->nr_failed    = 1;
  63                 f->nr_retries   = 0;
  64         } else if (p->idx != f->idx) {
  65                 f->idx          = p->idx;
  66                 f->nr_failed    = 1;
  67                 f->nr_retries   = 0;
  68         } else {
  69                 f->nr_failed++;
  70         }
  71 }
  72
  73 /*
  74  * returns true if p1 is better than p2:
  75  */
  76 static inline bool ptr_better(struct bch_fs *c,
  77                               const struct extent_ptr_decoded p1,
  78                               const struct extent_ptr_decoded p2)
  79 {
  80         if (likely(!p1.idx && !p2.idx)) {
  81                 struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
  82                 struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
  83
  84                 u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
  85                 u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
  86
  87                 /* Pick at random, biased in favor of the faster device: */
  88
  89                 return bch2_rand_range(l1 + l2) > l1;
  90         }
  91
  92         if (bch2_force_reconstruct_read)
  93                 return p1.idx > p2.idx;
  94
  95         return p1.idx < p2.idx;
  96 }
  97
  98 /*
  99  * This picks a non-stale pointer, preferably from a device other than @avoid.
 100  * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
 101  * other devices, it will still pick a pointer from avoid.
 102  */
 103 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 104                                struct bch_io_failures *failed,
 105                                struct extent_ptr_decoded *pick)
 106 {
 107         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 108         const union bch_extent_entry *entry;
 109         struct extent_ptr_decoded p;
 110         struct bch_dev_io_failures *f;
 111         struct bch_dev *ca;
 112         int ret = 0;
 113
 114         if (k.k->type == KEY_TYPE_error)
 115                 return -EIO;
 116
 117         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 118                 /*
 119                  * Unwritten extent: no need to actually read, treat it as a
 120                  * hole and return 0s:
 121                  */
 122                 if (p.ptr.unwritten)
 123                         return 0;
 124
 125                 ca = bch_dev_bkey_exists(c, p.ptr.dev);
 126
 127                 /*
 128                  * If there are any dirty pointers it's an error if we can't
 129                  * read:
 130                  */
 131                 if (!ret && !p.ptr.cached)
 132                         ret = -EIO;
 133
 134                 if (p.ptr.cached && ptr_stale(ca, &p.ptr))
 135                         continue;
 136
 137                 f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
 138                 if (f)
 139                         p.idx = f->nr_failed < f->nr_retries
 140                                 ? f->idx
 141                                 : f->idx + 1;
 142
 143                 if (!p.idx &&
 144                     !bch2_dev_is_readable(ca))
 145                         p.idx++;
 146
 147                 if (bch2_force_reconstruct_read &&
 148                     !p.idx && p.has_ec)
 149                         p.idx++;
 150
 151                 if (p.idx >= (unsigned) p.has_ec + 1)
 152                         continue;
 153
 154                 if (ret > 0 && !ptr_better(c, p, *pick))
 155                         continue;
 156
 157                 *pick = p;
 158                 ret = 1;
 159         }
 160
 161         return ret;
 162 }
 163
 164 /* KEY_TYPE_btree_ptr: */
 165
 166 int bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k,
 167                            unsigned flags, struct printbuf *err)
 168 {
 169         if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX) {
 170                 prt_printf(err, "value too big (%zu > %u)",
 171                        bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 172                 return -BCH_ERR_invalid_bkey;
 173         }
 174
 175         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 176 }
 177
 178 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 179                             struct bkey_s_c k)
 180 {
 181         bch2_bkey_ptrs_to_text(out, c, k);
 182 }
 183
 184 int bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 185                               unsigned flags, struct printbuf *err)
 186 {
 187         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 188
 189         if (bkey_val_bytes(k.k) <= sizeof(*bp.v)) {
 190                 prt_printf(err, "value too small (%zu <= %zu)",
 191                        bkey_val_bytes(k.k), sizeof(*bp.v));
 192                 return -BCH_ERR_invalid_bkey;
 193         }
 194
 195         if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) {
 196                 prt_printf(err, "value too big (%zu > %zu)",
 197                        bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 198                 return -BCH_ERR_invalid_bkey;
 199         }
 200
 201         if (c->sb.version < bcachefs_metadata_version_snapshot &&
 202             bp.v->min_key.snapshot) {
 203                 prt_printf(err, "invalid min_key.snapshot (%u != 0)",
 204                        bp.v->min_key.snapshot);
 205                 return -BCH_ERR_invalid_bkey;
 206         }
 207
 208         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 209 }
 210
 211 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 212                                struct bkey_s_c k)
 213 {
 214         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 215
 216         prt_printf(out, "seq %llx written %u min_key %s",
 217                le64_to_cpu(bp.v->seq),
 218                le16_to_cpu(bp.v->sectors_written),
 219                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 220
 221         bch2_bpos_to_text(out, bp.v->min_key);
 222         prt_printf(out, " ");
 223         bch2_bkey_ptrs_to_text(out, c, k);
 224 }
 225
 226 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 227                               unsigned big_endian, int write,
 228                               struct bkey_s k)
 229 {
 230         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 231
 232         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 233
 234         if (version < bcachefs_metadata_version_inode_btree_change &&
 235             btree_node_type_is_extents(btree_id) &&
 236             !bkey_eq(bp.v->min_key, POS_MIN))
 237                 bp.v->min_key = write
 238                         ? bpos_nosnap_predecessor(bp.v->min_key)
 239                         : bpos_nosnap_successor(bp.v->min_key);
 240 }
 241
 242 /* KEY_TYPE_extent: */
 243
 244 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 245 {
 246         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 247         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 248         union bch_extent_entry *en_l;
 249         const union bch_extent_entry *en_r;
 250         struct extent_ptr_decoded lp, rp;
 251         bool use_right_ptr;
 252         struct bch_dev *ca;
 253
 254         en_l = l_ptrs.start;
 255         en_r = r_ptrs.start;
 256         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 257                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 258                         return false;
 259
 260                 en_l = extent_entry_next(en_l);
 261                 en_r = extent_entry_next(en_r);
 262         }
 263
 264         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 265                 return false;
 266
 267         en_l = l_ptrs.start;
 268         en_r = r_ptrs.start;
 269         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 270         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 271
 272         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 273                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 274                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 275                     rp.ptr.offset + rp.crc.offset ||
 276                     lp.ptr.dev                  != rp.ptr.dev ||
 277                     lp.ptr.gen                  != rp.ptr.gen ||
 278                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 279                     lp.has_ec                   != rp.has_ec)
 280                         return false;
 281
 282                 /* Extents may not straddle buckets: */
 283                 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
 284                 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
 285                         return false;
 286
 287                 if (lp.has_ec                   != rp.has_ec ||
 288                     (lp.has_ec &&
 289                      (lp.ec.block               != rp.ec.block ||
 290                       lp.ec.redundancy          != rp.ec.redundancy ||
 291                       lp.ec.idx                 != rp.ec.idx)))
 292                         return false;
 293
 294                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 295                     lp.crc.nonce                != rp.crc.nonce)
 296                         return false;
 297
 298                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 299                     lp.crc.uncompressed_size) {
 300                         /* can use left extent's crc entry */
 301                 } else if (lp.crc.live_size <= rp.crc.offset) {
 302                         /* can use right extent's crc entry */
 303                 } else {
 304                         /* check if checksums can be merged: */
 305                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 306                             lp.crc.nonce                != rp.crc.nonce ||
 307                             crc_is_compressed(lp.crc) ||
 308                             !bch2_checksum_mergeable(lp.crc.csum_type))
 309                                 return false;
 310
 311                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 312                             rp.crc.offset)
 313                                 return false;
 314
 315                         if (lp.crc.csum_type &&
 316                             lp.crc.uncompressed_size +
 317                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 318                                 return false;
 319                 }
 320
 321                 en_l = extent_entry_next(en_l);
 322                 en_r = extent_entry_next(en_r);
 323         }
 324
 325         en_l = l_ptrs.start;
 326         en_r = r_ptrs.start;
 327         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 328                 if (extent_entry_is_crc(en_l)) {
 329                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 330                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 331
 332                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 333                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 334                                 return false;
 335                 }
 336
 337                 en_l = extent_entry_next(en_l);
 338                 en_r = extent_entry_next(en_r);
 339         }
 340
 341         use_right_ptr = false;
 342         en_l = l_ptrs.start;
 343         en_r = r_ptrs.start;
 344         while (en_l < l_ptrs.end) {
 345                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 346                     use_right_ptr)
 347                         en_l->ptr = en_r->ptr;
 348
 349                 if (extent_entry_is_crc(en_l)) {
 350                         struct bch_extent_crc_unpacked crc_l =
 351                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 352                         struct bch_extent_crc_unpacked crc_r =
 353                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 354
 355                         use_right_ptr = false;
 356
 357                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 358                             crc_l.uncompressed_size) {
 359                                 /* can use left extent's crc entry */
 360                         } else if (crc_l.live_size <= crc_r.offset) {
 361                                 /* can use right extent's crc entry */
 362                                 crc_r.offset -= crc_l.live_size;
 363                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 364                                                      extent_entry_type(en_l));
 365                                 use_right_ptr = true;
 366                         } else {
 367                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 368                                                                  crc_l.csum,
 369                                                                  crc_r.csum,
 370                                                                  crc_r.uncompressed_size << 9);
 371
 372                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 373                                 crc_l.compressed_size   += crc_r.compressed_size;
 374                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 375                                                      extent_entry_type(en_l));
 376                         }
 377                 }
 378
 379                 en_l = extent_entry_next(en_l);
 380                 en_r = extent_entry_next(en_r);
 381         }
 382
 383         bch2_key_resize(l.k, l.k->size + r.k->size);
 384         return true;
 385 }
 386
 387 /* KEY_TYPE_reservation: */
 388
 389 int bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k,
 390                              unsigned flags, struct printbuf *err)
 391 {
 392         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 393
 394         if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation)) {
 395                 prt_printf(err, "incorrect value size (%zu != %zu)",
 396                        bkey_val_bytes(k.k), sizeof(*r.v));
 397                 return -BCH_ERR_invalid_bkey;
 398         }
 399
 400         if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX) {
 401                 prt_printf(err, "invalid nr_replicas (%u)",
 402                        r.v->nr_replicas);
 403                 return -BCH_ERR_invalid_bkey;
 404         }
 405
 406         return 0;
 407 }
 408
 409 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 410                               struct bkey_s_c k)
 411 {
 412         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 413
 414         prt_printf(out, "generation %u replicas %u",
 415                le32_to_cpu(r.v->generation),
 416                r.v->nr_replicas);
 417 }
 418
 419 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 420 {
 421         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 422         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 423
 424         if (l.v->generation != r.v->generation ||
 425             l.v->nr_replicas != r.v->nr_replicas)
 426                 return false;
 427
 428         bch2_key_resize(l.k, l.k->size + r.k->size);
 429         return true;
 430 }
 431
 432 /* Extent checksum entries: */
 433
 434 /* returns true if not equal */
 435 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 436                                          struct bch_extent_crc_unpacked r)
 437 {
 438         return (l.csum_type             != r.csum_type ||
 439                 l.compression_type      != r.compression_type ||
 440                 l.compressed_size       != r.compressed_size ||
 441                 l.uncompressed_size     != r.uncompressed_size ||
 442                 l.offset                != r.offset ||
 443                 l.live_size             != r.live_size ||
 444                 l.nonce                 != r.nonce ||
 445                 bch2_crc_cmp(l.csum, r.csum));
 446 }
 447
 448 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 449                                   struct bch_extent_crc_unpacked n)
 450 {
 451         return !crc_is_compressed(u) &&
 452                 u.csum_type &&
 453                 u.uncompressed_size > u.live_size &&
 454                 bch2_csum_type_is_encryption(u.csum_type) ==
 455                 bch2_csum_type_is_encryption(n.csum_type);
 456 }
 457
 458 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 459                                  struct bch_extent_crc_unpacked n)
 460 {
 461         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 462         struct bch_extent_crc_unpacked crc;
 463         const union bch_extent_entry *i;
 464
 465         if (!n.csum_type)
 466                 return false;
 467
 468         bkey_for_each_crc(k.k, ptrs, crc, i)
 469                 if (can_narrow_crc(crc, n))
 470                         return true;
 471
 472         return false;
 473 }
 474
 475 /*
 476  * We're writing another replica for this extent, so while we've got the data in
 477  * memory we'll be computing a new checksum for the currently live data.
 478  *
 479  * If there are other replicas we aren't moving, and they are checksummed but
 480  * not compressed, we can modify them to point to only the data that is
 481  * currently live (so that readers won't have to bounce) while we've got the
 482  * checksum we need:
 483  */
 484 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 485 {
 486         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 487         struct bch_extent_crc_unpacked u;
 488         struct extent_ptr_decoded p;
 489         union bch_extent_entry *i;
 490         bool ret = false;
 491
 492         /* Find a checksum entry that covers only live data: */
 493         if (!n.csum_type) {
 494                 bkey_for_each_crc(&k->k, ptrs, u, i)
 495                         if (!crc_is_compressed(u) &&
 496                             u.csum_type &&
 497                             u.live_size == u.uncompressed_size) {
 498                                 n = u;
 499                                 goto found;
 500                         }
 501                 return false;
 502         }
 503 found:
 504         BUG_ON(crc_is_compressed(n));
 505         BUG_ON(n.offset);
 506         BUG_ON(n.live_size != k->k.size);
 507
 508 restart_narrow_pointers:
 509         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 510
 511         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 512                 if (can_narrow_crc(p.crc, n)) {
 513                         bch2_bkey_drop_ptr_noerror(bkey_i_to_s(k), &i->ptr);
 514                         p.ptr.offset += p.crc.offset;
 515                         p.crc = n;
 516                         bch2_extent_ptr_decoded_append(k, &p);
 517                         ret = true;
 518                         goto restart_narrow_pointers;
 519                 }
 520
 521         return ret;
 522 }
 523
 524 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 525                                  struct bch_extent_crc_unpacked src,
 526                                  enum bch_extent_entry_type type)
 527 {
 528 #define set_common_fields(_dst, _src)                                   \
 529                 _dst.type               = 1 << type;                    \
 530                 _dst.csum_type          = _src.csum_type,               \
 531                 _dst.compression_type   = _src.compression_type,        \
 532                 _dst._compressed_size   = _src.compressed_size - 1,     \
 533                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 534                 _dst.offset             = _src.offset
 535
 536         switch (type) {
 537         case BCH_EXTENT_ENTRY_crc32:
 538                 set_common_fields(dst->crc32, src);
 539                 dst->crc32.csum  = *((__le32 *) &src.csum.lo);
 540                 break;
 541         case BCH_EXTENT_ENTRY_crc64:
 542                 set_common_fields(dst->crc64, src);
 543                 dst->crc64.nonce        = src.nonce;
 544                 dst->crc64.csum_lo      = src.csum.lo;
 545                 dst->crc64.csum_hi      = *((__le16 *) &src.csum.hi);
 546                 break;
 547         case BCH_EXTENT_ENTRY_crc128:
 548                 set_common_fields(dst->crc128, src);
 549                 dst->crc128.nonce       = src.nonce;
 550                 dst->crc128.csum        = src.csum;
 551                 break;
 552         default:
 553                 BUG();
 554         }
 555 #undef set_common_fields
 556 }
 557
 558 void bch2_extent_crc_append(struct bkey_i *k,
 559                             struct bch_extent_crc_unpacked new)
 560 {
 561         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 562         union bch_extent_crc *crc = (void *) ptrs.end;
 563         enum bch_extent_entry_type type;
 564
 565         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 566             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 567             new.nonce                           <= CRC32_NONCE_MAX)
 568                 type = BCH_EXTENT_ENTRY_crc32;
 569         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 570                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 571                    new.nonce                    <= CRC64_NONCE_MAX)
 572                 type = BCH_EXTENT_ENTRY_crc64;
 573         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 574                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 575                    new.nonce                    <= CRC128_NONCE_MAX)
 576                 type = BCH_EXTENT_ENTRY_crc128;
 577         else
 578                 BUG();
 579
 580         bch2_extent_crc_pack(crc, new, type);
 581
 582         k->k.u64s += extent_entry_u64s(ptrs.end);
 583
 584         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 585 }
 586
 587 /* Generic code for keys with pointers: */
 588
 589 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 590 {
 591         return bch2_bkey_devs(k).nr;
 592 }
 593
 594 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 595 {
 596         return k.k->type == KEY_TYPE_reservation
 597                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 598                 : bch2_bkey_dirty_devs(k).nr;
 599 }
 600
 601 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 602 {
 603         unsigned ret = 0;
 604
 605         if (k.k->type == KEY_TYPE_reservation) {
 606                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 607         } else {
 608                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 609                 const union bch_extent_entry *entry;
 610                 struct extent_ptr_decoded p;
 611
 612                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 613                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 614         }
 615
 616         return ret;
 617 }
 618
 619 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 620 {
 621         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 622         const union bch_extent_entry *entry;
 623         struct extent_ptr_decoded p;
 624         unsigned ret = 0;
 625
 626         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 627                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 628                         ret += p.crc.compressed_size;
 629
 630         return ret;
 631 }
 632
 633 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
 634 {
 635         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 636         const union bch_extent_entry *entry;
 637         struct bch_extent_crc_unpacked crc;
 638
 639         bkey_for_each_crc(k.k, ptrs, crc, entry)
 640                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 641                         return true;
 642         return false;
 643 }
 644
 645 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 646 {
 647         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 648         const union bch_extent_entry *entry;
 649         struct extent_ptr_decoded p = { 0 };
 650         unsigned replicas = 0;
 651
 652         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 653                 if (p.ptr.cached)
 654                         continue;
 655
 656                 if (p.has_ec)
 657                         replicas += p.ec.redundancy;
 658
 659                 replicas++;
 660
 661         }
 662
 663         return replicas;
 664 }
 665
 666 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 667 {
 668         unsigned durability = 0;
 669         struct bch_dev *ca;
 670
 671         if (p->ptr.cached)
 672                 return 0;
 673
 674         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 675
 676         if (ca->mi.state != BCH_MEMBER_STATE_failed)
 677                 durability = max_t(unsigned, durability, ca->mi.durability);
 678
 679         if (p->has_ec)
 680                 durability += p->ec.redundancy;
 681
 682         return durability;
 683 }
 684
 685 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 686 {
 687         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 688         const union bch_extent_entry *entry;
 689         struct extent_ptr_decoded p;
 690         unsigned durability = 0;
 691
 692         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 693                 durability += bch2_extent_ptr_durability(c, &p);
 694
 695         return durability;
 696 }
 697
 698 static unsigned bch2_bkey_durability_safe(struct bch_fs *c, struct bkey_s_c k)
 699 {
 700         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 701         const union bch_extent_entry *entry;
 702         struct extent_ptr_decoded p;
 703         unsigned durability = 0;
 704
 705         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 706                 if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 707                         durability += bch2_extent_ptr_durability(c, &p);
 708
 709         return durability;
 710 }
 711
 712 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
 713 {
 714         union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
 715         union bch_extent_entry *next = extent_entry_next(entry);
 716
 717         memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
 718         k->k.u64s -= extent_entry_u64s(entry);
 719 }
 720
 721 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
 722                                     struct extent_ptr_decoded *p)
 723 {
 724         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 725         struct bch_extent_crc_unpacked crc =
 726                 bch2_extent_crc_unpack(&k->k, NULL);
 727         union bch_extent_entry *pos;
 728
 729         if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 730                 pos = ptrs.start;
 731                 goto found;
 732         }
 733
 734         bkey_for_each_crc(&k->k, ptrs, crc, pos)
 735                 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 736                         pos = extent_entry_next(pos);
 737                         goto found;
 738                 }
 739
 740         bch2_extent_crc_append(k, p->crc);
 741         pos = bkey_val_end(bkey_i_to_s(k));
 742 found:
 743         p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
 744         __extent_entry_insert(k, pos, to_entry(&p->ptr));
 745
 746         if (p->has_ec) {
 747                 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
 748                 __extent_entry_insert(k, pos, to_entry(&p->ec));
 749         }
 750 }
 751
 752 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 753                                           union bch_extent_entry *entry)
 754 {
 755         union bch_extent_entry *i = ptrs.start;
 756
 757         if (i == entry)
 758                 return NULL;
 759
 760         while (extent_entry_next(i) != entry)
 761                 i = extent_entry_next(i);
 762         return i;
 763 }
 764
 765 static void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
 766 {
 767         union bch_extent_entry *next = extent_entry_next(entry);
 768
 769         /* stripes have ptrs, but their layout doesn't work with this code */
 770         BUG_ON(k.k->type == KEY_TYPE_stripe);
 771
 772         memmove_u64s_down(entry, next,
 773                           (u64 *) bkey_val_end(k) - (u64 *) next);
 774         k.k->u64s -= (u64 *) next - (u64 *) entry;
 775 }
 776
 777 /*
 778  * Returns pointer to the next entry after the one being dropped:
 779  */
 780 union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
 781                                                    struct bch_extent_ptr *ptr)
 782 {
 783         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 784         union bch_extent_entry *entry = to_entry(ptr), *next;
 785         union bch_extent_entry *ret = entry;
 786         bool drop_crc = true;
 787
 788         EBUG_ON(ptr < &ptrs.start->ptr ||
 789                 ptr >= &ptrs.end->ptr);
 790         EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
 791
 792         for (next = extent_entry_next(entry);
 793              next != ptrs.end;
 794              next = extent_entry_next(next)) {
 795                 if (extent_entry_is_crc(next)) {
 796                         break;
 797                 } else if (extent_entry_is_ptr(next)) {
 798                         drop_crc = false;
 799                         break;
 800                 }
 801         }
 802
 803         extent_entry_drop(k, entry);
 804
 805         while ((entry = extent_entry_prev(ptrs, entry))) {
 806                 if (extent_entry_is_ptr(entry))
 807                         break;
 808
 809                 if ((extent_entry_is_crc(entry) && drop_crc) ||
 810                     extent_entry_is_stripe_ptr(entry)) {
 811                         ret = (void *) ret - extent_entry_bytes(entry);
 812                         extent_entry_drop(k, entry);
 813                 }
 814         }
 815
 816         return ret;
 817 }
 818
 819 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 820                                            struct bch_extent_ptr *ptr)
 821 {
 822         bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
 823         union bch_extent_entry *ret =
 824                 bch2_bkey_drop_ptr_noerror(k, ptr);
 825
 826         /*
 827          * If we deleted all the dirty pointers and there's still cached
 828          * pointers, we could set the cached pointers to dirty if they're not
 829          * stale - but to do that correctly we'd need to grab an open_bucket
 830          * reference so that we don't race with bucket reuse:
 831          */
 832         if (have_dirty &&
 833             !bch2_bkey_dirty_devs(k.s_c).nr) {
 834                 k.k->type = KEY_TYPE_error;
 835                 set_bkey_val_u64s(k.k, 0);
 836                 ret = NULL;
 837         } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 838                 k.k->type = KEY_TYPE_deleted;
 839                 set_bkey_val_u64s(k.k, 0);
 840                 ret = NULL;
 841         }
 842
 843         return ret;
 844 }
 845
 846 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 847 {
 848         struct bch_extent_ptr *ptr;
 849
 850         bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 851 }
 852
 853 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 854 {
 855         struct bch_extent_ptr *ptr = bch2_bkey_has_device(k, dev);
 856
 857         if (ptr)
 858                 bch2_bkey_drop_ptr_noerror(k, ptr);
 859 }
 860
 861 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
 862 {
 863         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 864         const struct bch_extent_ptr *ptr;
 865
 866         bkey_for_each_ptr(ptrs, ptr)
 867                 if (ptr->dev == dev)
 868                         return ptr;
 869
 870         return NULL;
 871 }
 872
 873 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 874 {
 875         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 876         const struct bch_extent_ptr *ptr;
 877
 878         bkey_for_each_ptr(ptrs, ptr)
 879                 if (bch2_dev_in_target(c, ptr->dev, target) &&
 880                     (!ptr->cached ||
 881                      !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
 882                         return true;
 883
 884         return false;
 885 }
 886
 887 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 888                            struct bch_extent_ptr m, u64 offset)
 889 {
 890         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 891         const union bch_extent_entry *entry;
 892         struct extent_ptr_decoded p;
 893
 894         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 895                 if (p.ptr.dev   == m.dev &&
 896                     p.ptr.gen   == m.gen &&
 897                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 898                     (s64) m.offset  - offset)
 899                         return true;
 900
 901         return false;
 902 }
 903
 904 /*
 905  * Returns true if two extents refer to the same data:
 906  */
 907 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 908 {
 909         if (k1.k->type != k2.k->type)
 910                 return false;
 911
 912         if (bkey_extent_is_direct_data(k1.k)) {
 913                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 914                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 915                 const union bch_extent_entry *entry1, *entry2;
 916                 struct extent_ptr_decoded p1, p2;
 917
 918                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 919                         return false;
 920
 921                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 922                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 923                         if (p1.ptr.dev          == p2.ptr.dev &&
 924                             p1.ptr.gen          == p2.ptr.gen &&
 925                             (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 926                             (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 927                                 return true;
 928
 929                 return false;
 930         } else {
 931                 /* KEY_TYPE_deleted, etc. */
 932                 return true;
 933         }
 934 }
 935
 936 struct bch_extent_ptr *
 937 bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bkey_s k2)
 938 {
 939         struct bkey_ptrs ptrs2 = bch2_bkey_ptrs(k2);
 940         union bch_extent_entry *entry2;
 941         struct extent_ptr_decoded p2;
 942
 943         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 944                 if (p1.ptr.dev          == p2.ptr.dev &&
 945                     p1.ptr.gen          == p2.ptr.gen &&
 946                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 947                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 948                         return &entry2->ptr;
 949
 950         return NULL;
 951 }
 952
 953 void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
 954 {
 955         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 956         union bch_extent_entry *entry;
 957         union bch_extent_entry *ec = NULL;
 958
 959         bkey_extent_entry_for_each(ptrs, entry) {
 960                 if (&entry->ptr == ptr) {
 961                         ptr->cached = true;
 962                         if (ec)
 963                                 extent_entry_drop(k, ec);
 964                         return;
 965                 }
 966
 967                 if (extent_entry_is_stripe_ptr(entry))
 968                         ec = entry;
 969                 else if (extent_entry_is_ptr(entry))
 970                         ec = NULL;
 971         }
 972
 973         BUG();
 974 }
 975
 976 /*
 977  * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 978  *
 979  * Returns true if @k should be dropped entirely
 980  *
 981  * For existing keys, only called when btree nodes are being rewritten, not when
 982  * they're merely being compacted/resorted in memory.
 983  */
 984 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 985 {
 986         struct bch_extent_ptr *ptr;
 987
 988         bch2_bkey_drop_ptrs(k, ptr,
 989                 ptr->cached &&
 990                 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
 991
 992         return bkey_deleted(k.k);
 993 }
 994
 995 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 996                             struct bkey_s_c k)
 997 {
 998         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 999         const union bch_extent_entry *entry;
1000         struct bch_extent_crc_unpacked crc;
1001         const struct bch_extent_ptr *ptr;
1002         const struct bch_extent_stripe_ptr *ec;
1003         struct bch_dev *ca;
1004         bool first = true;
1005
1006         if (c)
1007                 prt_printf(out, "durability: %u ", bch2_bkey_durability_safe(c, k));
1008
1009         bkey_extent_entry_for_each(ptrs, entry) {
1010                 if (!first)
1011                         prt_printf(out, " ");
1012
1013                 switch (__extent_entry_type(entry)) {
1014                 case BCH_EXTENT_ENTRY_ptr:
1015                         ptr = entry_to_ptr(entry);
1016                         ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
1017                                 ? bch_dev_bkey_exists(c, ptr->dev)
1018                                 : NULL;
1019
1020                         if (!ca) {
1021                                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1022                                        (u64) ptr->offset, ptr->gen,
1023                                        ptr->cached ? " cached" : "");
1024                         } else {
1025                                 u32 offset;
1026                                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1027
1028                                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1029                                            ptr->dev, b, offset, ptr->gen);
1030                                 if (ptr->cached)
1031                                         prt_str(out, " cached");
1032                                 if (ptr->unwritten)
1033                                         prt_str(out, " unwritten");
1034                                 if (ca && ptr_stale(ca, ptr))
1035                                         prt_printf(out, " stale");
1036                         }
1037                         break;
1038                 case BCH_EXTENT_ENTRY_crc32:
1039                 case BCH_EXTENT_ENTRY_crc64:
1040                 case BCH_EXTENT_ENTRY_crc128:
1041                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1042
1043                         prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
1044                                crc.compressed_size,
1045                                crc.uncompressed_size,
1046                                crc.offset, crc.nonce,
1047                                bch2_csum_types[crc.csum_type],
1048                                bch2_compression_types[crc.compression_type]);
1049                         break;
1050                 case BCH_EXTENT_ENTRY_stripe_ptr:
1051                         ec = &entry->stripe_ptr;
1052
1053                         prt_printf(out, "ec: idx %llu block %u",
1054                                (u64) ec->idx, ec->block);
1055                         break;
1056                 default:
1057                         prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1058                         return;
1059                 }
1060
1061                 first = false;
1062         }
1063 }
1064
1065 static int extent_ptr_invalid(const struct bch_fs *c,
1066                               struct bkey_s_c k,
1067                               const struct bch_extent_ptr *ptr,
1068                               unsigned size_ondisk,
1069                               bool metadata,
1070                               struct printbuf *err)
1071 {
1072         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1073         const struct bch_extent_ptr *ptr2;
1074         u64 bucket;
1075         u32 bucket_offset;
1076         struct bch_dev *ca;
1077
1078         if (!bch2_dev_exists2(c, ptr->dev)) {
1079                 prt_printf(err, "pointer to invalid device (%u)", ptr->dev);
1080                 return -BCH_ERR_invalid_bkey;
1081         }
1082
1083         ca = bch_dev_bkey_exists(c, ptr->dev);
1084         bkey_for_each_ptr(ptrs, ptr2)
1085                 if (ptr != ptr2 && ptr->dev == ptr2->dev) {
1086                         prt_printf(err, "multiple pointers to same device (%u)", ptr->dev);
1087                         return -BCH_ERR_invalid_bkey;
1088                 }
1089
1090         bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1091
1092         if (bucket >= ca->mi.nbuckets) {
1093                 prt_printf(err, "pointer past last bucket (%llu > %llu)",
1094                        bucket, ca->mi.nbuckets);
1095                 return -BCH_ERR_invalid_bkey;
1096         }
1097
1098         if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket)) {
1099                 prt_printf(err, "pointer before first bucket (%llu < %u)",
1100                        bucket, ca->mi.first_bucket);
1101                 return -BCH_ERR_invalid_bkey;
1102         }
1103
1104         if (bucket_offset + size_ondisk > ca->mi.bucket_size) {
1105                 prt_printf(err, "pointer spans multiple buckets (%u + %u > %u)",
1106                        bucket_offset, size_ondisk, ca->mi.bucket_size);
1107                 return -BCH_ERR_invalid_bkey;
1108         }
1109
1110         return 0;
1111 }
1112
1113 int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
1114                            unsigned flags, struct printbuf *err)
1115 {
1116         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1117         const union bch_extent_entry *entry;
1118         struct bch_extent_crc_unpacked crc;
1119         unsigned size_ondisk = k.k->size;
1120         unsigned nonce = UINT_MAX;
1121         unsigned nr_ptrs = 0;
1122         bool unwritten = false, have_ec = false, crc_since_last_ptr = false;
1123         int ret;
1124
1125         if (bkey_is_btree_ptr(k.k))
1126                 size_ondisk = btree_sectors(c);
1127
1128         bkey_extent_entry_for_each(ptrs, entry) {
1129                 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX) {
1130                         prt_printf(err, "invalid extent entry type (got %u, max %u)",
1131                                __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1132                         return -BCH_ERR_invalid_bkey;
1133                 }
1134
1135                 if (bkey_is_btree_ptr(k.k) &&
1136                     !extent_entry_is_ptr(entry)) {
1137                         prt_printf(err, "has non ptr field");
1138                         return -BCH_ERR_invalid_bkey;
1139                 }
1140
1141                 switch (extent_entry_type(entry)) {
1142                 case BCH_EXTENT_ENTRY_ptr:
1143                         ret = extent_ptr_invalid(c, k, &entry->ptr, size_ondisk,
1144                                                  false, err);
1145                         if (ret)
1146                                 return ret;
1147
1148                         if (nr_ptrs && unwritten != entry->ptr.unwritten) {
1149                                 prt_printf(err, "extent with unwritten and written ptrs");
1150                                 return -BCH_ERR_invalid_bkey;
1151                         }
1152
1153                         if (k.k->type != KEY_TYPE_extent && entry->ptr.unwritten) {
1154                                 prt_printf(err, "has unwritten ptrs");
1155                                 return -BCH_ERR_invalid_bkey;
1156                         }
1157
1158                         if (entry->ptr.cached && have_ec) {
1159                                 prt_printf(err, "cached, erasure coded ptr");
1160                                 return -BCH_ERR_invalid_bkey;
1161                         }
1162
1163                         unwritten = entry->ptr.unwritten;
1164                         have_ec = false;
1165                         crc_since_last_ptr = false;
1166                         nr_ptrs++;
1167                         break;
1168                 case BCH_EXTENT_ENTRY_crc32:
1169                 case BCH_EXTENT_ENTRY_crc64:
1170                 case BCH_EXTENT_ENTRY_crc128:
1171                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1172
1173                         if (crc.offset + crc.live_size >
1174                             crc.uncompressed_size) {
1175                                 prt_printf(err, "checksum offset + key size > uncompressed size");
1176                                 return -BCH_ERR_invalid_bkey;
1177                         }
1178
1179                         size_ondisk = crc.compressed_size;
1180
1181                         if (!bch2_checksum_type_valid(c, crc.csum_type)) {
1182                                 prt_printf(err, "invalid checksum type");
1183                                 return -BCH_ERR_invalid_bkey;
1184                         }
1185
1186                         if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR) {
1187                                 prt_printf(err, "invalid compression type");
1188                                 return -BCH_ERR_invalid_bkey;
1189                         }
1190
1191                         if (bch2_csum_type_is_encryption(crc.csum_type)) {
1192                                 if (nonce == UINT_MAX)
1193                                         nonce = crc.offset + crc.nonce;
1194                                 else if (nonce != crc.offset + crc.nonce) {
1195                                         prt_printf(err, "incorrect nonce");
1196                                         return -BCH_ERR_invalid_bkey;
1197                                 }
1198                         }
1199
1200                         if (crc_since_last_ptr) {
1201                                 prt_printf(err, "redundant crc entry");
1202                                 return -BCH_ERR_invalid_bkey;
1203                         }
1204                         crc_since_last_ptr = true;
1205                         break;
1206                 case BCH_EXTENT_ENTRY_stripe_ptr:
1207                         if (have_ec) {
1208                                 prt_printf(err, "redundant stripe entry");
1209                                 return -BCH_ERR_invalid_bkey;
1210                         }
1211                         have_ec = true;
1212                         break;
1213                 }
1214         }
1215
1216         if (!nr_ptrs) {
1217                 prt_str(err, "no ptrs");
1218                 return -BCH_ERR_invalid_bkey;
1219         }
1220
1221         if (nr_ptrs >= BCH_BKEY_PTRS_MAX) {
1222                 prt_str(err, "too many ptrs");
1223                 return -BCH_ERR_invalid_bkey;
1224         }
1225
1226         if (crc_since_last_ptr) {
1227                 prt_printf(err, "redundant crc entry");
1228                 return -BCH_ERR_invalid_bkey;
1229         }
1230
1231         if (have_ec) {
1232                 prt_printf(err, "redundant stripe entry");
1233                 return -BCH_ERR_invalid_bkey;
1234         }
1235
1236         return 0;
1237 }
1238
1239 void bch2_ptr_swab(struct bkey_s k)
1240 {
1241         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1242         union bch_extent_entry *entry;
1243         u64 *d;
1244
1245         for (d =  (u64 *) ptrs.start;
1246              d != (u64 *) ptrs.end;
1247              d++)
1248                 *d = swab64(*d);
1249
1250         for (entry = ptrs.start;
1251              entry < ptrs.end;
1252              entry = extent_entry_next(entry)) {
1253                 switch (extent_entry_type(entry)) {
1254                 case BCH_EXTENT_ENTRY_ptr:
1255                         break;
1256                 case BCH_EXTENT_ENTRY_crc32:
1257                         entry->crc32.csum = swab32(entry->crc32.csum);
1258                         break;
1259                 case BCH_EXTENT_ENTRY_crc64:
1260                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1261                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1262                         break;
1263                 case BCH_EXTENT_ENTRY_crc128:
1264                         entry->crc128.csum.hi = (__force __le64)
1265                                 swab64((__force u64) entry->crc128.csum.hi);
1266                         entry->crc128.csum.lo = (__force __le64)
1267                                 swab64((__force u64) entry->crc128.csum.lo);
1268                         break;
1269                 case BCH_EXTENT_ENTRY_stripe_ptr:
1270                         break;
1271                 }
1272         }
1273 }
1274
1275 /* Generic extent code: */
1276
1277 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1278 {
1279         unsigned new_val_u64s = bkey_val_u64s(k.k);
1280         int val_u64s_delta;
1281         u64 sub;
1282
1283         if (bkey_le(where, bkey_start_pos(k.k)))
1284                 return 0;
1285
1286         EBUG_ON(bkey_gt(where, k.k->p));
1287
1288         sub = where.offset - bkey_start_offset(k.k);
1289
1290         k.k->size -= sub;
1291
1292         if (!k.k->size) {
1293                 k.k->type = KEY_TYPE_deleted;
1294                 new_val_u64s = 0;
1295         }
1296
1297         switch (k.k->type) {
1298         case KEY_TYPE_extent:
1299         case KEY_TYPE_reflink_v: {
1300                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1301                 union bch_extent_entry *entry;
1302                 bool seen_crc = false;
1303
1304                 bkey_extent_entry_for_each(ptrs, entry) {
1305                         switch (extent_entry_type(entry)) {
1306                         case BCH_EXTENT_ENTRY_ptr:
1307                                 if (!seen_crc)
1308                                         entry->ptr.offset += sub;
1309                                 break;
1310                         case BCH_EXTENT_ENTRY_crc32:
1311                                 entry->crc32.offset += sub;
1312                                 break;
1313                         case BCH_EXTENT_ENTRY_crc64:
1314                                 entry->crc64.offset += sub;
1315                                 break;
1316                         case BCH_EXTENT_ENTRY_crc128:
1317                                 entry->crc128.offset += sub;
1318                                 break;
1319                         case BCH_EXTENT_ENTRY_stripe_ptr:
1320                                 break;
1321                         }
1322
1323                         if (extent_entry_is_crc(entry))
1324                                 seen_crc = true;
1325                 }
1326
1327                 break;
1328         }
1329         case KEY_TYPE_reflink_p: {
1330                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1331
1332                 le64_add_cpu(&p.v->idx, sub);
1333                 break;
1334         }
1335         case KEY_TYPE_inline_data:
1336         case KEY_TYPE_indirect_inline_data: {
1337                 void *p = bkey_inline_data_p(k);
1338                 unsigned bytes = bkey_inline_data_bytes(k.k);
1339
1340                 sub = min_t(u64, sub << 9, bytes);
1341
1342                 memmove(p, p + sub, bytes - sub);
1343
1344                 new_val_u64s -= sub >> 3;
1345                 break;
1346         }
1347         }
1348
1349         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1350         BUG_ON(val_u64s_delta < 0);
1351
1352         set_bkey_val_u64s(k.k, new_val_u64s);
1353         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1354         return -val_u64s_delta;
1355 }
1356
1357 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1358 {
1359         unsigned new_val_u64s = bkey_val_u64s(k.k);
1360         int val_u64s_delta;
1361         u64 len = 0;
1362
1363         if (bkey_ge(where, k.k->p))
1364                 return 0;
1365
1366         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1367
1368         len = where.offset - bkey_start_offset(k.k);
1369
1370         k.k->p.offset = where.offset;
1371         k.k->size = len;
1372
1373         if (!len) {
1374                 k.k->type = KEY_TYPE_deleted;
1375                 new_val_u64s = 0;
1376         }
1377
1378         switch (k.k->type) {
1379         case KEY_TYPE_inline_data:
1380         case KEY_TYPE_indirect_inline_data:
1381                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1382                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1383                 break;
1384         }
1385
1386         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1387         BUG_ON(val_u64s_delta < 0);
1388
1389         set_bkey_val_u64s(k.k, new_val_u64s);
1390         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1391         return -val_u64s_delta;
1392 }