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, FOLL_FORCE);
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,
280 if (copy_to_user(buf, page, nr_read)) {
293 * Command line (1 string) occupies ARGV and maybe
296 if (len1 + len2 <= *pos)
301 p = arg_start + *pos;
303 while (count > 0 && len > 0) {
304 unsigned int _count, l;
308 _count = min3(count, len, PAGE_SIZE);
309 nr_read = access_remote_vm(mm, p, page, _count,
317 * Command line can be shorter than whole ARGV
318 * even if last "marker" byte says it is not.
321 l = strnlen(page, nr_read);
327 if (copy_to_user(buf, page, nr_read)) {
343 * Command line (1 string) occupies ARGV and
347 p = env_start + *pos - len1;
348 len = len1 + len2 - *pos;
353 while (count > 0 && len > 0) {
354 unsigned int _count, l;
358 _count = min3(count, len, PAGE_SIZE);
359 nr_read = access_remote_vm(mm, p, page, _count,
368 l = strnlen(page, nr_read);
374 if (copy_to_user(buf, page, nr_read)) {
393 free_page((unsigned long)page);
401 static const struct file_operations proc_pid_cmdline_ops = {
402 .read = proc_pid_cmdline_read,
403 .llseek = generic_file_llseek,
406 #ifdef CONFIG_KALLSYMS
408 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
409 * Returns the resolved symbol. If that fails, simply return the address.
411 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
412 struct pid *pid, struct task_struct *task)
415 char symname[KSYM_NAME_LEN];
417 wchan = get_wchan(task);
419 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
420 && !lookup_symbol_name(wchan, symname))
421 seq_printf(m, "%s", symname);
427 #endif /* CONFIG_KALLSYMS */
429 static int lock_trace(struct task_struct *task)
431 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
434 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
435 mutex_unlock(&task->signal->cred_guard_mutex);
441 static void unlock_trace(struct task_struct *task)
443 mutex_unlock(&task->signal->cred_guard_mutex);
446 #ifdef CONFIG_STACKTRACE
448 #define MAX_STACK_TRACE_DEPTH 64
450 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
451 struct pid *pid, struct task_struct *task)
453 struct stack_trace trace;
454 unsigned long *entries;
458 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
462 trace.nr_entries = 0;
463 trace.max_entries = MAX_STACK_TRACE_DEPTH;
464 trace.entries = entries;
467 err = lock_trace(task);
469 save_stack_trace_tsk(task, &trace);
471 for (i = 0; i < trace.nr_entries; i++) {
472 seq_printf(m, "[<%pK>] %pB\n",
473 (void *)entries[i], (void *)entries[i]);
483 #ifdef CONFIG_SCHED_INFO
485 * Provides /proc/PID/schedstat
487 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
488 struct pid *pid, struct task_struct *task)
490 if (unlikely(!sched_info_on()))
491 seq_printf(m, "0 0 0\n");
493 seq_printf(m, "%llu %llu %lu\n",
494 (unsigned long long)task->se.sum_exec_runtime,
495 (unsigned long long)task->sched_info.run_delay,
496 task->sched_info.pcount);
502 #ifdef CONFIG_LATENCYTOP
503 static int lstats_show_proc(struct seq_file *m, void *v)
506 struct inode *inode = m->private;
507 struct task_struct *task = get_proc_task(inode);
511 seq_puts(m, "Latency Top version : v0.1\n");
512 for (i = 0; i < 32; i++) {
513 struct latency_record *lr = &task->latency_record[i];
514 if (lr->backtrace[0]) {
516 seq_printf(m, "%i %li %li",
517 lr->count, lr->time, lr->max);
518 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
519 unsigned long bt = lr->backtrace[q];
524 seq_printf(m, " %ps", (void *)bt);
530 put_task_struct(task);
534 static int lstats_open(struct inode *inode, struct file *file)
536 return single_open(file, lstats_show_proc, inode);
539 static ssize_t lstats_write(struct file *file, const char __user *buf,
540 size_t count, loff_t *offs)
542 struct task_struct *task = get_proc_task(file_inode(file));
546 clear_all_latency_tracing(task);
547 put_task_struct(task);
552 static const struct file_operations proc_lstats_operations = {
555 .write = lstats_write,
557 .release = single_release,
562 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
563 struct pid *pid, struct task_struct *task)
565 unsigned long totalpages = totalram_pages + total_swap_pages;
566 unsigned long points = 0;
568 points = oom_badness(task, NULL, NULL, totalpages) *
570 seq_printf(m, "%lu\n", points);
580 static const struct limit_names lnames[RLIM_NLIMITS] = {
581 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
582 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
583 [RLIMIT_DATA] = {"Max data size", "bytes"},
584 [RLIMIT_STACK] = {"Max stack size", "bytes"},
585 [RLIMIT_CORE] = {"Max core file size", "bytes"},
586 [RLIMIT_RSS] = {"Max resident set", "bytes"},
587 [RLIMIT_NPROC] = {"Max processes", "processes"},
588 [RLIMIT_NOFILE] = {"Max open files", "files"},
589 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
590 [RLIMIT_AS] = {"Max address space", "bytes"},
591 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
592 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
593 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
594 [RLIMIT_NICE] = {"Max nice priority", NULL},
595 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
596 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
599 /* Display limits for a process */
600 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
601 struct pid *pid, struct task_struct *task)
606 struct rlimit rlim[RLIM_NLIMITS];
608 if (!lock_task_sighand(task, &flags))
610 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
611 unlock_task_sighand(task, &flags);
614 * print the file header
616 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
617 "Limit", "Soft Limit", "Hard Limit", "Units");
619 for (i = 0; i < RLIM_NLIMITS; i++) {
620 if (rlim[i].rlim_cur == RLIM_INFINITY)
621 seq_printf(m, "%-25s %-20s ",
622 lnames[i].name, "unlimited");
624 seq_printf(m, "%-25s %-20lu ",
625 lnames[i].name, rlim[i].rlim_cur);
627 if (rlim[i].rlim_max == RLIM_INFINITY)
628 seq_printf(m, "%-20s ", "unlimited");
630 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
633 seq_printf(m, "%-10s\n", lnames[i].unit);
641 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
642 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
643 struct pid *pid, struct task_struct *task)
646 unsigned long args[6], sp, pc;
649 res = lock_trace(task);
653 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
654 seq_puts(m, "running\n");
656 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
659 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
661 args[0], args[1], args[2], args[3], args[4], args[5],
667 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
669 /************************************************************************/
670 /* Here the fs part begins */
671 /************************************************************************/
673 /* permission checks */
674 static int proc_fd_access_allowed(struct inode *inode)
676 struct task_struct *task;
678 /* Allow access to a task's file descriptors if it is us or we
679 * may use ptrace attach to the process and find out that
682 task = get_proc_task(inode);
684 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
685 put_task_struct(task);
690 int proc_setattr(struct dentry *dentry, struct iattr *attr)
693 struct inode *inode = d_inode(dentry);
695 if (attr->ia_valid & ATTR_MODE)
698 error = setattr_prepare(dentry, attr);
702 setattr_copy(inode, attr);
703 mark_inode_dirty(inode);
708 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
709 * or euid/egid (for hide_pid_min=2)?
711 static bool has_pid_permissions(struct pid_namespace *pid,
712 struct task_struct *task,
715 if (pid->hide_pid < hide_pid_min)
717 if (in_group_p(pid->pid_gid))
719 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
723 static int proc_pid_permission(struct inode *inode, int mask)
725 struct pid_namespace *pid = inode->i_sb->s_fs_info;
726 struct task_struct *task;
729 task = get_proc_task(inode);
732 has_perms = has_pid_permissions(pid, task, 1);
733 put_task_struct(task);
736 if (pid->hide_pid == 2) {
738 * Let's make getdents(), stat(), and open()
739 * consistent with each other. If a process
740 * may not stat() a file, it shouldn't be seen
748 return generic_permission(inode, mask);
753 static const struct inode_operations proc_def_inode_operations = {
754 .setattr = proc_setattr,
757 static int proc_single_show(struct seq_file *m, void *v)
759 struct inode *inode = m->private;
760 struct pid_namespace *ns;
762 struct task_struct *task;
765 ns = inode->i_sb->s_fs_info;
766 pid = proc_pid(inode);
767 task = get_pid_task(pid, PIDTYPE_PID);
771 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
773 put_task_struct(task);
777 static int proc_single_open(struct inode *inode, struct file *filp)
779 return single_open(filp, proc_single_show, inode);
782 static const struct file_operations proc_single_file_operations = {
783 .open = proc_single_open,
786 .release = single_release,
790 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
792 struct task_struct *task = get_proc_task(inode);
793 struct mm_struct *mm = ERR_PTR(-ESRCH);
796 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
797 put_task_struct(task);
799 if (!IS_ERR_OR_NULL(mm)) {
800 /* ensure this mm_struct can't be freed */
801 atomic_inc(&mm->mm_count);
802 /* but do not pin its memory */
810 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
812 struct mm_struct *mm = proc_mem_open(inode, mode);
817 file->private_data = mm;
821 static int mem_open(struct inode *inode, struct file *file)
823 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
825 /* OK to pass negative loff_t, we can catch out-of-range */
826 file->f_mode |= FMODE_UNSIGNED_OFFSET;
831 static ssize_t mem_rw(struct file *file, char __user *buf,
832 size_t count, loff_t *ppos, int write)
834 struct mm_struct *mm = file->private_data;
835 unsigned long addr = *ppos;
838 unsigned int flags = FOLL_FORCE;
843 page = (char *)__get_free_page(GFP_TEMPORARY);
848 if (!atomic_inc_not_zero(&mm->mm_users))
855 int this_len = min_t(int, count, PAGE_SIZE);
857 if (write && copy_from_user(page, buf, this_len)) {
862 this_len = access_remote_vm(mm, addr, page, this_len, flags);
869 if (!write && copy_to_user(buf, page, this_len)) {
883 free_page((unsigned long) page);
887 static ssize_t mem_read(struct file *file, char __user *buf,
888 size_t count, loff_t *ppos)
890 return mem_rw(file, buf, count, ppos, 0);
893 static ssize_t mem_write(struct file *file, const char __user *buf,
894 size_t count, loff_t *ppos)
896 return mem_rw(file, (char __user*)buf, count, ppos, 1);
899 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
903 file->f_pos = offset;
906 file->f_pos += offset;
911 force_successful_syscall_return();
915 static int mem_release(struct inode *inode, struct file *file)
917 struct mm_struct *mm = file->private_data;
923 static const struct file_operations proc_mem_operations = {
928 .release = mem_release,
931 static int environ_open(struct inode *inode, struct file *file)
933 return __mem_open(inode, file, PTRACE_MODE_READ);
936 static ssize_t environ_read(struct file *file, char __user *buf,
937 size_t count, loff_t *ppos)
940 unsigned long src = *ppos;
942 struct mm_struct *mm = file->private_data;
943 unsigned long env_start, env_end;
945 /* Ensure the process spawned far enough to have an environment. */
946 if (!mm || !mm->env_end)
949 page = (char *)__get_free_page(GFP_TEMPORARY);
954 if (!atomic_inc_not_zero(&mm->mm_users))
957 down_read(&mm->mmap_sem);
958 env_start = mm->env_start;
959 env_end = mm->env_end;
960 up_read(&mm->mmap_sem);
963 size_t this_len, max_len;
966 if (src >= (env_end - env_start))
969 this_len = env_end - (env_start + src);
971 max_len = min_t(size_t, PAGE_SIZE, count);
972 this_len = min(max_len, this_len);
974 retval = access_remote_vm(mm, (env_start + src),
975 page, this_len, FOLL_FORCE);
982 if (copy_to_user(buf, page, retval)) {
996 free_page((unsigned long) page);
1000 static const struct file_operations proc_environ_operations = {
1001 .open = environ_open,
1002 .read = environ_read,
1003 .llseek = generic_file_llseek,
1004 .release = mem_release,
1007 static int auxv_open(struct inode *inode, struct file *file)
1009 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1012 static ssize_t auxv_read(struct file *file, char __user *buf,
1013 size_t count, loff_t *ppos)
1015 struct mm_struct *mm = file->private_data;
1016 unsigned int nwords = 0;
1019 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1020 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1021 nwords * sizeof(mm->saved_auxv[0]));
1024 static const struct file_operations proc_auxv_operations = {
1027 .llseek = generic_file_llseek,
1028 .release = mem_release,
1031 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1034 struct task_struct *task = get_proc_task(file_inode(file));
1035 char buffer[PROC_NUMBUF];
1036 int oom_adj = OOM_ADJUST_MIN;
1041 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1042 oom_adj = OOM_ADJUST_MAX;
1044 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1046 put_task_struct(task);
1047 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1048 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1051 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1053 static DEFINE_MUTEX(oom_adj_mutex);
1054 struct mm_struct *mm = NULL;
1055 struct task_struct *task;
1058 task = get_proc_task(file_inode(file));
1062 mutex_lock(&oom_adj_mutex);
1064 if (oom_adj < task->signal->oom_score_adj &&
1065 !capable(CAP_SYS_RESOURCE)) {
1070 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1071 * /proc/pid/oom_score_adj instead.
1073 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1074 current->comm, task_pid_nr(current), task_pid_nr(task),
1077 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1078 !capable(CAP_SYS_RESOURCE)) {
1085 * Make sure we will check other processes sharing the mm if this is
1086 * not vfrok which wants its own oom_score_adj.
1087 * pin the mm so it doesn't go away and get reused after task_unlock
1089 if (!task->vfork_done) {
1090 struct task_struct *p = find_lock_task_mm(task);
1093 if (atomic_read(&p->mm->mm_users) > 1) {
1095 atomic_inc(&mm->mm_count);
1101 task->signal->oom_score_adj = oom_adj;
1102 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1103 task->signal->oom_score_adj_min = (short)oom_adj;
1104 trace_oom_score_adj_update(task);
1107 struct task_struct *p;
1110 for_each_process(p) {
1111 if (same_thread_group(task, p))
1114 /* do not touch kernel threads or the global init */
1115 if (p->flags & PF_KTHREAD || is_global_init(p))
1119 if (!p->vfork_done && process_shares_mm(p, mm)) {
1120 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",
1121 task_pid_nr(p), p->comm,
1122 p->signal->oom_score_adj, oom_adj,
1123 task_pid_nr(task), task->comm);
1124 p->signal->oom_score_adj = oom_adj;
1125 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1126 p->signal->oom_score_adj_min = (short)oom_adj;
1134 mutex_unlock(&oom_adj_mutex);
1135 put_task_struct(task);
1140 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1141 * kernels. The effective policy is defined by oom_score_adj, which has a
1142 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1143 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1144 * Processes that become oom disabled via oom_adj will still be oom disabled
1145 * with this implementation.
1147 * oom_adj cannot be removed since existing userspace binaries use it.
1149 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1150 size_t count, loff_t *ppos)
1152 char buffer[PROC_NUMBUF];
1156 memset(buffer, 0, sizeof(buffer));
1157 if (count > sizeof(buffer) - 1)
1158 count = sizeof(buffer) - 1;
1159 if (copy_from_user(buffer, buf, count)) {
1164 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1167 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1168 oom_adj != OOM_DISABLE) {
1174 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1175 * value is always attainable.
1177 if (oom_adj == OOM_ADJUST_MAX)
1178 oom_adj = OOM_SCORE_ADJ_MAX;
1180 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1182 err = __set_oom_adj(file, oom_adj, true);
1184 return err < 0 ? err : count;
1187 static const struct file_operations proc_oom_adj_operations = {
1188 .read = oom_adj_read,
1189 .write = oom_adj_write,
1190 .llseek = generic_file_llseek,
1193 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1194 size_t count, loff_t *ppos)
1196 struct task_struct *task = get_proc_task(file_inode(file));
1197 char buffer[PROC_NUMBUF];
1198 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1203 oom_score_adj = task->signal->oom_score_adj;
1204 put_task_struct(task);
1205 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1206 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1209 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1210 size_t count, loff_t *ppos)
1212 char buffer[PROC_NUMBUF];
1216 memset(buffer, 0, sizeof(buffer));
1217 if (count > sizeof(buffer) - 1)
1218 count = sizeof(buffer) - 1;
1219 if (copy_from_user(buffer, buf, count)) {
1224 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1227 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1228 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1233 err = __set_oom_adj(file, oom_score_adj, false);
1235 return err < 0 ? err : count;
1238 static const struct file_operations proc_oom_score_adj_operations = {
1239 .read = oom_score_adj_read,
1240 .write = oom_score_adj_write,
1241 .llseek = default_llseek,
1244 #ifdef CONFIG_AUDITSYSCALL
1245 #define TMPBUFLEN 21
1246 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1247 size_t count, loff_t *ppos)
1249 struct inode * inode = file_inode(file);
1250 struct task_struct *task = get_proc_task(inode);
1252 char tmpbuf[TMPBUFLEN];
1256 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1257 from_kuid(file->f_cred->user_ns,
1258 audit_get_loginuid(task)));
1259 put_task_struct(task);
1260 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1263 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1264 size_t count, loff_t *ppos)
1266 struct inode * inode = file_inode(file);
1272 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1279 /* No partial writes. */
1283 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1287 /* is userspace tring to explicitly UNSET the loginuid? */
1288 if (loginuid == AUDIT_UID_UNSET) {
1289 kloginuid = INVALID_UID;
1291 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1292 if (!uid_valid(kloginuid))
1296 rv = audit_set_loginuid(kloginuid);
1302 static const struct file_operations proc_loginuid_operations = {
1303 .read = proc_loginuid_read,
1304 .write = proc_loginuid_write,
1305 .llseek = generic_file_llseek,
1308 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1309 size_t count, loff_t *ppos)
1311 struct inode * inode = file_inode(file);
1312 struct task_struct *task = get_proc_task(inode);
1314 char tmpbuf[TMPBUFLEN];
1318 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1319 audit_get_sessionid(task));
1320 put_task_struct(task);
1321 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1324 static const struct file_operations proc_sessionid_operations = {
1325 .read = proc_sessionid_read,
1326 .llseek = generic_file_llseek,
1330 #ifdef CONFIG_FAULT_INJECTION
1331 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1332 size_t count, loff_t *ppos)
1334 struct task_struct *task = get_proc_task(file_inode(file));
1335 char buffer[PROC_NUMBUF];
1341 make_it_fail = task->make_it_fail;
1342 put_task_struct(task);
1344 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1346 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1349 static ssize_t proc_fault_inject_write(struct file * file,
1350 const char __user * buf, size_t count, loff_t *ppos)
1352 struct task_struct *task;
1353 char buffer[PROC_NUMBUF];
1357 if (!capable(CAP_SYS_RESOURCE))
1359 memset(buffer, 0, sizeof(buffer));
1360 if (count > sizeof(buffer) - 1)
1361 count = sizeof(buffer) - 1;
1362 if (copy_from_user(buffer, buf, count))
1364 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1367 if (make_it_fail < 0 || make_it_fail > 1)
1370 task = get_proc_task(file_inode(file));
1373 task->make_it_fail = make_it_fail;
1374 put_task_struct(task);
1379 static const struct file_operations proc_fault_inject_operations = {
1380 .read = proc_fault_inject_read,
1381 .write = proc_fault_inject_write,
1382 .llseek = generic_file_llseek,
1387 #ifdef CONFIG_SCHED_DEBUG
1389 * Print out various scheduling related per-task fields:
1391 static int sched_show(struct seq_file *m, void *v)
1393 struct inode *inode = m->private;
1394 struct task_struct *p;
1396 p = get_proc_task(inode);
1399 proc_sched_show_task(p, m);
1407 sched_write(struct file *file, const char __user *buf,
1408 size_t count, loff_t *offset)
1410 struct inode *inode = file_inode(file);
1411 struct task_struct *p;
1413 p = get_proc_task(inode);
1416 proc_sched_set_task(p);
1423 static int sched_open(struct inode *inode, struct file *filp)
1425 return single_open(filp, sched_show, inode);
1428 static const struct file_operations proc_pid_sched_operations = {
1431 .write = sched_write,
1432 .llseek = seq_lseek,
1433 .release = single_release,
1438 #ifdef CONFIG_SCHED_AUTOGROUP
1440 * Print out autogroup related information:
1442 static int sched_autogroup_show(struct seq_file *m, void *v)
1444 struct inode *inode = m->private;
1445 struct task_struct *p;
1447 p = get_proc_task(inode);
1450 proc_sched_autogroup_show_task(p, m);
1458 sched_autogroup_write(struct file *file, const char __user *buf,
1459 size_t count, loff_t *offset)
1461 struct inode *inode = file_inode(file);
1462 struct task_struct *p;
1463 char buffer[PROC_NUMBUF];
1467 memset(buffer, 0, sizeof(buffer));
1468 if (count > sizeof(buffer) - 1)
1469 count = sizeof(buffer) - 1;
1470 if (copy_from_user(buffer, buf, count))
1473 err = kstrtoint(strstrip(buffer), 0, &nice);
1477 p = get_proc_task(inode);
1481 err = proc_sched_autogroup_set_nice(p, nice);
1490 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1494 ret = single_open(filp, sched_autogroup_show, NULL);
1496 struct seq_file *m = filp->private_data;
1503 static const struct file_operations proc_pid_sched_autogroup_operations = {
1504 .open = sched_autogroup_open,
1506 .write = sched_autogroup_write,
1507 .llseek = seq_lseek,
1508 .release = single_release,
1511 #endif /* CONFIG_SCHED_AUTOGROUP */
1513 static ssize_t comm_write(struct file *file, const char __user *buf,
1514 size_t count, loff_t *offset)
1516 struct inode *inode = file_inode(file);
1517 struct task_struct *p;
1518 char buffer[TASK_COMM_LEN];
1519 const size_t maxlen = sizeof(buffer) - 1;
1521 memset(buffer, 0, sizeof(buffer));
1522 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1525 p = get_proc_task(inode);
1529 if (same_thread_group(current, p))
1530 set_task_comm(p, buffer);
1539 static int comm_show(struct seq_file *m, void *v)
1541 struct inode *inode = m->private;
1542 struct task_struct *p;
1544 p = get_proc_task(inode);
1549 seq_printf(m, "%s\n", p->comm);
1557 static int comm_open(struct inode *inode, struct file *filp)
1559 return single_open(filp, comm_show, inode);
1562 static const struct file_operations proc_pid_set_comm_operations = {
1565 .write = comm_write,
1566 .llseek = seq_lseek,
1567 .release = single_release,
1570 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1572 struct task_struct *task;
1573 struct file *exe_file;
1575 task = get_proc_task(d_inode(dentry));
1578 exe_file = get_task_exe_file(task);
1579 put_task_struct(task);
1581 *exe_path = exe_file->f_path;
1582 path_get(&exe_file->f_path);
1589 static const char *proc_pid_get_link(struct dentry *dentry,
1590 struct inode *inode,
1591 struct delayed_call *done)
1594 int error = -EACCES;
1597 return ERR_PTR(-ECHILD);
1599 /* Are we allowed to snoop on the tasks file descriptors? */
1600 if (!proc_fd_access_allowed(inode))
1603 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1607 nd_jump_link(&path);
1610 return ERR_PTR(error);
1613 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1615 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1622 pathname = d_path(path, tmp, PAGE_SIZE);
1623 len = PTR_ERR(pathname);
1624 if (IS_ERR(pathname))
1626 len = tmp + PAGE_SIZE - 1 - pathname;
1630 if (copy_to_user(buffer, pathname, len))
1633 free_page((unsigned long)tmp);
1637 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1639 int error = -EACCES;
1640 struct inode *inode = d_inode(dentry);
1643 /* Are we allowed to snoop on the tasks file descriptors? */
1644 if (!proc_fd_access_allowed(inode))
1647 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1651 error = do_proc_readlink(&path, buffer, buflen);
1657 const struct inode_operations proc_pid_link_inode_operations = {
1658 .readlink = proc_pid_readlink,
1659 .get_link = proc_pid_get_link,
1660 .setattr = proc_setattr,
1664 /* building an inode */
1666 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1668 struct inode * inode;
1669 struct proc_inode *ei;
1670 const struct cred *cred;
1672 /* We need a new inode */
1674 inode = new_inode(sb);
1680 inode->i_ino = get_next_ino();
1681 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1682 inode->i_op = &proc_def_inode_operations;
1685 * grab the reference to task.
1687 ei->pid = get_task_pid(task, PIDTYPE_PID);
1691 if (task_dumpable(task)) {
1693 cred = __task_cred(task);
1694 inode->i_uid = cred->euid;
1695 inode->i_gid = cred->egid;
1698 security_task_to_inode(task, inode);
1708 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1710 struct inode *inode = d_inode(dentry);
1711 struct task_struct *task;
1712 const struct cred *cred;
1713 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1715 generic_fillattr(inode, stat);
1718 stat->uid = GLOBAL_ROOT_UID;
1719 stat->gid = GLOBAL_ROOT_GID;
1720 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1722 if (!has_pid_permissions(pid, task, 2)) {
1725 * This doesn't prevent learning whether PID exists,
1726 * it only makes getattr() consistent with readdir().
1730 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1731 task_dumpable(task)) {
1732 cred = __task_cred(task);
1733 stat->uid = cred->euid;
1734 stat->gid = cred->egid;
1744 * Exceptional case: normally we are not allowed to unhash a busy
1745 * directory. In this case, however, we can do it - no aliasing problems
1746 * due to the way we treat inodes.
1748 * Rewrite the inode's ownerships here because the owning task may have
1749 * performed a setuid(), etc.
1751 * Before the /proc/pid/status file was created the only way to read
1752 * the effective uid of a /process was to stat /proc/pid. Reading
1753 * /proc/pid/status is slow enough that procps and other packages
1754 * kept stating /proc/pid. To keep the rules in /proc simple I have
1755 * made this apply to all per process world readable and executable
1758 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1760 struct inode *inode;
1761 struct task_struct *task;
1762 const struct cred *cred;
1764 if (flags & LOOKUP_RCU)
1767 inode = d_inode(dentry);
1768 task = get_proc_task(inode);
1771 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1772 task_dumpable(task)) {
1774 cred = __task_cred(task);
1775 inode->i_uid = cred->euid;
1776 inode->i_gid = cred->egid;
1779 inode->i_uid = GLOBAL_ROOT_UID;
1780 inode->i_gid = GLOBAL_ROOT_GID;
1782 inode->i_mode &= ~(S_ISUID | S_ISGID);
1783 security_task_to_inode(task, inode);
1784 put_task_struct(task);
1790 static inline bool proc_inode_is_dead(struct inode *inode)
1792 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1795 int pid_delete_dentry(const struct dentry *dentry)
1797 /* Is the task we represent dead?
1798 * If so, then don't put the dentry on the lru list,
1799 * kill it immediately.
1801 return proc_inode_is_dead(d_inode(dentry));
1804 const struct dentry_operations pid_dentry_operations =
1806 .d_revalidate = pid_revalidate,
1807 .d_delete = pid_delete_dentry,
1813 * Fill a directory entry.
1815 * If possible create the dcache entry and derive our inode number and
1816 * file type from dcache entry.
1818 * Since all of the proc inode numbers are dynamically generated, the inode
1819 * numbers do not exist until the inode is cache. This means creating the
1820 * the dcache entry in readdir is necessary to keep the inode numbers
1821 * reported by readdir in sync with the inode numbers reported
1824 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1825 const char *name, int len,
1826 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1828 struct dentry *child, *dir = file->f_path.dentry;
1829 struct qstr qname = QSTR_INIT(name, len);
1830 struct inode *inode;
1834 child = d_hash_and_lookup(dir, &qname);
1836 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1837 child = d_alloc_parallel(dir, &qname, &wq);
1839 goto end_instantiate;
1840 if (d_in_lookup(child)) {
1841 int err = instantiate(d_inode(dir), child, task, ptr);
1842 d_lookup_done(child);
1845 goto end_instantiate;
1849 inode = d_inode(child);
1851 type = inode->i_mode >> 12;
1853 return dir_emit(ctx, name, len, ino, type);
1856 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1860 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1861 * which represent vma start and end addresses.
1863 static int dname_to_vma_addr(struct dentry *dentry,
1864 unsigned long *start, unsigned long *end)
1866 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1872 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1874 unsigned long vm_start, vm_end;
1875 bool exact_vma_exists = false;
1876 struct mm_struct *mm = NULL;
1877 struct task_struct *task;
1878 const struct cred *cred;
1879 struct inode *inode;
1882 if (flags & LOOKUP_RCU)
1885 inode = d_inode(dentry);
1886 task = get_proc_task(inode);
1890 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1891 if (IS_ERR_OR_NULL(mm))
1894 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1895 down_read(&mm->mmap_sem);
1896 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1897 up_read(&mm->mmap_sem);
1902 if (exact_vma_exists) {
1903 if (task_dumpable(task)) {
1905 cred = __task_cred(task);
1906 inode->i_uid = cred->euid;
1907 inode->i_gid = cred->egid;
1910 inode->i_uid = GLOBAL_ROOT_UID;
1911 inode->i_gid = GLOBAL_ROOT_GID;
1913 security_task_to_inode(task, inode);
1918 put_task_struct(task);
1924 static const struct dentry_operations tid_map_files_dentry_operations = {
1925 .d_revalidate = map_files_d_revalidate,
1926 .d_delete = pid_delete_dentry,
1929 static int map_files_get_link(struct dentry *dentry, struct path *path)
1931 unsigned long vm_start, vm_end;
1932 struct vm_area_struct *vma;
1933 struct task_struct *task;
1934 struct mm_struct *mm;
1938 task = get_proc_task(d_inode(dentry));
1942 mm = get_task_mm(task);
1943 put_task_struct(task);
1947 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1952 down_read(&mm->mmap_sem);
1953 vma = find_exact_vma(mm, vm_start, vm_end);
1954 if (vma && vma->vm_file) {
1955 *path = vma->vm_file->f_path;
1959 up_read(&mm->mmap_sem);
1967 struct map_files_info {
1970 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1974 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1975 * symlinks may be used to bypass permissions on ancestor directories in the
1976 * path to the file in question.
1979 proc_map_files_get_link(struct dentry *dentry,
1980 struct inode *inode,
1981 struct delayed_call *done)
1983 if (!capable(CAP_SYS_ADMIN))
1984 return ERR_PTR(-EPERM);
1986 return proc_pid_get_link(dentry, inode, done);
1990 * Identical to proc_pid_link_inode_operations except for get_link()
1992 static const struct inode_operations proc_map_files_link_inode_operations = {
1993 .readlink = proc_pid_readlink,
1994 .get_link = proc_map_files_get_link,
1995 .setattr = proc_setattr,
1999 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2000 struct task_struct *task, const void *ptr)
2002 fmode_t mode = (fmode_t)(unsigned long)ptr;
2003 struct proc_inode *ei;
2004 struct inode *inode;
2006 inode = proc_pid_make_inode(dir->i_sb, task);
2011 ei->op.proc_get_link = map_files_get_link;
2013 inode->i_op = &proc_map_files_link_inode_operations;
2015 inode->i_mode = S_IFLNK;
2017 if (mode & FMODE_READ)
2018 inode->i_mode |= S_IRUSR;
2019 if (mode & FMODE_WRITE)
2020 inode->i_mode |= S_IWUSR;
2022 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2023 d_add(dentry, inode);
2028 static struct dentry *proc_map_files_lookup(struct inode *dir,
2029 struct dentry *dentry, unsigned int flags)
2031 unsigned long vm_start, vm_end;
2032 struct vm_area_struct *vma;
2033 struct task_struct *task;
2035 struct mm_struct *mm;
2038 task = get_proc_task(dir);
2043 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2047 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2050 mm = get_task_mm(task);
2054 down_read(&mm->mmap_sem);
2055 vma = find_exact_vma(mm, vm_start, vm_end);
2060 result = proc_map_files_instantiate(dir, dentry, task,
2061 (void *)(unsigned long)vma->vm_file->f_mode);
2064 up_read(&mm->mmap_sem);
2067 put_task_struct(task);
2069 return ERR_PTR(result);
2072 static const struct inode_operations proc_map_files_inode_operations = {
2073 .lookup = proc_map_files_lookup,
2074 .permission = proc_fd_permission,
2075 .setattr = proc_setattr,
2079 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2081 struct vm_area_struct *vma;
2082 struct task_struct *task;
2083 struct mm_struct *mm;
2084 unsigned long nr_files, pos, i;
2085 struct flex_array *fa = NULL;
2086 struct map_files_info info;
2087 struct map_files_info *p;
2091 task = get_proc_task(file_inode(file));
2096 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2100 if (!dir_emit_dots(file, ctx))
2103 mm = get_task_mm(task);
2106 down_read(&mm->mmap_sem);
2111 * We need two passes here:
2113 * 1) Collect vmas of mapped files with mmap_sem taken
2114 * 2) Release mmap_sem and instantiate entries
2116 * otherwise we get lockdep complained, since filldir()
2117 * routine might require mmap_sem taken in might_fault().
2120 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2121 if (vma->vm_file && ++pos > ctx->pos)
2126 fa = flex_array_alloc(sizeof(info), nr_files,
2128 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2132 flex_array_free(fa);
2133 up_read(&mm->mmap_sem);
2137 for (i = 0, vma = mm->mmap, pos = 2; vma;
2138 vma = vma->vm_next) {
2141 if (++pos <= ctx->pos)
2144 info.mode = vma->vm_file->f_mode;
2145 info.len = snprintf(info.name,
2146 sizeof(info.name), "%lx-%lx",
2147 vma->vm_start, vma->vm_end);
2148 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2152 up_read(&mm->mmap_sem);
2154 for (i = 0; i < nr_files; i++) {
2155 p = flex_array_get(fa, i);
2156 if (!proc_fill_cache(file, ctx,
2158 proc_map_files_instantiate,
2160 (void *)(unsigned long)p->mode))
2165 flex_array_free(fa);
2169 put_task_struct(task);
2174 static const struct file_operations proc_map_files_operations = {
2175 .read = generic_read_dir,
2176 .iterate_shared = proc_map_files_readdir,
2177 .llseek = generic_file_llseek,
2180 #ifdef CONFIG_CHECKPOINT_RESTORE
2181 struct timers_private {
2183 struct task_struct *task;
2184 struct sighand_struct *sighand;
2185 struct pid_namespace *ns;
2186 unsigned long flags;
2189 static void *timers_start(struct seq_file *m, loff_t *pos)
2191 struct timers_private *tp = m->private;
2193 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2195 return ERR_PTR(-ESRCH);
2197 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2199 return ERR_PTR(-ESRCH);
2201 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2204 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2206 struct timers_private *tp = m->private;
2207 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2210 static void timers_stop(struct seq_file *m, void *v)
2212 struct timers_private *tp = m->private;
2215 unlock_task_sighand(tp->task, &tp->flags);
2220 put_task_struct(tp->task);
2225 static int show_timer(struct seq_file *m, void *v)
2227 struct k_itimer *timer;
2228 struct timers_private *tp = m->private;
2230 static const char * const nstr[] = {
2231 [SIGEV_SIGNAL] = "signal",
2232 [SIGEV_NONE] = "none",
2233 [SIGEV_THREAD] = "thread",
2236 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2237 notify = timer->it_sigev_notify;
2239 seq_printf(m, "ID: %d\n", timer->it_id);
2240 seq_printf(m, "signal: %d/%p\n",
2241 timer->sigq->info.si_signo,
2242 timer->sigq->info.si_value.sival_ptr);
2243 seq_printf(m, "notify: %s/%s.%d\n",
2244 nstr[notify & ~SIGEV_THREAD_ID],
2245 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2246 pid_nr_ns(timer->it_pid, tp->ns));
2247 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2252 static const struct seq_operations proc_timers_seq_ops = {
2253 .start = timers_start,
2254 .next = timers_next,
2255 .stop = timers_stop,
2259 static int proc_timers_open(struct inode *inode, struct file *file)
2261 struct timers_private *tp;
2263 tp = __seq_open_private(file, &proc_timers_seq_ops,
2264 sizeof(struct timers_private));
2268 tp->pid = proc_pid(inode);
2269 tp->ns = inode->i_sb->s_fs_info;
2273 static const struct file_operations proc_timers_operations = {
2274 .open = proc_timers_open,
2276 .llseek = seq_lseek,
2277 .release = seq_release_private,
2281 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2282 size_t count, loff_t *offset)
2284 struct inode *inode = file_inode(file);
2285 struct task_struct *p;
2289 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2293 p = get_proc_task(inode);
2298 if (!capable(CAP_SYS_NICE)) {
2303 err = security_task_setscheduler(p);
2312 p->timer_slack_ns = p->default_timer_slack_ns;
2314 p->timer_slack_ns = slack_ns;
2323 static int timerslack_ns_show(struct seq_file *m, void *v)
2325 struct inode *inode = m->private;
2326 struct task_struct *p;
2329 p = get_proc_task(inode);
2335 if (!capable(CAP_SYS_NICE)) {
2339 err = security_task_getscheduler(p);
2345 seq_printf(m, "%llu\n", p->timer_slack_ns);
2354 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2356 return single_open(filp, timerslack_ns_show, inode);
2359 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2360 .open = timerslack_ns_open,
2362 .write = timerslack_ns_write,
2363 .llseek = seq_lseek,
2364 .release = single_release,
2367 static int proc_pident_instantiate(struct inode *dir,
2368 struct dentry *dentry, struct task_struct *task, const void *ptr)
2370 const struct pid_entry *p = ptr;
2371 struct inode *inode;
2372 struct proc_inode *ei;
2374 inode = proc_pid_make_inode(dir->i_sb, task);
2379 inode->i_mode = p->mode;
2380 if (S_ISDIR(inode->i_mode))
2381 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2383 inode->i_op = p->iop;
2385 inode->i_fop = p->fop;
2387 d_set_d_op(dentry, &pid_dentry_operations);
2388 d_add(dentry, inode);
2389 /* Close the race of the process dying before we return the dentry */
2390 if (pid_revalidate(dentry, 0))
2396 static struct dentry *proc_pident_lookup(struct inode *dir,
2397 struct dentry *dentry,
2398 const struct pid_entry *ents,
2402 struct task_struct *task = get_proc_task(dir);
2403 const struct pid_entry *p, *last;
2411 * Yes, it does not scale. And it should not. Don't add
2412 * new entries into /proc/<tgid>/ without very good reasons.
2414 last = &ents[nents - 1];
2415 for (p = ents; p <= last; p++) {
2416 if (p->len != dentry->d_name.len)
2418 if (!memcmp(dentry->d_name.name, p->name, p->len))
2424 error = proc_pident_instantiate(dir, dentry, task, p);
2426 put_task_struct(task);
2428 return ERR_PTR(error);
2431 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2432 const struct pid_entry *ents, unsigned int nents)
2434 struct task_struct *task = get_proc_task(file_inode(file));
2435 const struct pid_entry *p;
2440 if (!dir_emit_dots(file, ctx))
2443 if (ctx->pos >= nents + 2)
2446 for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2447 if (!proc_fill_cache(file, ctx, p->name, p->len,
2448 proc_pident_instantiate, task, p))
2453 put_task_struct(task);
2457 #ifdef CONFIG_SECURITY
2458 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2459 size_t count, loff_t *ppos)
2461 struct inode * inode = file_inode(file);
2464 struct task_struct *task = get_proc_task(inode);
2469 length = security_getprocattr(task,
2470 (char*)file->f_path.dentry->d_name.name,
2472 put_task_struct(task);
2474 length = simple_read_from_buffer(buf, count, ppos, p, length);
2479 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2480 size_t count, loff_t *ppos)
2482 struct inode * inode = file_inode(file);
2485 struct task_struct *task = get_proc_task(inode);
2490 if (count > PAGE_SIZE)
2493 /* No partial writes. */
2498 page = memdup_user(buf, count);
2500 length = PTR_ERR(page);
2504 /* Guard against adverse ptrace interaction */
2505 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2509 length = security_setprocattr(task,
2510 (char*)file->f_path.dentry->d_name.name,
2512 mutex_unlock(&task->signal->cred_guard_mutex);
2516 put_task_struct(task);
2521 static const struct file_operations proc_pid_attr_operations = {
2522 .read = proc_pid_attr_read,
2523 .write = proc_pid_attr_write,
2524 .llseek = generic_file_llseek,
2527 static const struct pid_entry attr_dir_stuff[] = {
2528 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2529 REG("prev", S_IRUGO, proc_pid_attr_operations),
2530 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2531 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2532 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2536 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2538 return proc_pident_readdir(file, ctx,
2539 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2542 static const struct file_operations proc_attr_dir_operations = {
2543 .read = generic_read_dir,
2544 .iterate_shared = proc_attr_dir_readdir,
2545 .llseek = generic_file_llseek,
2548 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2549 struct dentry *dentry, unsigned int flags)
2551 return proc_pident_lookup(dir, dentry,
2552 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2555 static const struct inode_operations proc_attr_dir_inode_operations = {
2556 .lookup = proc_attr_dir_lookup,
2557 .getattr = pid_getattr,
2558 .setattr = proc_setattr,
2563 #ifdef CONFIG_ELF_CORE
2564 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2565 size_t count, loff_t *ppos)
2567 struct task_struct *task = get_proc_task(file_inode(file));
2568 struct mm_struct *mm;
2569 char buffer[PROC_NUMBUF];
2577 mm = get_task_mm(task);
2579 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2580 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2581 MMF_DUMP_FILTER_SHIFT));
2583 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2586 put_task_struct(task);
2591 static ssize_t proc_coredump_filter_write(struct file *file,
2592 const char __user *buf,
2596 struct task_struct *task;
2597 struct mm_struct *mm;
2603 ret = kstrtouint_from_user(buf, count, 0, &val);
2608 task = get_proc_task(file_inode(file));
2612 mm = get_task_mm(task);
2617 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2619 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2621 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2626 put_task_struct(task);
2633 static const struct file_operations proc_coredump_filter_operations = {
2634 .read = proc_coredump_filter_read,
2635 .write = proc_coredump_filter_write,
2636 .llseek = generic_file_llseek,
2640 #ifdef CONFIG_TASK_IO_ACCOUNTING
2641 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2643 struct task_io_accounting acct = task->ioac;
2644 unsigned long flags;
2647 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2651 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2656 if (whole && lock_task_sighand(task, &flags)) {
2657 struct task_struct *t = task;
2659 task_io_accounting_add(&acct, &task->signal->ioac);
2660 while_each_thread(task, t)
2661 task_io_accounting_add(&acct, &t->ioac);
2663 unlock_task_sighand(task, &flags);
2670 "read_bytes: %llu\n"
2671 "write_bytes: %llu\n"
2672 "cancelled_write_bytes: %llu\n",
2673 (unsigned long long)acct.rchar,
2674 (unsigned long long)acct.wchar,
2675 (unsigned long long)acct.syscr,
2676 (unsigned long long)acct.syscw,
2677 (unsigned long long)acct.read_bytes,
2678 (unsigned long long)acct.write_bytes,
2679 (unsigned long long)acct.cancelled_write_bytes);
2683 mutex_unlock(&task->signal->cred_guard_mutex);
2687 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2688 struct pid *pid, struct task_struct *task)
2690 return do_io_accounting(task, m, 0);
2693 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2694 struct pid *pid, struct task_struct *task)
2696 return do_io_accounting(task, m, 1);
2698 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2700 #ifdef CONFIG_USER_NS
2701 static int proc_id_map_open(struct inode *inode, struct file *file,
2702 const struct seq_operations *seq_ops)
2704 struct user_namespace *ns = NULL;
2705 struct task_struct *task;
2706 struct seq_file *seq;
2709 task = get_proc_task(inode);
2712 ns = get_user_ns(task_cred_xxx(task, user_ns));
2714 put_task_struct(task);
2719 ret = seq_open(file, seq_ops);
2723 seq = file->private_data;
2733 static int proc_id_map_release(struct inode *inode, struct file *file)
2735 struct seq_file *seq = file->private_data;
2736 struct user_namespace *ns = seq->private;
2738 return seq_release(inode, file);
2741 static int proc_uid_map_open(struct inode *inode, struct file *file)
2743 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2746 static int proc_gid_map_open(struct inode *inode, struct file *file)
2748 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2751 static int proc_projid_map_open(struct inode *inode, struct file *file)
2753 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2756 static const struct file_operations proc_uid_map_operations = {
2757 .open = proc_uid_map_open,
2758 .write = proc_uid_map_write,
2760 .llseek = seq_lseek,
2761 .release = proc_id_map_release,
2764 static const struct file_operations proc_gid_map_operations = {
2765 .open = proc_gid_map_open,
2766 .write = proc_gid_map_write,
2768 .llseek = seq_lseek,
2769 .release = proc_id_map_release,
2772 static const struct file_operations proc_projid_map_operations = {
2773 .open = proc_projid_map_open,
2774 .write = proc_projid_map_write,
2776 .llseek = seq_lseek,
2777 .release = proc_id_map_release,
2780 static int proc_setgroups_open(struct inode *inode, struct file *file)
2782 struct user_namespace *ns = NULL;
2783 struct task_struct *task;
2787 task = get_proc_task(inode);
2790 ns = get_user_ns(task_cred_xxx(task, user_ns));
2792 put_task_struct(task);
2797 if (file->f_mode & FMODE_WRITE) {
2799 if (!ns_capable(ns, CAP_SYS_ADMIN))
2803 ret = single_open(file, &proc_setgroups_show, ns);
2814 static int proc_setgroups_release(struct inode *inode, struct file *file)
2816 struct seq_file *seq = file->private_data;
2817 struct user_namespace *ns = seq->private;
2818 int ret = single_release(inode, file);
2823 static const struct file_operations proc_setgroups_operations = {
2824 .open = proc_setgroups_open,
2825 .write = proc_setgroups_write,
2827 .llseek = seq_lseek,
2828 .release = proc_setgroups_release,
2830 #endif /* CONFIG_USER_NS */
2832 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2833 struct pid *pid, struct task_struct *task)
2835 int err = lock_trace(task);
2837 seq_printf(m, "%08x\n", task->personality);
2846 static const struct file_operations proc_task_operations;
2847 static const struct inode_operations proc_task_inode_operations;
2849 static const struct pid_entry tgid_base_stuff[] = {
2850 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2851 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2852 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2853 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2854 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2856 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2858 REG("environ", S_IRUSR, proc_environ_operations),
2859 REG("auxv", S_IRUSR, proc_auxv_operations),
2860 ONE("status", S_IRUGO, proc_pid_status),
2861 ONE("personality", S_IRUSR, proc_pid_personality),
2862 ONE("limits", S_IRUGO, proc_pid_limits),
2863 #ifdef CONFIG_SCHED_DEBUG
2864 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2866 #ifdef CONFIG_SCHED_AUTOGROUP
2867 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2869 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2870 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2871 ONE("syscall", S_IRUSR, proc_pid_syscall),
2873 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2874 ONE("stat", S_IRUGO, proc_tgid_stat),
2875 ONE("statm", S_IRUGO, proc_pid_statm),
2876 REG("maps", S_IRUGO, proc_pid_maps_operations),
2878 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2880 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2881 LNK("cwd", proc_cwd_link),
2882 LNK("root", proc_root_link),
2883 LNK("exe", proc_exe_link),
2884 REG("mounts", S_IRUGO, proc_mounts_operations),
2885 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2886 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2887 #ifdef CONFIG_PROC_PAGE_MONITOR
2888 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2889 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2890 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2892 #ifdef CONFIG_SECURITY
2893 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2895 #ifdef CONFIG_KALLSYMS
2896 ONE("wchan", S_IRUGO, proc_pid_wchan),
2898 #ifdef CONFIG_STACKTRACE
2899 ONE("stack", S_IRUSR, proc_pid_stack),
2901 #ifdef CONFIG_SCHED_INFO
2902 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2904 #ifdef CONFIG_LATENCYTOP
2905 REG("latency", S_IRUGO, proc_lstats_operations),
2907 #ifdef CONFIG_PROC_PID_CPUSET
2908 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2910 #ifdef CONFIG_CGROUPS
2911 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2913 ONE("oom_score", S_IRUGO, proc_oom_score),
2914 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2915 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2916 #ifdef CONFIG_AUDITSYSCALL
2917 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2918 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2920 #ifdef CONFIG_FAULT_INJECTION
2921 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2923 #ifdef CONFIG_ELF_CORE
2924 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2926 #ifdef CONFIG_TASK_IO_ACCOUNTING
2927 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2929 #ifdef CONFIG_HARDWALL
2930 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2932 #ifdef CONFIG_USER_NS
2933 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2934 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2935 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2936 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2938 #ifdef CONFIG_CHECKPOINT_RESTORE
2939 REG("timers", S_IRUGO, proc_timers_operations),
2941 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2944 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2946 return proc_pident_readdir(file, ctx,
2947 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2950 static const struct file_operations proc_tgid_base_operations = {
2951 .read = generic_read_dir,
2952 .iterate_shared = proc_tgid_base_readdir,
2953 .llseek = generic_file_llseek,
2956 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2958 return proc_pident_lookup(dir, dentry,
2959 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2962 static const struct inode_operations proc_tgid_base_inode_operations = {
2963 .lookup = proc_tgid_base_lookup,
2964 .getattr = pid_getattr,
2965 .setattr = proc_setattr,
2966 .permission = proc_pid_permission,
2969 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2971 struct dentry *dentry, *leader, *dir;
2972 char buf[PROC_NUMBUF];
2976 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2977 /* no ->d_hash() rejects on procfs */
2978 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2980 d_invalidate(dentry);
2988 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2989 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2994 name.len = strlen(name.name);
2995 dir = d_hash_and_lookup(leader, &name);
2997 goto out_put_leader;
3000 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3001 dentry = d_hash_and_lookup(dir, &name);
3003 d_invalidate(dentry);
3015 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3016 * @task: task that should be flushed.
3018 * When flushing dentries from proc, one needs to flush them from global
3019 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3020 * in. This call is supposed to do all of this job.
3022 * Looks in the dcache for
3024 * /proc/@tgid/task/@pid
3025 * if either directory is present flushes it and all of it'ts children
3028 * It is safe and reasonable to cache /proc entries for a task until
3029 * that task exits. After that they just clog up the dcache with
3030 * useless entries, possibly causing useful dcache entries to be
3031 * flushed instead. This routine is proved to flush those useless
3032 * dcache entries at process exit time.
3034 * NOTE: This routine is just an optimization so it does not guarantee
3035 * that no dcache entries will exist at process exit time it
3036 * just makes it very unlikely that any will persist.
3039 void proc_flush_task(struct task_struct *task)
3042 struct pid *pid, *tgid;
3045 pid = task_pid(task);
3046 tgid = task_tgid(task);
3048 for (i = 0; i <= pid->level; i++) {
3049 upid = &pid->numbers[i];
3050 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3051 tgid->numbers[i].nr);
3055 static int proc_pid_instantiate(struct inode *dir,
3056 struct dentry * dentry,
3057 struct task_struct *task, const void *ptr)
3059 struct inode *inode;
3061 inode = proc_pid_make_inode(dir->i_sb, task);
3065 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3066 inode->i_op = &proc_tgid_base_inode_operations;
3067 inode->i_fop = &proc_tgid_base_operations;
3068 inode->i_flags|=S_IMMUTABLE;
3070 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3071 ARRAY_SIZE(tgid_base_stuff)));
3073 d_set_d_op(dentry, &pid_dentry_operations);
3075 d_add(dentry, inode);
3076 /* Close the race of the process dying before we return the dentry */
3077 if (pid_revalidate(dentry, 0))
3083 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3085 int result = -ENOENT;
3086 struct task_struct *task;
3088 struct pid_namespace *ns;
3090 tgid = name_to_int(&dentry->d_name);
3094 ns = dentry->d_sb->s_fs_info;
3096 task = find_task_by_pid_ns(tgid, ns);
3098 get_task_struct(task);
3103 result = proc_pid_instantiate(dir, dentry, task, NULL);
3104 put_task_struct(task);
3106 return ERR_PTR(result);
3110 * Find the first task with tgid >= tgid
3115 struct task_struct *task;
3117 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3122 put_task_struct(iter.task);
3126 pid = find_ge_pid(iter.tgid, ns);
3128 iter.tgid = pid_nr_ns(pid, ns);
3129 iter.task = pid_task(pid, PIDTYPE_PID);
3130 /* What we to know is if the pid we have find is the
3131 * pid of a thread_group_leader. Testing for task
3132 * being a thread_group_leader is the obvious thing
3133 * todo but there is a window when it fails, due to
3134 * the pid transfer logic in de_thread.
3136 * So we perform the straight forward test of seeing
3137 * if the pid we have found is the pid of a thread
3138 * group leader, and don't worry if the task we have
3139 * found doesn't happen to be a thread group leader.
3140 * As we don't care in the case of readdir.
3142 if (!iter.task || !has_group_leader_pid(iter.task)) {
3146 get_task_struct(iter.task);
3152 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3154 /* for the /proc/ directory itself, after non-process stuff has been done */
3155 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3157 struct tgid_iter iter;
3158 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3159 loff_t pos = ctx->pos;
3161 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3164 if (pos == TGID_OFFSET - 2) {
3165 struct inode *inode = d_inode(ns->proc_self);
3166 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3168 ctx->pos = pos = pos + 1;
3170 if (pos == TGID_OFFSET - 1) {
3171 struct inode *inode = d_inode(ns->proc_thread_self);
3172 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3174 ctx->pos = pos = pos + 1;
3176 iter.tgid = pos - TGID_OFFSET;
3178 for (iter = next_tgid(ns, iter);
3180 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3181 char name[PROC_NUMBUF];
3183 if (!has_pid_permissions(ns, iter.task, 2))
3186 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3187 ctx->pos = iter.tgid + TGID_OFFSET;
3188 if (!proc_fill_cache(file, ctx, name, len,
3189 proc_pid_instantiate, iter.task, NULL)) {
3190 put_task_struct(iter.task);
3194 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3199 * proc_tid_comm_permission is a special permission function exclusively
3200 * used for the node /proc/<pid>/task/<tid>/comm.
3201 * It bypasses generic permission checks in the case where a task of the same
3202 * task group attempts to access the node.
3203 * The rationale behind this is that glibc and bionic access this node for
3204 * cross thread naming (pthread_set/getname_np(!self)). However, if
3205 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3206 * which locks out the cross thread naming implementation.
3207 * This function makes sure that the node is always accessible for members of
3208 * same thread group.
3210 static int proc_tid_comm_permission(struct inode *inode, int mask)
3212 bool is_same_tgroup;
3213 struct task_struct *task;
3215 task = get_proc_task(inode);
3218 is_same_tgroup = same_thread_group(current, task);
3219 put_task_struct(task);
3221 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3222 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3223 * read or written by the members of the corresponding
3229 return generic_permission(inode, mask);
3232 static const struct inode_operations proc_tid_comm_inode_operations = {
3233 .permission = proc_tid_comm_permission,
3239 static const struct pid_entry tid_base_stuff[] = {
3240 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3241 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3242 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3244 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3246 REG("environ", S_IRUSR, proc_environ_operations),
3247 REG("auxv", S_IRUSR, proc_auxv_operations),
3248 ONE("status", S_IRUGO, proc_pid_status),
3249 ONE("personality", S_IRUSR, proc_pid_personality),
3250 ONE("limits", S_IRUGO, proc_pid_limits),
3251 #ifdef CONFIG_SCHED_DEBUG
3252 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3254 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3255 &proc_tid_comm_inode_operations,
3256 &proc_pid_set_comm_operations, {}),
3257 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3258 ONE("syscall", S_IRUSR, proc_pid_syscall),
3260 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3261 ONE("stat", S_IRUGO, proc_tid_stat),
3262 ONE("statm", S_IRUGO, proc_pid_statm),
3263 REG("maps", S_IRUGO, proc_tid_maps_operations),
3264 #ifdef CONFIG_PROC_CHILDREN
3265 REG("children", S_IRUGO, proc_tid_children_operations),
3268 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3270 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3271 LNK("cwd", proc_cwd_link),
3272 LNK("root", proc_root_link),
3273 LNK("exe", proc_exe_link),
3274 REG("mounts", S_IRUGO, proc_mounts_operations),
3275 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3276 #ifdef CONFIG_PROC_PAGE_MONITOR
3277 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3278 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3279 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3281 #ifdef CONFIG_SECURITY
3282 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3284 #ifdef CONFIG_KALLSYMS
3285 ONE("wchan", S_IRUGO, proc_pid_wchan),
3287 #ifdef CONFIG_STACKTRACE
3288 ONE("stack", S_IRUSR, proc_pid_stack),
3290 #ifdef CONFIG_SCHED_INFO
3291 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3293 #ifdef CONFIG_LATENCYTOP
3294 REG("latency", S_IRUGO, proc_lstats_operations),
3296 #ifdef CONFIG_PROC_PID_CPUSET
3297 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3299 #ifdef CONFIG_CGROUPS
3300 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3302 ONE("oom_score", S_IRUGO, proc_oom_score),
3303 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3304 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3305 #ifdef CONFIG_AUDITSYSCALL
3306 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3307 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3309 #ifdef CONFIG_FAULT_INJECTION
3310 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3312 #ifdef CONFIG_TASK_IO_ACCOUNTING
3313 ONE("io", S_IRUSR, proc_tid_io_accounting),
3315 #ifdef CONFIG_HARDWALL
3316 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3318 #ifdef CONFIG_USER_NS
3319 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3320 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3321 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3322 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3326 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3328 return proc_pident_readdir(file, ctx,
3329 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3332 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3334 return proc_pident_lookup(dir, dentry,
3335 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3338 static const struct file_operations proc_tid_base_operations = {
3339 .read = generic_read_dir,
3340 .iterate_shared = proc_tid_base_readdir,
3341 .llseek = generic_file_llseek,
3344 static const struct inode_operations proc_tid_base_inode_operations = {
3345 .lookup = proc_tid_base_lookup,
3346 .getattr = pid_getattr,
3347 .setattr = proc_setattr,
3350 static int proc_task_instantiate(struct inode *dir,
3351 struct dentry *dentry, struct task_struct *task, const void *ptr)
3353 struct inode *inode;
3354 inode = proc_pid_make_inode(dir->i_sb, task);
3358 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3359 inode->i_op = &proc_tid_base_inode_operations;
3360 inode->i_fop = &proc_tid_base_operations;
3361 inode->i_flags|=S_IMMUTABLE;
3363 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3364 ARRAY_SIZE(tid_base_stuff)));
3366 d_set_d_op(dentry, &pid_dentry_operations);
3368 d_add(dentry, inode);
3369 /* Close the race of the process dying before we return the dentry */
3370 if (pid_revalidate(dentry, 0))
3376 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3378 int result = -ENOENT;
3379 struct task_struct *task;
3380 struct task_struct *leader = get_proc_task(dir);
3382 struct pid_namespace *ns;
3387 tid = name_to_int(&dentry->d_name);
3391 ns = dentry->d_sb->s_fs_info;
3393 task = find_task_by_pid_ns(tid, ns);
3395 get_task_struct(task);
3399 if (!same_thread_group(leader, task))
3402 result = proc_task_instantiate(dir, dentry, task, NULL);
3404 put_task_struct(task);
3406 put_task_struct(leader);
3408 return ERR_PTR(result);
3412 * Find the first tid of a thread group to return to user space.
3414 * Usually this is just the thread group leader, but if the users
3415 * buffer was too small or there was a seek into the middle of the
3416 * directory we have more work todo.
3418 * In the case of a short read we start with find_task_by_pid.
3420 * In the case of a seek we start with the leader and walk nr
3423 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3424 struct pid_namespace *ns)
3426 struct task_struct *pos, *task;
3427 unsigned long nr = f_pos;
3429 if (nr != f_pos) /* 32bit overflow? */
3433 task = pid_task(pid, PIDTYPE_PID);
3437 /* Attempt to start with the tid of a thread */
3439 pos = find_task_by_pid_ns(tid, ns);
3440 if (pos && same_thread_group(pos, task))
3444 /* If nr exceeds the number of threads there is nothing todo */
3445 if (nr >= get_nr_threads(task))
3448 /* If we haven't found our starting place yet start
3449 * with the leader and walk nr threads forward.
3451 pos = task = task->group_leader;
3455 } while_each_thread(task, pos);
3460 get_task_struct(pos);
3467 * Find the next thread in the thread list.
3468 * Return NULL if there is an error or no next thread.
3470 * The reference to the input task_struct is released.
3472 static struct task_struct *next_tid(struct task_struct *start)
3474 struct task_struct *pos = NULL;
3476 if (pid_alive(start)) {
3477 pos = next_thread(start);
3478 if (thread_group_leader(pos))
3481 get_task_struct(pos);
3484 put_task_struct(start);
3488 /* for the /proc/TGID/task/ directories */
3489 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3491 struct inode *inode = file_inode(file);
3492 struct task_struct *task;
3493 struct pid_namespace *ns;
3496 if (proc_inode_is_dead(inode))
3499 if (!dir_emit_dots(file, ctx))
3502 /* f_version caches the tgid value that the last readdir call couldn't
3503 * return. lseek aka telldir automagically resets f_version to 0.
3505 ns = inode->i_sb->s_fs_info;
3506 tid = (int)file->f_version;
3507 file->f_version = 0;
3508 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3510 task = next_tid(task), ctx->pos++) {
3511 char name[PROC_NUMBUF];
3513 tid = task_pid_nr_ns(task, ns);
3514 len = snprintf(name, sizeof(name), "%d", tid);
3515 if (!proc_fill_cache(file, ctx, name, len,
3516 proc_task_instantiate, task, NULL)) {
3517 /* returning this tgid failed, save it as the first
3518 * pid for the next readir call */
3519 file->f_version = (u64)tid;
3520 put_task_struct(task);
3528 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3530 struct inode *inode = d_inode(dentry);
3531 struct task_struct *p = get_proc_task(inode);
3532 generic_fillattr(inode, stat);
3535 stat->nlink += get_nr_threads(p);
3542 static const struct inode_operations proc_task_inode_operations = {
3543 .lookup = proc_task_lookup,
3544 .getattr = proc_task_getattr,
3545 .setattr = proc_setattr,
3546 .permission = proc_pid_permission,
3549 static const struct file_operations proc_task_operations = {
3550 .read = generic_read_dir,
3551 .iterate_shared = proc_task_readdir,
3552 .llseek = generic_file_llseek,