drivers/md/bcache/movinggc.c

   1 /*
   2  * Moving/copying garbage collector
   3  *
   4  * Copyright 2012 Google, Inc.
   5  */
   6
   7 #include "bcache.h"
   8 #include "btree.h"
   9 #include "debug.h"
  10 #include "request.h"
  11
  12 struct moving_io {
  13         struct keybuf_key       *w;
  14         struct search           s;
  15         struct bbio             bio;
  16 };
  17
  18 static bool moving_pred(struct keybuf *buf, struct bkey *k)
  19 {
  20         struct cache_set *c = container_of(buf, struct cache_set,
  21                                            moving_gc_keys);
  22         unsigned i;
  23
  24         for (i = 0; i < KEY_PTRS(k); i++) {
  25                 struct cache *ca = PTR_CACHE(c, k, i);
  26                 struct bucket *g = PTR_BUCKET(c, k, i);
  27
  28                 if (GC_SECTORS_USED(g) < ca->gc_move_threshold)
  29                         return true;
  30         }
  31
  32         return false;
  33 }
  34
  35 /* Moving GC - IO loop */
  36
  37 static void moving_io_destructor(struct closure *cl)
  38 {
  39         struct moving_io *io = container_of(cl, struct moving_io, s.cl);
  40         kfree(io);
  41 }
  42
  43 static void write_moving_finish(struct closure *cl)
  44 {
  45         struct moving_io *io = container_of(cl, struct moving_io, s.cl);
  46         struct bio *bio = &io->bio.bio;
  47         struct bio_vec *bv = bio_iovec_idx(bio, bio->bi_vcnt);
  48
  49         while (bv-- != bio->bi_io_vec)
  50                 __free_page(bv->bv_page);
  51
  52         pr_debug("%s %s", io->s.op.insert_collision
  53                  ? "collision moving" : "moved",
  54                  pkey(&io->w->key));
  55
  56         bch_keybuf_del(&io->s.op.c->moving_gc_keys, io->w);
  57
  58         atomic_dec_bug(&io->s.op.c->in_flight);
  59         closure_wake_up(&io->s.op.c->moving_gc_wait);
  60
  61         closure_return_with_destructor(cl, moving_io_destructor);
  62 }
  63
  64 static void read_moving_endio(struct bio *bio, int error)
  65 {
  66         struct moving_io *io = container_of(bio->bi_private,
  67                                             struct moving_io, s.cl);
  68
  69         if (error)
  70                 io->s.error = error;
  71
  72         bch_bbio_endio(io->s.op.c, bio, error, "reading data to move");
  73 }
  74
  75 static void moving_init(struct moving_io *io)
  76 {
  77         struct bio *bio = &io->bio.bio;
  78
  79         bio_init(bio);
  80         bio_get(bio);
  81         bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
  82
  83         bio->bi_size            = KEY_SIZE(&io->w->key) << 9;
  84         bio->bi_max_vecs        = DIV_ROUND_UP(KEY_SIZE(&io->w->key),
  85                                                PAGE_SECTORS);
  86         bio->bi_private         = &io->s.cl;
  87         bio->bi_io_vec          = bio->bi_inline_vecs;
  88         bch_bio_map(bio, NULL);
  89 }
  90
  91 static void write_moving(struct closure *cl)
  92 {
  93         struct search *s = container_of(cl, struct search, cl);
  94         struct moving_io *io = container_of(s, struct moving_io, s);
  95
  96         if (!s->error) {
  97                 trace_bcache_write_moving(&io->bio.bio);
  98
  99                 moving_init(io);
 100
 101                 io->bio.bio.bi_sector   = KEY_START(&io->w->key);
 102                 s->op.lock              = -1;
 103                 s->op.write_prio        = 1;
 104                 s->op.cache_bio         = &io->bio.bio;
 105
 106                 s->writeback            = KEY_DIRTY(&io->w->key);
 107                 s->op.csum              = KEY_CSUM(&io->w->key);
 108
 109                 s->op.type = BTREE_REPLACE;
 110                 bkey_copy(&s->op.replace, &io->w->key);
 111
 112                 closure_init(&s->op.cl, cl);
 113                 bch_insert_data(&s->op.cl);
 114         }
 115
 116         continue_at(cl, write_moving_finish, NULL);
 117 }
 118
 119 static void read_moving_submit(struct closure *cl)
 120 {
 121         struct search *s = container_of(cl, struct search, cl);
 122         struct moving_io *io = container_of(s, struct moving_io, s);
 123         struct bio *bio = &io->bio.bio;
 124
 125         trace_bcache_read_moving(bio);
 126         bch_submit_bbio(bio, s->op.c, &io->w->key, 0);
 127
 128         continue_at(cl, write_moving, bch_gc_wq);
 129 }
 130
 131 static void read_moving(struct closure *cl)
 132 {
 133         struct cache_set *c = container_of(cl, struct cache_set, moving_gc);
 134         struct keybuf_key *w;
 135         struct moving_io *io;
 136         struct bio *bio;
 137
 138         /* XXX: if we error, background writeback could stall indefinitely */
 139
 140         while (!test_bit(CACHE_SET_STOPPING, &c->flags)) {
 141                 w = bch_keybuf_next_rescan(c, &c->moving_gc_keys, &MAX_KEY);
 142                 if (!w)
 143                         break;
 144
 145                 io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec)
 146                              * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
 147                              GFP_KERNEL);
 148                 if (!io)
 149                         goto err;
 150
 151                 w->private      = io;
 152                 io->w           = w;
 153                 io->s.op.inode  = KEY_INODE(&w->key);
 154                 io->s.op.c      = c;
 155
 156                 moving_init(io);
 157                 bio = &io->bio.bio;
 158
 159                 bio->bi_rw      = READ;
 160                 bio->bi_end_io  = read_moving_endio;
 161
 162                 if (bch_bio_alloc_pages(bio, GFP_KERNEL))
 163                         goto err;
 164
 165                 pr_debug("%s", pkey(&w->key));
 166
 167                 closure_call(&io->s.cl, read_moving_submit, NULL, &c->gc.cl);
 168
 169                 if (atomic_inc_return(&c->in_flight) >= 64) {
 170                         closure_wait_event(&c->moving_gc_wait, cl,
 171                                            atomic_read(&c->in_flight) < 64);
 172                         continue_at(cl, read_moving, bch_gc_wq);
 173                 }
 174         }
 175
 176         if (0) {
 177 err:            if (!IS_ERR_OR_NULL(w->private))
 178                         kfree(w->private);
 179
 180                 bch_keybuf_del(&c->moving_gc_keys, w);
 181         }
 182
 183         closure_return(cl);
 184 }
 185
 186 static bool bucket_cmp(struct bucket *l, struct bucket *r)
 187 {
 188         return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
 189 }
 190
 191 static unsigned bucket_heap_top(struct cache *ca)
 192 {
 193         return GC_SECTORS_USED(heap_peek(&ca->heap));
 194 }
 195
 196 void bch_moving_gc(struct closure *cl)
 197 {
 198         struct cache_set *c = container_of(cl, struct cache_set, gc.cl);
 199         struct cache *ca;
 200         struct bucket *b;
 201         unsigned i;
 202
 203         if (!c->copy_gc_enabled)
 204                 closure_return(cl);
 205
 206         mutex_lock(&c->bucket_lock);
 207
 208         for_each_cache(ca, c, i) {
 209                 unsigned sectors_to_move = 0;
 210                 unsigned reserve_sectors = ca->sb.bucket_size *
 211                         min(fifo_used(&ca->free), ca->free.size / 2);
 212
 213                 ca->heap.used = 0;
 214
 215                 for_each_bucket(b, ca) {
 216                         if (!GC_SECTORS_USED(b))
 217                                 continue;
 218
 219                         if (!heap_full(&ca->heap)) {
 220                                 sectors_to_move += GC_SECTORS_USED(b);
 221                                 heap_add(&ca->heap, b, bucket_cmp);
 222                         } else if (bucket_cmp(b, heap_peek(&ca->heap))) {
 223                                 sectors_to_move -= bucket_heap_top(ca);
 224                                 sectors_to_move += GC_SECTORS_USED(b);
 225
 226                                 ca->heap.data[0] = b;
 227                                 heap_sift(&ca->heap, 0, bucket_cmp);
 228                         }
 229                 }
 230
 231                 while (sectors_to_move > reserve_sectors) {
 232                         heap_pop(&ca->heap, b, bucket_cmp);
 233                         sectors_to_move -= GC_SECTORS_USED(b);
 234                 }
 235
 236                 ca->gc_move_threshold = bucket_heap_top(ca);
 237
 238                 pr_debug("threshold %u", ca->gc_move_threshold);
 239         }
 240
 241         mutex_unlock(&c->bucket_lock);
 242
 243         c->moving_gc_keys.last_scanned = ZERO_KEY;
 244
 245         closure_init(&c->moving_gc, cl);
 246         read_moving(&c->moving_gc);
 247
 248         closure_return(cl);
 249 }
 250
 251 void bch_moving_init_cache_set(struct cache_set *c)
 252 {
 253         bch_keybuf_init(&c->moving_gc_keys, moving_pred);
 254 }