block/blk-cgroup.c

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