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(const struct bch_fs *c, struct bkey_s_c k,
 167                            enum bkey_invalid_flags flags,
 168                            struct printbuf *err)
 169 {
 170         if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX) {
 171                 prt_printf(err, "value too big (%zu > %u)",
 172                        bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 173                 return -BCH_ERR_invalid_bkey;
 174         }
 175
 176         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 177 }
 178
 179 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 180                             struct bkey_s_c k)
 181 {
 182         bch2_bkey_ptrs_to_text(out, c, k);
 183 }
 184
 185 int bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 186                               enum bkey_invalid_flags flags,
 187                               struct printbuf *err)
 188 {
 189         if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) {
 190                 prt_printf(err, "value too big (%zu > %zu)",
 191                        bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 192                 return -BCH_ERR_invalid_bkey;
 193         }
 194
 195         return bch2_bkey_ptrs_invalid(c, k, flags, err);
 196 }
 197
 198 void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
 199                                struct bkey_s_c k)
 200 {
 201         struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 202
 203         prt_printf(out, "seq %llx written %u min_key %s",
 204                le64_to_cpu(bp.v->seq),
 205                le16_to_cpu(bp.v->sectors_written),
 206                BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
 207
 208         bch2_bpos_to_text(out, bp.v->min_key);
 209         prt_printf(out, " ");
 210         bch2_bkey_ptrs_to_text(out, c, k);
 211 }
 212
 213 void bch2_btree_ptr_v2_compat(enum btree_id btree_id, unsigned version,
 214                               unsigned big_endian, int write,
 215                               struct bkey_s k)
 216 {
 217         struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(k);
 218
 219         compat_bpos(0, btree_id, version, big_endian, write, &bp.v->min_key);
 220
 221         if (version < bcachefs_metadata_version_inode_btree_change &&
 222             btree_id_is_extents(btree_id) &&
 223             !bkey_eq(bp.v->min_key, POS_MIN))
 224                 bp.v->min_key = write
 225                         ? bpos_nosnap_predecessor(bp.v->min_key)
 226                         : bpos_nosnap_successor(bp.v->min_key);
 227 }
 228
 229 /* KEY_TYPE_extent: */
 230
 231 bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 232 {
 233         struct bkey_ptrs   l_ptrs = bch2_bkey_ptrs(l);
 234         struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
 235         union bch_extent_entry *en_l;
 236         const union bch_extent_entry *en_r;
 237         struct extent_ptr_decoded lp, rp;
 238         bool use_right_ptr;
 239         struct bch_dev *ca;
 240
 241         en_l = l_ptrs.start;
 242         en_r = r_ptrs.start;
 243         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 244                 if (extent_entry_type(en_l) != extent_entry_type(en_r))
 245                         return false;
 246
 247                 en_l = extent_entry_next(en_l);
 248                 en_r = extent_entry_next(en_r);
 249         }
 250
 251         if (en_l < l_ptrs.end || en_r < r_ptrs.end)
 252                 return false;
 253
 254         en_l = l_ptrs.start;
 255         en_r = r_ptrs.start;
 256         lp.crc = bch2_extent_crc_unpack(l.k, NULL);
 257         rp.crc = bch2_extent_crc_unpack(r.k, NULL);
 258
 259         while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
 260                __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
 261                 if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
 262                     rp.ptr.offset + rp.crc.offset ||
 263                     lp.ptr.dev                  != rp.ptr.dev ||
 264                     lp.ptr.gen                  != rp.ptr.gen ||
 265                     lp.ptr.unwritten            != rp.ptr.unwritten ||
 266                     lp.has_ec                   != rp.has_ec)
 267                         return false;
 268
 269                 /* Extents may not straddle buckets: */
 270                 ca = bch_dev_bkey_exists(c, lp.ptr.dev);
 271                 if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
 272                         return false;
 273
 274                 if (lp.has_ec                   != rp.has_ec ||
 275                     (lp.has_ec &&
 276                      (lp.ec.block               != rp.ec.block ||
 277                       lp.ec.redundancy          != rp.ec.redundancy ||
 278                       lp.ec.idx                 != rp.ec.idx)))
 279                         return false;
 280
 281                 if (lp.crc.compression_type     != rp.crc.compression_type ||
 282                     lp.crc.nonce                != rp.crc.nonce)
 283                         return false;
 284
 285                 if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
 286                     lp.crc.uncompressed_size) {
 287                         /* can use left extent's crc entry */
 288                 } else if (lp.crc.live_size <= rp.crc.offset) {
 289                         /* can use right extent's crc entry */
 290                 } else {
 291                         /* check if checksums can be merged: */
 292                         if (lp.crc.csum_type            != rp.crc.csum_type ||
 293                             lp.crc.nonce                != rp.crc.nonce ||
 294                             crc_is_compressed(lp.crc) ||
 295                             !bch2_checksum_mergeable(lp.crc.csum_type))
 296                                 return false;
 297
 298                         if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
 299                             rp.crc.offset)
 300                                 return false;
 301
 302                         if (lp.crc.csum_type &&
 303                             lp.crc.uncompressed_size +
 304                             rp.crc.uncompressed_size > (c->opts.encoded_extent_max >> 9))
 305                                 return false;
 306                 }
 307
 308                 en_l = extent_entry_next(en_l);
 309                 en_r = extent_entry_next(en_r);
 310         }
 311
 312         en_l = l_ptrs.start;
 313         en_r = r_ptrs.start;
 314         while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
 315                 if (extent_entry_is_crc(en_l)) {
 316                         struct bch_extent_crc_unpacked crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 317                         struct bch_extent_crc_unpacked crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 318
 319                         if (crc_l.uncompressed_size + crc_r.uncompressed_size >
 320                             bch2_crc_field_size_max[extent_entry_type(en_l)])
 321                                 return false;
 322                 }
 323
 324                 en_l = extent_entry_next(en_l);
 325                 en_r = extent_entry_next(en_r);
 326         }
 327
 328         use_right_ptr = false;
 329         en_l = l_ptrs.start;
 330         en_r = r_ptrs.start;
 331         while (en_l < l_ptrs.end) {
 332                 if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
 333                     use_right_ptr)
 334                         en_l->ptr = en_r->ptr;
 335
 336                 if (extent_entry_is_crc(en_l)) {
 337                         struct bch_extent_crc_unpacked crc_l =
 338                                 bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
 339                         struct bch_extent_crc_unpacked crc_r =
 340                                 bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
 341
 342                         use_right_ptr = false;
 343
 344                         if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
 345                             crc_l.uncompressed_size) {
 346                                 /* can use left extent's crc entry */
 347                         } else if (crc_l.live_size <= crc_r.offset) {
 348                                 /* can use right extent's crc entry */
 349                                 crc_r.offset -= crc_l.live_size;
 350                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
 351                                                      extent_entry_type(en_l));
 352                                 use_right_ptr = true;
 353                         } else {
 354                                 crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
 355                                                                  crc_l.csum,
 356                                                                  crc_r.csum,
 357                                                                  crc_r.uncompressed_size << 9);
 358
 359                                 crc_l.uncompressed_size += crc_r.uncompressed_size;
 360                                 crc_l.compressed_size   += crc_r.compressed_size;
 361                                 bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
 362                                                      extent_entry_type(en_l));
 363                         }
 364                 }
 365
 366                 en_l = extent_entry_next(en_l);
 367                 en_r = extent_entry_next(en_r);
 368         }
 369
 370         bch2_key_resize(l.k, l.k->size + r.k->size);
 371         return true;
 372 }
 373
 374 /* KEY_TYPE_reservation: */
 375
 376 int bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k,
 377                              enum bkey_invalid_flags flags,
 378                              struct printbuf *err)
 379 {
 380         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 381
 382         if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX) {
 383                 prt_printf(err, "invalid nr_replicas (%u)",
 384                        r.v->nr_replicas);
 385                 return -BCH_ERR_invalid_bkey;
 386         }
 387
 388         return 0;
 389 }
 390
 391 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
 392                               struct bkey_s_c k)
 393 {
 394         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 395
 396         prt_printf(out, "generation %u replicas %u",
 397                le32_to_cpu(r.v->generation),
 398                r.v->nr_replicas);
 399 }
 400
 401 bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
 402 {
 403         struct bkey_s_reservation l = bkey_s_to_reservation(_l);
 404         struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
 405
 406         if (l.v->generation != r.v->generation ||
 407             l.v->nr_replicas != r.v->nr_replicas)
 408                 return false;
 409
 410         bch2_key_resize(l.k, l.k->size + r.k->size);
 411         return true;
 412 }
 413
 414 /* Extent checksum entries: */
 415
 416 /* returns true if not equal */
 417 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
 418                                          struct bch_extent_crc_unpacked r)
 419 {
 420         return (l.csum_type             != r.csum_type ||
 421                 l.compression_type      != r.compression_type ||
 422                 l.compressed_size       != r.compressed_size ||
 423                 l.uncompressed_size     != r.uncompressed_size ||
 424                 l.offset                != r.offset ||
 425                 l.live_size             != r.live_size ||
 426                 l.nonce                 != r.nonce ||
 427                 bch2_crc_cmp(l.csum, r.csum));
 428 }
 429
 430 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
 431                                   struct bch_extent_crc_unpacked n)
 432 {
 433         return !crc_is_compressed(u) &&
 434                 u.csum_type &&
 435                 u.uncompressed_size > u.live_size &&
 436                 bch2_csum_type_is_encryption(u.csum_type) ==
 437                 bch2_csum_type_is_encryption(n.csum_type);
 438 }
 439
 440 bool bch2_can_narrow_extent_crcs(struct bkey_s_c k,
 441                                  struct bch_extent_crc_unpacked n)
 442 {
 443         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 444         struct bch_extent_crc_unpacked crc;
 445         const union bch_extent_entry *i;
 446
 447         if (!n.csum_type)
 448                 return false;
 449
 450         bkey_for_each_crc(k.k, ptrs, crc, i)
 451                 if (can_narrow_crc(crc, n))
 452                         return true;
 453
 454         return false;
 455 }
 456
 457 /*
 458  * We're writing another replica for this extent, so while we've got the data in
 459  * memory we'll be computing a new checksum for the currently live data.
 460  *
 461  * If there are other replicas we aren't moving, and they are checksummed but
 462  * not compressed, we can modify them to point to only the data that is
 463  * currently live (so that readers won't have to bounce) while we've got the
 464  * checksum we need:
 465  */
 466 bool bch2_bkey_narrow_crcs(struct bkey_i *k, struct bch_extent_crc_unpacked n)
 467 {
 468         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 469         struct bch_extent_crc_unpacked u;
 470         struct extent_ptr_decoded p;
 471         union bch_extent_entry *i;
 472         bool ret = false;
 473
 474         /* Find a checksum entry that covers only live data: */
 475         if (!n.csum_type) {
 476                 bkey_for_each_crc(&k->k, ptrs, u, i)
 477                         if (!crc_is_compressed(u) &&
 478                             u.csum_type &&
 479                             u.live_size == u.uncompressed_size) {
 480                                 n = u;
 481                                 goto found;
 482                         }
 483                 return false;
 484         }
 485 found:
 486         BUG_ON(crc_is_compressed(n));
 487         BUG_ON(n.offset);
 488         BUG_ON(n.live_size != k->k.size);
 489
 490 restart_narrow_pointers:
 491         ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 492
 493         bkey_for_each_ptr_decode(&k->k, ptrs, p, i)
 494                 if (can_narrow_crc(p.crc, n)) {
 495                         bch2_bkey_drop_ptr_noerror(bkey_i_to_s(k), &i->ptr);
 496                         p.ptr.offset += p.crc.offset;
 497                         p.crc = n;
 498                         bch2_extent_ptr_decoded_append(k, &p);
 499                         ret = true;
 500                         goto restart_narrow_pointers;
 501                 }
 502
 503         return ret;
 504 }
 505
 506 static void bch2_extent_crc_pack(union bch_extent_crc *dst,
 507                                  struct bch_extent_crc_unpacked src,
 508                                  enum bch_extent_entry_type type)
 509 {
 510 #define set_common_fields(_dst, _src)                                   \
 511                 _dst.type               = 1 << type;                    \
 512                 _dst.csum_type          = _src.csum_type,               \
 513                 _dst.compression_type   = _src.compression_type,        \
 514                 _dst._compressed_size   = _src.compressed_size - 1,     \
 515                 _dst._uncompressed_size = _src.uncompressed_size - 1,   \
 516                 _dst.offset             = _src.offset
 517
 518         switch (type) {
 519         case BCH_EXTENT_ENTRY_crc32:
 520                 set_common_fields(dst->crc32, src);
 521                 dst->crc32.csum         = (u32 __force) *((__le32 *) &src.csum.lo);
 522                 break;
 523         case BCH_EXTENT_ENTRY_crc64:
 524                 set_common_fields(dst->crc64, src);
 525                 dst->crc64.nonce        = src.nonce;
 526                 dst->crc64.csum_lo      = (u64 __force) src.csum.lo;
 527                 dst->crc64.csum_hi      = (u64 __force) *((__le16 *) &src.csum.hi);
 528                 break;
 529         case BCH_EXTENT_ENTRY_crc128:
 530                 set_common_fields(dst->crc128, src);
 531                 dst->crc128.nonce       = src.nonce;
 532                 dst->crc128.csum        = src.csum;
 533                 break;
 534         default:
 535                 BUG();
 536         }
 537 #undef set_common_fields
 538 }
 539
 540 void bch2_extent_crc_append(struct bkey_i *k,
 541                             struct bch_extent_crc_unpacked new)
 542 {
 543         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 544         union bch_extent_crc *crc = (void *) ptrs.end;
 545         enum bch_extent_entry_type type;
 546
 547         if (bch_crc_bytes[new.csum_type]        <= 4 &&
 548             new.uncompressed_size               <= CRC32_SIZE_MAX &&
 549             new.nonce                           <= CRC32_NONCE_MAX)
 550                 type = BCH_EXTENT_ENTRY_crc32;
 551         else if (bch_crc_bytes[new.csum_type]   <= 10 &&
 552                    new.uncompressed_size        <= CRC64_SIZE_MAX &&
 553                    new.nonce                    <= CRC64_NONCE_MAX)
 554                 type = BCH_EXTENT_ENTRY_crc64;
 555         else if (bch_crc_bytes[new.csum_type]   <= 16 &&
 556                    new.uncompressed_size        <= CRC128_SIZE_MAX &&
 557                    new.nonce                    <= CRC128_NONCE_MAX)
 558                 type = BCH_EXTENT_ENTRY_crc128;
 559         else
 560                 BUG();
 561
 562         bch2_extent_crc_pack(crc, new, type);
 563
 564         k->k.u64s += extent_entry_u64s(ptrs.end);
 565
 566         EBUG_ON(bkey_val_u64s(&k->k) > BKEY_EXTENT_VAL_U64s_MAX);
 567 }
 568
 569 /* Generic code for keys with pointers: */
 570
 571 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
 572 {
 573         return bch2_bkey_devs(k).nr;
 574 }
 575
 576 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
 577 {
 578         return k.k->type == KEY_TYPE_reservation
 579                 ? bkey_s_c_to_reservation(k).v->nr_replicas
 580                 : bch2_bkey_dirty_devs(k).nr;
 581 }
 582
 583 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 584 {
 585         unsigned ret = 0;
 586
 587         if (k.k->type == KEY_TYPE_reservation) {
 588                 ret = bkey_s_c_to_reservation(k).v->nr_replicas;
 589         } else {
 590                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 591                 const union bch_extent_entry *entry;
 592                 struct extent_ptr_decoded p;
 593
 594                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 595                         ret += !p.ptr.cached && !crc_is_compressed(p.crc);
 596         }
 597
 598         return ret;
 599 }
 600
 601 unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 602 {
 603         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 604         const union bch_extent_entry *entry;
 605         struct extent_ptr_decoded p;
 606         unsigned ret = 0;
 607
 608         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 609                 if (!p.ptr.cached && crc_is_compressed(p.crc))
 610                         ret += p.crc.compressed_size;
 611
 612         return ret;
 613 }
 614
 615 bool bch2_bkey_is_incompressible(struct bkey_s_c k)
 616 {
 617         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 618         const union bch_extent_entry *entry;
 619         struct bch_extent_crc_unpacked crc;
 620
 621         bkey_for_each_crc(k.k, ptrs, crc, entry)
 622                 if (crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 623                         return true;
 624         return false;
 625 }
 626
 627 unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 628 {
 629         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 630         const union bch_extent_entry *entry;
 631         struct extent_ptr_decoded p = { 0 };
 632         unsigned replicas = 0;
 633
 634         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 635                 if (p.ptr.cached)
 636                         continue;
 637
 638                 if (p.has_ec)
 639                         replicas += p.ec.redundancy;
 640
 641                 replicas++;
 642
 643         }
 644
 645         return replicas;
 646 }
 647
 648 unsigned bch2_extent_ptr_desired_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 649 {
 650         struct bch_dev *ca;
 651
 652         if (p->ptr.cached)
 653                 return 0;
 654
 655         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 656
 657         return ca->mi.durability +
 658                 (p->has_ec
 659                  ? p->ec.redundancy
 660                  : 0);
 661 }
 662
 663 unsigned bch2_extent_ptr_durability(struct bch_fs *c, struct extent_ptr_decoded *p)
 664 {
 665         struct bch_dev *ca;
 666
 667         if (p->ptr.cached)
 668                 return 0;
 669
 670         ca = bch_dev_bkey_exists(c, p->ptr.dev);
 671
 672         if (ca->mi.state == BCH_MEMBER_STATE_failed)
 673                 return 0;
 674
 675         return ca->mi.durability +
 676                 (p->has_ec
 677                  ? p->ec.redundancy
 678                  : 0);
 679 }
 680
 681 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
 682 {
 683         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 684         const union bch_extent_entry *entry;
 685         struct extent_ptr_decoded p;
 686         unsigned durability = 0;
 687
 688         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 689                 durability += bch2_extent_ptr_durability(c, &p);
 690
 691         return durability;
 692 }
 693
 694 static unsigned bch2_bkey_durability_safe(struct bch_fs *c, struct bkey_s_c k)
 695 {
 696         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 697         const union bch_extent_entry *entry;
 698         struct extent_ptr_decoded p;
 699         unsigned durability = 0;
 700
 701         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 702                 if (p.ptr.dev < c->sb.nr_devices && c->devs[p.ptr.dev])
 703                         durability += bch2_extent_ptr_durability(c, &p);
 704
 705         return durability;
 706 }
 707
 708 void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
 709 {
 710         union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
 711         union bch_extent_entry *next = extent_entry_next(entry);
 712
 713         memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
 714         k->k.u64s -= extent_entry_u64s(entry);
 715 }
 716
 717 void bch2_extent_ptr_decoded_append(struct bkey_i *k,
 718                                     struct extent_ptr_decoded *p)
 719 {
 720         struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(k));
 721         struct bch_extent_crc_unpacked crc =
 722                 bch2_extent_crc_unpack(&k->k, NULL);
 723         union bch_extent_entry *pos;
 724
 725         if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 726                 pos = ptrs.start;
 727                 goto found;
 728         }
 729
 730         bkey_for_each_crc(&k->k, ptrs, crc, pos)
 731                 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
 732                         pos = extent_entry_next(pos);
 733                         goto found;
 734                 }
 735
 736         bch2_extent_crc_append(k, p->crc);
 737         pos = bkey_val_end(bkey_i_to_s(k));
 738 found:
 739         p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
 740         __extent_entry_insert(k, pos, to_entry(&p->ptr));
 741
 742         if (p->has_ec) {
 743                 p->ec.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
 744                 __extent_entry_insert(k, pos, to_entry(&p->ec));
 745         }
 746 }
 747
 748 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 749                                           union bch_extent_entry *entry)
 750 {
 751         union bch_extent_entry *i = ptrs.start;
 752
 753         if (i == entry)
 754                 return NULL;
 755
 756         while (extent_entry_next(i) != entry)
 757                 i = extent_entry_next(i);
 758         return i;
 759 }
 760
 761 /*
 762  * Returns pointer to the next entry after the one being dropped:
 763  */
 764 union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
 765                                                    struct bch_extent_ptr *ptr)
 766 {
 767         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 768         union bch_extent_entry *entry = to_entry(ptr), *next;
 769         union bch_extent_entry *ret = entry;
 770         bool drop_crc = true;
 771
 772         EBUG_ON(ptr < &ptrs.start->ptr ||
 773                 ptr >= &ptrs.end->ptr);
 774         EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
 775
 776         for (next = extent_entry_next(entry);
 777              next != ptrs.end;
 778              next = extent_entry_next(next)) {
 779                 if (extent_entry_is_crc(next)) {
 780                         break;
 781                 } else if (extent_entry_is_ptr(next)) {
 782                         drop_crc = false;
 783                         break;
 784                 }
 785         }
 786
 787         extent_entry_drop(k, entry);
 788
 789         while ((entry = extent_entry_prev(ptrs, entry))) {
 790                 if (extent_entry_is_ptr(entry))
 791                         break;
 792
 793                 if ((extent_entry_is_crc(entry) && drop_crc) ||
 794                     extent_entry_is_stripe_ptr(entry)) {
 795                         ret = (void *) ret - extent_entry_bytes(entry);
 796                         extent_entry_drop(k, entry);
 797                 }
 798         }
 799
 800         return ret;
 801 }
 802
 803 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 804                                            struct bch_extent_ptr *ptr)
 805 {
 806         bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
 807         union bch_extent_entry *ret =
 808                 bch2_bkey_drop_ptr_noerror(k, ptr);
 809
 810         /*
 811          * If we deleted all the dirty pointers and there's still cached
 812          * pointers, we could set the cached pointers to dirty if they're not
 813          * stale - but to do that correctly we'd need to grab an open_bucket
 814          * reference so that we don't race with bucket reuse:
 815          */
 816         if (have_dirty &&
 817             !bch2_bkey_dirty_devs(k.s_c).nr) {
 818                 k.k->type = KEY_TYPE_error;
 819                 set_bkey_val_u64s(k.k, 0);
 820                 ret = NULL;
 821         } else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 822                 k.k->type = KEY_TYPE_deleted;
 823                 set_bkey_val_u64s(k.k, 0);
 824                 ret = NULL;
 825         }
 826
 827         return ret;
 828 }
 829
 830 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
 831 {
 832         struct bch_extent_ptr *ptr;
 833
 834         bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
 835 }
 836
 837 void bch2_bkey_drop_device_noerror(struct bkey_s k, unsigned dev)
 838 {
 839         struct bch_extent_ptr *ptr = bch2_bkey_has_device(k, dev);
 840
 841         if (ptr)
 842                 bch2_bkey_drop_ptr_noerror(k, ptr);
 843 }
 844
 845 const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned dev)
 846 {
 847         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 848         const struct bch_extent_ptr *ptr;
 849
 850         bkey_for_each_ptr(ptrs, ptr)
 851                 if (ptr->dev == dev)
 852                         return ptr;
 853
 854         return NULL;
 855 }
 856
 857 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 858 {
 859         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 860         const struct bch_extent_ptr *ptr;
 861
 862         bkey_for_each_ptr(ptrs, ptr)
 863                 if (bch2_dev_in_target(c, ptr->dev, target) &&
 864                     (!ptr->cached ||
 865                      !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
 866                         return true;
 867
 868         return false;
 869 }
 870
 871 bool bch2_bkey_matches_ptr(struct bch_fs *c, struct bkey_s_c k,
 872                            struct bch_extent_ptr m, u64 offset)
 873 {
 874         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 875         const union bch_extent_entry *entry;
 876         struct extent_ptr_decoded p;
 877
 878         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 879                 if (p.ptr.dev   == m.dev &&
 880                     p.ptr.gen   == m.gen &&
 881                     (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(k.k) ==
 882                     (s64) m.offset  - offset)
 883                         return true;
 884
 885         return false;
 886 }
 887
 888 /*
 889  * Returns true if two extents refer to the same data:
 890  */
 891 bool bch2_extents_match(struct bkey_s_c k1, struct bkey_s_c k2)
 892 {
 893         if (k1.k->type != k2.k->type)
 894                 return false;
 895
 896         if (bkey_extent_is_direct_data(k1.k)) {
 897                 struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(k1);
 898                 struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(k2);
 899                 const union bch_extent_entry *entry1, *entry2;
 900                 struct extent_ptr_decoded p1, p2;
 901
 902                 if (bkey_extent_is_unwritten(k1) != bkey_extent_is_unwritten(k2))
 903                         return false;
 904
 905                 bkey_for_each_ptr_decode(k1.k, ptrs1, p1, entry1)
 906                         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 907                                 if (p1.ptr.dev          == p2.ptr.dev &&
 908                                     p1.ptr.gen          == p2.ptr.gen &&
 909                                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 910                                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 911                                         return true;
 912
 913                 return false;
 914         } else {
 915                 /* KEY_TYPE_deleted, etc. */
 916                 return true;
 917         }
 918 }
 919
 920 struct bch_extent_ptr *
 921 bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bkey_s k2)
 922 {
 923         struct bkey_ptrs ptrs2 = bch2_bkey_ptrs(k2);
 924         union bch_extent_entry *entry2;
 925         struct extent_ptr_decoded p2;
 926
 927         bkey_for_each_ptr_decode(k2.k, ptrs2, p2, entry2)
 928                 if (p1.ptr.dev          == p2.ptr.dev &&
 929                     p1.ptr.gen          == p2.ptr.gen &&
 930                     (s64) p1.ptr.offset + p1.crc.offset - bkey_start_offset(k1.k) ==
 931                     (s64) p2.ptr.offset + p2.crc.offset - bkey_start_offset(k2.k))
 932                         return &entry2->ptr;
 933
 934         return NULL;
 935 }
 936
 937 void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
 938 {
 939         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 940         union bch_extent_entry *entry;
 941         union bch_extent_entry *ec = NULL;
 942
 943         bkey_extent_entry_for_each(ptrs, entry) {
 944                 if (&entry->ptr == ptr) {
 945                         ptr->cached = true;
 946                         if (ec)
 947                                 extent_entry_drop(k, ec);
 948                         return;
 949                 }
 950
 951                 if (extent_entry_is_stripe_ptr(entry))
 952                         ec = entry;
 953                 else if (extent_entry_is_ptr(entry))
 954                         ec = NULL;
 955         }
 956
 957         BUG();
 958 }
 959
 960 /*
 961  * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 962  *
 963  * Returns true if @k should be dropped entirely
 964  *
 965  * For existing keys, only called when btree nodes are being rewritten, not when
 966  * they're merely being compacted/resorted in memory.
 967  */
 968 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 969 {
 970         struct bch_extent_ptr *ptr;
 971
 972         bch2_bkey_drop_ptrs(k, ptr,
 973                 ptr->cached &&
 974                 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
 975
 976         return bkey_deleted(k.k);
 977 }
 978
 979 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 980                             struct bkey_s_c k)
 981 {
 982         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 983         const union bch_extent_entry *entry;
 984         bool first = true;
 985
 986         if (c)
 987                 prt_printf(out, "durability: %u ", bch2_bkey_durability_safe(c, k));
 988
 989         bkey_extent_entry_for_each(ptrs, entry) {
 990                 if (!first)
 991                         prt_printf(out, " ");
 992
 993                 switch (__extent_entry_type(entry)) {
 994                 case BCH_EXTENT_ENTRY_ptr: {
 995                         const struct bch_extent_ptr *ptr = entry_to_ptr(entry);
 996                         struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
 997                                 ? bch_dev_bkey_exists(c, ptr->dev)
 998                                 : NULL;
 999
1000                         if (!ca) {
1001                                 prt_printf(out, "ptr: %u:%llu gen %u%s", ptr->dev,
1002                                        (u64) ptr->offset, ptr->gen,
1003                                        ptr->cached ? " cached" : "");
1004                         } else {
1005                                 u32 offset;
1006                                 u64 b = sector_to_bucket_and_offset(ca, ptr->offset, &offset);
1007
1008                                 prt_printf(out, "ptr: %u:%llu:%u gen %u",
1009                                            ptr->dev, b, offset, ptr->gen);
1010                                 if (ptr->cached)
1011                                         prt_str(out, " cached");
1012                                 if (ptr->unwritten)
1013                                         prt_str(out, " unwritten");
1014                                 if (ca && ptr_stale(ca, ptr))
1015                                         prt_printf(out, " stale");
1016                         }
1017                         break;
1018                 }
1019                 case BCH_EXTENT_ENTRY_crc32:
1020                 case BCH_EXTENT_ENTRY_crc64:
1021                 case BCH_EXTENT_ENTRY_crc128: {
1022                         struct bch_extent_crc_unpacked crc =
1023                                 bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1024
1025                         prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
1026                                crc.compressed_size,
1027                                crc.uncompressed_size,
1028                                crc.offset, crc.nonce,
1029                                bch2_csum_types[crc.csum_type],
1030                                bch2_compression_types[crc.compression_type]);
1031                         break;
1032                 }
1033                 case BCH_EXTENT_ENTRY_stripe_ptr: {
1034                         const struct bch_extent_stripe_ptr *ec = &entry->stripe_ptr;
1035
1036                         prt_printf(out, "ec: idx %llu block %u",
1037                                (u64) ec->idx, ec->block);
1038                         break;
1039                 }
1040                 case BCH_EXTENT_ENTRY_rebalance: {
1041                         const struct bch_extent_rebalance *r = &entry->rebalance;
1042
1043                         prt_str(out, "rebalance: target ");
1044                         if (c)
1045                                 bch2_target_to_text(out, c, r->target);
1046                         else
1047                                 prt_printf(out, "%u", r->target);
1048                         prt_str(out, " compression ");
1049                         bch2_compression_opt_to_text(out, r->compression);
1050                         break;
1051                 }
1052                 default:
1053                         prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
1054                         return;
1055                 }
1056
1057                 first = false;
1058         }
1059 }
1060
1061 static int extent_ptr_invalid(const struct bch_fs *c,
1062                               struct bkey_s_c k,
1063                               enum bkey_invalid_flags flags,
1064                               const struct bch_extent_ptr *ptr,
1065                               unsigned size_ondisk,
1066                               bool metadata,
1067                               struct printbuf *err)
1068 {
1069         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1070         const struct bch_extent_ptr *ptr2;
1071         u64 bucket;
1072         u32 bucket_offset;
1073         struct bch_dev *ca;
1074
1075         if (!bch2_dev_exists2(c, ptr->dev)) {
1076                 /*
1077                  * If we're in the write path this key might have already been
1078                  * overwritten, and we could be seeing a device that doesn't
1079                  * exist anymore due to racing with device removal:
1080                  */
1081                 if (flags & BKEY_INVALID_WRITE)
1082                         return 0;
1083
1084                 prt_printf(err, "pointer to invalid device (%u)", ptr->dev);
1085                 return -BCH_ERR_invalid_bkey;
1086         }
1087
1088         ca = bch_dev_bkey_exists(c, ptr->dev);
1089         bkey_for_each_ptr(ptrs, ptr2)
1090                 if (ptr != ptr2 && ptr->dev == ptr2->dev) {
1091                         prt_printf(err, "multiple pointers to same device (%u)", ptr->dev);
1092                         return -BCH_ERR_invalid_bkey;
1093                 }
1094
1095         bucket = sector_to_bucket_and_offset(ca, ptr->offset, &bucket_offset);
1096
1097         if (bucket >= ca->mi.nbuckets) {
1098                 prt_printf(err, "pointer past last bucket (%llu > %llu)",
1099                        bucket, ca->mi.nbuckets);
1100                 return -BCH_ERR_invalid_bkey;
1101         }
1102
1103         if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket)) {
1104                 prt_printf(err, "pointer before first bucket (%llu < %u)",
1105                        bucket, ca->mi.first_bucket);
1106                 return -BCH_ERR_invalid_bkey;
1107         }
1108
1109         if (bucket_offset + size_ondisk > ca->mi.bucket_size) {
1110                 prt_printf(err, "pointer spans multiple buckets (%u + %u > %u)",
1111                        bucket_offset, size_ondisk, ca->mi.bucket_size);
1112                 return -BCH_ERR_invalid_bkey;
1113         }
1114
1115         return 0;
1116 }
1117
1118 int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
1119                            enum bkey_invalid_flags flags,
1120                            struct printbuf *err)
1121 {
1122         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1123         const union bch_extent_entry *entry;
1124         struct bch_extent_crc_unpacked crc;
1125         unsigned size_ondisk = k.k->size;
1126         unsigned nonce = UINT_MAX;
1127         unsigned nr_ptrs = 0;
1128         bool unwritten = false, have_ec = false, crc_since_last_ptr = false;
1129         int ret;
1130
1131         if (bkey_is_btree_ptr(k.k))
1132                 size_ondisk = btree_sectors(c);
1133
1134         bkey_extent_entry_for_each(ptrs, entry) {
1135                 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX) {
1136                         prt_printf(err, "invalid extent entry type (got %u, max %u)",
1137                                __extent_entry_type(entry), BCH_EXTENT_ENTRY_MAX);
1138                         return -BCH_ERR_invalid_bkey;
1139                 }
1140
1141                 if (bkey_is_btree_ptr(k.k) &&
1142                     !extent_entry_is_ptr(entry)) {
1143                         prt_printf(err, "has non ptr field");
1144                         return -BCH_ERR_invalid_bkey;
1145                 }
1146
1147                 switch (extent_entry_type(entry)) {
1148                 case BCH_EXTENT_ENTRY_ptr:
1149                         ret = extent_ptr_invalid(c, k, flags, &entry->ptr,
1150                                                  size_ondisk, false, err);
1151                         if (ret)
1152                                 return ret;
1153
1154                         if (nr_ptrs && unwritten != entry->ptr.unwritten) {
1155                                 prt_printf(err, "extent with unwritten and written ptrs");
1156                                 return -BCH_ERR_invalid_bkey;
1157                         }
1158
1159                         if (k.k->type != KEY_TYPE_extent && entry->ptr.unwritten) {
1160                                 prt_printf(err, "has unwritten ptrs");
1161                                 return -BCH_ERR_invalid_bkey;
1162                         }
1163
1164                         if (entry->ptr.cached && have_ec) {
1165                                 prt_printf(err, "cached, erasure coded ptr");
1166                                 return -BCH_ERR_invalid_bkey;
1167                         }
1168
1169                         unwritten = entry->ptr.unwritten;
1170                         have_ec = false;
1171                         crc_since_last_ptr = false;
1172                         nr_ptrs++;
1173                         break;
1174                 case BCH_EXTENT_ENTRY_crc32:
1175                 case BCH_EXTENT_ENTRY_crc64:
1176                 case BCH_EXTENT_ENTRY_crc128:
1177                         crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
1178
1179                         if (crc.offset + crc.live_size >
1180                             crc.uncompressed_size) {
1181                                 prt_printf(err, "checksum offset + key size > uncompressed size");
1182                                 return -BCH_ERR_invalid_bkey;
1183                         }
1184
1185                         size_ondisk = crc.compressed_size;
1186
1187                         if (!bch2_checksum_type_valid(c, crc.csum_type)) {
1188                                 prt_printf(err, "invalid checksum type");
1189                                 return -BCH_ERR_invalid_bkey;
1190                         }
1191
1192                         if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR) {
1193                                 prt_printf(err, "invalid compression type");
1194                                 return -BCH_ERR_invalid_bkey;
1195                         }
1196
1197                         if (bch2_csum_type_is_encryption(crc.csum_type)) {
1198                                 if (nonce == UINT_MAX)
1199                                         nonce = crc.offset + crc.nonce;
1200                                 else if (nonce != crc.offset + crc.nonce) {
1201                                         prt_printf(err, "incorrect nonce");
1202                                         return -BCH_ERR_invalid_bkey;
1203                                 }
1204                         }
1205
1206                         if (crc_since_last_ptr) {
1207                                 prt_printf(err, "redundant crc entry");
1208                                 return -BCH_ERR_invalid_bkey;
1209                         }
1210                         crc_since_last_ptr = true;
1211
1212                         if (crc_is_encoded(crc) &&
1213                             (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
1214                             (flags & (BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT))) {
1215                                 prt_printf(err, "too large encoded extent");
1216                                 return -BCH_ERR_invalid_bkey;
1217                         }
1218
1219                         break;
1220                 case BCH_EXTENT_ENTRY_stripe_ptr:
1221                         if (have_ec) {
1222                                 prt_printf(err, "redundant stripe entry");
1223                                 return -BCH_ERR_invalid_bkey;
1224                         }
1225                         have_ec = true;
1226                         break;
1227                 case BCH_EXTENT_ENTRY_rebalance: {
1228                         const struct bch_extent_rebalance *r = &entry->rebalance;
1229
1230                         if (!bch2_compression_opt_valid(r->compression)) {
1231                                 struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
1232                                 prt_printf(err, "invalid compression opt %u:%u",
1233                                            opt.type, opt.level);
1234                                 return -BCH_ERR_invalid_bkey;
1235                         }
1236                         break;
1237                 }
1238                 }
1239         }
1240
1241         if (!nr_ptrs) {
1242                 prt_str(err, "no ptrs");
1243                 return -BCH_ERR_invalid_bkey;
1244         }
1245
1246         if (nr_ptrs >= BCH_BKEY_PTRS_MAX) {
1247                 prt_str(err, "too many ptrs");
1248                 return -BCH_ERR_invalid_bkey;
1249         }
1250
1251         if (crc_since_last_ptr) {
1252                 prt_printf(err, "redundant crc entry");
1253                 return -BCH_ERR_invalid_bkey;
1254         }
1255
1256         if (have_ec) {
1257                 prt_printf(err, "redundant stripe entry");
1258                 return -BCH_ERR_invalid_bkey;
1259         }
1260
1261         return 0;
1262 }
1263
1264 void bch2_ptr_swab(struct bkey_s k)
1265 {
1266         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1267         union bch_extent_entry *entry;
1268         u64 *d;
1269
1270         for (d =  (u64 *) ptrs.start;
1271              d != (u64 *) ptrs.end;
1272              d++)
1273                 *d = swab64(*d);
1274
1275         for (entry = ptrs.start;
1276              entry < ptrs.end;
1277              entry = extent_entry_next(entry)) {
1278                 switch (extent_entry_type(entry)) {
1279                 case BCH_EXTENT_ENTRY_ptr:
1280                         break;
1281                 case BCH_EXTENT_ENTRY_crc32:
1282                         entry->crc32.csum = swab32(entry->crc32.csum);
1283                         break;
1284                 case BCH_EXTENT_ENTRY_crc64:
1285                         entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
1286                         entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
1287                         break;
1288                 case BCH_EXTENT_ENTRY_crc128:
1289                         entry->crc128.csum.hi = (__force __le64)
1290                                 swab64((__force u64) entry->crc128.csum.hi);
1291                         entry->crc128.csum.lo = (__force __le64)
1292                                 swab64((__force u64) entry->crc128.csum.lo);
1293                         break;
1294                 case BCH_EXTENT_ENTRY_stripe_ptr:
1295                         break;
1296                 case BCH_EXTENT_ENTRY_rebalance:
1297                         break;
1298                 }
1299         }
1300 }
1301
1302 const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c k)
1303 {
1304         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1305         const union bch_extent_entry *entry;
1306
1307         bkey_extent_entry_for_each(ptrs, entry)
1308                 if (__extent_entry_type(entry) == BCH_EXTENT_ENTRY_rebalance)
1309                         return &entry->rebalance;
1310
1311         return NULL;
1312 }
1313
1314 unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *c, struct bkey_s_c k,
1315                                        unsigned target, unsigned compression)
1316 {
1317         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1318         unsigned rewrite_ptrs = 0;
1319
1320         if (compression) {
1321                 unsigned compression_type = bch2_compression_opt_to_type(compression);
1322                 const union bch_extent_entry *entry;
1323                 struct extent_ptr_decoded p;
1324                 unsigned i = 0;
1325
1326                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1327                         if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible) {
1328                                 rewrite_ptrs = 0;
1329                                 goto incompressible;
1330                         }
1331
1332                         if (!p.ptr.cached && p.crc.compression_type != compression_type)
1333                                 rewrite_ptrs |= 1U << i;
1334                         i++;
1335                 }
1336         }
1337 incompressible:
1338         if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
1339                 const struct bch_extent_ptr *ptr;
1340                 unsigned i = 0;
1341
1342                 bkey_for_each_ptr(ptrs, ptr) {
1343                         if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, target))
1344                                 rewrite_ptrs |= 1U << i;
1345                         i++;
1346                 }
1347         }
1348
1349         return rewrite_ptrs;
1350 }
1351
1352 bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k)
1353 {
1354         const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
1355
1356         /*
1357          * If it's an indirect extent, we don't delete the rebalance entry when
1358          * done so that we know what options were applied - check if it still
1359          * needs work done:
1360          */
1361         if (r &&
1362             k.k->type == KEY_TYPE_reflink_v &&
1363             !bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression))
1364                 r = NULL;
1365
1366         return r != NULL;
1367 }
1368
1369 int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
1370                                   unsigned target, unsigned compression)
1371 {
1372         struct bkey_s k = bkey_i_to_s(_k);
1373         struct bch_extent_rebalance *r;
1374         bool needs_rebalance;
1375
1376         if (!bkey_extent_is_direct_data(k.k))
1377                 return 0;
1378
1379         /* get existing rebalance entry: */
1380         r = (struct bch_extent_rebalance *) bch2_bkey_rebalance_opts(k.s_c);
1381         if (r) {
1382                 if (k.k->type == KEY_TYPE_reflink_v) {
1383                         /*
1384                          * indirect extents: existing options take precedence,
1385                          * so that we don't move extents back and forth if
1386                          * they're referenced by different inodes with different
1387                          * options:
1388                          */
1389                         if (r->target)
1390                                 target = r->target;
1391                         if (r->compression)
1392                                 compression = r->compression;
1393                 }
1394
1395                 r->target       = target;
1396                 r->compression  = compression;
1397         }
1398
1399         needs_rebalance = bch2_bkey_ptrs_need_rebalance(c, k.s_c, target, compression);
1400
1401         if (needs_rebalance && !r) {
1402                 union bch_extent_entry *new = bkey_val_end(k);
1403
1404                 new->rebalance.type             = 1U << BCH_EXTENT_ENTRY_rebalance;
1405                 new->rebalance.compression      = compression;
1406                 new->rebalance.target           = target;
1407                 new->rebalance.unused           = 0;
1408                 k.k->u64s += extent_entry_u64s(new);
1409         } else if (!needs_rebalance && r && k.k->type != KEY_TYPE_reflink_v) {
1410                 /*
1411                  * For indirect extents, don't delete the rebalance entry when
1412                  * we're finished so that we know we specifically moved it or
1413                  * compressed it to its current location/compression type
1414                  */
1415                 extent_entry_drop(k, (union bch_extent_entry *) r);
1416         }
1417
1418         return 0;
1419 }
1420
1421 /* Generic extent code: */
1422
1423 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
1424 {
1425         unsigned new_val_u64s = bkey_val_u64s(k.k);
1426         int val_u64s_delta;
1427         u64 sub;
1428
1429         if (bkey_le(where, bkey_start_pos(k.k)))
1430                 return 0;
1431
1432         EBUG_ON(bkey_gt(where, k.k->p));
1433
1434         sub = where.offset - bkey_start_offset(k.k);
1435
1436         k.k->size -= sub;
1437
1438         if (!k.k->size) {
1439                 k.k->type = KEY_TYPE_deleted;
1440                 new_val_u64s = 0;
1441         }
1442
1443         switch (k.k->type) {
1444         case KEY_TYPE_extent:
1445         case KEY_TYPE_reflink_v: {
1446                 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
1447                 union bch_extent_entry *entry;
1448                 bool seen_crc = false;
1449
1450                 bkey_extent_entry_for_each(ptrs, entry) {
1451                         switch (extent_entry_type(entry)) {
1452                         case BCH_EXTENT_ENTRY_ptr:
1453                                 if (!seen_crc)
1454                                         entry->ptr.offset += sub;
1455                                 break;
1456                         case BCH_EXTENT_ENTRY_crc32:
1457                                 entry->crc32.offset += sub;
1458                                 break;
1459                         case BCH_EXTENT_ENTRY_crc64:
1460                                 entry->crc64.offset += sub;
1461                                 break;
1462                         case BCH_EXTENT_ENTRY_crc128:
1463                                 entry->crc128.offset += sub;
1464                                 break;
1465                         case BCH_EXTENT_ENTRY_stripe_ptr:
1466                                 break;
1467                         case BCH_EXTENT_ENTRY_rebalance:
1468                                 break;
1469                         }
1470
1471                         if (extent_entry_is_crc(entry))
1472                                 seen_crc = true;
1473                 }
1474
1475                 break;
1476         }
1477         case KEY_TYPE_reflink_p: {
1478                 struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
1479
1480                 le64_add_cpu(&p.v->idx, sub);
1481                 break;
1482         }
1483         case KEY_TYPE_inline_data:
1484         case KEY_TYPE_indirect_inline_data: {
1485                 void *p = bkey_inline_data_p(k);
1486                 unsigned bytes = bkey_inline_data_bytes(k.k);
1487
1488                 sub = min_t(u64, sub << 9, bytes);
1489
1490                 memmove(p, p + sub, bytes - sub);
1491
1492                 new_val_u64s -= sub >> 3;
1493                 break;
1494         }
1495         }
1496
1497         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1498         BUG_ON(val_u64s_delta < 0);
1499
1500         set_bkey_val_u64s(k.k, new_val_u64s);
1501         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1502         return -val_u64s_delta;
1503 }
1504
1505 int bch2_cut_back_s(struct bpos where, struct bkey_s k)
1506 {
1507         unsigned new_val_u64s = bkey_val_u64s(k.k);
1508         int val_u64s_delta;
1509         u64 len = 0;
1510
1511         if (bkey_ge(where, k.k->p))
1512                 return 0;
1513
1514         EBUG_ON(bkey_lt(where, bkey_start_pos(k.k)));
1515
1516         len = where.offset - bkey_start_offset(k.k);
1517
1518         k.k->p.offset = where.offset;
1519         k.k->size = len;
1520
1521         if (!len) {
1522                 k.k->type = KEY_TYPE_deleted;
1523                 new_val_u64s = 0;
1524         }
1525
1526         switch (k.k->type) {
1527         case KEY_TYPE_inline_data:
1528         case KEY_TYPE_indirect_inline_data:
1529                 new_val_u64s = (bkey_inline_data_offset(k.k) +
1530                                 min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
1531                 break;
1532         }
1533
1534         val_u64s_delta = bkey_val_u64s(k.k) - new_val_u64s;
1535         BUG_ON(val_u64s_delta < 0);
1536
1537         set_bkey_val_u64s(k.k, new_val_u64s);
1538         memset(bkey_val_end(k), 0, val_u64s_delta * sizeof(u64));
1539         return -val_u64s_delta;
1540 }