drivers/md/bcache/movinggc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Moving/copying garbage collector
   4  *
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcache.h"
   9 #include "btree.h"
  10 #include "debug.h"
  11 #include "request.h"
  12
  13 #include <trace/events/bcache.h>
  14
  15 struct moving_io {
  16         struct closure          cl;
  17         struct keybuf_key       *w;
  18         struct data_insert_op   op;
  19         struct bbio             bio;
  20 };
  21
  22 static bool moving_pred(struct keybuf *buf, struct bkey *k)
  23 {
  24         struct cache_set *c = container_of(buf, struct cache_set,
  25                                            moving_gc_keys);
  26         unsigned int i;
  27
  28         for (i = 0; i < KEY_PTRS(k); i++)
  29                 if (ptr_available(c, k, i) &&
  30                     GC_MOVE(PTR_BUCKET(c, k, i)))
  31                         return true;
  32
  33         return false;
  34 }
  35
  36 /* Moving GC - IO loop */
  37
  38 static CLOSURE_CALLBACK(moving_io_destructor)
  39 {
  40         closure_type(io, struct moving_io, cl);
  41
  42         kfree(io);
  43 }
  44
  45 static CLOSURE_CALLBACK(write_moving_finish)
  46 {
  47         closure_type(io, struct moving_io, cl);
  48         struct bio *bio = &io->bio.bio;
  49
  50         bio_free_pages(bio);
  51
  52         if (io->op.replace_collision)
  53                 trace_bcache_gc_copy_collision(&io->w->key);
  54
  55         bch_keybuf_del(&io->op.c->moving_gc_keys, io->w);
  56
  57         up(&io->op.c->moving_in_flight);
  58
  59         closure_return_with_destructor(cl, moving_io_destructor);
  60 }
  61
  62 static void read_moving_endio(struct bio *bio)
  63 {
  64         struct bbio *b = container_of(bio, struct bbio, bio);
  65         struct moving_io *io = container_of(bio->bi_private,
  66                                             struct moving_io, cl);
  67
  68         if (bio->bi_status)
  69                 io->op.status = bio->bi_status;
  70         else if (!KEY_DIRTY(&b->key) &&
  71                  ptr_stale(io->op.c, &b->key, 0)) {
  72                 io->op.status = BLK_STS_IOERR;
  73         }
  74
  75         bch_bbio_endio(io->op.c, bio, bio->bi_status, "reading data to move");
  76 }
  77
  78 static void moving_init(struct moving_io *io)
  79 {
  80         struct bio *bio = &io->bio.bio;
  81
  82         bio_init(bio, NULL, bio->bi_inline_vecs,
  83                  DIV_ROUND_UP(KEY_SIZE(&io->w->key), PAGE_SECTORS), 0);
  84         bio_get(bio);
  85         bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
  86
  87         bio->bi_iter.bi_size    = KEY_SIZE(&io->w->key) << 9;
  88         bio->bi_private         = &io->cl;
  89         bch_bio_map(bio, NULL);
  90 }
  91
  92 static CLOSURE_CALLBACK(write_moving)
  93 {
  94         closure_type(io, struct moving_io, cl);
  95         struct data_insert_op *op = &io->op;
  96
  97         if (!op->status) {
  98                 moving_init(io);
  99
 100                 io->bio.bio.bi_iter.bi_sector = KEY_START(&io->w->key);
 101                 op->write_prio          = 1;
 102                 op->bio                 = &io->bio.bio;
 103
 104                 op->writeback           = KEY_DIRTY(&io->w->key);
 105                 op->csum                = KEY_CSUM(&io->w->key);
 106
 107                 bkey_copy(&op->replace_key, &io->w->key);
 108                 op->replace             = true;
 109
 110                 closure_call(&op->cl, bch_data_insert, NULL, cl);
 111         }
 112
 113         continue_at(cl, write_moving_finish, op->wq);
 114 }
 115
 116 static CLOSURE_CALLBACK(read_moving_submit)
 117 {
 118         closure_type(io, struct moving_io, cl);
 119         struct bio *bio = &io->bio.bio;
 120
 121         bch_submit_bbio(bio, io->op.c, &io->w->key, 0);
 122
 123         continue_at(cl, write_moving, io->op.wq);
 124 }
 125
 126 static void read_moving(struct cache_set *c)
 127 {
 128         struct keybuf_key *w;
 129         struct moving_io *io;
 130         struct bio *bio;
 131         struct closure cl;
 132
 133         closure_init_stack(&cl);
 134
 135         /* XXX: if we error, background writeback could stall indefinitely */
 136
 137         while (!test_bit(CACHE_SET_STOPPING, &c->flags)) {
 138                 w = bch_keybuf_next_rescan(c, &c->moving_gc_keys,
 139                                            &MAX_KEY, moving_pred);
 140                 if (!w)
 141                         break;
 142
 143                 if (ptr_stale(c, &w->key, 0)) {
 144                         bch_keybuf_del(&c->moving_gc_keys, w);
 145                         continue;
 146                 }
 147
 148                 io = kzalloc(struct_size(io, bio.bio.bi_inline_vecs,
 149                                          DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS)),
 150                              GFP_KERNEL);
 151                 if (!io)
 152                         goto err;
 153
 154                 w->private      = io;
 155                 io->w           = w;
 156                 io->op.inode    = KEY_INODE(&w->key);
 157                 io->op.c        = c;
 158                 io->op.wq       = c->moving_gc_wq;
 159
 160                 moving_init(io);
 161                 bio = &io->bio.bio;
 162
 163                 bio->bi_opf = REQ_OP_READ;
 164                 bio->bi_end_io  = read_moving_endio;
 165
 166                 if (bch_bio_alloc_pages(bio, GFP_KERNEL))
 167                         goto err;
 168
 169                 trace_bcache_gc_copy(&w->key);
 170
 171                 down(&c->moving_in_flight);
 172                 closure_call(&io->cl, read_moving_submit, NULL, &cl);
 173         }
 174
 175         if (0) {
 176 err:            if (!IS_ERR_OR_NULL(w->private))
 177                         kfree(w->private);
 178
 179                 bch_keybuf_del(&c->moving_gc_keys, w);
 180         }
 181
 182         closure_sync(&cl);
 183 }
 184
 185 static bool bucket_cmp(struct bucket *l, struct bucket *r)
 186 {
 187         return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
 188 }
 189
 190 static unsigned int bucket_heap_top(struct cache *ca)
 191 {
 192         struct bucket *b;
 193
 194         return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0;
 195 }
 196
 197 void bch_moving_gc(struct cache_set *c)
 198 {
 199         struct cache *ca = c->cache;
 200         struct bucket *b;
 201         unsigned long sectors_to_move, reserve_sectors;
 202
 203         if (!c->copy_gc_enabled)
 204                 return;
 205
 206         mutex_lock(&c->bucket_lock);
 207
 208         sectors_to_move = 0;
 209         reserve_sectors = ca->sb.bucket_size *
 210                              fifo_used(&ca->free[RESERVE_MOVINGGC]);
 211
 212         ca->heap.used = 0;
 213
 214         for_each_bucket(b, ca) {
 215                 if (GC_MARK(b) == GC_MARK_METADATA ||
 216                     !GC_SECTORS_USED(b) ||
 217                     GC_SECTORS_USED(b) == ca->sb.bucket_size ||
 218                     atomic_read(&b->pin))
 219                         continue;
 220
 221                 if (!heap_full(&ca->heap)) {
 222                         sectors_to_move += GC_SECTORS_USED(b);
 223                         heap_add(&ca->heap, b, bucket_cmp);
 224                 } else if (bucket_cmp(b, heap_peek(&ca->heap))) {
 225                         sectors_to_move -= bucket_heap_top(ca);
 226                         sectors_to_move += GC_SECTORS_USED(b);
 227
 228                         ca->heap.data[0] = b;
 229                         heap_sift(&ca->heap, 0, bucket_cmp);
 230                 }
 231         }
 232
 233         while (sectors_to_move > reserve_sectors) {
 234                 heap_pop(&ca->heap, b, bucket_cmp);
 235                 sectors_to_move -= GC_SECTORS_USED(b);
 236         }
 237
 238         while (heap_pop(&ca->heap, b, bucket_cmp))
 239                 SET_GC_MOVE(b, 1);
 240
 241         mutex_unlock(&c->bucket_lock);
 242
 243         c->moving_gc_keys.last_scanned = ZERO_KEY;
 244
 245         read_moving(c);
 246 }
 247
 248 void bch_moving_init_cache_set(struct cache_set *c)
 249 {
 250         bch_keybuf_init(&c->moving_gc_keys);
 251         sema_init(&c->moving_in_flight, 64);
 252 }