1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_SCHED_SIGNAL_H
3#define _LINUX_SCHED_SIGNAL_H
4
5#include <linux/rculist.h>
6#include <linux/signal.h>
7#include <linux/sched.h>
8#include <linux/sched/jobctl.h>
9#include <linux/sched/task.h>
10#include <linux/cred.h>
11
12/*
13 * Types defining task->signal and task->sighand and APIs using them:
14 */
15
16struct sighand_struct {
17 atomic_t count;
18 struct k_sigaction action[_NSIG];
19 spinlock_t siglock;
20 wait_queue_head_t signalfd_wqh;
21};
22
23/*
24 * Per-process accounting stats:
25 */
26struct pacct_struct {
27 int ac_flag;
28 long ac_exitcode;
29 unsigned long ac_mem;
30 u64 ac_utime, ac_stime;
31 unsigned long ac_minflt, ac_majflt;
32};
33
34struct cpu_itimer {
35 u64 expires;
36 u64 incr;
37};
38
39/*
40 * This is the atomic variant of task_cputime, which can be used for
41 * storing and updating task_cputime statistics without locking.
42 */
43struct task_cputime_atomic {
44 atomic64_t utime;
45 atomic64_t stime;
46 atomic64_t sum_exec_runtime;
47};
48
49#define INIT_CPUTIME_ATOMIC \
50 (struct task_cputime_atomic) { \
51 .utime = ATOMIC64_INIT(0), \
52 .stime = ATOMIC64_INIT(0), \
53 .sum_exec_runtime = ATOMIC64_INIT(0), \
54 }
55/**
56 * struct thread_group_cputimer - thread group interval timer counts
57 * @cputime_atomic: atomic thread group interval timers.
58 * @running: true when there are timers running and
59 * @cputime_atomic receives updates.
60 * @checking_timer: true when a thread in the group is in the
61 * process of checking for thread group timers.
62 *
63 * This structure contains the version of task_cputime, above, that is
64 * used for thread group CPU timer calculations.
65 */
66struct thread_group_cputimer {
67 struct task_cputime_atomic cputime_atomic;
68 bool running;
69 bool checking_timer;
70};
71
72struct multiprocess_signals {
73 sigset_t signal;
74 struct hlist_node node;
75};
76
77/*
78 * NOTE! "signal_struct" does not have its own
79 * locking, because a shared signal_struct always
80 * implies a shared sighand_struct, so locking
81 * sighand_struct is always a proper superset of
82 * the locking of signal_struct.
83 */
84struct signal_struct {
85 atomic_t sigcnt;
86 atomic_t live;
87 int nr_threads;
88 struct list_head thread_head;
89
90 wait_queue_head_t wait_chldexit; /* for wait4() */
91
92 /* current thread group signal load-balancing target: */
93 struct task_struct *curr_target;
94
95 /* shared signal handling: */
96 struct sigpending shared_pending;
97
98 /* For collecting multiprocess signals during fork */
99 struct hlist_head multiprocess;
100
101 /* thread group exit support */
102 int group_exit_code;
103 /* overloaded:
104 * - notify group_exit_task when ->count is equal to notify_count
105 * - everyone except group_exit_task is stopped during signal delivery
106 * of fatal signals, group_exit_task processes the signal.
107 */
108 int notify_count;
109 struct task_struct *group_exit_task;
110
111 /* thread group stop support, overloads group_exit_code too */
112 int group_stop_count;
113 unsigned int flags; /* see SIGNAL_* flags below */
114
115 /*
116 * PR_SET_CHILD_SUBREAPER marks a process, like a service
117 * manager, to re-parent orphan (double-forking) child processes
118 * to this process instead of 'init'. The service manager is
119 * able to receive SIGCHLD signals and is able to investigate
120 * the process until it calls wait(). All children of this
121 * process will inherit a flag if they should look for a
122 * child_subreaper process at exit.
123 */
124 unsigned int is_child_subreaper:1;
125 unsigned int has_child_subreaper:1;
126
127#ifdef CONFIG_POSIX_TIMERS
128
129 /* POSIX.1b Interval Timers */
130 int posix_timer_id;
131 struct list_head posix_timers;
132
133 /* ITIMER_REAL timer for the process */
134 struct hrtimer real_timer;
135 ktime_t it_real_incr;
136
137 /*
138 * ITIMER_PROF and ITIMER_VIRTUAL timers for the process, we use
139 * CPUCLOCK_PROF and CPUCLOCK_VIRT for indexing array as these
140 * values are defined to 0 and 1 respectively
141 */
142 struct cpu_itimer it[2];
143
144 /*
145 * Thread group totals for process CPU timers.
146 * See thread_group_cputimer(), et al, for details.
147 */
148 struct thread_group_cputimer cputimer;
149
150 /* Earliest-expiration cache. */
151 struct task_cputime cputime_expires;
152
153 struct list_head cpu_timers[3];
154
155#endif
156
157 /* PID/PID hash table linkage. */
158 struct pid *pids[PIDTYPE_MAX];
159
160#ifdef CONFIG_NO_HZ_FULL
161 atomic_t tick_dep_mask;
162#endif
163
164 struct pid *tty_old_pgrp;
165
166 /* boolean value for session group leader */
167 int leader;
168
169 struct tty_struct *tty; /* NULL if no tty */
170
171#ifdef CONFIG_SCHED_AUTOGROUP
172 struct autogroup *autogroup;
173#endif
174 /*
175 * Cumulative resource counters for dead threads in the group,
176 * and for reaped dead child processes forked by this group.
177 * Live threads maintain their own counters and add to these
178 * in __exit_signal, except for the group leader.
179 */
180 seqlock_t stats_lock;
181 u64 utime, stime, cutime, cstime;
182 u64 gtime;
183 u64 cgtime;
184 struct prev_cputime prev_cputime;
185 unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
186 unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
187 unsigned long inblock, oublock, cinblock, coublock;
188 unsigned long maxrss, cmaxrss;
189 struct task_io_accounting ioac;
190
191 /*
192 * Cumulative ns of schedule CPU time fo dead threads in the
193 * group, not including a zombie group leader, (This only differs
194 * from jiffies_to_ns(utime + stime) if sched_clock uses something
195 * other than jiffies.)
196 */
197 unsigned long long sum_sched_runtime;
198
199 /*
200 * We don't bother to synchronize most readers of this at all,
201 * because there is no reader checking a limit that actually needs
202 * to get both rlim_cur and rlim_max atomically, and either one
203 * alone is a single word that can safely be read normally.
204 * getrlimit/setrlimit use task_lock(current->group_leader) to
205 * protect this instead of the siglock, because they really
206 * have no need to disable irqs.
207 */
208 struct rlimit rlim[RLIM_NLIMITS];
209
210#ifdef CONFIG_BSD_PROCESS_ACCT
211 struct pacct_struct pacct; /* per-process accounting information */
212#endif
213#ifdef CONFIG_TASKSTATS
214 struct taskstats *stats;
215#endif
216#ifdef CONFIG_AUDIT
217 unsigned audit_tty;
218 struct tty_audit_buf *tty_audit_buf;
219#endif
220
221 /*
222 * Thread is the potential origin of an oom condition; kill first on
223 * oom
224 */
225 bool oom_flag_origin;
226 short oom_score_adj; /* OOM kill score adjustment */
227 short oom_score_adj_min; /* OOM kill score adjustment min value.
228 * Only settable by CAP_SYS_RESOURCE. */
229 struct mm_struct *oom_mm; /* recorded mm when the thread group got
230 * killed by the oom killer */
231
232 struct mutex cred_guard_mutex; /* guard against foreign influences on
233 * credential calculations
234 * (notably. ptrace) */
235} __randomize_layout;
236
237/*
238 * Bits in flags field of signal_struct.
239 */
240#define SIGNAL_STOP_STOPPED 0x00000001 /* job control stop in effect */
241#define SIGNAL_STOP_CONTINUED 0x00000002 /* SIGCONT since WCONTINUED reap */
242#define SIGNAL_GROUP_EXIT 0x00000004 /* group exit in progress */
243#define SIGNAL_GROUP_COREDUMP 0x00000008 /* coredump in progress */
244/*
245 * Pending notifications to parent.
246 */
247#define SIGNAL_CLD_STOPPED 0x00000010
248#define SIGNAL_CLD_CONTINUED 0x00000020
249#define SIGNAL_CLD_MASK (SIGNAL_CLD_STOPPED|SIGNAL_CLD_CONTINUED)
250
251#define SIGNAL_UNKILLABLE 0x00000040 /* for init: ignore fatal signals */
252
253#define SIGNAL_STOP_MASK (SIGNAL_CLD_MASK | SIGNAL_STOP_STOPPED | \
254 SIGNAL_STOP_CONTINUED)
255
256static inline void signal_set_stop_flags(struct signal_struct *sig,
257 unsigned int flags)
258{
259 WARN_ON(sig->flags & (SIGNAL_GROUP_EXIT|SIGNAL_GROUP_COREDUMP));
260 sig->flags = (sig->flags & ~SIGNAL_STOP_MASK) | flags;
261}
262
263/* If true, all threads except ->group_exit_task have pending SIGKILL */
264static inline int signal_group_exit(const struct signal_struct *sig)
265{
266 return (sig->flags & SIGNAL_GROUP_EXIT) ||
267 (sig->group_exit_task != NULL);
268}
269
270extern void flush_signals(struct task_struct *);
271extern void ignore_signals(struct task_struct *);
272extern void flush_signal_handlers(struct task_struct *, int force_default);
273extern int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info);
274
275static inline int kernel_dequeue_signal(siginfo_t *info)
276{
277 struct task_struct *tsk = current;
278 siginfo_t __info;
279 int ret;
280
281 spin_lock_irq(&tsk->sighand->siglock);
282 ret = dequeue_signal(tsk, &tsk->blocked, info ?: &__info);
283 spin_unlock_irq(&tsk->sighand->siglock);
284
285 return ret;
286}
287
288static inline void kernel_signal_stop(void)
289{
290 spin_lock_irq(&current->sighand->siglock);
291 if (current->jobctl & JOBCTL_STOP_DEQUEUED)
292 set_special_state(TASK_STOPPED);
293 spin_unlock_irq(&current->sighand->siglock);
294
295 schedule();
296}
297#ifdef __ARCH_SI_TRAPNO
298# define ___ARCH_SI_TRAPNO(_a1) , _a1
299#else
300# define ___ARCH_SI_TRAPNO(_a1)
301#endif
302#ifdef __ia64__
303# define ___ARCH_SI_IA64(_a1, _a2, _a3) , _a1, _a2, _a3
304#else
305# define ___ARCH_SI_IA64(_a1, _a2, _a3)
306#endif
307
308int force_sig_fault(int sig, int code, void __user *addr
309 ___ARCH_SI_TRAPNO(int trapno)
310 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
311 , struct task_struct *t);
312int send_sig_fault(int sig, int code, void __user *addr
313 ___ARCH_SI_TRAPNO(int trapno)
314 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
315 , struct task_struct *t);
316
317int force_sig_mceerr(int code, void __user *, short, struct task_struct *);
318int send_sig_mceerr(int code, void __user *, short, struct task_struct *);
319
320int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper);
321int force_sig_pkuerr(void __user *addr, u32 pkey);
322
323int force_sig_ptrace_errno_trap(int errno, void __user *addr);
324
325extern int send_sig_info(int, struct siginfo *, struct task_struct *);
326extern int force_sigsegv(int, struct task_struct *);
327extern int force_sig_info(int, struct siginfo *, struct task_struct *);
328extern int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp);
329extern int kill_pid_info(int sig, struct siginfo *info, struct pid *pid);
330extern int kill_pid_info_as_cred(int, struct siginfo *, struct pid *,
331 const struct cred *);
332extern int kill_pgrp(struct pid *pid, int sig, int priv);
333extern int kill_pid(struct pid *pid, int sig, int priv);
334extern __must_check bool do_notify_parent(struct task_struct *, int);
335extern void __wake_up_parent(struct task_struct *p, struct task_struct *parent);
336extern void force_sig(int, struct task_struct *);
337extern int send_sig(int, struct task_struct *, int);
338extern int zap_other_threads(struct task_struct *p);
339extern struct sigqueue *sigqueue_alloc(void);
340extern void sigqueue_free(struct sigqueue *);
341extern int send_sigqueue(struct sigqueue *, struct pid *, enum pid_type);
342extern int do_sigaction(int, struct k_sigaction *, struct k_sigaction *);
343
344static inline int restart_syscall(void)
345{
346 set_tsk_thread_flag(current, TIF_SIGPENDING);
347 return -ERESTARTNOINTR;
348}
349
350static inline int signal_pending(struct task_struct *p)
351{
352 return unlikely(test_tsk_thread_flag(p,TIF_SIGPENDING));
353}
354
355static inline int __fatal_signal_pending(struct task_struct *p)
356{
357 return unlikely(sigismember(&p->pending.signal, SIGKILL));
358}
359
360static inline int fatal_signal_pending(struct task_struct *p)
361{
362 return signal_pending(p) && __fatal_signal_pending(p);
363}
364
365static inline int signal_pending_state(long state, struct task_struct *p)
366{
367 if (!(state & (TASK_INTERRUPTIBLE | TASK_WAKEKILL)))
368 return 0;
369 if (!signal_pending(p))
370 return 0;
371
372 return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
373}
374
375/*
376 * Reevaluate whether the task has signals pending delivery.
377 * Wake the task if so.
378 * This is required every time the blocked sigset_t changes.
379 * callers must hold sighand->siglock.
380 */
381extern void recalc_sigpending_and_wake(struct task_struct *t);
382extern void recalc_sigpending(void);
383extern void calculate_sigpending(void);
384
385extern void signal_wake_up_state(struct task_struct *t, unsigned int state);
386
387static inline void signal_wake_up(struct task_struct *t, bool resume)
388{
389 signal_wake_up_state(t, resume ? TASK_WAKEKILL : 0);
390}
391static inline void ptrace_signal_wake_up(struct task_struct *t, bool resume)
392{
393 signal_wake_up_state(t, resume ? __TASK_TRACED : 0);
394}
395
396void task_join_group_stop(struct task_struct *task);
397
398#ifdef TIF_RESTORE_SIGMASK
399/*
400 * Legacy restore_sigmask accessors. These are inefficient on
401 * SMP architectures because they require atomic operations.
402 */
403
404/**
405 * set_restore_sigmask() - make sure saved_sigmask processing gets done
406 *
407 * This sets TIF_RESTORE_SIGMASK and ensures that the arch signal code
408 * will run before returning to user mode, to process the flag. For
409 * all callers, TIF_SIGPENDING is already set or it's no harm to set
410 * it. TIF_RESTORE_SIGMASK need not be in the set of bits that the
411 * arch code will notice on return to user mode, in case those bits
412 * are scarce. We set TIF_SIGPENDING here to ensure that the arch
413 * signal code always gets run when TIF_RESTORE_SIGMASK is set.
414 */
415static inline void set_restore_sigmask(void)
416{
417 set_thread_flag(TIF_RESTORE_SIGMASK);
418 WARN_ON(!test_thread_flag(TIF_SIGPENDING));
419}
420static inline void clear_restore_sigmask(void)
421{
422 clear_thread_flag(TIF_RESTORE_SIGMASK);
423}
424static inline bool test_restore_sigmask(void)
425{
426 return test_thread_flag(TIF_RESTORE_SIGMASK);
427}
428static inline bool test_and_clear_restore_sigmask(void)
429{
430 return test_and_clear_thread_flag(TIF_RESTORE_SIGMASK);
431}
432
433#else /* TIF_RESTORE_SIGMASK */
434
435/* Higher-quality implementation, used if TIF_RESTORE_SIGMASK doesn't exist. */
436static inline void set_restore_sigmask(void)
437{
438 current->restore_sigmask = true;
439 WARN_ON(!test_thread_flag(TIF_SIGPENDING));
440}
441static inline void clear_restore_sigmask(void)
442{
443 current->restore_sigmask = false;
444}
445static inline bool test_restore_sigmask(void)
446{
447 return current->restore_sigmask;
448}
449static inline bool test_and_clear_restore_sigmask(void)
450{
451 if (!current->restore_sigmask)
452 return false;
453 current->restore_sigmask = false;
454 return true;
455}
456#endif
457
458static inline void restore_saved_sigmask(void)
459{
460 if (test_and_clear_restore_sigmask())
461 __set_current_blocked(&current->saved_sigmask);
462}
463
464static inline sigset_t *sigmask_to_save(void)
465{
466 sigset_t *res = &current->blocked;
467 if (unlikely(test_restore_sigmask()))
468 res = &current->saved_sigmask;
469 return res;
470}
471
472static inline int kill_cad_pid(int sig, int priv)
473{
474 return kill_pid(cad_pid, sig, priv);
475}
476
477/* These can be the second arg to send_sig_info/send_group_sig_info. */
478#define SEND_SIG_NOINFO ((struct siginfo *) 0)
479#define SEND_SIG_PRIV ((struct siginfo *) 1)
480#define SEND_SIG_FORCED ((struct siginfo *) 2)
481
482/*
483 * True if we are on the alternate signal stack.
484 */
485static inline int on_sig_stack(unsigned long sp)
486{
487 /*
488 * If the signal stack is SS_AUTODISARM then, by construction, we
489 * can't be on the signal stack unless user code deliberately set
490 * SS_AUTODISARM when we were already on it.
491 *
492 * This improves reliability: if user state gets corrupted such that
493 * the stack pointer points very close to the end of the signal stack,
494 * then this check will enable the signal to be handled anyway.
495 */
496 if (current->sas_ss_flags & SS_AUTODISARM)
497 return 0;
498
499#ifdef CONFIG_STACK_GROWSUP
500 return sp >= current->sas_ss_sp &&
501 sp - current->sas_ss_sp < current->sas_ss_size;
502#else
503 return sp > current->sas_ss_sp &&
504 sp - current->sas_ss_sp <= current->sas_ss_size;
505#endif
506}
507
508static inline int sas_ss_flags(unsigned long sp)
509{
510 if (!current->sas_ss_size)
511 return SS_DISABLE;
512
513 return on_sig_stack(sp) ? SS_ONSTACK : 0;
514}
515
516static inline void sas_ss_reset(struct task_struct *p)
517{
518 p->sas_ss_sp = 0;
519 p->sas_ss_size = 0;
520 p->sas_ss_flags = SS_DISABLE;
521}
522
523static inline unsigned long sigsp(unsigned long sp, struct ksignal *ksig)
524{
525 if (unlikely((ksig->ka.sa.sa_flags & SA_ONSTACK)) && ! sas_ss_flags(sp))
526#ifdef CONFIG_STACK_GROWSUP
527 return current->sas_ss_sp;
528#else
529 return current->sas_ss_sp + current->sas_ss_size;
530#endif
531 return sp;
532}
533
534extern void __cleanup_sighand(struct sighand_struct *);
535extern void flush_itimer_signals(void);
536
537#define tasklist_empty() \
538 list_empty(&init_task.tasks)
539
540#define next_task(p) \
541 list_entry_rcu((p)->tasks.next, struct task_struct, tasks)
542
543#define for_each_process(p) \
544 for (p = &init_task ; (p = next_task(p)) != &init_task ; )
545
546extern bool current_is_single_threaded(void);
547
548/*
549 * Careful: do_each_thread/while_each_thread is a double loop so
550 * 'break' will not work as expected - use goto instead.
551 */
552#define do_each_thread(g, t) \
553 for (g = t = &init_task ; (g = t = next_task(g)) != &init_task ; ) do
554
555#define while_each_thread(g, t) \
556 while ((t = next_thread(t)) != g)
557
558#define __for_each_thread(signal, t) \
559 list_for_each_entry_rcu(t, &(signal)->thread_head, thread_node)
560
561#define for_each_thread(p, t) \
562 __for_each_thread((p)->signal, t)
563
564/* Careful: this is a double loop, 'break' won't work as expected. */
565#define for_each_process_thread(p, t) \
566 for_each_process(p) for_each_thread(p, t)
567
568typedef int (*proc_visitor)(struct task_struct *p, void *data);
569void walk_process_tree(struct task_struct *top, proc_visitor, void *);
570
571static inline
572struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
573{
574 struct pid *pid;
575 if (type == PIDTYPE_PID)
576 pid = task_pid(task);
577 else
578 pid = task->signal->pids[type];
579 return pid;
580}
581
582static inline struct pid *task_tgid(struct task_struct *task)
583{
584 return task->signal->pids[PIDTYPE_TGID];
585}
586
587/*
588 * Without tasklist or RCU lock it is not safe to dereference
589 * the result of task_pgrp/task_session even if task == current,
590 * we can race with another thread doing sys_setsid/sys_setpgid.
591 */
592static inline struct pid *task_pgrp(struct task_struct *task)
593{
594 return task->signal->pids[PIDTYPE_PGID];
595}
596
597static inline struct pid *task_session(struct task_struct *task)
598{
599 return task->signal->pids[PIDTYPE_SID];
600}
601
602static inline int get_nr_threads(struct task_struct *tsk)
603{
604 return tsk->signal->nr_threads;
605}
606
607static inline bool thread_group_leader(struct task_struct *p)
608{
609 return p->exit_signal >= 0;
610}
611
612/* Do to the insanities of de_thread it is possible for a process
613 * to have the pid of the thread group leader without actually being
614 * the thread group leader. For iteration through the pids in proc
615 * all we care about is that we have a task with the appropriate
616 * pid, we don't actually care if we have the right task.
617 */
618static inline bool has_group_leader_pid(struct task_struct *p)
619{
620 return task_pid(p) == task_tgid(p);
621}
622
623static inline
624bool same_thread_group(struct task_struct *p1, struct task_struct *p2)
625{
626 return p1->signal == p2->signal;
627}
628
629static inline struct task_struct *next_thread(const struct task_struct *p)
630{
631 return list_entry_rcu(p->thread_group.next,
632 struct task_struct, thread_group);
633}
634
635static inline int thread_group_empty(struct task_struct *p)
636{
637 return list_empty(&p->thread_group);
638}
639
640#define delay_group_leader(p) \
641 (thread_group_leader(p) && !thread_group_empty(p))
642
643extern struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
644 unsigned long *flags);
645
646static inline struct sighand_struct *lock_task_sighand(struct task_struct *tsk,
647 unsigned long *flags)
648{
649 struct sighand_struct *ret;
650
651 ret = __lock_task_sighand(tsk, flags);
652 (void)__cond_lock(&tsk->sighand->siglock, ret);
653 return ret;
654}
655
656static inline void unlock_task_sighand(struct task_struct *tsk,
657 unsigned long *flags)
658{
659 spin_unlock_irqrestore(&tsk->sighand->siglock, *flags);
660}
661
662static inline unsigned long task_rlimit(const struct task_struct *tsk,
663 unsigned int limit)
664{
665 return READ_ONCE(tsk->signal->rlim[limit].rlim_cur);
666}
667
668static inline unsigned long task_rlimit_max(const struct task_struct *tsk,
669 unsigned int limit)
670{
671 return READ_ONCE(tsk->signal->rlim[limit].rlim_max);
672}
673
674static inline unsigned long rlimit(unsigned int limit)
675{
676 return task_rlimit(current, limit);
677}
678
679static inline unsigned long rlimit_max(unsigned int limit)
680{
681 return task_rlimit_max(current, limit);
682}
683
684#endif /* _LINUX_SCHED_SIGNAL_H */
685