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 #ifdef CONFIG_KALLSYMS
405 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
406 * Returns the resolved symbol. If that fails, simply return the address.
408 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
409 struct pid *pid, struct task_struct *task)
412 char symname[KSYM_NAME_LEN];
414 wchan = get_wchan(task);
416 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
417 && !lookup_symbol_name(wchan, symname))
418 seq_printf(m, "%s", symname);
424 #endif /* CONFIG_KALLSYMS */
426 static int lock_trace(struct task_struct *task)
428 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
431 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
432 mutex_unlock(&task->signal->cred_guard_mutex);
438 static void unlock_trace(struct task_struct *task)
440 mutex_unlock(&task->signal->cred_guard_mutex);
443 #ifdef CONFIG_STACKTRACE
445 #define MAX_STACK_TRACE_DEPTH 64
447 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
448 struct pid *pid, struct task_struct *task)
450 struct stack_trace trace;
451 unsigned long *entries;
455 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
459 trace.nr_entries = 0;
460 trace.max_entries = MAX_STACK_TRACE_DEPTH;
461 trace.entries = entries;
464 err = lock_trace(task);
466 save_stack_trace_tsk(task, &trace);
468 for (i = 0; i < trace.nr_entries; i++) {
469 seq_printf(m, "[<%pK>] %pB\n",
470 (void *)entries[i], (void *)entries[i]);
480 #ifdef CONFIG_SCHED_INFO
482 * Provides /proc/PID/schedstat
484 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
485 struct pid *pid, struct task_struct *task)
487 if (unlikely(!sched_info_on()))
488 seq_printf(m, "0 0 0\n");
490 seq_printf(m, "%llu %llu %lu\n",
491 (unsigned long long)task->se.sum_exec_runtime,
492 (unsigned long long)task->sched_info.run_delay,
493 task->sched_info.pcount);
499 #ifdef CONFIG_LATENCYTOP
500 static int lstats_show_proc(struct seq_file *m, void *v)
503 struct inode *inode = m->private;
504 struct task_struct *task = get_proc_task(inode);
508 seq_puts(m, "Latency Top version : v0.1\n");
509 for (i = 0; i < 32; i++) {
510 struct latency_record *lr = &task->latency_record[i];
511 if (lr->backtrace[0]) {
513 seq_printf(m, "%i %li %li",
514 lr->count, lr->time, lr->max);
515 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
516 unsigned long bt = lr->backtrace[q];
521 seq_printf(m, " %ps", (void *)bt);
527 put_task_struct(task);
531 static int lstats_open(struct inode *inode, struct file *file)
533 return single_open(file, lstats_show_proc, inode);
536 static ssize_t lstats_write(struct file *file, const char __user *buf,
537 size_t count, loff_t *offs)
539 struct task_struct *task = get_proc_task(file_inode(file));
543 clear_all_latency_tracing(task);
544 put_task_struct(task);
549 static const struct file_operations proc_lstats_operations = {
552 .write = lstats_write,
554 .release = single_release,
559 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
560 struct pid *pid, struct task_struct *task)
562 unsigned long totalpages = totalram_pages + total_swap_pages;
563 unsigned long points = 0;
565 points = oom_badness(task, NULL, NULL, totalpages) *
567 seq_printf(m, "%lu\n", points);
577 static const struct limit_names lnames[RLIM_NLIMITS] = {
578 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
579 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
580 [RLIMIT_DATA] = {"Max data size", "bytes"},
581 [RLIMIT_STACK] = {"Max stack size", "bytes"},
582 [RLIMIT_CORE] = {"Max core file size", "bytes"},
583 [RLIMIT_RSS] = {"Max resident set", "bytes"},
584 [RLIMIT_NPROC] = {"Max processes", "processes"},
585 [RLIMIT_NOFILE] = {"Max open files", "files"},
586 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
587 [RLIMIT_AS] = {"Max address space", "bytes"},
588 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
589 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
590 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
591 [RLIMIT_NICE] = {"Max nice priority", NULL},
592 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
593 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
596 /* Display limits for a process */
597 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
598 struct pid *pid, struct task_struct *task)
603 struct rlimit rlim[RLIM_NLIMITS];
605 if (!lock_task_sighand(task, &flags))
607 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
608 unlock_task_sighand(task, &flags);
611 * print the file header
613 seq_printf(m, "%-25s %-20s %-20s %-10s\n",
614 "Limit", "Soft Limit", "Hard Limit", "Units");
616 for (i = 0; i < RLIM_NLIMITS; i++) {
617 if (rlim[i].rlim_cur == RLIM_INFINITY)
618 seq_printf(m, "%-25s %-20s ",
619 lnames[i].name, "unlimited");
621 seq_printf(m, "%-25s %-20lu ",
622 lnames[i].name, rlim[i].rlim_cur);
624 if (rlim[i].rlim_max == RLIM_INFINITY)
625 seq_printf(m, "%-20s ", "unlimited");
627 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
630 seq_printf(m, "%-10s\n", lnames[i].unit);
638 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
639 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
640 struct pid *pid, struct task_struct *task)
643 unsigned long args[6], sp, pc;
646 res = lock_trace(task);
650 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
651 seq_puts(m, "running\n");
653 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
656 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
658 args[0], args[1], args[2], args[3], args[4], args[5],
664 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
666 /************************************************************************/
667 /* Here the fs part begins */
668 /************************************************************************/
670 /* permission checks */
671 static int proc_fd_access_allowed(struct inode *inode)
673 struct task_struct *task;
675 /* Allow access to a task's file descriptors if it is us or we
676 * may use ptrace attach to the process and find out that
679 task = get_proc_task(inode);
681 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
682 put_task_struct(task);
687 int proc_setattr(struct dentry *dentry, struct iattr *attr)
690 struct inode *inode = d_inode(dentry);
692 if (attr->ia_valid & ATTR_MODE)
695 error = setattr_prepare(dentry, attr);
699 setattr_copy(inode, attr);
700 mark_inode_dirty(inode);
705 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
706 * or euid/egid (for hide_pid_min=2)?
708 static bool has_pid_permissions(struct pid_namespace *pid,
709 struct task_struct *task,
712 if (pid->hide_pid < hide_pid_min)
714 if (in_group_p(pid->pid_gid))
716 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
720 static int proc_pid_permission(struct inode *inode, int mask)
722 struct pid_namespace *pid = inode->i_sb->s_fs_info;
723 struct task_struct *task;
726 task = get_proc_task(inode);
729 has_perms = has_pid_permissions(pid, task, 1);
730 put_task_struct(task);
733 if (pid->hide_pid == 2) {
735 * Let's make getdents(), stat(), and open()
736 * consistent with each other. If a process
737 * may not stat() a file, it shouldn't be seen
745 return generic_permission(inode, mask);
750 static const struct inode_operations proc_def_inode_operations = {
751 .setattr = proc_setattr,
754 static int proc_single_show(struct seq_file *m, void *v)
756 struct inode *inode = m->private;
757 struct pid_namespace *ns;
759 struct task_struct *task;
762 ns = inode->i_sb->s_fs_info;
763 pid = proc_pid(inode);
764 task = get_pid_task(pid, PIDTYPE_PID);
768 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
770 put_task_struct(task);
774 static int proc_single_open(struct inode *inode, struct file *filp)
776 return single_open(filp, proc_single_show, inode);
779 static const struct file_operations proc_single_file_operations = {
780 .open = proc_single_open,
783 .release = single_release,
787 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
789 struct task_struct *task = get_proc_task(inode);
790 struct mm_struct *mm = ERR_PTR(-ESRCH);
793 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
794 put_task_struct(task);
796 if (!IS_ERR_OR_NULL(mm)) {
797 /* ensure this mm_struct can't be freed */
798 atomic_inc(&mm->mm_count);
799 /* but do not pin its memory */
807 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
809 struct mm_struct *mm = proc_mem_open(inode, mode);
814 file->private_data = mm;
818 static int mem_open(struct inode *inode, struct file *file)
820 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
822 /* OK to pass negative loff_t, we can catch out-of-range */
823 file->f_mode |= FMODE_UNSIGNED_OFFSET;
828 static ssize_t mem_rw(struct file *file, char __user *buf,
829 size_t count, loff_t *ppos, int write)
831 struct mm_struct *mm = file->private_data;
832 unsigned long addr = *ppos;
840 page = (char *)__get_free_page(GFP_TEMPORARY);
845 if (!atomic_inc_not_zero(&mm->mm_users))
848 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
854 int this_len = min_t(int, count, PAGE_SIZE);
856 if (write && copy_from_user(page, buf, this_len)) {
861 this_len = access_remote_vm(mm, addr, page, this_len, flags);
868 if (!write && copy_to_user(buf, page, this_len)) {
882 free_page((unsigned long) page);
886 static ssize_t mem_read(struct file *file, char __user *buf,
887 size_t count, loff_t *ppos)
889 return mem_rw(file, buf, count, ppos, 0);
892 static ssize_t mem_write(struct file *file, const char __user *buf,
893 size_t count, loff_t *ppos)
895 return mem_rw(file, (char __user*)buf, count, ppos, 1);
898 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
902 file->f_pos = offset;
905 file->f_pos += offset;
910 force_successful_syscall_return();
914 static int mem_release(struct inode *inode, struct file *file)
916 struct mm_struct *mm = file->private_data;
922 static const struct file_operations proc_mem_operations = {
927 .release = mem_release,
930 static int environ_open(struct inode *inode, struct file *file)
932 return __mem_open(inode, file, PTRACE_MODE_READ);
935 static ssize_t environ_read(struct file *file, char __user *buf,
936 size_t count, loff_t *ppos)
939 unsigned long src = *ppos;
941 struct mm_struct *mm = file->private_data;
942 unsigned long env_start, env_end;
944 /* Ensure the process spawned far enough to have an environment. */
945 if (!mm || !mm->env_end)
948 page = (char *)__get_free_page(GFP_TEMPORARY);
953 if (!atomic_inc_not_zero(&mm->mm_users))
956 down_read(&mm->mmap_sem);
957 env_start = mm->env_start;
958 env_end = mm->env_end;
959 up_read(&mm->mmap_sem);
962 size_t this_len, max_len;
965 if (src >= (env_end - env_start))
968 this_len = env_end - (env_start + src);
970 max_len = min_t(size_t, PAGE_SIZE, count);
971 this_len = min(max_len, this_len);
973 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
980 if (copy_to_user(buf, page, retval)) {
994 free_page((unsigned long) page);
998 static const struct file_operations proc_environ_operations = {
999 .open = environ_open,
1000 .read = environ_read,
1001 .llseek = generic_file_llseek,
1002 .release = mem_release,
1005 static int auxv_open(struct inode *inode, struct file *file)
1007 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1010 static ssize_t auxv_read(struct file *file, char __user *buf,
1011 size_t count, loff_t *ppos)
1013 struct mm_struct *mm = file->private_data;
1014 unsigned int nwords = 0;
1017 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1018 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1019 nwords * sizeof(mm->saved_auxv[0]));
1022 static const struct file_operations proc_auxv_operations = {
1025 .llseek = generic_file_llseek,
1026 .release = mem_release,
1029 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1032 struct task_struct *task = get_proc_task(file_inode(file));
1033 char buffer[PROC_NUMBUF];
1034 int oom_adj = OOM_ADJUST_MIN;
1039 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1040 oom_adj = OOM_ADJUST_MAX;
1042 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1044 put_task_struct(task);
1045 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1046 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1049 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1051 static DEFINE_MUTEX(oom_adj_mutex);
1052 struct mm_struct *mm = NULL;
1053 struct task_struct *task;
1056 task = get_proc_task(file_inode(file));
1060 mutex_lock(&oom_adj_mutex);
1062 if (oom_adj < task->signal->oom_score_adj &&
1063 !capable(CAP_SYS_RESOURCE)) {
1068 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1069 * /proc/pid/oom_score_adj instead.
1071 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1072 current->comm, task_pid_nr(current), task_pid_nr(task),
1075 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1076 !capable(CAP_SYS_RESOURCE)) {
1083 * Make sure we will check other processes sharing the mm if this is
1084 * not vfrok which wants its own oom_score_adj.
1085 * pin the mm so it doesn't go away and get reused after task_unlock
1087 if (!task->vfork_done) {
1088 struct task_struct *p = find_lock_task_mm(task);
1091 if (atomic_read(&p->mm->mm_users) > 1) {
1093 atomic_inc(&mm->mm_count);
1099 task->signal->oom_score_adj = oom_adj;
1100 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1101 task->signal->oom_score_adj_min = (short)oom_adj;
1102 trace_oom_score_adj_update(task);
1105 struct task_struct *p;
1108 for_each_process(p) {
1109 if (same_thread_group(task, p))
1112 /* do not touch kernel threads or the global init */
1113 if (p->flags & PF_KTHREAD || is_global_init(p))
1117 if (!p->vfork_done && process_shares_mm(p, mm)) {
1118 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",
1119 task_pid_nr(p), p->comm,
1120 p->signal->oom_score_adj, oom_adj,
1121 task_pid_nr(task), task->comm);
1122 p->signal->oom_score_adj = oom_adj;
1123 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1124 p->signal->oom_score_adj_min = (short)oom_adj;
1132 mutex_unlock(&oom_adj_mutex);
1133 put_task_struct(task);
1138 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1139 * kernels. The effective policy is defined by oom_score_adj, which has a
1140 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1141 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1142 * Processes that become oom disabled via oom_adj will still be oom disabled
1143 * with this implementation.
1145 * oom_adj cannot be removed since existing userspace binaries use it.
1147 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1148 size_t count, loff_t *ppos)
1150 char buffer[PROC_NUMBUF];
1154 memset(buffer, 0, sizeof(buffer));
1155 if (count > sizeof(buffer) - 1)
1156 count = sizeof(buffer) - 1;
1157 if (copy_from_user(buffer, buf, count)) {
1162 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1165 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1166 oom_adj != OOM_DISABLE) {
1172 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1173 * value is always attainable.
1175 if (oom_adj == OOM_ADJUST_MAX)
1176 oom_adj = OOM_SCORE_ADJ_MAX;
1178 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1180 err = __set_oom_adj(file, oom_adj, true);
1182 return err < 0 ? err : count;
1185 static const struct file_operations proc_oom_adj_operations = {
1186 .read = oom_adj_read,
1187 .write = oom_adj_write,
1188 .llseek = generic_file_llseek,
1191 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1192 size_t count, loff_t *ppos)
1194 struct task_struct *task = get_proc_task(file_inode(file));
1195 char buffer[PROC_NUMBUF];
1196 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1201 oom_score_adj = task->signal->oom_score_adj;
1202 put_task_struct(task);
1203 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1204 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1207 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1208 size_t count, loff_t *ppos)
1210 char buffer[PROC_NUMBUF];
1214 memset(buffer, 0, sizeof(buffer));
1215 if (count > sizeof(buffer) - 1)
1216 count = sizeof(buffer) - 1;
1217 if (copy_from_user(buffer, buf, count)) {
1222 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1225 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1226 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1231 err = __set_oom_adj(file, oom_score_adj, false);
1233 return err < 0 ? err : count;
1236 static const struct file_operations proc_oom_score_adj_operations = {
1237 .read = oom_score_adj_read,
1238 .write = oom_score_adj_write,
1239 .llseek = default_llseek,
1242 #ifdef CONFIG_AUDITSYSCALL
1243 #define TMPBUFLEN 21
1244 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1245 size_t count, loff_t *ppos)
1247 struct inode * inode = file_inode(file);
1248 struct task_struct *task = get_proc_task(inode);
1250 char tmpbuf[TMPBUFLEN];
1254 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1255 from_kuid(file->f_cred->user_ns,
1256 audit_get_loginuid(task)));
1257 put_task_struct(task);
1258 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1261 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1262 size_t count, loff_t *ppos)
1264 struct inode * inode = file_inode(file);
1270 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1277 /* No partial writes. */
1281 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1285 /* is userspace tring to explicitly UNSET the loginuid? */
1286 if (loginuid == AUDIT_UID_UNSET) {
1287 kloginuid = INVALID_UID;
1289 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1290 if (!uid_valid(kloginuid))
1294 rv = audit_set_loginuid(kloginuid);
1300 static const struct file_operations proc_loginuid_operations = {
1301 .read = proc_loginuid_read,
1302 .write = proc_loginuid_write,
1303 .llseek = generic_file_llseek,
1306 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1307 size_t count, loff_t *ppos)
1309 struct inode * inode = file_inode(file);
1310 struct task_struct *task = get_proc_task(inode);
1312 char tmpbuf[TMPBUFLEN];
1316 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1317 audit_get_sessionid(task));
1318 put_task_struct(task);
1319 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1322 static const struct file_operations proc_sessionid_operations = {
1323 .read = proc_sessionid_read,
1324 .llseek = generic_file_llseek,
1328 #ifdef CONFIG_FAULT_INJECTION
1329 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1330 size_t count, loff_t *ppos)
1332 struct task_struct *task = get_proc_task(file_inode(file));
1333 char buffer[PROC_NUMBUF];
1339 make_it_fail = task->make_it_fail;
1340 put_task_struct(task);
1342 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1344 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1347 static ssize_t proc_fault_inject_write(struct file * file,
1348 const char __user * buf, size_t count, loff_t *ppos)
1350 struct task_struct *task;
1351 char buffer[PROC_NUMBUF];
1355 if (!capable(CAP_SYS_RESOURCE))
1357 memset(buffer, 0, sizeof(buffer));
1358 if (count > sizeof(buffer) - 1)
1359 count = sizeof(buffer) - 1;
1360 if (copy_from_user(buffer, buf, count))
1362 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1365 if (make_it_fail < 0 || make_it_fail > 1)
1368 task = get_proc_task(file_inode(file));
1371 task->make_it_fail = make_it_fail;
1372 put_task_struct(task);
1377 static const struct file_operations proc_fault_inject_operations = {
1378 .read = proc_fault_inject_read,
1379 .write = proc_fault_inject_write,
1380 .llseek = generic_file_llseek,
1385 #ifdef CONFIG_SCHED_DEBUG
1387 * Print out various scheduling related per-task fields:
1389 static int sched_show(struct seq_file *m, void *v)
1391 struct inode *inode = m->private;
1392 struct task_struct *p;
1394 p = get_proc_task(inode);
1397 proc_sched_show_task(p, m);
1405 sched_write(struct file *file, const char __user *buf,
1406 size_t count, loff_t *offset)
1408 struct inode *inode = file_inode(file);
1409 struct task_struct *p;
1411 p = get_proc_task(inode);
1414 proc_sched_set_task(p);
1421 static int sched_open(struct inode *inode, struct file *filp)
1423 return single_open(filp, sched_show, inode);
1426 static const struct file_operations proc_pid_sched_operations = {
1429 .write = sched_write,
1430 .llseek = seq_lseek,
1431 .release = single_release,
1436 #ifdef CONFIG_SCHED_AUTOGROUP
1438 * Print out autogroup related information:
1440 static int sched_autogroup_show(struct seq_file *m, void *v)
1442 struct inode *inode = m->private;
1443 struct task_struct *p;
1445 p = get_proc_task(inode);
1448 proc_sched_autogroup_show_task(p, m);
1456 sched_autogroup_write(struct file *file, const char __user *buf,
1457 size_t count, loff_t *offset)
1459 struct inode *inode = file_inode(file);
1460 struct task_struct *p;
1461 char buffer[PROC_NUMBUF];
1465 memset(buffer, 0, sizeof(buffer));
1466 if (count > sizeof(buffer) - 1)
1467 count = sizeof(buffer) - 1;
1468 if (copy_from_user(buffer, buf, count))
1471 err = kstrtoint(strstrip(buffer), 0, &nice);
1475 p = get_proc_task(inode);
1479 err = proc_sched_autogroup_set_nice(p, nice);
1488 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1492 ret = single_open(filp, sched_autogroup_show, NULL);
1494 struct seq_file *m = filp->private_data;
1501 static const struct file_operations proc_pid_sched_autogroup_operations = {
1502 .open = sched_autogroup_open,
1504 .write = sched_autogroup_write,
1505 .llseek = seq_lseek,
1506 .release = single_release,
1509 #endif /* CONFIG_SCHED_AUTOGROUP */
1511 static ssize_t comm_write(struct file *file, const char __user *buf,
1512 size_t count, loff_t *offset)
1514 struct inode *inode = file_inode(file);
1515 struct task_struct *p;
1516 char buffer[TASK_COMM_LEN];
1517 const size_t maxlen = sizeof(buffer) - 1;
1519 memset(buffer, 0, sizeof(buffer));
1520 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1523 p = get_proc_task(inode);
1527 if (same_thread_group(current, p))
1528 set_task_comm(p, buffer);
1537 static int comm_show(struct seq_file *m, void *v)
1539 struct inode *inode = m->private;
1540 struct task_struct *p;
1542 p = get_proc_task(inode);
1547 seq_printf(m, "%s\n", p->comm);
1555 static int comm_open(struct inode *inode, struct file *filp)
1557 return single_open(filp, comm_show, inode);
1560 static const struct file_operations proc_pid_set_comm_operations = {
1563 .write = comm_write,
1564 .llseek = seq_lseek,
1565 .release = single_release,
1568 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1570 struct task_struct *task;
1571 struct file *exe_file;
1573 task = get_proc_task(d_inode(dentry));
1576 exe_file = get_task_exe_file(task);
1577 put_task_struct(task);
1579 *exe_path = exe_file->f_path;
1580 path_get(&exe_file->f_path);
1587 static const char *proc_pid_get_link(struct dentry *dentry,
1588 struct inode *inode,
1589 struct delayed_call *done)
1592 int error = -EACCES;
1595 return ERR_PTR(-ECHILD);
1597 /* Are we allowed to snoop on the tasks file descriptors? */
1598 if (!proc_fd_access_allowed(inode))
1601 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1605 nd_jump_link(&path);
1608 return ERR_PTR(error);
1611 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1613 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1620 pathname = d_path(path, tmp, PAGE_SIZE);
1621 len = PTR_ERR(pathname);
1622 if (IS_ERR(pathname))
1624 len = tmp + PAGE_SIZE - 1 - pathname;
1628 if (copy_to_user(buffer, pathname, len))
1631 free_page((unsigned long)tmp);
1635 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1637 int error = -EACCES;
1638 struct inode *inode = d_inode(dentry);
1641 /* Are we allowed to snoop on the tasks file descriptors? */
1642 if (!proc_fd_access_allowed(inode))
1645 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1649 error = do_proc_readlink(&path, buffer, buflen);
1655 const struct inode_operations proc_pid_link_inode_operations = {
1656 .readlink = proc_pid_readlink,
1657 .get_link = proc_pid_get_link,
1658 .setattr = proc_setattr,
1662 /* building an inode */
1664 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1666 struct inode * inode;
1667 struct proc_inode *ei;
1668 const struct cred *cred;
1670 /* We need a new inode */
1672 inode = new_inode(sb);
1678 inode->i_ino = get_next_ino();
1679 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1680 inode->i_op = &proc_def_inode_operations;
1683 * grab the reference to task.
1685 ei->pid = get_task_pid(task, PIDTYPE_PID);
1689 if (task_dumpable(task)) {
1691 cred = __task_cred(task);
1692 inode->i_uid = cred->euid;
1693 inode->i_gid = cred->egid;
1696 security_task_to_inode(task, inode);
1706 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1708 struct inode *inode = d_inode(dentry);
1709 struct task_struct *task;
1710 const struct cred *cred;
1711 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1713 generic_fillattr(inode, stat);
1716 stat->uid = GLOBAL_ROOT_UID;
1717 stat->gid = GLOBAL_ROOT_GID;
1718 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1720 if (!has_pid_permissions(pid, task, 2)) {
1723 * This doesn't prevent learning whether PID exists,
1724 * it only makes getattr() consistent with readdir().
1728 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1729 task_dumpable(task)) {
1730 cred = __task_cred(task);
1731 stat->uid = cred->euid;
1732 stat->gid = cred->egid;
1742 * Exceptional case: normally we are not allowed to unhash a busy
1743 * directory. In this case, however, we can do it - no aliasing problems
1744 * due to the way we treat inodes.
1746 * Rewrite the inode's ownerships here because the owning task may have
1747 * performed a setuid(), etc.
1749 * Before the /proc/pid/status file was created the only way to read
1750 * the effective uid of a /process was to stat /proc/pid. Reading
1751 * /proc/pid/status is slow enough that procps and other packages
1752 * kept stating /proc/pid. To keep the rules in /proc simple I have
1753 * made this apply to all per process world readable and executable
1756 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1758 struct inode *inode;
1759 struct task_struct *task;
1760 const struct cred *cred;
1762 if (flags & LOOKUP_RCU)
1765 inode = d_inode(dentry);
1766 task = get_proc_task(inode);
1769 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1770 task_dumpable(task)) {
1772 cred = __task_cred(task);
1773 inode->i_uid = cred->euid;
1774 inode->i_gid = cred->egid;
1777 inode->i_uid = GLOBAL_ROOT_UID;
1778 inode->i_gid = GLOBAL_ROOT_GID;
1780 inode->i_mode &= ~(S_ISUID | S_ISGID);
1781 security_task_to_inode(task, inode);
1782 put_task_struct(task);
1788 static inline bool proc_inode_is_dead(struct inode *inode)
1790 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1793 int pid_delete_dentry(const struct dentry *dentry)
1795 /* Is the task we represent dead?
1796 * If so, then don't put the dentry on the lru list,
1797 * kill it immediately.
1799 return proc_inode_is_dead(d_inode(dentry));
1802 const struct dentry_operations pid_dentry_operations =
1804 .d_revalidate = pid_revalidate,
1805 .d_delete = pid_delete_dentry,
1811 * Fill a directory entry.
1813 * If possible create the dcache entry and derive our inode number and
1814 * file type from dcache entry.
1816 * Since all of the proc inode numbers are dynamically generated, the inode
1817 * numbers do not exist until the inode is cache. This means creating the
1818 * the dcache entry in readdir is necessary to keep the inode numbers
1819 * reported by readdir in sync with the inode numbers reported
1822 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1823 const char *name, int len,
1824 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1826 struct dentry *child, *dir = file->f_path.dentry;
1827 struct qstr qname = QSTR_INIT(name, len);
1828 struct inode *inode;
1832 child = d_hash_and_lookup(dir, &qname);
1834 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1835 child = d_alloc_parallel(dir, &qname, &wq);
1837 goto end_instantiate;
1838 if (d_in_lookup(child)) {
1839 int err = instantiate(d_inode(dir), child, task, ptr);
1840 d_lookup_done(child);
1843 goto end_instantiate;
1847 inode = d_inode(child);
1849 type = inode->i_mode >> 12;
1851 return dir_emit(ctx, name, len, ino, type);
1854 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1858 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1859 * which represent vma start and end addresses.
1861 static int dname_to_vma_addr(struct dentry *dentry,
1862 unsigned long *start, unsigned long *end)
1864 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1870 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1872 unsigned long vm_start, vm_end;
1873 bool exact_vma_exists = false;
1874 struct mm_struct *mm = NULL;
1875 struct task_struct *task;
1876 const struct cred *cred;
1877 struct inode *inode;
1880 if (flags & LOOKUP_RCU)
1883 inode = d_inode(dentry);
1884 task = get_proc_task(inode);
1888 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1889 if (IS_ERR_OR_NULL(mm))
1892 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1893 down_read(&mm->mmap_sem);
1894 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1895 up_read(&mm->mmap_sem);
1900 if (exact_vma_exists) {
1901 if (task_dumpable(task)) {
1903 cred = __task_cred(task);
1904 inode->i_uid = cred->euid;
1905 inode->i_gid = cred->egid;
1908 inode->i_uid = GLOBAL_ROOT_UID;
1909 inode->i_gid = GLOBAL_ROOT_GID;
1911 security_task_to_inode(task, inode);
1916 put_task_struct(task);
1922 static const struct dentry_operations tid_map_files_dentry_operations = {
1923 .d_revalidate = map_files_d_revalidate,
1924 .d_delete = pid_delete_dentry,
1927 static int map_files_get_link(struct dentry *dentry, struct path *path)
1929 unsigned long vm_start, vm_end;
1930 struct vm_area_struct *vma;
1931 struct task_struct *task;
1932 struct mm_struct *mm;
1936 task = get_proc_task(d_inode(dentry));
1940 mm = get_task_mm(task);
1941 put_task_struct(task);
1945 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1950 down_read(&mm->mmap_sem);
1951 vma = find_exact_vma(mm, vm_start, vm_end);
1952 if (vma && vma->vm_file) {
1953 *path = vma->vm_file->f_path;
1957 up_read(&mm->mmap_sem);
1965 struct map_files_info {
1968 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1972 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1973 * symlinks may be used to bypass permissions on ancestor directories in the
1974 * path to the file in question.
1977 proc_map_files_get_link(struct dentry *dentry,
1978 struct inode *inode,
1979 struct delayed_call *done)
1981 if (!capable(CAP_SYS_ADMIN))
1982 return ERR_PTR(-EPERM);
1984 return proc_pid_get_link(dentry, inode, done);
1988 * Identical to proc_pid_link_inode_operations except for get_link()
1990 static const struct inode_operations proc_map_files_link_inode_operations = {
1991 .readlink = proc_pid_readlink,
1992 .get_link = proc_map_files_get_link,
1993 .setattr = proc_setattr,
1997 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1998 struct task_struct *task, const void *ptr)
2000 fmode_t mode = (fmode_t)(unsigned long)ptr;
2001 struct proc_inode *ei;
2002 struct inode *inode;
2004 inode = proc_pid_make_inode(dir->i_sb, task);
2009 ei->op.proc_get_link = map_files_get_link;
2011 inode->i_op = &proc_map_files_link_inode_operations;
2013 inode->i_mode = S_IFLNK;
2015 if (mode & FMODE_READ)
2016 inode->i_mode |= S_IRUSR;
2017 if (mode & FMODE_WRITE)
2018 inode->i_mode |= S_IWUSR;
2020 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2021 d_add(dentry, inode);
2026 static struct dentry *proc_map_files_lookup(struct inode *dir,
2027 struct dentry *dentry, unsigned int flags)
2029 unsigned long vm_start, vm_end;
2030 struct vm_area_struct *vma;
2031 struct task_struct *task;
2033 struct mm_struct *mm;
2036 task = get_proc_task(dir);
2041 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2045 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2048 mm = get_task_mm(task);
2052 down_read(&mm->mmap_sem);
2053 vma = find_exact_vma(mm, vm_start, vm_end);
2058 result = proc_map_files_instantiate(dir, dentry, task,
2059 (void *)(unsigned long)vma->vm_file->f_mode);
2062 up_read(&mm->mmap_sem);
2065 put_task_struct(task);
2067 return ERR_PTR(result);
2070 static const struct inode_operations proc_map_files_inode_operations = {
2071 .lookup = proc_map_files_lookup,
2072 .permission = proc_fd_permission,
2073 .setattr = proc_setattr,
2077 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2079 struct vm_area_struct *vma;
2080 struct task_struct *task;
2081 struct mm_struct *mm;
2082 unsigned long nr_files, pos, i;
2083 struct flex_array *fa = NULL;
2084 struct map_files_info info;
2085 struct map_files_info *p;
2089 task = get_proc_task(file_inode(file));
2094 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2098 if (!dir_emit_dots(file, ctx))
2101 mm = get_task_mm(task);
2104 down_read(&mm->mmap_sem);
2109 * We need two passes here:
2111 * 1) Collect vmas of mapped files with mmap_sem taken
2112 * 2) Release mmap_sem and instantiate entries
2114 * otherwise we get lockdep complained, since filldir()
2115 * routine might require mmap_sem taken in might_fault().
2118 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2119 if (vma->vm_file && ++pos > ctx->pos)
2124 fa = flex_array_alloc(sizeof(info), nr_files,
2126 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2130 flex_array_free(fa);
2131 up_read(&mm->mmap_sem);
2135 for (i = 0, vma = mm->mmap, pos = 2; vma;
2136 vma = vma->vm_next) {
2139 if (++pos <= ctx->pos)
2142 info.mode = vma->vm_file->f_mode;
2143 info.len = snprintf(info.name,
2144 sizeof(info.name), "%lx-%lx",
2145 vma->vm_start, vma->vm_end);
2146 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2150 up_read(&mm->mmap_sem);
2152 for (i = 0; i < nr_files; i++) {
2153 p = flex_array_get(fa, i);
2154 if (!proc_fill_cache(file, ctx,
2156 proc_map_files_instantiate,
2158 (void *)(unsigned long)p->mode))
2163 flex_array_free(fa);
2167 put_task_struct(task);
2172 static const struct file_operations proc_map_files_operations = {
2173 .read = generic_read_dir,
2174 .iterate_shared = proc_map_files_readdir,
2175 .llseek = generic_file_llseek,
2178 #ifdef CONFIG_CHECKPOINT_RESTORE
2179 struct timers_private {
2181 struct task_struct *task;
2182 struct sighand_struct *sighand;
2183 struct pid_namespace *ns;
2184 unsigned long flags;
2187 static void *timers_start(struct seq_file *m, loff_t *pos)
2189 struct timers_private *tp = m->private;
2191 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2193 return ERR_PTR(-ESRCH);
2195 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2197 return ERR_PTR(-ESRCH);
2199 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2202 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2204 struct timers_private *tp = m->private;
2205 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2208 static void timers_stop(struct seq_file *m, void *v)
2210 struct timers_private *tp = m->private;
2213 unlock_task_sighand(tp->task, &tp->flags);
2218 put_task_struct(tp->task);
2223 static int show_timer(struct seq_file *m, void *v)
2225 struct k_itimer *timer;
2226 struct timers_private *tp = m->private;
2228 static const char * const nstr[] = {
2229 [SIGEV_SIGNAL] = "signal",
2230 [SIGEV_NONE] = "none",
2231 [SIGEV_THREAD] = "thread",
2234 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2235 notify = timer->it_sigev_notify;
2237 seq_printf(m, "ID: %d\n", timer->it_id);
2238 seq_printf(m, "signal: %d/%p\n",
2239 timer->sigq->info.si_signo,
2240 timer->sigq->info.si_value.sival_ptr);
2241 seq_printf(m, "notify: %s/%s.%d\n",
2242 nstr[notify & ~SIGEV_THREAD_ID],
2243 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2244 pid_nr_ns(timer->it_pid, tp->ns));
2245 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2250 static const struct seq_operations proc_timers_seq_ops = {
2251 .start = timers_start,
2252 .next = timers_next,
2253 .stop = timers_stop,
2257 static int proc_timers_open(struct inode *inode, struct file *file)
2259 struct timers_private *tp;
2261 tp = __seq_open_private(file, &proc_timers_seq_ops,
2262 sizeof(struct timers_private));
2266 tp->pid = proc_pid(inode);
2267 tp->ns = inode->i_sb->s_fs_info;
2271 static const struct file_operations proc_timers_operations = {
2272 .open = proc_timers_open,
2274 .llseek = seq_lseek,
2275 .release = seq_release_private,
2279 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2280 size_t count, loff_t *offset)
2282 struct inode *inode = file_inode(file);
2283 struct task_struct *p;
2287 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2291 p = get_proc_task(inode);
2296 if (!capable(CAP_SYS_NICE)) {
2301 err = security_task_setscheduler(p);
2310 p->timer_slack_ns = p->default_timer_slack_ns;
2312 p->timer_slack_ns = slack_ns;
2321 static int timerslack_ns_show(struct seq_file *m, void *v)
2323 struct inode *inode = m->private;
2324 struct task_struct *p;
2327 p = get_proc_task(inode);
2333 if (!capable(CAP_SYS_NICE)) {
2337 err = security_task_getscheduler(p);
2343 seq_printf(m, "%llu\n", p->timer_slack_ns);
2352 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2354 return single_open(filp, timerslack_ns_show, inode);
2357 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2358 .open = timerslack_ns_open,
2360 .write = timerslack_ns_write,
2361 .llseek = seq_lseek,
2362 .release = single_release,
2365 static int proc_pident_instantiate(struct inode *dir,
2366 struct dentry *dentry, struct task_struct *task, const void *ptr)
2368 const struct pid_entry *p = ptr;
2369 struct inode *inode;
2370 struct proc_inode *ei;
2372 inode = proc_pid_make_inode(dir->i_sb, task);
2377 inode->i_mode = p->mode;
2378 if (S_ISDIR(inode->i_mode))
2379 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2381 inode->i_op = p->iop;
2383 inode->i_fop = p->fop;
2385 d_set_d_op(dentry, &pid_dentry_operations);
2386 d_add(dentry, inode);
2387 /* Close the race of the process dying before we return the dentry */
2388 if (pid_revalidate(dentry, 0))
2394 static struct dentry *proc_pident_lookup(struct inode *dir,
2395 struct dentry *dentry,
2396 const struct pid_entry *ents,
2400 struct task_struct *task = get_proc_task(dir);
2401 const struct pid_entry *p, *last;
2409 * Yes, it does not scale. And it should not. Don't add
2410 * new entries into /proc/<tgid>/ without very good reasons.
2412 last = &ents[nents - 1];
2413 for (p = ents; p <= last; p++) {
2414 if (p->len != dentry->d_name.len)
2416 if (!memcmp(dentry->d_name.name, p->name, p->len))
2422 error = proc_pident_instantiate(dir, dentry, task, p);
2424 put_task_struct(task);
2426 return ERR_PTR(error);
2429 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2430 const struct pid_entry *ents, unsigned int nents)
2432 struct task_struct *task = get_proc_task(file_inode(file));
2433 const struct pid_entry *p;
2438 if (!dir_emit_dots(file, ctx))
2441 if (ctx->pos >= nents + 2)
2444 for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2445 if (!proc_fill_cache(file, ctx, p->name, p->len,
2446 proc_pident_instantiate, task, p))
2451 put_task_struct(task);
2455 #ifdef CONFIG_SECURITY
2456 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2457 size_t count, loff_t *ppos)
2459 struct inode * inode = file_inode(file);
2462 struct task_struct *task = get_proc_task(inode);
2467 length = security_getprocattr(task,
2468 (char*)file->f_path.dentry->d_name.name,
2470 put_task_struct(task);
2472 length = simple_read_from_buffer(buf, count, ppos, p, length);
2477 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2478 size_t count, loff_t *ppos)
2480 struct inode * inode = file_inode(file);
2483 struct task_struct *task = get_proc_task(inode);
2488 if (count > PAGE_SIZE)
2491 /* No partial writes. */
2496 page = memdup_user(buf, count);
2498 length = PTR_ERR(page);
2502 /* Guard against adverse ptrace interaction */
2503 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2507 length = security_setprocattr(task,
2508 (char*)file->f_path.dentry->d_name.name,
2510 mutex_unlock(&task->signal->cred_guard_mutex);
2514 put_task_struct(task);
2519 static const struct file_operations proc_pid_attr_operations = {
2520 .read = proc_pid_attr_read,
2521 .write = proc_pid_attr_write,
2522 .llseek = generic_file_llseek,
2525 static const struct pid_entry attr_dir_stuff[] = {
2526 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2527 REG("prev", S_IRUGO, proc_pid_attr_operations),
2528 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2529 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2530 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2531 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2534 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2536 return proc_pident_readdir(file, ctx,
2537 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2540 static const struct file_operations proc_attr_dir_operations = {
2541 .read = generic_read_dir,
2542 .iterate_shared = proc_attr_dir_readdir,
2543 .llseek = generic_file_llseek,
2546 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2547 struct dentry *dentry, unsigned int flags)
2549 return proc_pident_lookup(dir, dentry,
2550 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2553 static const struct inode_operations proc_attr_dir_inode_operations = {
2554 .lookup = proc_attr_dir_lookup,
2555 .getattr = pid_getattr,
2556 .setattr = proc_setattr,
2561 #ifdef CONFIG_ELF_CORE
2562 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2563 size_t count, loff_t *ppos)
2565 struct task_struct *task = get_proc_task(file_inode(file));
2566 struct mm_struct *mm;
2567 char buffer[PROC_NUMBUF];
2575 mm = get_task_mm(task);
2577 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2578 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2579 MMF_DUMP_FILTER_SHIFT));
2581 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2584 put_task_struct(task);
2589 static ssize_t proc_coredump_filter_write(struct file *file,
2590 const char __user *buf,
2594 struct task_struct *task;
2595 struct mm_struct *mm;
2601 ret = kstrtouint_from_user(buf, count, 0, &val);
2606 task = get_proc_task(file_inode(file));
2610 mm = get_task_mm(task);
2615 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2617 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2619 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2624 put_task_struct(task);
2631 static const struct file_operations proc_coredump_filter_operations = {
2632 .read = proc_coredump_filter_read,
2633 .write = proc_coredump_filter_write,
2634 .llseek = generic_file_llseek,
2638 #ifdef CONFIG_TASK_IO_ACCOUNTING
2639 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2641 struct task_io_accounting acct = task->ioac;
2642 unsigned long flags;
2645 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2649 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2654 if (whole && lock_task_sighand(task, &flags)) {
2655 struct task_struct *t = task;
2657 task_io_accounting_add(&acct, &task->signal->ioac);
2658 while_each_thread(task, t)
2659 task_io_accounting_add(&acct, &t->ioac);
2661 unlock_task_sighand(task, &flags);
2668 "read_bytes: %llu\n"
2669 "write_bytes: %llu\n"
2670 "cancelled_write_bytes: %llu\n",
2671 (unsigned long long)acct.rchar,
2672 (unsigned long long)acct.wchar,
2673 (unsigned long long)acct.syscr,
2674 (unsigned long long)acct.syscw,
2675 (unsigned long long)acct.read_bytes,
2676 (unsigned long long)acct.write_bytes,
2677 (unsigned long long)acct.cancelled_write_bytes);
2681 mutex_unlock(&task->signal->cred_guard_mutex);
2685 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2686 struct pid *pid, struct task_struct *task)
2688 return do_io_accounting(task, m, 0);
2691 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2692 struct pid *pid, struct task_struct *task)
2694 return do_io_accounting(task, m, 1);
2696 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2698 #ifdef CONFIG_USER_NS
2699 static int proc_id_map_open(struct inode *inode, struct file *file,
2700 const struct seq_operations *seq_ops)
2702 struct user_namespace *ns = NULL;
2703 struct task_struct *task;
2704 struct seq_file *seq;
2707 task = get_proc_task(inode);
2710 ns = get_user_ns(task_cred_xxx(task, user_ns));
2712 put_task_struct(task);
2717 ret = seq_open(file, seq_ops);
2721 seq = file->private_data;
2731 static int proc_id_map_release(struct inode *inode, struct file *file)
2733 struct seq_file *seq = file->private_data;
2734 struct user_namespace *ns = seq->private;
2736 return seq_release(inode, file);
2739 static int proc_uid_map_open(struct inode *inode, struct file *file)
2741 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2744 static int proc_gid_map_open(struct inode *inode, struct file *file)
2746 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2749 static int proc_projid_map_open(struct inode *inode, struct file *file)
2751 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2754 static const struct file_operations proc_uid_map_operations = {
2755 .open = proc_uid_map_open,
2756 .write = proc_uid_map_write,
2758 .llseek = seq_lseek,
2759 .release = proc_id_map_release,
2762 static const struct file_operations proc_gid_map_operations = {
2763 .open = proc_gid_map_open,
2764 .write = proc_gid_map_write,
2766 .llseek = seq_lseek,
2767 .release = proc_id_map_release,
2770 static const struct file_operations proc_projid_map_operations = {
2771 .open = proc_projid_map_open,
2772 .write = proc_projid_map_write,
2774 .llseek = seq_lseek,
2775 .release = proc_id_map_release,
2778 static int proc_setgroups_open(struct inode *inode, struct file *file)
2780 struct user_namespace *ns = NULL;
2781 struct task_struct *task;
2785 task = get_proc_task(inode);
2788 ns = get_user_ns(task_cred_xxx(task, user_ns));
2790 put_task_struct(task);
2795 if (file->f_mode & FMODE_WRITE) {
2797 if (!ns_capable(ns, CAP_SYS_ADMIN))
2801 ret = single_open(file, &proc_setgroups_show, ns);
2812 static int proc_setgroups_release(struct inode *inode, struct file *file)
2814 struct seq_file *seq = file->private_data;
2815 struct user_namespace *ns = seq->private;
2816 int ret = single_release(inode, file);
2821 static const struct file_operations proc_setgroups_operations = {
2822 .open = proc_setgroups_open,
2823 .write = proc_setgroups_write,
2825 .llseek = seq_lseek,
2826 .release = proc_setgroups_release,
2828 #endif /* CONFIG_USER_NS */
2830 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2831 struct pid *pid, struct task_struct *task)
2833 int err = lock_trace(task);
2835 seq_printf(m, "%08x\n", task->personality);
2844 static const struct file_operations proc_task_operations;
2845 static const struct inode_operations proc_task_inode_operations;
2847 static const struct pid_entry tgid_base_stuff[] = {
2848 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2849 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2850 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2851 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2852 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2854 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2856 REG("environ", S_IRUSR, proc_environ_operations),
2857 REG("auxv", S_IRUSR, proc_auxv_operations),
2858 ONE("status", S_IRUGO, proc_pid_status),
2859 ONE("personality", S_IRUSR, proc_pid_personality),
2860 ONE("limits", S_IRUGO, proc_pid_limits),
2861 #ifdef CONFIG_SCHED_DEBUG
2862 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2864 #ifdef CONFIG_SCHED_AUTOGROUP
2865 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2867 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2868 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2869 ONE("syscall", S_IRUSR, proc_pid_syscall),
2871 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2872 ONE("stat", S_IRUGO, proc_tgid_stat),
2873 ONE("statm", S_IRUGO, proc_pid_statm),
2874 REG("maps", S_IRUGO, proc_pid_maps_operations),
2876 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2878 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2879 LNK("cwd", proc_cwd_link),
2880 LNK("root", proc_root_link),
2881 LNK("exe", proc_exe_link),
2882 REG("mounts", S_IRUGO, proc_mounts_operations),
2883 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2884 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2885 #ifdef CONFIG_PROC_PAGE_MONITOR
2886 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2887 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2888 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2890 #ifdef CONFIG_SECURITY
2891 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2893 #ifdef CONFIG_KALLSYMS
2894 ONE("wchan", S_IRUGO, proc_pid_wchan),
2896 #ifdef CONFIG_STACKTRACE
2897 ONE("stack", S_IRUSR, proc_pid_stack),
2899 #ifdef CONFIG_SCHED_INFO
2900 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2902 #ifdef CONFIG_LATENCYTOP
2903 REG("latency", S_IRUGO, proc_lstats_operations),
2905 #ifdef CONFIG_PROC_PID_CPUSET
2906 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2908 #ifdef CONFIG_CGROUPS
2909 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2911 ONE("oom_score", S_IRUGO, proc_oom_score),
2912 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2913 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2914 #ifdef CONFIG_AUDITSYSCALL
2915 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2916 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2918 #ifdef CONFIG_FAULT_INJECTION
2919 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2921 #ifdef CONFIG_ELF_CORE
2922 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2924 #ifdef CONFIG_TASK_IO_ACCOUNTING
2925 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2927 #ifdef CONFIG_HARDWALL
2928 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2930 #ifdef CONFIG_USER_NS
2931 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2932 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2933 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2934 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2936 #ifdef CONFIG_CHECKPOINT_RESTORE
2937 REG("timers", S_IRUGO, proc_timers_operations),
2939 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2942 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2944 return proc_pident_readdir(file, ctx,
2945 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2948 static const struct file_operations proc_tgid_base_operations = {
2949 .read = generic_read_dir,
2950 .iterate_shared = proc_tgid_base_readdir,
2951 .llseek = generic_file_llseek,
2954 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2956 return proc_pident_lookup(dir, dentry,
2957 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2960 static const struct inode_operations proc_tgid_base_inode_operations = {
2961 .lookup = proc_tgid_base_lookup,
2962 .getattr = pid_getattr,
2963 .setattr = proc_setattr,
2964 .permission = proc_pid_permission,
2967 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2969 struct dentry *dentry, *leader, *dir;
2970 char buf[PROC_NUMBUF];
2974 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2975 /* no ->d_hash() rejects on procfs */
2976 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2978 d_invalidate(dentry);
2986 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2987 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2992 name.len = strlen(name.name);
2993 dir = d_hash_and_lookup(leader, &name);
2995 goto out_put_leader;
2998 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2999 dentry = d_hash_and_lookup(dir, &name);
3001 d_invalidate(dentry);
3013 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3014 * @task: task that should be flushed.
3016 * When flushing dentries from proc, one needs to flush them from global
3017 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3018 * in. This call is supposed to do all of this job.
3020 * Looks in the dcache for
3022 * /proc/@tgid/task/@pid
3023 * if either directory is present flushes it and all of it'ts children
3026 * It is safe and reasonable to cache /proc entries for a task until
3027 * that task exits. After that they just clog up the dcache with
3028 * useless entries, possibly causing useful dcache entries to be
3029 * flushed instead. This routine is proved to flush those useless
3030 * dcache entries at process exit time.
3032 * NOTE: This routine is just an optimization so it does not guarantee
3033 * that no dcache entries will exist at process exit time it
3034 * just makes it very unlikely that any will persist.
3037 void proc_flush_task(struct task_struct *task)
3040 struct pid *pid, *tgid;
3043 pid = task_pid(task);
3044 tgid = task_tgid(task);
3046 for (i = 0; i <= pid->level; i++) {
3047 upid = &pid->numbers[i];
3048 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3049 tgid->numbers[i].nr);
3053 static int proc_pid_instantiate(struct inode *dir,
3054 struct dentry * dentry,
3055 struct task_struct *task, const void *ptr)
3057 struct inode *inode;
3059 inode = proc_pid_make_inode(dir->i_sb, task);
3063 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3064 inode->i_op = &proc_tgid_base_inode_operations;
3065 inode->i_fop = &proc_tgid_base_operations;
3066 inode->i_flags|=S_IMMUTABLE;
3068 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3069 ARRAY_SIZE(tgid_base_stuff)));
3071 d_set_d_op(dentry, &pid_dentry_operations);
3073 d_add(dentry, inode);
3074 /* Close the race of the process dying before we return the dentry */
3075 if (pid_revalidate(dentry, 0))
3081 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3083 int result = -ENOENT;
3084 struct task_struct *task;
3086 struct pid_namespace *ns;
3088 tgid = name_to_int(&dentry->d_name);
3092 ns = dentry->d_sb->s_fs_info;
3094 task = find_task_by_pid_ns(tgid, ns);
3096 get_task_struct(task);
3101 result = proc_pid_instantiate(dir, dentry, task, NULL);
3102 put_task_struct(task);
3104 return ERR_PTR(result);
3108 * Find the first task with tgid >= tgid
3113 struct task_struct *task;
3115 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3120 put_task_struct(iter.task);
3124 pid = find_ge_pid(iter.tgid, ns);
3126 iter.tgid = pid_nr_ns(pid, ns);
3127 iter.task = pid_task(pid, PIDTYPE_PID);
3128 /* What we to know is if the pid we have find is the
3129 * pid of a thread_group_leader. Testing for task
3130 * being a thread_group_leader is the obvious thing
3131 * todo but there is a window when it fails, due to
3132 * the pid transfer logic in de_thread.
3134 * So we perform the straight forward test of seeing
3135 * if the pid we have found is the pid of a thread
3136 * group leader, and don't worry if the task we have
3137 * found doesn't happen to be a thread group leader.
3138 * As we don't care in the case of readdir.
3140 if (!iter.task || !has_group_leader_pid(iter.task)) {
3144 get_task_struct(iter.task);
3150 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3152 /* for the /proc/ directory itself, after non-process stuff has been done */
3153 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3155 struct tgid_iter iter;
3156 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3157 loff_t pos = ctx->pos;
3159 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3162 if (pos == TGID_OFFSET - 2) {
3163 struct inode *inode = d_inode(ns->proc_self);
3164 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3166 ctx->pos = pos = pos + 1;
3168 if (pos == TGID_OFFSET - 1) {
3169 struct inode *inode = d_inode(ns->proc_thread_self);
3170 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3172 ctx->pos = pos = pos + 1;
3174 iter.tgid = pos - TGID_OFFSET;
3176 for (iter = next_tgid(ns, iter);
3178 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3179 char name[PROC_NUMBUF];
3181 if (!has_pid_permissions(ns, iter.task, 2))
3184 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3185 ctx->pos = iter.tgid + TGID_OFFSET;
3186 if (!proc_fill_cache(file, ctx, name, len,
3187 proc_pid_instantiate, iter.task, NULL)) {
3188 put_task_struct(iter.task);
3192 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3197 * proc_tid_comm_permission is a special permission function exclusively
3198 * used for the node /proc/<pid>/task/<tid>/comm.
3199 * It bypasses generic permission checks in the case where a task of the same
3200 * task group attempts to access the node.
3201 * The rationale behind this is that glibc and bionic access this node for
3202 * cross thread naming (pthread_set/getname_np(!self)). However, if
3203 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3204 * which locks out the cross thread naming implementation.
3205 * This function makes sure that the node is always accessible for members of
3206 * same thread group.
3208 static int proc_tid_comm_permission(struct inode *inode, int mask)
3210 bool is_same_tgroup;
3211 struct task_struct *task;
3213 task = get_proc_task(inode);
3216 is_same_tgroup = same_thread_group(current, task);
3217 put_task_struct(task);
3219 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3220 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3221 * read or written by the members of the corresponding
3227 return generic_permission(inode, mask);
3230 static const struct inode_operations proc_tid_comm_inode_operations = {
3231 .permission = proc_tid_comm_permission,
3237 static const struct pid_entry tid_base_stuff[] = {
3238 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3239 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3240 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3242 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3244 REG("environ", S_IRUSR, proc_environ_operations),
3245 REG("auxv", S_IRUSR, proc_auxv_operations),
3246 ONE("status", S_IRUGO, proc_pid_status),
3247 ONE("personality", S_IRUSR, proc_pid_personality),
3248 ONE("limits", S_IRUGO, proc_pid_limits),
3249 #ifdef CONFIG_SCHED_DEBUG
3250 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3252 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3253 &proc_tid_comm_inode_operations,
3254 &proc_pid_set_comm_operations, {}),
3255 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3256 ONE("syscall", S_IRUSR, proc_pid_syscall),
3258 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3259 ONE("stat", S_IRUGO, proc_tid_stat),
3260 ONE("statm", S_IRUGO, proc_pid_statm),
3261 REG("maps", S_IRUGO, proc_tid_maps_operations),
3262 #ifdef CONFIG_PROC_CHILDREN
3263 REG("children", S_IRUGO, proc_tid_children_operations),
3266 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3268 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3269 LNK("cwd", proc_cwd_link),
3270 LNK("root", proc_root_link),
3271 LNK("exe", proc_exe_link),
3272 REG("mounts", S_IRUGO, proc_mounts_operations),
3273 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3274 #ifdef CONFIG_PROC_PAGE_MONITOR
3275 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3276 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3277 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3279 #ifdef CONFIG_SECURITY
3280 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3282 #ifdef CONFIG_KALLSYMS
3283 ONE("wchan", S_IRUGO, proc_pid_wchan),
3285 #ifdef CONFIG_STACKTRACE
3286 ONE("stack", S_IRUSR, proc_pid_stack),
3288 #ifdef CONFIG_SCHED_INFO
3289 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3291 #ifdef CONFIG_LATENCYTOP
3292 REG("latency", S_IRUGO, proc_lstats_operations),
3294 #ifdef CONFIG_PROC_PID_CPUSET
3295 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3297 #ifdef CONFIG_CGROUPS
3298 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3300 ONE("oom_score", S_IRUGO, proc_oom_score),
3301 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3302 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3303 #ifdef CONFIG_AUDITSYSCALL
3304 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3305 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3307 #ifdef CONFIG_FAULT_INJECTION
3308 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3310 #ifdef CONFIG_TASK_IO_ACCOUNTING
3311 ONE("io", S_IRUSR, proc_tid_io_accounting),
3313 #ifdef CONFIG_HARDWALL
3314 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3316 #ifdef CONFIG_USER_NS
3317 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3318 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3319 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3320 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3324 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3326 return proc_pident_readdir(file, ctx,
3327 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3330 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3332 return proc_pident_lookup(dir, dentry,
3333 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3336 static const struct file_operations proc_tid_base_operations = {
3337 .read = generic_read_dir,
3338 .iterate_shared = proc_tid_base_readdir,
3339 .llseek = generic_file_llseek,
3342 static const struct inode_operations proc_tid_base_inode_operations = {
3343 .lookup = proc_tid_base_lookup,
3344 .getattr = pid_getattr,
3345 .setattr = proc_setattr,
3348 static int proc_task_instantiate(struct inode *dir,
3349 struct dentry *dentry, struct task_struct *task, const void *ptr)
3351 struct inode *inode;
3352 inode = proc_pid_make_inode(dir->i_sb, task);
3356 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3357 inode->i_op = &proc_tid_base_inode_operations;
3358 inode->i_fop = &proc_tid_base_operations;
3359 inode->i_flags|=S_IMMUTABLE;
3361 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3362 ARRAY_SIZE(tid_base_stuff)));
3364 d_set_d_op(dentry, &pid_dentry_operations);
3366 d_add(dentry, inode);
3367 /* Close the race of the process dying before we return the dentry */
3368 if (pid_revalidate(dentry, 0))
3374 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3376 int result = -ENOENT;
3377 struct task_struct *task;
3378 struct task_struct *leader = get_proc_task(dir);
3380 struct pid_namespace *ns;
3385 tid = name_to_int(&dentry->d_name);
3389 ns = dentry->d_sb->s_fs_info;
3391 task = find_task_by_pid_ns(tid, ns);
3393 get_task_struct(task);
3397 if (!same_thread_group(leader, task))
3400 result = proc_task_instantiate(dir, dentry, task, NULL);
3402 put_task_struct(task);
3404 put_task_struct(leader);
3406 return ERR_PTR(result);
3410 * Find the first tid of a thread group to return to user space.
3412 * Usually this is just the thread group leader, but if the users
3413 * buffer was too small or there was a seek into the middle of the
3414 * directory we have more work todo.
3416 * In the case of a short read we start with find_task_by_pid.
3418 * In the case of a seek we start with the leader and walk nr
3421 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3422 struct pid_namespace *ns)
3424 struct task_struct *pos, *task;
3425 unsigned long nr = f_pos;
3427 if (nr != f_pos) /* 32bit overflow? */
3431 task = pid_task(pid, PIDTYPE_PID);
3435 /* Attempt to start with the tid of a thread */
3437 pos = find_task_by_pid_ns(tid, ns);
3438 if (pos && same_thread_group(pos, task))
3442 /* If nr exceeds the number of threads there is nothing todo */
3443 if (nr >= get_nr_threads(task))
3446 /* If we haven't found our starting place yet start
3447 * with the leader and walk nr threads forward.
3449 pos = task = task->group_leader;
3453 } while_each_thread(task, pos);
3458 get_task_struct(pos);
3465 * Find the next thread in the thread list.
3466 * Return NULL if there is an error or no next thread.
3468 * The reference to the input task_struct is released.
3470 static struct task_struct *next_tid(struct task_struct *start)
3472 struct task_struct *pos = NULL;
3474 if (pid_alive(start)) {
3475 pos = next_thread(start);
3476 if (thread_group_leader(pos))
3479 get_task_struct(pos);
3482 put_task_struct(start);
3486 /* for the /proc/TGID/task/ directories */
3487 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3489 struct inode *inode = file_inode(file);
3490 struct task_struct *task;
3491 struct pid_namespace *ns;
3494 if (proc_inode_is_dead(inode))
3497 if (!dir_emit_dots(file, ctx))
3500 /* f_version caches the tgid value that the last readdir call couldn't
3501 * return. lseek aka telldir automagically resets f_version to 0.
3503 ns = inode->i_sb->s_fs_info;
3504 tid = (int)file->f_version;
3505 file->f_version = 0;
3506 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3508 task = next_tid(task), ctx->pos++) {
3509 char name[PROC_NUMBUF];
3511 tid = task_pid_nr_ns(task, ns);
3512 len = snprintf(name, sizeof(name), "%d", tid);
3513 if (!proc_fill_cache(file, ctx, name, len,
3514 proc_task_instantiate, task, NULL)) {
3515 /* returning this tgid failed, save it as the first
3516 * pid for the next readir call */
3517 file->f_version = (u64)tid;
3518 put_task_struct(task);
3526 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3528 struct inode *inode = d_inode(dentry);
3529 struct task_struct *p = get_proc_task(inode);
3530 generic_fillattr(inode, stat);
3533 stat->nlink += get_nr_threads(p);
3540 static const struct inode_operations proc_task_inode_operations = {
3541 .lookup = proc_task_lookup,
3542 .getattr = proc_task_getattr,
3543 .setattr = proc_setattr,
3544 .permission = proc_pid_permission,
3547 static const struct file_operations proc_task_operations = {
3548 .read = generic_read_dir,
3549 .iterate_shared = proc_task_readdir,
3550 .llseek = generic_file_llseek,