| 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/seq_file.h> |
| 15 | #include <linux/kdev_t.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/err.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include "blk-cgroup.h" |
| 20 | |
| 21 | static DEFINE_SPINLOCK(blkio_list_lock); |
| 22 | static LIST_HEAD(blkio_list); |
| 23 | |
| 24 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
| 25 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
| 26 | |
| 27 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
| 28 | struct cgroup *); |
| 29 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
| 30 | struct task_struct *, bool); |
| 31 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, |
| 32 | struct cgroup *, struct task_struct *, bool); |
| 33 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
| 34 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); |
| 35 | |
| 36 | struct cgroup_subsys blkio_subsys = { |
| 37 | .name = "blkio", |
| 38 | .create = blkiocg_create, |
| 39 | .can_attach = blkiocg_can_attach, |
| 40 | .attach = blkiocg_attach, |
| 41 | .destroy = blkiocg_destroy, |
| 42 | .populate = blkiocg_populate, |
| 43 | #ifdef CONFIG_BLK_CGROUP |
| 44 | /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */ |
| 45 | .subsys_id = blkio_subsys_id, |
| 46 | #endif |
| 47 | .use_id = 1, |
| 48 | .module = THIS_MODULE, |
| 49 | }; |
| 50 | EXPORT_SYMBOL_GPL(blkio_subsys); |
| 51 | |
| 52 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
| 53 | { |
| 54 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), |
| 55 | struct blkio_cgroup, css); |
| 56 | } |
| 57 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
| 58 | |
| 59 | void blkiocg_update_blkio_group_stats(struct blkio_group *blkg, |
| 60 | unsigned long time, unsigned long sectors) |
| 61 | { |
| 62 | blkg->time += time; |
| 63 | blkg->sectors += sectors; |
| 64 | } |
| 65 | EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_stats); |
| 66 | |
| 67 | void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
| 68 | struct blkio_group *blkg, void *key, dev_t dev) |
| 69 | { |
| 70 | unsigned long flags; |
| 71 | |
| 72 | spin_lock_irqsave(&blkcg->lock, flags); |
| 73 | rcu_assign_pointer(blkg->key, key); |
| 74 | blkg->blkcg_id = css_id(&blkcg->css); |
| 75 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
| 76 | spin_unlock_irqrestore(&blkcg->lock, flags); |
| 77 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 78 | /* Need to take css reference ? */ |
| 79 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
| 80 | #endif |
| 81 | blkg->dev = dev; |
| 82 | } |
| 83 | EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group); |
| 84 | |
| 85 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
| 86 | { |
| 87 | hlist_del_init_rcu(&blkg->blkcg_node); |
| 88 | blkg->blkcg_id = 0; |
| 89 | } |
| 90 | |
| 91 | /* |
| 92 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 |
| 93 | * indicating that blk_group was unhashed by the time we got to it. |
| 94 | */ |
| 95 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
| 96 | { |
| 97 | struct blkio_cgroup *blkcg; |
| 98 | unsigned long flags; |
| 99 | struct cgroup_subsys_state *css; |
| 100 | int ret = 1; |
| 101 | |
| 102 | rcu_read_lock(); |
| 103 | css = css_lookup(&blkio_subsys, blkg->blkcg_id); |
| 104 | if (!css) |
| 105 | goto out; |
| 106 | |
| 107 | blkcg = container_of(css, struct blkio_cgroup, css); |
| 108 | spin_lock_irqsave(&blkcg->lock, flags); |
| 109 | if (!hlist_unhashed(&blkg->blkcg_node)) { |
| 110 | __blkiocg_del_blkio_group(blkg); |
| 111 | ret = 0; |
| 112 | } |
| 113 | spin_unlock_irqrestore(&blkcg->lock, flags); |
| 114 | out: |
| 115 | rcu_read_unlock(); |
| 116 | return ret; |
| 117 | } |
| 118 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
| 119 | |
| 120 | /* called under rcu_read_lock(). */ |
| 121 | struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) |
| 122 | { |
| 123 | struct blkio_group *blkg; |
| 124 | struct hlist_node *n; |
| 125 | void *__key; |
| 126 | |
| 127 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { |
| 128 | __key = blkg->key; |
| 129 | if (__key == key) |
| 130 | return blkg; |
| 131 | } |
| 132 | |
| 133 | return NULL; |
| 134 | } |
| 135 | EXPORT_SYMBOL_GPL(blkiocg_lookup_group); |
| 136 | |
| 137 | #define SHOW_FUNCTION(__VAR) \ |
| 138 | static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \ |
| 139 | struct cftype *cftype) \ |
| 140 | { \ |
| 141 | struct blkio_cgroup *blkcg; \ |
| 142 | \ |
| 143 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ |
| 144 | return (u64)blkcg->__VAR; \ |
| 145 | } |
| 146 | |
| 147 | SHOW_FUNCTION(weight); |
| 148 | #undef SHOW_FUNCTION |
| 149 | |
| 150 | static int |
| 151 | blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
| 152 | { |
| 153 | struct blkio_cgroup *blkcg; |
| 154 | struct blkio_group *blkg; |
| 155 | struct hlist_node *n; |
| 156 | struct blkio_policy_type *blkiop; |
| 157 | |
| 158 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) |
| 159 | return -EINVAL; |
| 160 | |
| 161 | blkcg = cgroup_to_blkio_cgroup(cgroup); |
| 162 | spin_lock(&blkio_list_lock); |
| 163 | spin_lock_irq(&blkcg->lock); |
| 164 | blkcg->weight = (unsigned int)val; |
| 165 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
| 166 | list_for_each_entry(blkiop, &blkio_list, list) |
| 167 | blkiop->ops.blkio_update_group_weight_fn(blkg, |
| 168 | blkcg->weight); |
| 169 | } |
| 170 | spin_unlock_irq(&blkcg->lock); |
| 171 | spin_unlock(&blkio_list_lock); |
| 172 | return 0; |
| 173 | } |
| 174 | |
| 175 | #define SHOW_FUNCTION_PER_GROUP(__VAR) \ |
| 176 | static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \ |
| 177 | struct cftype *cftype, struct seq_file *m) \ |
| 178 | { \ |
| 179 | struct blkio_cgroup *blkcg; \ |
| 180 | struct blkio_group *blkg; \ |
| 181 | struct hlist_node *n; \ |
| 182 | \ |
| 183 | if (!cgroup_lock_live_group(cgroup)) \ |
| 184 | return -ENODEV; \ |
| 185 | \ |
| 186 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ |
| 187 | rcu_read_lock(); \ |
| 188 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\ |
| 189 | if (blkg->dev) \ |
| 190 | seq_printf(m, "%u:%u %lu\n", MAJOR(blkg->dev), \ |
| 191 | MINOR(blkg->dev), blkg->__VAR); \ |
| 192 | } \ |
| 193 | rcu_read_unlock(); \ |
| 194 | cgroup_unlock(); \ |
| 195 | return 0; \ |
| 196 | } |
| 197 | |
| 198 | SHOW_FUNCTION_PER_GROUP(time); |
| 199 | SHOW_FUNCTION_PER_GROUP(sectors); |
| 200 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 201 | SHOW_FUNCTION_PER_GROUP(dequeue); |
| 202 | #endif |
| 203 | #undef SHOW_FUNCTION_PER_GROUP |
| 204 | |
| 205 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 206 | void blkiocg_update_blkio_group_dequeue_stats(struct blkio_group *blkg, |
| 207 | unsigned long dequeue) |
| 208 | { |
| 209 | blkg->dequeue += dequeue; |
| 210 | } |
| 211 | EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_dequeue_stats); |
| 212 | #endif |
| 213 | |
| 214 | struct cftype blkio_files[] = { |
| 215 | { |
| 216 | .name = "weight", |
| 217 | .read_u64 = blkiocg_weight_read, |
| 218 | .write_u64 = blkiocg_weight_write, |
| 219 | }, |
| 220 | { |
| 221 | .name = "time", |
| 222 | .read_seq_string = blkiocg_time_read, |
| 223 | }, |
| 224 | { |
| 225 | .name = "sectors", |
| 226 | .read_seq_string = blkiocg_sectors_read, |
| 227 | }, |
| 228 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 229 | { |
| 230 | .name = "dequeue", |
| 231 | .read_seq_string = blkiocg_dequeue_read, |
| 232 | }, |
| 233 | #endif |
| 234 | }; |
| 235 | |
| 236 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) |
| 237 | { |
| 238 | return cgroup_add_files(cgroup, subsys, blkio_files, |
| 239 | ARRAY_SIZE(blkio_files)); |
| 240 | } |
| 241 | |
| 242 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) |
| 243 | { |
| 244 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
| 245 | unsigned long flags; |
| 246 | struct blkio_group *blkg; |
| 247 | void *key; |
| 248 | struct blkio_policy_type *blkiop; |
| 249 | |
| 250 | rcu_read_lock(); |
| 251 | remove_entry: |
| 252 | spin_lock_irqsave(&blkcg->lock, flags); |
| 253 | |
| 254 | if (hlist_empty(&blkcg->blkg_list)) { |
| 255 | spin_unlock_irqrestore(&blkcg->lock, flags); |
| 256 | goto done; |
| 257 | } |
| 258 | |
| 259 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
| 260 | blkcg_node); |
| 261 | key = rcu_dereference(blkg->key); |
| 262 | __blkiocg_del_blkio_group(blkg); |
| 263 | |
| 264 | spin_unlock_irqrestore(&blkcg->lock, flags); |
| 265 | |
| 266 | /* |
| 267 | * This blkio_group is being unlinked as associated cgroup is going |
| 268 | * away. Let all the IO controlling policies know about this event. |
| 269 | * |
| 270 | * Currently this is static call to one io controlling policy. Once |
| 271 | * we have more policies in place, we need some dynamic registration |
| 272 | * of callback function. |
| 273 | */ |
| 274 | spin_lock(&blkio_list_lock); |
| 275 | list_for_each_entry(blkiop, &blkio_list, list) |
| 276 | blkiop->ops.blkio_unlink_group_fn(key, blkg); |
| 277 | spin_unlock(&blkio_list_lock); |
| 278 | goto remove_entry; |
| 279 | done: |
| 280 | free_css_id(&blkio_subsys, &blkcg->css); |
| 281 | rcu_read_unlock(); |
| 282 | if (blkcg != &blkio_root_cgroup) |
| 283 | kfree(blkcg); |
| 284 | } |
| 285 | |
| 286 | static struct cgroup_subsys_state * |
| 287 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) |
| 288 | { |
| 289 | struct blkio_cgroup *blkcg, *parent_blkcg; |
| 290 | |
| 291 | if (!cgroup->parent) { |
| 292 | blkcg = &blkio_root_cgroup; |
| 293 | goto done; |
| 294 | } |
| 295 | |
| 296 | /* Currently we do not support hierarchy deeper than two level (0,1) */ |
| 297 | parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent); |
| 298 | if (css_depth(&parent_blkcg->css) > 0) |
| 299 | return ERR_PTR(-EINVAL); |
| 300 | |
| 301 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
| 302 | if (!blkcg) |
| 303 | return ERR_PTR(-ENOMEM); |
| 304 | |
| 305 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; |
| 306 | done: |
| 307 | spin_lock_init(&blkcg->lock); |
| 308 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
| 309 | |
| 310 | return &blkcg->css; |
| 311 | } |
| 312 | |
| 313 | /* |
| 314 | * We cannot support shared io contexts, as we have no mean to support |
| 315 | * two tasks with the same ioc in two different groups without major rework |
| 316 | * of the main cic data structures. For now we allow a task to change |
| 317 | * its cgroup only if it's the only owner of its ioc. |
| 318 | */ |
| 319 | static int blkiocg_can_attach(struct cgroup_subsys *subsys, |
| 320 | struct cgroup *cgroup, struct task_struct *tsk, |
| 321 | bool threadgroup) |
| 322 | { |
| 323 | struct io_context *ioc; |
| 324 | int ret = 0; |
| 325 | |
| 326 | /* task_lock() is needed to avoid races with exit_io_context() */ |
| 327 | task_lock(tsk); |
| 328 | ioc = tsk->io_context; |
| 329 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) |
| 330 | ret = -EINVAL; |
| 331 | task_unlock(tsk); |
| 332 | |
| 333 | return ret; |
| 334 | } |
| 335 | |
| 336 | static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup, |
| 337 | struct cgroup *prev, struct task_struct *tsk, |
| 338 | bool threadgroup) |
| 339 | { |
| 340 | struct io_context *ioc; |
| 341 | |
| 342 | task_lock(tsk); |
| 343 | ioc = tsk->io_context; |
| 344 | if (ioc) |
| 345 | ioc->cgroup_changed = 1; |
| 346 | task_unlock(tsk); |
| 347 | } |
| 348 | |
| 349 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
| 350 | { |
| 351 | spin_lock(&blkio_list_lock); |
| 352 | list_add_tail(&blkiop->list, &blkio_list); |
| 353 | spin_unlock(&blkio_list_lock); |
| 354 | } |
| 355 | EXPORT_SYMBOL_GPL(blkio_policy_register); |
| 356 | |
| 357 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) |
| 358 | { |
| 359 | spin_lock(&blkio_list_lock); |
| 360 | list_del_init(&blkiop->list); |
| 361 | spin_unlock(&blkio_list_lock); |
| 362 | } |
| 363 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |
| 364 | |
| 365 | static int __init init_cgroup_blkio(void) |
| 366 | { |
| 367 | return cgroup_load_subsys(&blkio_subsys); |
| 368 | } |
| 369 | |
| 370 | static void __exit exit_cgroup_blkio(void) |
| 371 | { |
| 372 | cgroup_unload_subsys(&blkio_subsys); |
| 373 | } |
| 374 | |
| 375 | module_init(init_cgroup_blkio); |
| 376 | module_exit(exit_cgroup_blkio); |
| 377 | MODULE_LICENSE("GPL"); |