| 1 | /* |
| 2 | * Common Block IO controller cgroup interface |
| 3 | * |
| 4 | * Based on ideas and code from CFQ, CFS and BFQ: |
| 5 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
| 6 | * |
| 7 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> |
| 8 | * Paolo Valente <paolo.valente@unimore.it> |
| 9 | * |
| 10 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> |
| 11 | * Nauman Rafique <nauman@google.com> |
| 12 | */ |
| 13 | #include <linux/ioprio.h> |
| 14 | #include <linux/kdev_t.h> |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/err.h> |
| 17 | #include <linux/blkdev.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/genhd.h> |
| 20 | #include <linux/delay.h> |
| 21 | #include <linux/atomic.h> |
| 22 | #include "blk-cgroup.h" |
| 23 | #include "blk.h" |
| 24 | |
| 25 | #define MAX_KEY_LEN 100 |
| 26 | |
| 27 | static DEFINE_MUTEX(blkcg_pol_mutex); |
| 28 | static DEFINE_MUTEX(all_q_mutex); |
| 29 | static LIST_HEAD(all_q_list); |
| 30 | |
| 31 | struct blkio_cgroup blkio_root_cgroup = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT }; |
| 32 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
| 33 | |
| 34 | static struct blkio_policy_type *blkio_policy[BLKCG_MAX_POLS]; |
| 35 | |
| 36 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
| 37 | { |
| 38 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), |
| 39 | struct blkio_cgroup, css); |
| 40 | } |
| 41 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
| 42 | |
| 43 | static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
| 44 | { |
| 45 | return container_of(task_subsys_state(tsk, blkio_subsys_id), |
| 46 | struct blkio_cgroup, css); |
| 47 | } |
| 48 | |
| 49 | struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio) |
| 50 | { |
| 51 | if (bio && bio->bi_css) |
| 52 | return container_of(bio->bi_css, struct blkio_cgroup, css); |
| 53 | return task_blkio_cgroup(current); |
| 54 | } |
| 55 | EXPORT_SYMBOL_GPL(bio_blkio_cgroup); |
| 56 | |
| 57 | /** |
| 58 | * blkg_free - free a blkg |
| 59 | * @blkg: blkg to free |
| 60 | * |
| 61 | * Free @blkg which may be partially allocated. |
| 62 | */ |
| 63 | static void blkg_free(struct blkio_group *blkg) |
| 64 | { |
| 65 | int i; |
| 66 | |
| 67 | if (!blkg) |
| 68 | return; |
| 69 | |
| 70 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 71 | struct blkio_policy_type *pol = blkio_policy[i]; |
| 72 | struct blkg_policy_data *pd = blkg->pd[i]; |
| 73 | |
| 74 | if (!pd) |
| 75 | continue; |
| 76 | |
| 77 | if (pol && pol->ops.blkio_exit_group_fn) |
| 78 | pol->ops.blkio_exit_group_fn(blkg); |
| 79 | |
| 80 | kfree(pd); |
| 81 | } |
| 82 | |
| 83 | kfree(blkg); |
| 84 | } |
| 85 | |
| 86 | /** |
| 87 | * blkg_alloc - allocate a blkg |
| 88 | * @blkcg: block cgroup the new blkg is associated with |
| 89 | * @q: request_queue the new blkg is associated with |
| 90 | * |
| 91 | * Allocate a new blkg assocating @blkcg and @q. |
| 92 | */ |
| 93 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, |
| 94 | struct request_queue *q) |
| 95 | { |
| 96 | struct blkio_group *blkg; |
| 97 | int i; |
| 98 | |
| 99 | /* alloc and init base part */ |
| 100 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); |
| 101 | if (!blkg) |
| 102 | return NULL; |
| 103 | |
| 104 | blkg->q = q; |
| 105 | INIT_LIST_HEAD(&blkg->q_node); |
| 106 | blkg->blkcg = blkcg; |
| 107 | blkg->refcnt = 1; |
| 108 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
| 109 | |
| 110 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 111 | struct blkio_policy_type *pol = blkio_policy[i]; |
| 112 | struct blkg_policy_data *pd; |
| 113 | |
| 114 | if (!pol) |
| 115 | continue; |
| 116 | |
| 117 | /* alloc per-policy data and attach it to blkg */ |
| 118 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, |
| 119 | q->node); |
| 120 | if (!pd) { |
| 121 | blkg_free(blkg); |
| 122 | return NULL; |
| 123 | } |
| 124 | |
| 125 | blkg->pd[i] = pd; |
| 126 | pd->blkg = blkg; |
| 127 | } |
| 128 | |
| 129 | /* invoke per-policy init */ |
| 130 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 131 | struct blkio_policy_type *pol = blkio_policy[i]; |
| 132 | |
| 133 | if (pol) |
| 134 | pol->ops.blkio_init_group_fn(blkg); |
| 135 | } |
| 136 | |
| 137 | return blkg; |
| 138 | } |
| 139 | |
| 140 | static struct blkio_group *__blkg_lookup(struct blkio_cgroup *blkcg, |
| 141 | struct request_queue *q) |
| 142 | { |
| 143 | struct blkio_group *blkg; |
| 144 | struct hlist_node *n; |
| 145 | |
| 146 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
| 147 | if (blkg->q == q) |
| 148 | return blkg; |
| 149 | return NULL; |
| 150 | } |
| 151 | |
| 152 | /** |
| 153 | * blkg_lookup - lookup blkg for the specified blkcg - q pair |
| 154 | * @blkcg: blkcg of interest |
| 155 | * @q: request_queue of interest |
| 156 | * |
| 157 | * Lookup blkg for the @blkcg - @q pair. This function should be called |
| 158 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing |
| 159 | * - see blk_queue_bypass_start() for details. |
| 160 | */ |
| 161 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
| 162 | struct request_queue *q) |
| 163 | { |
| 164 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 165 | |
| 166 | if (unlikely(blk_queue_bypass(q))) |
| 167 | return NULL; |
| 168 | return __blkg_lookup(blkcg, q); |
| 169 | } |
| 170 | EXPORT_SYMBOL_GPL(blkg_lookup); |
| 171 | |
| 172 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
| 173 | struct request_queue *q, |
| 174 | bool for_root) |
| 175 | __releases(q->queue_lock) __acquires(q->queue_lock) |
| 176 | { |
| 177 | struct blkio_group *blkg; |
| 178 | |
| 179 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 180 | lockdep_assert_held(q->queue_lock); |
| 181 | |
| 182 | /* |
| 183 | * This could be the first entry point of blkcg implementation and |
| 184 | * we shouldn't allow anything to go through for a bypassing queue. |
| 185 | */ |
| 186 | if (unlikely(blk_queue_bypass(q)) && !for_root) |
| 187 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); |
| 188 | |
| 189 | blkg = __blkg_lookup(blkcg, q); |
| 190 | if (blkg) |
| 191 | return blkg; |
| 192 | |
| 193 | /* blkg holds a reference to blkcg */ |
| 194 | if (!css_tryget(&blkcg->css)) |
| 195 | return ERR_PTR(-EINVAL); |
| 196 | |
| 197 | /* |
| 198 | * Allocate and initialize. |
| 199 | */ |
| 200 | blkg = blkg_alloc(blkcg, q); |
| 201 | |
| 202 | /* did alloc fail? */ |
| 203 | if (unlikely(!blkg)) { |
| 204 | blkg = ERR_PTR(-ENOMEM); |
| 205 | goto out; |
| 206 | } |
| 207 | |
| 208 | /* insert */ |
| 209 | spin_lock(&blkcg->lock); |
| 210 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
| 211 | list_add(&blkg->q_node, &q->blkg_list); |
| 212 | spin_unlock(&blkcg->lock); |
| 213 | out: |
| 214 | return blkg; |
| 215 | } |
| 216 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
| 217 | |
| 218 | static void blkg_destroy(struct blkio_group *blkg) |
| 219 | { |
| 220 | struct request_queue *q = blkg->q; |
| 221 | struct blkio_cgroup *blkcg = blkg->blkcg; |
| 222 | |
| 223 | lockdep_assert_held(q->queue_lock); |
| 224 | lockdep_assert_held(&blkcg->lock); |
| 225 | |
| 226 | /* Something wrong if we are trying to remove same group twice */ |
| 227 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
| 228 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
| 229 | list_del_init(&blkg->q_node); |
| 230 | hlist_del_init_rcu(&blkg->blkcg_node); |
| 231 | |
| 232 | /* |
| 233 | * Put the reference taken at the time of creation so that when all |
| 234 | * queues are gone, group can be destroyed. |
| 235 | */ |
| 236 | blkg_put(blkg); |
| 237 | } |
| 238 | |
| 239 | /* |
| 240 | * XXX: This updates blkg policy data in-place for root blkg, which is |
| 241 | * necessary across elevator switch and policy registration as root blkgs |
| 242 | * aren't shot down. This broken and racy implementation is temporary. |
| 243 | * Eventually, blkg shoot down will be replaced by proper in-place update. |
| 244 | */ |
| 245 | void update_root_blkg_pd(struct request_queue *q, |
| 246 | const struct blkio_policy_type *pol) |
| 247 | { |
| 248 | struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q); |
| 249 | struct blkg_policy_data *pd; |
| 250 | |
| 251 | if (!blkg) |
| 252 | return; |
| 253 | |
| 254 | kfree(blkg->pd[pol->plid]); |
| 255 | blkg->pd[pol->plid] = NULL; |
| 256 | |
| 257 | if (!pol) |
| 258 | return; |
| 259 | |
| 260 | pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL); |
| 261 | WARN_ON_ONCE(!pd); |
| 262 | |
| 263 | blkg->pd[pol->plid] = pd; |
| 264 | pd->blkg = blkg; |
| 265 | pol->ops.blkio_init_group_fn(blkg); |
| 266 | } |
| 267 | EXPORT_SYMBOL_GPL(update_root_blkg_pd); |
| 268 | |
| 269 | /** |
| 270 | * blkg_destroy_all - destroy all blkgs associated with a request_queue |
| 271 | * @q: request_queue of interest |
| 272 | * @destroy_root: whether to destroy root blkg or not |
| 273 | * |
| 274 | * Destroy blkgs associated with @q. If @destroy_root is %true, all are |
| 275 | * destroyed; otherwise, root blkg is left alone. |
| 276 | */ |
| 277 | void blkg_destroy_all(struct request_queue *q, bool destroy_root) |
| 278 | { |
| 279 | struct blkio_group *blkg, *n; |
| 280 | |
| 281 | spin_lock_irq(q->queue_lock); |
| 282 | |
| 283 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
| 284 | struct blkio_cgroup *blkcg = blkg->blkcg; |
| 285 | |
| 286 | /* skip root? */ |
| 287 | if (!destroy_root && blkg->blkcg == &blkio_root_cgroup) |
| 288 | continue; |
| 289 | |
| 290 | spin_lock(&blkcg->lock); |
| 291 | blkg_destroy(blkg); |
| 292 | spin_unlock(&blkcg->lock); |
| 293 | } |
| 294 | |
| 295 | spin_unlock_irq(q->queue_lock); |
| 296 | } |
| 297 | EXPORT_SYMBOL_GPL(blkg_destroy_all); |
| 298 | |
| 299 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
| 300 | { |
| 301 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); |
| 302 | } |
| 303 | |
| 304 | void __blkg_release(struct blkio_group *blkg) |
| 305 | { |
| 306 | /* release the extra blkcg reference this blkg has been holding */ |
| 307 | css_put(&blkg->blkcg->css); |
| 308 | |
| 309 | /* |
| 310 | * A group is freed in rcu manner. But having an rcu lock does not |
| 311 | * mean that one can access all the fields of blkg and assume these |
| 312 | * are valid. For example, don't try to follow throtl_data and |
| 313 | * request queue links. |
| 314 | * |
| 315 | * Having a reference to blkg under an rcu allows acess to only |
| 316 | * values local to groups like group stats and group rate limits |
| 317 | */ |
| 318 | call_rcu(&blkg->rcu_head, blkg_rcu_free); |
| 319 | } |
| 320 | EXPORT_SYMBOL_GPL(__blkg_release); |
| 321 | |
| 322 | static int |
| 323 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
| 324 | { |
| 325 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
| 326 | struct blkio_group *blkg; |
| 327 | struct hlist_node *n; |
| 328 | int i; |
| 329 | |
| 330 | mutex_lock(&blkcg_pol_mutex); |
| 331 | spin_lock_irq(&blkcg->lock); |
| 332 | |
| 333 | /* |
| 334 | * Note that stat reset is racy - it doesn't synchronize against |
| 335 | * stat updates. This is a debug feature which shouldn't exist |
| 336 | * anyway. If you get hit by a race, retry. |
| 337 | */ |
| 338 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
| 339 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 340 | struct blkio_policy_type *pol = blkio_policy[i]; |
| 341 | |
| 342 | if (pol && pol->ops.blkio_reset_group_stats_fn) |
| 343 | pol->ops.blkio_reset_group_stats_fn(blkg); |
| 344 | } |
| 345 | } |
| 346 | |
| 347 | spin_unlock_irq(&blkcg->lock); |
| 348 | mutex_unlock(&blkcg_pol_mutex); |
| 349 | return 0; |
| 350 | } |
| 351 | |
| 352 | static const char *blkg_dev_name(struct blkio_group *blkg) |
| 353 | { |
| 354 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
| 355 | if (blkg->q->backing_dev_info.dev) |
| 356 | return dev_name(blkg->q->backing_dev_info.dev); |
| 357 | return NULL; |
| 358 | } |
| 359 | |
| 360 | /** |
| 361 | * blkcg_print_blkgs - helper for printing per-blkg data |
| 362 | * @sf: seq_file to print to |
| 363 | * @blkcg: blkcg of interest |
| 364 | * @prfill: fill function to print out a blkg |
| 365 | * @pol: policy in question |
| 366 | * @data: data to be passed to @prfill |
| 367 | * @show_total: to print out sum of prfill return values or not |
| 368 | * |
| 369 | * This function invokes @prfill on each blkg of @blkcg if pd for the |
| 370 | * policy specified by @pol exists. @prfill is invoked with @sf, the |
| 371 | * policy data and @data. If @show_total is %true, the sum of the return |
| 372 | * values from @prfill is printed with "Total" label at the end. |
| 373 | * |
| 374 | * This is to be used to construct print functions for |
| 375 | * cftype->read_seq_string method. |
| 376 | */ |
| 377 | void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg, |
| 378 | u64 (*prfill)(struct seq_file *, void *, int), |
| 379 | const struct blkio_policy_type *pol, int data, |
| 380 | bool show_total) |
| 381 | { |
| 382 | struct blkio_group *blkg; |
| 383 | struct hlist_node *n; |
| 384 | u64 total = 0; |
| 385 | |
| 386 | spin_lock_irq(&blkcg->lock); |
| 387 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) |
| 388 | if (blkg->pd[pol->plid]) |
| 389 | total += prfill(sf, blkg->pd[pol->plid]->pdata, data); |
| 390 | spin_unlock_irq(&blkcg->lock); |
| 391 | |
| 392 | if (show_total) |
| 393 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); |
| 394 | } |
| 395 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
| 396 | |
| 397 | /** |
| 398 | * __blkg_prfill_u64 - prfill helper for a single u64 value |
| 399 | * @sf: seq_file to print to |
| 400 | * @pdata: policy private data of interest |
| 401 | * @v: value to print |
| 402 | * |
| 403 | * Print @v to @sf for the device assocaited with @pdata. |
| 404 | */ |
| 405 | u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v) |
| 406 | { |
| 407 | const char *dname = blkg_dev_name(pdata_to_blkg(pdata)); |
| 408 | |
| 409 | if (!dname) |
| 410 | return 0; |
| 411 | |
| 412 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); |
| 413 | return v; |
| 414 | } |
| 415 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
| 416 | |
| 417 | /** |
| 418 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat |
| 419 | * @sf: seq_file to print to |
| 420 | * @pdata: policy private data of interest |
| 421 | * @rwstat: rwstat to print |
| 422 | * |
| 423 | * Print @rwstat to @sf for the device assocaited with @pdata. |
| 424 | */ |
| 425 | u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata, |
| 426 | const struct blkg_rwstat *rwstat) |
| 427 | { |
| 428 | static const char *rwstr[] = { |
| 429 | [BLKG_RWSTAT_READ] = "Read", |
| 430 | [BLKG_RWSTAT_WRITE] = "Write", |
| 431 | [BLKG_RWSTAT_SYNC] = "Sync", |
| 432 | [BLKG_RWSTAT_ASYNC] = "Async", |
| 433 | }; |
| 434 | const char *dname = blkg_dev_name(pdata_to_blkg(pdata)); |
| 435 | u64 v; |
| 436 | int i; |
| 437 | |
| 438 | if (!dname) |
| 439 | return 0; |
| 440 | |
| 441 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
| 442 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], |
| 443 | (unsigned long long)rwstat->cnt[i]); |
| 444 | |
| 445 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; |
| 446 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
| 447 | return v; |
| 448 | } |
| 449 | |
| 450 | /** |
| 451 | * blkg_prfill_stat - prfill callback for blkg_stat |
| 452 | * @sf: seq_file to print to |
| 453 | * @pdata: policy private data of interest |
| 454 | * @off: offset to the blkg_stat in @pdata |
| 455 | * |
| 456 | * prfill callback for printing a blkg_stat. |
| 457 | */ |
| 458 | u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off) |
| 459 | { |
| 460 | return __blkg_prfill_u64(sf, pdata, blkg_stat_read(pdata + off)); |
| 461 | } |
| 462 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
| 463 | |
| 464 | /** |
| 465 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat |
| 466 | * @sf: seq_file to print to |
| 467 | * @pdata: policy private data of interest |
| 468 | * @off: offset to the blkg_rwstat in @pdata |
| 469 | * |
| 470 | * prfill callback for printing a blkg_rwstat. |
| 471 | */ |
| 472 | u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off) |
| 473 | { |
| 474 | struct blkg_rwstat rwstat = blkg_rwstat_read(pdata + off); |
| 475 | |
| 476 | return __blkg_prfill_rwstat(sf, pdata, &rwstat); |
| 477 | } |
| 478 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
| 479 | |
| 480 | /** |
| 481 | * blkg_conf_prep - parse and prepare for per-blkg config update |
| 482 | * @blkcg: target block cgroup |
| 483 | * @pol: target policy |
| 484 | * @input: input string |
| 485 | * @ctx: blkg_conf_ctx to be filled |
| 486 | * |
| 487 | * Parse per-blkg config update from @input and initialize @ctx with the |
| 488 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new |
| 489 | * value. This function returns with RCU read lock and queue lock held and |
| 490 | * must be paired with blkg_conf_finish(). |
| 491 | */ |
| 492 | int blkg_conf_prep(struct blkio_cgroup *blkcg, |
| 493 | const struct blkio_policy_type *pol, const char *input, |
| 494 | struct blkg_conf_ctx *ctx) |
| 495 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
| 496 | { |
| 497 | struct gendisk *disk; |
| 498 | struct blkio_group *blkg; |
| 499 | unsigned int major, minor; |
| 500 | unsigned long long v; |
| 501 | int part, ret; |
| 502 | |
| 503 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
| 504 | return -EINVAL; |
| 505 | |
| 506 | disk = get_gendisk(MKDEV(major, minor), &part); |
| 507 | if (!disk || part) |
| 508 | return -EINVAL; |
| 509 | |
| 510 | rcu_read_lock(); |
| 511 | spin_lock_irq(disk->queue->queue_lock); |
| 512 | |
| 513 | blkg = blkg_lookup_create(blkcg, disk->queue, false); |
| 514 | |
| 515 | if (IS_ERR(blkg)) { |
| 516 | ret = PTR_ERR(blkg); |
| 517 | rcu_read_unlock(); |
| 518 | spin_unlock_irq(disk->queue->queue_lock); |
| 519 | put_disk(disk); |
| 520 | /* |
| 521 | * If queue was bypassing, we should retry. Do so after a |
| 522 | * short msleep(). It isn't strictly necessary but queue |
| 523 | * can be bypassing for some time and it's always nice to |
| 524 | * avoid busy looping. |
| 525 | */ |
| 526 | if (ret == -EBUSY) { |
| 527 | msleep(10); |
| 528 | ret = restart_syscall(); |
| 529 | } |
| 530 | return ret; |
| 531 | } |
| 532 | |
| 533 | ctx->disk = disk; |
| 534 | ctx->blkg = blkg; |
| 535 | ctx->v = v; |
| 536 | return 0; |
| 537 | } |
| 538 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
| 539 | |
| 540 | /** |
| 541 | * blkg_conf_finish - finish up per-blkg config update |
| 542 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() |
| 543 | * |
| 544 | * Finish up after per-blkg config update. This function must be paired |
| 545 | * with blkg_conf_prep(). |
| 546 | */ |
| 547 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
| 548 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
| 549 | { |
| 550 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
| 551 | rcu_read_unlock(); |
| 552 | put_disk(ctx->disk); |
| 553 | } |
| 554 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
| 555 | |
| 556 | struct cftype blkio_files[] = { |
| 557 | { |
| 558 | .name = "reset_stats", |
| 559 | .write_u64 = blkiocg_reset_stats, |
| 560 | }, |
| 561 | { } /* terminate */ |
| 562 | }; |
| 563 | |
| 564 | /** |
| 565 | * blkiocg_pre_destroy - cgroup pre_destroy callback |
| 566 | * @cgroup: cgroup of interest |
| 567 | * |
| 568 | * This function is called when @cgroup is about to go away and responsible |
| 569 | * for shooting down all blkgs associated with @cgroup. blkgs should be |
| 570 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
| 571 | * inside q lock, this function performs reverse double lock dancing. |
| 572 | * |
| 573 | * This is the blkcg counterpart of ioc_release_fn(). |
| 574 | */ |
| 575 | static int blkiocg_pre_destroy(struct cgroup *cgroup) |
| 576 | { |
| 577 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
| 578 | |
| 579 | spin_lock_irq(&blkcg->lock); |
| 580 | |
| 581 | while (!hlist_empty(&blkcg->blkg_list)) { |
| 582 | struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first, |
| 583 | struct blkio_group, blkcg_node); |
| 584 | struct request_queue *q = blkg->q; |
| 585 | |
| 586 | if (spin_trylock(q->queue_lock)) { |
| 587 | blkg_destroy(blkg); |
| 588 | spin_unlock(q->queue_lock); |
| 589 | } else { |
| 590 | spin_unlock_irq(&blkcg->lock); |
| 591 | cpu_relax(); |
| 592 | spin_lock_irq(&blkcg->lock); |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | spin_unlock_irq(&blkcg->lock); |
| 597 | return 0; |
| 598 | } |
| 599 | |
| 600 | static void blkiocg_destroy(struct cgroup *cgroup) |
| 601 | { |
| 602 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
| 603 | |
| 604 | if (blkcg != &blkio_root_cgroup) |
| 605 | kfree(blkcg); |
| 606 | } |
| 607 | |
| 608 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup) |
| 609 | { |
| 610 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
| 611 | struct blkio_cgroup *blkcg; |
| 612 | struct cgroup *parent = cgroup->parent; |
| 613 | |
| 614 | if (!parent) { |
| 615 | blkcg = &blkio_root_cgroup; |
| 616 | goto done; |
| 617 | } |
| 618 | |
| 619 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
| 620 | if (!blkcg) |
| 621 | return ERR_PTR(-ENOMEM); |
| 622 | |
| 623 | blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT; |
| 624 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
| 625 | done: |
| 626 | spin_lock_init(&blkcg->lock); |
| 627 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
| 628 | |
| 629 | return &blkcg->css; |
| 630 | } |
| 631 | |
| 632 | /** |
| 633 | * blkcg_init_queue - initialize blkcg part of request queue |
| 634 | * @q: request_queue to initialize |
| 635 | * |
| 636 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg |
| 637 | * part of new request_queue @q. |
| 638 | * |
| 639 | * RETURNS: |
| 640 | * 0 on success, -errno on failure. |
| 641 | */ |
| 642 | int blkcg_init_queue(struct request_queue *q) |
| 643 | { |
| 644 | int ret; |
| 645 | |
| 646 | might_sleep(); |
| 647 | |
| 648 | ret = blk_throtl_init(q); |
| 649 | if (ret) |
| 650 | return ret; |
| 651 | |
| 652 | mutex_lock(&all_q_mutex); |
| 653 | INIT_LIST_HEAD(&q->all_q_node); |
| 654 | list_add_tail(&q->all_q_node, &all_q_list); |
| 655 | mutex_unlock(&all_q_mutex); |
| 656 | |
| 657 | return 0; |
| 658 | } |
| 659 | |
| 660 | /** |
| 661 | * blkcg_drain_queue - drain blkcg part of request_queue |
| 662 | * @q: request_queue to drain |
| 663 | * |
| 664 | * Called from blk_drain_queue(). Responsible for draining blkcg part. |
| 665 | */ |
| 666 | void blkcg_drain_queue(struct request_queue *q) |
| 667 | { |
| 668 | lockdep_assert_held(q->queue_lock); |
| 669 | |
| 670 | blk_throtl_drain(q); |
| 671 | } |
| 672 | |
| 673 | /** |
| 674 | * blkcg_exit_queue - exit and release blkcg part of request_queue |
| 675 | * @q: request_queue being released |
| 676 | * |
| 677 | * Called from blk_release_queue(). Responsible for exiting blkcg part. |
| 678 | */ |
| 679 | void blkcg_exit_queue(struct request_queue *q) |
| 680 | { |
| 681 | mutex_lock(&all_q_mutex); |
| 682 | list_del_init(&q->all_q_node); |
| 683 | mutex_unlock(&all_q_mutex); |
| 684 | |
| 685 | blkg_destroy_all(q, true); |
| 686 | |
| 687 | blk_throtl_exit(q); |
| 688 | } |
| 689 | |
| 690 | /* |
| 691 | * We cannot support shared io contexts, as we have no mean to support |
| 692 | * two tasks with the same ioc in two different groups without major rework |
| 693 | * of the main cic data structures. For now we allow a task to change |
| 694 | * its cgroup only if it's the only owner of its ioc. |
| 695 | */ |
| 696 | static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
| 697 | { |
| 698 | struct task_struct *task; |
| 699 | struct io_context *ioc; |
| 700 | int ret = 0; |
| 701 | |
| 702 | /* task_lock() is needed to avoid races with exit_io_context() */ |
| 703 | cgroup_taskset_for_each(task, cgrp, tset) { |
| 704 | task_lock(task); |
| 705 | ioc = task->io_context; |
| 706 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) |
| 707 | ret = -EINVAL; |
| 708 | task_unlock(task); |
| 709 | if (ret) |
| 710 | break; |
| 711 | } |
| 712 | return ret; |
| 713 | } |
| 714 | |
| 715 | static void blkcg_bypass_start(void) |
| 716 | __acquires(&all_q_mutex) |
| 717 | { |
| 718 | struct request_queue *q; |
| 719 | |
| 720 | mutex_lock(&all_q_mutex); |
| 721 | |
| 722 | list_for_each_entry(q, &all_q_list, all_q_node) { |
| 723 | blk_queue_bypass_start(q); |
| 724 | blkg_destroy_all(q, false); |
| 725 | } |
| 726 | } |
| 727 | |
| 728 | static void blkcg_bypass_end(void) |
| 729 | __releases(&all_q_mutex) |
| 730 | { |
| 731 | struct request_queue *q; |
| 732 | |
| 733 | list_for_each_entry(q, &all_q_list, all_q_node) |
| 734 | blk_queue_bypass_end(q); |
| 735 | |
| 736 | mutex_unlock(&all_q_mutex); |
| 737 | } |
| 738 | |
| 739 | struct cgroup_subsys blkio_subsys = { |
| 740 | .name = "blkio", |
| 741 | .create = blkiocg_create, |
| 742 | .can_attach = blkiocg_can_attach, |
| 743 | .pre_destroy = blkiocg_pre_destroy, |
| 744 | .destroy = blkiocg_destroy, |
| 745 | .subsys_id = blkio_subsys_id, |
| 746 | .base_cftypes = blkio_files, |
| 747 | .module = THIS_MODULE, |
| 748 | }; |
| 749 | EXPORT_SYMBOL_GPL(blkio_subsys); |
| 750 | |
| 751 | /** |
| 752 | * blkio_policy_register - register a blkcg policy |
| 753 | * @blkiop: blkcg policy to register |
| 754 | * |
| 755 | * Register @blkiop with blkcg core. Might sleep and @blkiop may be |
| 756 | * modified on successful registration. Returns 0 on success and -errno on |
| 757 | * failure. |
| 758 | */ |
| 759 | int blkio_policy_register(struct blkio_policy_type *blkiop) |
| 760 | { |
| 761 | struct request_queue *q; |
| 762 | int i, ret; |
| 763 | |
| 764 | mutex_lock(&blkcg_pol_mutex); |
| 765 | |
| 766 | /* find an empty slot */ |
| 767 | ret = -ENOSPC; |
| 768 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
| 769 | if (!blkio_policy[i]) |
| 770 | break; |
| 771 | if (i >= BLKCG_MAX_POLS) |
| 772 | goto out_unlock; |
| 773 | |
| 774 | /* register and update blkgs */ |
| 775 | blkiop->plid = i; |
| 776 | blkio_policy[i] = blkiop; |
| 777 | |
| 778 | blkcg_bypass_start(); |
| 779 | list_for_each_entry(q, &all_q_list, all_q_node) |
| 780 | update_root_blkg_pd(q, blkiop); |
| 781 | blkcg_bypass_end(); |
| 782 | |
| 783 | /* everything is in place, add intf files for the new policy */ |
| 784 | if (blkiop->cftypes) |
| 785 | WARN_ON(cgroup_add_cftypes(&blkio_subsys, blkiop->cftypes)); |
| 786 | ret = 0; |
| 787 | out_unlock: |
| 788 | mutex_unlock(&blkcg_pol_mutex); |
| 789 | return ret; |
| 790 | } |
| 791 | EXPORT_SYMBOL_GPL(blkio_policy_register); |
| 792 | |
| 793 | /** |
| 794 | * blkiop_policy_unregister - unregister a blkcg policy |
| 795 | * @blkiop: blkcg policy to unregister |
| 796 | * |
| 797 | * Undo blkio_policy_register(@blkiop). Might sleep. |
| 798 | */ |
| 799 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) |
| 800 | { |
| 801 | struct request_queue *q; |
| 802 | |
| 803 | mutex_lock(&blkcg_pol_mutex); |
| 804 | |
| 805 | if (WARN_ON(blkio_policy[blkiop->plid] != blkiop)) |
| 806 | goto out_unlock; |
| 807 | |
| 808 | /* kill the intf files first */ |
| 809 | if (blkiop->cftypes) |
| 810 | cgroup_rm_cftypes(&blkio_subsys, blkiop->cftypes); |
| 811 | |
| 812 | /* unregister and update blkgs */ |
| 813 | blkio_policy[blkiop->plid] = NULL; |
| 814 | |
| 815 | blkcg_bypass_start(); |
| 816 | list_for_each_entry(q, &all_q_list, all_q_node) |
| 817 | update_root_blkg_pd(q, blkiop); |
| 818 | blkcg_bypass_end(); |
| 819 | out_unlock: |
| 820 | mutex_unlock(&blkcg_pol_mutex); |
| 821 | } |
| 822 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |