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 union bch_extent_entry *__bch2_bkey_drop_ptr(struct bkey_s, struct bch_extent_ptr *);
  30
  31 static unsigned bch2_crc_field_size_max[] = {
  32         [BCH_EXTENT_ENTRY_crc32] = CRC32_SIZE_MAX,
  33         [BCH_EXTENT_ENTRY_crc64] = CRC64_SIZE_MAX,
  34         [BCH_EXTENT_ENTRY_crc128] = CRC128_SIZE_MAX,
  35 };
  36
  37 static void bch2_extent_crc_pack(union bch_extent_crc *,
  38                                  struct bch_extent_crc_unpacked,
  39                                  enum bch_extent_entry_type);
  40
  41 static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
  42                                                    unsigned dev)
  43 {
  44         struct bch_dev_io_failures *i;
  45
  46         for (i = f->devs; i < f->devs + f->nr; i++)
  47                 if (i->dev == dev)
  48                         return i;
  49
  50         return NULL;
  51 }
  52
  53 void bch2_mark_io_failure(struct bch_io_failures *failed,
  54                           struct extent_ptr_decoded *p)
  55 {
  56         struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
  57
  58         if (!f) {
  59                 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
  60
  61                 f = &failed->devs[failed->nr++];
  62                 f->dev          = p->ptr.dev;
  63                 f->idx          = p->idx;
  64                 f->nr_failed    = 1;
  65                 f->nr_retries   = 0;
  66         } else if (p->idx != f->idx) {
  67                 f->idx          = p->idx;
  68                 f->nr_failed    = 1;
  69                 f->nr_retries   = 0;
  70         } else {
  71                 f->nr_failed++;
  72         }
  73 }
  74
  75 /*
  76  * returns true if p1 is better than p2:
  77  */
  78 static inline bool ptr_better(struct bch_fs *c,
  79                               const struct extent_ptr_decoded p1,
  80                               const struct extent_ptr_decoded p2)
  81 {
  82         if (likely(!p1.idx && !p2.idx)) {
  83                 struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
  84                 struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
  85
  86                 u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
  87                 u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
  88
  89                 /* Pick at random, biased in favor of the faster device: */
  90
  91                 return bch2_rand_range(l1 + l2) > l1;
  92         }
  93
  94         if (bch2_force_reconstruct_read)
  95                 return p1.idx > p2.idx;
  96
  97         return p1.idx < p2.idx;
  98 }
  99
 100 /*
 101  * This picks a non-stale pointer, preferably from a device other than @avoid.
 102  * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
 103  * other devices, it will still pick a pointer from avoid.
 104  */
 105 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 106                                struct bch_io_failures *failed,
 107                                struct extent_ptr_decoded *pick)
 108 {
 109         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 110         const union bch_extent_entry *entry;
 111         struct extent_ptr_decoded p;
 112         struct bch_dev_io_failures *f;
 113         struct bch_dev *ca;
 114         int ret = 0;
 115
 116         if (k.k->type == KEY_TYPE_error)
 117                 return -EIO;
 118
 119         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 120                 /*
 121                  * Unwritten extent: no need to actually read, treat it as a
 122                  * hole and return 0s:
 123                  */
 124                 if (p.ptr.unwritten)
 125                         return 0;
 126
 127                 ca = bch_dev_bkey_exists(c, p.ptr.dev);
 128
 129                 /*
 130                  * If there are any dirty pointers it's an error if we can't
 131                  * read:
 132                  */
 133                 if (!ret && !p.ptr.cached)
 134                         ret = -EIO;
 135
 136                 if (p.ptr.cached && ptr_stale(ca, &p.ptr))
 137                         continue;
 138
 139                 f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
 140                 if (f)
 141                         p.idx = f->nr_failed < f->nr_retries
 142                                 ? f->idx
 143                                 : f->idx + 1;
 144
 145                 if (!p.idx &&
 146                     !bch2_dev_is_readable(ca))
 147                         p.idx++;
 148
 149                 if (bch2_force_reconstruct_read &&
 150                     !p.idx && p.has_ec)
 151                         p.idx++;
 152
 153                 if (p.idx >= (unsigned) p.has_ec + 1)
 154                         continue;
 155
 156                 if (ret > 0 && !ptr_better(c, p, *pick))
 157                         continue;
 158
 159                 *pick = p;
 160                 ret = 1;
 161         }
 162
 163         return ret;
 164 }
 165
 166 /* KEY_TYPE_btree_ptr: */
 167
 168 int bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k,
 169                            unsigned flags, struct printbuf *err)
 170 {
 171         if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX) {
 172                 prt_printf(err, "value too big (%zu > %u)",
 173                        bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 174                 return -BCH_ERR_invalid_bkey;
 175         }
 176
 177         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 178 }
 179
 180 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 181                             struct bkey_s_c k)
 182 {
 183         bch2_bkey_ptrs_to_text(out, c, k);
 184 }
 185
 186 int bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 187                               unsigned flags, struct printbuf *err)
 188 {
 189         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 190
 191         if (bkey_val_bytes(k.k) <= sizeof(*bp.v)) {
 192                 prt_printf(err, "value too small (%zu <= %zu)",
 193                        bkey_val_bytes(k.k), sizeof(*bp.v));
 194                 return -BCH_ERR_invalid_bkey;
 195         }
 196
 197         if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) {
 198                 prt_printf(err, "value too big (%zu > %zu)",
 199                        bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 200                 return -BCH_ERR_invalid_bkey;
 201         }
 202
 203         if (c->sb.version < bcachefs_metadata_version_snapshot &&
 204             bp.v->min_key.snapshot) {
 205                 prt_printf(err, "invalid min_key.snapshot (%u != 0)",
 206                        bp.v->min_key.snapshot);
 207                 return -BCH_ERR_invalid_bkey;
 208         }
 209
 210         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 211 }
 212
 213 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 214                                struct bkey_s_c k)
 215 {
 216         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 217
 218         prt_printf(out, "seq %llx written %u min_key %s",
 219                le64_to_cpu(bp.v->seq),
 220                le16_to_cpu(bp.v->sectors_written),
 221                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 222
 223         bch2_bpos_to_text(out, bp.v->min_key);
 224         prt_printf(out, " ");
 225         bch2_bkey_ptrs_to_text(out, c, k);
 226 }
 227
 228 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 229                               unsigned big_endian, int write,
 230                               struct bkey_s k)
 231 {
 232         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 233
 234         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 235
 236         if (version < bcachefs_metadata_version_inode_btree_change &&
 237             btree_node_type_is_extents(btree_id) &&
 238             !bkey_eq(bp.v->min_key, POS_MIN))
 239                 bp.v->min_key = write
 240                         ? bpos_nosnap_predecessor(bp.v->min_key)
 241                         : bpos_nosnap_successor(bp.v->min_key);
 242 }
 243
 244 /* KEY_TYPE_extent: */
 245
 246 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 247 {
 248         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 249         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 250         union bch_extent_entry *en_l;
 251         const union bch_extent_entry *en_r;
 252         struct extent_ptr_decoded lp, rp;
 253         bool use_right_ptr;
 254         struct bch_dev *ca;
 255
 256         en_l = l_ptrs.start;
 257         en_r = r_ptrs.start;
 258         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 259                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 260                         return false;
 261
 262                 en_l = extent_entry_next(en_l);
 263                 en_r = extent_entry_next(en_r);
 264         }
 265
 266         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 267                 return false;
 268
 269         en_l = l_ptrs.start;
 270         en_r = r_ptrs.start;
 271         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 272         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 273
 274         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 275                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 276                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 277                     rp.ptr.offset + rp.crc.offset ||
 278                     lp.ptr.dev                  != rp.ptr.dev ||
 279                     lp.ptr.gen                  != rp.ptr.gen ||
 280                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 281                     lp.has_ec                   != rp.has_ec)
 282                         return false;
 283
 284                 /* Extents may not straddle buckets: */
 285                 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
 286                 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
 287                         return false;
 288
 289                 if (lp.has_ec                   != rp.has_ec ||
 290                     (lp.has_ec &&
 291                      (lp.ec.block               != rp.ec.block ||
 292                       lp.ec.redundancy          != rp.ec.redundancy ||
 293                       lp.ec.idx                 != rp.ec.idx)))
 294                         return false;
 295
 296                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 297                     lp.crc.nonce                != rp.crc.nonce)
 298                         return false;
 299
 300                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 301                     lp.crc.uncompressed_size) {
 302                         /* can use left extent's crc entry */
 303                 } else if (lp.crc.live_size <= rp.crc.offset) {
 304                         /* can use right extent's crc entry */
 305                 } else {
 306                         /* check if checksums can be merged: */
 307                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 308                             lp.crc.nonce                != rp.crc.nonce ||
 309                             crc_is_compressed(lp.crc) ||
 310                             !bch2_checksum_mergeable(lp.crc.csum_type))
 311                                 return false;
 312
 313                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 314                             rp.crc.offset)
 315                                 return false;
 316
 317                         if (lp.crc.csum_type &&
 318                             lp.crc.uncompressed_size +
 319                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 320                                 return false;
 321                 }
 322
 323                 en_l = extent_entry_next(en_l);
 324                 en_r = extent_entry_next(en_r);
 325         }
 326
 327         en_l = l_ptrs.start;
 328         en_r = r_ptrs.start;
 329         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 330                 if (extent_entry_is_crc(en_l)) {
 331                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 332                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 333
 334                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 335                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 336                                 return false;
 337                 }
 338
 339                 en_l = extent_entry_next(en_l);
 340                 en_r = extent_entry_next(en_r);
 341         }
 342
 343         use_right_ptr = false;
 344         en_l = l_ptrs.start;
 345         en_r = r_ptrs.start;
 346         while (en_l < l_ptrs.end) {
 347                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 348                     use_right_ptr)
 349                         en_l->ptr = en_r->ptr;
 350
 351                 if (extent_entry_is_crc(en_l)) {
 352                         struct bch_extent_crc_unpacked crc_l =
 353                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 354                         struct bch_extent_crc_unpacked crc_r =
 355                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 356
 357                         use_right_ptr = false;
 358
 359                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 360                             crc_l.uncompressed_size) {
 361                                 /* can use left extent's crc entry */
 362                         } else if (crc_l.live_size <= crc_r.offset) {
 363                                 /* can use right extent's crc entry */
 364                                 crc_r.offset -= crc_l.live_size;
 365                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 366                                                      extent_entry_type(en_l));
 367                                 use_right_ptr = true;
 368                         } else {
 369                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 370                                                                  crc_l.csum,
 371                                                                  crc_r.csum,
 372                                                                  crc_r.uncompressed_size << 9);
 373
 374                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 375                                 crc_l.compressed_size   += crc_r.compressed_size;
 376                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 377                                                      extent_entry_type(en_l));
 378                         }
 379                 }
 380
 381                 en_l = extent_entry_next(en_l);
 382                 en_r = extent_entry_next(en_r);
 383         }
 384
 385         bch2_key_resize(l.k, l.k->size + r.k->size);
 386         return true;
 387 }
 388
 389 /* KEY_TYPE_reservation: */
 390
 391 int bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k,
 392                              unsigned flags, struct printbuf *err)
 393 {
 394         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 395
 396         if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation)) {
 397                 prt_printf(err, "incorrect value size (%zu != %zu)",
 398                        bkey_val_bytes(k.k), sizeof(*r.v));
 399                 return -BCH_ERR_invalid_bkey;
 400         }
 401
 402         if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX) {
 403                 prt_printf(err, "invalid nr_replicas (%u)",
 404                        r.v->nr_replicas);
 405                 return -BCH_ERR_invalid_bkey;
 406         }
 407
 408         return 0;
 409 }
 410
 411 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 412                               struct bkey_s_c k)
 413 {
 414         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 415
 416         prt_printf(out, "generation %u replicas %u",
 417                le32_to_cpu(r.v->generation),
 418                r.v->nr_replicas);
 419 }
 420
 421 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 422 {
 423         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 424         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 425
 426         if (l.v->generation != r.v->generation ||
 427             l.v->nr_replicas != r.v->nr_replicas)
 428                 return false;
 429
 430         bch2_key_resize(l.k, l.k->size + r.k->size);
 431         return true;
 432 }
 433
 434 /* Extent checksum entries: */
 435
 436 /* returns true if not equal */
 437 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 438                                          struct bch_extent_crc_unpacked r)
 439 {
 440         return (l.csum_type             != r.csum_type ||
 441                 l.compression_type      != r.compression_type ||
 442                 l.compressed_size       != r.compressed_size ||
 443                 l.uncompressed_size     != r.uncompressed_size ||
 444                 l.offset                != r.offset ||
 445                 l.live_size             != r.live_size ||
 446                 l.nonce                 != r.nonce ||
 447                 bch2_crc_cmp(l.csum, r.csum));
 448 }
 449
 450 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 451                                   struct bch_extent_crc_unpacked n)
 452 {
 453         return !crc_is_compressed(u) &&
 454                 u.csum_type &&
 455                 u.uncompressed_size > u.live_size &&
 456                 bch2_csum_type_is_encryption(u.csum_type) ==
 457                 bch2_csum_type_is_encryption(n.csum_type);
 458 }
 459
 460 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 461                                  struct bch_extent_crc_unpacked n)
 462 {
 463         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 464         struct bch_extent_crc_unpacked crc;
 465         const union bch_extent_entry *i;
 466
 467         if (!n.csum_type)
 468                 return false;
 469
 470         bkey_for_each_crc(k.k, ptrs, crc, i)
 471                 if (can_narrow_crc(crc, n))
 472                         return true;
 473
 474         return false;
 475 }
 476
 477 /*
 478  * We're writing another replica for this extent, so while we've got the data in
 479  * memory we'll be computing a new checksum for the currently live data.
 480  *
 481  * If there are other replicas we aren't moving, and they are checksummed but
 482  * not compressed, we can modify them to point to only the data that is
 483  * currently live (so that readers won't have to bounce) while we've got the
 484  * checksum we need:
 485  */
 486 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 487 {
 488         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 489         struct bch_extent_crc_unpacked u;
 490         struct extent_ptr_decoded p;
 491         union bch_extent_entry *i;
 492         bool ret = false;
 493
 494         /* Find a checksum entry that covers only live data: */
 495         if (!n.csum_type) {
 496                 bkey_for_each_crc(&k->k, ptrs, u, i)
 497                         if (!crc_is_compressed(u) &&
 498                             u.csum_type &&
 499                             u.live_size == u.uncompressed_size) {
 500                                 n = u;
 501                                 goto found;
 502                         }
 503                 return false;
 504         }
 505 found:
 506         BUG_ON(crc_is_compressed(n));
 507         BUG_ON(n.offset);
 508         BUG_ON(n.live_size != k->k.size);
 509
 510 restart_narrow_pointers:
 511         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 512
 513         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 514                 if (can_narrow_crc(p.crc, n)) {
 515                         __bch2_bkey_drop_ptr(bkey_i_to_s(k), &i->ptr);
 516                         p.ptr.offset += p.crc.offset;
 517                         p.crc = n;
 518                         bch2_extent_ptr_decoded_append(k, &p);
 519                         ret = true;
 520                         goto restart_narrow_pointers;
 521                 }
 522
 523         return ret;
 524 }
 525
 526 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 527                                  struct bch_extent_crc_unpacked src,
 528                                  enum bch_extent_entry_type type)
 529 {
 530 #define set_common_fields(_dst, _src)                                   \
 531                 _dst.type               = 1 << type;                    \
 532                 _dst.csum_type          = _src.csum_type,               \
 533                 _dst.compression_type   = _src.compression_type,        \
 534                 _dst._compressed_size   = _src.compressed_size - 1,     \
 535                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 536                 _dst.offset             = _src.offset
 537
 538         switch (type) {
 539         case BCH_EXTENT_ENTRY_crc32:
 540                 set_common_fields(dst->crc32, src);
 541                 dst->crc32.csum  = *((__le32 *) &src.csum.lo);
 542                 break;
 543         case BCH_EXTENT_ENTRY_crc64:
 544                 set_common_fields(dst->crc64, src);
 545                 dst->crc64.nonce        = src.nonce;
 546                 dst->crc64.csum_lo      = src.csum.lo;
 547                 dst->crc64.csum_hi      = *((__le16 *) &src.csum.hi);
 548                 break;
 549         case BCH_EXTENT_ENTRY_crc128:
 550                 set_common_fields(dst->crc128, src);
 551                 dst->crc128.nonce       = src.nonce;
 552                 dst->crc128.csum        = src.csum;
 553                 break;
 554         default:
 555                 BUG();
 556         }
 557 #undef set_common_fields
 558 }
 559
 560 void bch2_extent_crc_append(struct bkey_i *k,
 561                             struct bch_extent_crc_unpacked new)
 562 {
 563         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 564         union bch_extent_crc *crc = (void *) ptrs.end;
 565         enum bch_extent_entry_type type;
 566
 567         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 568             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 569             new.nonce                           <= CRC32_NONCE_MAX)
 570                 type = BCH_EXTENT_ENTRY_crc32;
 571         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 572                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 573                    new.nonce                    <= CRC64_NONCE_MAX)
 574                 type = BCH_EXTENT_ENTRY_crc64;
 575         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 576                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 577                    new.nonce                    <= CRC128_NONCE_MAX)
 578                 type = BCH_EXTENT_ENTRY_crc128;
 579         else
 580                 BUG();
 581
 582         bch2_extent_crc_pack(crc, new, type);
 583
 584         k->k.u64s += extent_entry_u64s(ptrs.end);
 585
 586         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 587 }
 588
 589 /* Generic code for keys with pointers: */
 590
 591 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 592 {
 593         return bch2_bkey_devs(k).nr;
 594 }
 595
 596 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 597 {
 598         return k.k->type == KEY_TYPE_reservation
 599                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 600                 : bch2_bkey_dirty_devs(k).nr;
 601 }
 602
 603 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 604 {
 605         unsigned ret = 0;
 606
 607         if (k.k->type == KEY_TYPE_reservation) {
 608                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 609         } else {
 610                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 611                 const union bch_extent_entry *entry;
 612                 struct extent_ptr_decoded p;
 613
 614                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 615                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 616         }
 617
 618         return ret;
 619 }
 620
 621 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 622 {
 623         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 624         const union bch_extent_entry *entry;
 625         struct extent_ptr_decoded p;
 626         unsigned ret = 0;
 627
 628         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 629                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 630                         ret += p.crc.compressed_size;
 631
 632         return ret;
 633 }
 634
 635 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
 636 {
 637         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 638         const union bch_extent_entry *entry;
 639         struct bch_extent_crc_unpacked crc;
 640
 641         bkey_for_each_crc(k.k, ptrs, crc, entry)
 642                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 643                         return true;
 644         return false;
 645 }
 646
 647 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 648 {
 649         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 650         const union bch_extent_entry *entry;
 651         struct extent_ptr_decoded p = { 0 };
 652         unsigned replicas = 0;
 653
 654         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 655                 if (p.ptr.cached)
 656                         continue;
 657
 658                 if (p.has_ec)
 659                         replicas += p.ec.redundancy;
 660
 661                 replicas++;
 662
 663         }
 664
 665         return replicas;
 666 }
 667
 668 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 669 {
 670         unsigned durability = 0;
 671         struct bch_dev *ca;
 672
 673         if (p->ptr.cached)
 674                 return 0;
 675
 676         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 677
 678         if (ca->mi.state != BCH_MEMBER_STATE_failed)
 679                 durability = max_t(unsigned, durability, ca->mi.durability);
 680
 681         if (p->has_ec)
 682                 durability += p->ec.redundancy;
 683
 684         return durability;
 685 }
 686
 687 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 688 {
 689         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 690         const union bch_extent_entry *entry;
 691         struct extent_ptr_decoded p;
 692         unsigned durability = 0;
 693
 694         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 695                 durability += bch2_extent_ptr_durability(c,& p);
 696
 697         return durability;
 698 }
 699
 700 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
 701 {
 702         union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
 703         union bch_extent_entry *next = extent_entry_next(entry);
 704
 705         memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
 706         k->k.u64s -= extent_entry_u64s(entry);
 707 }
 708
 709 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
 710                                     struct extent_ptr_decoded *p)
 711 {
 712         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 713         struct bch_extent_crc_unpacked crc =
 714                 bch2_extent_crc_unpack(&k->k, NULL);
 715         union bch_extent_entry *pos;
 716
 717         if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 718                 pos = ptrs.start;
 719                 goto found;
 720         }
 721
 722         bkey_for_each_crc(&k->k, ptrs, crc, pos)
 723                 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 724                         pos = extent_entry_next(pos);
 725                         goto found;
 726                 }
 727
 728         bch2_extent_crc_append(k, p->crc);
 729         pos = bkey_val_end(bkey_i_to_s(k));
 730 found:
 731         p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
 732         __extent_entry_insert(k, pos, to_entry(&p->ptr));
 733
 734         if (p->has_ec) {
 735                 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
 736                 __extent_entry_insert(k, pos, to_entry(&p->ec));
 737         }
 738 }
 739
 740 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 741                                           union bch_extent_entry *entry)
 742 {
 743         union bch_extent_entry *i = ptrs.start;
 744
 745         if (i == entry)
 746                 return NULL;
 747
 748         while (extent_entry_next(i) != entry)
 749                 i = extent_entry_next(i);
 750         return i;
 751 }
 752
 753 static void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
 754 {
 755         union bch_extent_entry *next = extent_entry_next(entry);
 756
 757         /* stripes have ptrs, but their layout doesn't work with this code */
 758         BUG_ON(k.k->type == KEY_TYPE_stripe);
 759
 760         memmove_u64s_down(entry, next,
 761                           (u64 *) bkey_val_end(k) - (u64 *) next);
 762         k.k->u64s -= (u64 *) next - (u64 *) entry;
 763 }
 764
 765 /*
 766  * Returns pointer to the next entry after the one being dropped:
 767  */
 768 static union bch_extent_entry *__bch2_bkey_drop_ptr(struct bkey_s k,
 769                                            struct bch_extent_ptr *ptr)
 770 {
 771         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 772         union bch_extent_entry *entry = to_entry(ptr), *next;
 773         union bch_extent_entry *ret = entry;
 774         bool drop_crc = true;
 775
 776         EBUG_ON(ptr < &ptrs.start->ptr ||
 777                 ptr >= &ptrs.end->ptr);
 778         EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
 779
 780         for (next = extent_entry_next(entry);
 781              next != ptrs.end;
 782              next = extent_entry_next(next)) {
 783                 if (extent_entry_is_crc(next)) {
 784                         break;
 785                 } else if (extent_entry_is_ptr(next)) {
 786                         drop_crc = false;
 787                         break;
 788                 }
 789         }
 790
 791         extent_entry_drop(k, entry);
 792
 793         while ((entry = extent_entry_prev(ptrs, entry))) {
 794                 if (extent_entry_is_ptr(entry))
 795                         break;
 796
 797                 if ((extent_entry_is_crc(entry) && drop_crc) ||
 798                     extent_entry_is_stripe_ptr(entry)) {
 799                         ret = (void *) ret - extent_entry_bytes(entry);
 800                         extent_entry_drop(k, entry);
 801                 }
 802         }
 803
 804         return ret;
 805 }
 806
 807 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 808                                            struct bch_extent_ptr *ptr)
 809 {
 810         bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
 811         union bch_extent_entry *ret =
 812                 __bch2_bkey_drop_ptr(k, ptr);
 813
 814         /*
 815          * If we deleted all the dirty pointers and there's still cached
 816          * pointers, we could set the cached pointers to dirty if they're not
 817          * stale - but to do that correctly we'd need to grab an open_bucket
 818          * reference so that we don't race with bucket reuse:
 819          */
 820         if (have_dirty &&
 821             !bch2_bkey_dirty_devs(k.s_c).nr) {
 822                 k.k->type = KEY_TYPE_error;
 823                 set_bkey_val_u64s(k.k, 0);
 824                 ret = NULL;
 825         } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 826                 k.k->type = KEY_TYPE_deleted;
 827                 set_bkey_val_u64s(k.k, 0);
 828                 ret = NULL;
 829         }
 830
 831         return ret;
 832 }
 833
 834 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 835 {
 836         struct bch_extent_ptr *ptr;
 837
 838         bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 839 }
 840
 841 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 842 {
 843         struct bch_extent_ptr *ptr = (void *) bch2_bkey_has_device(k.s_c, dev);
 844
 845         if (ptr)
 846                 __bch2_bkey_drop_ptr(k, ptr);
 847 }
 848
 849 const struct bch_extent_ptr *
 850 bch2_bkey_has_device(struct bkey_s_c k, unsigned dev)
 851 {
 852         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 853         const struct bch_extent_ptr *ptr;
 854
 855         bkey_for_each_ptr(ptrs, ptr)
 856                 if (ptr->dev == dev)
 857                         return ptr;
 858
 859         return NULL;
 860 }
 861
 862 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 863 {
 864         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 865         const struct bch_extent_ptr *ptr;
 866
 867         bkey_for_each_ptr(ptrs, ptr)
 868                 if (bch2_dev_in_target(c, ptr->dev, target) &&
 869                     (!ptr->cached ||
 870                      !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
 871                         return true;
 872
 873         return false;
 874 }
 875
 876 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 877                            struct bch_extent_ptr m, u64 offset)
 878 {
 879         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 880         const union bch_extent_entry *entry;
 881         struct extent_ptr_decoded p;
 882
 883         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 884                 if (p.ptr.dev   == m.dev &&
 885                     p.ptr.gen   == m.gen &&
 886                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 887                     (s64) m.offset  - offset)
 888                         return true;
 889
 890         return false;
 891 }
 892
 893 /*
 894  * Returns true if two extents refer to the same data:
 895  */
 896 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 897 {
 898         if (k1.k->type != k2.k->type)
 899                 return false;
 900
 901         if (bkey_extent_is_direct_data(k1.k)) {
 902                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 903                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 904                 const union bch_extent_entry *entry1, *entry2;
 905                 struct extent_ptr_decoded p1, p2;
 906
 907                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 908                         return false;
 909
 910                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 911                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 912                         if (p1.ptr.dev          == p2.ptr.dev &&
 913                             p1.ptr.gen          == p2.ptr.gen &&
 914                             (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 915                             (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 916                                 return true;
 917
 918                 return false;
 919         } else {
 920                 /* KEY_TYPE_deleted, etc. */
 921                 return true;
 922         }
 923 }
 924
 925 bool bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1,
 926                          struct bkey_s_c k2)
 927 {
 928         struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 929         const union bch_extent_entry *entry2;
 930         struct extent_ptr_decoded p2;
 931
 932         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 933                 if (p1.ptr.dev          == p2.ptr.dev &&
 934                     p1.ptr.gen          == p2.ptr.gen &&
 935                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 936                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 937                         return true;
 938
 939         return false;
 940 }
 941
 942 void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
 943 {
 944         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 945         union bch_extent_entry *entry;
 946         union bch_extent_entry *ec = NULL;
 947
 948         bkey_extent_entry_for_each(ptrs, entry) {
 949                 if (&entry->ptr == ptr) {
 950                         ptr->cached = true;
 951                         if (ec)
 952                                 extent_entry_drop(k, ec);
 953                         return;
 954                 }
 955
 956                 if (extent_entry_is_stripe_ptr(entry))
 957                         ec = entry;
 958                 else if (extent_entry_is_ptr(entry))
 959                         ec = NULL;
 960         }
 961
 962         BUG();
 963 }
 964
 965 /*
 966  * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 967  *
 968  * Returns true if @k should be dropped entirely
 969  *
 970  * For existing keys, only called when btree nodes are being rewritten, not when
 971  * they're merely being compacted/resorted in memory.
 972  */
 973 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 974 {
 975         struct bch_extent_ptr *ptr;
 976
 977         bch2_bkey_drop_ptrs(k, ptr,
 978                 ptr->cached &&
 979                 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
 980
 981         return bkey_deleted(k.k);
 982 }
 983
 984 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 985                             struct bkey_s_c k)
 986 {
 987         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 988         const union bch_extent_entry *entry;
 989         struct bch_extent_crc_unpacked crc;
 990         const struct bch_extent_ptr *ptr;
 991         const struct bch_extent_stripe_ptr *ec;
 992         struct bch_dev *ca;
 993         bool first = true;
 994
 995         bkey_extent_entry_for_each(ptrs, entry) {
 996                 if (!first)
 997                         prt_printf(out, " ");
 998
 999                 switch (__extent_entry_type(entry)) {
1000                 case BCH_EXTENT_ENTRY_ptr:
1001                         ptr = entry_to_ptr(entry);
1002                         ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
1003                                 ? bch_dev_bkey_exists(c, ptr->dev)
1004                                 : NULL;
1005
1006                         if (!ca) {
1007                                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1008                                        (u64) ptr->offset, ptr->gen,
1009                                        ptr->cached ? " cached" : "");
1010                         } else {
1011                                 u32 offset;
1012                                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1013
1014                                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1015                                            ptr->dev, b, offset, ptr->gen);
1016                                 if (ptr->cached)
1017                                         prt_str(out, " cached");
1018                                 if (ptr->unwritten)
1019                                         prt_str(out, " unwritten");
1020                                 if (ca && ptr_stale(ca, ptr))
1021                                         prt_printf(out, " stale");
1022                         }
1023                         break;
1024                 case BCH_EXTENT_ENTRY_crc32:
1025                 case BCH_EXTENT_ENTRY_crc64:
1026                 case BCH_EXTENT_ENTRY_crc128:
1027                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1028
1029                         prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
1030                                crc.compressed_size,
1031                                crc.uncompressed_size,
1032                                crc.offset, crc.nonce,
1033                                bch2_csum_types[crc.csum_type],
1034                                bch2_compression_types[crc.compression_type]);
1035                         break;
1036                 case BCH_EXTENT_ENTRY_stripe_ptr:
1037                         ec = &entry->stripe_ptr;
1038
1039                         prt_printf(out, "ec: idx %llu block %u",
1040                                (u64) ec->idx, ec->block);
1041                         break;
1042                 default:
1043                         prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1044                         return;
1045                 }
1046
1047                 first = false;
1048         }
1049 }
1050
1051 static int extent_ptr_invalid(const struct bch_fs *c,
1052                               struct bkey_s_c k,
1053                               const struct bch_extent_ptr *ptr,
1054                               unsigned size_ondisk,
1055                               bool metadata,
1056                               struct printbuf *err)
1057 {
1058         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1059         const struct bch_extent_ptr *ptr2;
1060         u64 bucket;
1061         u32 bucket_offset;
1062         struct bch_dev *ca;
1063
1064         if (!bch2_dev_exists2(c, ptr->dev)) {
1065                 prt_printf(err, "pointer to invalid device (%u)", ptr->dev);
1066                 return -BCH_ERR_invalid_bkey;
1067         }
1068
1069         ca = bch_dev_bkey_exists(c, ptr->dev);
1070         bkey_for_each_ptr(ptrs, ptr2)
1071                 if (ptr != ptr2 && ptr->dev == ptr2->dev) {
1072                         prt_printf(err, "multiple pointers to same device (%u)", ptr->dev);
1073                         return -BCH_ERR_invalid_bkey;
1074                 }
1075
1076         bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1077
1078         if (bucket >= ca->mi.nbuckets) {
1079                 prt_printf(err, "pointer past last bucket (%llu > %llu)",
1080                        bucket, ca->mi.nbuckets);
1081                 return -BCH_ERR_invalid_bkey;
1082         }
1083
1084         if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket)) {
1085                 prt_printf(err, "pointer before first bucket (%llu < %u)",
1086                        bucket, ca->mi.first_bucket);
1087                 return -BCH_ERR_invalid_bkey;
1088         }
1089
1090         if (bucket_offset + size_ondisk > ca->mi.bucket_size) {
1091                 prt_printf(err, "pointer spans multiple buckets (%u + %u > %u)",
1092                        bucket_offset, size_ondisk, ca->mi.bucket_size);
1093                 return -BCH_ERR_invalid_bkey;
1094         }
1095
1096         return 0;
1097 }
1098
1099 int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
1100                            unsigned flags, struct printbuf *err)
1101 {
1102         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1103         const union bch_extent_entry *entry;
1104         struct bch_extent_crc_unpacked crc;
1105         unsigned size_ondisk = k.k->size;
1106         unsigned nonce = UINT_MAX;
1107         unsigned nr_ptrs = 0;
1108         bool unwritten = false, have_ec = false, crc_since_last_ptr = false;
1109         int ret;
1110
1111         if (bkey_is_btree_ptr(k.k))
1112                 size_ondisk = btree_sectors(c);
1113
1114         bkey_extent_entry_for_each(ptrs, entry) {
1115                 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX) {
1116                         prt_printf(err, "invalid extent entry type (got %u, max %u)",
1117                                __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1118                         return -BCH_ERR_invalid_bkey;
1119                 }
1120
1121                 if (bkey_is_btree_ptr(k.k) &&
1122                     !extent_entry_is_ptr(entry)) {
1123                         prt_printf(err, "has non ptr field");
1124                         return -BCH_ERR_invalid_bkey;
1125                 }
1126
1127                 switch (extent_entry_type(entry)) {
1128                 case BCH_EXTENT_ENTRY_ptr:
1129                         ret = extent_ptr_invalid(c, k, &entry->ptr, size_ondisk,
1130                                                  false, err);
1131                         if (ret)
1132                                 return ret;
1133
1134                         if (nr_ptrs && unwritten != entry->ptr.unwritten) {
1135                                 prt_printf(err, "extent with unwritten and written ptrs");
1136                                 return -BCH_ERR_invalid_bkey;
1137                         }
1138
1139                         if (k.k->type != KEY_TYPE_extent && entry->ptr.unwritten) {
1140                                 prt_printf(err, "has unwritten ptrs");
1141                                 return -BCH_ERR_invalid_bkey;
1142                         }
1143
1144                         if (entry->ptr.cached && have_ec) {
1145                                 prt_printf(err, "cached, erasure coded ptr");
1146                                 return -BCH_ERR_invalid_bkey;
1147                         }
1148
1149                         unwritten = entry->ptr.unwritten;
1150                         have_ec = false;
1151                         crc_since_last_ptr = false;
1152                         nr_ptrs++;
1153                         break;
1154                 case BCH_EXTENT_ENTRY_crc32:
1155                 case BCH_EXTENT_ENTRY_crc64:
1156                 case BCH_EXTENT_ENTRY_crc128:
1157                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1158
1159                         if (crc.offset + crc.live_size >
1160                             crc.uncompressed_size) {
1161                                 prt_printf(err, "checksum offset + key size > uncompressed size");
1162                                 return -BCH_ERR_invalid_bkey;
1163                         }
1164
1165                         size_ondisk = crc.compressed_size;
1166
1167                         if (!bch2_checksum_type_valid(c, crc.csum_type)) {
1168                                 prt_printf(err, "invalid checksum type");
1169                                 return -BCH_ERR_invalid_bkey;
1170                         }
1171
1172                         if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR) {
1173                                 prt_printf(err, "invalid compression type");
1174                                 return -BCH_ERR_invalid_bkey;
1175                         }
1176
1177                         if (bch2_csum_type_is_encryption(crc.csum_type)) {
1178                                 if (nonce == UINT_MAX)
1179                                         nonce = crc.offset + crc.nonce;
1180                                 else if (nonce != crc.offset + crc.nonce) {
1181                                         prt_printf(err, "incorrect nonce");
1182                                         return -BCH_ERR_invalid_bkey;
1183                                 }
1184                         }
1185
1186                         if (crc_since_last_ptr) {
1187                                 prt_printf(err, "redundant crc entry");
1188                                 return -BCH_ERR_invalid_bkey;
1189                         }
1190                         crc_since_last_ptr = true;
1191                         break;
1192                 case BCH_EXTENT_ENTRY_stripe_ptr:
1193                         if (have_ec) {
1194                                 prt_printf(err, "redundant stripe entry");
1195                                 return -BCH_ERR_invalid_bkey;
1196                         }
1197                         have_ec = true;
1198                         break;
1199                 }
1200         }
1201
1202         if (!nr_ptrs) {
1203                 prt_str(err, "no ptrs");
1204                 return -BCH_ERR_invalid_bkey;
1205         }
1206
1207         if (nr_ptrs >= BCH_BKEY_PTRS_MAX) {
1208                 prt_str(err, "too many ptrs");
1209                 return -BCH_ERR_invalid_bkey;
1210         }
1211
1212         if (crc_since_last_ptr) {
1213                 prt_printf(err, "redundant crc entry");
1214                 return -BCH_ERR_invalid_bkey;
1215         }
1216
1217         if (have_ec) {
1218                 prt_printf(err, "redundant stripe entry");
1219                 return -BCH_ERR_invalid_bkey;
1220         }
1221
1222         return 0;
1223 }
1224
1225 void bch2_ptr_swab(struct bkey_s k)
1226 {
1227         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1228         union bch_extent_entry *entry;
1229         u64 *d;
1230
1231         for (d =  (u64 *) ptrs.start;
1232              d != (u64 *) ptrs.end;
1233              d++)
1234                 *d = swab64(*d);
1235
1236         for (entry = ptrs.start;
1237              entry < ptrs.end;
1238              entry = extent_entry_next(entry)) {
1239                 switch (extent_entry_type(entry)) {
1240                 case BCH_EXTENT_ENTRY_ptr:
1241                         break;
1242                 case BCH_EXTENT_ENTRY_crc32:
1243                         entry->crc32.csum = swab32(entry->crc32.csum);
1244                         break;
1245                 case BCH_EXTENT_ENTRY_crc64:
1246                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1247                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1248                         break;
1249                 case BCH_EXTENT_ENTRY_crc128:
1250                         entry->crc128.csum.hi = (__force __le64)
1251                                 swab64((__force u64) entry->crc128.csum.hi);
1252                         entry->crc128.csum.lo = (__force __le64)
1253                                 swab64((__force u64) entry->crc128.csum.lo);
1254                         break;
1255                 case BCH_EXTENT_ENTRY_stripe_ptr:
1256                         break;
1257                 }
1258         }
1259 }
1260
1261 /* Generic extent code: */
1262
1263 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1264 {
1265         unsigned new_val_u64s = bkey_val_u64s(k.k);
1266         int val_u64s_delta;
1267         u64 sub;
1268
1269         if (bkey_le(where, bkey_start_pos(k.k)))
1270                 return 0;
1271
1272         EBUG_ON(bkey_gt(where, k.k->p));
1273
1274         sub = where.offset - bkey_start_offset(k.k);
1275
1276         k.k->size -= sub;
1277
1278         if (!k.k->size) {
1279                 k.k->type = KEY_TYPE_deleted;
1280                 new_val_u64s = 0;
1281         }
1282
1283         switch (k.k->type) {
1284         case KEY_TYPE_extent:
1285         case KEY_TYPE_reflink_v: {
1286                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1287                 union bch_extent_entry *entry;
1288                 bool seen_crc = false;
1289
1290                 bkey_extent_entry_for_each(ptrs, entry) {
1291                         switch (extent_entry_type(entry)) {
1292                         case BCH_EXTENT_ENTRY_ptr:
1293                                 if (!seen_crc)
1294                                         entry->ptr.offset += sub;
1295                                 break;
1296                         case BCH_EXTENT_ENTRY_crc32:
1297                                 entry->crc32.offset += sub;
1298                                 break;
1299                         case BCH_EXTENT_ENTRY_crc64:
1300                                 entry->crc64.offset += sub;
1301                                 break;
1302                         case BCH_EXTENT_ENTRY_crc128:
1303                                 entry->crc128.offset += sub;
1304                                 break;
1305                         case BCH_EXTENT_ENTRY_stripe_ptr:
1306                                 break;
1307                         }
1308
1309                         if (extent_entry_is_crc(entry))
1310                                 seen_crc = true;
1311                 }
1312
1313                 break;
1314         }
1315         case KEY_TYPE_reflink_p: {
1316                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1317
1318                 le64_add_cpu(&p.v->idx, sub);
1319                 break;
1320         }
1321         case KEY_TYPE_inline_data:
1322         case KEY_TYPE_indirect_inline_data: {
1323                 void *p = bkey_inline_data_p(k);
1324                 unsigned bytes = bkey_inline_data_bytes(k.k);
1325
1326                 sub = min_t(u64, sub << 9, bytes);
1327
1328                 memmove(p, p + sub, bytes - sub);
1329
1330                 new_val_u64s -= sub >> 3;
1331                 break;
1332         }
1333         }
1334
1335         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1336         BUG_ON(val_u64s_delta < 0);
1337
1338         set_bkey_val_u64s(k.k, new_val_u64s);
1339         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1340         return -val_u64s_delta;
1341 }
1342
1343 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1344 {
1345         unsigned new_val_u64s = bkey_val_u64s(k.k);
1346         int val_u64s_delta;
1347         u64 len = 0;
1348
1349         if (bkey_ge(where, k.k->p))
1350                 return 0;
1351
1352         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1353
1354         len = where.offset - bkey_start_offset(k.k);
1355
1356         k.k->p.offset = where.offset;
1357         k.k->size = len;
1358
1359         if (!len) {
1360                 k.k->type = KEY_TYPE_deleted;
1361                 new_val_u64s = 0;
1362         }
1363
1364         switch (k.k->type) {
1365         case KEY_TYPE_inline_data:
1366         case KEY_TYPE_indirect_inline_data:
1367                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1368                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1369                 break;
1370         }
1371
1372         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1373         BUG_ON(val_u64s_delta < 0);
1374
1375         set_bkey_val_u64s(k.k, new_val_u64s);
1376         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1377         return -val_u64s_delta;
1378 }