Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 1 | #include "sched.h" |
| 2 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 3 | #include <linux/proc_fs.h> |
| 4 | #include <linux/seq_file.h> |
| 5 | #include <linux/kallsyms.h> |
| 6 | #include <linux/utsname.h> |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7 | #include <linux/security.h> |
| 8 | #include <linux/export.h> |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 9 | |
| 10 | unsigned int __read_mostly sysctl_sched_autogroup_enabled = 1; |
| 11 | static struct autogroup autogroup_default; |
| 12 | static atomic_t autogroup_seq_nr; |
| 13 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 14 | void __init autogroup_init(struct task_struct *init_task) |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 15 | { |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 16 | autogroup_default.tg = &root_task_group; |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 17 | kref_init(&autogroup_default.kref); |
| 18 | init_rwsem(&autogroup_default.lock); |
| 19 | init_task->signal->autogroup = &autogroup_default; |
| 20 | } |
| 21 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 22 | void autogroup_free(struct task_group *tg) |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 23 | { |
| 24 | kfree(tg->autogroup); |
| 25 | } |
| 26 | |
| 27 | static inline void autogroup_destroy(struct kref *kref) |
| 28 | { |
| 29 | struct autogroup *ag = container_of(kref, struct autogroup, kref); |
| 30 | |
Mike Galbraith | f449377 | 2011-01-13 04:54:50 +0100 | [diff] [blame] | 31 | #ifdef CONFIG_RT_GROUP_SCHED |
| 32 | /* We've redirected RT tasks to the root task group... */ |
| 33 | ag->tg->rt_se = NULL; |
| 34 | ag->tg->rt_rq = NULL; |
| 35 | #endif |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 36 | sched_offline_group(ag->tg); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 37 | sched_destroy_group(ag->tg); |
| 38 | } |
| 39 | |
| 40 | static inline void autogroup_kref_put(struct autogroup *ag) |
| 41 | { |
| 42 | kref_put(&ag->kref, autogroup_destroy); |
| 43 | } |
| 44 | |
| 45 | static inline struct autogroup *autogroup_kref_get(struct autogroup *ag) |
| 46 | { |
| 47 | kref_get(&ag->kref); |
| 48 | return ag; |
| 49 | } |
| 50 | |
Mike Galbraith | 4f82198 | 2010-12-16 15:09:52 +0100 | [diff] [blame] | 51 | static inline struct autogroup *autogroup_task_get(struct task_struct *p) |
| 52 | { |
| 53 | struct autogroup *ag; |
| 54 | unsigned long flags; |
| 55 | |
| 56 | if (!lock_task_sighand(p, &flags)) |
| 57 | return autogroup_kref_get(&autogroup_default); |
| 58 | |
| 59 | ag = autogroup_kref_get(p->signal->autogroup); |
| 60 | unlock_task_sighand(p, &flags); |
| 61 | |
| 62 | return ag; |
| 63 | } |
| 64 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 65 | static inline struct autogroup *autogroup_create(void) |
| 66 | { |
| 67 | struct autogroup *ag = kzalloc(sizeof(*ag), GFP_KERNEL); |
| 68 | struct task_group *tg; |
| 69 | |
| 70 | if (!ag) |
| 71 | goto out_fail; |
| 72 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 73 | tg = sched_create_group(&root_task_group); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 74 | |
| 75 | if (IS_ERR(tg)) |
| 76 | goto out_free; |
| 77 | |
| 78 | kref_init(&ag->kref); |
| 79 | init_rwsem(&ag->lock); |
| 80 | ag->id = atomic_inc_return(&autogroup_seq_nr); |
| 81 | ag->tg = tg; |
Mike Galbraith | f449377 | 2011-01-13 04:54:50 +0100 | [diff] [blame] | 82 | #ifdef CONFIG_RT_GROUP_SCHED |
| 83 | /* |
| 84 | * Autogroup RT tasks are redirected to the root task group |
| 85 | * so we don't have to move tasks around upon policy change, |
| 86 | * or flail around trying to allocate bandwidth on the fly. |
| 87 | * A bandwidth exception in __sched_setscheduler() allows |
Peter Zijlstra | 1fe89e1 | 2015-02-09 11:53:18 +0100 | [diff] [blame] | 88 | * the policy change to proceed. |
Mike Galbraith | f449377 | 2011-01-13 04:54:50 +0100 | [diff] [blame] | 89 | */ |
| 90 | free_rt_sched_group(tg); |
| 91 | tg->rt_se = root_task_group.rt_se; |
| 92 | tg->rt_rq = root_task_group.rt_rq; |
| 93 | #endif |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 94 | tg->autogroup = ag; |
| 95 | |
Gerald Schaefer | 41261b6 | 2013-05-24 18:07:49 +0200 | [diff] [blame] | 96 | sched_online_group(tg, &root_task_group); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 97 | return ag; |
| 98 | |
| 99 | out_free: |
| 100 | kfree(ag); |
| 101 | out_fail: |
| 102 | if (printk_ratelimit()) { |
| 103 | printk(KERN_WARNING "autogroup_create: %s failure.\n", |
| 104 | ag ? "sched_create_group()" : "kmalloc()"); |
| 105 | } |
| 106 | |
| 107 | return autogroup_kref_get(&autogroup_default); |
| 108 | } |
| 109 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 110 | bool task_wants_autogroup(struct task_struct *p, struct task_group *tg) |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 111 | { |
| 112 | if (tg != &root_task_group) |
| 113 | return false; |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 114 | /* |
Oleg Nesterov | 18f649e | 2016-11-14 19:46:09 +0100 | [diff] [blame] | 115 | * If we race with autogroup_move_group() the caller can use the old |
| 116 | * value of signal->autogroup but in this case sched_move_task() will |
| 117 | * be called again before autogroup_kref_put(). |
Oleg Nesterov | 8e5bfa8 | 2016-11-14 19:46:12 +0100 | [diff] [blame] | 118 | * |
| 119 | * However, there is no way sched_autogroup_exit_task() could tell us |
| 120 | * to avoid autogroup->tg, so we abuse PF_EXITING flag for this case. |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 121 | */ |
Oleg Nesterov | 8e5bfa8 | 2016-11-14 19:46:12 +0100 | [diff] [blame] | 122 | if (p->flags & PF_EXITING) |
| 123 | return false; |
| 124 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 125 | return true; |
| 126 | } |
| 127 | |
Oleg Nesterov | 8e5bfa8 | 2016-11-14 19:46:12 +0100 | [diff] [blame] | 128 | void sched_autogroup_exit_task(struct task_struct *p) |
| 129 | { |
| 130 | /* |
| 131 | * We are going to call exit_notify() and autogroup_move_group() can't |
| 132 | * see this thread after that: we can no longer use signal->autogroup. |
| 133 | * See the PF_EXITING check in task_wants_autogroup(). |
| 134 | */ |
| 135 | sched_move_task(p); |
| 136 | } |
| 137 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 138 | static void |
| 139 | autogroup_move_group(struct task_struct *p, struct autogroup *ag) |
| 140 | { |
| 141 | struct autogroup *prev; |
| 142 | struct task_struct *t; |
| 143 | unsigned long flags; |
| 144 | |
| 145 | BUG_ON(!lock_task_sighand(p, &flags)); |
| 146 | |
| 147 | prev = p->signal->autogroup; |
| 148 | if (prev == ag) { |
| 149 | unlock_task_sighand(p, &flags); |
| 150 | return; |
| 151 | } |
| 152 | |
| 153 | p->signal->autogroup = autogroup_kref_get(ag); |
Oleg Nesterov | 18f649e | 2016-11-14 19:46:09 +0100 | [diff] [blame] | 154 | /* |
| 155 | * We can't avoid sched_move_task() after we changed signal->autogroup, |
| 156 | * this process can already run with task_group() == prev->tg or we can |
| 157 | * race with cgroup code which can read autogroup = prev under rq->lock. |
| 158 | * In the latter case for_each_thread() can not miss a migrating thread, |
| 159 | * cpu_cgroup_attach() must not be possible after cgroup_exit() and it |
| 160 | * can't be removed from thread list, we hold ->siglock. |
Oleg Nesterov | 8e5bfa8 | 2016-11-14 19:46:12 +0100 | [diff] [blame] | 161 | * |
| 162 | * If an exiting thread was already removed from thread list we rely on |
| 163 | * sched_autogroup_exit_task(). |
Oleg Nesterov | 18f649e | 2016-11-14 19:46:09 +0100 | [diff] [blame] | 164 | */ |
Oleg Nesterov | 5aface5 | 2014-08-13 21:20:03 +0200 | [diff] [blame] | 165 | for_each_thread(p, t) |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 166 | sched_move_task(t); |
Oleg Nesterov | 18f649e | 2016-11-14 19:46:09 +0100 | [diff] [blame] | 167 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 168 | unlock_task_sighand(p, &flags); |
| 169 | autogroup_kref_put(prev); |
| 170 | } |
| 171 | |
| 172 | /* Allocates GFP_KERNEL, cannot be called under any spinlock */ |
| 173 | void sched_autogroup_create_attach(struct task_struct *p) |
| 174 | { |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 175 | struct autogroup *ag = autogroup_create(); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 176 | |
| 177 | autogroup_move_group(p, ag); |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 178 | /* drop extra reference added by autogroup_create() */ |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 179 | autogroup_kref_put(ag); |
| 180 | } |
| 181 | EXPORT_SYMBOL(sched_autogroup_create_attach); |
| 182 | |
| 183 | /* Cannot be called under siglock. Currently has no users */ |
| 184 | void sched_autogroup_detach(struct task_struct *p) |
| 185 | { |
| 186 | autogroup_move_group(p, &autogroup_default); |
| 187 | } |
| 188 | EXPORT_SYMBOL(sched_autogroup_detach); |
| 189 | |
| 190 | void sched_autogroup_fork(struct signal_struct *sig) |
| 191 | { |
Mike Galbraith | 4f82198 | 2010-12-16 15:09:52 +0100 | [diff] [blame] | 192 | sig->autogroup = autogroup_task_get(current); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 193 | } |
| 194 | |
| 195 | void sched_autogroup_exit(struct signal_struct *sig) |
| 196 | { |
| 197 | autogroup_kref_put(sig->autogroup); |
| 198 | } |
| 199 | |
| 200 | static int __init setup_autogroup(char *str) |
| 201 | { |
| 202 | sysctl_sched_autogroup_enabled = 0; |
| 203 | |
| 204 | return 1; |
| 205 | } |
| 206 | |
| 207 | __setup("noautogroup", setup_autogroup); |
| 208 | |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 209 | #ifdef CONFIG_PROC_FS |
| 210 | |
| 211 | int proc_sched_autogroup_set_nice(struct task_struct *p, int nice) |
| 212 | { |
| 213 | static unsigned long next = INITIAL_JIFFIES; |
| 214 | struct autogroup *ag; |
Mike Galbraith | 83929cce | 2016-11-23 11:33:37 +0100 | [diff] [blame] | 215 | unsigned long shares; |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 216 | int err; |
| 217 | |
Dongsheng Yang | 75e45d5 | 2014-02-11 15:34:50 +0800 | [diff] [blame] | 218 | if (nice < MIN_NICE || nice > MAX_NICE) |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 219 | return -EINVAL; |
| 220 | |
| 221 | err = security_task_setnice(current, nice); |
| 222 | if (err) |
| 223 | return err; |
| 224 | |
| 225 | if (nice < 0 && !can_nice(current, nice)) |
| 226 | return -EPERM; |
| 227 | |
| 228 | /* this is a heavy operation taking global locks.. */ |
| 229 | if (!capable(CAP_SYS_ADMIN) && time_before(jiffies, next)) |
| 230 | return -EAGAIN; |
| 231 | |
| 232 | next = HZ / 10 + jiffies; |
| 233 | ag = autogroup_task_get(p); |
Mike Galbraith | 83929cce | 2016-11-23 11:33:37 +0100 | [diff] [blame] | 234 | shares = scale_load(sched_prio_to_weight[nice + 20]); |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 235 | |
| 236 | down_write(&ag->lock); |
Mike Galbraith | 83929cce | 2016-11-23 11:33:37 +0100 | [diff] [blame] | 237 | err = sched_group_set_shares(ag->tg, shares); |
Ingo Molnar | c1ad41f | 2012-12-11 10:23:45 +0100 | [diff] [blame] | 238 | if (!err) |
| 239 | ag->nice = nice; |
| 240 | up_write(&ag->lock); |
| 241 | |
| 242 | autogroup_kref_put(ag); |
| 243 | |
| 244 | return err; |
| 245 | } |
| 246 | |
| 247 | void proc_sched_autogroup_show_task(struct task_struct *p, struct seq_file *m) |
| 248 | { |
| 249 | struct autogroup *ag = autogroup_task_get(p); |
| 250 | |
| 251 | if (!task_group_is_autogroup(ag->tg)) |
| 252 | goto out; |
| 253 | |
| 254 | down_read(&ag->lock); |
| 255 | seq_printf(m, "/autogroup-%ld nice %d\n", ag->id, ag->nice); |
| 256 | up_read(&ag->lock); |
| 257 | |
| 258 | out: |
| 259 | autogroup_kref_put(ag); |
| 260 | } |
| 261 | #endif /* CONFIG_PROC_FS */ |
| 262 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 263 | #ifdef CONFIG_SCHED_DEBUG |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 264 | int autogroup_path(struct task_group *tg, char *buf, int buflen) |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 265 | { |
Mike Galbraith | 511f67a | 2011-02-22 15:02:00 +0100 | [diff] [blame] | 266 | if (!task_group_is_autogroup(tg)) |
Bharata B Rao | 8ecedd7 | 2011-01-11 15:42:57 +0530 | [diff] [blame] | 267 | return 0; |
| 268 | |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 269 | return snprintf(buf, buflen, "%s-%ld", "/autogroup", tg->autogroup->id); |
| 270 | } |
| 271 | #endif /* CONFIG_SCHED_DEBUG */ |