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 "compress.h"
  17 #include "debug.h"
  18 #include "disk_groups.h"
  19 #include "error.h"
  20 #include "extents.h"
  21 #include "inode.h"
  22 #include "journal.h"
  23 #include "replicas.h"
  24 #include "super.h"
  25 #include "super-io.h"
  26 #include "trace.h"
  27 #include "util.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(struct bch_fs *c, struct bkey_s_c k,
 167                            enum bkey_invalid_flags flags,
 168                            struct printbuf *err)
 169 {
 170         int ret = 0;
 171
 172         bkey_fsck_err_on(bkey_val_u64s(k.k) > BCH_REPLICAS_MAX, c, err,
 173                          btree_ptr_val_too_big,
 174                          "value too big (%zu > %u)", bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 175
 176         ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
 177 fsck_err:
 178         return ret;
 179 }
 180
 181 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 182                             struct bkey_s_c k)
 183 {
 184         bch2_bkey_ptrs_to_text(out, c, k);
 185 }
 186
 187 int bch2_btree_ptr_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 188                               enum bkey_invalid_flags flags,
 189                               struct printbuf *err)
 190 {
 191         int ret = 0;
 192
 193         bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX, c, err,
 194                          btree_ptr_v2_val_too_big,
 195                          "value too big (%zu > %zu)",
 196                          bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 197
 198         ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
 199 fsck_err:
 200         return ret;
 201 }
 202
 203 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 204                                struct bkey_s_c k)
 205 {
 206         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 207
 208         prt_printf(out, "seq %llx written %u min_key %s",
 209                le64_to_cpu(bp.v->seq),
 210                le16_to_cpu(bp.v->sectors_written),
 211                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 212
 213         bch2_bpos_to_text(out, bp.v->min_key);
 214         prt_printf(out, " ");
 215         bch2_bkey_ptrs_to_text(out, c, k);
 216 }
 217
 218 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 219                               unsigned big_endian, int write,
 220                               struct bkey_s k)
 221 {
 222         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 223
 224         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 225
 226         if (version < bcachefs_metadata_version_inode_btree_change &&
 227             btree_id_is_extents(btree_id) &&
 228             !bkey_eq(bp.v->min_key, POS_MIN))
 229                 bp.v->min_key = write
 230                         ? bpos_nosnap_predecessor(bp.v->min_key)
 231                         : bpos_nosnap_successor(bp.v->min_key);
 232 }
 233
 234 /* KEY_TYPE_extent: */
 235
 236 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 237 {
 238         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 239         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 240         union bch_extent_entry *en_l;
 241         const union bch_extent_entry *en_r;
 242         struct extent_ptr_decoded lp, rp;
 243         bool use_right_ptr;
 244         struct bch_dev *ca;
 245
 246         en_l = l_ptrs.start;
 247         en_r = r_ptrs.start;
 248         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 249                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 250                         return false;
 251
 252                 en_l = extent_entry_next(en_l);
 253                 en_r = extent_entry_next(en_r);
 254         }
 255
 256         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 257                 return false;
 258
 259         en_l = l_ptrs.start;
 260         en_r = r_ptrs.start;
 261         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 262         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 263
 264         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 265                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 266                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 267                     rp.ptr.offset + rp.crc.offset ||
 268                     lp.ptr.dev                  != rp.ptr.dev ||
 269                     lp.ptr.gen                  != rp.ptr.gen ||
 270                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 271                     lp.has_ec                   != rp.has_ec)
 272                         return false;
 273
 274                 /* Extents may not straddle buckets: */
 275                 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
 276                 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
 277                         return false;
 278
 279                 if (lp.has_ec                   != rp.has_ec ||
 280                     (lp.has_ec &&
 281                      (lp.ec.block               != rp.ec.block ||
 282                       lp.ec.redundancy          != rp.ec.redundancy ||
 283                       lp.ec.idx                 != rp.ec.idx)))
 284                         return false;
 285
 286                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 287                     lp.crc.nonce                != rp.crc.nonce)
 288                         return false;
 289
 290                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 291                     lp.crc.uncompressed_size) {
 292                         /* can use left extent's crc entry */
 293                 } else if (lp.crc.live_size <= rp.crc.offset) {
 294                         /* can use right extent's crc entry */
 295                 } else {
 296                         /* check if checksums can be merged: */
 297                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 298                             lp.crc.nonce                != rp.crc.nonce ||
 299                             crc_is_compressed(lp.crc) ||
 300                             !bch2_checksum_mergeable(lp.crc.csum_type))
 301                                 return false;
 302
 303                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 304                             rp.crc.offset)
 305                                 return false;
 306
 307                         if (lp.crc.csum_type &&
 308                             lp.crc.uncompressed_size +
 309                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 310                                 return false;
 311                 }
 312
 313                 en_l = extent_entry_next(en_l);
 314                 en_r = extent_entry_next(en_r);
 315         }
 316
 317         en_l = l_ptrs.start;
 318         en_r = r_ptrs.start;
 319         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 320                 if (extent_entry_is_crc(en_l)) {
 321                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 322                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 323
 324                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 325                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 326                                 return false;
 327                 }
 328
 329                 en_l = extent_entry_next(en_l);
 330                 en_r = extent_entry_next(en_r);
 331         }
 332
 333         use_right_ptr = false;
 334         en_l = l_ptrs.start;
 335         en_r = r_ptrs.start;
 336         while (en_l < l_ptrs.end) {
 337                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 338                     use_right_ptr)
 339                         en_l->ptr = en_r->ptr;
 340
 341                 if (extent_entry_is_crc(en_l)) {
 342                         struct bch_extent_crc_unpacked crc_l =
 343                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 344                         struct bch_extent_crc_unpacked crc_r =
 345                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 346
 347                         use_right_ptr = false;
 348
 349                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 350                             crc_l.uncompressed_size) {
 351                                 /* can use left extent's crc entry */
 352                         } else if (crc_l.live_size <= crc_r.offset) {
 353                                 /* can use right extent's crc entry */
 354                                 crc_r.offset -= crc_l.live_size;
 355                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 356                                                      extent_entry_type(en_l));
 357                                 use_right_ptr = true;
 358                         } else {
 359                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 360                                                                  crc_l.csum,
 361                                                                  crc_r.csum,
 362                                                                  crc_r.uncompressed_size << 9);
 363
 364                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 365                                 crc_l.compressed_size   += crc_r.compressed_size;
 366                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 367                                                      extent_entry_type(en_l));
 368                         }
 369                 }
 370
 371                 en_l = extent_entry_next(en_l);
 372                 en_r = extent_entry_next(en_r);
 373         }
 374
 375         bch2_key_resize(l.k, l.k->size + r.k->size);
 376         return true;
 377 }
 378
 379 /* KEY_TYPE_reservation: */
 380
 381 int bch2_reservation_invalid(struct bch_fs *c, struct bkey_s_c k,
 382                              enum bkey_invalid_flags flags,
 383                              struct printbuf *err)
 384 {
 385         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 386         int ret = 0;
 387
 388         bkey_fsck_err_on(!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX, c, err,
 389                          reservation_key_nr_replicas_invalid,
 390                          "invalid nr_replicas (%u)", r.v->nr_replicas);
 391 fsck_err:
 392         return ret;
 393 }
 394
 395 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 396                               struct bkey_s_c k)
 397 {
 398         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 399
 400         prt_printf(out, "generation %u replicas %u",
 401                le32_to_cpu(r.v->generation),
 402                r.v->nr_replicas);
 403 }
 404
 405 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 406 {
 407         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 408         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 409
 410         if (l.v->generation != r.v->generation ||
 411             l.v->nr_replicas != r.v->nr_replicas)
 412                 return false;
 413
 414         bch2_key_resize(l.k, l.k->size + r.k->size);
 415         return true;
 416 }
 417
 418 /* Extent checksum entries: */
 419
 420 /* returns true if not equal */
 421 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 422                                          struct bch_extent_crc_unpacked r)
 423 {
 424         return (l.csum_type             != r.csum_type ||
 425                 l.compression_type      != r.compression_type ||
 426                 l.compressed_size       != r.compressed_size ||
 427                 l.uncompressed_size     != r.uncompressed_size ||
 428                 l.offset                != r.offset ||
 429                 l.live_size             != r.live_size ||
 430                 l.nonce                 != r.nonce ||
 431                 bch2_crc_cmp(l.csum, r.csum));
 432 }
 433
 434 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 435                                   struct bch_extent_crc_unpacked n)
 436 {
 437         return !crc_is_compressed(u) &&
 438                 u.csum_type &&
 439                 u.uncompressed_size > u.live_size &&
 440                 bch2_csum_type_is_encryption(u.csum_type) ==
 441                 bch2_csum_type_is_encryption(n.csum_type);
 442 }
 443
 444 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 445                                  struct bch_extent_crc_unpacked n)
 446 {
 447         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 448         struct bch_extent_crc_unpacked crc;
 449         const union bch_extent_entry *i;
 450
 451         if (!n.csum_type)
 452                 return false;
 453
 454         bkey_for_each_crc(k.k, ptrs, crc, i)
 455                 if (can_narrow_crc(crc, n))
 456                         return true;
 457
 458         return false;
 459 }
 460
 461 /*
 462  * We're writing another replica for this extent, so while we've got the data in
 463  * memory we'll be computing a new checksum for the currently live data.
 464  *
 465  * If there are other replicas we aren't moving, and they are checksummed but
 466  * not compressed, we can modify them to point to only the data that is
 467  * currently live (so that readers won't have to bounce) while we've got the
 468  * checksum we need:
 469  */
 470 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 471 {
 472         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 473         struct bch_extent_crc_unpacked u;
 474         struct extent_ptr_decoded p;
 475         union bch_extent_entry *i;
 476         bool ret = false;
 477
 478         /* Find a checksum entry that covers only live data: */
 479         if (!n.csum_type) {
 480                 bkey_for_each_crc(&k->k, ptrs, u, i)
 481                         if (!crc_is_compressed(u) &&
 482                             u.csum_type &&
 483                             u.live_size == u.uncompressed_size) {
 484                                 n = u;
 485                                 goto found;
 486                         }
 487                 return false;
 488         }
 489 found:
 490         BUG_ON(crc_is_compressed(n));
 491         BUG_ON(n.offset);
 492         BUG_ON(n.live_size != k->k.size);
 493
 494 restart_narrow_pointers:
 495         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 496
 497         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 498                 if (can_narrow_crc(p.crc, n)) {
 499                         bch2_bkey_drop_ptr_noerror(bkey_i_to_s(k), &i->ptr);
 500                         p.ptr.offset += p.crc.offset;
 501                         p.crc = n;
 502                         bch2_extent_ptr_decoded_append(k, &p);
 503                         ret = true;
 504                         goto restart_narrow_pointers;
 505                 }
 506
 507         return ret;
 508 }
 509
 510 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 511                                  struct bch_extent_crc_unpacked src,
 512                                  enum bch_extent_entry_type type)
 513 {
 514 #define set_common_fields(_dst, _src)                                   \
 515                 _dst.type               = 1 << type;                    \
 516                 _dst.csum_type          = _src.csum_type,               \
 517                 _dst.compression_type   = _src.compression_type,        \
 518                 _dst._compressed_size   = _src.compressed_size - 1,     \
 519                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 520                 _dst.offset             = _src.offset
 521
 522         switch (type) {
 523         case BCH_EXTENT_ENTRY_crc32:
 524                 set_common_fields(dst->crc32, src);
 525                 dst->crc32.csum         = (u32 __force) *((__le32 *) &src.csum.lo);
 526                 break;
 527         case BCH_EXTENT_ENTRY_crc64:
 528                 set_common_fields(dst->crc64, src);
 529                 dst->crc64.nonce        = src.nonce;
 530                 dst->crc64.csum_lo      = (u64 __force) src.csum.lo;
 531                 dst->crc64.csum_hi      = (u64 __force) *((__le16 *) &src.csum.hi);
 532                 break;
 533         case BCH_EXTENT_ENTRY_crc128:
 534                 set_common_fields(dst->crc128, src);
 535                 dst->crc128.nonce       = src.nonce;
 536                 dst->crc128.csum        = src.csum;
 537                 break;
 538         default:
 539                 BUG();
 540         }
 541 #undef set_common_fields
 542 }
 543
 544 void bch2_extent_crc_append(struct bkey_i *k,
 545                             struct bch_extent_crc_unpacked new)
 546 {
 547         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 548         union bch_extent_crc *crc = (void *) ptrs.end;
 549         enum bch_extent_entry_type type;
 550
 551         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 552             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 553             new.nonce                           <= CRC32_NONCE_MAX)
 554                 type = BCH_EXTENT_ENTRY_crc32;
 555         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 556                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 557                    new.nonce                    <= CRC64_NONCE_MAX)
 558                 type = BCH_EXTENT_ENTRY_crc64;
 559         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 560                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 561                    new.nonce                    <= CRC128_NONCE_MAX)
 562                 type = BCH_EXTENT_ENTRY_crc128;
 563         else
 564                 BUG();
 565
 566         bch2_extent_crc_pack(crc, new, type);
 567
 568         k->k.u64s += extent_entry_u64s(ptrs.end);
 569
 570         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 571 }
 572
 573 /* Generic code for keys with pointers: */
 574
 575 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 576 {
 577         return bch2_bkey_devs(k).nr;
 578 }
 579
 580 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 581 {
 582         return k.k->type == KEY_TYPE_reservation
 583                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 584                 : bch2_bkey_dirty_devs(k).nr;
 585 }
 586
 587 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 588 {
 589         unsigned ret = 0;
 590
 591         if (k.k->type == KEY_TYPE_reservation) {
 592                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 593         } else {
 594                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 595                 const union bch_extent_entry *entry;
 596                 struct extent_ptr_decoded p;
 597
 598                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 599                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 600         }
 601
 602         return ret;
 603 }
 604
 605 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 606 {
 607         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 608         const union bch_extent_entry *entry;
 609         struct extent_ptr_decoded p;
 610         unsigned ret = 0;
 611
 612         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 613                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 614                         ret += p.crc.compressed_size;
 615
 616         return ret;
 617 }
 618
 619 bool bch2_bkey_is_incompressible(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 bch_extent_crc_unpacked crc;
 624
 625         bkey_for_each_crc(k.k, ptrs, crc, entry)
 626                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 627                         return true;
 628         return false;
 629 }
 630
 631 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 632 {
 633         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 634         const union bch_extent_entry *entry;
 635         struct extent_ptr_decoded p = { 0 };
 636         unsigned replicas = 0;
 637
 638         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 639                 if (p.ptr.cached)
 640                         continue;
 641
 642                 if (p.has_ec)
 643                         replicas += p.ec.redundancy;
 644
 645                 replicas++;
 646
 647         }
 648
 649         return replicas;
 650 }
 651
 652 unsigned bch2_extent_ptr_desired_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 653 {
 654         struct bch_dev *ca;
 655
 656         if (p->ptr.cached)
 657                 return 0;
 658
 659         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 660
 661         return ca->mi.durability +
 662                 (p->has_ec
 663                  ? p->ec.redundancy
 664                  : 0);
 665 }
 666
 667 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 668 {
 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                 return 0;
 678
 679         return ca->mi.durability +
 680                 (p->has_ec
 681                  ? p->ec.redundancy
 682                  : 0);
 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 /*
 766  * Returns pointer to the next entry after the one being dropped:
 767  */
 768 union bch_extent_entry *bch2_bkey_drop_ptr_noerror(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_noerror(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 = bch2_bkey_has_device(k, dev);
 844
 845         if (ptr)
 846                 bch2_bkey_drop_ptr_noerror(k, ptr);
 847 }
 848
 849 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
 850 {
 851         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 852         const struct bch_extent_ptr *ptr;
 853
 854         bkey_for_each_ptr(ptrs, ptr)
 855                 if (ptr->dev == dev)
 856                         return ptr;
 857
 858         return NULL;
 859 }
 860
 861 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 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 (bch2_dev_in_target(c, ptr->dev, target) &&
 868                     (!ptr->cached ||
 869                      !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
 870                         return true;
 871
 872         return false;
 873 }
 874
 875 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 876                            struct bch_extent_ptr m, u64 offset)
 877 {
 878         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 879         const union bch_extent_entry *entry;
 880         struct extent_ptr_decoded p;
 881
 882         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 883                 if (p.ptr.dev   == m.dev &&
 884                     p.ptr.gen   == m.gen &&
 885                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 886                     (s64) m.offset  - offset)
 887                         return true;
 888
 889         return false;
 890 }
 891
 892 /*
 893  * Returns true if two extents refer to the same data:
 894  */
 895 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 896 {
 897         if (k1.k->type != k2.k->type)
 898                 return false;
 899
 900         if (bkey_extent_is_direct_data(k1.k)) {
 901                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 902                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 903                 const union bch_extent_entry *entry1, *entry2;
 904                 struct extent_ptr_decoded p1, p2;
 905
 906                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 907                         return false;
 908
 909                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 910                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 911                                 if (p1.ptr.dev          == p2.ptr.dev &&
 912                                     p1.ptr.gen          == p2.ptr.gen &&
 913                                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 914                                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 915                                         return true;
 916
 917                 return false;
 918         } else {
 919                 /* KEY_TYPE_deleted, etc. */
 920                 return true;
 921         }
 922 }
 923
 924 struct bch_extent_ptr *
 925 bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bkey_s k2)
 926 {
 927         struct bkey_ptrs ptrs2 = bch2_bkey_ptrs(k2);
 928         union bch_extent_entry *entry2;
 929         struct extent_ptr_decoded p2;
 930
 931         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 932                 if (p1.ptr.dev          == p2.ptr.dev &&
 933                     p1.ptr.gen          == p2.ptr.gen &&
 934                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 935                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 936                         return &entry2->ptr;
 937
 938         return NULL;
 939 }
 940
 941 void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
 942 {
 943         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 944         union bch_extent_entry *entry;
 945         union bch_extent_entry *ec = NULL;
 946
 947         bkey_extent_entry_for_each(ptrs, entry) {
 948                 if (&entry->ptr == ptr) {
 949                         ptr->cached = true;
 950                         if (ec)
 951                                 extent_entry_drop(k, ec);
 952                         return;
 953                 }
 954
 955                 if (extent_entry_is_stripe_ptr(entry))
 956                         ec = entry;
 957                 else if (extent_entry_is_ptr(entry))
 958                         ec = NULL;
 959         }
 960
 961         BUG();
 962 }
 963
 964 /*
 965  * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 966  *
 967  * Returns true if @k should be dropped entirely
 968  *
 969  * For existing keys, only called when btree nodes are being rewritten, not when
 970  * they're merely being compacted/resorted in memory.
 971  */
 972 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 973 {
 974         struct bch_extent_ptr *ptr;
 975
 976         bch2_bkey_drop_ptrs(k, ptr,
 977                 ptr->cached &&
 978                 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
 979
 980         return bkey_deleted(k.k);
 981 }
 982
 983 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 984                             struct bkey_s_c k)
 985 {
 986         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 987         const union bch_extent_entry *entry;
 988         bool first = true;
 989
 990         if (c)
 991                 prt_printf(out, "durability: %u ", bch2_bkey_durability_safe(c, k));
 992
 993         bkey_extent_entry_for_each(ptrs, entry) {
 994                 if (!first)
 995                         prt_printf(out, " ");
 996
 997                 switch (__extent_entry_type(entry)) {
 998                 case BCH_EXTENT_ENTRY_ptr: {
 999                         const struct bch_extent_ptr *ptr = entry_to_ptr(entry);
1000                         struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
1001                                 ? bch_dev_bkey_exists(c, ptr->dev)
1002                                 : NULL;
1003
1004                         if (!ca) {
1005                                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1006                                        (u64) ptr->offset, ptr->gen,
1007                                        ptr->cached ? " cached" : "");
1008                         } else {
1009                                 u32 offset;
1010                                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1011
1012                                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1013                                            ptr->dev, b, offset, ptr->gen);
1014                                 if (ptr->cached)
1015                                         prt_str(out, " cached");
1016                                 if (ptr->unwritten)
1017                                         prt_str(out, " unwritten");
1018                                 if (ca && ptr_stale(ca, ptr))
1019                                         prt_printf(out, " stale");
1020                         }
1021                         break;
1022                 }
1023                 case BCH_EXTENT_ENTRY_crc32:
1024                 case BCH_EXTENT_ENTRY_crc64:
1025                 case BCH_EXTENT_ENTRY_crc128: {
1026                         struct bch_extent_crc_unpacked crc =
1027                                 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                 }
1037                 case BCH_EXTENT_ENTRY_stripe_ptr: {
1038                         const struct bch_extent_stripe_ptr *ec = &entry->stripe_ptr;
1039
1040                         prt_printf(out, "ec: idx %llu block %u",
1041                                (u64) ec->idx, ec->block);
1042                         break;
1043                 }
1044                 case BCH_EXTENT_ENTRY_rebalance: {
1045                         const struct bch_extent_rebalance *r = &entry->rebalance;
1046
1047                         prt_str(out, "rebalance: target ");
1048                         if (c)
1049                                 bch2_target_to_text(out, c, r->target);
1050                         else
1051                                 prt_printf(out, "%u", r->target);
1052                         prt_str(out, " compression ");
1053                         bch2_compression_opt_to_text(out, r->compression);
1054                         break;
1055                 }
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(struct bch_fs *c,
1066                               struct bkey_s_c k,
1067                               enum bkey_invalid_flags flags,
1068                               const struct bch_extent_ptr *ptr,
1069                               unsigned size_ondisk,
1070                               bool metadata,
1071                               struct printbuf *err)
1072 {
1073         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1074         const struct bch_extent_ptr *ptr2;
1075         u64 bucket;
1076         u32 bucket_offset;
1077         struct bch_dev *ca;
1078         int ret = 0;
1079
1080         if (!bch2_dev_exists2(c, ptr->dev)) {
1081                 /*
1082                  * If we're in the write path this key might have already been
1083                  * overwritten, and we could be seeing a device that doesn't
1084                  * exist anymore due to racing with device removal:
1085                  */
1086                 if (flags & BKEY_INVALID_WRITE)
1087                         return 0;
1088
1089                 bkey_fsck_err(c, err, ptr_to_invalid_device,
1090                            "pointer to invalid device (%u)", ptr->dev);
1091         }
1092
1093         ca = bch_dev_bkey_exists(c, ptr->dev);
1094         bkey_for_each_ptr(ptrs, ptr2)
1095                 bkey_fsck_err_on(ptr != ptr2 && ptr->dev == ptr2->dev, c, err,
1096                                  ptr_to_duplicate_device,
1097                                  "multiple pointers to same device (%u)", ptr->dev);
1098
1099         bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1100
1101         bkey_fsck_err_on(bucket >= ca->mi.nbuckets, c, err,
1102                          ptr_after_last_bucket,
1103                          "pointer past last bucket (%llu > %llu)", bucket, ca->mi.nbuckets);
1104         bkey_fsck_err_on(ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket), c, err,
1105                          ptr_before_first_bucket,
1106                          "pointer before first bucket (%llu < %u)", bucket, ca->mi.first_bucket);
1107         bkey_fsck_err_on(bucket_offset + size_ondisk > ca->mi.bucket_size, c, err,
1108                          ptr_spans_multiple_buckets,
1109                          "pointer spans multiple buckets (%u + %u > %u)",
1110                        bucket_offset, size_ondisk, ca->mi.bucket_size);
1111 fsck_err:
1112         return ret;
1113 }
1114
1115 int bch2_bkey_ptrs_invalid(struct bch_fs *c, struct bkey_s_c k,
1116                            enum bkey_invalid_flags flags,
1117                            struct printbuf *err)
1118 {
1119         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1120         const union bch_extent_entry *entry;
1121         struct bch_extent_crc_unpacked crc;
1122         unsigned size_ondisk = k.k->size;
1123         unsigned nonce = UINT_MAX;
1124         unsigned nr_ptrs = 0;
1125         bool have_written = false, have_unwritten = false, have_ec = false, crc_since_last_ptr = false;
1126         int ret = 0;
1127
1128         if (bkey_is_btree_ptr(k.k))
1129                 size_ondisk = btree_sectors(c);
1130
1131         bkey_extent_entry_for_each(ptrs, entry) {
1132                 bkey_fsck_err_on(__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX, c, err,
1133                         extent_ptrs_invalid_entry,
1134                         "invalid extent entry type (got %u, max %u)",
1135                         __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1136
1137                 bkey_fsck_err_on(bkey_is_btree_ptr(k.k) &&
1138                                  !extent_entry_is_ptr(entry), c, err,
1139                                  btree_ptr_has_non_ptr,
1140                                  "has non ptr field");
1141
1142                 switch (extent_entry_type(entry)) {
1143                 case BCH_EXTENT_ENTRY_ptr:
1144                         ret = extent_ptr_invalid(c, k, flags, &entry->ptr,
1145                                                  size_ondisk, false, err);
1146                         if (ret)
1147                                 return ret;
1148
1149                         bkey_fsck_err_on(entry->ptr.cached && have_ec, c, err,
1150                                          ptr_cached_and_erasure_coded,
1151                                          "cached, erasure coded ptr");
1152
1153                         if (!entry->ptr.unwritten)
1154                                 have_written = true;
1155                         else
1156                                 have_unwritten = true;
1157
1158                         have_ec = false;
1159                         crc_since_last_ptr = false;
1160                         nr_ptrs++;
1161                         break;
1162                 case BCH_EXTENT_ENTRY_crc32:
1163                 case BCH_EXTENT_ENTRY_crc64:
1164                 case BCH_EXTENT_ENTRY_crc128:
1165                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1166
1167                         bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size, c, err,
1168                                          ptr_crc_uncompressed_size_too_small,
1169                                          "checksum offset + key size > uncompressed size");
1170                         bkey_fsck_err_on(!bch2_checksum_type_valid(c, crc.csum_type), c, err,
1171                                          ptr_crc_csum_type_unknown,
1172                                          "invalid checksum type");
1173                         bkey_fsck_err_on(crc.compression_type >= BCH_COMPRESSION_TYPE_NR, c, err,
1174                                          ptr_crc_compression_type_unknown,
1175                                          "invalid compression type");
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                                         bkey_fsck_err(c, err, ptr_crc_nonce_mismatch,
1182                                                       "incorrect nonce");
1183                         }
1184
1185                         bkey_fsck_err_on(crc_since_last_ptr, c, err,
1186                                          ptr_crc_redundant,
1187                                          "redundant crc entry");
1188                         crc_since_last_ptr = true;
1189
1190                         bkey_fsck_err_on(crc_is_encoded(crc) &&
1191                                          (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
1192                                          (flags & (BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT)), c, err,
1193                                          ptr_crc_uncompressed_size_too_big,
1194                                          "too large encoded extent");
1195
1196                         size_ondisk = crc.compressed_size;
1197                         break;
1198                 case BCH_EXTENT_ENTRY_stripe_ptr:
1199                         bkey_fsck_err_on(have_ec, c, err,
1200                                          ptr_stripe_redundant,
1201                                          "redundant stripe entry");
1202                         have_ec = true;
1203                         break;
1204                 case BCH_EXTENT_ENTRY_rebalance: {
1205                         const struct bch_extent_rebalance *r = &entry->rebalance;
1206
1207                         if (!bch2_compression_opt_valid(r->compression)) {
1208                                 struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
1209                                 prt_printf(err, "invalid compression opt %u:%u",
1210                                            opt.type, opt.level);
1211                                 return -BCH_ERR_invalid_bkey;
1212                         }
1213                         break;
1214                 }
1215                 }
1216         }
1217
1218         bkey_fsck_err_on(!nr_ptrs, c, err,
1219                          extent_ptrs_no_ptrs,
1220                          "no ptrs");
1221         bkey_fsck_err_on(nr_ptrs > BCH_BKEY_PTRS_MAX, c, err,
1222                          extent_ptrs_too_many_ptrs,
1223                          "too many ptrs: %u > %u", nr_ptrs, BCH_BKEY_PTRS_MAX);
1224         bkey_fsck_err_on(have_written && have_unwritten, c, err,
1225                          extent_ptrs_written_and_unwritten,
1226                          "extent with unwritten and written ptrs");
1227         bkey_fsck_err_on(k.k->type != KEY_TYPE_extent && have_unwritten, c, err,
1228                          extent_ptrs_unwritten,
1229                          "has unwritten ptrs");
1230         bkey_fsck_err_on(crc_since_last_ptr, c, err,
1231                          extent_ptrs_redundant_crc,
1232                          "redundant crc entry");
1233         bkey_fsck_err_on(have_ec, c, err,
1234                          extent_ptrs_redundant_stripe,
1235                          "redundant stripe entry");
1236 fsck_err:
1237         return ret;
1238 }
1239
1240 void bch2_ptr_swab(struct bkey_s k)
1241 {
1242         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1243         union bch_extent_entry *entry;
1244         u64 *d;
1245
1246         for (d =  (u64 *) ptrs.start;
1247              d != (u64 *) ptrs.end;
1248              d++)
1249                 *d = swab64(*d);
1250
1251         for (entry = ptrs.start;
1252              entry < ptrs.end;
1253              entry = extent_entry_next(entry)) {
1254                 switch (extent_entry_type(entry)) {
1255                 case BCH_EXTENT_ENTRY_ptr:
1256                         break;
1257                 case BCH_EXTENT_ENTRY_crc32:
1258                         entry->crc32.csum = swab32(entry->crc32.csum);
1259                         break;
1260                 case BCH_EXTENT_ENTRY_crc64:
1261                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1262                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1263                         break;
1264                 case BCH_EXTENT_ENTRY_crc128:
1265                         entry->crc128.csum.hi = (__force __le64)
1266                                 swab64((__force u64) entry->crc128.csum.hi);
1267                         entry->crc128.csum.lo = (__force __le64)
1268                                 swab64((__force u64) entry->crc128.csum.lo);
1269                         break;
1270                 case BCH_EXTENT_ENTRY_stripe_ptr:
1271                         break;
1272                 case BCH_EXTENT_ENTRY_rebalance:
1273                         break;
1274                 }
1275         }
1276 }
1277
1278 const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c k)
1279 {
1280         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1281         const union bch_extent_entry *entry;
1282
1283         bkey_extent_entry_for_each(ptrs, entry)
1284                 if (__extent_entry_type(entry) == BCH_EXTENT_ENTRY_rebalance)
1285                         return &entry->rebalance;
1286
1287         return NULL;
1288 }
1289
1290 unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *c, struct bkey_s_c k,
1291                                        unsigned target, unsigned compression)
1292 {
1293         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1294         unsigned rewrite_ptrs = 0;
1295
1296         if (compression) {
1297                 unsigned compression_type = bch2_compression_opt_to_type(compression);
1298                 const union bch_extent_entry *entry;
1299                 struct extent_ptr_decoded p;
1300                 unsigned i = 0;
1301
1302                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1303                         if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible) {
1304                                 rewrite_ptrs = 0;
1305                                 goto incompressible;
1306                         }
1307
1308                         if (!p.ptr.cached && p.crc.compression_type != compression_type)
1309                                 rewrite_ptrs |= 1U << i;
1310                         i++;
1311                 }
1312         }
1313 incompressible:
1314         if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
1315                 const struct bch_extent_ptr *ptr;
1316                 unsigned i = 0;
1317
1318                 bkey_for_each_ptr(ptrs, ptr) {
1319                         if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, target))
1320                                 rewrite_ptrs |= 1U << i;
1321                         i++;
1322                 }
1323         }
1324
1325         return rewrite_ptrs;
1326 }
1327
1328 bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k)
1329 {
1330         const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
1331
1332         /*
1333          * If it's an indirect extent, we don't delete the rebalance entry when
1334          * done so that we know what options were applied - check if it still
1335          * needs work done:
1336          */
1337         if (r &&
1338             k.k->type == KEY_TYPE_reflink_v &&
1339             !bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression))
1340                 r = NULL;
1341
1342         return r != NULL;
1343 }
1344
1345 int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
1346                                   unsigned target, unsigned compression)
1347 {
1348         struct bkey_s k = bkey_i_to_s(_k);
1349         struct bch_extent_rebalance *r;
1350         bool needs_rebalance;
1351
1352         if (!bkey_extent_is_direct_data(k.k))
1353                 return 0;
1354
1355         /* get existing rebalance entry: */
1356         r = (struct bch_extent_rebalance *) bch2_bkey_rebalance_opts(k.s_c);
1357         if (r) {
1358                 if (k.k->type == KEY_TYPE_reflink_v) {
1359                         /*
1360                          * indirect extents: existing options take precedence,
1361                          * so that we don't move extents back and forth if
1362                          * they're referenced by different inodes with different
1363                          * options:
1364                          */
1365                         if (r->target)
1366                                 target = r->target;
1367                         if (r->compression)
1368                                 compression = r->compression;
1369                 }
1370
1371                 r->target       = target;
1372                 r->compression  = compression;
1373         }
1374
1375         needs_rebalance = bch2_bkey_ptrs_need_rebalance(c, k.s_c, target, compression);
1376
1377         if (needs_rebalance && !r) {
1378                 union bch_extent_entry *new = bkey_val_end(k);
1379
1380                 new->rebalance.type             = 1U << BCH_EXTENT_ENTRY_rebalance;
1381                 new->rebalance.compression      = compression;
1382                 new->rebalance.target           = target;
1383                 new->rebalance.unused           = 0;
1384                 k.k->u64s += extent_entry_u64s(new);
1385         } else if (!needs_rebalance && r && k.k->type != KEY_TYPE_reflink_v) {
1386                 /*
1387                  * For indirect extents, don't delete the rebalance entry when
1388                  * we're finished so that we know we specifically moved it or
1389                  * compressed it to its current location/compression type
1390                  */
1391                 extent_entry_drop(k, (union bch_extent_entry *) r);
1392         }
1393
1394         return 0;
1395 }
1396
1397 /* Generic extent code: */
1398
1399 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1400 {
1401         unsigned new_val_u64s = bkey_val_u64s(k.k);
1402         int val_u64s_delta;
1403         u64 sub;
1404
1405         if (bkey_le(where, bkey_start_pos(k.k)))
1406                 return 0;
1407
1408         EBUG_ON(bkey_gt(where, k.k->p));
1409
1410         sub = where.offset - bkey_start_offset(k.k);
1411
1412         k.k->size -= sub;
1413
1414         if (!k.k->size) {
1415                 k.k->type = KEY_TYPE_deleted;
1416                 new_val_u64s = 0;
1417         }
1418
1419         switch (k.k->type) {
1420         case KEY_TYPE_extent:
1421         case KEY_TYPE_reflink_v: {
1422                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1423                 union bch_extent_entry *entry;
1424                 bool seen_crc = false;
1425
1426                 bkey_extent_entry_for_each(ptrs, entry) {
1427                         switch (extent_entry_type(entry)) {
1428                         case BCH_EXTENT_ENTRY_ptr:
1429                                 if (!seen_crc)
1430                                         entry->ptr.offset += sub;
1431                                 break;
1432                         case BCH_EXTENT_ENTRY_crc32:
1433                                 entry->crc32.offset += sub;
1434                                 break;
1435                         case BCH_EXTENT_ENTRY_crc64:
1436                                 entry->crc64.offset += sub;
1437                                 break;
1438                         case BCH_EXTENT_ENTRY_crc128:
1439                                 entry->crc128.offset += sub;
1440                                 break;
1441                         case BCH_EXTENT_ENTRY_stripe_ptr:
1442                                 break;
1443                         case BCH_EXTENT_ENTRY_rebalance:
1444                                 break;
1445                         }
1446
1447                         if (extent_entry_is_crc(entry))
1448                                 seen_crc = true;
1449                 }
1450
1451                 break;
1452         }
1453         case KEY_TYPE_reflink_p: {
1454                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1455
1456                 le64_add_cpu(&p.v->idx, sub);
1457                 break;
1458         }
1459         case KEY_TYPE_inline_data:
1460         case KEY_TYPE_indirect_inline_data: {
1461                 void *p = bkey_inline_data_p(k);
1462                 unsigned bytes = bkey_inline_data_bytes(k.k);
1463
1464                 sub = min_t(u64, sub << 9, bytes);
1465
1466                 memmove(p, p + sub, bytes - sub);
1467
1468                 new_val_u64s -= sub >> 3;
1469                 break;
1470         }
1471         }
1472
1473         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1474         BUG_ON(val_u64s_delta < 0);
1475
1476         set_bkey_val_u64s(k.k, new_val_u64s);
1477         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1478         return -val_u64s_delta;
1479 }
1480
1481 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1482 {
1483         unsigned new_val_u64s = bkey_val_u64s(k.k);
1484         int val_u64s_delta;
1485         u64 len = 0;
1486
1487         if (bkey_ge(where, k.k->p))
1488                 return 0;
1489
1490         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1491
1492         len = where.offset - bkey_start_offset(k.k);
1493
1494         k.k->p.offset = where.offset;
1495         k.k->size = len;
1496
1497         if (!len) {
1498                 k.k->type = KEY_TYPE_deleted;
1499                 new_val_u64s = 0;
1500         }
1501
1502         switch (k.k->type) {
1503         case KEY_TYPE_inline_data:
1504         case KEY_TYPE_indirect_inline_data:
1505                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1506                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1507                 break;
1508         }
1509
1510         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1511         BUG_ON(val_u64s_delta < 0);
1512
1513         set_bkey_val_u64s(k.k, new_val_u64s);
1514         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1515         return -val_u64s_delta;
1516 }