4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * proc base directory handling functions
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
50 #include <linux/uaccess.h>
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/flex_array.h>
89 #include <linux/posix-timers.h>
90 #ifdef CONFIG_HARDWALL
91 #include <asm/hardwall.h>
93 #include <trace/events/oom.h>
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
108 static u8 nlink_tgid;
114 const struct inode_operations *iop;
115 const struct file_operations *fop;
119 #define NOD(NAME, MODE, IOP, FOP, OP) { \
121 .len = sizeof(NAME) - 1, \
128 #define DIR(NAME, MODE, iops, fops) \
129 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
130 #define LNK(NAME, get_link) \
131 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
132 &proc_pid_link_inode_operations, NULL, \
133 { .proc_get_link = get_link } )
134 #define REG(NAME, MODE, fops) \
135 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
136 #define ONE(NAME, MODE, show) \
137 NOD(NAME, (S_IFREG|(MODE)), \
138 NULL, &proc_single_file_operations, \
139 { .proc_show = show } )
142 * Count the number of hardlinks for the pid_entry table, excluding the .
145 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
160 static int get_task_root(struct task_struct *task, struct path *root)
162 int result = -ENOENT;
166 get_fs_root(task->fs, root);
173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
175 struct task_struct *task = get_proc_task(d_inode(dentry));
176 int result = -ENOENT;
181 get_fs_pwd(task->fs, path);
185 put_task_struct(task);
190 static int proc_root_link(struct dentry *dentry, struct path *path)
192 struct task_struct *task = get_proc_task(d_inode(dentry));
193 int result = -ENOENT;
196 result = get_task_root(task, path);
197 put_task_struct(task);
202 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
203 size_t _count, loff_t *pos)
205 struct task_struct *tsk;
206 struct mm_struct *mm;
208 unsigned long count = _count;
209 unsigned long arg_start, arg_end, env_start, env_end;
210 unsigned long len1, len2, len;
217 tsk = get_proc_task(file_inode(file));
220 mm = get_task_mm(tsk);
221 put_task_struct(tsk);
224 /* Check if process spawned far enough to have cmdline. */
230 page = (char *)__get_free_page(GFP_TEMPORARY);
236 down_read(&mm->mmap_sem);
237 arg_start = mm->arg_start;
238 arg_end = mm->arg_end;
239 env_start = mm->env_start;
240 env_end = mm->env_end;
241 up_read(&mm->mmap_sem);
243 BUG_ON(arg_start > arg_end);
244 BUG_ON(env_start > env_end);
246 len1 = arg_end - arg_start;
247 len2 = env_end - env_start;
255 * Inherently racy -- command line shares address space
256 * with code and data.
258 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
265 /* Command line (set of strings) occupies whole ARGV. */
269 p = arg_start + *pos;
271 while (count > 0 && len > 0) {
275 _count = min3(count, len, PAGE_SIZE);
276 nr_read = access_remote_vm(mm, p, page, _count, 0);
282 if (copy_to_user(buf, page, nr_read)) {
295 * Command line (1 string) occupies ARGV and
302 { .p = arg_start, .len = len1 },
303 { .p = env_start, .len = len2 },
309 while (i < 2 && pos1 >= cmdline[i].len) {
310 pos1 -= cmdline[i].len;
314 p = cmdline[i].p + pos1;
315 len = cmdline[i].len - pos1;
316 while (count > 0 && len > 0) {
317 unsigned int _count, l;
321 _count = min3(count, len, PAGE_SIZE);
322 nr_read = access_remote_vm(mm, p, page, _count, 0);
329 * Command line can be shorter than whole ARGV
330 * even if last "marker" byte says it is not.
333 l = strnlen(page, nr_read);
339 if (copy_to_user(buf, page, nr_read)) {
354 /* Only first chunk can be read partially. */
361 free_page((unsigned long)page);
369 static const struct file_operations proc_pid_cmdline_ops = {
370 .read = proc_pid_cmdline_read,
371 .llseek = generic_file_llseek,
374 #ifdef CONFIG_KALLSYMS
376 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
377 * Returns the resolved symbol. If that fails, simply return the address.
379 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
380 struct pid *pid, struct task_struct *task)
383 char symname[KSYM_NAME_LEN];
385 wchan = get_wchan(task);
387 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
388 && !lookup_symbol_name(wchan, symname))
389 seq_printf(m, "%s", symname);
395 #endif /* CONFIG_KALLSYMS */
397 static int lock_trace(struct task_struct *task)
399 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
402 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
403 mutex_unlock(&task->signal->cred_guard_mutex);
409 static void unlock_trace(struct task_struct *task)
411 mutex_unlock(&task->signal->cred_guard_mutex);
414 #ifdef CONFIG_STACKTRACE
416 #define MAX_STACK_TRACE_DEPTH 64
418 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
419 struct pid *pid, struct task_struct *task)
421 struct stack_trace trace;
422 unsigned long *entries;
426 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
430 trace.nr_entries = 0;
431 trace.max_entries = MAX_STACK_TRACE_DEPTH;
432 trace.entries = entries;
435 err = lock_trace(task);
437 save_stack_trace_tsk(task, &trace);
439 for (i = 0; i < trace.nr_entries; i++) {
440 seq_printf(m, "[<%pK>] %pB\n",
441 (void *)entries[i], (void *)entries[i]);
451 #ifdef CONFIG_SCHED_INFO
453 * Provides /proc/PID/schedstat
455 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
456 struct pid *pid, struct task_struct *task)
458 if (unlikely(!sched_info_on()))
459 seq_printf(m, "0 0 0\n");
461 seq_printf(m, "%llu %llu %lu\n",
462 (unsigned long long)task->se.sum_exec_runtime,
463 (unsigned long long)task->sched_info.run_delay,
464 task->sched_info.pcount);
470 #ifdef CONFIG_LATENCYTOP
471 static int lstats_show_proc(struct seq_file *m, void *v)
474 struct inode *inode = m->private;
475 struct task_struct *task = get_proc_task(inode);
479 seq_puts(m, "Latency Top version : v0.1\n");
480 for (i = 0; i < 32; i++) {
481 struct latency_record *lr = &task->latency_record[i];
482 if (lr->backtrace[0]) {
484 seq_printf(m, "%i %li %li",
485 lr->count, lr->time, lr->max);
486 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
487 unsigned long bt = lr->backtrace[q];
492 seq_printf(m, " %ps", (void *)bt);
498 put_task_struct(task);
502 static int lstats_open(struct inode *inode, struct file *file)
504 return single_open(file, lstats_show_proc, inode);
507 static ssize_t lstats_write(struct file *file, const char __user *buf,
508 size_t count, loff_t *offs)
510 struct task_struct *task = get_proc_task(file_inode(file));
514 clear_all_latency_tracing(task);
515 put_task_struct(task);
520 static const struct file_operations proc_lstats_operations = {
523 .write = lstats_write,
525 .release = single_release,
530 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
531 struct pid *pid, struct task_struct *task)
533 unsigned long totalpages = totalram_pages + total_swap_pages;
534 unsigned long points = 0;
536 points = oom_badness(task, NULL, NULL, totalpages) *
538 seq_printf(m, "%lu\n", points);
548 static const struct limit_names lnames[RLIM_NLIMITS] = {
549 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
550 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
551 [RLIMIT_DATA] = {"Max data size", "bytes"},
552 [RLIMIT_STACK] = {"Max stack size", "bytes"},
553 [RLIMIT_CORE] = {"Max core file size", "bytes"},
554 [RLIMIT_RSS] = {"Max resident set", "bytes"},
555 [RLIMIT_NPROC] = {"Max processes", "processes"},
556 [RLIMIT_NOFILE] = {"Max open files", "files"},
557 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
558 [RLIMIT_AS] = {"Max address space", "bytes"},
559 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
560 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
561 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
562 [RLIMIT_NICE] = {"Max nice priority", NULL},
563 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
564 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
567 /* Display limits for a process */
568 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
569 struct pid *pid, struct task_struct *task)
574 struct rlimit rlim[RLIM_NLIMITS];
576 if (!lock_task_sighand(task, &flags))
578 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
579 unlock_task_sighand(task, &flags);
582 * print the file header
584 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
585 "Limit", "Soft Limit", "Hard Limit", "Units");
587 for (i = 0; i < RLIM_NLIMITS; i++) {
588 if (rlim[i].rlim_cur == RLIM_INFINITY)
589 seq_printf(m, "%-25s %-20s ",
590 lnames[i].name, "unlimited");
592 seq_printf(m, "%-25s %-20lu ",
593 lnames[i].name, rlim[i].rlim_cur);
595 if (rlim[i].rlim_max == RLIM_INFINITY)
596 seq_printf(m, "%-20s ", "unlimited");
598 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
601 seq_printf(m, "%-10s\n", lnames[i].unit);
609 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
610 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
611 struct pid *pid, struct task_struct *task)
614 unsigned long args[6], sp, pc;
617 res = lock_trace(task);
621 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
622 seq_puts(m, "running\n");
624 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
627 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
629 args[0], args[1], args[2], args[3], args[4], args[5],
635 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
637 /************************************************************************/
638 /* Here the fs part begins */
639 /************************************************************************/
641 /* permission checks */
642 static int proc_fd_access_allowed(struct inode *inode)
644 struct task_struct *task;
646 /* Allow access to a task's file descriptors if it is us or we
647 * may use ptrace attach to the process and find out that
650 task = get_proc_task(inode);
652 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
653 put_task_struct(task);
658 int proc_setattr(struct dentry *dentry, struct iattr *attr)
661 struct inode *inode = d_inode(dentry);
663 if (attr->ia_valid & ATTR_MODE)
666 error = setattr_prepare(dentry, attr);
670 setattr_copy(inode, attr);
671 mark_inode_dirty(inode);
676 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
677 * or euid/egid (for hide_pid_min=2)?
679 static bool has_pid_permissions(struct pid_namespace *pid,
680 struct task_struct *task,
683 if (pid->hide_pid < hide_pid_min)
685 if (in_group_p(pid->pid_gid))
687 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
691 static int proc_pid_permission(struct inode *inode, int mask)
693 struct pid_namespace *pid = inode->i_sb->s_fs_info;
694 struct task_struct *task;
697 task = get_proc_task(inode);
700 has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
701 put_task_struct(task);
704 if (pid->hide_pid == HIDEPID_INVISIBLE) {
706 * Let's make getdents(), stat(), and open()
707 * consistent with each other. If a process
708 * may not stat() a file, it shouldn't be seen
716 return generic_permission(inode, mask);
721 static const struct inode_operations proc_def_inode_operations = {
722 .setattr = proc_setattr,
725 static int proc_single_show(struct seq_file *m, void *v)
727 struct inode *inode = m->private;
728 struct pid_namespace *ns;
730 struct task_struct *task;
733 ns = inode->i_sb->s_fs_info;
734 pid = proc_pid(inode);
735 task = get_pid_task(pid, PIDTYPE_PID);
739 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
741 put_task_struct(task);
745 static int proc_single_open(struct inode *inode, struct file *filp)
747 return single_open(filp, proc_single_show, inode);
750 static const struct file_operations proc_single_file_operations = {
751 .open = proc_single_open,
754 .release = single_release,
758 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
760 struct task_struct *task = get_proc_task(inode);
761 struct mm_struct *mm = ERR_PTR(-ESRCH);
764 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
765 put_task_struct(task);
767 if (!IS_ERR_OR_NULL(mm)) {
768 /* ensure this mm_struct can't be freed */
770 /* but do not pin its memory */
778 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
780 struct mm_struct *mm = proc_mem_open(inode, mode);
785 file->private_data = mm;
789 static int mem_open(struct inode *inode, struct file *file)
791 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
793 /* OK to pass negative loff_t, we can catch out-of-range */
794 file->f_mode |= FMODE_UNSIGNED_OFFSET;
799 static ssize_t mem_rw(struct file *file, char __user *buf,
800 size_t count, loff_t *ppos, int write)
802 struct mm_struct *mm = file->private_data;
803 unsigned long addr = *ppos;
811 page = (char *)__get_free_page(GFP_TEMPORARY);
816 if (!mmget_not_zero(mm))
819 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
825 int this_len = min_t(int, count, PAGE_SIZE);
827 if (write && copy_from_user(page, buf, this_len)) {
832 this_len = access_remote_vm(mm, addr, page, this_len, flags);
839 if (!write && copy_to_user(buf, page, this_len)) {
853 free_page((unsigned long) page);
857 static ssize_t mem_read(struct file *file, char __user *buf,
858 size_t count, loff_t *ppos)
860 return mem_rw(file, buf, count, ppos, 0);
863 static ssize_t mem_write(struct file *file, const char __user *buf,
864 size_t count, loff_t *ppos)
866 return mem_rw(file, (char __user*)buf, count, ppos, 1);
869 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
873 file->f_pos = offset;
876 file->f_pos += offset;
881 force_successful_syscall_return();
885 static int mem_release(struct inode *inode, struct file *file)
887 struct mm_struct *mm = file->private_data;
893 static const struct file_operations proc_mem_operations = {
898 .release = mem_release,
901 static int environ_open(struct inode *inode, struct file *file)
903 return __mem_open(inode, file, PTRACE_MODE_READ);
906 static ssize_t environ_read(struct file *file, char __user *buf,
907 size_t count, loff_t *ppos)
910 unsigned long src = *ppos;
912 struct mm_struct *mm = file->private_data;
913 unsigned long env_start, env_end;
915 /* Ensure the process spawned far enough to have an environment. */
916 if (!mm || !mm->env_end)
919 page = (char *)__get_free_page(GFP_TEMPORARY);
924 if (!mmget_not_zero(mm))
927 down_read(&mm->mmap_sem);
928 env_start = mm->env_start;
929 env_end = mm->env_end;
930 up_read(&mm->mmap_sem);
933 size_t this_len, max_len;
936 if (src >= (env_end - env_start))
939 this_len = env_end - (env_start + src);
941 max_len = min_t(size_t, PAGE_SIZE, count);
942 this_len = min(max_len, this_len);
944 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
951 if (copy_to_user(buf, page, retval)) {
965 free_page((unsigned long) page);
969 static const struct file_operations proc_environ_operations = {
970 .open = environ_open,
971 .read = environ_read,
972 .llseek = generic_file_llseek,
973 .release = mem_release,
976 static int auxv_open(struct inode *inode, struct file *file)
978 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
981 static ssize_t auxv_read(struct file *file, char __user *buf,
982 size_t count, loff_t *ppos)
984 struct mm_struct *mm = file->private_data;
985 unsigned int nwords = 0;
991 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
992 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
993 nwords * sizeof(mm->saved_auxv[0]));
996 static const struct file_operations proc_auxv_operations = {
999 .llseek = generic_file_llseek,
1000 .release = mem_release,
1003 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006 struct task_struct *task = get_proc_task(file_inode(file));
1007 char buffer[PROC_NUMBUF];
1008 int oom_adj = OOM_ADJUST_MIN;
1013 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1014 oom_adj = OOM_ADJUST_MAX;
1016 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1018 put_task_struct(task);
1019 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1020 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1025 static DEFINE_MUTEX(oom_adj_mutex);
1026 struct mm_struct *mm = NULL;
1027 struct task_struct *task;
1030 task = get_proc_task(file_inode(file));
1034 mutex_lock(&oom_adj_mutex);
1036 if (oom_adj < task->signal->oom_score_adj &&
1037 !capable(CAP_SYS_RESOURCE)) {
1042 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1043 * /proc/pid/oom_score_adj instead.
1045 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1046 current->comm, task_pid_nr(current), task_pid_nr(task),
1049 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1050 !capable(CAP_SYS_RESOURCE)) {
1057 * Make sure we will check other processes sharing the mm if this is
1058 * not vfrok which wants its own oom_score_adj.
1059 * pin the mm so it doesn't go away and get reused after task_unlock
1061 if (!task->vfork_done) {
1062 struct task_struct *p = find_lock_task_mm(task);
1065 if (atomic_read(&p->mm->mm_users) > 1) {
1073 task->signal->oom_score_adj = oom_adj;
1074 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1075 task->signal->oom_score_adj_min = (short)oom_adj;
1076 trace_oom_score_adj_update(task);
1079 struct task_struct *p;
1082 for_each_process(p) {
1083 if (same_thread_group(task, p))
1086 /* do not touch kernel threads or the global init */
1087 if (p->flags & PF_KTHREAD || is_global_init(p))
1091 if (!p->vfork_done && process_shares_mm(p, mm)) {
1092 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1093 task_pid_nr(p), p->comm,
1094 p->signal->oom_score_adj, oom_adj,
1095 task_pid_nr(task), task->comm);
1096 p->signal->oom_score_adj = oom_adj;
1097 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1098 p->signal->oom_score_adj_min = (short)oom_adj;
1106 mutex_unlock(&oom_adj_mutex);
1107 put_task_struct(task);
1112 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1113 * kernels. The effective policy is defined by oom_score_adj, which has a
1114 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1115 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1116 * Processes that become oom disabled via oom_adj will still be oom disabled
1117 * with this implementation.
1119 * oom_adj cannot be removed since existing userspace binaries use it.
1121 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1122 size_t count, loff_t *ppos)
1124 char buffer[PROC_NUMBUF];
1128 memset(buffer, 0, sizeof(buffer));
1129 if (count > sizeof(buffer) - 1)
1130 count = sizeof(buffer) - 1;
1131 if (copy_from_user(buffer, buf, count)) {
1136 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1140 oom_adj != OOM_DISABLE) {
1146 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1147 * value is always attainable.
1149 if (oom_adj == OOM_ADJUST_MAX)
1150 oom_adj = OOM_SCORE_ADJ_MAX;
1152 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1154 err = __set_oom_adj(file, oom_adj, true);
1156 return err < 0 ? err : count;
1159 static const struct file_operations proc_oom_adj_operations = {
1160 .read = oom_adj_read,
1161 .write = oom_adj_write,
1162 .llseek = generic_file_llseek,
1165 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1166 size_t count, loff_t *ppos)
1168 struct task_struct *task = get_proc_task(file_inode(file));
1169 char buffer[PROC_NUMBUF];
1170 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1175 oom_score_adj = task->signal->oom_score_adj;
1176 put_task_struct(task);
1177 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1178 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1182 size_t count, loff_t *ppos)
1184 char buffer[PROC_NUMBUF];
1188 memset(buffer, 0, sizeof(buffer));
1189 if (count > sizeof(buffer) - 1)
1190 count = sizeof(buffer) - 1;
1191 if (copy_from_user(buffer, buf, count)) {
1196 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1200 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1205 err = __set_oom_adj(file, oom_score_adj, false);
1207 return err < 0 ? err : count;
1210 static const struct file_operations proc_oom_score_adj_operations = {
1211 .read = oom_score_adj_read,
1212 .write = oom_score_adj_write,
1213 .llseek = default_llseek,
1216 #ifdef CONFIG_AUDITSYSCALL
1217 #define TMPBUFLEN 11
1218 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1219 size_t count, loff_t *ppos)
1221 struct inode * inode = file_inode(file);
1222 struct task_struct *task = get_proc_task(inode);
1224 char tmpbuf[TMPBUFLEN];
1228 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1229 from_kuid(file->f_cred->user_ns,
1230 audit_get_loginuid(task)));
1231 put_task_struct(task);
1232 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1236 size_t count, loff_t *ppos)
1238 struct inode * inode = file_inode(file);
1244 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1251 /* No partial writes. */
1255 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1259 /* is userspace tring to explicitly UNSET the loginuid? */
1260 if (loginuid == AUDIT_UID_UNSET) {
1261 kloginuid = INVALID_UID;
1263 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1264 if (!uid_valid(kloginuid))
1268 rv = audit_set_loginuid(kloginuid);
1274 static const struct file_operations proc_loginuid_operations = {
1275 .read = proc_loginuid_read,
1276 .write = proc_loginuid_write,
1277 .llseek = generic_file_llseek,
1280 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1281 size_t count, loff_t *ppos)
1283 struct inode * inode = file_inode(file);
1284 struct task_struct *task = get_proc_task(inode);
1286 char tmpbuf[TMPBUFLEN];
1290 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1291 audit_get_sessionid(task));
1292 put_task_struct(task);
1293 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 static const struct file_operations proc_sessionid_operations = {
1297 .read = proc_sessionid_read,
1298 .llseek = generic_file_llseek,
1302 #ifdef CONFIG_FAULT_INJECTION
1303 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1304 size_t count, loff_t *ppos)
1306 struct task_struct *task = get_proc_task(file_inode(file));
1307 char buffer[PROC_NUMBUF];
1313 make_it_fail = task->make_it_fail;
1314 put_task_struct(task);
1316 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1318 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 static ssize_t proc_fault_inject_write(struct file * file,
1322 const char __user * buf, size_t count, loff_t *ppos)
1324 struct task_struct *task;
1325 char buffer[PROC_NUMBUF];
1329 if (!capable(CAP_SYS_RESOURCE))
1331 memset(buffer, 0, sizeof(buffer));
1332 if (count > sizeof(buffer) - 1)
1333 count = sizeof(buffer) - 1;
1334 if (copy_from_user(buffer, buf, count))
1336 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339 if (make_it_fail < 0 || make_it_fail > 1)
1342 task = get_proc_task(file_inode(file));
1345 task->make_it_fail = make_it_fail;
1346 put_task_struct(task);
1351 static const struct file_operations proc_fault_inject_operations = {
1352 .read = proc_fault_inject_read,
1353 .write = proc_fault_inject_write,
1354 .llseek = generic_file_llseek,
1359 #ifdef CONFIG_SCHED_DEBUG
1361 * Print out various scheduling related per-task fields:
1363 static int sched_show(struct seq_file *m, void *v)
1365 struct inode *inode = m->private;
1366 struct task_struct *p;
1368 p = get_proc_task(inode);
1371 proc_sched_show_task(p, m);
1379 sched_write(struct file *file, const char __user *buf,
1380 size_t count, loff_t *offset)
1382 struct inode *inode = file_inode(file);
1383 struct task_struct *p;
1385 p = get_proc_task(inode);
1388 proc_sched_set_task(p);
1395 static int sched_open(struct inode *inode, struct file *filp)
1397 return single_open(filp, sched_show, inode);
1400 static const struct file_operations proc_pid_sched_operations = {
1403 .write = sched_write,
1404 .llseek = seq_lseek,
1405 .release = single_release,
1410 #ifdef CONFIG_SCHED_AUTOGROUP
1412 * Print out autogroup related information:
1414 static int sched_autogroup_show(struct seq_file *m, void *v)
1416 struct inode *inode = m->private;
1417 struct task_struct *p;
1419 p = get_proc_task(inode);
1422 proc_sched_autogroup_show_task(p, m);
1430 sched_autogroup_write(struct file *file, const char __user *buf,
1431 size_t count, loff_t *offset)
1433 struct inode *inode = file_inode(file);
1434 struct task_struct *p;
1435 char buffer[PROC_NUMBUF];
1439 memset(buffer, 0, sizeof(buffer));
1440 if (count > sizeof(buffer) - 1)
1441 count = sizeof(buffer) - 1;
1442 if (copy_from_user(buffer, buf, count))
1445 err = kstrtoint(strstrip(buffer), 0, &nice);
1449 p = get_proc_task(inode);
1453 err = proc_sched_autogroup_set_nice(p, nice);
1462 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1466 ret = single_open(filp, sched_autogroup_show, NULL);
1468 struct seq_file *m = filp->private_data;
1475 static const struct file_operations proc_pid_sched_autogroup_operations = {
1476 .open = sched_autogroup_open,
1478 .write = sched_autogroup_write,
1479 .llseek = seq_lseek,
1480 .release = single_release,
1483 #endif /* CONFIG_SCHED_AUTOGROUP */
1485 static ssize_t comm_write(struct file *file, const char __user *buf,
1486 size_t count, loff_t *offset)
1488 struct inode *inode = file_inode(file);
1489 struct task_struct *p;
1490 char buffer[TASK_COMM_LEN];
1491 const size_t maxlen = sizeof(buffer) - 1;
1493 memset(buffer, 0, sizeof(buffer));
1494 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1497 p = get_proc_task(inode);
1501 if (same_thread_group(current, p))
1502 set_task_comm(p, buffer);
1511 static int comm_show(struct seq_file *m, void *v)
1513 struct inode *inode = m->private;
1514 struct task_struct *p;
1516 p = get_proc_task(inode);
1521 seq_printf(m, "%s\n", p->comm);
1529 static int comm_open(struct inode *inode, struct file *filp)
1531 return single_open(filp, comm_show, inode);
1534 static const struct file_operations proc_pid_set_comm_operations = {
1537 .write = comm_write,
1538 .llseek = seq_lseek,
1539 .release = single_release,
1542 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1544 struct task_struct *task;
1545 struct file *exe_file;
1547 task = get_proc_task(d_inode(dentry));
1550 exe_file = get_task_exe_file(task);
1551 put_task_struct(task);
1553 *exe_path = exe_file->f_path;
1554 path_get(&exe_file->f_path);
1561 static const char *proc_pid_get_link(struct dentry *dentry,
1562 struct inode *inode,
1563 struct delayed_call *done)
1566 int error = -EACCES;
1569 return ERR_PTR(-ECHILD);
1571 /* Are we allowed to snoop on the tasks file descriptors? */
1572 if (!proc_fd_access_allowed(inode))
1575 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1579 nd_jump_link(&path);
1582 return ERR_PTR(error);
1585 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1587 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1594 pathname = d_path(path, tmp, PAGE_SIZE);
1595 len = PTR_ERR(pathname);
1596 if (IS_ERR(pathname))
1598 len = tmp + PAGE_SIZE - 1 - pathname;
1602 if (copy_to_user(buffer, pathname, len))
1605 free_page((unsigned long)tmp);
1609 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1611 int error = -EACCES;
1612 struct inode *inode = d_inode(dentry);
1615 /* Are we allowed to snoop on the tasks file descriptors? */
1616 if (!proc_fd_access_allowed(inode))
1619 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1623 error = do_proc_readlink(&path, buffer, buflen);
1629 const struct inode_operations proc_pid_link_inode_operations = {
1630 .readlink = proc_pid_readlink,
1631 .get_link = proc_pid_get_link,
1632 .setattr = proc_setattr,
1636 /* building an inode */
1638 void task_dump_owner(struct task_struct *task, mode_t mode,
1639 kuid_t *ruid, kgid_t *rgid)
1641 /* Depending on the state of dumpable compute who should own a
1642 * proc file for a task.
1644 const struct cred *cred;
1648 /* Default to the tasks effective ownership */
1650 cred = __task_cred(task);
1656 * Before the /proc/pid/status file was created the only way to read
1657 * the effective uid of a /process was to stat /proc/pid. Reading
1658 * /proc/pid/status is slow enough that procps and other packages
1659 * kept stating /proc/pid. To keep the rules in /proc simple I have
1660 * made this apply to all per process world readable and executable
1663 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1664 struct mm_struct *mm;
1667 /* Make non-dumpable tasks owned by some root */
1669 if (get_dumpable(mm) != SUID_DUMP_USER) {
1670 struct user_namespace *user_ns = mm->user_ns;
1672 uid = make_kuid(user_ns, 0);
1673 if (!uid_valid(uid))
1674 uid = GLOBAL_ROOT_UID;
1676 gid = make_kgid(user_ns, 0);
1677 if (!gid_valid(gid))
1678 gid = GLOBAL_ROOT_GID;
1681 uid = GLOBAL_ROOT_UID;
1682 gid = GLOBAL_ROOT_GID;
1690 struct inode *proc_pid_make_inode(struct super_block * sb,
1691 struct task_struct *task, umode_t mode)
1693 struct inode * inode;
1694 struct proc_inode *ei;
1696 /* We need a new inode */
1698 inode = new_inode(sb);
1704 inode->i_mode = mode;
1705 inode->i_ino = get_next_ino();
1706 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1707 inode->i_op = &proc_def_inode_operations;
1710 * grab the reference to task.
1712 ei->pid = get_task_pid(task, PIDTYPE_PID);
1716 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1717 security_task_to_inode(task, inode);
1727 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1729 struct inode *inode = d_inode(dentry);
1730 struct task_struct *task;
1731 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1733 generic_fillattr(inode, stat);
1736 stat->uid = GLOBAL_ROOT_UID;
1737 stat->gid = GLOBAL_ROOT_GID;
1738 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1740 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1743 * This doesn't prevent learning whether PID exists,
1744 * it only makes getattr() consistent with readdir().
1748 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1757 * Exceptional case: normally we are not allowed to unhash a busy
1758 * directory. In this case, however, we can do it - no aliasing problems
1759 * due to the way we treat inodes.
1761 * Rewrite the inode's ownerships here because the owning task may have
1762 * performed a setuid(), etc.
1765 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1767 struct inode *inode;
1768 struct task_struct *task;
1770 if (flags & LOOKUP_RCU)
1773 inode = d_inode(dentry);
1774 task = get_proc_task(inode);
1777 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1779 inode->i_mode &= ~(S_ISUID | S_ISGID);
1780 security_task_to_inode(task, inode);
1781 put_task_struct(task);
1787 static inline bool proc_inode_is_dead(struct inode *inode)
1789 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1792 int pid_delete_dentry(const struct dentry *dentry)
1794 /* Is the task we represent dead?
1795 * If so, then don't put the dentry on the lru list,
1796 * kill it immediately.
1798 return proc_inode_is_dead(d_inode(dentry));
1801 const struct dentry_operations pid_dentry_operations =
1803 .d_revalidate = pid_revalidate,
1804 .d_delete = pid_delete_dentry,
1810 * Fill a directory entry.
1812 * If possible create the dcache entry and derive our inode number and
1813 * file type from dcache entry.
1815 * Since all of the proc inode numbers are dynamically generated, the inode
1816 * numbers do not exist until the inode is cache. This means creating the
1817 * the dcache entry in readdir is necessary to keep the inode numbers
1818 * reported by readdir in sync with the inode numbers reported
1821 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1822 const char *name, int len,
1823 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1825 struct dentry *child, *dir = file->f_path.dentry;
1826 struct qstr qname = QSTR_INIT(name, len);
1827 struct inode *inode;
1831 child = d_hash_and_lookup(dir, &qname);
1833 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1834 child = d_alloc_parallel(dir, &qname, &wq);
1836 goto end_instantiate;
1837 if (d_in_lookup(child)) {
1838 int err = instantiate(d_inode(dir), child, task, ptr);
1839 d_lookup_done(child);
1842 goto end_instantiate;
1846 inode = d_inode(child);
1848 type = inode->i_mode >> 12;
1850 return dir_emit(ctx, name, len, ino, type);
1853 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1857 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1858 * which represent vma start and end addresses.
1860 static int dname_to_vma_addr(struct dentry *dentry,
1861 unsigned long *start, unsigned long *end)
1863 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1869 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1871 unsigned long vm_start, vm_end;
1872 bool exact_vma_exists = false;
1873 struct mm_struct *mm = NULL;
1874 struct task_struct *task;
1875 struct inode *inode;
1878 if (flags & LOOKUP_RCU)
1881 inode = d_inode(dentry);
1882 task = get_proc_task(inode);
1886 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1887 if (IS_ERR_OR_NULL(mm))
1890 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1891 down_read(&mm->mmap_sem);
1892 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1893 up_read(&mm->mmap_sem);
1898 if (exact_vma_exists) {
1899 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1901 security_task_to_inode(task, inode);
1906 put_task_struct(task);
1912 static const struct dentry_operations tid_map_files_dentry_operations = {
1913 .d_revalidate = map_files_d_revalidate,
1914 .d_delete = pid_delete_dentry,
1917 static int map_files_get_link(struct dentry *dentry, struct path *path)
1919 unsigned long vm_start, vm_end;
1920 struct vm_area_struct *vma;
1921 struct task_struct *task;
1922 struct mm_struct *mm;
1926 task = get_proc_task(d_inode(dentry));
1930 mm = get_task_mm(task);
1931 put_task_struct(task);
1935 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1940 down_read(&mm->mmap_sem);
1941 vma = find_exact_vma(mm, vm_start, vm_end);
1942 if (vma && vma->vm_file) {
1943 *path = vma->vm_file->f_path;
1947 up_read(&mm->mmap_sem);
1955 struct map_files_info {
1958 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1962 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1963 * symlinks may be used to bypass permissions on ancestor directories in the
1964 * path to the file in question.
1967 proc_map_files_get_link(struct dentry *dentry,
1968 struct inode *inode,
1969 struct delayed_call *done)
1971 if (!capable(CAP_SYS_ADMIN))
1972 return ERR_PTR(-EPERM);
1974 return proc_pid_get_link(dentry, inode, done);
1978 * Identical to proc_pid_link_inode_operations except for get_link()
1980 static const struct inode_operations proc_map_files_link_inode_operations = {
1981 .readlink = proc_pid_readlink,
1982 .get_link = proc_map_files_get_link,
1983 .setattr = proc_setattr,
1987 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1988 struct task_struct *task, const void *ptr)
1990 fmode_t mode = (fmode_t)(unsigned long)ptr;
1991 struct proc_inode *ei;
1992 struct inode *inode;
1994 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
1995 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
1996 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2001 ei->op.proc_get_link = map_files_get_link;
2003 inode->i_op = &proc_map_files_link_inode_operations;
2006 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2007 d_add(dentry, inode);
2012 static struct dentry *proc_map_files_lookup(struct inode *dir,
2013 struct dentry *dentry, unsigned int flags)
2015 unsigned long vm_start, vm_end;
2016 struct vm_area_struct *vma;
2017 struct task_struct *task;
2019 struct mm_struct *mm;
2022 task = get_proc_task(dir);
2027 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2031 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2034 mm = get_task_mm(task);
2038 down_read(&mm->mmap_sem);
2039 vma = find_exact_vma(mm, vm_start, vm_end);
2044 result = proc_map_files_instantiate(dir, dentry, task,
2045 (void *)(unsigned long)vma->vm_file->f_mode);
2048 up_read(&mm->mmap_sem);
2051 put_task_struct(task);
2053 return ERR_PTR(result);
2056 static const struct inode_operations proc_map_files_inode_operations = {
2057 .lookup = proc_map_files_lookup,
2058 .permission = proc_fd_permission,
2059 .setattr = proc_setattr,
2063 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2065 struct vm_area_struct *vma;
2066 struct task_struct *task;
2067 struct mm_struct *mm;
2068 unsigned long nr_files, pos, i;
2069 struct flex_array *fa = NULL;
2070 struct map_files_info info;
2071 struct map_files_info *p;
2075 task = get_proc_task(file_inode(file));
2080 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2084 if (!dir_emit_dots(file, ctx))
2087 mm = get_task_mm(task);
2090 down_read(&mm->mmap_sem);
2095 * We need two passes here:
2097 * 1) Collect vmas of mapped files with mmap_sem taken
2098 * 2) Release mmap_sem and instantiate entries
2100 * otherwise we get lockdep complained, since filldir()
2101 * routine might require mmap_sem taken in might_fault().
2104 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2105 if (vma->vm_file && ++pos > ctx->pos)
2110 fa = flex_array_alloc(sizeof(info), nr_files,
2112 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2116 flex_array_free(fa);
2117 up_read(&mm->mmap_sem);
2121 for (i = 0, vma = mm->mmap, pos = 2; vma;
2122 vma = vma->vm_next) {
2125 if (++pos <= ctx->pos)
2128 info.mode = vma->vm_file->f_mode;
2129 info.len = snprintf(info.name,
2130 sizeof(info.name), "%lx-%lx",
2131 vma->vm_start, vma->vm_end);
2132 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2136 up_read(&mm->mmap_sem);
2138 for (i = 0; i < nr_files; i++) {
2139 p = flex_array_get(fa, i);
2140 if (!proc_fill_cache(file, ctx,
2142 proc_map_files_instantiate,
2144 (void *)(unsigned long)p->mode))
2149 flex_array_free(fa);
2153 put_task_struct(task);
2158 static const struct file_operations proc_map_files_operations = {
2159 .read = generic_read_dir,
2160 .iterate_shared = proc_map_files_readdir,
2161 .llseek = generic_file_llseek,
2164 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2165 struct timers_private {
2167 struct task_struct *task;
2168 struct sighand_struct *sighand;
2169 struct pid_namespace *ns;
2170 unsigned long flags;
2173 static void *timers_start(struct seq_file *m, loff_t *pos)
2175 struct timers_private *tp = m->private;
2177 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2179 return ERR_PTR(-ESRCH);
2181 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2183 return ERR_PTR(-ESRCH);
2185 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2188 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2190 struct timers_private *tp = m->private;
2191 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2194 static void timers_stop(struct seq_file *m, void *v)
2196 struct timers_private *tp = m->private;
2199 unlock_task_sighand(tp->task, &tp->flags);
2204 put_task_struct(tp->task);
2209 static int show_timer(struct seq_file *m, void *v)
2211 struct k_itimer *timer;
2212 struct timers_private *tp = m->private;
2214 static const char * const nstr[] = {
2215 [SIGEV_SIGNAL] = "signal",
2216 [SIGEV_NONE] = "none",
2217 [SIGEV_THREAD] = "thread",
2220 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2221 notify = timer->it_sigev_notify;
2223 seq_printf(m, "ID: %d\n", timer->it_id);
2224 seq_printf(m, "signal: %d/%p\n",
2225 timer->sigq->info.si_signo,
2226 timer->sigq->info.si_value.sival_ptr);
2227 seq_printf(m, "notify: %s/%s.%d\n",
2228 nstr[notify & ~SIGEV_THREAD_ID],
2229 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2230 pid_nr_ns(timer->it_pid, tp->ns));
2231 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2236 static const struct seq_operations proc_timers_seq_ops = {
2237 .start = timers_start,
2238 .next = timers_next,
2239 .stop = timers_stop,
2243 static int proc_timers_open(struct inode *inode, struct file *file)
2245 struct timers_private *tp;
2247 tp = __seq_open_private(file, &proc_timers_seq_ops,
2248 sizeof(struct timers_private));
2252 tp->pid = proc_pid(inode);
2253 tp->ns = inode->i_sb->s_fs_info;
2257 static const struct file_operations proc_timers_operations = {
2258 .open = proc_timers_open,
2260 .llseek = seq_lseek,
2261 .release = seq_release_private,
2265 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2266 size_t count, loff_t *offset)
2268 struct inode *inode = file_inode(file);
2269 struct task_struct *p;
2273 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2277 p = get_proc_task(inode);
2282 if (!capable(CAP_SYS_NICE)) {
2287 err = security_task_setscheduler(p);
2296 p->timer_slack_ns = p->default_timer_slack_ns;
2298 p->timer_slack_ns = slack_ns;
2307 static int timerslack_ns_show(struct seq_file *m, void *v)
2309 struct inode *inode = m->private;
2310 struct task_struct *p;
2313 p = get_proc_task(inode);
2319 if (!capable(CAP_SYS_NICE)) {
2323 err = security_task_getscheduler(p);
2329 seq_printf(m, "%llu\n", p->timer_slack_ns);
2338 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2340 return single_open(filp, timerslack_ns_show, inode);
2343 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2344 .open = timerslack_ns_open,
2346 .write = timerslack_ns_write,
2347 .llseek = seq_lseek,
2348 .release = single_release,
2351 static int proc_pident_instantiate(struct inode *dir,
2352 struct dentry *dentry, struct task_struct *task, const void *ptr)
2354 const struct pid_entry *p = ptr;
2355 struct inode *inode;
2356 struct proc_inode *ei;
2358 inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2363 if (S_ISDIR(inode->i_mode))
2364 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2366 inode->i_op = p->iop;
2368 inode->i_fop = p->fop;
2370 d_set_d_op(dentry, &pid_dentry_operations);
2371 d_add(dentry, inode);
2372 /* Close the race of the process dying before we return the dentry */
2373 if (pid_revalidate(dentry, 0))
2379 static struct dentry *proc_pident_lookup(struct inode *dir,
2380 struct dentry *dentry,
2381 const struct pid_entry *ents,
2385 struct task_struct *task = get_proc_task(dir);
2386 const struct pid_entry *p, *last;
2394 * Yes, it does not scale. And it should not. Don't add
2395 * new entries into /proc/<tgid>/ without very good reasons.
2397 last = &ents[nents];
2398 for (p = ents; p < last; p++) {
2399 if (p->len != dentry->d_name.len)
2401 if (!memcmp(dentry->d_name.name, p->name, p->len))
2407 error = proc_pident_instantiate(dir, dentry, task, p);
2409 put_task_struct(task);
2411 return ERR_PTR(error);
2414 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2415 const struct pid_entry *ents, unsigned int nents)
2417 struct task_struct *task = get_proc_task(file_inode(file));
2418 const struct pid_entry *p;
2423 if (!dir_emit_dots(file, ctx))
2426 if (ctx->pos >= nents + 2)
2429 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2430 if (!proc_fill_cache(file, ctx, p->name, p->len,
2431 proc_pident_instantiate, task, p))
2436 put_task_struct(task);
2440 #ifdef CONFIG_SECURITY
2441 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2442 size_t count, loff_t *ppos)
2444 struct inode * inode = file_inode(file);
2447 struct task_struct *task = get_proc_task(inode);
2452 length = security_getprocattr(task,
2453 (char*)file->f_path.dentry->d_name.name,
2455 put_task_struct(task);
2457 length = simple_read_from_buffer(buf, count, ppos, p, length);
2462 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2463 size_t count, loff_t *ppos)
2465 struct inode * inode = file_inode(file);
2468 struct task_struct *task = get_proc_task(inode);
2474 /* A task may only write its own attributes. */
2476 if (current != task)
2479 if (count > PAGE_SIZE)
2482 /* No partial writes. */
2487 page = memdup_user(buf, count);
2489 length = PTR_ERR(page);
2493 /* Guard against adverse ptrace interaction */
2494 length = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2498 length = security_setprocattr(file->f_path.dentry->d_name.name,
2500 mutex_unlock(¤t->signal->cred_guard_mutex);
2504 put_task_struct(task);
2509 static const struct file_operations proc_pid_attr_operations = {
2510 .read = proc_pid_attr_read,
2511 .write = proc_pid_attr_write,
2512 .llseek = generic_file_llseek,
2515 static const struct pid_entry attr_dir_stuff[] = {
2516 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2517 REG("prev", S_IRUGO, proc_pid_attr_operations),
2518 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2519 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2520 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2521 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2524 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2526 return proc_pident_readdir(file, ctx,
2527 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2530 static const struct file_operations proc_attr_dir_operations = {
2531 .read = generic_read_dir,
2532 .iterate_shared = proc_attr_dir_readdir,
2533 .llseek = generic_file_llseek,
2536 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2537 struct dentry *dentry, unsigned int flags)
2539 return proc_pident_lookup(dir, dentry,
2540 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2543 static const struct inode_operations proc_attr_dir_inode_operations = {
2544 .lookup = proc_attr_dir_lookup,
2545 .getattr = pid_getattr,
2546 .setattr = proc_setattr,
2551 #ifdef CONFIG_ELF_CORE
2552 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2553 size_t count, loff_t *ppos)
2555 struct task_struct *task = get_proc_task(file_inode(file));
2556 struct mm_struct *mm;
2557 char buffer[PROC_NUMBUF];
2565 mm = get_task_mm(task);
2567 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2568 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2569 MMF_DUMP_FILTER_SHIFT));
2571 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2574 put_task_struct(task);
2579 static ssize_t proc_coredump_filter_write(struct file *file,
2580 const char __user *buf,
2584 struct task_struct *task;
2585 struct mm_struct *mm;
2591 ret = kstrtouint_from_user(buf, count, 0, &val);
2596 task = get_proc_task(file_inode(file));
2600 mm = get_task_mm(task);
2605 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2607 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2609 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2614 put_task_struct(task);
2621 static const struct file_operations proc_coredump_filter_operations = {
2622 .read = proc_coredump_filter_read,
2623 .write = proc_coredump_filter_write,
2624 .llseek = generic_file_llseek,
2628 #ifdef CONFIG_TASK_IO_ACCOUNTING
2629 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2631 struct task_io_accounting acct = task->ioac;
2632 unsigned long flags;
2635 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2639 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2644 if (whole && lock_task_sighand(task, &flags)) {
2645 struct task_struct *t = task;
2647 task_io_accounting_add(&acct, &task->signal->ioac);
2648 while_each_thread(task, t)
2649 task_io_accounting_add(&acct, &t->ioac);
2651 unlock_task_sighand(task, &flags);
2658 "read_bytes: %llu\n"
2659 "write_bytes: %llu\n"
2660 "cancelled_write_bytes: %llu\n",
2661 (unsigned long long)acct.rchar,
2662 (unsigned long long)acct.wchar,
2663 (unsigned long long)acct.syscr,
2664 (unsigned long long)acct.syscw,
2665 (unsigned long long)acct.read_bytes,
2666 (unsigned long long)acct.write_bytes,
2667 (unsigned long long)acct.cancelled_write_bytes);
2671 mutex_unlock(&task->signal->cred_guard_mutex);
2675 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2676 struct pid *pid, struct task_struct *task)
2678 return do_io_accounting(task, m, 0);
2681 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2682 struct pid *pid, struct task_struct *task)
2684 return do_io_accounting(task, m, 1);
2686 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2688 #ifdef CONFIG_USER_NS
2689 static int proc_id_map_open(struct inode *inode, struct file *file,
2690 const struct seq_operations *seq_ops)
2692 struct user_namespace *ns = NULL;
2693 struct task_struct *task;
2694 struct seq_file *seq;
2697 task = get_proc_task(inode);
2700 ns = get_user_ns(task_cred_xxx(task, user_ns));
2702 put_task_struct(task);
2707 ret = seq_open(file, seq_ops);
2711 seq = file->private_data;
2721 static int proc_id_map_release(struct inode *inode, struct file *file)
2723 struct seq_file *seq = file->private_data;
2724 struct user_namespace *ns = seq->private;
2726 return seq_release(inode, file);
2729 static int proc_uid_map_open(struct inode *inode, struct file *file)
2731 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2734 static int proc_gid_map_open(struct inode *inode, struct file *file)
2736 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2739 static int proc_projid_map_open(struct inode *inode, struct file *file)
2741 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2744 static const struct file_operations proc_uid_map_operations = {
2745 .open = proc_uid_map_open,
2746 .write = proc_uid_map_write,
2748 .llseek = seq_lseek,
2749 .release = proc_id_map_release,
2752 static const struct file_operations proc_gid_map_operations = {
2753 .open = proc_gid_map_open,
2754 .write = proc_gid_map_write,
2756 .llseek = seq_lseek,
2757 .release = proc_id_map_release,
2760 static const struct file_operations proc_projid_map_operations = {
2761 .open = proc_projid_map_open,
2762 .write = proc_projid_map_write,
2764 .llseek = seq_lseek,
2765 .release = proc_id_map_release,
2768 static int proc_setgroups_open(struct inode *inode, struct file *file)
2770 struct user_namespace *ns = NULL;
2771 struct task_struct *task;
2775 task = get_proc_task(inode);
2778 ns = get_user_ns(task_cred_xxx(task, user_ns));
2780 put_task_struct(task);
2785 if (file->f_mode & FMODE_WRITE) {
2787 if (!ns_capable(ns, CAP_SYS_ADMIN))
2791 ret = single_open(file, &proc_setgroups_show, ns);
2802 static int proc_setgroups_release(struct inode *inode, struct file *file)
2804 struct seq_file *seq = file->private_data;
2805 struct user_namespace *ns = seq->private;
2806 int ret = single_release(inode, file);
2811 static const struct file_operations proc_setgroups_operations = {
2812 .open = proc_setgroups_open,
2813 .write = proc_setgroups_write,
2815 .llseek = seq_lseek,
2816 .release = proc_setgroups_release,
2818 #endif /* CONFIG_USER_NS */
2820 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2821 struct pid *pid, struct task_struct *task)
2823 int err = lock_trace(task);
2825 seq_printf(m, "%08x\n", task->personality);
2834 static const struct file_operations proc_task_operations;
2835 static const struct inode_operations proc_task_inode_operations;
2837 static const struct pid_entry tgid_base_stuff[] = {
2838 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2839 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2840 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2841 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2842 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2844 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2846 REG("environ", S_IRUSR, proc_environ_operations),
2847 REG("auxv", S_IRUSR, proc_auxv_operations),
2848 ONE("status", S_IRUGO, proc_pid_status),
2849 ONE("personality", S_IRUSR, proc_pid_personality),
2850 ONE("limits", S_IRUGO, proc_pid_limits),
2851 #ifdef CONFIG_SCHED_DEBUG
2852 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2854 #ifdef CONFIG_SCHED_AUTOGROUP
2855 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2857 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2858 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2859 ONE("syscall", S_IRUSR, proc_pid_syscall),
2861 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2862 ONE("stat", S_IRUGO, proc_tgid_stat),
2863 ONE("statm", S_IRUGO, proc_pid_statm),
2864 REG("maps", S_IRUGO, proc_pid_maps_operations),
2866 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2868 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2869 LNK("cwd", proc_cwd_link),
2870 LNK("root", proc_root_link),
2871 LNK("exe", proc_exe_link),
2872 REG("mounts", S_IRUGO, proc_mounts_operations),
2873 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2874 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2875 #ifdef CONFIG_PROC_PAGE_MONITOR
2876 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2877 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2878 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2880 #ifdef CONFIG_SECURITY
2881 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2883 #ifdef CONFIG_KALLSYMS
2884 ONE("wchan", S_IRUGO, proc_pid_wchan),
2886 #ifdef CONFIG_STACKTRACE
2887 ONE("stack", S_IRUSR, proc_pid_stack),
2889 #ifdef CONFIG_SCHED_INFO
2890 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2892 #ifdef CONFIG_LATENCYTOP
2893 REG("latency", S_IRUGO, proc_lstats_operations),
2895 #ifdef CONFIG_PROC_PID_CPUSET
2896 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2898 #ifdef CONFIG_CGROUPS
2899 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2901 ONE("oom_score", S_IRUGO, proc_oom_score),
2902 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2903 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2904 #ifdef CONFIG_AUDITSYSCALL
2905 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2906 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2908 #ifdef CONFIG_FAULT_INJECTION
2909 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2911 #ifdef CONFIG_ELF_CORE
2912 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2914 #ifdef CONFIG_TASK_IO_ACCOUNTING
2915 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2917 #ifdef CONFIG_HARDWALL
2918 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2920 #ifdef CONFIG_USER_NS
2921 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2922 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2923 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2924 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2926 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2927 REG("timers", S_IRUGO, proc_timers_operations),
2929 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2932 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2934 return proc_pident_readdir(file, ctx,
2935 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2938 static const struct file_operations proc_tgid_base_operations = {
2939 .read = generic_read_dir,
2940 .iterate_shared = proc_tgid_base_readdir,
2941 .llseek = generic_file_llseek,
2944 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2946 return proc_pident_lookup(dir, dentry,
2947 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2950 static const struct inode_operations proc_tgid_base_inode_operations = {
2951 .lookup = proc_tgid_base_lookup,
2952 .getattr = pid_getattr,
2953 .setattr = proc_setattr,
2954 .permission = proc_pid_permission,
2957 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2959 struct dentry *dentry, *leader, *dir;
2960 char buf[PROC_NUMBUF];
2964 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2965 /* no ->d_hash() rejects on procfs */
2966 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2968 d_invalidate(dentry);
2976 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2977 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2982 name.len = strlen(name.name);
2983 dir = d_hash_and_lookup(leader, &name);
2985 goto out_put_leader;
2988 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2989 dentry = d_hash_and_lookup(dir, &name);
2991 d_invalidate(dentry);
3003 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3004 * @task: task that should be flushed.
3006 * When flushing dentries from proc, one needs to flush them from global
3007 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3008 * in. This call is supposed to do all of this job.
3010 * Looks in the dcache for
3012 * /proc/@tgid/task/@pid
3013 * if either directory is present flushes it and all of it'ts children
3016 * It is safe and reasonable to cache /proc entries for a task until
3017 * that task exits. After that they just clog up the dcache with
3018 * useless entries, possibly causing useful dcache entries to be
3019 * flushed instead. This routine is proved to flush those useless
3020 * dcache entries at process exit time.
3022 * NOTE: This routine is just an optimization so it does not guarantee
3023 * that no dcache entries will exist at process exit time it
3024 * just makes it very unlikely that any will persist.
3027 void proc_flush_task(struct task_struct *task)
3030 struct pid *pid, *tgid;
3033 pid = task_pid(task);
3034 tgid = task_tgid(task);
3036 for (i = 0; i <= pid->level; i++) {
3037 upid = &pid->numbers[i];
3038 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3039 tgid->numbers[i].nr);
3043 static int proc_pid_instantiate(struct inode *dir,
3044 struct dentry * dentry,
3045 struct task_struct *task, const void *ptr)
3047 struct inode *inode;
3049 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3053 inode->i_op = &proc_tgid_base_inode_operations;
3054 inode->i_fop = &proc_tgid_base_operations;
3055 inode->i_flags|=S_IMMUTABLE;
3057 set_nlink(inode, nlink_tgid);
3059 d_set_d_op(dentry, &pid_dentry_operations);
3061 d_add(dentry, inode);
3062 /* Close the race of the process dying before we return the dentry */
3063 if (pid_revalidate(dentry, 0))
3069 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3071 int result = -ENOENT;
3072 struct task_struct *task;
3074 struct pid_namespace *ns;
3076 tgid = name_to_int(&dentry->d_name);
3080 ns = dentry->d_sb->s_fs_info;
3082 task = find_task_by_pid_ns(tgid, ns);
3084 get_task_struct(task);
3089 result = proc_pid_instantiate(dir, dentry, task, NULL);
3090 put_task_struct(task);
3092 return ERR_PTR(result);
3096 * Find the first task with tgid >= tgid
3101 struct task_struct *task;
3103 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3108 put_task_struct(iter.task);
3112 pid = find_ge_pid(iter.tgid, ns);
3114 iter.tgid = pid_nr_ns(pid, ns);
3115 iter.task = pid_task(pid, PIDTYPE_PID);
3116 /* What we to know is if the pid we have find is the
3117 * pid of a thread_group_leader. Testing for task
3118 * being a thread_group_leader is the obvious thing
3119 * todo but there is a window when it fails, due to
3120 * the pid transfer logic in de_thread.
3122 * So we perform the straight forward test of seeing
3123 * if the pid we have found is the pid of a thread
3124 * group leader, and don't worry if the task we have
3125 * found doesn't happen to be a thread group leader.
3126 * As we don't care in the case of readdir.
3128 if (!iter.task || !has_group_leader_pid(iter.task)) {
3132 get_task_struct(iter.task);
3138 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3140 /* for the /proc/ directory itself, after non-process stuff has been done */
3141 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3143 struct tgid_iter iter;
3144 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3145 loff_t pos = ctx->pos;
3147 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3150 if (pos == TGID_OFFSET - 2) {
3151 struct inode *inode = d_inode(ns->proc_self);
3152 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3154 ctx->pos = pos = pos + 1;
3156 if (pos == TGID_OFFSET - 1) {
3157 struct inode *inode = d_inode(ns->proc_thread_self);
3158 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3160 ctx->pos = pos = pos + 1;
3162 iter.tgid = pos - TGID_OFFSET;
3164 for (iter = next_tgid(ns, iter);
3166 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3167 char name[PROC_NUMBUF];
3171 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3174 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3175 ctx->pos = iter.tgid + TGID_OFFSET;
3176 if (!proc_fill_cache(file, ctx, name, len,
3177 proc_pid_instantiate, iter.task, NULL)) {
3178 put_task_struct(iter.task);
3182 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3187 * proc_tid_comm_permission is a special permission function exclusively
3188 * used for the node /proc/<pid>/task/<tid>/comm.
3189 * It bypasses generic permission checks in the case where a task of the same
3190 * task group attempts to access the node.
3191 * The rationale behind this is that glibc and bionic access this node for
3192 * cross thread naming (pthread_set/getname_np(!self)). However, if
3193 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3194 * which locks out the cross thread naming implementation.
3195 * This function makes sure that the node is always accessible for members of
3196 * same thread group.
3198 static int proc_tid_comm_permission(struct inode *inode, int mask)
3200 bool is_same_tgroup;
3201 struct task_struct *task;
3203 task = get_proc_task(inode);
3206 is_same_tgroup = same_thread_group(current, task);
3207 put_task_struct(task);
3209 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3210 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3211 * read or written by the members of the corresponding
3217 return generic_permission(inode, mask);
3220 static const struct inode_operations proc_tid_comm_inode_operations = {
3221 .permission = proc_tid_comm_permission,
3227 static const struct pid_entry tid_base_stuff[] = {
3228 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3229 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3230 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3232 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3234 REG("environ", S_IRUSR, proc_environ_operations),
3235 REG("auxv", S_IRUSR, proc_auxv_operations),
3236 ONE("status", S_IRUGO, proc_pid_status),
3237 ONE("personality", S_IRUSR, proc_pid_personality),
3238 ONE("limits", S_IRUGO, proc_pid_limits),
3239 #ifdef CONFIG_SCHED_DEBUG
3240 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3242 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3243 &proc_tid_comm_inode_operations,
3244 &proc_pid_set_comm_operations, {}),
3245 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3246 ONE("syscall", S_IRUSR, proc_pid_syscall),
3248 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3249 ONE("stat", S_IRUGO, proc_tid_stat),
3250 ONE("statm", S_IRUGO, proc_pid_statm),
3251 REG("maps", S_IRUGO, proc_tid_maps_operations),
3252 #ifdef CONFIG_PROC_CHILDREN
3253 REG("children", S_IRUGO, proc_tid_children_operations),
3256 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3258 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3259 LNK("cwd", proc_cwd_link),
3260 LNK("root", proc_root_link),
3261 LNK("exe", proc_exe_link),
3262 REG("mounts", S_IRUGO, proc_mounts_operations),
3263 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3264 #ifdef CONFIG_PROC_PAGE_MONITOR
3265 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3266 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3267 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3269 #ifdef CONFIG_SECURITY
3270 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3272 #ifdef CONFIG_KALLSYMS
3273 ONE("wchan", S_IRUGO, proc_pid_wchan),
3275 #ifdef CONFIG_STACKTRACE
3276 ONE("stack", S_IRUSR, proc_pid_stack),
3278 #ifdef CONFIG_SCHED_INFO
3279 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3281 #ifdef CONFIG_LATENCYTOP
3282 REG("latency", S_IRUGO, proc_lstats_operations),
3284 #ifdef CONFIG_PROC_PID_CPUSET
3285 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3287 #ifdef CONFIG_CGROUPS
3288 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3290 ONE("oom_score", S_IRUGO, proc_oom_score),
3291 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3292 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3293 #ifdef CONFIG_AUDITSYSCALL
3294 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3295 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3297 #ifdef CONFIG_FAULT_INJECTION
3298 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3300 #ifdef CONFIG_TASK_IO_ACCOUNTING
3301 ONE("io", S_IRUSR, proc_tid_io_accounting),
3303 #ifdef CONFIG_HARDWALL
3304 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3306 #ifdef CONFIG_USER_NS
3307 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3308 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3309 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3310 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3314 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3316 return proc_pident_readdir(file, ctx,
3317 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3320 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3322 return proc_pident_lookup(dir, dentry,
3323 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3326 static const struct file_operations proc_tid_base_operations = {
3327 .read = generic_read_dir,
3328 .iterate_shared = proc_tid_base_readdir,
3329 .llseek = generic_file_llseek,
3332 static const struct inode_operations proc_tid_base_inode_operations = {
3333 .lookup = proc_tid_base_lookup,
3334 .getattr = pid_getattr,
3335 .setattr = proc_setattr,
3338 static int proc_task_instantiate(struct inode *dir,
3339 struct dentry *dentry, struct task_struct *task, const void *ptr)
3341 struct inode *inode;
3342 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3346 inode->i_op = &proc_tid_base_inode_operations;
3347 inode->i_fop = &proc_tid_base_operations;
3348 inode->i_flags|=S_IMMUTABLE;
3350 set_nlink(inode, nlink_tid);
3352 d_set_d_op(dentry, &pid_dentry_operations);
3354 d_add(dentry, inode);
3355 /* Close the race of the process dying before we return the dentry */
3356 if (pid_revalidate(dentry, 0))
3362 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3364 int result = -ENOENT;
3365 struct task_struct *task;
3366 struct task_struct *leader = get_proc_task(dir);
3368 struct pid_namespace *ns;
3373 tid = name_to_int(&dentry->d_name);
3377 ns = dentry->d_sb->s_fs_info;
3379 task = find_task_by_pid_ns(tid, ns);
3381 get_task_struct(task);
3385 if (!same_thread_group(leader, task))
3388 result = proc_task_instantiate(dir, dentry, task, NULL);
3390 put_task_struct(task);
3392 put_task_struct(leader);
3394 return ERR_PTR(result);
3398 * Find the first tid of a thread group to return to user space.
3400 * Usually this is just the thread group leader, but if the users
3401 * buffer was too small or there was a seek into the middle of the
3402 * directory we have more work todo.
3404 * In the case of a short read we start with find_task_by_pid.
3406 * In the case of a seek we start with the leader and walk nr
3409 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3410 struct pid_namespace *ns)
3412 struct task_struct *pos, *task;
3413 unsigned long nr = f_pos;
3415 if (nr != f_pos) /* 32bit overflow? */
3419 task = pid_task(pid, PIDTYPE_PID);
3423 /* Attempt to start with the tid of a thread */
3425 pos = find_task_by_pid_ns(tid, ns);
3426 if (pos && same_thread_group(pos, task))
3430 /* If nr exceeds the number of threads there is nothing todo */
3431 if (nr >= get_nr_threads(task))
3434 /* If we haven't found our starting place yet start
3435 * with the leader and walk nr threads forward.
3437 pos = task = task->group_leader;
3441 } while_each_thread(task, pos);
3446 get_task_struct(pos);
3453 * Find the next thread in the thread list.
3454 * Return NULL if there is an error or no next thread.
3456 * The reference to the input task_struct is released.
3458 static struct task_struct *next_tid(struct task_struct *start)
3460 struct task_struct *pos = NULL;
3462 if (pid_alive(start)) {
3463 pos = next_thread(start);
3464 if (thread_group_leader(pos))
3467 get_task_struct(pos);
3470 put_task_struct(start);
3474 /* for the /proc/TGID/task/ directories */
3475 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3477 struct inode *inode = file_inode(file);
3478 struct task_struct *task;
3479 struct pid_namespace *ns;
3482 if (proc_inode_is_dead(inode))
3485 if (!dir_emit_dots(file, ctx))
3488 /* f_version caches the tgid value that the last readdir call couldn't
3489 * return. lseek aka telldir automagically resets f_version to 0.
3491 ns = inode->i_sb->s_fs_info;
3492 tid = (int)file->f_version;
3493 file->f_version = 0;
3494 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3496 task = next_tid(task), ctx->pos++) {
3497 char name[PROC_NUMBUF];
3499 tid = task_pid_nr_ns(task, ns);
3500 len = snprintf(name, sizeof(name), "%d", tid);
3501 if (!proc_fill_cache(file, ctx, name, len,
3502 proc_task_instantiate, task, NULL)) {
3503 /* returning this tgid failed, save it as the first
3504 * pid for the next readir call */
3505 file->f_version = (u64)tid;
3506 put_task_struct(task);
3514 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3516 struct inode *inode = d_inode(dentry);
3517 struct task_struct *p = get_proc_task(inode);
3518 generic_fillattr(inode, stat);
3521 stat->nlink += get_nr_threads(p);
3528 static const struct inode_operations proc_task_inode_operations = {
3529 .lookup = proc_task_lookup,
3530 .getattr = proc_task_getattr,
3531 .setattr = proc_setattr,
3532 .permission = proc_pid_permission,
3535 static const struct file_operations proc_task_operations = {
3536 .read = generic_read_dir,
3537 .iterate_shared = proc_task_readdir,
3538 .llseek = generic_file_llseek,
3541 void __init set_proc_pid_nlink(void)
3543 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3544 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));