kernel/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 #include <trace/events/bcache.h>
  13
  14 struct moving_io {
  15         struct closure          cl;
  16         struct keybuf_key       *w;
  17         struct data_insert_op   op;
  18         struct bbio             bio;
  19 };
  20
  21 static bool moving_pred(struct keybuf *buf, struct bkey *k)
  22 {
  23         struct cache_set *c = container_of(buf, struct cache_set,
  24                                            moving_gc_keys);
  25         unsigned i;
  26
  27         for (i = 0; i < KEY_PTRS(k); i++)
  28                 if (ptr_available(c, k, i) &&
  29                     GC_MOVE(PTR_BUCKET(c, k, i)))
  30                         return true;
  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, 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, cl);
  46         struct bio *bio = &io->bio.bio;
  47         struct bio_vec *bv;
  48         int i;
  49
  50         bio_for_each_segment_all(bv, bio, i)
  51                 __free_page(bv->bv_page);
  52
  53         if (io->op.replace_collision)
  54                 trace_bcache_gc_copy_collision(&io->w->key);
  55
  56         bch_keybuf_del(&io->op.c->moving_gc_keys, io->w);
  57
  58         up(&io->op.c->moving_in_flight);
  59
  60         closure_return_with_destructor(cl, moving_io_destructor);
  61 }
  62
  63 static void read_moving_endio(struct bio *bio, int error)
  64 {
  65         struct bbio *b = container_of(bio, struct bbio, bio);
  66         struct moving_io *io = container_of(bio->bi_private,
  67                                             struct moving_io, cl);
  68
  69         if (error)
  70                 io->op.error = error;
  71         else if (!KEY_DIRTY(&b->key) &&
  72                  ptr_stale(io->op.c, &b->key, 0)) {
  73                 io->op.error = -EINTR;
  74         }
  75
  76         bch_bbio_endio(io->op.c, bio, error, "reading data to move");
  77 }
  78
  79 static void moving_init(struct moving_io *io)
  80 {
  81         struct bio *bio = &io->bio.bio;
  82
  83         bio_init(bio);
  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_max_vecs        = DIV_ROUND_UP(KEY_SIZE(&io->w->key),
  89                                                PAGE_SECTORS);
  90         bio->bi_private         = &io->cl;
  91         bio->bi_io_vec          = bio->bi_inline_vecs;
  92         bch_bio_map(bio, NULL);
  93 }
  94
  95 static void write_moving(struct closure *cl)
  96 {
  97         struct moving_io *io = container_of(cl, struct moving_io, cl);
  98         struct data_insert_op *op = &io->op;
  99
 100         if (!op->error) {
 101                 moving_init(io);
 102
 103                 io->bio.bio.bi_iter.bi_sector = KEY_START(&io->w->key);
 104                 op->write_prio          = 1;
 105                 op->bio                 = &io->bio.bio;
 106
 107                 op->writeback           = KEY_DIRTY(&io->w->key);
 108                 op->csum                = KEY_CSUM(&io->w->key);
 109
 110                 bkey_copy(&op->replace_key, &io->w->key);
 111                 op->replace             = true;
 112
 113                 closure_call(&op->cl, bch_data_insert, NULL, cl);
 114         }
 115
 116         continue_at(cl, write_moving_finish, op->wq);
 117 }
 118
 119 static void read_moving_submit(struct closure *cl)
 120 {
 121         struct moving_io *io = container_of(cl, struct moving_io, cl);
 122         struct bio *bio = &io->bio.bio;
 123
 124         bch_submit_bbio(bio, io->op.c, &io->w->key, 0);
 125
 126         continue_at(cl, write_moving, io->op.wq);
 127 }
 128
 129 static void read_moving(struct cache_set *c)
 130 {
 131         struct keybuf_key *w;
 132         struct moving_io *io;
 133         struct bio *bio;
 134         struct closure cl;
 135
 136         closure_init_stack(&cl);
 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,
 142                                            &MAX_KEY, moving_pred);
 143                 if (!w)
 144                         break;
 145
 146                 if (ptr_stale(c, &w->key, 0)) {
 147                         bch_keybuf_del(&c->moving_gc_keys, w);
 148                         continue;
 149                 }
 150
 151                 io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec)
 152                              * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
 153                              GFP_KERNEL);
 154                 if (!io)
 155                         goto err;
 156
 157                 w->private      = io;
 158                 io->w           = w;
 159                 io->op.inode    = KEY_INODE(&w->key);
 160                 io->op.c        = c;
 161                 io->op.wq       = c->moving_gc_wq;
 162
 163                 moving_init(io);
 164                 bio = &io->bio.bio;
 165
 166                 bio->bi_rw      = READ;
 167                 bio->bi_end_io  = read_moving_endio;
 168
 169                 if (bio_alloc_pages(bio, GFP_KERNEL))
 170                         goto err;
 171
 172                 trace_bcache_gc_copy(&w->key);
 173
 174                 down(&c->moving_in_flight);
 175                 closure_call(&io->cl, read_moving_submit, NULL, &cl);
 176         }
 177
 178         if (0) {
 179 err:            if (!IS_ERR_OR_NULL(w->private))
 180                         kfree(w->private);
 181
 182                 bch_keybuf_del(&c->moving_gc_keys, w);
 183         }
 184
 185         closure_sync(&cl);
 186 }
 187
 188 static bool bucket_cmp(struct bucket *l, struct bucket *r)
 189 {
 190         return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
 191 }
 192
 193 static unsigned bucket_heap_top(struct cache *ca)
 194 {
 195         struct bucket *b;
 196         return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0;
 197 }
 198
 199 void bch_moving_gc(struct cache_set *c)
 200 {
 201         struct cache *ca;
 202         struct bucket *b;
 203         unsigned i;
 204
 205         if (!c->copy_gc_enabled)
 206                 return;
 207
 208         mutex_lock(&c->bucket_lock);
 209
 210         for_each_cache(ca, c, i) {
 211                 unsigned sectors_to_move = 0;
 212                 unsigned reserve_sectors = ca->sb.bucket_size *
 213                         fifo_used(&ca->free[RESERVE_MOVINGGC]);
 214
 215                 ca->heap.used = 0;
 216
 217                 for_each_bucket(b, ca) {
 218                         if (GC_MARK(b) == GC_MARK_METADATA ||
 219                             !GC_SECTORS_USED(b) ||
 220                             GC_SECTORS_USED(b) == ca->sb.bucket_size ||
 221                             atomic_read(&b->pin))
 222                                 continue;
 223
 224                         if (!heap_full(&ca->heap)) {
 225                                 sectors_to_move += GC_SECTORS_USED(b);
 226                                 heap_add(&ca->heap, b, bucket_cmp);
 227                         } else if (bucket_cmp(b, heap_peek(&ca->heap))) {
 228                                 sectors_to_move -= bucket_heap_top(ca);
 229                                 sectors_to_move += GC_SECTORS_USED(b);
 230
 231                                 ca->heap.data[0] = b;
 232                                 heap_sift(&ca->heap, 0, bucket_cmp);
 233                         }
 234                 }
 235
 236                 while (sectors_to_move > reserve_sectors) {
 237                         heap_pop(&ca->heap, b, bucket_cmp);
 238                         sectors_to_move -= GC_SECTORS_USED(b);
 239                 }
 240
 241                 while (heap_pop(&ca->heap, b, bucket_cmp))
 242                         SET_GC_MOVE(b, 1);
 243         }
 244
 245         mutex_unlock(&c->bucket_lock);
 246
 247         c->moving_gc_keys.last_scanned = ZERO_KEY;
 248
 249         read_moving(c);
 250 }
 251
 252 void bch_moving_init_cache_set(struct cache_set *c)
 253 {
 254         bch_keybuf_init(&c->moving_gc_keys);
 255         sema_init(&c->moving_in_flight, 64);
 256 }