fs/bcachefs/journal.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_JOURNAL_H
   3 #define _BCACHEFS_JOURNAL_H
   4
   5 /*
   6  * THE JOURNAL:
   7  *
   8  * The primary purpose of the journal is to log updates (insertions) to the
   9  * b-tree, to avoid having to do synchronous updates to the b-tree on disk.
  10  *
  11  * Without the journal, the b-tree is always internally consistent on
  12  * disk - and in fact, in the earliest incarnations bcache didn't have a journal
  13  * but did handle unclean shutdowns by doing all index updates synchronously
  14  * (with coalescing).
  15  *
  16  * Updates to interior nodes still happen synchronously and without the journal
  17  * (for simplicity) - this may change eventually but updates to interior nodes
  18  * are rare enough it's not a huge priority.
  19  *
  20  * This means the journal is relatively separate from the b-tree; it consists of
  21  * just a list of keys and journal replay consists of just redoing those
  22  * insertions in same order that they appear in the journal.
  23  *
  24  * PERSISTENCE:
  25  *
  26  * For synchronous updates (where we're waiting on the index update to hit
  27  * disk), the journal entry will be written out immediately (or as soon as
  28  * possible, if the write for the previous journal entry was still in flight).
  29  *
  30  * Synchronous updates are specified by passing a closure (@flush_cl) to
  31  * bch2_btree_insert() or bch_btree_insert_node(), which then pass that parameter
  32  * down to the journalling code. That closure will wait on the journal write to
  33  * complete (via closure_wait()).
  34  *
  35  * If the index update wasn't synchronous, the journal entry will be
  36  * written out after 10 ms have elapsed, by default (the delay_ms field
  37  * in struct journal).
  38  *
  39  * JOURNAL ENTRIES:
  40  *
  41  * A journal entry is variable size (struct jset), it's got a fixed length
  42  * header and then a variable number of struct jset_entry entries.
  43  *
  44  * Journal entries are identified by monotonically increasing 64 bit sequence
  45  * numbers - jset->seq; other places in the code refer to this sequence number.
  46  *
  47  * A jset_entry entry contains one or more bkeys (which is what gets inserted
  48  * into the b-tree). We need a container to indicate which b-tree the key is
  49  * for; also, the roots of the various b-trees are stored in jset_entry entries
  50  * (one for each b-tree) - this lets us add new b-tree types without changing
  51  * the on disk format.
  52  *
  53  * We also keep some things in the journal header that are logically part of the
  54  * superblock - all the things that are frequently updated. This is for future
  55  * bcache on raw flash support; the superblock (which will become another
  56  * journal) can't be moved or wear leveled, so it contains just enough
  57  * information to find the main journal, and the superblock only has to be
  58  * rewritten when we want to move/wear level the main journal.
  59  *
  60  * JOURNAL LAYOUT ON DISK:
  61  *
  62  * The journal is written to a ringbuffer of buckets (which is kept in the
  63  * superblock); the individual buckets are not necessarily contiguous on disk
  64  * which means that journal entries are not allowed to span buckets, but also
  65  * that we can resize the journal at runtime if desired (unimplemented).
  66  *
  67  * The journal buckets exist in the same pool as all the other buckets that are
  68  * managed by the allocator and garbage collection - garbage collection marks
  69  * the journal buckets as metadata buckets.
  70  *
  71  * OPEN/DIRTY JOURNAL ENTRIES:
  72  *
  73  * Open/dirty journal entries are journal entries that contain b-tree updates
  74  * that have not yet been written out to the b-tree on disk. We have to track
  75  * which journal entries are dirty, and we also have to avoid wrapping around
  76  * the journal and overwriting old but still dirty journal entries with new
  77  * journal entries.
  78  *
  79  * On disk, this is represented with the "last_seq" field of struct jset;
  80  * last_seq is the first sequence number that journal replay has to replay.
  81  *
  82  * To avoid overwriting dirty journal entries on disk, we keep a mapping (in
  83  * journal_device->seq) of for each journal bucket, the highest sequence number
  84  * any journal entry it contains. Then, by comparing that against last_seq we
  85  * can determine whether that journal bucket contains dirty journal entries or
  86  * not.
  87  *
  88  * To track which journal entries are dirty, we maintain a fifo of refcounts
  89  * (where each entry corresponds to a specific sequence number) - when a ref
  90  * goes to 0, that journal entry is no longer dirty.
  91  *
  92  * Journalling of index updates is done at the same time as the b-tree itself is
  93  * being modified (see btree_insert_key()); when we add the key to the journal
  94  * the pending b-tree write takes a ref on the journal entry the key was added
  95  * to. If a pending b-tree write would need to take refs on multiple dirty
  96  * journal entries, it only keeps the ref on the oldest one (since a newer
  97  * journal entry will still be replayed if an older entry was dirty).
  98  *
  99  * JOURNAL FILLING UP:
 100  *
 101  * There are two ways the journal could fill up; either we could run out of
 102  * space to write to, or we could have too many open journal entries and run out
 103  * of room in the fifo of refcounts. Since those refcounts are decremented
 104  * without any locking we can't safely resize that fifo, so we handle it the
 105  * same way.
 106  *
 107  * If the journal fills up, we start flushing dirty btree nodes until we can
 108  * allocate space for a journal write again - preferentially flushing btree
 109  * nodes that are pinning the oldest journal entries first.
 110  */
 111
 112 #include <linux/hash.h>
 113
 114 #include "journal_types.h"
 115
 116 struct bch_fs;
 117
 118 static inline void journal_wake(struct journal *j)
 119 {
 120         wake_up(&j->wait);
 121         closure_wake_up(&j->async_wait);
 122         closure_wake_up(&j->preres_wait);
 123 }
 124
 125 static inline struct journal_buf *journal_cur_buf(struct journal *j)
 126 {
 127         return j->buf + j->reservations.idx;
 128 }
 129
 130 /* Sequence number of oldest dirty journal entry */
 131
 132 static inline u64 journal_last_seq(struct journal *j)
 133 {
 134         return j->pin.front;
 135 }
 136
 137 static inline u64 journal_cur_seq(struct journal *j)
 138 {
 139         return atomic64_read(&j->seq);
 140 }
 141
 142 static inline u64 journal_last_unwritten_seq(struct journal *j)
 143 {
 144         return j->seq_ondisk + 1;
 145 }
 146
 147 static inline int journal_state_count(union journal_res_state s, int idx)
 148 {
 149         switch (idx) {
 150         case 0: return s.buf0_count;
 151         case 1: return s.buf1_count;
 152         case 2: return s.buf2_count;
 153         case 3: return s.buf3_count;
 154         }
 155         BUG();
 156 }
 157
 158 static inline void journal_state_inc(union journal_res_state *s)
 159 {
 160         s->buf0_count += s->idx == 0;
 161         s->buf1_count += s->idx == 1;
 162         s->buf2_count += s->idx == 2;
 163         s->buf3_count += s->idx == 3;
 164 }
 165
 166 /*
 167  * Amount of space that will be taken up by some keys in the journal (i.e.
 168  * including the jset header)
 169  */
 170 static inline unsigned jset_u64s(unsigned u64s)
 171 {
 172         return u64s + sizeof(struct jset_entry) / sizeof(u64);
 173 }
 174
 175 static inline int journal_entry_overhead(struct journal *j)
 176 {
 177         return sizeof(struct jset) / sizeof(u64) + j->entry_u64s_reserved;
 178 }
 179
 180 static inline struct jset_entry *
 181 bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s)
 182 {
 183         struct jset *jset = buf->data;
 184         struct jset_entry *entry = vstruct_idx(jset, le32_to_cpu(jset->u64s));
 185
 186         memset(entry, 0, sizeof(*entry));
 187         entry->u64s = cpu_to_le16(u64s);
 188
 189         le32_add_cpu(&jset->u64s, jset_u64s(u64s));
 190
 191         return entry;
 192 }
 193
 194 static inline struct jset_entry *
 195 journal_res_entry(struct journal *j, struct journal_res *res)
 196 {
 197         return vstruct_idx(j->buf[res->idx].data, res->offset);
 198 }
 199
 200 static inline unsigned journal_entry_init(struct jset_entry *entry, unsigned type,
 201                                           enum btree_id id, unsigned level,
 202                                           unsigned u64s)
 203 {
 204         entry->u64s     = cpu_to_le16(u64s);
 205         entry->btree_id = id;
 206         entry->level    = level;
 207         entry->type     = type;
 208         entry->pad[0]   = 0;
 209         entry->pad[1]   = 0;
 210         entry->pad[2]   = 0;
 211         return jset_u64s(u64s);
 212 }
 213
 214 static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type,
 215                                           enum btree_id id, unsigned level,
 216                                           const void *data, unsigned u64s)
 217 {
 218         unsigned ret = journal_entry_init(entry, type, id, level, u64s);
 219
 220         memcpy_u64s_small(entry->_data, data, u64s);
 221         return ret;
 222 }
 223
 224 static inline struct jset_entry *
 225 bch2_journal_add_entry(struct journal *j, struct journal_res *res,
 226                          unsigned type, enum btree_id id,
 227                          unsigned level, unsigned u64s)
 228 {
 229         struct jset_entry *entry = journal_res_entry(j, res);
 230         unsigned actual = journal_entry_init(entry, type, id, level, u64s);
 231
 232         EBUG_ON(!res->ref);
 233         EBUG_ON(actual > res->u64s);
 234
 235         res->offset     += actual;
 236         res->u64s       -= actual;
 237         return entry;
 238 }
 239
 240 static inline bool journal_entry_empty(struct jset *j)
 241 {
 242         struct jset_entry *i;
 243
 244         if (j->seq != j->last_seq)
 245                 return false;
 246
 247         vstruct_for_each(j, i)
 248                 if (i->type == BCH_JSET_ENTRY_btree_keys && i->u64s)
 249                         return false;
 250         return true;
 251 }
 252
 253 /*
 254  * Drop reference on a buffer index and return true if the count has hit zero.
 255  */
 256 static inline union journal_res_state journal_state_buf_put(struct journal *j, unsigned idx)
 257 {
 258         union journal_res_state s;
 259
 260         s.v = atomic64_sub_return(((union journal_res_state) {
 261                                     .buf0_count = idx == 0,
 262                                     .buf1_count = idx == 1,
 263                                     .buf2_count = idx == 2,
 264                                     .buf3_count = idx == 3,
 265                                     }).v, &j->reservations.counter);
 266         return s;
 267 }
 268
 269 bool bch2_journal_entry_close(struct journal *);
 270 void bch2_journal_buf_put_final(struct journal *, u64, bool);
 271
 272 static inline void __bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq)
 273 {
 274         union journal_res_state s;
 275
 276         s = journal_state_buf_put(j, idx);
 277         if (!journal_state_count(s, idx))
 278                 bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx);
 279 }
 280
 281 static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq)
 282 {
 283         union journal_res_state s;
 284
 285         s = journal_state_buf_put(j, idx);
 286         if (!journal_state_count(s, idx)) {
 287                 spin_lock(&j->lock);
 288                 bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx);
 289                 spin_unlock(&j->lock);
 290         }
 291 }
 292
 293 /*
 294  * This function releases the journal write structure so other threads can
 295  * then proceed to add their keys as well.
 296  */
 297 static inline void bch2_journal_res_put(struct journal *j,
 298                                        struct journal_res *res)
 299 {
 300         if (!res->ref)
 301                 return;
 302
 303         lock_release(&j->res_map, _THIS_IP_);
 304
 305         while (res->u64s)
 306                 bch2_journal_add_entry(j, res,
 307                                        BCH_JSET_ENTRY_btree_keys,
 308                                        0, 0, 0);
 309
 310         bch2_journal_buf_put(j, res->idx, res->seq);
 311
 312         res->ref = 0;
 313 }
 314
 315 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *,
 316                                   unsigned);
 317
 318 /* First bits for BCH_WATERMARK: */
 319 enum journal_res_flags {
 320         __JOURNAL_RES_GET_NONBLOCK      = BCH_WATERMARK_BITS,
 321         __JOURNAL_RES_GET_CHECK,
 322 };
 323
 324 #define JOURNAL_RES_GET_NONBLOCK        (1 << __JOURNAL_RES_GET_NONBLOCK)
 325 #define JOURNAL_RES_GET_CHECK           (1 << __JOURNAL_RES_GET_CHECK)
 326
 327 static inline int journal_res_get_fast(struct journal *j,
 328                                        struct journal_res *res,
 329                                        unsigned flags)
 330 {
 331         union journal_res_state old, new;
 332         u64 v = atomic64_read(&j->reservations.counter);
 333
 334         do {
 335                 old.v = new.v = v;
 336
 337                 /*
 338                  * Check if there is still room in the current journal
 339                  * entry:
 340                  */
 341                 if (new.cur_entry_offset + res->u64s > j->cur_entry_u64s)
 342                         return 0;
 343
 344                 EBUG_ON(!journal_state_count(new, new.idx));
 345
 346                 if ((flags & BCH_WATERMARK_MASK) < j->watermark)
 347                         return 0;
 348
 349                 new.cur_entry_offset += res->u64s;
 350                 journal_state_inc(&new);
 351
 352                 /*
 353                  * If the refcount would overflow, we have to wait:
 354                  * XXX - tracepoint this:
 355                  */
 356                 if (!journal_state_count(new, new.idx))
 357                         return 0;
 358
 359                 if (flags & JOURNAL_RES_GET_CHECK)
 360                         return 1;
 361         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 362                                        old.v, new.v)) != old.v);
 363
 364         res->ref        = true;
 365         res->idx        = old.idx;
 366         res->offset     = old.cur_entry_offset;
 367         res->seq        = le64_to_cpu(j->buf[old.idx].data->seq);
 368         return 1;
 369 }
 370
 371 static inline int bch2_journal_res_get(struct journal *j, struct journal_res *res,
 372                                        unsigned u64s, unsigned flags)
 373 {
 374         int ret;
 375
 376         EBUG_ON(res->ref);
 377         EBUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
 378
 379         res->u64s = u64s;
 380
 381         if (journal_res_get_fast(j, res, flags))
 382                 goto out;
 383
 384         ret = bch2_journal_res_get_slowpath(j, res, flags);
 385         if (ret)
 386                 return ret;
 387 out:
 388         if (!(flags & JOURNAL_RES_GET_CHECK)) {
 389                 lock_acquire_shared(&j->res_map, 0,
 390                                     (flags & JOURNAL_RES_GET_NONBLOCK) != 0,
 391                                     NULL, _THIS_IP_);
 392                 EBUG_ON(!res->ref);
 393         }
 394         return 0;
 395 }
 396
 397 /* journal_entry_res: */
 398
 399 void bch2_journal_entry_res_resize(struct journal *,
 400                                    struct journal_entry_res *,
 401                                    unsigned);
 402
 403 int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
 404 void bch2_journal_flush_async(struct journal *, struct closure *);
 405
 406 int bch2_journal_flush_seq(struct journal *, u64);
 407 int bch2_journal_flush(struct journal *);
 408 bool bch2_journal_noflush_seq(struct journal *, u64);
 409 int bch2_journal_meta(struct journal *);
 410
 411 void bch2_journal_halt(struct journal *);
 412
 413 static inline int bch2_journal_error(struct journal *j)
 414 {
 415         return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL
 416                 ? -EIO : 0;
 417 }
 418
 419 struct bch_dev;
 420
 421 static inline void bch2_journal_set_replay_done(struct journal *j)
 422 {
 423         BUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
 424         set_bit(JOURNAL_REPLAY_DONE, &j->flags);
 425 }
 426
 427 void bch2_journal_unblock(struct journal *);
 428 void bch2_journal_block(struct journal *);
 429
 430 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 431 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 432 void bch2_journal_pins_to_text(struct printbuf *, struct journal *);
 433 bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *);
 434
 435 int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
 436                                 unsigned nr);
 437 int bch2_dev_journal_alloc(struct bch_dev *);
 438 int bch2_fs_journal_alloc(struct bch_fs *);
 439
 440 void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
 441
 442 void bch2_fs_journal_stop(struct journal *);
 443 int bch2_fs_journal_start(struct journal *, u64);
 444
 445 void bch2_dev_journal_exit(struct bch_dev *);
 446 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *);
 447 void bch2_fs_journal_exit(struct journal *);
 448 int bch2_fs_journal_init(struct journal *);
 449
 450 #endif /* _BCACHEFS_JOURNAL_H */