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 <asm/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.
111 const struct inode_operations *iop;
112 const struct file_operations *fop;
116 #define NOD(NAME, MODE, IOP, FOP, OP) { \
118 .len = sizeof(NAME) - 1, \
125 #define DIR(NAME, MODE, iops, fops) \
126 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
127 #define LNK(NAME, get_link) \
128 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
129 &proc_pid_link_inode_operations, NULL, \
130 { .proc_get_link = get_link } )
131 #define REG(NAME, MODE, fops) \
132 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
133 #define ONE(NAME, MODE, show) \
134 NOD(NAME, (S_IFREG|(MODE)), \
135 NULL, &proc_single_file_operations, \
136 { .proc_show = show } )
139 * Count the number of hardlinks for the pid_entry table, excluding the .
142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
149 for (i = 0; i < n; ++i) {
150 if (S_ISDIR(entries[i].mode))
157 static int get_task_root(struct task_struct *task, struct path *root)
159 int result = -ENOENT;
163 get_fs_root(task->fs, root);
170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
172 struct task_struct *task = get_proc_task(d_inode(dentry));
173 int result = -ENOENT;
178 get_fs_pwd(task->fs, path);
182 put_task_struct(task);
187 static int proc_root_link(struct dentry *dentry, struct path *path)
189 struct task_struct *task = get_proc_task(d_inode(dentry));
190 int result = -ENOENT;
193 result = get_task_root(task, path);
194 put_task_struct(task);
199 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
200 size_t _count, loff_t *pos)
202 struct task_struct *tsk;
203 struct mm_struct *mm;
205 unsigned long count = _count;
206 unsigned long arg_start, arg_end, env_start, env_end;
207 unsigned long len1, len2, len;
214 tsk = get_proc_task(file_inode(file));
217 mm = get_task_mm(tsk);
218 put_task_struct(tsk);
221 /* Check if process spawned far enough to have cmdline. */
227 page = (char *)__get_free_page(GFP_TEMPORARY);
233 down_read(&mm->mmap_sem);
234 arg_start = mm->arg_start;
235 arg_end = mm->arg_end;
236 env_start = mm->env_start;
237 env_end = mm->env_end;
238 up_read(&mm->mmap_sem);
240 BUG_ON(arg_start > arg_end);
241 BUG_ON(env_start > env_end);
243 len1 = arg_end - arg_start;
244 len2 = env_end - env_start;
252 * Inherently racy -- command line shares address space
253 * with code and data.
255 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
262 /* Command line (set of strings) occupies whole ARGV. */
266 p = arg_start + *pos;
268 while (count > 0 && len > 0) {
272 _count = min3(count, len, PAGE_SIZE);
273 nr_read = access_remote_vm(mm, p, page, _count, 0);
279 if (copy_to_user(buf, page, nr_read)) {
292 * Command line (1 string) occupies ARGV and maybe
295 if (len1 + len2 <= *pos)
300 p = arg_start + *pos;
302 while (count > 0 && len > 0) {
303 unsigned int _count, l;
307 _count = min3(count, len, PAGE_SIZE);
308 nr_read = access_remote_vm(mm, p, page, _count, 0);
315 * Command line can be shorter than whole ARGV
316 * even if last "marker" byte says it is not.
319 l = strnlen(page, nr_read);
325 if (copy_to_user(buf, page, nr_read)) {
341 * Command line (1 string) occupies ARGV and
345 p = env_start + *pos - len1;
346 len = len1 + len2 - *pos;
351 while (count > 0 && len > 0) {
352 unsigned int _count, l;
356 _count = min3(count, len, PAGE_SIZE);
357 nr_read = access_remote_vm(mm, p, page, _count, 0);
365 l = strnlen(page, nr_read);
371 if (copy_to_user(buf, page, nr_read)) {
390 free_page((unsigned long)page);
398 static const struct file_operations proc_pid_cmdline_ops = {
399 .read = proc_pid_cmdline_read,
400 .llseek = generic_file_llseek,
403 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
404 struct pid *pid, struct task_struct *task)
406 struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
407 if (mm && !IS_ERR(mm)) {
408 unsigned int nwords = 0;
411 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
412 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
420 #ifdef CONFIG_KALLSYMS
422 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
423 * Returns the resolved symbol. If that fails, simply return the address.
425 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
426 struct pid *pid, struct task_struct *task)
429 char symname[KSYM_NAME_LEN];
431 wchan = get_wchan(task);
433 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
434 && !lookup_symbol_name(wchan, symname))
435 seq_printf(m, "%s", symname);
441 #endif /* CONFIG_KALLSYMS */
443 static int lock_trace(struct task_struct *task)
445 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
448 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
449 mutex_unlock(&task->signal->cred_guard_mutex);
455 static void unlock_trace(struct task_struct *task)
457 mutex_unlock(&task->signal->cred_guard_mutex);
460 #ifdef CONFIG_STACKTRACE
462 #define MAX_STACK_TRACE_DEPTH 64
464 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
465 struct pid *pid, struct task_struct *task)
467 struct stack_trace trace;
468 unsigned long *entries;
472 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
476 trace.nr_entries = 0;
477 trace.max_entries = MAX_STACK_TRACE_DEPTH;
478 trace.entries = entries;
481 err = lock_trace(task);
483 save_stack_trace_tsk(task, &trace);
485 for (i = 0; i < trace.nr_entries; i++) {
486 seq_printf(m, "[<%pK>] %pB\n",
487 (void *)entries[i], (void *)entries[i]);
497 #ifdef CONFIG_SCHED_INFO
499 * Provides /proc/PID/schedstat
501 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
502 struct pid *pid, struct task_struct *task)
504 if (unlikely(!sched_info_on()))
505 seq_printf(m, "0 0 0\n");
507 seq_printf(m, "%llu %llu %lu\n",
508 (unsigned long long)task->se.sum_exec_runtime,
509 (unsigned long long)task->sched_info.run_delay,
510 task->sched_info.pcount);
516 #ifdef CONFIG_LATENCYTOP
517 static int lstats_show_proc(struct seq_file *m, void *v)
520 struct inode *inode = m->private;
521 struct task_struct *task = get_proc_task(inode);
525 seq_puts(m, "Latency Top version : v0.1\n");
526 for (i = 0; i < 32; i++) {
527 struct latency_record *lr = &task->latency_record[i];
528 if (lr->backtrace[0]) {
530 seq_printf(m, "%i %li %li",
531 lr->count, lr->time, lr->max);
532 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
533 unsigned long bt = lr->backtrace[q];
538 seq_printf(m, " %ps", (void *)bt);
544 put_task_struct(task);
548 static int lstats_open(struct inode *inode, struct file *file)
550 return single_open(file, lstats_show_proc, inode);
553 static ssize_t lstats_write(struct file *file, const char __user *buf,
554 size_t count, loff_t *offs)
556 struct task_struct *task = get_proc_task(file_inode(file));
560 clear_all_latency_tracing(task);
561 put_task_struct(task);
566 static const struct file_operations proc_lstats_operations = {
569 .write = lstats_write,
571 .release = single_release,
576 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
577 struct pid *pid, struct task_struct *task)
579 unsigned long totalpages = totalram_pages + total_swap_pages;
580 unsigned long points = 0;
582 points = oom_badness(task, NULL, NULL, totalpages) *
584 seq_printf(m, "%lu\n", points);
594 static const struct limit_names lnames[RLIM_NLIMITS] = {
595 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
596 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
597 [RLIMIT_DATA] = {"Max data size", "bytes"},
598 [RLIMIT_STACK] = {"Max stack size", "bytes"},
599 [RLIMIT_CORE] = {"Max core file size", "bytes"},
600 [RLIMIT_RSS] = {"Max resident set", "bytes"},
601 [RLIMIT_NPROC] = {"Max processes", "processes"},
602 [RLIMIT_NOFILE] = {"Max open files", "files"},
603 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
604 [RLIMIT_AS] = {"Max address space", "bytes"},
605 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
606 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
607 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
608 [RLIMIT_NICE] = {"Max nice priority", NULL},
609 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
610 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
613 /* Display limits for a process */
614 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
615 struct pid *pid, struct task_struct *task)
620 struct rlimit rlim[RLIM_NLIMITS];
622 if (!lock_task_sighand(task, &flags))
624 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
625 unlock_task_sighand(task, &flags);
628 * print the file header
630 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
631 "Limit", "Soft Limit", "Hard Limit", "Units");
633 for (i = 0; i < RLIM_NLIMITS; i++) {
634 if (rlim[i].rlim_cur == RLIM_INFINITY)
635 seq_printf(m, "%-25s %-20s ",
636 lnames[i].name, "unlimited");
638 seq_printf(m, "%-25s %-20lu ",
639 lnames[i].name, rlim[i].rlim_cur);
641 if (rlim[i].rlim_max == RLIM_INFINITY)
642 seq_printf(m, "%-20s ", "unlimited");
644 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
647 seq_printf(m, "%-10s\n", lnames[i].unit);
655 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
656 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
657 struct pid *pid, struct task_struct *task)
660 unsigned long args[6], sp, pc;
663 res = lock_trace(task);
667 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
668 seq_puts(m, "running\n");
670 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
673 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
675 args[0], args[1], args[2], args[3], args[4], args[5],
681 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
683 /************************************************************************/
684 /* Here the fs part begins */
685 /************************************************************************/
687 /* permission checks */
688 static int proc_fd_access_allowed(struct inode *inode)
690 struct task_struct *task;
692 /* Allow access to a task's file descriptors if it is us or we
693 * may use ptrace attach to the process and find out that
696 task = get_proc_task(inode);
698 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
699 put_task_struct(task);
704 int proc_setattr(struct dentry *dentry, struct iattr *attr)
707 struct inode *inode = d_inode(dentry);
709 if (attr->ia_valid & ATTR_MODE)
712 error = inode_change_ok(inode, attr);
716 setattr_copy(inode, attr);
717 mark_inode_dirty(inode);
722 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
723 * or euid/egid (for hide_pid_min=2)?
725 static bool has_pid_permissions(struct pid_namespace *pid,
726 struct task_struct *task,
729 if (pid->hide_pid < hide_pid_min)
731 if (in_group_p(pid->pid_gid))
733 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
737 static int proc_pid_permission(struct inode *inode, int mask)
739 struct pid_namespace *pid = inode->i_sb->s_fs_info;
740 struct task_struct *task;
743 task = get_proc_task(inode);
746 has_perms = has_pid_permissions(pid, task, 1);
747 put_task_struct(task);
750 if (pid->hide_pid == 2) {
752 * Let's make getdents(), stat(), and open()
753 * consistent with each other. If a process
754 * may not stat() a file, it shouldn't be seen
762 return generic_permission(inode, mask);
767 static const struct inode_operations proc_def_inode_operations = {
768 .setattr = proc_setattr,
771 static int proc_single_show(struct seq_file *m, void *v)
773 struct inode *inode = m->private;
774 struct pid_namespace *ns;
776 struct task_struct *task;
779 ns = inode->i_sb->s_fs_info;
780 pid = proc_pid(inode);
781 task = get_pid_task(pid, PIDTYPE_PID);
785 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
787 put_task_struct(task);
791 static int proc_single_open(struct inode *inode, struct file *filp)
793 return single_open(filp, proc_single_show, inode);
796 static const struct file_operations proc_single_file_operations = {
797 .open = proc_single_open,
800 .release = single_release,
804 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
806 struct task_struct *task = get_proc_task(inode);
807 struct mm_struct *mm = ERR_PTR(-ESRCH);
810 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
811 put_task_struct(task);
813 if (!IS_ERR_OR_NULL(mm)) {
814 /* ensure this mm_struct can't be freed */
815 atomic_inc(&mm->mm_count);
816 /* but do not pin its memory */
824 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
826 struct mm_struct *mm = proc_mem_open(inode, mode);
831 file->private_data = mm;
835 static int mem_open(struct inode *inode, struct file *file)
837 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
839 /* OK to pass negative loff_t, we can catch out-of-range */
840 file->f_mode |= FMODE_UNSIGNED_OFFSET;
845 static ssize_t mem_rw(struct file *file, char __user *buf,
846 size_t count, loff_t *ppos, int write)
848 struct mm_struct *mm = file->private_data;
849 unsigned long addr = *ppos;
856 page = (char *)__get_free_page(GFP_TEMPORARY);
861 if (!atomic_inc_not_zero(&mm->mm_users))
865 int this_len = min_t(int, count, PAGE_SIZE);
867 if (write && copy_from_user(page, buf, this_len)) {
872 this_len = access_remote_vm(mm, addr, page, this_len, write);
879 if (!write && copy_to_user(buf, page, this_len)) {
893 free_page((unsigned long) page);
897 static ssize_t mem_read(struct file *file, char __user *buf,
898 size_t count, loff_t *ppos)
900 return mem_rw(file, buf, count, ppos, 0);
903 static ssize_t mem_write(struct file *file, const char __user *buf,
904 size_t count, loff_t *ppos)
906 return mem_rw(file, (char __user*)buf, count, ppos, 1);
909 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
913 file->f_pos = offset;
916 file->f_pos += offset;
921 force_successful_syscall_return();
925 static int mem_release(struct inode *inode, struct file *file)
927 struct mm_struct *mm = file->private_data;
933 static const struct file_operations proc_mem_operations = {
938 .release = mem_release,
941 static int environ_open(struct inode *inode, struct file *file)
943 return __mem_open(inode, file, PTRACE_MODE_READ);
946 static ssize_t environ_read(struct file *file, char __user *buf,
947 size_t count, loff_t *ppos)
950 unsigned long src = *ppos;
952 struct mm_struct *mm = file->private_data;
953 unsigned long env_start, env_end;
955 /* Ensure the process spawned far enough to have an environment. */
956 if (!mm || !mm->env_end)
959 page = (char *)__get_free_page(GFP_TEMPORARY);
964 if (!atomic_inc_not_zero(&mm->mm_users))
967 down_read(&mm->mmap_sem);
968 env_start = mm->env_start;
969 env_end = mm->env_end;
970 up_read(&mm->mmap_sem);
973 size_t this_len, max_len;
976 if (src >= (env_end - env_start))
979 this_len = env_end - (env_start + src);
981 max_len = min_t(size_t, PAGE_SIZE, count);
982 this_len = min(max_len, this_len);
984 retval = access_remote_vm(mm, (env_start + src),
992 if (copy_to_user(buf, page, retval)) {
1006 free_page((unsigned long) page);
1010 static const struct file_operations proc_environ_operations = {
1011 .open = environ_open,
1012 .read = environ_read,
1013 .llseek = generic_file_llseek,
1014 .release = mem_release,
1017 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1020 struct task_struct *task = get_proc_task(file_inode(file));
1021 char buffer[PROC_NUMBUF];
1022 int oom_adj = OOM_ADJUST_MIN;
1027 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1028 oom_adj = OOM_ADJUST_MAX;
1030 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1032 put_task_struct(task);
1033 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1034 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1037 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1039 static DEFINE_MUTEX(oom_adj_mutex);
1040 struct mm_struct *mm = NULL;
1041 struct task_struct *task;
1044 task = get_proc_task(file_inode(file));
1048 mutex_lock(&oom_adj_mutex);
1050 if (oom_adj < task->signal->oom_score_adj &&
1051 !capable(CAP_SYS_RESOURCE)) {
1056 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1057 * /proc/pid/oom_score_adj instead.
1059 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1060 current->comm, task_pid_nr(current), task_pid_nr(task),
1063 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1064 !capable(CAP_SYS_RESOURCE)) {
1071 * Make sure we will check other processes sharing the mm if this is
1072 * not vfrok which wants its own oom_score_adj.
1073 * pin the mm so it doesn't go away and get reused after task_unlock
1075 if (!task->vfork_done) {
1076 struct task_struct *p = find_lock_task_mm(task);
1079 if (atomic_read(&p->mm->mm_users) > 1) {
1081 atomic_inc(&mm->mm_count);
1087 task->signal->oom_score_adj = oom_adj;
1088 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1089 task->signal->oom_score_adj_min = (short)oom_adj;
1090 trace_oom_score_adj_update(task);
1093 struct task_struct *p;
1096 for_each_process(p) {
1097 if (same_thread_group(task, p))
1100 /* do not touch kernel threads or the global init */
1101 if (p->flags & PF_KTHREAD || is_global_init(p))
1105 if (!p->vfork_done && process_shares_mm(p, mm)) {
1106 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",
1107 task_pid_nr(p), p->comm,
1108 p->signal->oom_score_adj, oom_adj,
1109 task_pid_nr(task), task->comm);
1110 p->signal->oom_score_adj = oom_adj;
1111 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1112 p->signal->oom_score_adj_min = (short)oom_adj;
1120 mutex_unlock(&oom_adj_mutex);
1121 put_task_struct(task);
1126 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1127 * kernels. The effective policy is defined by oom_score_adj, which has a
1128 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1129 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1130 * Processes that become oom disabled via oom_adj will still be oom disabled
1131 * with this implementation.
1133 * oom_adj cannot be removed since existing userspace binaries use it.
1135 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1136 size_t count, loff_t *ppos)
1138 char buffer[PROC_NUMBUF];
1142 memset(buffer, 0, sizeof(buffer));
1143 if (count > sizeof(buffer) - 1)
1144 count = sizeof(buffer) - 1;
1145 if (copy_from_user(buffer, buf, count)) {
1150 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1153 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1154 oom_adj != OOM_DISABLE) {
1160 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1161 * value is always attainable.
1163 if (oom_adj == OOM_ADJUST_MAX)
1164 oom_adj = OOM_SCORE_ADJ_MAX;
1166 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1168 err = __set_oom_adj(file, oom_adj, true);
1170 return err < 0 ? err : count;
1173 static const struct file_operations proc_oom_adj_operations = {
1174 .read = oom_adj_read,
1175 .write = oom_adj_write,
1176 .llseek = generic_file_llseek,
1179 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1180 size_t count, loff_t *ppos)
1182 struct task_struct *task = get_proc_task(file_inode(file));
1183 char buffer[PROC_NUMBUF];
1184 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1189 oom_score_adj = task->signal->oom_score_adj;
1190 put_task_struct(task);
1191 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1192 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1195 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1196 size_t count, loff_t *ppos)
1198 char buffer[PROC_NUMBUF];
1202 memset(buffer, 0, sizeof(buffer));
1203 if (count > sizeof(buffer) - 1)
1204 count = sizeof(buffer) - 1;
1205 if (copy_from_user(buffer, buf, count)) {
1210 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1213 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1214 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1219 err = __set_oom_adj(file, oom_score_adj, false);
1221 return err < 0 ? err : count;
1224 static const struct file_operations proc_oom_score_adj_operations = {
1225 .read = oom_score_adj_read,
1226 .write = oom_score_adj_write,
1227 .llseek = default_llseek,
1230 #ifdef CONFIG_AUDITSYSCALL
1231 #define TMPBUFLEN 21
1232 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1233 size_t count, loff_t *ppos)
1235 struct inode * inode = file_inode(file);
1236 struct task_struct *task = get_proc_task(inode);
1238 char tmpbuf[TMPBUFLEN];
1242 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1243 from_kuid(file->f_cred->user_ns,
1244 audit_get_loginuid(task)));
1245 put_task_struct(task);
1246 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1249 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1250 size_t count, loff_t *ppos)
1252 struct inode * inode = file_inode(file);
1258 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1265 /* No partial writes. */
1269 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1273 /* is userspace tring to explicitly UNSET the loginuid? */
1274 if (loginuid == AUDIT_UID_UNSET) {
1275 kloginuid = INVALID_UID;
1277 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1278 if (!uid_valid(kloginuid))
1282 rv = audit_set_loginuid(kloginuid);
1288 static const struct file_operations proc_loginuid_operations = {
1289 .read = proc_loginuid_read,
1290 .write = proc_loginuid_write,
1291 .llseek = generic_file_llseek,
1294 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1295 size_t count, loff_t *ppos)
1297 struct inode * inode = file_inode(file);
1298 struct task_struct *task = get_proc_task(inode);
1300 char tmpbuf[TMPBUFLEN];
1304 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1305 audit_get_sessionid(task));
1306 put_task_struct(task);
1307 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1310 static const struct file_operations proc_sessionid_operations = {
1311 .read = proc_sessionid_read,
1312 .llseek = generic_file_llseek,
1316 #ifdef CONFIG_FAULT_INJECTION
1317 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1318 size_t count, loff_t *ppos)
1320 struct task_struct *task = get_proc_task(file_inode(file));
1321 char buffer[PROC_NUMBUF];
1327 make_it_fail = task->make_it_fail;
1328 put_task_struct(task);
1330 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1332 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1335 static ssize_t proc_fault_inject_write(struct file * file,
1336 const char __user * buf, size_t count, loff_t *ppos)
1338 struct task_struct *task;
1339 char buffer[PROC_NUMBUF];
1343 if (!capable(CAP_SYS_RESOURCE))
1345 memset(buffer, 0, sizeof(buffer));
1346 if (count > sizeof(buffer) - 1)
1347 count = sizeof(buffer) - 1;
1348 if (copy_from_user(buffer, buf, count))
1350 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1353 if (make_it_fail < 0 || make_it_fail > 1)
1356 task = get_proc_task(file_inode(file));
1359 task->make_it_fail = make_it_fail;
1360 put_task_struct(task);
1365 static const struct file_operations proc_fault_inject_operations = {
1366 .read = proc_fault_inject_read,
1367 .write = proc_fault_inject_write,
1368 .llseek = generic_file_llseek,
1373 #ifdef CONFIG_SCHED_DEBUG
1375 * Print out various scheduling related per-task fields:
1377 static int sched_show(struct seq_file *m, void *v)
1379 struct inode *inode = m->private;
1380 struct task_struct *p;
1382 p = get_proc_task(inode);
1385 proc_sched_show_task(p, m);
1393 sched_write(struct file *file, const char __user *buf,
1394 size_t count, loff_t *offset)
1396 struct inode *inode = file_inode(file);
1397 struct task_struct *p;
1399 p = get_proc_task(inode);
1402 proc_sched_set_task(p);
1409 static int sched_open(struct inode *inode, struct file *filp)
1411 return single_open(filp, sched_show, inode);
1414 static const struct file_operations proc_pid_sched_operations = {
1417 .write = sched_write,
1418 .llseek = seq_lseek,
1419 .release = single_release,
1424 #ifdef CONFIG_SCHED_AUTOGROUP
1426 * Print out autogroup related information:
1428 static int sched_autogroup_show(struct seq_file *m, void *v)
1430 struct inode *inode = m->private;
1431 struct task_struct *p;
1433 p = get_proc_task(inode);
1436 proc_sched_autogroup_show_task(p, m);
1444 sched_autogroup_write(struct file *file, const char __user *buf,
1445 size_t count, loff_t *offset)
1447 struct inode *inode = file_inode(file);
1448 struct task_struct *p;
1449 char buffer[PROC_NUMBUF];
1453 memset(buffer, 0, sizeof(buffer));
1454 if (count > sizeof(buffer) - 1)
1455 count = sizeof(buffer) - 1;
1456 if (copy_from_user(buffer, buf, count))
1459 err = kstrtoint(strstrip(buffer), 0, &nice);
1463 p = get_proc_task(inode);
1467 err = proc_sched_autogroup_set_nice(p, nice);
1476 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1480 ret = single_open(filp, sched_autogroup_show, NULL);
1482 struct seq_file *m = filp->private_data;
1489 static const struct file_operations proc_pid_sched_autogroup_operations = {
1490 .open = sched_autogroup_open,
1492 .write = sched_autogroup_write,
1493 .llseek = seq_lseek,
1494 .release = single_release,
1497 #endif /* CONFIG_SCHED_AUTOGROUP */
1499 static ssize_t comm_write(struct file *file, const char __user *buf,
1500 size_t count, loff_t *offset)
1502 struct inode *inode = file_inode(file);
1503 struct task_struct *p;
1504 char buffer[TASK_COMM_LEN];
1505 const size_t maxlen = sizeof(buffer) - 1;
1507 memset(buffer, 0, sizeof(buffer));
1508 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1511 p = get_proc_task(inode);
1515 if (same_thread_group(current, p))
1516 set_task_comm(p, buffer);
1525 static int comm_show(struct seq_file *m, void *v)
1527 struct inode *inode = m->private;
1528 struct task_struct *p;
1530 p = get_proc_task(inode);
1535 seq_printf(m, "%s\n", p->comm);
1543 static int comm_open(struct inode *inode, struct file *filp)
1545 return single_open(filp, comm_show, inode);
1548 static const struct file_operations proc_pid_set_comm_operations = {
1551 .write = comm_write,
1552 .llseek = seq_lseek,
1553 .release = single_release,
1556 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1558 struct task_struct *task;
1559 struct file *exe_file;
1561 task = get_proc_task(d_inode(dentry));
1564 exe_file = get_task_exe_file(task);
1565 put_task_struct(task);
1567 *exe_path = exe_file->f_path;
1568 path_get(&exe_file->f_path);
1575 static const char *proc_pid_get_link(struct dentry *dentry,
1576 struct inode *inode,
1577 struct delayed_call *done)
1580 int error = -EACCES;
1583 return ERR_PTR(-ECHILD);
1585 /* Are we allowed to snoop on the tasks file descriptors? */
1586 if (!proc_fd_access_allowed(inode))
1589 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1593 nd_jump_link(&path);
1596 return ERR_PTR(error);
1599 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1601 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1608 pathname = d_path(path, tmp, PAGE_SIZE);
1609 len = PTR_ERR(pathname);
1610 if (IS_ERR(pathname))
1612 len = tmp + PAGE_SIZE - 1 - pathname;
1616 if (copy_to_user(buffer, pathname, len))
1619 free_page((unsigned long)tmp);
1623 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1625 int error = -EACCES;
1626 struct inode *inode = d_inode(dentry);
1629 /* Are we allowed to snoop on the tasks file descriptors? */
1630 if (!proc_fd_access_allowed(inode))
1633 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1637 error = do_proc_readlink(&path, buffer, buflen);
1643 const struct inode_operations proc_pid_link_inode_operations = {
1644 .readlink = proc_pid_readlink,
1645 .get_link = proc_pid_get_link,
1646 .setattr = proc_setattr,
1650 /* building an inode */
1652 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1654 struct inode * inode;
1655 struct proc_inode *ei;
1656 const struct cred *cred;
1658 /* We need a new inode */
1660 inode = new_inode(sb);
1666 inode->i_ino = get_next_ino();
1667 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1668 inode->i_op = &proc_def_inode_operations;
1671 * grab the reference to task.
1673 ei->pid = get_task_pid(task, PIDTYPE_PID);
1677 if (task_dumpable(task)) {
1679 cred = __task_cred(task);
1680 inode->i_uid = cred->euid;
1681 inode->i_gid = cred->egid;
1684 security_task_to_inode(task, inode);
1694 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1696 struct inode *inode = d_inode(dentry);
1697 struct task_struct *task;
1698 const struct cred *cred;
1699 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1701 generic_fillattr(inode, stat);
1704 stat->uid = GLOBAL_ROOT_UID;
1705 stat->gid = GLOBAL_ROOT_GID;
1706 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1708 if (!has_pid_permissions(pid, task, 2)) {
1711 * This doesn't prevent learning whether PID exists,
1712 * it only makes getattr() consistent with readdir().
1716 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1717 task_dumpable(task)) {
1718 cred = __task_cred(task);
1719 stat->uid = cred->euid;
1720 stat->gid = cred->egid;
1730 * Exceptional case: normally we are not allowed to unhash a busy
1731 * directory. In this case, however, we can do it - no aliasing problems
1732 * due to the way we treat inodes.
1734 * Rewrite the inode's ownerships here because the owning task may have
1735 * performed a setuid(), etc.
1737 * Before the /proc/pid/status file was created the only way to read
1738 * the effective uid of a /process was to stat /proc/pid. Reading
1739 * /proc/pid/status is slow enough that procps and other packages
1740 * kept stating /proc/pid. To keep the rules in /proc simple I have
1741 * made this apply to all per process world readable and executable
1744 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1746 struct inode *inode;
1747 struct task_struct *task;
1748 const struct cred *cred;
1750 if (flags & LOOKUP_RCU)
1753 inode = d_inode(dentry);
1754 task = get_proc_task(inode);
1757 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1758 task_dumpable(task)) {
1760 cred = __task_cred(task);
1761 inode->i_uid = cred->euid;
1762 inode->i_gid = cred->egid;
1765 inode->i_uid = GLOBAL_ROOT_UID;
1766 inode->i_gid = GLOBAL_ROOT_GID;
1768 inode->i_mode &= ~(S_ISUID | S_ISGID);
1769 security_task_to_inode(task, inode);
1770 put_task_struct(task);
1776 static inline bool proc_inode_is_dead(struct inode *inode)
1778 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1781 int pid_delete_dentry(const struct dentry *dentry)
1783 /* Is the task we represent dead?
1784 * If so, then don't put the dentry on the lru list,
1785 * kill it immediately.
1787 return proc_inode_is_dead(d_inode(dentry));
1790 const struct dentry_operations pid_dentry_operations =
1792 .d_revalidate = pid_revalidate,
1793 .d_delete = pid_delete_dentry,
1799 * Fill a directory entry.
1801 * If possible create the dcache entry and derive our inode number and
1802 * file type from dcache entry.
1804 * Since all of the proc inode numbers are dynamically generated, the inode
1805 * numbers do not exist until the inode is cache. This means creating the
1806 * the dcache entry in readdir is necessary to keep the inode numbers
1807 * reported by readdir in sync with the inode numbers reported
1810 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1811 const char *name, int len,
1812 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1814 struct dentry *child, *dir = file->f_path.dentry;
1815 struct qstr qname = QSTR_INIT(name, len);
1816 struct inode *inode;
1820 child = d_hash_and_lookup(dir, &qname);
1822 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1823 child = d_alloc_parallel(dir, &qname, &wq);
1825 goto end_instantiate;
1826 if (d_in_lookup(child)) {
1827 int err = instantiate(d_inode(dir), child, task, ptr);
1828 d_lookup_done(child);
1831 goto end_instantiate;
1835 inode = d_inode(child);
1837 type = inode->i_mode >> 12;
1839 return dir_emit(ctx, name, len, ino, type);
1842 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1846 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1847 * which represent vma start and end addresses.
1849 static int dname_to_vma_addr(struct dentry *dentry,
1850 unsigned long *start, unsigned long *end)
1852 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1858 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1860 unsigned long vm_start, vm_end;
1861 bool exact_vma_exists = false;
1862 struct mm_struct *mm = NULL;
1863 struct task_struct *task;
1864 const struct cred *cred;
1865 struct inode *inode;
1868 if (flags & LOOKUP_RCU)
1871 inode = d_inode(dentry);
1872 task = get_proc_task(inode);
1876 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1877 if (IS_ERR_OR_NULL(mm))
1880 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1881 down_read(&mm->mmap_sem);
1882 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1883 up_read(&mm->mmap_sem);
1888 if (exact_vma_exists) {
1889 if (task_dumpable(task)) {
1891 cred = __task_cred(task);
1892 inode->i_uid = cred->euid;
1893 inode->i_gid = cred->egid;
1896 inode->i_uid = GLOBAL_ROOT_UID;
1897 inode->i_gid = GLOBAL_ROOT_GID;
1899 security_task_to_inode(task, inode);
1904 put_task_struct(task);
1910 static const struct dentry_operations tid_map_files_dentry_operations = {
1911 .d_revalidate = map_files_d_revalidate,
1912 .d_delete = pid_delete_dentry,
1915 static int map_files_get_link(struct dentry *dentry, struct path *path)
1917 unsigned long vm_start, vm_end;
1918 struct vm_area_struct *vma;
1919 struct task_struct *task;
1920 struct mm_struct *mm;
1924 task = get_proc_task(d_inode(dentry));
1928 mm = get_task_mm(task);
1929 put_task_struct(task);
1933 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1938 down_read(&mm->mmap_sem);
1939 vma = find_exact_vma(mm, vm_start, vm_end);
1940 if (vma && vma->vm_file) {
1941 *path = vma->vm_file->f_path;
1945 up_read(&mm->mmap_sem);
1953 struct map_files_info {
1956 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1960 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1961 * symlinks may be used to bypass permissions on ancestor directories in the
1962 * path to the file in question.
1965 proc_map_files_get_link(struct dentry *dentry,
1966 struct inode *inode,
1967 struct delayed_call *done)
1969 if (!capable(CAP_SYS_ADMIN))
1970 return ERR_PTR(-EPERM);
1972 return proc_pid_get_link(dentry, inode, done);
1976 * Identical to proc_pid_link_inode_operations except for get_link()
1978 static const struct inode_operations proc_map_files_link_inode_operations = {
1979 .readlink = proc_pid_readlink,
1980 .get_link = proc_map_files_get_link,
1981 .setattr = proc_setattr,
1985 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1986 struct task_struct *task, const void *ptr)
1988 fmode_t mode = (fmode_t)(unsigned long)ptr;
1989 struct proc_inode *ei;
1990 struct inode *inode;
1992 inode = proc_pid_make_inode(dir->i_sb, task);
1997 ei->op.proc_get_link = map_files_get_link;
1999 inode->i_op = &proc_map_files_link_inode_operations;
2001 inode->i_mode = S_IFLNK;
2003 if (mode & FMODE_READ)
2004 inode->i_mode |= S_IRUSR;
2005 if (mode & FMODE_WRITE)
2006 inode->i_mode |= S_IWUSR;
2008 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2009 d_add(dentry, inode);
2014 static struct dentry *proc_map_files_lookup(struct inode *dir,
2015 struct dentry *dentry, unsigned int flags)
2017 unsigned long vm_start, vm_end;
2018 struct vm_area_struct *vma;
2019 struct task_struct *task;
2021 struct mm_struct *mm;
2024 task = get_proc_task(dir);
2029 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2033 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2036 mm = get_task_mm(task);
2040 down_read(&mm->mmap_sem);
2041 vma = find_exact_vma(mm, vm_start, vm_end);
2046 result = proc_map_files_instantiate(dir, dentry, task,
2047 (void *)(unsigned long)vma->vm_file->f_mode);
2050 up_read(&mm->mmap_sem);
2053 put_task_struct(task);
2055 return ERR_PTR(result);
2058 static const struct inode_operations proc_map_files_inode_operations = {
2059 .lookup = proc_map_files_lookup,
2060 .permission = proc_fd_permission,
2061 .setattr = proc_setattr,
2065 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2067 struct vm_area_struct *vma;
2068 struct task_struct *task;
2069 struct mm_struct *mm;
2070 unsigned long nr_files, pos, i;
2071 struct flex_array *fa = NULL;
2072 struct map_files_info info;
2073 struct map_files_info *p;
2077 task = get_proc_task(file_inode(file));
2082 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2086 if (!dir_emit_dots(file, ctx))
2089 mm = get_task_mm(task);
2092 down_read(&mm->mmap_sem);
2097 * We need two passes here:
2099 * 1) Collect vmas of mapped files with mmap_sem taken
2100 * 2) Release mmap_sem and instantiate entries
2102 * otherwise we get lockdep complained, since filldir()
2103 * routine might require mmap_sem taken in might_fault().
2106 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2107 if (vma->vm_file && ++pos > ctx->pos)
2112 fa = flex_array_alloc(sizeof(info), nr_files,
2114 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2118 flex_array_free(fa);
2119 up_read(&mm->mmap_sem);
2123 for (i = 0, vma = mm->mmap, pos = 2; vma;
2124 vma = vma->vm_next) {
2127 if (++pos <= ctx->pos)
2130 info.mode = vma->vm_file->f_mode;
2131 info.len = snprintf(info.name,
2132 sizeof(info.name), "%lx-%lx",
2133 vma->vm_start, vma->vm_end);
2134 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2138 up_read(&mm->mmap_sem);
2140 for (i = 0; i < nr_files; i++) {
2141 p = flex_array_get(fa, i);
2142 if (!proc_fill_cache(file, ctx,
2144 proc_map_files_instantiate,
2146 (void *)(unsigned long)p->mode))
2151 flex_array_free(fa);
2155 put_task_struct(task);
2160 static const struct file_operations proc_map_files_operations = {
2161 .read = generic_read_dir,
2162 .iterate_shared = proc_map_files_readdir,
2163 .llseek = generic_file_llseek,
2166 #ifdef CONFIG_CHECKPOINT_RESTORE
2167 struct timers_private {
2169 struct task_struct *task;
2170 struct sighand_struct *sighand;
2171 struct pid_namespace *ns;
2172 unsigned long flags;
2175 static void *timers_start(struct seq_file *m, loff_t *pos)
2177 struct timers_private *tp = m->private;
2179 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2181 return ERR_PTR(-ESRCH);
2183 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2185 return ERR_PTR(-ESRCH);
2187 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2190 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2192 struct timers_private *tp = m->private;
2193 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2196 static void timers_stop(struct seq_file *m, void *v)
2198 struct timers_private *tp = m->private;
2201 unlock_task_sighand(tp->task, &tp->flags);
2206 put_task_struct(tp->task);
2211 static int show_timer(struct seq_file *m, void *v)
2213 struct k_itimer *timer;
2214 struct timers_private *tp = m->private;
2216 static const char * const nstr[] = {
2217 [SIGEV_SIGNAL] = "signal",
2218 [SIGEV_NONE] = "none",
2219 [SIGEV_THREAD] = "thread",
2222 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2223 notify = timer->it_sigev_notify;
2225 seq_printf(m, "ID: %d\n", timer->it_id);
2226 seq_printf(m, "signal: %d/%p\n",
2227 timer->sigq->info.si_signo,
2228 timer->sigq->info.si_value.sival_ptr);
2229 seq_printf(m, "notify: %s/%s.%d\n",
2230 nstr[notify & ~SIGEV_THREAD_ID],
2231 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2232 pid_nr_ns(timer->it_pid, tp->ns));
2233 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2238 static const struct seq_operations proc_timers_seq_ops = {
2239 .start = timers_start,
2240 .next = timers_next,
2241 .stop = timers_stop,
2245 static int proc_timers_open(struct inode *inode, struct file *file)
2247 struct timers_private *tp;
2249 tp = __seq_open_private(file, &proc_timers_seq_ops,
2250 sizeof(struct timers_private));
2254 tp->pid = proc_pid(inode);
2255 tp->ns = inode->i_sb->s_fs_info;
2259 static const struct file_operations proc_timers_operations = {
2260 .open = proc_timers_open,
2262 .llseek = seq_lseek,
2263 .release = seq_release_private,
2267 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2268 size_t count, loff_t *offset)
2270 struct inode *inode = file_inode(file);
2271 struct task_struct *p;
2275 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2279 p = get_proc_task(inode);
2284 if (!capable(CAP_SYS_NICE)) {
2289 err = security_task_setscheduler(p);
2298 p->timer_slack_ns = p->default_timer_slack_ns;
2300 p->timer_slack_ns = slack_ns;
2309 static int timerslack_ns_show(struct seq_file *m, void *v)
2311 struct inode *inode = m->private;
2312 struct task_struct *p;
2315 p = get_proc_task(inode);
2321 if (!capable(CAP_SYS_NICE)) {
2325 err = security_task_getscheduler(p);
2331 seq_printf(m, "%llu\n", p->timer_slack_ns);
2340 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2342 return single_open(filp, timerslack_ns_show, inode);
2345 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2346 .open = timerslack_ns_open,
2348 .write = timerslack_ns_write,
2349 .llseek = seq_lseek,
2350 .release = single_release,
2353 static int proc_pident_instantiate(struct inode *dir,
2354 struct dentry *dentry, struct task_struct *task, const void *ptr)
2356 const struct pid_entry *p = ptr;
2357 struct inode *inode;
2358 struct proc_inode *ei;
2360 inode = proc_pid_make_inode(dir->i_sb, task);
2365 inode->i_mode = p->mode;
2366 if (S_ISDIR(inode->i_mode))
2367 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2369 inode->i_op = p->iop;
2371 inode->i_fop = p->fop;
2373 d_set_d_op(dentry, &pid_dentry_operations);
2374 d_add(dentry, inode);
2375 /* Close the race of the process dying before we return the dentry */
2376 if (pid_revalidate(dentry, 0))
2382 static struct dentry *proc_pident_lookup(struct inode *dir,
2383 struct dentry *dentry,
2384 const struct pid_entry *ents,
2388 struct task_struct *task = get_proc_task(dir);
2389 const struct pid_entry *p, *last;
2397 * Yes, it does not scale. And it should not. Don't add
2398 * new entries into /proc/<tgid>/ without very good reasons.
2400 last = &ents[nents - 1];
2401 for (p = ents; p <= last; p++) {
2402 if (p->len != dentry->d_name.len)
2404 if (!memcmp(dentry->d_name.name, p->name, p->len))
2410 error = proc_pident_instantiate(dir, dentry, task, p);
2412 put_task_struct(task);
2414 return ERR_PTR(error);
2417 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2418 const struct pid_entry *ents, unsigned int nents)
2420 struct task_struct *task = get_proc_task(file_inode(file));
2421 const struct pid_entry *p;
2426 if (!dir_emit_dots(file, ctx))
2429 if (ctx->pos >= nents + 2)
2432 for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2433 if (!proc_fill_cache(file, ctx, p->name, p->len,
2434 proc_pident_instantiate, task, p))
2439 put_task_struct(task);
2443 #ifdef CONFIG_SECURITY
2444 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2445 size_t count, loff_t *ppos)
2447 struct inode * inode = file_inode(file);
2450 struct task_struct *task = get_proc_task(inode);
2455 length = security_getprocattr(task,
2456 (char*)file->f_path.dentry->d_name.name,
2458 put_task_struct(task);
2460 length = simple_read_from_buffer(buf, count, ppos, p, length);
2465 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2466 size_t count, loff_t *ppos)
2468 struct inode * inode = file_inode(file);
2471 struct task_struct *task = get_proc_task(inode);
2476 if (count > PAGE_SIZE)
2479 /* No partial writes. */
2484 page = memdup_user(buf, count);
2486 length = PTR_ERR(page);
2490 /* Guard against adverse ptrace interaction */
2491 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2495 length = security_setprocattr(task,
2496 (char*)file->f_path.dentry->d_name.name,
2498 mutex_unlock(&task->signal->cred_guard_mutex);
2502 put_task_struct(task);
2507 static const struct file_operations proc_pid_attr_operations = {
2508 .read = proc_pid_attr_read,
2509 .write = proc_pid_attr_write,
2510 .llseek = generic_file_llseek,
2513 static const struct pid_entry attr_dir_stuff[] = {
2514 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2515 REG("prev", S_IRUGO, proc_pid_attr_operations),
2516 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2517 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2518 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2519 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2522 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2524 return proc_pident_readdir(file, ctx,
2525 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2528 static const struct file_operations proc_attr_dir_operations = {
2529 .read = generic_read_dir,
2530 .iterate_shared = proc_attr_dir_readdir,
2531 .llseek = generic_file_llseek,
2534 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2535 struct dentry *dentry, unsigned int flags)
2537 return proc_pident_lookup(dir, dentry,
2538 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2541 static const struct inode_operations proc_attr_dir_inode_operations = {
2542 .lookup = proc_attr_dir_lookup,
2543 .getattr = pid_getattr,
2544 .setattr = proc_setattr,
2549 #ifdef CONFIG_ELF_CORE
2550 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2551 size_t count, loff_t *ppos)
2553 struct task_struct *task = get_proc_task(file_inode(file));
2554 struct mm_struct *mm;
2555 char buffer[PROC_NUMBUF];
2563 mm = get_task_mm(task);
2565 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2566 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2567 MMF_DUMP_FILTER_SHIFT));
2569 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2572 put_task_struct(task);
2577 static ssize_t proc_coredump_filter_write(struct file *file,
2578 const char __user *buf,
2582 struct task_struct *task;
2583 struct mm_struct *mm;
2589 ret = kstrtouint_from_user(buf, count, 0, &val);
2594 task = get_proc_task(file_inode(file));
2598 mm = get_task_mm(task);
2603 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2605 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2607 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2612 put_task_struct(task);
2619 static const struct file_operations proc_coredump_filter_operations = {
2620 .read = proc_coredump_filter_read,
2621 .write = proc_coredump_filter_write,
2622 .llseek = generic_file_llseek,
2626 #ifdef CONFIG_TASK_IO_ACCOUNTING
2627 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2629 struct task_io_accounting acct = task->ioac;
2630 unsigned long flags;
2633 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2637 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2642 if (whole && lock_task_sighand(task, &flags)) {
2643 struct task_struct *t = task;
2645 task_io_accounting_add(&acct, &task->signal->ioac);
2646 while_each_thread(task, t)
2647 task_io_accounting_add(&acct, &t->ioac);
2649 unlock_task_sighand(task, &flags);
2656 "read_bytes: %llu\n"
2657 "write_bytes: %llu\n"
2658 "cancelled_write_bytes: %llu\n",
2659 (unsigned long long)acct.rchar,
2660 (unsigned long long)acct.wchar,
2661 (unsigned long long)acct.syscr,
2662 (unsigned long long)acct.syscw,
2663 (unsigned long long)acct.read_bytes,
2664 (unsigned long long)acct.write_bytes,
2665 (unsigned long long)acct.cancelled_write_bytes);
2669 mutex_unlock(&task->signal->cred_guard_mutex);
2673 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2674 struct pid *pid, struct task_struct *task)
2676 return do_io_accounting(task, m, 0);
2679 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2680 struct pid *pid, struct task_struct *task)
2682 return do_io_accounting(task, m, 1);
2684 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2686 #ifdef CONFIG_USER_NS
2687 static int proc_id_map_open(struct inode *inode, struct file *file,
2688 const struct seq_operations *seq_ops)
2690 struct user_namespace *ns = NULL;
2691 struct task_struct *task;
2692 struct seq_file *seq;
2695 task = get_proc_task(inode);
2698 ns = get_user_ns(task_cred_xxx(task, user_ns));
2700 put_task_struct(task);
2705 ret = seq_open(file, seq_ops);
2709 seq = file->private_data;
2719 static int proc_id_map_release(struct inode *inode, struct file *file)
2721 struct seq_file *seq = file->private_data;
2722 struct user_namespace *ns = seq->private;
2724 return seq_release(inode, file);
2727 static int proc_uid_map_open(struct inode *inode, struct file *file)
2729 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2732 static int proc_gid_map_open(struct inode *inode, struct file *file)
2734 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2737 static int proc_projid_map_open(struct inode *inode, struct file *file)
2739 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2742 static const struct file_operations proc_uid_map_operations = {
2743 .open = proc_uid_map_open,
2744 .write = proc_uid_map_write,
2746 .llseek = seq_lseek,
2747 .release = proc_id_map_release,
2750 static const struct file_operations proc_gid_map_operations = {
2751 .open = proc_gid_map_open,
2752 .write = proc_gid_map_write,
2754 .llseek = seq_lseek,
2755 .release = proc_id_map_release,
2758 static const struct file_operations proc_projid_map_operations = {
2759 .open = proc_projid_map_open,
2760 .write = proc_projid_map_write,
2762 .llseek = seq_lseek,
2763 .release = proc_id_map_release,
2766 static int proc_setgroups_open(struct inode *inode, struct file *file)
2768 struct user_namespace *ns = NULL;
2769 struct task_struct *task;
2773 task = get_proc_task(inode);
2776 ns = get_user_ns(task_cred_xxx(task, user_ns));
2778 put_task_struct(task);
2783 if (file->f_mode & FMODE_WRITE) {
2785 if (!ns_capable(ns, CAP_SYS_ADMIN))
2789 ret = single_open(file, &proc_setgroups_show, ns);
2800 static int proc_setgroups_release(struct inode *inode, struct file *file)
2802 struct seq_file *seq = file->private_data;
2803 struct user_namespace *ns = seq->private;
2804 int ret = single_release(inode, file);
2809 static const struct file_operations proc_setgroups_operations = {
2810 .open = proc_setgroups_open,
2811 .write = proc_setgroups_write,
2813 .llseek = seq_lseek,
2814 .release = proc_setgroups_release,
2816 #endif /* CONFIG_USER_NS */
2818 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2819 struct pid *pid, struct task_struct *task)
2821 int err = lock_trace(task);
2823 seq_printf(m, "%08x\n", task->personality);
2832 static const struct file_operations proc_task_operations;
2833 static const struct inode_operations proc_task_inode_operations;
2835 static const struct pid_entry tgid_base_stuff[] = {
2836 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2837 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2838 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2839 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2840 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2842 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2844 REG("environ", S_IRUSR, proc_environ_operations),
2845 ONE("auxv", S_IRUSR, proc_pid_auxv),
2846 ONE("status", S_IRUGO, proc_pid_status),
2847 ONE("personality", S_IRUSR, proc_pid_personality),
2848 ONE("limits", S_IRUGO, proc_pid_limits),
2849 #ifdef CONFIG_SCHED_DEBUG
2850 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2852 #ifdef CONFIG_SCHED_AUTOGROUP
2853 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2855 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2856 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2857 ONE("syscall", S_IRUSR, proc_pid_syscall),
2859 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2860 ONE("stat", S_IRUGO, proc_tgid_stat),
2861 ONE("statm", S_IRUGO, proc_pid_statm),
2862 REG("maps", S_IRUGO, proc_pid_maps_operations),
2864 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2866 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2867 LNK("cwd", proc_cwd_link),
2868 LNK("root", proc_root_link),
2869 LNK("exe", proc_exe_link),
2870 REG("mounts", S_IRUGO, proc_mounts_operations),
2871 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2872 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2873 #ifdef CONFIG_PROC_PAGE_MONITOR
2874 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2875 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2876 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2878 #ifdef CONFIG_SECURITY
2879 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2881 #ifdef CONFIG_KALLSYMS
2882 ONE("wchan", S_IRUGO, proc_pid_wchan),
2884 #ifdef CONFIG_STACKTRACE
2885 ONE("stack", S_IRUSR, proc_pid_stack),
2887 #ifdef CONFIG_SCHED_INFO
2888 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2890 #ifdef CONFIG_LATENCYTOP
2891 REG("latency", S_IRUGO, proc_lstats_operations),
2893 #ifdef CONFIG_PROC_PID_CPUSET
2894 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2896 #ifdef CONFIG_CGROUPS
2897 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2899 ONE("oom_score", S_IRUGO, proc_oom_score),
2900 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2901 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2902 #ifdef CONFIG_AUDITSYSCALL
2903 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2904 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2906 #ifdef CONFIG_FAULT_INJECTION
2907 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2909 #ifdef CONFIG_ELF_CORE
2910 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2912 #ifdef CONFIG_TASK_IO_ACCOUNTING
2913 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2915 #ifdef CONFIG_HARDWALL
2916 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2918 #ifdef CONFIG_USER_NS
2919 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2920 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2921 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2922 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2924 #ifdef CONFIG_CHECKPOINT_RESTORE
2925 REG("timers", S_IRUGO, proc_timers_operations),
2927 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2930 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2932 return proc_pident_readdir(file, ctx,
2933 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2936 static const struct file_operations proc_tgid_base_operations = {
2937 .read = generic_read_dir,
2938 .iterate_shared = proc_tgid_base_readdir,
2939 .llseek = generic_file_llseek,
2942 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2944 return proc_pident_lookup(dir, dentry,
2945 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2948 static const struct inode_operations proc_tgid_base_inode_operations = {
2949 .lookup = proc_tgid_base_lookup,
2950 .getattr = pid_getattr,
2951 .setattr = proc_setattr,
2952 .permission = proc_pid_permission,
2955 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2957 struct dentry *dentry, *leader, *dir;
2958 char buf[PROC_NUMBUF];
2962 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2963 /* no ->d_hash() rejects on procfs */
2964 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2966 d_invalidate(dentry);
2974 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2975 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2980 name.len = strlen(name.name);
2981 dir = d_hash_and_lookup(leader, &name);
2983 goto out_put_leader;
2986 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2987 dentry = d_hash_and_lookup(dir, &name);
2989 d_invalidate(dentry);
3001 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3002 * @task: task that should be flushed.
3004 * When flushing dentries from proc, one needs to flush them from global
3005 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3006 * in. This call is supposed to do all of this job.
3008 * Looks in the dcache for
3010 * /proc/@tgid/task/@pid
3011 * if either directory is present flushes it and all of it'ts children
3014 * It is safe and reasonable to cache /proc entries for a task until
3015 * that task exits. After that they just clog up the dcache with
3016 * useless entries, possibly causing useful dcache entries to be
3017 * flushed instead. This routine is proved to flush those useless
3018 * dcache entries at process exit time.
3020 * NOTE: This routine is just an optimization so it does not guarantee
3021 * that no dcache entries will exist at process exit time it
3022 * just makes it very unlikely that any will persist.
3025 void proc_flush_task(struct task_struct *task)
3028 struct pid *pid, *tgid;
3031 pid = task_pid(task);
3032 tgid = task_tgid(task);
3034 for (i = 0; i <= pid->level; i++) {
3035 upid = &pid->numbers[i];
3036 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3037 tgid->numbers[i].nr);
3041 static int proc_pid_instantiate(struct inode *dir,
3042 struct dentry * dentry,
3043 struct task_struct *task, const void *ptr)
3045 struct inode *inode;
3047 inode = proc_pid_make_inode(dir->i_sb, task);
3051 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3052 inode->i_op = &proc_tgid_base_inode_operations;
3053 inode->i_fop = &proc_tgid_base_operations;
3054 inode->i_flags|=S_IMMUTABLE;
3056 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3057 ARRAY_SIZE(tgid_base_stuff)));
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];
3169 if (!has_pid_permissions(ns, iter.task, 2))
3172 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3173 ctx->pos = iter.tgid + TGID_OFFSET;
3174 if (!proc_fill_cache(file, ctx, name, len,
3175 proc_pid_instantiate, iter.task, NULL)) {
3176 put_task_struct(iter.task);
3180 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3185 * proc_tid_comm_permission is a special permission function exclusively
3186 * used for the node /proc/<pid>/task/<tid>/comm.
3187 * It bypasses generic permission checks in the case where a task of the same
3188 * task group attempts to access the node.
3189 * The rationale behind this is that glibc and bionic access this node for
3190 * cross thread naming (pthread_set/getname_np(!self)). However, if
3191 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3192 * which locks out the cross thread naming implementation.
3193 * This function makes sure that the node is always accessible for members of
3194 * same thread group.
3196 static int proc_tid_comm_permission(struct inode *inode, int mask)
3198 bool is_same_tgroup;
3199 struct task_struct *task;
3201 task = get_proc_task(inode);
3204 is_same_tgroup = same_thread_group(current, task);
3205 put_task_struct(task);
3207 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3208 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3209 * read or written by the members of the corresponding
3215 return generic_permission(inode, mask);
3218 static const struct inode_operations proc_tid_comm_inode_operations = {
3219 .permission = proc_tid_comm_permission,
3225 static const struct pid_entry tid_base_stuff[] = {
3226 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3227 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3228 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3230 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3232 REG("environ", S_IRUSR, proc_environ_operations),
3233 ONE("auxv", S_IRUSR, proc_pid_auxv),
3234 ONE("status", S_IRUGO, proc_pid_status),
3235 ONE("personality", S_IRUSR, proc_pid_personality),
3236 ONE("limits", S_IRUGO, proc_pid_limits),
3237 #ifdef CONFIG_SCHED_DEBUG
3238 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3240 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3241 &proc_tid_comm_inode_operations,
3242 &proc_pid_set_comm_operations, {}),
3243 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3244 ONE("syscall", S_IRUSR, proc_pid_syscall),
3246 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3247 ONE("stat", S_IRUGO, proc_tid_stat),
3248 ONE("statm", S_IRUGO, proc_pid_statm),
3249 REG("maps", S_IRUGO, proc_tid_maps_operations),
3250 #ifdef CONFIG_PROC_CHILDREN
3251 REG("children", S_IRUGO, proc_tid_children_operations),
3254 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3256 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3257 LNK("cwd", proc_cwd_link),
3258 LNK("root", proc_root_link),
3259 LNK("exe", proc_exe_link),
3260 REG("mounts", S_IRUGO, proc_mounts_operations),
3261 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3262 #ifdef CONFIG_PROC_PAGE_MONITOR
3263 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3264 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3265 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3267 #ifdef CONFIG_SECURITY
3268 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3270 #ifdef CONFIG_KALLSYMS
3271 ONE("wchan", S_IRUGO, proc_pid_wchan),
3273 #ifdef CONFIG_STACKTRACE
3274 ONE("stack", S_IRUSR, proc_pid_stack),
3276 #ifdef CONFIG_SCHED_INFO
3277 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3279 #ifdef CONFIG_LATENCYTOP
3280 REG("latency", S_IRUGO, proc_lstats_operations),
3282 #ifdef CONFIG_PROC_PID_CPUSET
3283 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3285 #ifdef CONFIG_CGROUPS
3286 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3288 ONE("oom_score", S_IRUGO, proc_oom_score),
3289 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3290 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3291 #ifdef CONFIG_AUDITSYSCALL
3292 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3293 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3295 #ifdef CONFIG_FAULT_INJECTION
3296 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3298 #ifdef CONFIG_TASK_IO_ACCOUNTING
3299 ONE("io", S_IRUSR, proc_tid_io_accounting),
3301 #ifdef CONFIG_HARDWALL
3302 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3304 #ifdef CONFIG_USER_NS
3305 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3306 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3307 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3308 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3312 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3314 return proc_pident_readdir(file, ctx,
3315 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3318 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3320 return proc_pident_lookup(dir, dentry,
3321 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3324 static const struct file_operations proc_tid_base_operations = {
3325 .read = generic_read_dir,
3326 .iterate_shared = proc_tid_base_readdir,
3327 .llseek = generic_file_llseek,
3330 static const struct inode_operations proc_tid_base_inode_operations = {
3331 .lookup = proc_tid_base_lookup,
3332 .getattr = pid_getattr,
3333 .setattr = proc_setattr,
3336 static int proc_task_instantiate(struct inode *dir,
3337 struct dentry *dentry, struct task_struct *task, const void *ptr)
3339 struct inode *inode;
3340 inode = proc_pid_make_inode(dir->i_sb, task);
3344 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3345 inode->i_op = &proc_tid_base_inode_operations;
3346 inode->i_fop = &proc_tid_base_operations;
3347 inode->i_flags|=S_IMMUTABLE;
3349 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3350 ARRAY_SIZE(tid_base_stuff)));
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,