4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * proc base directory handling functions
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
50 #include <linux/uaccess.h>
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/flex_array.h>
89 #include <linux/posix-timers.h>
90 #ifdef CONFIG_HARDWALL
91 #include <asm/hardwall.h>
93 #include <trace/events/oom.h>
98 * Implementing inode permission operations in /proc is almost
99 * certainly an error. Permission checks need to happen during
100 * each system call not at open time. The reason is that most of
101 * what we wish to check for permissions in /proc varies at runtime.
103 * The classic example of a problem is opening file descriptors
104 * in /proc for a task before it execs a suid executable.
108 static u8 nlink_tgid;
114 const struct inode_operations *iop;
115 const struct file_operations *fop;
119 #define NOD(NAME, MODE, IOP, FOP, OP) { \
121 .len = sizeof(NAME) - 1, \
128 #define DIR(NAME, MODE, iops, fops) \
129 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
130 #define LNK(NAME, get_link) \
131 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
132 &proc_pid_link_inode_operations, NULL, \
133 { .proc_get_link = get_link } )
134 #define REG(NAME, MODE, fops) \
135 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
136 #define ONE(NAME, MODE, show) \
137 NOD(NAME, (S_IFREG|(MODE)), \
138 NULL, &proc_single_file_operations, \
139 { .proc_show = show } )
142 * Count the number of hardlinks for the pid_entry table, excluding the .
145 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
160 static int get_task_root(struct task_struct *task, struct path *root)
162 int result = -ENOENT;
166 get_fs_root(task->fs, root);
173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
175 struct task_struct *task = get_proc_task(d_inode(dentry));
176 int result = -ENOENT;
181 get_fs_pwd(task->fs, path);
185 put_task_struct(task);
190 static int proc_root_link(struct dentry *dentry, struct path *path)
192 struct task_struct *task = get_proc_task(d_inode(dentry));
193 int result = -ENOENT;
196 result = get_task_root(task, path);
197 put_task_struct(task);
202 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
203 size_t _count, loff_t *pos)
205 struct task_struct *tsk;
206 struct mm_struct *mm;
208 unsigned long count = _count;
209 unsigned long arg_start, arg_end, env_start, env_end;
210 unsigned long len1, len2, len;
217 tsk = get_proc_task(file_inode(file));
220 mm = get_task_mm(tsk);
221 put_task_struct(tsk);
224 /* Check if process spawned far enough to have cmdline. */
230 page = (char *)__get_free_page(GFP_TEMPORARY);
236 down_read(&mm->mmap_sem);
237 arg_start = mm->arg_start;
238 arg_end = mm->arg_end;
239 env_start = mm->env_start;
240 env_end = mm->env_end;
241 up_read(&mm->mmap_sem);
243 BUG_ON(arg_start > arg_end);
244 BUG_ON(env_start > env_end);
246 len1 = arg_end - arg_start;
247 len2 = env_end - env_start;
255 * Inherently racy -- command line shares address space
256 * with code and data.
258 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
265 /* Command line (set of strings) occupies whole ARGV. */
269 p = arg_start + *pos;
271 while (count > 0 && len > 0) {
275 _count = min3(count, len, PAGE_SIZE);
276 nr_read = access_remote_vm(mm, p, page, _count, 0);
282 if (copy_to_user(buf, page, nr_read)) {
295 * Command line (1 string) occupies ARGV and maybe
298 if (len1 + len2 <= *pos)
303 p = arg_start + *pos;
305 while (count > 0 && len > 0) {
306 unsigned int _count, l;
310 _count = min3(count, len, PAGE_SIZE);
311 nr_read = access_remote_vm(mm, p, page, _count, 0);
318 * Command line can be shorter than whole ARGV
319 * even if last "marker" byte says it is not.
322 l = strnlen(page, nr_read);
328 if (copy_to_user(buf, page, nr_read)) {
344 * Command line (1 string) occupies ARGV and
348 p = env_start + *pos - len1;
349 len = len1 + len2 - *pos;
354 while (count > 0 && len > 0) {
355 unsigned int _count, l;
359 _count = min3(count, len, PAGE_SIZE);
360 nr_read = access_remote_vm(mm, p, page, _count, 0);
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;
843 page = (char *)__get_free_page(GFP_TEMPORARY);
848 if (!atomic_inc_not_zero(&mm->mm_users))
851 /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
857 int this_len = min_t(int, count, PAGE_SIZE);
859 if (write && copy_from_user(page, buf, this_len)) {
864 this_len = access_remote_vm(mm, addr, page, this_len, flags);
871 if (!write && copy_to_user(buf, page, this_len)) {
885 free_page((unsigned long) page);
889 static ssize_t mem_read(struct file *file, char __user *buf,
890 size_t count, loff_t *ppos)
892 return mem_rw(file, buf, count, ppos, 0);
895 static ssize_t mem_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
898 return mem_rw(file, (char __user*)buf, count, ppos, 1);
901 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
905 file->f_pos = offset;
908 file->f_pos += offset;
913 force_successful_syscall_return();
917 static int mem_release(struct inode *inode, struct file *file)
919 struct mm_struct *mm = file->private_data;
925 static const struct file_operations proc_mem_operations = {
930 .release = mem_release,
933 static int environ_open(struct inode *inode, struct file *file)
935 return __mem_open(inode, file, PTRACE_MODE_READ);
938 static ssize_t environ_read(struct file *file, char __user *buf,
939 size_t count, loff_t *ppos)
942 unsigned long src = *ppos;
944 struct mm_struct *mm = file->private_data;
945 unsigned long env_start, env_end;
947 /* Ensure the process spawned far enough to have an environment. */
948 if (!mm || !mm->env_end)
951 page = (char *)__get_free_page(GFP_TEMPORARY);
956 if (!atomic_inc_not_zero(&mm->mm_users))
959 down_read(&mm->mmap_sem);
960 env_start = mm->env_start;
961 env_end = mm->env_end;
962 up_read(&mm->mmap_sem);
965 size_t this_len, max_len;
968 if (src >= (env_end - env_start))
971 this_len = env_end - (env_start + src);
973 max_len = min_t(size_t, PAGE_SIZE, count);
974 this_len = min(max_len, this_len);
976 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
983 if (copy_to_user(buf, page, retval)) {
997 free_page((unsigned long) page);
1001 static const struct file_operations proc_environ_operations = {
1002 .open = environ_open,
1003 .read = environ_read,
1004 .llseek = generic_file_llseek,
1005 .release = mem_release,
1008 static int auxv_open(struct inode *inode, struct file *file)
1010 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1013 static ssize_t auxv_read(struct file *file, char __user *buf,
1014 size_t count, loff_t *ppos)
1016 struct mm_struct *mm = file->private_data;
1017 unsigned int nwords = 0;
1023 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1024 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1025 nwords * sizeof(mm->saved_auxv[0]));
1028 static const struct file_operations proc_auxv_operations = {
1031 .llseek = generic_file_llseek,
1032 .release = mem_release,
1035 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1038 struct task_struct *task = get_proc_task(file_inode(file));
1039 char buffer[PROC_NUMBUF];
1040 int oom_adj = OOM_ADJUST_MIN;
1045 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1046 oom_adj = OOM_ADJUST_MAX;
1048 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1050 put_task_struct(task);
1051 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1052 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1055 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1057 static DEFINE_MUTEX(oom_adj_mutex);
1058 struct mm_struct *mm = NULL;
1059 struct task_struct *task;
1062 task = get_proc_task(file_inode(file));
1066 mutex_lock(&oom_adj_mutex);
1068 if (oom_adj < task->signal->oom_score_adj &&
1069 !capable(CAP_SYS_RESOURCE)) {
1074 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1075 * /proc/pid/oom_score_adj instead.
1077 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1078 current->comm, task_pid_nr(current), task_pid_nr(task),
1081 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1082 !capable(CAP_SYS_RESOURCE)) {
1089 * Make sure we will check other processes sharing the mm if this is
1090 * not vfrok which wants its own oom_score_adj.
1091 * pin the mm so it doesn't go away and get reused after task_unlock
1093 if (!task->vfork_done) {
1094 struct task_struct *p = find_lock_task_mm(task);
1097 if (atomic_read(&p->mm->mm_users) > 1) {
1099 atomic_inc(&mm->mm_count);
1105 task->signal->oom_score_adj = oom_adj;
1106 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1107 task->signal->oom_score_adj_min = (short)oom_adj;
1108 trace_oom_score_adj_update(task);
1111 struct task_struct *p;
1114 for_each_process(p) {
1115 if (same_thread_group(task, p))
1118 /* do not touch kernel threads or the global init */
1119 if (p->flags & PF_KTHREAD || is_global_init(p))
1123 if (!p->vfork_done && process_shares_mm(p, mm)) {
1124 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",
1125 task_pid_nr(p), p->comm,
1126 p->signal->oom_score_adj, oom_adj,
1127 task_pid_nr(task), task->comm);
1128 p->signal->oom_score_adj = oom_adj;
1129 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1130 p->signal->oom_score_adj_min = (short)oom_adj;
1138 mutex_unlock(&oom_adj_mutex);
1139 put_task_struct(task);
1144 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1145 * kernels. The effective policy is defined by oom_score_adj, which has a
1146 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1147 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1148 * Processes that become oom disabled via oom_adj will still be oom disabled
1149 * with this implementation.
1151 * oom_adj cannot be removed since existing userspace binaries use it.
1153 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1154 size_t count, loff_t *ppos)
1156 char buffer[PROC_NUMBUF];
1160 memset(buffer, 0, sizeof(buffer));
1161 if (count > sizeof(buffer) - 1)
1162 count = sizeof(buffer) - 1;
1163 if (copy_from_user(buffer, buf, count)) {
1168 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1171 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1172 oom_adj != OOM_DISABLE) {
1178 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1179 * value is always attainable.
1181 if (oom_adj == OOM_ADJUST_MAX)
1182 oom_adj = OOM_SCORE_ADJ_MAX;
1184 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1186 err = __set_oom_adj(file, oom_adj, true);
1188 return err < 0 ? err : count;
1191 static const struct file_operations proc_oom_adj_operations = {
1192 .read = oom_adj_read,
1193 .write = oom_adj_write,
1194 .llseek = generic_file_llseek,
1197 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1198 size_t count, loff_t *ppos)
1200 struct task_struct *task = get_proc_task(file_inode(file));
1201 char buffer[PROC_NUMBUF];
1202 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1207 oom_score_adj = task->signal->oom_score_adj;
1208 put_task_struct(task);
1209 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1210 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1213 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1214 size_t count, loff_t *ppos)
1216 char buffer[PROC_NUMBUF];
1220 memset(buffer, 0, sizeof(buffer));
1221 if (count > sizeof(buffer) - 1)
1222 count = sizeof(buffer) - 1;
1223 if (copy_from_user(buffer, buf, count)) {
1228 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1231 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1232 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1237 err = __set_oom_adj(file, oom_score_adj, false);
1239 return err < 0 ? err : count;
1242 static const struct file_operations proc_oom_score_adj_operations = {
1243 .read = oom_score_adj_read,
1244 .write = oom_score_adj_write,
1245 .llseek = default_llseek,
1248 #ifdef CONFIG_AUDITSYSCALL
1249 #define TMPBUFLEN 11
1250 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1251 size_t count, loff_t *ppos)
1253 struct inode * inode = file_inode(file);
1254 struct task_struct *task = get_proc_task(inode);
1256 char tmpbuf[TMPBUFLEN];
1260 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1261 from_kuid(file->f_cred->user_ns,
1262 audit_get_loginuid(task)));
1263 put_task_struct(task);
1264 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1267 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1268 size_t count, loff_t *ppos)
1270 struct inode * inode = file_inode(file);
1276 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1283 /* No partial writes. */
1287 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1291 /* is userspace tring to explicitly UNSET the loginuid? */
1292 if (loginuid == AUDIT_UID_UNSET) {
1293 kloginuid = INVALID_UID;
1295 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1296 if (!uid_valid(kloginuid))
1300 rv = audit_set_loginuid(kloginuid);
1306 static const struct file_operations proc_loginuid_operations = {
1307 .read = proc_loginuid_read,
1308 .write = proc_loginuid_write,
1309 .llseek = generic_file_llseek,
1312 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1313 size_t count, loff_t *ppos)
1315 struct inode * inode = file_inode(file);
1316 struct task_struct *task = get_proc_task(inode);
1318 char tmpbuf[TMPBUFLEN];
1322 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1323 audit_get_sessionid(task));
1324 put_task_struct(task);
1325 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1328 static const struct file_operations proc_sessionid_operations = {
1329 .read = proc_sessionid_read,
1330 .llseek = generic_file_llseek,
1334 #ifdef CONFIG_FAULT_INJECTION
1335 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1336 size_t count, loff_t *ppos)
1338 struct task_struct *task = get_proc_task(file_inode(file));
1339 char buffer[PROC_NUMBUF];
1345 make_it_fail = task->make_it_fail;
1346 put_task_struct(task);
1348 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1350 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1353 static ssize_t proc_fault_inject_write(struct file * file,
1354 const char __user * buf, size_t count, loff_t *ppos)
1356 struct task_struct *task;
1357 char buffer[PROC_NUMBUF];
1361 if (!capable(CAP_SYS_RESOURCE))
1363 memset(buffer, 0, sizeof(buffer));
1364 if (count > sizeof(buffer) - 1)
1365 count = sizeof(buffer) - 1;
1366 if (copy_from_user(buffer, buf, count))
1368 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1371 if (make_it_fail < 0 || make_it_fail > 1)
1374 task = get_proc_task(file_inode(file));
1377 task->make_it_fail = make_it_fail;
1378 put_task_struct(task);
1383 static const struct file_operations proc_fault_inject_operations = {
1384 .read = proc_fault_inject_read,
1385 .write = proc_fault_inject_write,
1386 .llseek = generic_file_llseek,
1391 #ifdef CONFIG_SCHED_DEBUG
1393 * Print out various scheduling related per-task fields:
1395 static int sched_show(struct seq_file *m, void *v)
1397 struct inode *inode = m->private;
1398 struct task_struct *p;
1400 p = get_proc_task(inode);
1403 proc_sched_show_task(p, m);
1411 sched_write(struct file *file, const char __user *buf,
1412 size_t count, loff_t *offset)
1414 struct inode *inode = file_inode(file);
1415 struct task_struct *p;
1417 p = get_proc_task(inode);
1420 proc_sched_set_task(p);
1427 static int sched_open(struct inode *inode, struct file *filp)
1429 return single_open(filp, sched_show, inode);
1432 static const struct file_operations proc_pid_sched_operations = {
1435 .write = sched_write,
1436 .llseek = seq_lseek,
1437 .release = single_release,
1442 #ifdef CONFIG_SCHED_AUTOGROUP
1444 * Print out autogroup related information:
1446 static int sched_autogroup_show(struct seq_file *m, void *v)
1448 struct inode *inode = m->private;
1449 struct task_struct *p;
1451 p = get_proc_task(inode);
1454 proc_sched_autogroup_show_task(p, m);
1462 sched_autogroup_write(struct file *file, const char __user *buf,
1463 size_t count, loff_t *offset)
1465 struct inode *inode = file_inode(file);
1466 struct task_struct *p;
1467 char buffer[PROC_NUMBUF];
1471 memset(buffer, 0, sizeof(buffer));
1472 if (count > sizeof(buffer) - 1)
1473 count = sizeof(buffer) - 1;
1474 if (copy_from_user(buffer, buf, count))
1477 err = kstrtoint(strstrip(buffer), 0, &nice);
1481 p = get_proc_task(inode);
1485 err = proc_sched_autogroup_set_nice(p, nice);
1494 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1498 ret = single_open(filp, sched_autogroup_show, NULL);
1500 struct seq_file *m = filp->private_data;
1507 static const struct file_operations proc_pid_sched_autogroup_operations = {
1508 .open = sched_autogroup_open,
1510 .write = sched_autogroup_write,
1511 .llseek = seq_lseek,
1512 .release = single_release,
1515 #endif /* CONFIG_SCHED_AUTOGROUP */
1517 static ssize_t comm_write(struct file *file, const char __user *buf,
1518 size_t count, loff_t *offset)
1520 struct inode *inode = file_inode(file);
1521 struct task_struct *p;
1522 char buffer[TASK_COMM_LEN];
1523 const size_t maxlen = sizeof(buffer) - 1;
1525 memset(buffer, 0, sizeof(buffer));
1526 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1529 p = get_proc_task(inode);
1533 if (same_thread_group(current, p))
1534 set_task_comm(p, buffer);
1543 static int comm_show(struct seq_file *m, void *v)
1545 struct inode *inode = m->private;
1546 struct task_struct *p;
1548 p = get_proc_task(inode);
1553 seq_printf(m, "%s\n", p->comm);
1561 static int comm_open(struct inode *inode, struct file *filp)
1563 return single_open(filp, comm_show, inode);
1566 static const struct file_operations proc_pid_set_comm_operations = {
1569 .write = comm_write,
1570 .llseek = seq_lseek,
1571 .release = single_release,
1574 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1576 struct task_struct *task;
1577 struct file *exe_file;
1579 task = get_proc_task(d_inode(dentry));
1582 exe_file = get_task_exe_file(task);
1583 put_task_struct(task);
1585 *exe_path = exe_file->f_path;
1586 path_get(&exe_file->f_path);
1593 static const char *proc_pid_get_link(struct dentry *dentry,
1594 struct inode *inode,
1595 struct delayed_call *done)
1598 int error = -EACCES;
1601 return ERR_PTR(-ECHILD);
1603 /* Are we allowed to snoop on the tasks file descriptors? */
1604 if (!proc_fd_access_allowed(inode))
1607 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1611 nd_jump_link(&path);
1614 return ERR_PTR(error);
1617 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1619 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1626 pathname = d_path(path, tmp, PAGE_SIZE);
1627 len = PTR_ERR(pathname);
1628 if (IS_ERR(pathname))
1630 len = tmp + PAGE_SIZE - 1 - pathname;
1634 if (copy_to_user(buffer, pathname, len))
1637 free_page((unsigned long)tmp);
1641 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1643 int error = -EACCES;
1644 struct inode *inode = d_inode(dentry);
1647 /* Are we allowed to snoop on the tasks file descriptors? */
1648 if (!proc_fd_access_allowed(inode))
1651 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1655 error = do_proc_readlink(&path, buffer, buflen);
1661 const struct inode_operations proc_pid_link_inode_operations = {
1662 .readlink = proc_pid_readlink,
1663 .get_link = proc_pid_get_link,
1664 .setattr = proc_setattr,
1668 /* building an inode */
1670 struct inode *proc_pid_make_inode(struct super_block * sb,
1671 struct task_struct *task, umode_t mode)
1673 struct inode * inode;
1674 struct proc_inode *ei;
1675 const struct cred *cred;
1677 /* We need a new inode */
1679 inode = new_inode(sb);
1685 inode->i_mode = mode;
1686 inode->i_ino = get_next_ino();
1687 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1688 inode->i_op = &proc_def_inode_operations;
1691 * grab the reference to task.
1693 ei->pid = get_task_pid(task, PIDTYPE_PID);
1697 if (task_dumpable(task)) {
1699 cred = __task_cred(task);
1700 inode->i_uid = cred->euid;
1701 inode->i_gid = cred->egid;
1704 security_task_to_inode(task, inode);
1714 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1716 struct inode *inode = d_inode(dentry);
1717 struct task_struct *task;
1718 const struct cred *cred;
1719 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1721 generic_fillattr(inode, stat);
1724 stat->uid = GLOBAL_ROOT_UID;
1725 stat->gid = GLOBAL_ROOT_GID;
1726 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1728 if (!has_pid_permissions(pid, task, 2)) {
1731 * This doesn't prevent learning whether PID exists,
1732 * it only makes getattr() consistent with readdir().
1736 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1737 task_dumpable(task)) {
1738 cred = __task_cred(task);
1739 stat->uid = cred->euid;
1740 stat->gid = cred->egid;
1750 * Exceptional case: normally we are not allowed to unhash a busy
1751 * directory. In this case, however, we can do it - no aliasing problems
1752 * due to the way we treat inodes.
1754 * Rewrite the inode's ownerships here because the owning task may have
1755 * performed a setuid(), etc.
1757 * Before the /proc/pid/status file was created the only way to read
1758 * the effective uid of a /process was to stat /proc/pid. Reading
1759 * /proc/pid/status is slow enough that procps and other packages
1760 * kept stating /proc/pid. To keep the rules in /proc simple I have
1761 * made this apply to all per process world readable and executable
1764 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1766 struct inode *inode;
1767 struct task_struct *task;
1768 const struct cred *cred;
1770 if (flags & LOOKUP_RCU)
1773 inode = d_inode(dentry);
1774 task = get_proc_task(inode);
1777 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1778 task_dumpable(task)) {
1780 cred = __task_cred(task);
1781 inode->i_uid = cred->euid;
1782 inode->i_gid = cred->egid;
1785 inode->i_uid = GLOBAL_ROOT_UID;
1786 inode->i_gid = GLOBAL_ROOT_GID;
1788 inode->i_mode &= ~(S_ISUID | S_ISGID);
1789 security_task_to_inode(task, inode);
1790 put_task_struct(task);
1796 static inline bool proc_inode_is_dead(struct inode *inode)
1798 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1801 int pid_delete_dentry(const struct dentry *dentry)
1803 /* Is the task we represent dead?
1804 * If so, then don't put the dentry on the lru list,
1805 * kill it immediately.
1807 return proc_inode_is_dead(d_inode(dentry));
1810 const struct dentry_operations pid_dentry_operations =
1812 .d_revalidate = pid_revalidate,
1813 .d_delete = pid_delete_dentry,
1819 * Fill a directory entry.
1821 * If possible create the dcache entry and derive our inode number and
1822 * file type from dcache entry.
1824 * Since all of the proc inode numbers are dynamically generated, the inode
1825 * numbers do not exist until the inode is cache. This means creating the
1826 * the dcache entry in readdir is necessary to keep the inode numbers
1827 * reported by readdir in sync with the inode numbers reported
1830 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1831 const char *name, int len,
1832 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1834 struct dentry *child, *dir = file->f_path.dentry;
1835 struct qstr qname = QSTR_INIT(name, len);
1836 struct inode *inode;
1840 child = d_hash_and_lookup(dir, &qname);
1842 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1843 child = d_alloc_parallel(dir, &qname, &wq);
1845 goto end_instantiate;
1846 if (d_in_lookup(child)) {
1847 int err = instantiate(d_inode(dir), child, task, ptr);
1848 d_lookup_done(child);
1851 goto end_instantiate;
1855 inode = d_inode(child);
1857 type = inode->i_mode >> 12;
1859 return dir_emit(ctx, name, len, ino, type);
1862 return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1866 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1867 * which represent vma start and end addresses.
1869 static int dname_to_vma_addr(struct dentry *dentry,
1870 unsigned long *start, unsigned long *end)
1872 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1878 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1880 unsigned long vm_start, vm_end;
1881 bool exact_vma_exists = false;
1882 struct mm_struct *mm = NULL;
1883 struct task_struct *task;
1884 const struct cred *cred;
1885 struct inode *inode;
1888 if (flags & LOOKUP_RCU)
1891 inode = d_inode(dentry);
1892 task = get_proc_task(inode);
1896 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1897 if (IS_ERR_OR_NULL(mm))
1900 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1901 down_read(&mm->mmap_sem);
1902 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1903 up_read(&mm->mmap_sem);
1908 if (exact_vma_exists) {
1909 if (task_dumpable(task)) {
1911 cred = __task_cred(task);
1912 inode->i_uid = cred->euid;
1913 inode->i_gid = cred->egid;
1916 inode->i_uid = GLOBAL_ROOT_UID;
1917 inode->i_gid = GLOBAL_ROOT_GID;
1919 security_task_to_inode(task, inode);
1924 put_task_struct(task);
1930 static const struct dentry_operations tid_map_files_dentry_operations = {
1931 .d_revalidate = map_files_d_revalidate,
1932 .d_delete = pid_delete_dentry,
1935 static int map_files_get_link(struct dentry *dentry, struct path *path)
1937 unsigned long vm_start, vm_end;
1938 struct vm_area_struct *vma;
1939 struct task_struct *task;
1940 struct mm_struct *mm;
1944 task = get_proc_task(d_inode(dentry));
1948 mm = get_task_mm(task);
1949 put_task_struct(task);
1953 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1958 down_read(&mm->mmap_sem);
1959 vma = find_exact_vma(mm, vm_start, vm_end);
1960 if (vma && vma->vm_file) {
1961 *path = vma->vm_file->f_path;
1965 up_read(&mm->mmap_sem);
1973 struct map_files_info {
1976 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1980 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1981 * symlinks may be used to bypass permissions on ancestor directories in the
1982 * path to the file in question.
1985 proc_map_files_get_link(struct dentry *dentry,
1986 struct inode *inode,
1987 struct delayed_call *done)
1989 if (!capable(CAP_SYS_ADMIN))
1990 return ERR_PTR(-EPERM);
1992 return proc_pid_get_link(dentry, inode, done);
1996 * Identical to proc_pid_link_inode_operations except for get_link()
1998 static const struct inode_operations proc_map_files_link_inode_operations = {
1999 .readlink = proc_pid_readlink,
2000 .get_link = proc_map_files_get_link,
2001 .setattr = proc_setattr,
2005 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2006 struct task_struct *task, const void *ptr)
2008 fmode_t mode = (fmode_t)(unsigned long)ptr;
2009 struct proc_inode *ei;
2010 struct inode *inode;
2012 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2013 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2014 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2019 ei->op.proc_get_link = map_files_get_link;
2021 inode->i_op = &proc_map_files_link_inode_operations;
2024 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2025 d_add(dentry, inode);
2030 static struct dentry *proc_map_files_lookup(struct inode *dir,
2031 struct dentry *dentry, unsigned int flags)
2033 unsigned long vm_start, vm_end;
2034 struct vm_area_struct *vma;
2035 struct task_struct *task;
2037 struct mm_struct *mm;
2040 task = get_proc_task(dir);
2045 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2049 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2052 mm = get_task_mm(task);
2056 down_read(&mm->mmap_sem);
2057 vma = find_exact_vma(mm, vm_start, vm_end);
2062 result = proc_map_files_instantiate(dir, dentry, task,
2063 (void *)(unsigned long)vma->vm_file->f_mode);
2066 up_read(&mm->mmap_sem);
2069 put_task_struct(task);
2071 return ERR_PTR(result);
2074 static const struct inode_operations proc_map_files_inode_operations = {
2075 .lookup = proc_map_files_lookup,
2076 .permission = proc_fd_permission,
2077 .setattr = proc_setattr,
2081 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2083 struct vm_area_struct *vma;
2084 struct task_struct *task;
2085 struct mm_struct *mm;
2086 unsigned long nr_files, pos, i;
2087 struct flex_array *fa = NULL;
2088 struct map_files_info info;
2089 struct map_files_info *p;
2093 task = get_proc_task(file_inode(file));
2098 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2102 if (!dir_emit_dots(file, ctx))
2105 mm = get_task_mm(task);
2108 down_read(&mm->mmap_sem);
2113 * We need two passes here:
2115 * 1) Collect vmas of mapped files with mmap_sem taken
2116 * 2) Release mmap_sem and instantiate entries
2118 * otherwise we get lockdep complained, since filldir()
2119 * routine might require mmap_sem taken in might_fault().
2122 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2123 if (vma->vm_file && ++pos > ctx->pos)
2128 fa = flex_array_alloc(sizeof(info), nr_files,
2130 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2134 flex_array_free(fa);
2135 up_read(&mm->mmap_sem);
2139 for (i = 0, vma = mm->mmap, pos = 2; vma;
2140 vma = vma->vm_next) {
2143 if (++pos <= ctx->pos)
2146 info.mode = vma->vm_file->f_mode;
2147 info.len = snprintf(info.name,
2148 sizeof(info.name), "%lx-%lx",
2149 vma->vm_start, vma->vm_end);
2150 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2154 up_read(&mm->mmap_sem);
2156 for (i = 0; i < nr_files; i++) {
2157 p = flex_array_get(fa, i);
2158 if (!proc_fill_cache(file, ctx,
2160 proc_map_files_instantiate,
2162 (void *)(unsigned long)p->mode))
2167 flex_array_free(fa);
2171 put_task_struct(task);
2176 static const struct file_operations proc_map_files_operations = {
2177 .read = generic_read_dir,
2178 .iterate_shared = proc_map_files_readdir,
2179 .llseek = generic_file_llseek,
2182 #ifdef CONFIG_CHECKPOINT_RESTORE
2183 struct timers_private {
2185 struct task_struct *task;
2186 struct sighand_struct *sighand;
2187 struct pid_namespace *ns;
2188 unsigned long flags;
2191 static void *timers_start(struct seq_file *m, loff_t *pos)
2193 struct timers_private *tp = m->private;
2195 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2197 return ERR_PTR(-ESRCH);
2199 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2201 return ERR_PTR(-ESRCH);
2203 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2206 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2208 struct timers_private *tp = m->private;
2209 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2212 static void timers_stop(struct seq_file *m, void *v)
2214 struct timers_private *tp = m->private;
2217 unlock_task_sighand(tp->task, &tp->flags);
2222 put_task_struct(tp->task);
2227 static int show_timer(struct seq_file *m, void *v)
2229 struct k_itimer *timer;
2230 struct timers_private *tp = m->private;
2232 static const char * const nstr[] = {
2233 [SIGEV_SIGNAL] = "signal",
2234 [SIGEV_NONE] = "none",
2235 [SIGEV_THREAD] = "thread",
2238 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2239 notify = timer->it_sigev_notify;
2241 seq_printf(m, "ID: %d\n", timer->it_id);
2242 seq_printf(m, "signal: %d/%p\n",
2243 timer->sigq->info.si_signo,
2244 timer->sigq->info.si_value.sival_ptr);
2245 seq_printf(m, "notify: %s/%s.%d\n",
2246 nstr[notify & ~SIGEV_THREAD_ID],
2247 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2248 pid_nr_ns(timer->it_pid, tp->ns));
2249 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2254 static const struct seq_operations proc_timers_seq_ops = {
2255 .start = timers_start,
2256 .next = timers_next,
2257 .stop = timers_stop,
2261 static int proc_timers_open(struct inode *inode, struct file *file)
2263 struct timers_private *tp;
2265 tp = __seq_open_private(file, &proc_timers_seq_ops,
2266 sizeof(struct timers_private));
2270 tp->pid = proc_pid(inode);
2271 tp->ns = inode->i_sb->s_fs_info;
2275 static const struct file_operations proc_timers_operations = {
2276 .open = proc_timers_open,
2278 .llseek = seq_lseek,
2279 .release = seq_release_private,
2283 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2284 size_t count, loff_t *offset)
2286 struct inode *inode = file_inode(file);
2287 struct task_struct *p;
2291 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2295 p = get_proc_task(inode);
2300 if (!capable(CAP_SYS_NICE)) {
2305 err = security_task_setscheduler(p);
2314 p->timer_slack_ns = p->default_timer_slack_ns;
2316 p->timer_slack_ns = slack_ns;
2325 static int timerslack_ns_show(struct seq_file *m, void *v)
2327 struct inode *inode = m->private;
2328 struct task_struct *p;
2331 p = get_proc_task(inode);
2337 if (!capable(CAP_SYS_NICE)) {
2341 err = security_task_getscheduler(p);
2347 seq_printf(m, "%llu\n", p->timer_slack_ns);
2356 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2358 return single_open(filp, timerslack_ns_show, inode);
2361 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2362 .open = timerslack_ns_open,
2364 .write = timerslack_ns_write,
2365 .llseek = seq_lseek,
2366 .release = single_release,
2369 static int proc_pident_instantiate(struct inode *dir,
2370 struct dentry *dentry, struct task_struct *task, const void *ptr)
2372 const struct pid_entry *p = ptr;
2373 struct inode *inode;
2374 struct proc_inode *ei;
2376 inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2381 if (S_ISDIR(inode->i_mode))
2382 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2384 inode->i_op = p->iop;
2386 inode->i_fop = p->fop;
2388 d_set_d_op(dentry, &pid_dentry_operations);
2389 d_add(dentry, inode);
2390 /* Close the race of the process dying before we return the dentry */
2391 if (pid_revalidate(dentry, 0))
2397 static struct dentry *proc_pident_lookup(struct inode *dir,
2398 struct dentry *dentry,
2399 const struct pid_entry *ents,
2403 struct task_struct *task = get_proc_task(dir);
2404 const struct pid_entry *p, *last;
2412 * Yes, it does not scale. And it should not. Don't add
2413 * new entries into /proc/<tgid>/ without very good reasons.
2415 last = &ents[nents];
2416 for (p = ents; p < last; p++) {
2417 if (p->len != dentry->d_name.len)
2419 if (!memcmp(dentry->d_name.name, p->name, p->len))
2425 error = proc_pident_instantiate(dir, dentry, task, p);
2427 put_task_struct(task);
2429 return ERR_PTR(error);
2432 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2433 const struct pid_entry *ents, unsigned int nents)
2435 struct task_struct *task = get_proc_task(file_inode(file));
2436 const struct pid_entry *p;
2441 if (!dir_emit_dots(file, ctx))
2444 if (ctx->pos >= nents + 2)
2447 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2448 if (!proc_fill_cache(file, ctx, p->name, p->len,
2449 proc_pident_instantiate, task, p))
2454 put_task_struct(task);
2458 #ifdef CONFIG_SECURITY
2459 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2460 size_t count, loff_t *ppos)
2462 struct inode * inode = file_inode(file);
2465 struct task_struct *task = get_proc_task(inode);
2470 length = security_getprocattr(task,
2471 (char*)file->f_path.dentry->d_name.name,
2473 put_task_struct(task);
2475 length = simple_read_from_buffer(buf, count, ppos, p, length);
2480 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2481 size_t count, loff_t *ppos)
2483 struct inode * inode = file_inode(file);
2486 struct task_struct *task = get_proc_task(inode);
2491 if (count > PAGE_SIZE)
2494 /* No partial writes. */
2499 page = memdup_user(buf, count);
2501 length = PTR_ERR(page);
2505 /* Guard against adverse ptrace interaction */
2506 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2510 length = security_setprocattr(task,
2511 (char*)file->f_path.dentry->d_name.name,
2513 mutex_unlock(&task->signal->cred_guard_mutex);
2517 put_task_struct(task);
2522 static const struct file_operations proc_pid_attr_operations = {
2523 .read = proc_pid_attr_read,
2524 .write = proc_pid_attr_write,
2525 .llseek = generic_file_llseek,
2528 static const struct pid_entry attr_dir_stuff[] = {
2529 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2530 REG("prev", S_IRUGO, proc_pid_attr_operations),
2531 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2532 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2534 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2537 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2539 return proc_pident_readdir(file, ctx,
2540 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2543 static const struct file_operations proc_attr_dir_operations = {
2544 .read = generic_read_dir,
2545 .iterate_shared = proc_attr_dir_readdir,
2546 .llseek = generic_file_llseek,
2549 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2550 struct dentry *dentry, unsigned int flags)
2552 return proc_pident_lookup(dir, dentry,
2553 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2556 static const struct inode_operations proc_attr_dir_inode_operations = {
2557 .lookup = proc_attr_dir_lookup,
2558 .getattr = pid_getattr,
2559 .setattr = proc_setattr,
2564 #ifdef CONFIG_ELF_CORE
2565 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2566 size_t count, loff_t *ppos)
2568 struct task_struct *task = get_proc_task(file_inode(file));
2569 struct mm_struct *mm;
2570 char buffer[PROC_NUMBUF];
2578 mm = get_task_mm(task);
2580 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2581 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2582 MMF_DUMP_FILTER_SHIFT));
2584 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2587 put_task_struct(task);
2592 static ssize_t proc_coredump_filter_write(struct file *file,
2593 const char __user *buf,
2597 struct task_struct *task;
2598 struct mm_struct *mm;
2604 ret = kstrtouint_from_user(buf, count, 0, &val);
2609 task = get_proc_task(file_inode(file));
2613 mm = get_task_mm(task);
2618 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2620 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2622 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2627 put_task_struct(task);
2634 static const struct file_operations proc_coredump_filter_operations = {
2635 .read = proc_coredump_filter_read,
2636 .write = proc_coredump_filter_write,
2637 .llseek = generic_file_llseek,
2641 #ifdef CONFIG_TASK_IO_ACCOUNTING
2642 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2644 struct task_io_accounting acct = task->ioac;
2645 unsigned long flags;
2648 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2652 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2657 if (whole && lock_task_sighand(task, &flags)) {
2658 struct task_struct *t = task;
2660 task_io_accounting_add(&acct, &task->signal->ioac);
2661 while_each_thread(task, t)
2662 task_io_accounting_add(&acct, &t->ioac);
2664 unlock_task_sighand(task, &flags);
2671 "read_bytes: %llu\n"
2672 "write_bytes: %llu\n"
2673 "cancelled_write_bytes: %llu\n",
2674 (unsigned long long)acct.rchar,
2675 (unsigned long long)acct.wchar,
2676 (unsigned long long)acct.syscr,
2677 (unsigned long long)acct.syscw,
2678 (unsigned long long)acct.read_bytes,
2679 (unsigned long long)acct.write_bytes,
2680 (unsigned long long)acct.cancelled_write_bytes);
2684 mutex_unlock(&task->signal->cred_guard_mutex);
2688 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2689 struct pid *pid, struct task_struct *task)
2691 return do_io_accounting(task, m, 0);
2694 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2695 struct pid *pid, struct task_struct *task)
2697 return do_io_accounting(task, m, 1);
2699 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2701 #ifdef CONFIG_USER_NS
2702 static int proc_id_map_open(struct inode *inode, struct file *file,
2703 const struct seq_operations *seq_ops)
2705 struct user_namespace *ns = NULL;
2706 struct task_struct *task;
2707 struct seq_file *seq;
2710 task = get_proc_task(inode);
2713 ns = get_user_ns(task_cred_xxx(task, user_ns));
2715 put_task_struct(task);
2720 ret = seq_open(file, seq_ops);
2724 seq = file->private_data;
2734 static int proc_id_map_release(struct inode *inode, struct file *file)
2736 struct seq_file *seq = file->private_data;
2737 struct user_namespace *ns = seq->private;
2739 return seq_release(inode, file);
2742 static int proc_uid_map_open(struct inode *inode, struct file *file)
2744 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2747 static int proc_gid_map_open(struct inode *inode, struct file *file)
2749 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2752 static int proc_projid_map_open(struct inode *inode, struct file *file)
2754 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2757 static const struct file_operations proc_uid_map_operations = {
2758 .open = proc_uid_map_open,
2759 .write = proc_uid_map_write,
2761 .llseek = seq_lseek,
2762 .release = proc_id_map_release,
2765 static const struct file_operations proc_gid_map_operations = {
2766 .open = proc_gid_map_open,
2767 .write = proc_gid_map_write,
2769 .llseek = seq_lseek,
2770 .release = proc_id_map_release,
2773 static const struct file_operations proc_projid_map_operations = {
2774 .open = proc_projid_map_open,
2775 .write = proc_projid_map_write,
2777 .llseek = seq_lseek,
2778 .release = proc_id_map_release,
2781 static int proc_setgroups_open(struct inode *inode, struct file *file)
2783 struct user_namespace *ns = NULL;
2784 struct task_struct *task;
2788 task = get_proc_task(inode);
2791 ns = get_user_ns(task_cred_xxx(task, user_ns));
2793 put_task_struct(task);
2798 if (file->f_mode & FMODE_WRITE) {
2800 if (!ns_capable(ns, CAP_SYS_ADMIN))
2804 ret = single_open(file, &proc_setgroups_show, ns);
2815 static int proc_setgroups_release(struct inode *inode, struct file *file)
2817 struct seq_file *seq = file->private_data;
2818 struct user_namespace *ns = seq->private;
2819 int ret = single_release(inode, file);
2824 static const struct file_operations proc_setgroups_operations = {
2825 .open = proc_setgroups_open,
2826 .write = proc_setgroups_write,
2828 .llseek = seq_lseek,
2829 .release = proc_setgroups_release,
2831 #endif /* CONFIG_USER_NS */
2833 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2834 struct pid *pid, struct task_struct *task)
2836 int err = lock_trace(task);
2838 seq_printf(m, "%08x\n", task->personality);
2847 static const struct file_operations proc_task_operations;
2848 static const struct inode_operations proc_task_inode_operations;
2850 static const struct pid_entry tgid_base_stuff[] = {
2851 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2852 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2853 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2854 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2855 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2857 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2859 REG("environ", S_IRUSR, proc_environ_operations),
2860 REG("auxv", S_IRUSR, proc_auxv_operations),
2861 ONE("status", S_IRUGO, proc_pid_status),
2862 ONE("personality", S_IRUSR, proc_pid_personality),
2863 ONE("limits", S_IRUGO, proc_pid_limits),
2864 #ifdef CONFIG_SCHED_DEBUG
2865 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2867 #ifdef CONFIG_SCHED_AUTOGROUP
2868 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2870 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2871 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2872 ONE("syscall", S_IRUSR, proc_pid_syscall),
2874 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
2875 ONE("stat", S_IRUGO, proc_tgid_stat),
2876 ONE("statm", S_IRUGO, proc_pid_statm),
2877 REG("maps", S_IRUGO, proc_pid_maps_operations),
2879 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
2881 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2882 LNK("cwd", proc_cwd_link),
2883 LNK("root", proc_root_link),
2884 LNK("exe", proc_exe_link),
2885 REG("mounts", S_IRUGO, proc_mounts_operations),
2886 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2887 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2888 #ifdef CONFIG_PROC_PAGE_MONITOR
2889 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2890 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
2891 REG("pagemap", S_IRUSR, proc_pagemap_operations),
2893 #ifdef CONFIG_SECURITY
2894 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2896 #ifdef CONFIG_KALLSYMS
2897 ONE("wchan", S_IRUGO, proc_pid_wchan),
2899 #ifdef CONFIG_STACKTRACE
2900 ONE("stack", S_IRUSR, proc_pid_stack),
2902 #ifdef CONFIG_SCHED_INFO
2903 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2905 #ifdef CONFIG_LATENCYTOP
2906 REG("latency", S_IRUGO, proc_lstats_operations),
2908 #ifdef CONFIG_PROC_PID_CPUSET
2909 ONE("cpuset", S_IRUGO, proc_cpuset_show),
2911 #ifdef CONFIG_CGROUPS
2912 ONE("cgroup", S_IRUGO, proc_cgroup_show),
2914 ONE("oom_score", S_IRUGO, proc_oom_score),
2915 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2916 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2917 #ifdef CONFIG_AUDITSYSCALL
2918 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2919 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2921 #ifdef CONFIG_FAULT_INJECTION
2922 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2924 #ifdef CONFIG_ELF_CORE
2925 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2927 #ifdef CONFIG_TASK_IO_ACCOUNTING
2928 ONE("io", S_IRUSR, proc_tgid_io_accounting),
2930 #ifdef CONFIG_HARDWALL
2931 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
2933 #ifdef CONFIG_USER_NS
2934 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
2935 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
2936 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2937 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
2939 #ifdef CONFIG_CHECKPOINT_RESTORE
2940 REG("timers", S_IRUGO, proc_timers_operations),
2942 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2945 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2947 return proc_pident_readdir(file, ctx,
2948 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2951 static const struct file_operations proc_tgid_base_operations = {
2952 .read = generic_read_dir,
2953 .iterate_shared = proc_tgid_base_readdir,
2954 .llseek = generic_file_llseek,
2957 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2959 return proc_pident_lookup(dir, dentry,
2960 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2963 static const struct inode_operations proc_tgid_base_inode_operations = {
2964 .lookup = proc_tgid_base_lookup,
2965 .getattr = pid_getattr,
2966 .setattr = proc_setattr,
2967 .permission = proc_pid_permission,
2970 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2972 struct dentry *dentry, *leader, *dir;
2973 char buf[PROC_NUMBUF];
2977 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2978 /* no ->d_hash() rejects on procfs */
2979 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2981 d_invalidate(dentry);
2989 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2990 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2995 name.len = strlen(name.name);
2996 dir = d_hash_and_lookup(leader, &name);
2998 goto out_put_leader;
3001 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3002 dentry = d_hash_and_lookup(dir, &name);
3004 d_invalidate(dentry);
3016 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3017 * @task: task that should be flushed.
3019 * When flushing dentries from proc, one needs to flush them from global
3020 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3021 * in. This call is supposed to do all of this job.
3023 * Looks in the dcache for
3025 * /proc/@tgid/task/@pid
3026 * if either directory is present flushes it and all of it'ts children
3029 * It is safe and reasonable to cache /proc entries for a task until
3030 * that task exits. After that they just clog up the dcache with
3031 * useless entries, possibly causing useful dcache entries to be
3032 * flushed instead. This routine is proved to flush those useless
3033 * dcache entries at process exit time.
3035 * NOTE: This routine is just an optimization so it does not guarantee
3036 * that no dcache entries will exist at process exit time it
3037 * just makes it very unlikely that any will persist.
3040 void proc_flush_task(struct task_struct *task)
3043 struct pid *pid, *tgid;
3046 pid = task_pid(task);
3047 tgid = task_tgid(task);
3049 for (i = 0; i <= pid->level; i++) {
3050 upid = &pid->numbers[i];
3051 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3052 tgid->numbers[i].nr);
3056 static int proc_pid_instantiate(struct inode *dir,
3057 struct dentry * dentry,
3058 struct task_struct *task, const void *ptr)
3060 struct inode *inode;
3062 inode = proc_pid_make_inode(dir->i_sb, task, 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, nlink_tgid);
3072 d_set_d_op(dentry, &pid_dentry_operations);
3074 d_add(dentry, inode);
3075 /* Close the race of the process dying before we return the dentry */
3076 if (pid_revalidate(dentry, 0))
3082 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3084 int result = -ENOENT;
3085 struct task_struct *task;
3087 struct pid_namespace *ns;
3089 tgid = name_to_int(&dentry->d_name);
3093 ns = dentry->d_sb->s_fs_info;
3095 task = find_task_by_pid_ns(tgid, ns);
3097 get_task_struct(task);
3102 result = proc_pid_instantiate(dir, dentry, task, NULL);
3103 put_task_struct(task);
3105 return ERR_PTR(result);
3109 * Find the first task with tgid >= tgid
3114 struct task_struct *task;
3116 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3121 put_task_struct(iter.task);
3125 pid = find_ge_pid(iter.tgid, ns);
3127 iter.tgid = pid_nr_ns(pid, ns);
3128 iter.task = pid_task(pid, PIDTYPE_PID);
3129 /* What we to know is if the pid we have find is the
3130 * pid of a thread_group_leader. Testing for task
3131 * being a thread_group_leader is the obvious thing
3132 * todo but there is a window when it fails, due to
3133 * the pid transfer logic in de_thread.
3135 * So we perform the straight forward test of seeing
3136 * if the pid we have found is the pid of a thread
3137 * group leader, and don't worry if the task we have
3138 * found doesn't happen to be a thread group leader.
3139 * As we don't care in the case of readdir.
3141 if (!iter.task || !has_group_leader_pid(iter.task)) {
3145 get_task_struct(iter.task);
3151 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3153 /* for the /proc/ directory itself, after non-process stuff has been done */
3154 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3156 struct tgid_iter iter;
3157 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3158 loff_t pos = ctx->pos;
3160 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3163 if (pos == TGID_OFFSET - 2) {
3164 struct inode *inode = d_inode(ns->proc_self);
3165 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3167 ctx->pos = pos = pos + 1;
3169 if (pos == TGID_OFFSET - 1) {
3170 struct inode *inode = d_inode(ns->proc_thread_self);
3171 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3173 ctx->pos = pos = pos + 1;
3175 iter.tgid = pos - TGID_OFFSET;
3177 for (iter = next_tgid(ns, iter);
3179 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3180 char name[PROC_NUMBUF];
3182 if (!has_pid_permissions(ns, iter.task, 2))
3185 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3186 ctx->pos = iter.tgid + TGID_OFFSET;
3187 if (!proc_fill_cache(file, ctx, name, len,
3188 proc_pid_instantiate, iter.task, NULL)) {
3189 put_task_struct(iter.task);
3193 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3198 * proc_tid_comm_permission is a special permission function exclusively
3199 * used for the node /proc/<pid>/task/<tid>/comm.
3200 * It bypasses generic permission checks in the case where a task of the same
3201 * task group attempts to access the node.
3202 * The rationale behind this is that glibc and bionic access this node for
3203 * cross thread naming (pthread_set/getname_np(!self)). However, if
3204 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3205 * which locks out the cross thread naming implementation.
3206 * This function makes sure that the node is always accessible for members of
3207 * same thread group.
3209 static int proc_tid_comm_permission(struct inode *inode, int mask)
3211 bool is_same_tgroup;
3212 struct task_struct *task;
3214 task = get_proc_task(inode);
3217 is_same_tgroup = same_thread_group(current, task);
3218 put_task_struct(task);
3220 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3221 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3222 * read or written by the members of the corresponding
3228 return generic_permission(inode, mask);
3231 static const struct inode_operations proc_tid_comm_inode_operations = {
3232 .permission = proc_tid_comm_permission,
3238 static const struct pid_entry tid_base_stuff[] = {
3239 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3240 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3241 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3243 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3245 REG("environ", S_IRUSR, proc_environ_operations),
3246 REG("auxv", S_IRUSR, proc_auxv_operations),
3247 ONE("status", S_IRUGO, proc_pid_status),
3248 ONE("personality", S_IRUSR, proc_pid_personality),
3249 ONE("limits", S_IRUGO, proc_pid_limits),
3250 #ifdef CONFIG_SCHED_DEBUG
3251 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3253 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3254 &proc_tid_comm_inode_operations,
3255 &proc_pid_set_comm_operations, {}),
3256 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3257 ONE("syscall", S_IRUSR, proc_pid_syscall),
3259 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3260 ONE("stat", S_IRUGO, proc_tid_stat),
3261 ONE("statm", S_IRUGO, proc_pid_statm),
3262 REG("maps", S_IRUGO, proc_tid_maps_operations),
3263 #ifdef CONFIG_PROC_CHILDREN
3264 REG("children", S_IRUGO, proc_tid_children_operations),
3267 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3269 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3270 LNK("cwd", proc_cwd_link),
3271 LNK("root", proc_root_link),
3272 LNK("exe", proc_exe_link),
3273 REG("mounts", S_IRUGO, proc_mounts_operations),
3274 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3275 #ifdef CONFIG_PROC_PAGE_MONITOR
3276 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3277 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3278 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3280 #ifdef CONFIG_SECURITY
3281 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3283 #ifdef CONFIG_KALLSYMS
3284 ONE("wchan", S_IRUGO, proc_pid_wchan),
3286 #ifdef CONFIG_STACKTRACE
3287 ONE("stack", S_IRUSR, proc_pid_stack),
3289 #ifdef CONFIG_SCHED_INFO
3290 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3292 #ifdef CONFIG_LATENCYTOP
3293 REG("latency", S_IRUGO, proc_lstats_operations),
3295 #ifdef CONFIG_PROC_PID_CPUSET
3296 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3298 #ifdef CONFIG_CGROUPS
3299 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3301 ONE("oom_score", S_IRUGO, proc_oom_score),
3302 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3303 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3304 #ifdef CONFIG_AUDITSYSCALL
3305 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3306 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3308 #ifdef CONFIG_FAULT_INJECTION
3309 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3311 #ifdef CONFIG_TASK_IO_ACCOUNTING
3312 ONE("io", S_IRUSR, proc_tid_io_accounting),
3314 #ifdef CONFIG_HARDWALL
3315 ONE("hardwall", S_IRUGO, proc_pid_hardwall),
3317 #ifdef CONFIG_USER_NS
3318 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3319 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3320 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3321 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3325 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3327 return proc_pident_readdir(file, ctx,
3328 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3331 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3333 return proc_pident_lookup(dir, dentry,
3334 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3337 static const struct file_operations proc_tid_base_operations = {
3338 .read = generic_read_dir,
3339 .iterate_shared = proc_tid_base_readdir,
3340 .llseek = generic_file_llseek,
3343 static const struct inode_operations proc_tid_base_inode_operations = {
3344 .lookup = proc_tid_base_lookup,
3345 .getattr = pid_getattr,
3346 .setattr = proc_setattr,
3349 static int proc_task_instantiate(struct inode *dir,
3350 struct dentry *dentry, struct task_struct *task, const void *ptr)
3352 struct inode *inode;
3353 inode = proc_pid_make_inode(dir->i_sb, task, 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, nlink_tid);
3363 d_set_d_op(dentry, &pid_dentry_operations);
3365 d_add(dentry, inode);
3366 /* Close the race of the process dying before we return the dentry */
3367 if (pid_revalidate(dentry, 0))
3373 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3375 int result = -ENOENT;
3376 struct task_struct *task;
3377 struct task_struct *leader = get_proc_task(dir);
3379 struct pid_namespace *ns;
3384 tid = name_to_int(&dentry->d_name);
3388 ns = dentry->d_sb->s_fs_info;
3390 task = find_task_by_pid_ns(tid, ns);
3392 get_task_struct(task);
3396 if (!same_thread_group(leader, task))
3399 result = proc_task_instantiate(dir, dentry, task, NULL);
3401 put_task_struct(task);
3403 put_task_struct(leader);
3405 return ERR_PTR(result);
3409 * Find the first tid of a thread group to return to user space.
3411 * Usually this is just the thread group leader, but if the users
3412 * buffer was too small or there was a seek into the middle of the
3413 * directory we have more work todo.
3415 * In the case of a short read we start with find_task_by_pid.
3417 * In the case of a seek we start with the leader and walk nr
3420 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3421 struct pid_namespace *ns)
3423 struct task_struct *pos, *task;
3424 unsigned long nr = f_pos;
3426 if (nr != f_pos) /* 32bit overflow? */
3430 task = pid_task(pid, PIDTYPE_PID);
3434 /* Attempt to start with the tid of a thread */
3436 pos = find_task_by_pid_ns(tid, ns);
3437 if (pos && same_thread_group(pos, task))
3441 /* If nr exceeds the number of threads there is nothing todo */
3442 if (nr >= get_nr_threads(task))
3445 /* If we haven't found our starting place yet start
3446 * with the leader and walk nr threads forward.
3448 pos = task = task->group_leader;
3452 } while_each_thread(task, pos);
3457 get_task_struct(pos);
3464 * Find the next thread in the thread list.
3465 * Return NULL if there is an error or no next thread.
3467 * The reference to the input task_struct is released.
3469 static struct task_struct *next_tid(struct task_struct *start)
3471 struct task_struct *pos = NULL;
3473 if (pid_alive(start)) {
3474 pos = next_thread(start);
3475 if (thread_group_leader(pos))
3478 get_task_struct(pos);
3481 put_task_struct(start);
3485 /* for the /proc/TGID/task/ directories */
3486 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3488 struct inode *inode = file_inode(file);
3489 struct task_struct *task;
3490 struct pid_namespace *ns;
3493 if (proc_inode_is_dead(inode))
3496 if (!dir_emit_dots(file, ctx))
3499 /* f_version caches the tgid value that the last readdir call couldn't
3500 * return. lseek aka telldir automagically resets f_version to 0.
3502 ns = inode->i_sb->s_fs_info;
3503 tid = (int)file->f_version;
3504 file->f_version = 0;
3505 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3507 task = next_tid(task), ctx->pos++) {
3508 char name[PROC_NUMBUF];
3510 tid = task_pid_nr_ns(task, ns);
3511 len = snprintf(name, sizeof(name), "%d", tid);
3512 if (!proc_fill_cache(file, ctx, name, len,
3513 proc_task_instantiate, task, NULL)) {
3514 /* returning this tgid failed, save it as the first
3515 * pid for the next readir call */
3516 file->f_version = (u64)tid;
3517 put_task_struct(task);
3525 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3527 struct inode *inode = d_inode(dentry);
3528 struct task_struct *p = get_proc_task(inode);
3529 generic_fillattr(inode, stat);
3532 stat->nlink += get_nr_threads(p);
3539 static const struct inode_operations proc_task_inode_operations = {
3540 .lookup = proc_task_lookup,
3541 .getattr = proc_task_getattr,
3542 .setattr = proc_setattr,
3543 .permission = proc_pid_permission,
3546 static const struct file_operations proc_task_operations = {
3547 .read = generic_read_dir,
3548 .iterate_shared = proc_task_readdir,
3549 .llseek = generic_file_llseek,
3552 void __init set_proc_pid_nlink(void)
3554 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3555 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));