Merge branch 'for-4.18' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
[muen/linux.git] / fs / proc / base.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/proc/base.c
4  *
5  *  Copyright (C) 1991, 1992 Linus Torvalds
6  *
7  *  proc base directory handling functions
8  *
9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10  *  Instead of using magical inumbers to determine the kind of object
11  *  we allocate and fill in-core inodes upon lookup. They don't even
12  *  go into icache. We cache the reference to task_struct upon lookup too.
13  *  Eventually it should become a filesystem in its own. We don't use the
14  *  rest of procfs anymore.
15  *
16  *
17  *  Changelog:
18  *  17-Jan-2005
19  *  Allan Bezerra
20  *  Bruna Moreira <bruna.moreira@indt.org.br>
21  *  Edjard Mota <edjard.mota@indt.org.br>
22  *  Ilias Biris <ilias.biris@indt.org.br>
23  *  Mauricio Lin <mauricio.lin@indt.org.br>
24  *
25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26  *
27  *  A new process specific entry (smaps) included in /proc. It shows the
28  *  size of rss for each memory area. The maps entry lacks information
29  *  about physical memory size (rss) for each mapped file, i.e.,
30  *  rss information for executables and library files.
31  *  This additional information is useful for any tools that need to know
32  *  about physical memory consumption for a process specific library.
33  *
34  *  Changelog:
35  *  21-Feb-2005
36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37  *  Pud inclusion in the page table walking.
38  *
39  *  ChangeLog:
40  *  10-Mar-2005
41  *  10LE Instituto Nokia de Tecnologia - INdT:
42  *  A better way to walks through the page table as suggested by Hugh Dickins.
43  *
44  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
45  *  Smaps information related to shared, private, clean and dirty pages.
46  *
47  *  Paul Mundt <paul.mundt@nokia.com>:
48  *  Overall revision about smaps.
49  */
50
51 #include <linux/uaccess.h>
52
53 #include <linux/errno.h>
54 #include <linux/time.h>
55 #include <linux/proc_fs.h>
56 #include <linux/stat.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/init.h>
59 #include <linux/capability.h>
60 #include <linux/file.h>
61 #include <linux/fdtable.h>
62 #include <linux/string.h>
63 #include <linux/seq_file.h>
64 #include <linux/namei.h>
65 #include <linux/mnt_namespace.h>
66 #include <linux/mm.h>
67 #include <linux/swap.h>
68 #include <linux/rcupdate.h>
69 #include <linux/kallsyms.h>
70 #include <linux/stacktrace.h>
71 #include <linux/resource.h>
72 #include <linux/module.h>
73 #include <linux/mount.h>
74 #include <linux/security.h>
75 #include <linux/ptrace.h>
76 #include <linux/tracehook.h>
77 #include <linux/printk.h>
78 #include <linux/cache.h>
79 #include <linux/cgroup.h>
80 #include <linux/cpuset.h>
81 #include <linux/audit.h>
82 #include <linux/poll.h>
83 #include <linux/nsproxy.h>
84 #include <linux/oom.h>
85 #include <linux/elf.h>
86 #include <linux/pid_namespace.h>
87 #include <linux/user_namespace.h>
88 #include <linux/fs_struct.h>
89 #include <linux/slab.h>
90 #include <linux/sched/autogroup.h>
91 #include <linux/sched/mm.h>
92 #include <linux/sched/coredump.h>
93 #include <linux/sched/debug.h>
94 #include <linux/sched/stat.h>
95 #include <linux/flex_array.h>
96 #include <linux/posix-timers.h>
97 #include <trace/events/oom.h>
98 #include "internal.h"
99 #include "fd.h"
100
101 #include "../../lib/kstrtox.h"
102
103 /* NOTE:
104  *      Implementing inode permission operations in /proc is almost
105  *      certainly an error.  Permission checks need to happen during
106  *      each system call not at open time.  The reason is that most of
107  *      what we wish to check for permissions in /proc varies at runtime.
108  *
109  *      The classic example of a problem is opening file descriptors
110  *      in /proc for a task before it execs a suid executable.
111  */
112
113 static u8 nlink_tid __ro_after_init;
114 static u8 nlink_tgid __ro_after_init;
115
116 struct pid_entry {
117         const char *name;
118         unsigned int len;
119         umode_t mode;
120         const struct inode_operations *iop;
121         const struct file_operations *fop;
122         union proc_op op;
123 };
124
125 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
126         .name = (NAME),                                 \
127         .len  = sizeof(NAME) - 1,                       \
128         .mode = MODE,                                   \
129         .iop  = IOP,                                    \
130         .fop  = FOP,                                    \
131         .op   = OP,                                     \
132 }
133
134 #define DIR(NAME, MODE, iops, fops)     \
135         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
136 #define LNK(NAME, get_link)                                     \
137         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
138                 &proc_pid_link_inode_operations, NULL,          \
139                 { .proc_get_link = get_link } )
140 #define REG(NAME, MODE, fops)                           \
141         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
142 #define ONE(NAME, MODE, show)                           \
143         NOD(NAME, (S_IFREG|(MODE)),                     \
144                 NULL, &proc_single_file_operations,     \
145                 { .proc_show = show } )
146
147 /*
148  * Count the number of hardlinks for the pid_entry table, excluding the .
149  * and .. links.
150  */
151 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
152         unsigned int n)
153 {
154         unsigned int i;
155         unsigned int count;
156
157         count = 2;
158         for (i = 0; i < n; ++i) {
159                 if (S_ISDIR(entries[i].mode))
160                         ++count;
161         }
162
163         return count;
164 }
165
166 static int get_task_root(struct task_struct *task, struct path *root)
167 {
168         int result = -ENOENT;
169
170         task_lock(task);
171         if (task->fs) {
172                 get_fs_root(task->fs, root);
173                 result = 0;
174         }
175         task_unlock(task);
176         return result;
177 }
178
179 static int proc_cwd_link(struct dentry *dentry, struct path *path)
180 {
181         struct task_struct *task = get_proc_task(d_inode(dentry));
182         int result = -ENOENT;
183
184         if (task) {
185                 task_lock(task);
186                 if (task->fs) {
187                         get_fs_pwd(task->fs, path);
188                         result = 0;
189                 }
190                 task_unlock(task);
191                 put_task_struct(task);
192         }
193         return result;
194 }
195
196 static int proc_root_link(struct dentry *dentry, struct path *path)
197 {
198         struct task_struct *task = get_proc_task(d_inode(dentry));
199         int result = -ENOENT;
200
201         if (task) {
202                 result = get_task_root(task, path);
203                 put_task_struct(task);
204         }
205         return result;
206 }
207
208 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
209                                      size_t _count, loff_t *pos)
210 {
211         struct task_struct *tsk;
212         struct mm_struct *mm;
213         char *page;
214         unsigned long count = _count;
215         unsigned long arg_start, arg_end, env_start, env_end;
216         unsigned long len1, len2, len;
217         unsigned long p;
218         char c;
219         ssize_t rv;
220
221         BUG_ON(*pos < 0);
222
223         tsk = get_proc_task(file_inode(file));
224         if (!tsk)
225                 return -ESRCH;
226         mm = get_task_mm(tsk);
227         put_task_struct(tsk);
228         if (!mm)
229                 return 0;
230         /* Check if process spawned far enough to have cmdline. */
231         if (!mm->env_end) {
232                 rv = 0;
233                 goto out_mmput;
234         }
235
236         page = (char *)__get_free_page(GFP_KERNEL);
237         if (!page) {
238                 rv = -ENOMEM;
239                 goto out_mmput;
240         }
241
242         down_read(&mm->mmap_sem);
243         arg_start = mm->arg_start;
244         arg_end = mm->arg_end;
245         env_start = mm->env_start;
246         env_end = mm->env_end;
247         up_read(&mm->mmap_sem);
248
249         BUG_ON(arg_start > arg_end);
250         BUG_ON(env_start > env_end);
251
252         len1 = arg_end - arg_start;
253         len2 = env_end - env_start;
254
255         /* Empty ARGV. */
256         if (len1 == 0) {
257                 rv = 0;
258                 goto out_free_page;
259         }
260         /*
261          * Inherently racy -- command line shares address space
262          * with code and data.
263          */
264         rv = access_remote_vm(mm, arg_end - 1, &c, 1, FOLL_ANON);
265         if (rv <= 0)
266                 goto out_free_page;
267
268         rv = 0;
269
270         if (c == '\0') {
271                 /* Command line (set of strings) occupies whole ARGV. */
272                 if (len1 <= *pos)
273                         goto out_free_page;
274
275                 p = arg_start + *pos;
276                 len = len1 - *pos;
277                 while (count > 0 && len > 0) {
278                         unsigned int _count;
279                         int nr_read;
280
281                         _count = min3(count, len, PAGE_SIZE);
282                         nr_read = access_remote_vm(mm, p, page, _count, FOLL_ANON);
283                         if (nr_read < 0)
284                                 rv = nr_read;
285                         if (nr_read <= 0)
286                                 goto out_free_page;
287
288                         if (copy_to_user(buf, page, nr_read)) {
289                                 rv = -EFAULT;
290                                 goto out_free_page;
291                         }
292
293                         p       += nr_read;
294                         len     -= nr_read;
295                         buf     += nr_read;
296                         count   -= nr_read;
297                         rv      += nr_read;
298                 }
299         } else {
300                 /*
301                  * Command line (1 string) occupies ARGV and
302                  * extends into ENVP.
303                  */
304                 struct {
305                         unsigned long p;
306                         unsigned long len;
307                 } cmdline[2] = {
308                         { .p = arg_start, .len = len1 },
309                         { .p = env_start, .len = len2 },
310                 };
311                 loff_t pos1 = *pos;
312                 unsigned int i;
313
314                 i = 0;
315                 while (i < 2 && pos1 >= cmdline[i].len) {
316                         pos1 -= cmdline[i].len;
317                         i++;
318                 }
319                 while (i < 2) {
320                         p = cmdline[i].p + pos1;
321                         len = cmdline[i].len - pos1;
322                         while (count > 0 && len > 0) {
323                                 unsigned int _count, l;
324                                 int nr_read;
325                                 bool final;
326
327                                 _count = min3(count, len, PAGE_SIZE);
328                                 nr_read = access_remote_vm(mm, p, page, _count, FOLL_ANON);
329                                 if (nr_read < 0)
330                                         rv = nr_read;
331                                 if (nr_read <= 0)
332                                         goto out_free_page;
333
334                                 /*
335                                  * Command line can be shorter than whole ARGV
336                                  * even if last "marker" byte says it is not.
337                                  */
338                                 final = false;
339                                 l = strnlen(page, nr_read);
340                                 if (l < nr_read) {
341                                         nr_read = l;
342                                         final = true;
343                                 }
344
345                                 if (copy_to_user(buf, page, nr_read)) {
346                                         rv = -EFAULT;
347                                         goto out_free_page;
348                                 }
349
350                                 p       += nr_read;
351                                 len     -= nr_read;
352                                 buf     += nr_read;
353                                 count   -= nr_read;
354                                 rv      += nr_read;
355
356                                 if (final)
357                                         goto out_free_page;
358                         }
359
360                         /* Only first chunk can be read partially. */
361                         pos1 = 0;
362                         i++;
363                 }
364         }
365
366 out_free_page:
367         free_page((unsigned long)page);
368 out_mmput:
369         mmput(mm);
370         if (rv > 0)
371                 *pos += rv;
372         return rv;
373 }
374
375 static const struct file_operations proc_pid_cmdline_ops = {
376         .read   = proc_pid_cmdline_read,
377         .llseek = generic_file_llseek,
378 };
379
380 #ifdef CONFIG_KALLSYMS
381 /*
382  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
383  * Returns the resolved symbol.  If that fails, simply return the address.
384  */
385 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
386                           struct pid *pid, struct task_struct *task)
387 {
388         unsigned long wchan;
389         char symname[KSYM_NAME_LEN];
390
391         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
392                 goto print0;
393
394         wchan = get_wchan(task);
395         if (wchan && !lookup_symbol_name(wchan, symname)) {
396                 seq_puts(m, symname);
397                 return 0;
398         }
399
400 print0:
401         seq_putc(m, '0');
402         return 0;
403 }
404 #endif /* CONFIG_KALLSYMS */
405
406 static int lock_trace(struct task_struct *task)
407 {
408         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
409         if (err)
410                 return err;
411         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
412                 mutex_unlock(&task->signal->cred_guard_mutex);
413                 return -EPERM;
414         }
415         return 0;
416 }
417
418 static void unlock_trace(struct task_struct *task)
419 {
420         mutex_unlock(&task->signal->cred_guard_mutex);
421 }
422
423 #ifdef CONFIG_STACKTRACE
424
425 #define MAX_STACK_TRACE_DEPTH   64
426
427 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
428                           struct pid *pid, struct task_struct *task)
429 {
430         struct stack_trace trace;
431         unsigned long *entries;
432         int err;
433         int i;
434
435         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
436         if (!entries)
437                 return -ENOMEM;
438
439         trace.nr_entries        = 0;
440         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
441         trace.entries           = entries;
442         trace.skip              = 0;
443
444         err = lock_trace(task);
445         if (!err) {
446                 save_stack_trace_tsk(task, &trace);
447
448                 for (i = 0; i < trace.nr_entries; i++) {
449                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
450                 }
451                 unlock_trace(task);
452         }
453         kfree(entries);
454
455         return err;
456 }
457 #endif
458
459 #ifdef CONFIG_SCHED_INFO
460 /*
461  * Provides /proc/PID/schedstat
462  */
463 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
464                               struct pid *pid, struct task_struct *task)
465 {
466         if (unlikely(!sched_info_on()))
467                 seq_printf(m, "0 0 0\n");
468         else
469                 seq_printf(m, "%llu %llu %lu\n",
470                    (unsigned long long)task->se.sum_exec_runtime,
471                    (unsigned long long)task->sched_info.run_delay,
472                    task->sched_info.pcount);
473
474         return 0;
475 }
476 #endif
477
478 #ifdef CONFIG_LATENCYTOP
479 static int lstats_show_proc(struct seq_file *m, void *v)
480 {
481         int i;
482         struct inode *inode = m->private;
483         struct task_struct *task = get_proc_task(inode);
484
485         if (!task)
486                 return -ESRCH;
487         seq_puts(m, "Latency Top version : v0.1\n");
488         for (i = 0; i < 32; i++) {
489                 struct latency_record *lr = &task->latency_record[i];
490                 if (lr->backtrace[0]) {
491                         int q;
492                         seq_printf(m, "%i %li %li",
493                                    lr->count, lr->time, lr->max);
494                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
495                                 unsigned long bt = lr->backtrace[q];
496                                 if (!bt)
497                                         break;
498                                 if (bt == ULONG_MAX)
499                                         break;
500                                 seq_printf(m, " %ps", (void *)bt);
501                         }
502                         seq_putc(m, '\n');
503                 }
504
505         }
506         put_task_struct(task);
507         return 0;
508 }
509
510 static int lstats_open(struct inode *inode, struct file *file)
511 {
512         return single_open(file, lstats_show_proc, inode);
513 }
514
515 static ssize_t lstats_write(struct file *file, const char __user *buf,
516                             size_t count, loff_t *offs)
517 {
518         struct task_struct *task = get_proc_task(file_inode(file));
519
520         if (!task)
521                 return -ESRCH;
522         clear_all_latency_tracing(task);
523         put_task_struct(task);
524
525         return count;
526 }
527
528 static const struct file_operations proc_lstats_operations = {
529         .open           = lstats_open,
530         .read           = seq_read,
531         .write          = lstats_write,
532         .llseek         = seq_lseek,
533         .release        = single_release,
534 };
535
536 #endif
537
538 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
539                           struct pid *pid, struct task_struct *task)
540 {
541         unsigned long totalpages = totalram_pages + total_swap_pages;
542         unsigned long points = 0;
543
544         points = oom_badness(task, NULL, NULL, totalpages) *
545                                         1000 / totalpages;
546         seq_printf(m, "%lu\n", points);
547
548         return 0;
549 }
550
551 struct limit_names {
552         const char *name;
553         const char *unit;
554 };
555
556 static const struct limit_names lnames[RLIM_NLIMITS] = {
557         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
558         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
559         [RLIMIT_DATA] = {"Max data size", "bytes"},
560         [RLIMIT_STACK] = {"Max stack size", "bytes"},
561         [RLIMIT_CORE] = {"Max core file size", "bytes"},
562         [RLIMIT_RSS] = {"Max resident set", "bytes"},
563         [RLIMIT_NPROC] = {"Max processes", "processes"},
564         [RLIMIT_NOFILE] = {"Max open files", "files"},
565         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
566         [RLIMIT_AS] = {"Max address space", "bytes"},
567         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
568         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
569         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
570         [RLIMIT_NICE] = {"Max nice priority", NULL},
571         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
572         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
573 };
574
575 /* Display limits for a process */
576 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
577                            struct pid *pid, struct task_struct *task)
578 {
579         unsigned int i;
580         unsigned long flags;
581
582         struct rlimit rlim[RLIM_NLIMITS];
583
584         if (!lock_task_sighand(task, &flags))
585                 return 0;
586         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
587         unlock_task_sighand(task, &flags);
588
589         /*
590          * print the file header
591          */
592        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
593                   "Limit", "Soft Limit", "Hard Limit", "Units");
594
595         for (i = 0; i < RLIM_NLIMITS; i++) {
596                 if (rlim[i].rlim_cur == RLIM_INFINITY)
597                         seq_printf(m, "%-25s %-20s ",
598                                    lnames[i].name, "unlimited");
599                 else
600                         seq_printf(m, "%-25s %-20lu ",
601                                    lnames[i].name, rlim[i].rlim_cur);
602
603                 if (rlim[i].rlim_max == RLIM_INFINITY)
604                         seq_printf(m, "%-20s ", "unlimited");
605                 else
606                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
607
608                 if (lnames[i].unit)
609                         seq_printf(m, "%-10s\n", lnames[i].unit);
610                 else
611                         seq_putc(m, '\n');
612         }
613
614         return 0;
615 }
616
617 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
618 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
619                             struct pid *pid, struct task_struct *task)
620 {
621         long nr;
622         unsigned long args[6], sp, pc;
623         int res;
624
625         res = lock_trace(task);
626         if (res)
627                 return res;
628
629         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
630                 seq_puts(m, "running\n");
631         else if (nr < 0)
632                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
633         else
634                 seq_printf(m,
635                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
636                        nr,
637                        args[0], args[1], args[2], args[3], args[4], args[5],
638                        sp, pc);
639         unlock_trace(task);
640
641         return 0;
642 }
643 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
644
645 /************************************************************************/
646 /*                       Here the fs part begins                        */
647 /************************************************************************/
648
649 /* permission checks */
650 static int proc_fd_access_allowed(struct inode *inode)
651 {
652         struct task_struct *task;
653         int allowed = 0;
654         /* Allow access to a task's file descriptors if it is us or we
655          * may use ptrace attach to the process and find out that
656          * information.
657          */
658         task = get_proc_task(inode);
659         if (task) {
660                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
661                 put_task_struct(task);
662         }
663         return allowed;
664 }
665
666 int proc_setattr(struct dentry *dentry, struct iattr *attr)
667 {
668         int error;
669         struct inode *inode = d_inode(dentry);
670
671         if (attr->ia_valid & ATTR_MODE)
672                 return -EPERM;
673
674         error = setattr_prepare(dentry, attr);
675         if (error)
676                 return error;
677
678         setattr_copy(inode, attr);
679         mark_inode_dirty(inode);
680         return 0;
681 }
682
683 /*
684  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
685  * or euid/egid (for hide_pid_min=2)?
686  */
687 static bool has_pid_permissions(struct pid_namespace *pid,
688                                  struct task_struct *task,
689                                  int hide_pid_min)
690 {
691         if (pid->hide_pid < hide_pid_min)
692                 return true;
693         if (in_group_p(pid->pid_gid))
694                 return true;
695         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
696 }
697
698
699 static int proc_pid_permission(struct inode *inode, int mask)
700 {
701         struct pid_namespace *pid = proc_pid_ns(inode);
702         struct task_struct *task;
703         bool has_perms;
704
705         task = get_proc_task(inode);
706         if (!task)
707                 return -ESRCH;
708         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
709         put_task_struct(task);
710
711         if (!has_perms) {
712                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
713                         /*
714                          * Let's make getdents(), stat(), and open()
715                          * consistent with each other.  If a process
716                          * may not stat() a file, it shouldn't be seen
717                          * in procfs at all.
718                          */
719                         return -ENOENT;
720                 }
721
722                 return -EPERM;
723         }
724         return generic_permission(inode, mask);
725 }
726
727
728
729 static const struct inode_operations proc_def_inode_operations = {
730         .setattr        = proc_setattr,
731 };
732
733 static int proc_single_show(struct seq_file *m, void *v)
734 {
735         struct inode *inode = m->private;
736         struct pid_namespace *ns = proc_pid_ns(inode);
737         struct pid *pid = proc_pid(inode);
738         struct task_struct *task;
739         int ret;
740
741         task = get_pid_task(pid, PIDTYPE_PID);
742         if (!task)
743                 return -ESRCH;
744
745         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
746
747         put_task_struct(task);
748         return ret;
749 }
750
751 static int proc_single_open(struct inode *inode, struct file *filp)
752 {
753         return single_open(filp, proc_single_show, inode);
754 }
755
756 static const struct file_operations proc_single_file_operations = {
757         .open           = proc_single_open,
758         .read           = seq_read,
759         .llseek         = seq_lseek,
760         .release        = single_release,
761 };
762
763
764 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
765 {
766         struct task_struct *task = get_proc_task(inode);
767         struct mm_struct *mm = ERR_PTR(-ESRCH);
768
769         if (task) {
770                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
771                 put_task_struct(task);
772
773                 if (!IS_ERR_OR_NULL(mm)) {
774                         /* ensure this mm_struct can't be freed */
775                         mmgrab(mm);
776                         /* but do not pin its memory */
777                         mmput(mm);
778                 }
779         }
780
781         return mm;
782 }
783
784 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
785 {
786         struct mm_struct *mm = proc_mem_open(inode, mode);
787
788         if (IS_ERR(mm))
789                 return PTR_ERR(mm);
790
791         file->private_data = mm;
792         return 0;
793 }
794
795 static int mem_open(struct inode *inode, struct file *file)
796 {
797         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
798
799         /* OK to pass negative loff_t, we can catch out-of-range */
800         file->f_mode |= FMODE_UNSIGNED_OFFSET;
801
802         return ret;
803 }
804
805 static ssize_t mem_rw(struct file *file, char __user *buf,
806                         size_t count, loff_t *ppos, int write)
807 {
808         struct mm_struct *mm = file->private_data;
809         unsigned long addr = *ppos;
810         ssize_t copied;
811         char *page;
812         unsigned int flags;
813
814         if (!mm)
815                 return 0;
816
817         page = (char *)__get_free_page(GFP_KERNEL);
818         if (!page)
819                 return -ENOMEM;
820
821         copied = 0;
822         if (!mmget_not_zero(mm))
823                 goto free;
824
825         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
826
827         while (count > 0) {
828                 int this_len = min_t(int, count, PAGE_SIZE);
829
830                 if (write && copy_from_user(page, buf, this_len)) {
831                         copied = -EFAULT;
832                         break;
833                 }
834
835                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
836                 if (!this_len) {
837                         if (!copied)
838                                 copied = -EIO;
839                         break;
840                 }
841
842                 if (!write && copy_to_user(buf, page, this_len)) {
843                         copied = -EFAULT;
844                         break;
845                 }
846
847                 buf += this_len;
848                 addr += this_len;
849                 copied += this_len;
850                 count -= this_len;
851         }
852         *ppos = addr;
853
854         mmput(mm);
855 free:
856         free_page((unsigned long) page);
857         return copied;
858 }
859
860 static ssize_t mem_read(struct file *file, char __user *buf,
861                         size_t count, loff_t *ppos)
862 {
863         return mem_rw(file, buf, count, ppos, 0);
864 }
865
866 static ssize_t mem_write(struct file *file, const char __user *buf,
867                          size_t count, loff_t *ppos)
868 {
869         return mem_rw(file, (char __user*)buf, count, ppos, 1);
870 }
871
872 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
873 {
874         switch (orig) {
875         case 0:
876                 file->f_pos = offset;
877                 break;
878         case 1:
879                 file->f_pos += offset;
880                 break;
881         default:
882                 return -EINVAL;
883         }
884         force_successful_syscall_return();
885         return file->f_pos;
886 }
887
888 static int mem_release(struct inode *inode, struct file *file)
889 {
890         struct mm_struct *mm = file->private_data;
891         if (mm)
892                 mmdrop(mm);
893         return 0;
894 }
895
896 static const struct file_operations proc_mem_operations = {
897         .llseek         = mem_lseek,
898         .read           = mem_read,
899         .write          = mem_write,
900         .open           = mem_open,
901         .release        = mem_release,
902 };
903
904 static int environ_open(struct inode *inode, struct file *file)
905 {
906         return __mem_open(inode, file, PTRACE_MODE_READ);
907 }
908
909 static ssize_t environ_read(struct file *file, char __user *buf,
910                         size_t count, loff_t *ppos)
911 {
912         char *page;
913         unsigned long src = *ppos;
914         int ret = 0;
915         struct mm_struct *mm = file->private_data;
916         unsigned long env_start, env_end;
917
918         /* Ensure the process spawned far enough to have an environment. */
919         if (!mm || !mm->env_end)
920                 return 0;
921
922         page = (char *)__get_free_page(GFP_KERNEL);
923         if (!page)
924                 return -ENOMEM;
925
926         ret = 0;
927         if (!mmget_not_zero(mm))
928                 goto free;
929
930         down_read(&mm->mmap_sem);
931         env_start = mm->env_start;
932         env_end = mm->env_end;
933         up_read(&mm->mmap_sem);
934
935         while (count > 0) {
936                 size_t this_len, max_len;
937                 int retval;
938
939                 if (src >= (env_end - env_start))
940                         break;
941
942                 this_len = env_end - (env_start + src);
943
944                 max_len = min_t(size_t, PAGE_SIZE, count);
945                 this_len = min(max_len, this_len);
946
947                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
948
949                 if (retval <= 0) {
950                         ret = retval;
951                         break;
952                 }
953
954                 if (copy_to_user(buf, page, retval)) {
955                         ret = -EFAULT;
956                         break;
957                 }
958
959                 ret += retval;
960                 src += retval;
961                 buf += retval;
962                 count -= retval;
963         }
964         *ppos = src;
965         mmput(mm);
966
967 free:
968         free_page((unsigned long) page);
969         return ret;
970 }
971
972 static const struct file_operations proc_environ_operations = {
973         .open           = environ_open,
974         .read           = environ_read,
975         .llseek         = generic_file_llseek,
976         .release        = mem_release,
977 };
978
979 static int auxv_open(struct inode *inode, struct file *file)
980 {
981         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
982 }
983
984 static ssize_t auxv_read(struct file *file, char __user *buf,
985                         size_t count, loff_t *ppos)
986 {
987         struct mm_struct *mm = file->private_data;
988         unsigned int nwords = 0;
989
990         if (!mm)
991                 return 0;
992         do {
993                 nwords += 2;
994         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
995         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
996                                        nwords * sizeof(mm->saved_auxv[0]));
997 }
998
999 static const struct file_operations proc_auxv_operations = {
1000         .open           = auxv_open,
1001         .read           = auxv_read,
1002         .llseek         = generic_file_llseek,
1003         .release        = mem_release,
1004 };
1005
1006 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1007                             loff_t *ppos)
1008 {
1009         struct task_struct *task = get_proc_task(file_inode(file));
1010         char buffer[PROC_NUMBUF];
1011         int oom_adj = OOM_ADJUST_MIN;
1012         size_t len;
1013
1014         if (!task)
1015                 return -ESRCH;
1016         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1017                 oom_adj = OOM_ADJUST_MAX;
1018         else
1019                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1020                           OOM_SCORE_ADJ_MAX;
1021         put_task_struct(task);
1022         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1023         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1024 }
1025
1026 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1027 {
1028         static DEFINE_MUTEX(oom_adj_mutex);
1029         struct mm_struct *mm = NULL;
1030         struct task_struct *task;
1031         int err = 0;
1032
1033         task = get_proc_task(file_inode(file));
1034         if (!task)
1035                 return -ESRCH;
1036
1037         mutex_lock(&oom_adj_mutex);
1038         if (legacy) {
1039                 if (oom_adj < task->signal->oom_score_adj &&
1040                                 !capable(CAP_SYS_RESOURCE)) {
1041                         err = -EACCES;
1042                         goto err_unlock;
1043                 }
1044                 /*
1045                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1046                  * /proc/pid/oom_score_adj instead.
1047                  */
1048                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1049                           current->comm, task_pid_nr(current), task_pid_nr(task),
1050                           task_pid_nr(task));
1051         } else {
1052                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1053                                 !capable(CAP_SYS_RESOURCE)) {
1054                         err = -EACCES;
1055                         goto err_unlock;
1056                 }
1057         }
1058
1059         /*
1060          * Make sure we will check other processes sharing the mm if this is
1061          * not vfrok which wants its own oom_score_adj.
1062          * pin the mm so it doesn't go away and get reused after task_unlock
1063          */
1064         if (!task->vfork_done) {
1065                 struct task_struct *p = find_lock_task_mm(task);
1066
1067                 if (p) {
1068                         if (atomic_read(&p->mm->mm_users) > 1) {
1069                                 mm = p->mm;
1070                                 mmgrab(mm);
1071                         }
1072                         task_unlock(p);
1073                 }
1074         }
1075
1076         task->signal->oom_score_adj = oom_adj;
1077         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1078                 task->signal->oom_score_adj_min = (short)oom_adj;
1079         trace_oom_score_adj_update(task);
1080
1081         if (mm) {
1082                 struct task_struct *p;
1083
1084                 rcu_read_lock();
1085                 for_each_process(p) {
1086                         if (same_thread_group(task, p))
1087                                 continue;
1088
1089                         /* do not touch kernel threads or the global init */
1090                         if (p->flags & PF_KTHREAD || is_global_init(p))
1091                                 continue;
1092
1093                         task_lock(p);
1094                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1095                                 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",
1096                                                 task_pid_nr(p), p->comm,
1097                                                 p->signal->oom_score_adj, oom_adj,
1098                                                 task_pid_nr(task), task->comm);
1099                                 p->signal->oom_score_adj = oom_adj;
1100                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1101                                         p->signal->oom_score_adj_min = (short)oom_adj;
1102                         }
1103                         task_unlock(p);
1104                 }
1105                 rcu_read_unlock();
1106                 mmdrop(mm);
1107         }
1108 err_unlock:
1109         mutex_unlock(&oom_adj_mutex);
1110         put_task_struct(task);
1111         return err;
1112 }
1113
1114 /*
1115  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1116  * kernels.  The effective policy is defined by oom_score_adj, which has a
1117  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1118  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1119  * Processes that become oom disabled via oom_adj will still be oom disabled
1120  * with this implementation.
1121  *
1122  * oom_adj cannot be removed since existing userspace binaries use it.
1123  */
1124 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1125                              size_t count, loff_t *ppos)
1126 {
1127         char buffer[PROC_NUMBUF];
1128         int oom_adj;
1129         int err;
1130
1131         memset(buffer, 0, sizeof(buffer));
1132         if (count > sizeof(buffer) - 1)
1133                 count = sizeof(buffer) - 1;
1134         if (copy_from_user(buffer, buf, count)) {
1135                 err = -EFAULT;
1136                 goto out;
1137         }
1138
1139         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1140         if (err)
1141                 goto out;
1142         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1143              oom_adj != OOM_DISABLE) {
1144                 err = -EINVAL;
1145                 goto out;
1146         }
1147
1148         /*
1149          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1150          * value is always attainable.
1151          */
1152         if (oom_adj == OOM_ADJUST_MAX)
1153                 oom_adj = OOM_SCORE_ADJ_MAX;
1154         else
1155                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1156
1157         err = __set_oom_adj(file, oom_adj, true);
1158 out:
1159         return err < 0 ? err : count;
1160 }
1161
1162 static const struct file_operations proc_oom_adj_operations = {
1163         .read           = oom_adj_read,
1164         .write          = oom_adj_write,
1165         .llseek         = generic_file_llseek,
1166 };
1167
1168 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1169                                         size_t count, loff_t *ppos)
1170 {
1171         struct task_struct *task = get_proc_task(file_inode(file));
1172         char buffer[PROC_NUMBUF];
1173         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1174         size_t len;
1175
1176         if (!task)
1177                 return -ESRCH;
1178         oom_score_adj = task->signal->oom_score_adj;
1179         put_task_struct(task);
1180         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1181         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1182 }
1183
1184 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1185                                         size_t count, loff_t *ppos)
1186 {
1187         char buffer[PROC_NUMBUF];
1188         int oom_score_adj;
1189         int err;
1190
1191         memset(buffer, 0, sizeof(buffer));
1192         if (count > sizeof(buffer) - 1)
1193                 count = sizeof(buffer) - 1;
1194         if (copy_from_user(buffer, buf, count)) {
1195                 err = -EFAULT;
1196                 goto out;
1197         }
1198
1199         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1200         if (err)
1201                 goto out;
1202         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1203                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1204                 err = -EINVAL;
1205                 goto out;
1206         }
1207
1208         err = __set_oom_adj(file, oom_score_adj, false);
1209 out:
1210         return err < 0 ? err : count;
1211 }
1212
1213 static const struct file_operations proc_oom_score_adj_operations = {
1214         .read           = oom_score_adj_read,
1215         .write          = oom_score_adj_write,
1216         .llseek         = default_llseek,
1217 };
1218
1219 #ifdef CONFIG_AUDITSYSCALL
1220 #define TMPBUFLEN 11
1221 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1222                                   size_t count, loff_t *ppos)
1223 {
1224         struct inode * inode = file_inode(file);
1225         struct task_struct *task = get_proc_task(inode);
1226         ssize_t length;
1227         char tmpbuf[TMPBUFLEN];
1228
1229         if (!task)
1230                 return -ESRCH;
1231         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1232                            from_kuid(file->f_cred->user_ns,
1233                                      audit_get_loginuid(task)));
1234         put_task_struct(task);
1235         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1236 }
1237
1238 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1239                                    size_t count, loff_t *ppos)
1240 {
1241         struct inode * inode = file_inode(file);
1242         uid_t loginuid;
1243         kuid_t kloginuid;
1244         int rv;
1245
1246         rcu_read_lock();
1247         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1248                 rcu_read_unlock();
1249                 return -EPERM;
1250         }
1251         rcu_read_unlock();
1252
1253         if (*ppos != 0) {
1254                 /* No partial writes. */
1255                 return -EINVAL;
1256         }
1257
1258         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1259         if (rv < 0)
1260                 return rv;
1261
1262         /* is userspace tring to explicitly UNSET the loginuid? */
1263         if (loginuid == AUDIT_UID_UNSET) {
1264                 kloginuid = INVALID_UID;
1265         } else {
1266                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1267                 if (!uid_valid(kloginuid))
1268                         return -EINVAL;
1269         }
1270
1271         rv = audit_set_loginuid(kloginuid);
1272         if (rv < 0)
1273                 return rv;
1274         return count;
1275 }
1276
1277 static const struct file_operations proc_loginuid_operations = {
1278         .read           = proc_loginuid_read,
1279         .write          = proc_loginuid_write,
1280         .llseek         = generic_file_llseek,
1281 };
1282
1283 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1284                                   size_t count, loff_t *ppos)
1285 {
1286         struct inode * inode = file_inode(file);
1287         struct task_struct *task = get_proc_task(inode);
1288         ssize_t length;
1289         char tmpbuf[TMPBUFLEN];
1290
1291         if (!task)
1292                 return -ESRCH;
1293         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1294                                 audit_get_sessionid(task));
1295         put_task_struct(task);
1296         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1297 }
1298
1299 static const struct file_operations proc_sessionid_operations = {
1300         .read           = proc_sessionid_read,
1301         .llseek         = generic_file_llseek,
1302 };
1303 #endif
1304
1305 #ifdef CONFIG_FAULT_INJECTION
1306 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1307                                       size_t count, loff_t *ppos)
1308 {
1309         struct task_struct *task = get_proc_task(file_inode(file));
1310         char buffer[PROC_NUMBUF];
1311         size_t len;
1312         int make_it_fail;
1313
1314         if (!task)
1315                 return -ESRCH;
1316         make_it_fail = task->make_it_fail;
1317         put_task_struct(task);
1318
1319         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1320
1321         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1322 }
1323
1324 static ssize_t proc_fault_inject_write(struct file * file,
1325                         const char __user * buf, size_t count, loff_t *ppos)
1326 {
1327         struct task_struct *task;
1328         char buffer[PROC_NUMBUF];
1329         int make_it_fail;
1330         int rv;
1331
1332         if (!capable(CAP_SYS_RESOURCE))
1333                 return -EPERM;
1334         memset(buffer, 0, sizeof(buffer));
1335         if (count > sizeof(buffer) - 1)
1336                 count = sizeof(buffer) - 1;
1337         if (copy_from_user(buffer, buf, count))
1338                 return -EFAULT;
1339         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1340         if (rv < 0)
1341                 return rv;
1342         if (make_it_fail < 0 || make_it_fail > 1)
1343                 return -EINVAL;
1344
1345         task = get_proc_task(file_inode(file));
1346         if (!task)
1347                 return -ESRCH;
1348         task->make_it_fail = make_it_fail;
1349         put_task_struct(task);
1350
1351         return count;
1352 }
1353
1354 static const struct file_operations proc_fault_inject_operations = {
1355         .read           = proc_fault_inject_read,
1356         .write          = proc_fault_inject_write,
1357         .llseek         = generic_file_llseek,
1358 };
1359
1360 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1361                                    size_t count, loff_t *ppos)
1362 {
1363         struct task_struct *task;
1364         int err;
1365         unsigned int n;
1366
1367         err = kstrtouint_from_user(buf, count, 0, &n);
1368         if (err)
1369                 return err;
1370
1371         task = get_proc_task(file_inode(file));
1372         if (!task)
1373                 return -ESRCH;
1374         task->fail_nth = n;
1375         put_task_struct(task);
1376
1377         return count;
1378 }
1379
1380 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1381                                   size_t count, loff_t *ppos)
1382 {
1383         struct task_struct *task;
1384         char numbuf[PROC_NUMBUF];
1385         ssize_t len;
1386
1387         task = get_proc_task(file_inode(file));
1388         if (!task)
1389                 return -ESRCH;
1390         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1391         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1392         put_task_struct(task);
1393
1394         return len;
1395 }
1396
1397 static const struct file_operations proc_fail_nth_operations = {
1398         .read           = proc_fail_nth_read,
1399         .write          = proc_fail_nth_write,
1400 };
1401 #endif
1402
1403
1404 #ifdef CONFIG_SCHED_DEBUG
1405 /*
1406  * Print out various scheduling related per-task fields:
1407  */
1408 static int sched_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct pid_namespace *ns = proc_pid_ns(inode);
1412         struct task_struct *p;
1413
1414         p = get_proc_task(inode);
1415         if (!p)
1416                 return -ESRCH;
1417         proc_sched_show_task(p, ns, m);
1418
1419         put_task_struct(p);
1420
1421         return 0;
1422 }
1423
1424 static ssize_t
1425 sched_write(struct file *file, const char __user *buf,
1426             size_t count, loff_t *offset)
1427 {
1428         struct inode *inode = file_inode(file);
1429         struct task_struct *p;
1430
1431         p = get_proc_task(inode);
1432         if (!p)
1433                 return -ESRCH;
1434         proc_sched_set_task(p);
1435
1436         put_task_struct(p);
1437
1438         return count;
1439 }
1440
1441 static int sched_open(struct inode *inode, struct file *filp)
1442 {
1443         return single_open(filp, sched_show, inode);
1444 }
1445
1446 static const struct file_operations proc_pid_sched_operations = {
1447         .open           = sched_open,
1448         .read           = seq_read,
1449         .write          = sched_write,
1450         .llseek         = seq_lseek,
1451         .release        = single_release,
1452 };
1453
1454 #endif
1455
1456 #ifdef CONFIG_SCHED_AUTOGROUP
1457 /*
1458  * Print out autogroup related information:
1459  */
1460 static int sched_autogroup_show(struct seq_file *m, void *v)
1461 {
1462         struct inode *inode = m->private;
1463         struct task_struct *p;
1464
1465         p = get_proc_task(inode);
1466         if (!p)
1467                 return -ESRCH;
1468         proc_sched_autogroup_show_task(p, m);
1469
1470         put_task_struct(p);
1471
1472         return 0;
1473 }
1474
1475 static ssize_t
1476 sched_autogroup_write(struct file *file, const char __user *buf,
1477             size_t count, loff_t *offset)
1478 {
1479         struct inode *inode = file_inode(file);
1480         struct task_struct *p;
1481         char buffer[PROC_NUMBUF];
1482         int nice;
1483         int err;
1484
1485         memset(buffer, 0, sizeof(buffer));
1486         if (count > sizeof(buffer) - 1)
1487                 count = sizeof(buffer) - 1;
1488         if (copy_from_user(buffer, buf, count))
1489                 return -EFAULT;
1490
1491         err = kstrtoint(strstrip(buffer), 0, &nice);
1492         if (err < 0)
1493                 return err;
1494
1495         p = get_proc_task(inode);
1496         if (!p)
1497                 return -ESRCH;
1498
1499         err = proc_sched_autogroup_set_nice(p, nice);
1500         if (err)
1501                 count = err;
1502
1503         put_task_struct(p);
1504
1505         return count;
1506 }
1507
1508 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1509 {
1510         int ret;
1511
1512         ret = single_open(filp, sched_autogroup_show, NULL);
1513         if (!ret) {
1514                 struct seq_file *m = filp->private_data;
1515
1516                 m->private = inode;
1517         }
1518         return ret;
1519 }
1520
1521 static const struct file_operations proc_pid_sched_autogroup_operations = {
1522         .open           = sched_autogroup_open,
1523         .read           = seq_read,
1524         .write          = sched_autogroup_write,
1525         .llseek         = seq_lseek,
1526         .release        = single_release,
1527 };
1528
1529 #endif /* CONFIG_SCHED_AUTOGROUP */
1530
1531 static ssize_t comm_write(struct file *file, const char __user *buf,
1532                                 size_t count, loff_t *offset)
1533 {
1534         struct inode *inode = file_inode(file);
1535         struct task_struct *p;
1536         char buffer[TASK_COMM_LEN];
1537         const size_t maxlen = sizeof(buffer) - 1;
1538
1539         memset(buffer, 0, sizeof(buffer));
1540         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1541                 return -EFAULT;
1542
1543         p = get_proc_task(inode);
1544         if (!p)
1545                 return -ESRCH;
1546
1547         if (same_thread_group(current, p))
1548                 set_task_comm(p, buffer);
1549         else
1550                 count = -EINVAL;
1551
1552         put_task_struct(p);
1553
1554         return count;
1555 }
1556
1557 static int comm_show(struct seq_file *m, void *v)
1558 {
1559         struct inode *inode = m->private;
1560         struct task_struct *p;
1561
1562         p = get_proc_task(inode);
1563         if (!p)
1564                 return -ESRCH;
1565
1566         proc_task_name(m, p, false);
1567         seq_putc(m, '\n');
1568
1569         put_task_struct(p);
1570
1571         return 0;
1572 }
1573
1574 static int comm_open(struct inode *inode, struct file *filp)
1575 {
1576         return single_open(filp, comm_show, inode);
1577 }
1578
1579 static const struct file_operations proc_pid_set_comm_operations = {
1580         .open           = comm_open,
1581         .read           = seq_read,
1582         .write          = comm_write,
1583         .llseek         = seq_lseek,
1584         .release        = single_release,
1585 };
1586
1587 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1588 {
1589         struct task_struct *task;
1590         struct file *exe_file;
1591
1592         task = get_proc_task(d_inode(dentry));
1593         if (!task)
1594                 return -ENOENT;
1595         exe_file = get_task_exe_file(task);
1596         put_task_struct(task);
1597         if (exe_file) {
1598                 *exe_path = exe_file->f_path;
1599                 path_get(&exe_file->f_path);
1600                 fput(exe_file);
1601                 return 0;
1602         } else
1603                 return -ENOENT;
1604 }
1605
1606 static const char *proc_pid_get_link(struct dentry *dentry,
1607                                      struct inode *inode,
1608                                      struct delayed_call *done)
1609 {
1610         struct path path;
1611         int error = -EACCES;
1612
1613         if (!dentry)
1614                 return ERR_PTR(-ECHILD);
1615
1616         /* Are we allowed to snoop on the tasks file descriptors? */
1617         if (!proc_fd_access_allowed(inode))
1618                 goto out;
1619
1620         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1621         if (error)
1622                 goto out;
1623
1624         nd_jump_link(&path);
1625         return NULL;
1626 out:
1627         return ERR_PTR(error);
1628 }
1629
1630 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1631 {
1632         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1633         char *pathname;
1634         int len;
1635
1636         if (!tmp)
1637                 return -ENOMEM;
1638
1639         pathname = d_path(path, tmp, PAGE_SIZE);
1640         len = PTR_ERR(pathname);
1641         if (IS_ERR(pathname))
1642                 goto out;
1643         len = tmp + PAGE_SIZE - 1 - pathname;
1644
1645         if (len > buflen)
1646                 len = buflen;
1647         if (copy_to_user(buffer, pathname, len))
1648                 len = -EFAULT;
1649  out:
1650         free_page((unsigned long)tmp);
1651         return len;
1652 }
1653
1654 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1655 {
1656         int error = -EACCES;
1657         struct inode *inode = d_inode(dentry);
1658         struct path path;
1659
1660         /* Are we allowed to snoop on the tasks file descriptors? */
1661         if (!proc_fd_access_allowed(inode))
1662                 goto out;
1663
1664         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1665         if (error)
1666                 goto out;
1667
1668         error = do_proc_readlink(&path, buffer, buflen);
1669         path_put(&path);
1670 out:
1671         return error;
1672 }
1673
1674 const struct inode_operations proc_pid_link_inode_operations = {
1675         .readlink       = proc_pid_readlink,
1676         .get_link       = proc_pid_get_link,
1677         .setattr        = proc_setattr,
1678 };
1679
1680
1681 /* building an inode */
1682
1683 void task_dump_owner(struct task_struct *task, umode_t mode,
1684                      kuid_t *ruid, kgid_t *rgid)
1685 {
1686         /* Depending on the state of dumpable compute who should own a
1687          * proc file for a task.
1688          */
1689         const struct cred *cred;
1690         kuid_t uid;
1691         kgid_t gid;
1692
1693         if (unlikely(task->flags & PF_KTHREAD)) {
1694                 *ruid = GLOBAL_ROOT_UID;
1695                 *rgid = GLOBAL_ROOT_GID;
1696                 return;
1697         }
1698
1699         /* Default to the tasks effective ownership */
1700         rcu_read_lock();
1701         cred = __task_cred(task);
1702         uid = cred->euid;
1703         gid = cred->egid;
1704         rcu_read_unlock();
1705
1706         /*
1707          * Before the /proc/pid/status file was created the only way to read
1708          * the effective uid of a /process was to stat /proc/pid.  Reading
1709          * /proc/pid/status is slow enough that procps and other packages
1710          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1711          * made this apply to all per process world readable and executable
1712          * directories.
1713          */
1714         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1715                 struct mm_struct *mm;
1716                 task_lock(task);
1717                 mm = task->mm;
1718                 /* Make non-dumpable tasks owned by some root */
1719                 if (mm) {
1720                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1721                                 struct user_namespace *user_ns = mm->user_ns;
1722
1723                                 uid = make_kuid(user_ns, 0);
1724                                 if (!uid_valid(uid))
1725                                         uid = GLOBAL_ROOT_UID;
1726
1727                                 gid = make_kgid(user_ns, 0);
1728                                 if (!gid_valid(gid))
1729                                         gid = GLOBAL_ROOT_GID;
1730                         }
1731                 } else {
1732                         uid = GLOBAL_ROOT_UID;
1733                         gid = GLOBAL_ROOT_GID;
1734                 }
1735                 task_unlock(task);
1736         }
1737         *ruid = uid;
1738         *rgid = gid;
1739 }
1740
1741 struct inode *proc_pid_make_inode(struct super_block * sb,
1742                                   struct task_struct *task, umode_t mode)
1743 {
1744         struct inode * inode;
1745         struct proc_inode *ei;
1746
1747         /* We need a new inode */
1748
1749         inode = new_inode(sb);
1750         if (!inode)
1751                 goto out;
1752
1753         /* Common stuff */
1754         ei = PROC_I(inode);
1755         inode->i_mode = mode;
1756         inode->i_ino = get_next_ino();
1757         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1758         inode->i_op = &proc_def_inode_operations;
1759
1760         /*
1761          * grab the reference to task.
1762          */
1763         ei->pid = get_task_pid(task, PIDTYPE_PID);
1764         if (!ei->pid)
1765                 goto out_unlock;
1766
1767         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1768         security_task_to_inode(task, inode);
1769
1770 out:
1771         return inode;
1772
1773 out_unlock:
1774         iput(inode);
1775         return NULL;
1776 }
1777
1778 int pid_getattr(const struct path *path, struct kstat *stat,
1779                 u32 request_mask, unsigned int query_flags)
1780 {
1781         struct inode *inode = d_inode(path->dentry);
1782         struct pid_namespace *pid = proc_pid_ns(inode);
1783         struct task_struct *task;
1784
1785         generic_fillattr(inode, stat);
1786
1787         rcu_read_lock();
1788         stat->uid = GLOBAL_ROOT_UID;
1789         stat->gid = GLOBAL_ROOT_GID;
1790         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1791         if (task) {
1792                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1793                         rcu_read_unlock();
1794                         /*
1795                          * This doesn't prevent learning whether PID exists,
1796                          * it only makes getattr() consistent with readdir().
1797                          */
1798                         return -ENOENT;
1799                 }
1800                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1801         }
1802         rcu_read_unlock();
1803         return 0;
1804 }
1805
1806 /* dentry stuff */
1807
1808 /*
1809  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1810  */
1811 void pid_update_inode(struct task_struct *task, struct inode *inode)
1812 {
1813         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1814
1815         inode->i_mode &= ~(S_ISUID | S_ISGID);
1816         security_task_to_inode(task, inode);
1817 }
1818
1819 /*
1820  * Rewrite the inode's ownerships here because the owning task may have
1821  * performed a setuid(), etc.
1822  *
1823  */
1824 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1825 {
1826         struct inode *inode;
1827         struct task_struct *task;
1828
1829         if (flags & LOOKUP_RCU)
1830                 return -ECHILD;
1831
1832         inode = d_inode(dentry);
1833         task = get_proc_task(inode);
1834
1835         if (task) {
1836                 pid_update_inode(task, inode);
1837                 put_task_struct(task);
1838                 return 1;
1839         }
1840         return 0;
1841 }
1842
1843 static inline bool proc_inode_is_dead(struct inode *inode)
1844 {
1845         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1846 }
1847
1848 int pid_delete_dentry(const struct dentry *dentry)
1849 {
1850         /* Is the task we represent dead?
1851          * If so, then don't put the dentry on the lru list,
1852          * kill it immediately.
1853          */
1854         return proc_inode_is_dead(d_inode(dentry));
1855 }
1856
1857 const struct dentry_operations pid_dentry_operations =
1858 {
1859         .d_revalidate   = pid_revalidate,
1860         .d_delete       = pid_delete_dentry,
1861 };
1862
1863 /* Lookups */
1864
1865 /*
1866  * Fill a directory entry.
1867  *
1868  * If possible create the dcache entry and derive our inode number and
1869  * file type from dcache entry.
1870  *
1871  * Since all of the proc inode numbers are dynamically generated, the inode
1872  * numbers do not exist until the inode is cache.  This means creating the
1873  * the dcache entry in readdir is necessary to keep the inode numbers
1874  * reported by readdir in sync with the inode numbers reported
1875  * by stat.
1876  */
1877 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1878         const char *name, int len,
1879         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1880 {
1881         struct dentry *child, *dir = file->f_path.dentry;
1882         struct qstr qname = QSTR_INIT(name, len);
1883         struct inode *inode;
1884         unsigned type = DT_UNKNOWN;
1885         ino_t ino = 1;
1886
1887         child = d_hash_and_lookup(dir, &qname);
1888         if (!child) {
1889                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1890                 child = d_alloc_parallel(dir, &qname, &wq);
1891                 if (IS_ERR(child))
1892                         goto end_instantiate;
1893                 if (d_in_lookup(child)) {
1894                         struct dentry *res;
1895                         res = instantiate(child, task, ptr);
1896                         d_lookup_done(child);
1897                         if (IS_ERR(res))
1898                                 goto end_instantiate;
1899                         if (unlikely(res)) {
1900                                 dput(child);
1901                                 child = res;
1902                         }
1903                 }
1904         }
1905         inode = d_inode(child);
1906         ino = inode->i_ino;
1907         type = inode->i_mode >> 12;
1908 end_instantiate:
1909         dput(child);
1910         return dir_emit(ctx, name, len, ino, type);
1911 }
1912
1913 /*
1914  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1915  * which represent vma start and end addresses.
1916  */
1917 static int dname_to_vma_addr(struct dentry *dentry,
1918                              unsigned long *start, unsigned long *end)
1919 {
1920         const char *str = dentry->d_name.name;
1921         unsigned long long sval, eval;
1922         unsigned int len;
1923
1924         if (str[0] == '0' && str[1] != '-')
1925                 return -EINVAL;
1926         len = _parse_integer(str, 16, &sval);
1927         if (len & KSTRTOX_OVERFLOW)
1928                 return -EINVAL;
1929         if (sval != (unsigned long)sval)
1930                 return -EINVAL;
1931         str += len;
1932
1933         if (*str != '-')
1934                 return -EINVAL;
1935         str++;
1936
1937         if (str[0] == '0' && str[1])
1938                 return -EINVAL;
1939         len = _parse_integer(str, 16, &eval);
1940         if (len & KSTRTOX_OVERFLOW)
1941                 return -EINVAL;
1942         if (eval != (unsigned long)eval)
1943                 return -EINVAL;
1944         str += len;
1945
1946         if (*str != '\0')
1947                 return -EINVAL;
1948
1949         *start = sval;
1950         *end = eval;
1951
1952         return 0;
1953 }
1954
1955 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1956 {
1957         unsigned long vm_start, vm_end;
1958         bool exact_vma_exists = false;
1959         struct mm_struct *mm = NULL;
1960         struct task_struct *task;
1961         struct inode *inode;
1962         int status = 0;
1963
1964         if (flags & LOOKUP_RCU)
1965                 return -ECHILD;
1966
1967         inode = d_inode(dentry);
1968         task = get_proc_task(inode);
1969         if (!task)
1970                 goto out_notask;
1971
1972         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1973         if (IS_ERR_OR_NULL(mm))
1974                 goto out;
1975
1976         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1977                 down_read(&mm->mmap_sem);
1978                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1979                 up_read(&mm->mmap_sem);
1980         }
1981
1982         mmput(mm);
1983
1984         if (exact_vma_exists) {
1985                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1986
1987                 security_task_to_inode(task, inode);
1988                 status = 1;
1989         }
1990
1991 out:
1992         put_task_struct(task);
1993
1994 out_notask:
1995         return status;
1996 }
1997
1998 static const struct dentry_operations tid_map_files_dentry_operations = {
1999         .d_revalidate   = map_files_d_revalidate,
2000         .d_delete       = pid_delete_dentry,
2001 };
2002
2003 static int map_files_get_link(struct dentry *dentry, struct path *path)
2004 {
2005         unsigned long vm_start, vm_end;
2006         struct vm_area_struct *vma;
2007         struct task_struct *task;
2008         struct mm_struct *mm;
2009         int rc;
2010
2011         rc = -ENOENT;
2012         task = get_proc_task(d_inode(dentry));
2013         if (!task)
2014                 goto out;
2015
2016         mm = get_task_mm(task);
2017         put_task_struct(task);
2018         if (!mm)
2019                 goto out;
2020
2021         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2022         if (rc)
2023                 goto out_mmput;
2024
2025         rc = -ENOENT;
2026         down_read(&mm->mmap_sem);
2027         vma = find_exact_vma(mm, vm_start, vm_end);
2028         if (vma && vma->vm_file) {
2029                 *path = vma->vm_file->f_path;
2030                 path_get(path);
2031                 rc = 0;
2032         }
2033         up_read(&mm->mmap_sem);
2034
2035 out_mmput:
2036         mmput(mm);
2037 out:
2038         return rc;
2039 }
2040
2041 struct map_files_info {
2042         unsigned long   start;
2043         unsigned long   end;
2044         fmode_t         mode;
2045 };
2046
2047 /*
2048  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2049  * symlinks may be used to bypass permissions on ancestor directories in the
2050  * path to the file in question.
2051  */
2052 static const char *
2053 proc_map_files_get_link(struct dentry *dentry,
2054                         struct inode *inode,
2055                         struct delayed_call *done)
2056 {
2057         if (!capable(CAP_SYS_ADMIN))
2058                 return ERR_PTR(-EPERM);
2059
2060         return proc_pid_get_link(dentry, inode, done);
2061 }
2062
2063 /*
2064  * Identical to proc_pid_link_inode_operations except for get_link()
2065  */
2066 static const struct inode_operations proc_map_files_link_inode_operations = {
2067         .readlink       = proc_pid_readlink,
2068         .get_link       = proc_map_files_get_link,
2069         .setattr        = proc_setattr,
2070 };
2071
2072 static struct dentry *
2073 proc_map_files_instantiate(struct dentry *dentry,
2074                            struct task_struct *task, const void *ptr)
2075 {
2076         fmode_t mode = (fmode_t)(unsigned long)ptr;
2077         struct proc_inode *ei;
2078         struct inode *inode;
2079
2080         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2081                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2082                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2083         if (!inode)
2084                 return ERR_PTR(-ENOENT);
2085
2086         ei = PROC_I(inode);
2087         ei->op.proc_get_link = map_files_get_link;
2088
2089         inode->i_op = &proc_map_files_link_inode_operations;
2090         inode->i_size = 64;
2091
2092         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2093         return d_splice_alias(inode, dentry);
2094 }
2095
2096 static struct dentry *proc_map_files_lookup(struct inode *dir,
2097                 struct dentry *dentry, unsigned int flags)
2098 {
2099         unsigned long vm_start, vm_end;
2100         struct vm_area_struct *vma;
2101         struct task_struct *task;
2102         struct dentry *result;
2103         struct mm_struct *mm;
2104
2105         result = ERR_PTR(-ENOENT);
2106         task = get_proc_task(dir);
2107         if (!task)
2108                 goto out;
2109
2110         result = ERR_PTR(-EACCES);
2111         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2112                 goto out_put_task;
2113
2114         result = ERR_PTR(-ENOENT);
2115         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2116                 goto out_put_task;
2117
2118         mm = get_task_mm(task);
2119         if (!mm)
2120                 goto out_put_task;
2121
2122         down_read(&mm->mmap_sem);
2123         vma = find_exact_vma(mm, vm_start, vm_end);
2124         if (!vma)
2125                 goto out_no_vma;
2126
2127         if (vma->vm_file)
2128                 result = proc_map_files_instantiate(dentry, task,
2129                                 (void *)(unsigned long)vma->vm_file->f_mode);
2130
2131 out_no_vma:
2132         up_read(&mm->mmap_sem);
2133         mmput(mm);
2134 out_put_task:
2135         put_task_struct(task);
2136 out:
2137         return result;
2138 }
2139
2140 static const struct inode_operations proc_map_files_inode_operations = {
2141         .lookup         = proc_map_files_lookup,
2142         .permission     = proc_fd_permission,
2143         .setattr        = proc_setattr,
2144 };
2145
2146 static int
2147 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2148 {
2149         struct vm_area_struct *vma;
2150         struct task_struct *task;
2151         struct mm_struct *mm;
2152         unsigned long nr_files, pos, i;
2153         struct flex_array *fa = NULL;
2154         struct map_files_info info;
2155         struct map_files_info *p;
2156         int ret;
2157
2158         ret = -ENOENT;
2159         task = get_proc_task(file_inode(file));
2160         if (!task)
2161                 goto out;
2162
2163         ret = -EACCES;
2164         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2165                 goto out_put_task;
2166
2167         ret = 0;
2168         if (!dir_emit_dots(file, ctx))
2169                 goto out_put_task;
2170
2171         mm = get_task_mm(task);
2172         if (!mm)
2173                 goto out_put_task;
2174         down_read(&mm->mmap_sem);
2175
2176         nr_files = 0;
2177
2178         /*
2179          * We need two passes here:
2180          *
2181          *  1) Collect vmas of mapped files with mmap_sem taken
2182          *  2) Release mmap_sem and instantiate entries
2183          *
2184          * otherwise we get lockdep complained, since filldir()
2185          * routine might require mmap_sem taken in might_fault().
2186          */
2187
2188         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2189                 if (vma->vm_file && ++pos > ctx->pos)
2190                         nr_files++;
2191         }
2192
2193         if (nr_files) {
2194                 fa = flex_array_alloc(sizeof(info), nr_files,
2195                                         GFP_KERNEL);
2196                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2197                                                 GFP_KERNEL)) {
2198                         ret = -ENOMEM;
2199                         if (fa)
2200                                 flex_array_free(fa);
2201                         up_read(&mm->mmap_sem);
2202                         mmput(mm);
2203                         goto out_put_task;
2204                 }
2205                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2206                                 vma = vma->vm_next) {
2207                         if (!vma->vm_file)
2208                                 continue;
2209                         if (++pos <= ctx->pos)
2210                                 continue;
2211
2212                         info.start = vma->vm_start;
2213                         info.end = vma->vm_end;
2214                         info.mode = vma->vm_file->f_mode;
2215                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2216                                 BUG();
2217                 }
2218         }
2219         up_read(&mm->mmap_sem);
2220         mmput(mm);
2221
2222         for (i = 0; i < nr_files; i++) {
2223                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2224                 unsigned int len;
2225
2226                 p = flex_array_get(fa, i);
2227                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2228                 if (!proc_fill_cache(file, ctx,
2229                                       buf, len,
2230                                       proc_map_files_instantiate,
2231                                       task,
2232                                       (void *)(unsigned long)p->mode))
2233                         break;
2234                 ctx->pos++;
2235         }
2236         if (fa)
2237                 flex_array_free(fa);
2238
2239 out_put_task:
2240         put_task_struct(task);
2241 out:
2242         return ret;
2243 }
2244
2245 static const struct file_operations proc_map_files_operations = {
2246         .read           = generic_read_dir,
2247         .iterate_shared = proc_map_files_readdir,
2248         .llseek         = generic_file_llseek,
2249 };
2250
2251 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2252 struct timers_private {
2253         struct pid *pid;
2254         struct task_struct *task;
2255         struct sighand_struct *sighand;
2256         struct pid_namespace *ns;
2257         unsigned long flags;
2258 };
2259
2260 static void *timers_start(struct seq_file *m, loff_t *pos)
2261 {
2262         struct timers_private *tp = m->private;
2263
2264         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2265         if (!tp->task)
2266                 return ERR_PTR(-ESRCH);
2267
2268         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2269         if (!tp->sighand)
2270                 return ERR_PTR(-ESRCH);
2271
2272         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2273 }
2274
2275 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2276 {
2277         struct timers_private *tp = m->private;
2278         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2279 }
2280
2281 static void timers_stop(struct seq_file *m, void *v)
2282 {
2283         struct timers_private *tp = m->private;
2284
2285         if (tp->sighand) {
2286                 unlock_task_sighand(tp->task, &tp->flags);
2287                 tp->sighand = NULL;
2288         }
2289
2290         if (tp->task) {
2291                 put_task_struct(tp->task);
2292                 tp->task = NULL;
2293         }
2294 }
2295
2296 static int show_timer(struct seq_file *m, void *v)
2297 {
2298         struct k_itimer *timer;
2299         struct timers_private *tp = m->private;
2300         int notify;
2301         static const char * const nstr[] = {
2302                 [SIGEV_SIGNAL] = "signal",
2303                 [SIGEV_NONE] = "none",
2304                 [SIGEV_THREAD] = "thread",
2305         };
2306
2307         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2308         notify = timer->it_sigev_notify;
2309
2310         seq_printf(m, "ID: %d\n", timer->it_id);
2311         seq_printf(m, "signal: %d/%px\n",
2312                    timer->sigq->info.si_signo,
2313                    timer->sigq->info.si_value.sival_ptr);
2314         seq_printf(m, "notify: %s/%s.%d\n",
2315                    nstr[notify & ~SIGEV_THREAD_ID],
2316                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2317                    pid_nr_ns(timer->it_pid, tp->ns));
2318         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2319
2320         return 0;
2321 }
2322
2323 static const struct seq_operations proc_timers_seq_ops = {
2324         .start  = timers_start,
2325         .next   = timers_next,
2326         .stop   = timers_stop,
2327         .show   = show_timer,
2328 };
2329
2330 static int proc_timers_open(struct inode *inode, struct file *file)
2331 {
2332         struct timers_private *tp;
2333
2334         tp = __seq_open_private(file, &proc_timers_seq_ops,
2335                         sizeof(struct timers_private));
2336         if (!tp)
2337                 return -ENOMEM;
2338
2339         tp->pid = proc_pid(inode);
2340         tp->ns = proc_pid_ns(inode);
2341         return 0;
2342 }
2343
2344 static const struct file_operations proc_timers_operations = {
2345         .open           = proc_timers_open,
2346         .read           = seq_read,
2347         .llseek         = seq_lseek,
2348         .release        = seq_release_private,
2349 };
2350 #endif
2351
2352 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2353                                         size_t count, loff_t *offset)
2354 {
2355         struct inode *inode = file_inode(file);
2356         struct task_struct *p;
2357         u64 slack_ns;
2358         int err;
2359
2360         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2361         if (err < 0)
2362                 return err;
2363
2364         p = get_proc_task(inode);
2365         if (!p)
2366                 return -ESRCH;
2367
2368         if (p != current) {
2369                 if (!capable(CAP_SYS_NICE)) {
2370                         count = -EPERM;
2371                         goto out;
2372                 }
2373
2374                 err = security_task_setscheduler(p);
2375                 if (err) {
2376                         count = err;
2377                         goto out;
2378                 }
2379         }
2380
2381         task_lock(p);
2382         if (slack_ns == 0)
2383                 p->timer_slack_ns = p->default_timer_slack_ns;
2384         else
2385                 p->timer_slack_ns = slack_ns;
2386         task_unlock(p);
2387
2388 out:
2389         put_task_struct(p);
2390
2391         return count;
2392 }
2393
2394 static int timerslack_ns_show(struct seq_file *m, void *v)
2395 {
2396         struct inode *inode = m->private;
2397         struct task_struct *p;
2398         int err = 0;
2399
2400         p = get_proc_task(inode);
2401         if (!p)
2402                 return -ESRCH;
2403
2404         if (p != current) {
2405
2406                 if (!capable(CAP_SYS_NICE)) {
2407                         err = -EPERM;
2408                         goto out;
2409                 }
2410                 err = security_task_getscheduler(p);
2411                 if (err)
2412                         goto out;
2413         }
2414
2415         task_lock(p);
2416         seq_printf(m, "%llu\n", p->timer_slack_ns);
2417         task_unlock(p);
2418
2419 out:
2420         put_task_struct(p);
2421
2422         return err;
2423 }
2424
2425 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2426 {
2427         return single_open(filp, timerslack_ns_show, inode);
2428 }
2429
2430 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2431         .open           = timerslack_ns_open,
2432         .read           = seq_read,
2433         .write          = timerslack_ns_write,
2434         .llseek         = seq_lseek,
2435         .release        = single_release,
2436 };
2437
2438 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2439         struct task_struct *task, const void *ptr)
2440 {
2441         const struct pid_entry *p = ptr;
2442         struct inode *inode;
2443         struct proc_inode *ei;
2444
2445         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2446         if (!inode)
2447                 return ERR_PTR(-ENOENT);
2448
2449         ei = PROC_I(inode);
2450         if (S_ISDIR(inode->i_mode))
2451                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2452         if (p->iop)
2453                 inode->i_op = p->iop;
2454         if (p->fop)
2455                 inode->i_fop = p->fop;
2456         ei->op = p->op;
2457         pid_update_inode(task, inode);
2458         d_set_d_op(dentry, &pid_dentry_operations);
2459         return d_splice_alias(inode, dentry);
2460 }
2461
2462 static struct dentry *proc_pident_lookup(struct inode *dir, 
2463                                          struct dentry *dentry,
2464                                          const struct pid_entry *ents,
2465                                          unsigned int nents)
2466 {
2467         struct task_struct *task = get_proc_task(dir);
2468         const struct pid_entry *p, *last;
2469         struct dentry *res = ERR_PTR(-ENOENT);
2470
2471         if (!task)
2472                 goto out_no_task;
2473
2474         /*
2475          * Yes, it does not scale. And it should not. Don't add
2476          * new entries into /proc/<tgid>/ without very good reasons.
2477          */
2478         last = &ents[nents];
2479         for (p = ents; p < last; p++) {
2480                 if (p->len != dentry->d_name.len)
2481                         continue;
2482                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2483                         break;
2484         }
2485         if (p >= last)
2486                 goto out;
2487
2488         res = proc_pident_instantiate(dentry, task, p);
2489 out:
2490         put_task_struct(task);
2491 out_no_task:
2492         return res;
2493 }
2494
2495 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2496                 const struct pid_entry *ents, unsigned int nents)
2497 {
2498         struct task_struct *task = get_proc_task(file_inode(file));
2499         const struct pid_entry *p;
2500
2501         if (!task)
2502                 return -ENOENT;
2503
2504         if (!dir_emit_dots(file, ctx))
2505                 goto out;
2506
2507         if (ctx->pos >= nents + 2)
2508                 goto out;
2509
2510         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2511                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2512                                 proc_pident_instantiate, task, p))
2513                         break;
2514                 ctx->pos++;
2515         }
2516 out:
2517         put_task_struct(task);
2518         return 0;
2519 }
2520
2521 #ifdef CONFIG_SECURITY
2522 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2523                                   size_t count, loff_t *ppos)
2524 {
2525         struct inode * inode = file_inode(file);
2526         char *p = NULL;
2527         ssize_t length;
2528         struct task_struct *task = get_proc_task(inode);
2529
2530         if (!task)
2531                 return -ESRCH;
2532
2533         length = security_getprocattr(task,
2534                                       (char*)file->f_path.dentry->d_name.name,
2535                                       &p);
2536         put_task_struct(task);
2537         if (length > 0)
2538                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2539         kfree(p);
2540         return length;
2541 }
2542
2543 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2544                                    size_t count, loff_t *ppos)
2545 {
2546         struct inode * inode = file_inode(file);
2547         void *page;
2548         ssize_t length;
2549         struct task_struct *task = get_proc_task(inode);
2550
2551         length = -ESRCH;
2552         if (!task)
2553                 goto out_no_task;
2554
2555         /* A task may only write its own attributes. */
2556         length = -EACCES;
2557         if (current != task)
2558                 goto out;
2559
2560         if (count > PAGE_SIZE)
2561                 count = PAGE_SIZE;
2562
2563         /* No partial writes. */
2564         length = -EINVAL;
2565         if (*ppos != 0)
2566                 goto out;
2567
2568         page = memdup_user(buf, count);
2569         if (IS_ERR(page)) {
2570                 length = PTR_ERR(page);
2571                 goto out;
2572         }
2573
2574         /* Guard against adverse ptrace interaction */
2575         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2576         if (length < 0)
2577                 goto out_free;
2578
2579         length = security_setprocattr(file->f_path.dentry->d_name.name,
2580                                       page, count);
2581         mutex_unlock(&current->signal->cred_guard_mutex);
2582 out_free:
2583         kfree(page);
2584 out:
2585         put_task_struct(task);
2586 out_no_task:
2587         return length;
2588 }
2589
2590 static const struct file_operations proc_pid_attr_operations = {
2591         .read           = proc_pid_attr_read,
2592         .write          = proc_pid_attr_write,
2593         .llseek         = generic_file_llseek,
2594 };
2595
2596 static const struct pid_entry attr_dir_stuff[] = {
2597         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2598         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2599         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2600         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2601         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2602         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2603 };
2604
2605 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2606 {
2607         return proc_pident_readdir(file, ctx, 
2608                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2609 }
2610
2611 static const struct file_operations proc_attr_dir_operations = {
2612         .read           = generic_read_dir,
2613         .iterate_shared = proc_attr_dir_readdir,
2614         .llseek         = generic_file_llseek,
2615 };
2616
2617 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2618                                 struct dentry *dentry, unsigned int flags)
2619 {
2620         return proc_pident_lookup(dir, dentry,
2621                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2622 }
2623
2624 static const struct inode_operations proc_attr_dir_inode_operations = {
2625         .lookup         = proc_attr_dir_lookup,
2626         .getattr        = pid_getattr,
2627         .setattr        = proc_setattr,
2628 };
2629
2630 #endif
2631
2632 #ifdef CONFIG_ELF_CORE
2633 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2634                                          size_t count, loff_t *ppos)
2635 {
2636         struct task_struct *task = get_proc_task(file_inode(file));
2637         struct mm_struct *mm;
2638         char buffer[PROC_NUMBUF];
2639         size_t len;
2640         int ret;
2641
2642         if (!task)
2643                 return -ESRCH;
2644
2645         ret = 0;
2646         mm = get_task_mm(task);
2647         if (mm) {
2648                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2649                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2650                                 MMF_DUMP_FILTER_SHIFT));
2651                 mmput(mm);
2652                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2653         }
2654
2655         put_task_struct(task);
2656
2657         return ret;
2658 }
2659
2660 static ssize_t proc_coredump_filter_write(struct file *file,
2661                                           const char __user *buf,
2662                                           size_t count,
2663                                           loff_t *ppos)
2664 {
2665         struct task_struct *task;
2666         struct mm_struct *mm;
2667         unsigned int val;
2668         int ret;
2669         int i;
2670         unsigned long mask;
2671
2672         ret = kstrtouint_from_user(buf, count, 0, &val);
2673         if (ret < 0)
2674                 return ret;
2675
2676         ret = -ESRCH;
2677         task = get_proc_task(file_inode(file));
2678         if (!task)
2679                 goto out_no_task;
2680
2681         mm = get_task_mm(task);
2682         if (!mm)
2683                 goto out_no_mm;
2684         ret = 0;
2685
2686         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2687                 if (val & mask)
2688                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2689                 else
2690                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2691         }
2692
2693         mmput(mm);
2694  out_no_mm:
2695         put_task_struct(task);
2696  out_no_task:
2697         if (ret < 0)
2698                 return ret;
2699         return count;
2700 }
2701
2702 static const struct file_operations proc_coredump_filter_operations = {
2703         .read           = proc_coredump_filter_read,
2704         .write          = proc_coredump_filter_write,
2705         .llseek         = generic_file_llseek,
2706 };
2707 #endif
2708
2709 #ifdef CONFIG_TASK_IO_ACCOUNTING
2710 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2711 {
2712         struct task_io_accounting acct = task->ioac;
2713         unsigned long flags;
2714         int result;
2715
2716         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2717         if (result)
2718                 return result;
2719
2720         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2721                 result = -EACCES;
2722                 goto out_unlock;
2723         }
2724
2725         if (whole && lock_task_sighand(task, &flags)) {
2726                 struct task_struct *t = task;
2727
2728                 task_io_accounting_add(&acct, &task->signal->ioac);
2729                 while_each_thread(task, t)
2730                         task_io_accounting_add(&acct, &t->ioac);
2731
2732                 unlock_task_sighand(task, &flags);
2733         }
2734         seq_printf(m,
2735                    "rchar: %llu\n"
2736                    "wchar: %llu\n"
2737                    "syscr: %llu\n"
2738                    "syscw: %llu\n"
2739                    "read_bytes: %llu\n"
2740                    "write_bytes: %llu\n"
2741                    "cancelled_write_bytes: %llu\n",
2742                    (unsigned long long)acct.rchar,
2743                    (unsigned long long)acct.wchar,
2744                    (unsigned long long)acct.syscr,
2745                    (unsigned long long)acct.syscw,
2746                    (unsigned long long)acct.read_bytes,
2747                    (unsigned long long)acct.write_bytes,
2748                    (unsigned long long)acct.cancelled_write_bytes);
2749         result = 0;
2750
2751 out_unlock:
2752         mutex_unlock(&task->signal->cred_guard_mutex);
2753         return result;
2754 }
2755
2756 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2757                                   struct pid *pid, struct task_struct *task)
2758 {
2759         return do_io_accounting(task, m, 0);
2760 }
2761
2762 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2763                                    struct pid *pid, struct task_struct *task)
2764 {
2765         return do_io_accounting(task, m, 1);
2766 }
2767 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2768
2769 #ifdef CONFIG_USER_NS
2770 static int proc_id_map_open(struct inode *inode, struct file *file,
2771         const struct seq_operations *seq_ops)
2772 {
2773         struct user_namespace *ns = NULL;
2774         struct task_struct *task;
2775         struct seq_file *seq;
2776         int ret = -EINVAL;
2777
2778         task = get_proc_task(inode);
2779         if (task) {
2780                 rcu_read_lock();
2781                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2782                 rcu_read_unlock();
2783                 put_task_struct(task);
2784         }
2785         if (!ns)
2786                 goto err;
2787
2788         ret = seq_open(file, seq_ops);
2789         if (ret)
2790                 goto err_put_ns;
2791
2792         seq = file->private_data;
2793         seq->private = ns;
2794
2795         return 0;
2796 err_put_ns:
2797         put_user_ns(ns);
2798 err:
2799         return ret;
2800 }
2801
2802 static int proc_id_map_release(struct inode *inode, struct file *file)
2803 {
2804         struct seq_file *seq = file->private_data;
2805         struct user_namespace *ns = seq->private;
2806         put_user_ns(ns);
2807         return seq_release(inode, file);
2808 }
2809
2810 static int proc_uid_map_open(struct inode *inode, struct file *file)
2811 {
2812         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2813 }
2814
2815 static int proc_gid_map_open(struct inode *inode, struct file *file)
2816 {
2817         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2818 }
2819
2820 static int proc_projid_map_open(struct inode *inode, struct file *file)
2821 {
2822         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2823 }
2824
2825 static const struct file_operations proc_uid_map_operations = {
2826         .open           = proc_uid_map_open,
2827         .write          = proc_uid_map_write,
2828         .read           = seq_read,
2829         .llseek         = seq_lseek,
2830         .release        = proc_id_map_release,
2831 };
2832
2833 static const struct file_operations proc_gid_map_operations = {
2834         .open           = proc_gid_map_open,
2835         .write          = proc_gid_map_write,
2836         .read           = seq_read,
2837         .llseek         = seq_lseek,
2838         .release        = proc_id_map_release,
2839 };
2840
2841 static const struct file_operations proc_projid_map_operations = {
2842         .open           = proc_projid_map_open,
2843         .write          = proc_projid_map_write,
2844         .read           = seq_read,
2845         .llseek         = seq_lseek,
2846         .release        = proc_id_map_release,
2847 };
2848
2849 static int proc_setgroups_open(struct inode *inode, struct file *file)
2850 {
2851         struct user_namespace *ns = NULL;
2852         struct task_struct *task;
2853         int ret;
2854
2855         ret = -ESRCH;
2856         task = get_proc_task(inode);
2857         if (task) {
2858                 rcu_read_lock();
2859                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2860                 rcu_read_unlock();
2861                 put_task_struct(task);
2862         }
2863         if (!ns)
2864                 goto err;
2865
2866         if (file->f_mode & FMODE_WRITE) {
2867                 ret = -EACCES;
2868                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2869                         goto err_put_ns;
2870         }
2871
2872         ret = single_open(file, &proc_setgroups_show, ns);
2873         if (ret)
2874                 goto err_put_ns;
2875
2876         return 0;
2877 err_put_ns:
2878         put_user_ns(ns);
2879 err:
2880         return ret;
2881 }
2882
2883 static int proc_setgroups_release(struct inode *inode, struct file *file)
2884 {
2885         struct seq_file *seq = file->private_data;
2886         struct user_namespace *ns = seq->private;
2887         int ret = single_release(inode, file);
2888         put_user_ns(ns);
2889         return ret;
2890 }
2891
2892 static const struct file_operations proc_setgroups_operations = {
2893         .open           = proc_setgroups_open,
2894         .write          = proc_setgroups_write,
2895         .read           = seq_read,
2896         .llseek         = seq_lseek,
2897         .release        = proc_setgroups_release,
2898 };
2899 #endif /* CONFIG_USER_NS */
2900
2901 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2902                                 struct pid *pid, struct task_struct *task)
2903 {
2904         int err = lock_trace(task);
2905         if (!err) {
2906                 seq_printf(m, "%08x\n", task->personality);
2907                 unlock_trace(task);
2908         }
2909         return err;
2910 }
2911
2912 #ifdef CONFIG_LIVEPATCH
2913 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2914                                 struct pid *pid, struct task_struct *task)
2915 {
2916         seq_printf(m, "%d\n", task->patch_state);
2917         return 0;
2918 }
2919 #endif /* CONFIG_LIVEPATCH */
2920
2921 /*
2922  * Thread groups
2923  */
2924 static const struct file_operations proc_task_operations;
2925 static const struct inode_operations proc_task_inode_operations;
2926
2927 static const struct pid_entry tgid_base_stuff[] = {
2928         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2929         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2930         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2931         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2932         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2933 #ifdef CONFIG_NET
2934         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2935 #endif
2936         REG("environ",    S_IRUSR, proc_environ_operations),
2937         REG("auxv",       S_IRUSR, proc_auxv_operations),
2938         ONE("status",     S_IRUGO, proc_pid_status),
2939         ONE("personality", S_IRUSR, proc_pid_personality),
2940         ONE("limits",     S_IRUGO, proc_pid_limits),
2941 #ifdef CONFIG_SCHED_DEBUG
2942         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2943 #endif
2944 #ifdef CONFIG_SCHED_AUTOGROUP
2945         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2946 #endif
2947         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2948 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2949         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2950 #endif
2951         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2952         ONE("stat",       S_IRUGO, proc_tgid_stat),
2953         ONE("statm",      S_IRUGO, proc_pid_statm),
2954         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2955 #ifdef CONFIG_NUMA
2956         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2957 #endif
2958         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2959         LNK("cwd",        proc_cwd_link),
2960         LNK("root",       proc_root_link),
2961         LNK("exe",        proc_exe_link),
2962         REG("mounts",     S_IRUGO, proc_mounts_operations),
2963         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2964         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2965 #ifdef CONFIG_PROC_PAGE_MONITOR
2966         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2967         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2968         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2969         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2970 #endif
2971 #ifdef CONFIG_SECURITY
2972         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2973 #endif
2974 #ifdef CONFIG_KALLSYMS
2975         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2976 #endif
2977 #ifdef CONFIG_STACKTRACE
2978         ONE("stack",      S_IRUSR, proc_pid_stack),
2979 #endif
2980 #ifdef CONFIG_SCHED_INFO
2981         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2982 #endif
2983 #ifdef CONFIG_LATENCYTOP
2984         REG("latency",  S_IRUGO, proc_lstats_operations),
2985 #endif
2986 #ifdef CONFIG_PROC_PID_CPUSET
2987         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2988 #endif
2989 #ifdef CONFIG_CGROUPS
2990         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2991 #endif
2992         ONE("oom_score",  S_IRUGO, proc_oom_score),
2993         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2994         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2995 #ifdef CONFIG_AUDITSYSCALL
2996         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2997         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2998 #endif
2999 #ifdef CONFIG_FAULT_INJECTION
3000         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3001         REG("fail-nth", 0644, proc_fail_nth_operations),
3002 #endif
3003 #ifdef CONFIG_ELF_CORE
3004         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3005 #endif
3006 #ifdef CONFIG_TASK_IO_ACCOUNTING
3007         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3008 #endif
3009 #ifdef CONFIG_USER_NS
3010         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3011         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3012         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3013         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3014 #endif
3015 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3016         REG("timers",     S_IRUGO, proc_timers_operations),
3017 #endif
3018         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3019 #ifdef CONFIG_LIVEPATCH
3020         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3021 #endif
3022 };
3023
3024 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3025 {
3026         return proc_pident_readdir(file, ctx,
3027                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3028 }
3029
3030 static const struct file_operations proc_tgid_base_operations = {
3031         .read           = generic_read_dir,
3032         .iterate_shared = proc_tgid_base_readdir,
3033         .llseek         = generic_file_llseek,
3034 };
3035
3036 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3037 {
3038         return proc_pident_lookup(dir, dentry,
3039                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3040 }
3041
3042 static const struct inode_operations proc_tgid_base_inode_operations = {
3043         .lookup         = proc_tgid_base_lookup,
3044         .getattr        = pid_getattr,
3045         .setattr        = proc_setattr,
3046         .permission     = proc_pid_permission,
3047 };
3048
3049 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3050 {
3051         struct dentry *dentry, *leader, *dir;
3052         char buf[10 + 1];
3053         struct qstr name;
3054
3055         name.name = buf;
3056         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3057         /* no ->d_hash() rejects on procfs */
3058         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3059         if (dentry) {
3060                 d_invalidate(dentry);
3061                 dput(dentry);
3062         }
3063
3064         if (pid == tgid)
3065                 return;
3066
3067         name.name = buf;
3068         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3069         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3070         if (!leader)
3071                 goto out;
3072
3073         name.name = "task";
3074         name.len = strlen(name.name);
3075         dir = d_hash_and_lookup(leader, &name);
3076         if (!dir)
3077                 goto out_put_leader;
3078
3079         name.name = buf;
3080         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3081         dentry = d_hash_and_lookup(dir, &name);
3082         if (dentry) {
3083                 d_invalidate(dentry);
3084                 dput(dentry);
3085         }
3086
3087         dput(dir);
3088 out_put_leader:
3089         dput(leader);
3090 out:
3091         return;
3092 }
3093
3094 /**
3095  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3096  * @task: task that should be flushed.
3097  *
3098  * When flushing dentries from proc, one needs to flush them from global
3099  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3100  * in. This call is supposed to do all of this job.
3101  *
3102  * Looks in the dcache for
3103  * /proc/@pid
3104  * /proc/@tgid/task/@pid
3105  * if either directory is present flushes it and all of it'ts children
3106  * from the dcache.
3107  *
3108  * It is safe and reasonable to cache /proc entries for a task until
3109  * that task exits.  After that they just clog up the dcache with
3110  * useless entries, possibly causing useful dcache entries to be
3111  * flushed instead.  This routine is proved to flush those useless
3112  * dcache entries at process exit time.
3113  *
3114  * NOTE: This routine is just an optimization so it does not guarantee
3115  *       that no dcache entries will exist at process exit time it
3116  *       just makes it very unlikely that any will persist.
3117  */
3118
3119 void proc_flush_task(struct task_struct *task)
3120 {
3121         int i;
3122         struct pid *pid, *tgid;
3123         struct upid *upid;
3124
3125         pid = task_pid(task);
3126         tgid = task_tgid(task);
3127
3128         for (i = 0; i <= pid->level; i++) {
3129                 upid = &pid->numbers[i];
3130                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3131                                         tgid->numbers[i].nr);
3132         }
3133 }
3134
3135 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3136                                    struct task_struct *task, const void *ptr)
3137 {
3138         struct inode *inode;
3139
3140         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3141         if (!inode)
3142                 return ERR_PTR(-ENOENT);
3143
3144         inode->i_op = &proc_tgid_base_inode_operations;
3145         inode->i_fop = &proc_tgid_base_operations;
3146         inode->i_flags|=S_IMMUTABLE;
3147
3148         set_nlink(inode, nlink_tgid);
3149         pid_update_inode(task, inode);
3150
3151         d_set_d_op(dentry, &pid_dentry_operations);
3152         return d_splice_alias(inode, dentry);
3153 }
3154
3155 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3156 {
3157         struct task_struct *task;
3158         unsigned tgid;
3159         struct pid_namespace *ns;
3160         struct dentry *result = ERR_PTR(-ENOENT);
3161
3162         tgid = name_to_int(&dentry->d_name);
3163         if (tgid == ~0U)
3164                 goto out;
3165
3166         ns = dentry->d_sb->s_fs_info;
3167         rcu_read_lock();
3168         task = find_task_by_pid_ns(tgid, ns);
3169         if (task)
3170                 get_task_struct(task);
3171         rcu_read_unlock();
3172         if (!task)
3173                 goto out;
3174
3175         result = proc_pid_instantiate(dentry, task, NULL);
3176         put_task_struct(task);
3177 out:
3178         return result;
3179 }
3180
3181 /*
3182  * Find the first task with tgid >= tgid
3183  *
3184  */
3185 struct tgid_iter {
3186         unsigned int tgid;
3187         struct task_struct *task;
3188 };
3189 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3190 {
3191         struct pid *pid;
3192
3193         if (iter.task)
3194                 put_task_struct(iter.task);
3195         rcu_read_lock();
3196 retry:
3197         iter.task = NULL;
3198         pid = find_ge_pid(iter.tgid, ns);
3199         if (pid) {
3200                 iter.tgid = pid_nr_ns(pid, ns);
3201                 iter.task = pid_task(pid, PIDTYPE_PID);
3202                 /* What we to know is if the pid we have find is the
3203                  * pid of a thread_group_leader.  Testing for task
3204                  * being a thread_group_leader is the obvious thing
3205                  * todo but there is a window when it fails, due to
3206                  * the pid transfer logic in de_thread.
3207                  *
3208                  * So we perform the straight forward test of seeing
3209                  * if the pid we have found is the pid of a thread
3210                  * group leader, and don't worry if the task we have
3211                  * found doesn't happen to be a thread group leader.
3212                  * As we don't care in the case of readdir.
3213                  */
3214                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3215                         iter.tgid += 1;
3216                         goto retry;
3217                 }
3218                 get_task_struct(iter.task);
3219         }
3220         rcu_read_unlock();
3221         return iter;
3222 }
3223
3224 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3225
3226 /* for the /proc/ directory itself, after non-process stuff has been done */
3227 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3228 {
3229         struct tgid_iter iter;
3230         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3231         loff_t pos = ctx->pos;
3232
3233         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3234                 return 0;
3235
3236         if (pos == TGID_OFFSET - 2) {
3237                 struct inode *inode = d_inode(ns->proc_self);
3238                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3239                         return 0;
3240                 ctx->pos = pos = pos + 1;
3241         }
3242         if (pos == TGID_OFFSET - 1) {
3243                 struct inode *inode = d_inode(ns->proc_thread_self);
3244                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3245                         return 0;
3246                 ctx->pos = pos = pos + 1;
3247         }
3248         iter.tgid = pos - TGID_OFFSET;
3249         iter.task = NULL;
3250         for (iter = next_tgid(ns, iter);
3251              iter.task;
3252              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3253                 char name[10 + 1];
3254                 int len;
3255
3256                 cond_resched();
3257                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3258                         continue;
3259
3260                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3261                 ctx->pos = iter.tgid + TGID_OFFSET;
3262                 if (!proc_fill_cache(file, ctx, name, len,
3263                                      proc_pid_instantiate, iter.task, NULL)) {
3264                         put_task_struct(iter.task);
3265                         return 0;
3266                 }
3267         }
3268         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3269         return 0;
3270 }
3271
3272 /*
3273  * proc_tid_comm_permission is a special permission function exclusively
3274  * used for the node /proc/<pid>/task/<tid>/comm.
3275  * It bypasses generic permission checks in the case where a task of the same
3276  * task group attempts to access the node.
3277  * The rationale behind this is that glibc and bionic access this node for
3278  * cross thread naming (pthread_set/getname_np(!self)). However, if
3279  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3280  * which locks out the cross thread naming implementation.
3281  * This function makes sure that the node is always accessible for members of
3282  * same thread group.
3283  */
3284 static int proc_tid_comm_permission(struct inode *inode, int mask)
3285 {
3286         bool is_same_tgroup;
3287         struct task_struct *task;
3288
3289         task = get_proc_task(inode);
3290         if (!task)
3291                 return -ESRCH;
3292         is_same_tgroup = same_thread_group(current, task);
3293         put_task_struct(task);
3294
3295         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3296                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3297                  * read or written by the members of the corresponding
3298                  * thread group.
3299                  */
3300                 return 0;
3301         }
3302
3303         return generic_permission(inode, mask);
3304 }
3305
3306 static const struct inode_operations proc_tid_comm_inode_operations = {
3307                 .permission = proc_tid_comm_permission,
3308 };
3309
3310 /*
3311  * Tasks
3312  */
3313 static const struct pid_entry tid_base_stuff[] = {
3314         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3315         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3316         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3317 #ifdef CONFIG_NET
3318         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3319 #endif
3320         REG("environ",   S_IRUSR, proc_environ_operations),
3321         REG("auxv",      S_IRUSR, proc_auxv_operations),
3322         ONE("status",    S_IRUGO, proc_pid_status),
3323         ONE("personality", S_IRUSR, proc_pid_personality),
3324         ONE("limits",    S_IRUGO, proc_pid_limits),
3325 #ifdef CONFIG_SCHED_DEBUG
3326         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3327 #endif
3328         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3329                          &proc_tid_comm_inode_operations,
3330                          &proc_pid_set_comm_operations, {}),
3331 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3332         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3333 #endif
3334         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3335         ONE("stat",      S_IRUGO, proc_tid_stat),
3336         ONE("statm",     S_IRUGO, proc_pid_statm),
3337         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3338 #ifdef CONFIG_PROC_CHILDREN
3339         REG("children",  S_IRUGO, proc_tid_children_operations),
3340 #endif
3341 #ifdef CONFIG_NUMA
3342         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3343 #endif
3344         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3345         LNK("cwd",       proc_cwd_link),
3346         LNK("root",      proc_root_link),
3347         LNK("exe",       proc_exe_link),
3348         REG("mounts",    S_IRUGO, proc_mounts_operations),
3349         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3350 #ifdef CONFIG_PROC_PAGE_MONITOR
3351         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3352         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3353         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3354         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3355 #endif
3356 #ifdef CONFIG_SECURITY
3357         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3358 #endif
3359 #ifdef CONFIG_KALLSYMS
3360         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3361 #endif
3362 #ifdef CONFIG_STACKTRACE
3363         ONE("stack",      S_IRUSR, proc_pid_stack),
3364 #endif
3365 #ifdef CONFIG_SCHED_INFO
3366         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3367 #endif
3368 #ifdef CONFIG_LATENCYTOP
3369         REG("latency",  S_IRUGO, proc_lstats_operations),
3370 #endif
3371 #ifdef CONFIG_PROC_PID_CPUSET
3372         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3373 #endif
3374 #ifdef CONFIG_CGROUPS
3375         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3376 #endif
3377         ONE("oom_score", S_IRUGO, proc_oom_score),
3378         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3379         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3380 #ifdef CONFIG_AUDITSYSCALL
3381         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3382         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3383 #endif
3384 #ifdef CONFIG_FAULT_INJECTION
3385         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3386         REG("fail-nth", 0644, proc_fail_nth_operations),
3387 #endif
3388 #ifdef CONFIG_TASK_IO_ACCOUNTING
3389         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3390 #endif
3391 #ifdef CONFIG_USER_NS
3392         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3393         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3394         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3395         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3396 #endif
3397 #ifdef CONFIG_LIVEPATCH
3398         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3399 #endif
3400 };
3401
3402 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3403 {
3404         return proc_pident_readdir(file, ctx,
3405                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3406 }
3407
3408 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3409 {
3410         return proc_pident_lookup(dir, dentry,
3411                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3412 }
3413
3414 static const struct file_operations proc_tid_base_operations = {
3415         .read           = generic_read_dir,
3416         .iterate_shared = proc_tid_base_readdir,
3417         .llseek         = generic_file_llseek,
3418 };
3419
3420 static const struct inode_operations proc_tid_base_inode_operations = {
3421         .lookup         = proc_tid_base_lookup,
3422         .getattr        = pid_getattr,
3423         .setattr        = proc_setattr,
3424 };
3425
3426 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3427         struct task_struct *task, const void *ptr)
3428 {
3429         struct inode *inode;
3430         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3431         if (!inode)
3432                 return ERR_PTR(-ENOENT);
3433
3434         inode->i_op = &proc_tid_base_inode_operations;
3435         inode->i_fop = &proc_tid_base_operations;
3436         inode->i_flags |= S_IMMUTABLE;
3437
3438         set_nlink(inode, nlink_tid);
3439         pid_update_inode(task, inode);
3440
3441         d_set_d_op(dentry, &pid_dentry_operations);
3442         return d_splice_alias(inode, dentry);
3443 }
3444
3445 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3446 {
3447         struct task_struct *task;
3448         struct task_struct *leader = get_proc_task(dir);
3449         unsigned tid;
3450         struct pid_namespace *ns;
3451         struct dentry *result = ERR_PTR(-ENOENT);
3452
3453         if (!leader)
3454                 goto out_no_task;
3455
3456         tid = name_to_int(&dentry->d_name);
3457         if (tid == ~0U)
3458                 goto out;
3459
3460         ns = dentry->d_sb->s_fs_info;
3461         rcu_read_lock();
3462         task = find_task_by_pid_ns(tid, ns);
3463         if (task)
3464                 get_task_struct(task);
3465         rcu_read_unlock();
3466         if (!task)
3467                 goto out;
3468         if (!same_thread_group(leader, task))
3469                 goto out_drop_task;
3470
3471         result = proc_task_instantiate(dentry, task, NULL);
3472 out_drop_task:
3473         put_task_struct(task);
3474 out:
3475         put_task_struct(leader);
3476 out_no_task:
3477         return result;
3478 }
3479
3480 /*
3481  * Find the first tid of a thread group to return to user space.
3482  *
3483  * Usually this is just the thread group leader, but if the users
3484  * buffer was too small or there was a seek into the middle of the
3485  * directory we have more work todo.
3486  *
3487  * In the case of a short read we start with find_task_by_pid.
3488  *
3489  * In the case of a seek we start with the leader and walk nr
3490  * threads past it.
3491  */
3492 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3493                                         struct pid_namespace *ns)
3494 {
3495         struct task_struct *pos, *task;
3496         unsigned long nr = f_pos;
3497
3498         if (nr != f_pos)        /* 32bit overflow? */
3499                 return NULL;
3500
3501         rcu_read_lock();
3502         task = pid_task(pid, PIDTYPE_PID);
3503         if (!task)
3504                 goto fail;
3505
3506         /* Attempt to start with the tid of a thread */
3507         if (tid && nr) {
3508                 pos = find_task_by_pid_ns(tid, ns);
3509                 if (pos && same_thread_group(pos, task))
3510                         goto found;
3511         }
3512
3513         /* If nr exceeds the number of threads there is nothing todo */
3514         if (nr >= get_nr_threads(task))
3515                 goto fail;
3516
3517         /* If we haven't found our starting place yet start
3518          * with the leader and walk nr threads forward.
3519          */
3520         pos = task = task->group_leader;
3521         do {
3522                 if (!nr--)
3523                         goto found;
3524         } while_each_thread(task, pos);
3525 fail:
3526         pos = NULL;
3527         goto out;
3528 found:
3529         get_task_struct(pos);
3530 out:
3531         rcu_read_unlock();
3532         return pos;
3533 }
3534
3535 /*
3536  * Find the next thread in the thread list.
3537  * Return NULL if there is an error or no next thread.
3538  *
3539  * The reference to the input task_struct is released.
3540  */
3541 static struct task_struct *next_tid(struct task_struct *start)
3542 {
3543         struct task_struct *pos = NULL;
3544         rcu_read_lock();
3545         if (pid_alive(start)) {
3546                 pos = next_thread(start);
3547                 if (thread_group_leader(pos))
3548                         pos = NULL;
3549                 else
3550                         get_task_struct(pos);
3551         }
3552         rcu_read_unlock();
3553         put_task_struct(start);
3554         return pos;
3555 }
3556
3557 /* for the /proc/TGID/task/ directories */
3558 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3559 {
3560         struct inode *inode = file_inode(file);
3561         struct task_struct *task;
3562         struct pid_namespace *ns;
3563         int tid;
3564
3565         if (proc_inode_is_dead(inode))
3566                 return -ENOENT;
3567
3568         if (!dir_emit_dots(file, ctx))
3569                 return 0;
3570
3571         /* f_version caches the tgid value that the last readdir call couldn't
3572          * return. lseek aka telldir automagically resets f_version to 0.
3573          */
3574         ns = proc_pid_ns(inode);
3575         tid = (int)file->f_version;
3576         file->f_version = 0;
3577         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3578              task;
3579              task = next_tid(task), ctx->pos++) {
3580                 char name[10 + 1];
3581                 int len;
3582                 tid = task_pid_nr_ns(task, ns);
3583                 len = snprintf(name, sizeof(name), "%u", tid);
3584                 if (!proc_fill_cache(file, ctx, name, len,
3585                                 proc_task_instantiate, task, NULL)) {
3586                         /* returning this tgid failed, save it as the first
3587                          * pid for the next readir call */
3588                         file->f_version = (u64)tid;
3589                         put_task_struct(task);
3590                         break;
3591                 }
3592         }
3593
3594         return 0;
3595 }
3596
3597 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3598                              u32 request_mask, unsigned int query_flags)
3599 {
3600         struct inode *inode = d_inode(path->dentry);
3601         struct task_struct *p = get_proc_task(inode);
3602         generic_fillattr(inode, stat);
3603
3604         if (p) {
3605                 stat->nlink += get_nr_threads(p);
3606                 put_task_struct(p);
3607         }
3608
3609         return 0;
3610 }
3611
3612 static const struct inode_operations proc_task_inode_operations = {
3613         .lookup         = proc_task_lookup,
3614         .getattr        = proc_task_getattr,
3615         .setattr        = proc_setattr,
3616         .permission     = proc_pid_permission,
3617 };
3618
3619 static const struct file_operations proc_task_operations = {
3620         .read           = generic_read_dir,
3621         .iterate_shared = proc_task_readdir,
3622         .llseek         = generic_file_llseek,
3623 };
3624
3625 void __init set_proc_pid_nlink(void)
3626 {
3627         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3628         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3629 }