Merge branch 'work.lookup' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[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         task_lock(p);
1567         seq_printf(m, "%s\n", p->comm);
1568         task_unlock(p);
1569
1570         put_task_struct(p);
1571
1572         return 0;
1573 }
1574
1575 static int comm_open(struct inode *inode, struct file *filp)
1576 {
1577         return single_open(filp, comm_show, inode);
1578 }
1579
1580 static const struct file_operations proc_pid_set_comm_operations = {
1581         .open           = comm_open,
1582         .read           = seq_read,
1583         .write          = comm_write,
1584         .llseek         = seq_lseek,
1585         .release        = single_release,
1586 };
1587
1588 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1589 {
1590         struct task_struct *task;
1591         struct file *exe_file;
1592
1593         task = get_proc_task(d_inode(dentry));
1594         if (!task)
1595                 return -ENOENT;
1596         exe_file = get_task_exe_file(task);
1597         put_task_struct(task);
1598         if (exe_file) {
1599                 *exe_path = exe_file->f_path;
1600                 path_get(&exe_file->f_path);
1601                 fput(exe_file);
1602                 return 0;
1603         } else
1604                 return -ENOENT;
1605 }
1606
1607 static const char *proc_pid_get_link(struct dentry *dentry,
1608                                      struct inode *inode,
1609                                      struct delayed_call *done)
1610 {
1611         struct path path;
1612         int error = -EACCES;
1613
1614         if (!dentry)
1615                 return ERR_PTR(-ECHILD);
1616
1617         /* Are we allowed to snoop on the tasks file descriptors? */
1618         if (!proc_fd_access_allowed(inode))
1619                 goto out;
1620
1621         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1622         if (error)
1623                 goto out;
1624
1625         nd_jump_link(&path);
1626         return NULL;
1627 out:
1628         return ERR_PTR(error);
1629 }
1630
1631 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1632 {
1633         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1634         char *pathname;
1635         int len;
1636
1637         if (!tmp)
1638                 return -ENOMEM;
1639
1640         pathname = d_path(path, tmp, PAGE_SIZE);
1641         len = PTR_ERR(pathname);
1642         if (IS_ERR(pathname))
1643                 goto out;
1644         len = tmp + PAGE_SIZE - 1 - pathname;
1645
1646         if (len > buflen)
1647                 len = buflen;
1648         if (copy_to_user(buffer, pathname, len))
1649                 len = -EFAULT;
1650  out:
1651         free_page((unsigned long)tmp);
1652         return len;
1653 }
1654
1655 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1656 {
1657         int error = -EACCES;
1658         struct inode *inode = d_inode(dentry);
1659         struct path path;
1660
1661         /* Are we allowed to snoop on the tasks file descriptors? */
1662         if (!proc_fd_access_allowed(inode))
1663                 goto out;
1664
1665         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1666         if (error)
1667                 goto out;
1668
1669         error = do_proc_readlink(&path, buffer, buflen);
1670         path_put(&path);
1671 out:
1672         return error;
1673 }
1674
1675 const struct inode_operations proc_pid_link_inode_operations = {
1676         .readlink       = proc_pid_readlink,
1677         .get_link       = proc_pid_get_link,
1678         .setattr        = proc_setattr,
1679 };
1680
1681
1682 /* building an inode */
1683
1684 void task_dump_owner(struct task_struct *task, umode_t mode,
1685                      kuid_t *ruid, kgid_t *rgid)
1686 {
1687         /* Depending on the state of dumpable compute who should own a
1688          * proc file for a task.
1689          */
1690         const struct cred *cred;
1691         kuid_t uid;
1692         kgid_t gid;
1693
1694         if (unlikely(task->flags & PF_KTHREAD)) {
1695                 *ruid = GLOBAL_ROOT_UID;
1696                 *rgid = GLOBAL_ROOT_GID;
1697                 return;
1698         }
1699
1700         /* Default to the tasks effective ownership */
1701         rcu_read_lock();
1702         cred = __task_cred(task);
1703         uid = cred->euid;
1704         gid = cred->egid;
1705         rcu_read_unlock();
1706
1707         /*
1708          * Before the /proc/pid/status file was created the only way to read
1709          * the effective uid of a /process was to stat /proc/pid.  Reading
1710          * /proc/pid/status is slow enough that procps and other packages
1711          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1712          * made this apply to all per process world readable and executable
1713          * directories.
1714          */
1715         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1716                 struct mm_struct *mm;
1717                 task_lock(task);
1718                 mm = task->mm;
1719                 /* Make non-dumpable tasks owned by some root */
1720                 if (mm) {
1721                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1722                                 struct user_namespace *user_ns = mm->user_ns;
1723
1724                                 uid = make_kuid(user_ns, 0);
1725                                 if (!uid_valid(uid))
1726                                         uid = GLOBAL_ROOT_UID;
1727
1728                                 gid = make_kgid(user_ns, 0);
1729                                 if (!gid_valid(gid))
1730                                         gid = GLOBAL_ROOT_GID;
1731                         }
1732                 } else {
1733                         uid = GLOBAL_ROOT_UID;
1734                         gid = GLOBAL_ROOT_GID;
1735                 }
1736                 task_unlock(task);
1737         }
1738         *ruid = uid;
1739         *rgid = gid;
1740 }
1741
1742 struct inode *proc_pid_make_inode(struct super_block * sb,
1743                                   struct task_struct *task, umode_t mode)
1744 {
1745         struct inode * inode;
1746         struct proc_inode *ei;
1747
1748         /* We need a new inode */
1749
1750         inode = new_inode(sb);
1751         if (!inode)
1752                 goto out;
1753
1754         /* Common stuff */
1755         ei = PROC_I(inode);
1756         inode->i_mode = mode;
1757         inode->i_ino = get_next_ino();
1758         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1759         inode->i_op = &proc_def_inode_operations;
1760
1761         /*
1762          * grab the reference to task.
1763          */
1764         ei->pid = get_task_pid(task, PIDTYPE_PID);
1765         if (!ei->pid)
1766                 goto out_unlock;
1767
1768         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1769         security_task_to_inode(task, inode);
1770
1771 out:
1772         return inode;
1773
1774 out_unlock:
1775         iput(inode);
1776         return NULL;
1777 }
1778
1779 int pid_getattr(const struct path *path, struct kstat *stat,
1780                 u32 request_mask, unsigned int query_flags)
1781 {
1782         struct inode *inode = d_inode(path->dentry);
1783         struct pid_namespace *pid = proc_pid_ns(inode);
1784         struct task_struct *task;
1785
1786         generic_fillattr(inode, stat);
1787
1788         rcu_read_lock();
1789         stat->uid = GLOBAL_ROOT_UID;
1790         stat->gid = GLOBAL_ROOT_GID;
1791         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1792         if (task) {
1793                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1794                         rcu_read_unlock();
1795                         /*
1796                          * This doesn't prevent learning whether PID exists,
1797                          * it only makes getattr() consistent with readdir().
1798                          */
1799                         return -ENOENT;
1800                 }
1801                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1802         }
1803         rcu_read_unlock();
1804         return 0;
1805 }
1806
1807 /* dentry stuff */
1808
1809 /*
1810  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1811  */
1812 void pid_update_inode(struct task_struct *task, struct inode *inode)
1813 {
1814         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1815
1816         inode->i_mode &= ~(S_ISUID | S_ISGID);
1817         security_task_to_inode(task, inode);
1818 }
1819
1820 /*
1821  * Rewrite the inode's ownerships here because the owning task may have
1822  * performed a setuid(), etc.
1823  *
1824  */
1825 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1826 {
1827         struct inode *inode;
1828         struct task_struct *task;
1829
1830         if (flags & LOOKUP_RCU)
1831                 return -ECHILD;
1832
1833         inode = d_inode(dentry);
1834         task = get_proc_task(inode);
1835
1836         if (task) {
1837                 pid_update_inode(task, inode);
1838                 put_task_struct(task);
1839                 return 1;
1840         }
1841         return 0;
1842 }
1843
1844 static inline bool proc_inode_is_dead(struct inode *inode)
1845 {
1846         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1847 }
1848
1849 int pid_delete_dentry(const struct dentry *dentry)
1850 {
1851         /* Is the task we represent dead?
1852          * If so, then don't put the dentry on the lru list,
1853          * kill it immediately.
1854          */
1855         return proc_inode_is_dead(d_inode(dentry));
1856 }
1857
1858 const struct dentry_operations pid_dentry_operations =
1859 {
1860         .d_revalidate   = pid_revalidate,
1861         .d_delete       = pid_delete_dentry,
1862 };
1863
1864 /* Lookups */
1865
1866 /*
1867  * Fill a directory entry.
1868  *
1869  * If possible create the dcache entry and derive our inode number and
1870  * file type from dcache entry.
1871  *
1872  * Since all of the proc inode numbers are dynamically generated, the inode
1873  * numbers do not exist until the inode is cache.  This means creating the
1874  * the dcache entry in readdir is necessary to keep the inode numbers
1875  * reported by readdir in sync with the inode numbers reported
1876  * by stat.
1877  */
1878 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1879         const char *name, int len,
1880         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1881 {
1882         struct dentry *child, *dir = file->f_path.dentry;
1883         struct qstr qname = QSTR_INIT(name, len);
1884         struct inode *inode;
1885         unsigned type = DT_UNKNOWN;
1886         ino_t ino = 1;
1887
1888         child = d_hash_and_lookup(dir, &qname);
1889         if (!child) {
1890                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1891                 child = d_alloc_parallel(dir, &qname, &wq);
1892                 if (IS_ERR(child))
1893                         goto end_instantiate;
1894                 if (d_in_lookup(child)) {
1895                         struct dentry *res;
1896                         res = instantiate(child, task, ptr);
1897                         d_lookup_done(child);
1898                         if (IS_ERR(res))
1899                                 goto end_instantiate;
1900                         if (unlikely(res)) {
1901                                 dput(child);
1902                                 child = res;
1903                         }
1904                 }
1905         }
1906         inode = d_inode(child);
1907         ino = inode->i_ino;
1908         type = inode->i_mode >> 12;
1909 end_instantiate:
1910         dput(child);
1911         return dir_emit(ctx, name, len, ino, type);
1912 }
1913
1914 /*
1915  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1916  * which represent vma start and end addresses.
1917  */
1918 static int dname_to_vma_addr(struct dentry *dentry,
1919                              unsigned long *start, unsigned long *end)
1920 {
1921         const char *str = dentry->d_name.name;
1922         unsigned long long sval, eval;
1923         unsigned int len;
1924
1925         if (str[0] == '0' && str[1] != '-')
1926                 return -EINVAL;
1927         len = _parse_integer(str, 16, &sval);
1928         if (len & KSTRTOX_OVERFLOW)
1929                 return -EINVAL;
1930         if (sval != (unsigned long)sval)
1931                 return -EINVAL;
1932         str += len;
1933
1934         if (*str != '-')
1935                 return -EINVAL;
1936         str++;
1937
1938         if (str[0] == '0' && str[1])
1939                 return -EINVAL;
1940         len = _parse_integer(str, 16, &eval);
1941         if (len & KSTRTOX_OVERFLOW)
1942                 return -EINVAL;
1943         if (eval != (unsigned long)eval)
1944                 return -EINVAL;
1945         str += len;
1946
1947         if (*str != '\0')
1948                 return -EINVAL;
1949
1950         *start = sval;
1951         *end = eval;
1952
1953         return 0;
1954 }
1955
1956 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1957 {
1958         unsigned long vm_start, vm_end;
1959         bool exact_vma_exists = false;
1960         struct mm_struct *mm = NULL;
1961         struct task_struct *task;
1962         struct inode *inode;
1963         int status = 0;
1964
1965         if (flags & LOOKUP_RCU)
1966                 return -ECHILD;
1967
1968         inode = d_inode(dentry);
1969         task = get_proc_task(inode);
1970         if (!task)
1971                 goto out_notask;
1972
1973         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1974         if (IS_ERR_OR_NULL(mm))
1975                 goto out;
1976
1977         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1978                 down_read(&mm->mmap_sem);
1979                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1980                 up_read(&mm->mmap_sem);
1981         }
1982
1983         mmput(mm);
1984
1985         if (exact_vma_exists) {
1986                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1987
1988                 security_task_to_inode(task, inode);
1989                 status = 1;
1990         }
1991
1992 out:
1993         put_task_struct(task);
1994
1995 out_notask:
1996         return status;
1997 }
1998
1999 static const struct dentry_operations tid_map_files_dentry_operations = {
2000         .d_revalidate   = map_files_d_revalidate,
2001         .d_delete       = pid_delete_dentry,
2002 };
2003
2004 static int map_files_get_link(struct dentry *dentry, struct path *path)
2005 {
2006         unsigned long vm_start, vm_end;
2007         struct vm_area_struct *vma;
2008         struct task_struct *task;
2009         struct mm_struct *mm;
2010         int rc;
2011
2012         rc = -ENOENT;
2013         task = get_proc_task(d_inode(dentry));
2014         if (!task)
2015                 goto out;
2016
2017         mm = get_task_mm(task);
2018         put_task_struct(task);
2019         if (!mm)
2020                 goto out;
2021
2022         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2023         if (rc)
2024                 goto out_mmput;
2025
2026         rc = -ENOENT;
2027         down_read(&mm->mmap_sem);
2028         vma = find_exact_vma(mm, vm_start, vm_end);
2029         if (vma && vma->vm_file) {
2030                 *path = vma->vm_file->f_path;
2031                 path_get(path);
2032                 rc = 0;
2033         }
2034         up_read(&mm->mmap_sem);
2035
2036 out_mmput:
2037         mmput(mm);
2038 out:
2039         return rc;
2040 }
2041
2042 struct map_files_info {
2043         unsigned long   start;
2044         unsigned long   end;
2045         fmode_t         mode;
2046 };
2047
2048 /*
2049  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2050  * symlinks may be used to bypass permissions on ancestor directories in the
2051  * path to the file in question.
2052  */
2053 static const char *
2054 proc_map_files_get_link(struct dentry *dentry,
2055                         struct inode *inode,
2056                         struct delayed_call *done)
2057 {
2058         if (!capable(CAP_SYS_ADMIN))
2059                 return ERR_PTR(-EPERM);
2060
2061         return proc_pid_get_link(dentry, inode, done);
2062 }
2063
2064 /*
2065  * Identical to proc_pid_link_inode_operations except for get_link()
2066  */
2067 static const struct inode_operations proc_map_files_link_inode_operations = {
2068         .readlink       = proc_pid_readlink,
2069         .get_link       = proc_map_files_get_link,
2070         .setattr        = proc_setattr,
2071 };
2072
2073 static struct dentry *
2074 proc_map_files_instantiate(struct dentry *dentry,
2075                            struct task_struct *task, const void *ptr)
2076 {
2077         fmode_t mode = (fmode_t)(unsigned long)ptr;
2078         struct proc_inode *ei;
2079         struct inode *inode;
2080
2081         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2082                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2083                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2084         if (!inode)
2085                 return ERR_PTR(-ENOENT);
2086
2087         ei = PROC_I(inode);
2088         ei->op.proc_get_link = map_files_get_link;
2089
2090         inode->i_op = &proc_map_files_link_inode_operations;
2091         inode->i_size = 64;
2092
2093         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2094         return d_splice_alias(inode, dentry);
2095 }
2096
2097 static struct dentry *proc_map_files_lookup(struct inode *dir,
2098                 struct dentry *dentry, unsigned int flags)
2099 {
2100         unsigned long vm_start, vm_end;
2101         struct vm_area_struct *vma;
2102         struct task_struct *task;
2103         struct dentry *result;
2104         struct mm_struct *mm;
2105
2106         result = ERR_PTR(-ENOENT);
2107         task = get_proc_task(dir);
2108         if (!task)
2109                 goto out;
2110
2111         result = ERR_PTR(-EACCES);
2112         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2113                 goto out_put_task;
2114
2115         result = ERR_PTR(-ENOENT);
2116         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2117                 goto out_put_task;
2118
2119         mm = get_task_mm(task);
2120         if (!mm)
2121                 goto out_put_task;
2122
2123         down_read(&mm->mmap_sem);
2124         vma = find_exact_vma(mm, vm_start, vm_end);
2125         if (!vma)
2126                 goto out_no_vma;
2127
2128         if (vma->vm_file)
2129                 result = proc_map_files_instantiate(dentry, task,
2130                                 (void *)(unsigned long)vma->vm_file->f_mode);
2131
2132 out_no_vma:
2133         up_read(&mm->mmap_sem);
2134         mmput(mm);
2135 out_put_task:
2136         put_task_struct(task);
2137 out:
2138         return result;
2139 }
2140
2141 static const struct inode_operations proc_map_files_inode_operations = {
2142         .lookup         = proc_map_files_lookup,
2143         .permission     = proc_fd_permission,
2144         .setattr        = proc_setattr,
2145 };
2146
2147 static int
2148 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2149 {
2150         struct vm_area_struct *vma;
2151         struct task_struct *task;
2152         struct mm_struct *mm;
2153         unsigned long nr_files, pos, i;
2154         struct flex_array *fa = NULL;
2155         struct map_files_info info;
2156         struct map_files_info *p;
2157         int ret;
2158
2159         ret = -ENOENT;
2160         task = get_proc_task(file_inode(file));
2161         if (!task)
2162                 goto out;
2163
2164         ret = -EACCES;
2165         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2166                 goto out_put_task;
2167
2168         ret = 0;
2169         if (!dir_emit_dots(file, ctx))
2170                 goto out_put_task;
2171
2172         mm = get_task_mm(task);
2173         if (!mm)
2174                 goto out_put_task;
2175         down_read(&mm->mmap_sem);
2176
2177         nr_files = 0;
2178
2179         /*
2180          * We need two passes here:
2181          *
2182          *  1) Collect vmas of mapped files with mmap_sem taken
2183          *  2) Release mmap_sem and instantiate entries
2184          *
2185          * otherwise we get lockdep complained, since filldir()
2186          * routine might require mmap_sem taken in might_fault().
2187          */
2188
2189         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2190                 if (vma->vm_file && ++pos > ctx->pos)
2191                         nr_files++;
2192         }
2193
2194         if (nr_files) {
2195                 fa = flex_array_alloc(sizeof(info), nr_files,
2196                                         GFP_KERNEL);
2197                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2198                                                 GFP_KERNEL)) {
2199                         ret = -ENOMEM;
2200                         if (fa)
2201                                 flex_array_free(fa);
2202                         up_read(&mm->mmap_sem);
2203                         mmput(mm);
2204                         goto out_put_task;
2205                 }
2206                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2207                                 vma = vma->vm_next) {
2208                         if (!vma->vm_file)
2209                                 continue;
2210                         if (++pos <= ctx->pos)
2211                                 continue;
2212
2213                         info.start = vma->vm_start;
2214                         info.end = vma->vm_end;
2215                         info.mode = vma->vm_file->f_mode;
2216                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2217                                 BUG();
2218                 }
2219         }
2220         up_read(&mm->mmap_sem);
2221         mmput(mm);
2222
2223         for (i = 0; i < nr_files; i++) {
2224                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2225                 unsigned int len;
2226
2227                 p = flex_array_get(fa, i);
2228                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2229                 if (!proc_fill_cache(file, ctx,
2230                                       buf, len,
2231                                       proc_map_files_instantiate,
2232                                       task,
2233                                       (void *)(unsigned long)p->mode))
2234                         break;
2235                 ctx->pos++;
2236         }
2237         if (fa)
2238                 flex_array_free(fa);
2239
2240 out_put_task:
2241         put_task_struct(task);
2242 out:
2243         return ret;
2244 }
2245
2246 static const struct file_operations proc_map_files_operations = {
2247         .read           = generic_read_dir,
2248         .iterate_shared = proc_map_files_readdir,
2249         .llseek         = generic_file_llseek,
2250 };
2251
2252 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2253 struct timers_private {
2254         struct pid *pid;
2255         struct task_struct *task;
2256         struct sighand_struct *sighand;
2257         struct pid_namespace *ns;
2258         unsigned long flags;
2259 };
2260
2261 static void *timers_start(struct seq_file *m, loff_t *pos)
2262 {
2263         struct timers_private *tp = m->private;
2264
2265         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2266         if (!tp->task)
2267                 return ERR_PTR(-ESRCH);
2268
2269         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2270         if (!tp->sighand)
2271                 return ERR_PTR(-ESRCH);
2272
2273         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2274 }
2275
2276 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2277 {
2278         struct timers_private *tp = m->private;
2279         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2280 }
2281
2282 static void timers_stop(struct seq_file *m, void *v)
2283 {
2284         struct timers_private *tp = m->private;
2285
2286         if (tp->sighand) {
2287                 unlock_task_sighand(tp->task, &tp->flags);
2288                 tp->sighand = NULL;
2289         }
2290
2291         if (tp->task) {
2292                 put_task_struct(tp->task);
2293                 tp->task = NULL;
2294         }
2295 }
2296
2297 static int show_timer(struct seq_file *m, void *v)
2298 {
2299         struct k_itimer *timer;
2300         struct timers_private *tp = m->private;
2301         int notify;
2302         static const char * const nstr[] = {
2303                 [SIGEV_SIGNAL] = "signal",
2304                 [SIGEV_NONE] = "none",
2305                 [SIGEV_THREAD] = "thread",
2306         };
2307
2308         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2309         notify = timer->it_sigev_notify;
2310
2311         seq_printf(m, "ID: %d\n", timer->it_id);
2312         seq_printf(m, "signal: %d/%px\n",
2313                    timer->sigq->info.si_signo,
2314                    timer->sigq->info.si_value.sival_ptr);
2315         seq_printf(m, "notify: %s/%s.%d\n",
2316                    nstr[notify & ~SIGEV_THREAD_ID],
2317                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2318                    pid_nr_ns(timer->it_pid, tp->ns));
2319         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2320
2321         return 0;
2322 }
2323
2324 static const struct seq_operations proc_timers_seq_ops = {
2325         .start  = timers_start,
2326         .next   = timers_next,
2327         .stop   = timers_stop,
2328         .show   = show_timer,
2329 };
2330
2331 static int proc_timers_open(struct inode *inode, struct file *file)
2332 {
2333         struct timers_private *tp;
2334
2335         tp = __seq_open_private(file, &proc_timers_seq_ops,
2336                         sizeof(struct timers_private));
2337         if (!tp)
2338                 return -ENOMEM;
2339
2340         tp->pid = proc_pid(inode);
2341         tp->ns = proc_pid_ns(inode);
2342         return 0;
2343 }
2344
2345 static const struct file_operations proc_timers_operations = {
2346         .open           = proc_timers_open,
2347         .read           = seq_read,
2348         .llseek         = seq_lseek,
2349         .release        = seq_release_private,
2350 };
2351 #endif
2352
2353 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2354                                         size_t count, loff_t *offset)
2355 {
2356         struct inode *inode = file_inode(file);
2357         struct task_struct *p;
2358         u64 slack_ns;
2359         int err;
2360
2361         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2362         if (err < 0)
2363                 return err;
2364
2365         p = get_proc_task(inode);
2366         if (!p)
2367                 return -ESRCH;
2368
2369         if (p != current) {
2370                 if (!capable(CAP_SYS_NICE)) {
2371                         count = -EPERM;
2372                         goto out;
2373                 }
2374
2375                 err = security_task_setscheduler(p);
2376                 if (err) {
2377                         count = err;
2378                         goto out;
2379                 }
2380         }
2381
2382         task_lock(p);
2383         if (slack_ns == 0)
2384                 p->timer_slack_ns = p->default_timer_slack_ns;
2385         else
2386                 p->timer_slack_ns = slack_ns;
2387         task_unlock(p);
2388
2389 out:
2390         put_task_struct(p);
2391
2392         return count;
2393 }
2394
2395 static int timerslack_ns_show(struct seq_file *m, void *v)
2396 {
2397         struct inode *inode = m->private;
2398         struct task_struct *p;
2399         int err = 0;
2400
2401         p = get_proc_task(inode);
2402         if (!p)
2403                 return -ESRCH;
2404
2405         if (p != current) {
2406
2407                 if (!capable(CAP_SYS_NICE)) {
2408                         err = -EPERM;
2409                         goto out;
2410                 }
2411                 err = security_task_getscheduler(p);
2412                 if (err)
2413                         goto out;
2414         }
2415
2416         task_lock(p);
2417         seq_printf(m, "%llu\n", p->timer_slack_ns);
2418         task_unlock(p);
2419
2420 out:
2421         put_task_struct(p);
2422
2423         return err;
2424 }
2425
2426 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2427 {
2428         return single_open(filp, timerslack_ns_show, inode);
2429 }
2430
2431 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2432         .open           = timerslack_ns_open,
2433         .read           = seq_read,
2434         .write          = timerslack_ns_write,
2435         .llseek         = seq_lseek,
2436         .release        = single_release,
2437 };
2438
2439 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2440         struct task_struct *task, const void *ptr)
2441 {
2442         const struct pid_entry *p = ptr;
2443         struct inode *inode;
2444         struct proc_inode *ei;
2445
2446         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2447         if (!inode)
2448                 return ERR_PTR(-ENOENT);
2449
2450         ei = PROC_I(inode);
2451         if (S_ISDIR(inode->i_mode))
2452                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2453         if (p->iop)
2454                 inode->i_op = p->iop;
2455         if (p->fop)
2456                 inode->i_fop = p->fop;
2457         ei->op = p->op;
2458         pid_update_inode(task, inode);
2459         d_set_d_op(dentry, &pid_dentry_operations);
2460         return d_splice_alias(inode, dentry);
2461 }
2462
2463 static struct dentry *proc_pident_lookup(struct inode *dir, 
2464                                          struct dentry *dentry,
2465                                          const struct pid_entry *ents,
2466                                          unsigned int nents)
2467 {
2468         struct task_struct *task = get_proc_task(dir);
2469         const struct pid_entry *p, *last;
2470         struct dentry *res = ERR_PTR(-ENOENT);
2471
2472         if (!task)
2473                 goto out_no_task;
2474
2475         /*
2476          * Yes, it does not scale. And it should not. Don't add
2477          * new entries into /proc/<tgid>/ without very good reasons.
2478          */
2479         last = &ents[nents];
2480         for (p = ents; p < last; p++) {
2481                 if (p->len != dentry->d_name.len)
2482                         continue;
2483                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2484                         break;
2485         }
2486         if (p >= last)
2487                 goto out;
2488
2489         res = proc_pident_instantiate(dentry, task, p);
2490 out:
2491         put_task_struct(task);
2492 out_no_task:
2493         return res;
2494 }
2495
2496 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2497                 const struct pid_entry *ents, unsigned int nents)
2498 {
2499         struct task_struct *task = get_proc_task(file_inode(file));
2500         const struct pid_entry *p;
2501
2502         if (!task)
2503                 return -ENOENT;
2504
2505         if (!dir_emit_dots(file, ctx))
2506                 goto out;
2507
2508         if (ctx->pos >= nents + 2)
2509                 goto out;
2510
2511         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2512                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2513                                 proc_pident_instantiate, task, p))
2514                         break;
2515                 ctx->pos++;
2516         }
2517 out:
2518         put_task_struct(task);
2519         return 0;
2520 }
2521
2522 #ifdef CONFIG_SECURITY
2523 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2524                                   size_t count, loff_t *ppos)
2525 {
2526         struct inode * inode = file_inode(file);
2527         char *p = NULL;
2528         ssize_t length;
2529         struct task_struct *task = get_proc_task(inode);
2530
2531         if (!task)
2532                 return -ESRCH;
2533
2534         length = security_getprocattr(task,
2535                                       (char*)file->f_path.dentry->d_name.name,
2536                                       &p);
2537         put_task_struct(task);
2538         if (length > 0)
2539                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2540         kfree(p);
2541         return length;
2542 }
2543
2544 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2545                                    size_t count, loff_t *ppos)
2546 {
2547         struct inode * inode = file_inode(file);
2548         void *page;
2549         ssize_t length;
2550         struct task_struct *task = get_proc_task(inode);
2551
2552         length = -ESRCH;
2553         if (!task)
2554                 goto out_no_task;
2555
2556         /* A task may only write its own attributes. */
2557         length = -EACCES;
2558         if (current != task)
2559                 goto out;
2560
2561         if (count > PAGE_SIZE)
2562                 count = PAGE_SIZE;
2563
2564         /* No partial writes. */
2565         length = -EINVAL;
2566         if (*ppos != 0)
2567                 goto out;
2568
2569         page = memdup_user(buf, count);
2570         if (IS_ERR(page)) {
2571                 length = PTR_ERR(page);
2572                 goto out;
2573         }
2574
2575         /* Guard against adverse ptrace interaction */
2576         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2577         if (length < 0)
2578                 goto out_free;
2579
2580         length = security_setprocattr(file->f_path.dentry->d_name.name,
2581                                       page, count);
2582         mutex_unlock(&current->signal->cred_guard_mutex);
2583 out_free:
2584         kfree(page);
2585 out:
2586         put_task_struct(task);
2587 out_no_task:
2588         return length;
2589 }
2590
2591 static const struct file_operations proc_pid_attr_operations = {
2592         .read           = proc_pid_attr_read,
2593         .write          = proc_pid_attr_write,
2594         .llseek         = generic_file_llseek,
2595 };
2596
2597 static const struct pid_entry attr_dir_stuff[] = {
2598         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2599         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2600         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2601         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2602         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2603         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2604 };
2605
2606 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2607 {
2608         return proc_pident_readdir(file, ctx, 
2609                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2610 }
2611
2612 static const struct file_operations proc_attr_dir_operations = {
2613         .read           = generic_read_dir,
2614         .iterate_shared = proc_attr_dir_readdir,
2615         .llseek         = generic_file_llseek,
2616 };
2617
2618 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2619                                 struct dentry *dentry, unsigned int flags)
2620 {
2621         return proc_pident_lookup(dir, dentry,
2622                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2623 }
2624
2625 static const struct inode_operations proc_attr_dir_inode_operations = {
2626         .lookup         = proc_attr_dir_lookup,
2627         .getattr        = pid_getattr,
2628         .setattr        = proc_setattr,
2629 };
2630
2631 #endif
2632
2633 #ifdef CONFIG_ELF_CORE
2634 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2635                                          size_t count, loff_t *ppos)
2636 {
2637         struct task_struct *task = get_proc_task(file_inode(file));
2638         struct mm_struct *mm;
2639         char buffer[PROC_NUMBUF];
2640         size_t len;
2641         int ret;
2642
2643         if (!task)
2644                 return -ESRCH;
2645
2646         ret = 0;
2647         mm = get_task_mm(task);
2648         if (mm) {
2649                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2650                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2651                                 MMF_DUMP_FILTER_SHIFT));
2652                 mmput(mm);
2653                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2654         }
2655
2656         put_task_struct(task);
2657
2658         return ret;
2659 }
2660
2661 static ssize_t proc_coredump_filter_write(struct file *file,
2662                                           const char __user *buf,
2663                                           size_t count,
2664                                           loff_t *ppos)
2665 {
2666         struct task_struct *task;
2667         struct mm_struct *mm;
2668         unsigned int val;
2669         int ret;
2670         int i;
2671         unsigned long mask;
2672
2673         ret = kstrtouint_from_user(buf, count, 0, &val);
2674         if (ret < 0)
2675                 return ret;
2676
2677         ret = -ESRCH;
2678         task = get_proc_task(file_inode(file));
2679         if (!task)
2680                 goto out_no_task;
2681
2682         mm = get_task_mm(task);
2683         if (!mm)
2684                 goto out_no_mm;
2685         ret = 0;
2686
2687         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2688                 if (val & mask)
2689                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2690                 else
2691                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2692         }
2693
2694         mmput(mm);
2695  out_no_mm:
2696         put_task_struct(task);
2697  out_no_task:
2698         if (ret < 0)
2699                 return ret;
2700         return count;
2701 }
2702
2703 static const struct file_operations proc_coredump_filter_operations = {
2704         .read           = proc_coredump_filter_read,
2705         .write          = proc_coredump_filter_write,
2706         .llseek         = generic_file_llseek,
2707 };
2708 #endif
2709
2710 #ifdef CONFIG_TASK_IO_ACCOUNTING
2711 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2712 {
2713         struct task_io_accounting acct = task->ioac;
2714         unsigned long flags;
2715         int result;
2716
2717         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2718         if (result)
2719                 return result;
2720
2721         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2722                 result = -EACCES;
2723                 goto out_unlock;
2724         }
2725
2726         if (whole && lock_task_sighand(task, &flags)) {
2727                 struct task_struct *t = task;
2728
2729                 task_io_accounting_add(&acct, &task->signal->ioac);
2730                 while_each_thread(task, t)
2731                         task_io_accounting_add(&acct, &t->ioac);
2732
2733                 unlock_task_sighand(task, &flags);
2734         }
2735         seq_printf(m,
2736                    "rchar: %llu\n"
2737                    "wchar: %llu\n"
2738                    "syscr: %llu\n"
2739                    "syscw: %llu\n"
2740                    "read_bytes: %llu\n"
2741                    "write_bytes: %llu\n"
2742                    "cancelled_write_bytes: %llu\n",
2743                    (unsigned long long)acct.rchar,
2744                    (unsigned long long)acct.wchar,
2745                    (unsigned long long)acct.syscr,
2746                    (unsigned long long)acct.syscw,
2747                    (unsigned long long)acct.read_bytes,
2748                    (unsigned long long)acct.write_bytes,
2749                    (unsigned long long)acct.cancelled_write_bytes);
2750         result = 0;
2751
2752 out_unlock:
2753         mutex_unlock(&task->signal->cred_guard_mutex);
2754         return result;
2755 }
2756
2757 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2758                                   struct pid *pid, struct task_struct *task)
2759 {
2760         return do_io_accounting(task, m, 0);
2761 }
2762
2763 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2764                                    struct pid *pid, struct task_struct *task)
2765 {
2766         return do_io_accounting(task, m, 1);
2767 }
2768 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2769
2770 #ifdef CONFIG_USER_NS
2771 static int proc_id_map_open(struct inode *inode, struct file *file,
2772         const struct seq_operations *seq_ops)
2773 {
2774         struct user_namespace *ns = NULL;
2775         struct task_struct *task;
2776         struct seq_file *seq;
2777         int ret = -EINVAL;
2778
2779         task = get_proc_task(inode);
2780         if (task) {
2781                 rcu_read_lock();
2782                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2783                 rcu_read_unlock();
2784                 put_task_struct(task);
2785         }
2786         if (!ns)
2787                 goto err;
2788
2789         ret = seq_open(file, seq_ops);
2790         if (ret)
2791                 goto err_put_ns;
2792
2793         seq = file->private_data;
2794         seq->private = ns;
2795
2796         return 0;
2797 err_put_ns:
2798         put_user_ns(ns);
2799 err:
2800         return ret;
2801 }
2802
2803 static int proc_id_map_release(struct inode *inode, struct file *file)
2804 {
2805         struct seq_file *seq = file->private_data;
2806         struct user_namespace *ns = seq->private;
2807         put_user_ns(ns);
2808         return seq_release(inode, file);
2809 }
2810
2811 static int proc_uid_map_open(struct inode *inode, struct file *file)
2812 {
2813         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2814 }
2815
2816 static int proc_gid_map_open(struct inode *inode, struct file *file)
2817 {
2818         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2819 }
2820
2821 static int proc_projid_map_open(struct inode *inode, struct file *file)
2822 {
2823         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2824 }
2825
2826 static const struct file_operations proc_uid_map_operations = {
2827         .open           = proc_uid_map_open,
2828         .write          = proc_uid_map_write,
2829         .read           = seq_read,
2830         .llseek         = seq_lseek,
2831         .release        = proc_id_map_release,
2832 };
2833
2834 static const struct file_operations proc_gid_map_operations = {
2835         .open           = proc_gid_map_open,
2836         .write          = proc_gid_map_write,
2837         .read           = seq_read,
2838         .llseek         = seq_lseek,
2839         .release        = proc_id_map_release,
2840 };
2841
2842 static const struct file_operations proc_projid_map_operations = {
2843         .open           = proc_projid_map_open,
2844         .write          = proc_projid_map_write,
2845         .read           = seq_read,
2846         .llseek         = seq_lseek,
2847         .release        = proc_id_map_release,
2848 };
2849
2850 static int proc_setgroups_open(struct inode *inode, struct file *file)
2851 {
2852         struct user_namespace *ns = NULL;
2853         struct task_struct *task;
2854         int ret;
2855
2856         ret = -ESRCH;
2857         task = get_proc_task(inode);
2858         if (task) {
2859                 rcu_read_lock();
2860                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2861                 rcu_read_unlock();
2862                 put_task_struct(task);
2863         }
2864         if (!ns)
2865                 goto err;
2866
2867         if (file->f_mode & FMODE_WRITE) {
2868                 ret = -EACCES;
2869                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2870                         goto err_put_ns;
2871         }
2872
2873         ret = single_open(file, &proc_setgroups_show, ns);
2874         if (ret)
2875                 goto err_put_ns;
2876
2877         return 0;
2878 err_put_ns:
2879         put_user_ns(ns);
2880 err:
2881         return ret;
2882 }
2883
2884 static int proc_setgroups_release(struct inode *inode, struct file *file)
2885 {
2886         struct seq_file *seq = file->private_data;
2887         struct user_namespace *ns = seq->private;
2888         int ret = single_release(inode, file);
2889         put_user_ns(ns);
2890         return ret;
2891 }
2892
2893 static const struct file_operations proc_setgroups_operations = {
2894         .open           = proc_setgroups_open,
2895         .write          = proc_setgroups_write,
2896         .read           = seq_read,
2897         .llseek         = seq_lseek,
2898         .release        = proc_setgroups_release,
2899 };
2900 #endif /* CONFIG_USER_NS */
2901
2902 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2903                                 struct pid *pid, struct task_struct *task)
2904 {
2905         int err = lock_trace(task);
2906         if (!err) {
2907                 seq_printf(m, "%08x\n", task->personality);
2908                 unlock_trace(task);
2909         }
2910         return err;
2911 }
2912
2913 #ifdef CONFIG_LIVEPATCH
2914 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2915                                 struct pid *pid, struct task_struct *task)
2916 {
2917         seq_printf(m, "%d\n", task->patch_state);
2918         return 0;
2919 }
2920 #endif /* CONFIG_LIVEPATCH */
2921
2922 /*
2923  * Thread groups
2924  */
2925 static const struct file_operations proc_task_operations;
2926 static const struct inode_operations proc_task_inode_operations;
2927
2928 static const struct pid_entry tgid_base_stuff[] = {
2929         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2930         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2931         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2932         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2933         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2934 #ifdef CONFIG_NET
2935         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2936 #endif
2937         REG("environ",    S_IRUSR, proc_environ_operations),
2938         REG("auxv",       S_IRUSR, proc_auxv_operations),
2939         ONE("status",     S_IRUGO, proc_pid_status),
2940         ONE("personality", S_IRUSR, proc_pid_personality),
2941         ONE("limits",     S_IRUGO, proc_pid_limits),
2942 #ifdef CONFIG_SCHED_DEBUG
2943         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2944 #endif
2945 #ifdef CONFIG_SCHED_AUTOGROUP
2946         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2947 #endif
2948         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2949 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2950         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2951 #endif
2952         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2953         ONE("stat",       S_IRUGO, proc_tgid_stat),
2954         ONE("statm",      S_IRUGO, proc_pid_statm),
2955         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2956 #ifdef CONFIG_NUMA
2957         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2958 #endif
2959         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2960         LNK("cwd",        proc_cwd_link),
2961         LNK("root",       proc_root_link),
2962         LNK("exe",        proc_exe_link),
2963         REG("mounts",     S_IRUGO, proc_mounts_operations),
2964         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2965         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2966 #ifdef CONFIG_PROC_PAGE_MONITOR
2967         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2968         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2969         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2970         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2971 #endif
2972 #ifdef CONFIG_SECURITY
2973         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2974 #endif
2975 #ifdef CONFIG_KALLSYMS
2976         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2977 #endif
2978 #ifdef CONFIG_STACKTRACE
2979         ONE("stack",      S_IRUSR, proc_pid_stack),
2980 #endif
2981 #ifdef CONFIG_SCHED_INFO
2982         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2983 #endif
2984 #ifdef CONFIG_LATENCYTOP
2985         REG("latency",  S_IRUGO, proc_lstats_operations),
2986 #endif
2987 #ifdef CONFIG_PROC_PID_CPUSET
2988         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2989 #endif
2990 #ifdef CONFIG_CGROUPS
2991         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2992 #endif
2993         ONE("oom_score",  S_IRUGO, proc_oom_score),
2994         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2995         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2996 #ifdef CONFIG_AUDITSYSCALL
2997         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2998         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2999 #endif
3000 #ifdef CONFIG_FAULT_INJECTION
3001         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3002         REG("fail-nth", 0644, proc_fail_nth_operations),
3003 #endif
3004 #ifdef CONFIG_ELF_CORE
3005         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3006 #endif
3007 #ifdef CONFIG_TASK_IO_ACCOUNTING
3008         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3009 #endif
3010 #ifdef CONFIG_USER_NS
3011         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3012         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3013         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3014         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3015 #endif
3016 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3017         REG("timers",     S_IRUGO, proc_timers_operations),
3018 #endif
3019         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3020 #ifdef CONFIG_LIVEPATCH
3021         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3022 #endif
3023 };
3024
3025 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3026 {
3027         return proc_pident_readdir(file, ctx,
3028                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3029 }
3030
3031 static const struct file_operations proc_tgid_base_operations = {
3032         .read           = generic_read_dir,
3033         .iterate_shared = proc_tgid_base_readdir,
3034         .llseek         = generic_file_llseek,
3035 };
3036
3037 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3038 {
3039         return proc_pident_lookup(dir, dentry,
3040                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3041 }
3042
3043 static const struct inode_operations proc_tgid_base_inode_operations = {
3044         .lookup         = proc_tgid_base_lookup,
3045         .getattr        = pid_getattr,
3046         .setattr        = proc_setattr,
3047         .permission     = proc_pid_permission,
3048 };
3049
3050 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3051 {
3052         struct dentry *dentry, *leader, *dir;
3053         char buf[10 + 1];
3054         struct qstr name;
3055
3056         name.name = buf;
3057         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3058         /* no ->d_hash() rejects on procfs */
3059         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3060         if (dentry) {
3061                 d_invalidate(dentry);
3062                 dput(dentry);
3063         }
3064
3065         if (pid == tgid)
3066                 return;
3067
3068         name.name = buf;
3069         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3070         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3071         if (!leader)
3072                 goto out;
3073
3074         name.name = "task";
3075         name.len = strlen(name.name);
3076         dir = d_hash_and_lookup(leader, &name);
3077         if (!dir)
3078                 goto out_put_leader;
3079
3080         name.name = buf;
3081         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3082         dentry = d_hash_and_lookup(dir, &name);
3083         if (dentry) {
3084                 d_invalidate(dentry);
3085                 dput(dentry);
3086         }
3087
3088         dput(dir);
3089 out_put_leader:
3090         dput(leader);
3091 out:
3092         return;
3093 }
3094
3095 /**
3096  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3097  * @task: task that should be flushed.
3098  *
3099  * When flushing dentries from proc, one needs to flush them from global
3100  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3101  * in. This call is supposed to do all of this job.
3102  *
3103  * Looks in the dcache for
3104  * /proc/@pid
3105  * /proc/@tgid/task/@pid
3106  * if either directory is present flushes it and all of it'ts children
3107  * from the dcache.
3108  *
3109  * It is safe and reasonable to cache /proc entries for a task until
3110  * that task exits.  After that they just clog up the dcache with
3111  * useless entries, possibly causing useful dcache entries to be
3112  * flushed instead.  This routine is proved to flush those useless
3113  * dcache entries at process exit time.
3114  *
3115  * NOTE: This routine is just an optimization so it does not guarantee
3116  *       that no dcache entries will exist at process exit time it
3117  *       just makes it very unlikely that any will persist.
3118  */
3119
3120 void proc_flush_task(struct task_struct *task)
3121 {
3122         int i;
3123         struct pid *pid, *tgid;
3124         struct upid *upid;
3125
3126         pid = task_pid(task);
3127         tgid = task_tgid(task);
3128
3129         for (i = 0; i <= pid->level; i++) {
3130                 upid = &pid->numbers[i];
3131                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3132                                         tgid->numbers[i].nr);
3133         }
3134 }
3135
3136 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3137                                    struct task_struct *task, const void *ptr)
3138 {
3139         struct inode *inode;
3140
3141         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3142         if (!inode)
3143                 return ERR_PTR(-ENOENT);
3144
3145         inode->i_op = &proc_tgid_base_inode_operations;
3146         inode->i_fop = &proc_tgid_base_operations;
3147         inode->i_flags|=S_IMMUTABLE;
3148
3149         set_nlink(inode, nlink_tgid);
3150         pid_update_inode(task, inode);
3151
3152         d_set_d_op(dentry, &pid_dentry_operations);
3153         return d_splice_alias(inode, dentry);
3154 }
3155
3156 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3157 {
3158         struct task_struct *task;
3159         unsigned tgid;
3160         struct pid_namespace *ns;
3161         struct dentry *result = ERR_PTR(-ENOENT);
3162
3163         tgid = name_to_int(&dentry->d_name);
3164         if (tgid == ~0U)
3165                 goto out;
3166
3167         ns = dentry->d_sb->s_fs_info;
3168         rcu_read_lock();
3169         task = find_task_by_pid_ns(tgid, ns);
3170         if (task)
3171                 get_task_struct(task);
3172         rcu_read_unlock();
3173         if (!task)
3174                 goto out;
3175
3176         result = proc_pid_instantiate(dentry, task, NULL);
3177         put_task_struct(task);
3178 out:
3179         return result;
3180 }
3181
3182 /*
3183  * Find the first task with tgid >= tgid
3184  *
3185  */
3186 struct tgid_iter {
3187         unsigned int tgid;
3188         struct task_struct *task;
3189 };
3190 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3191 {
3192         struct pid *pid;
3193
3194         if (iter.task)
3195                 put_task_struct(iter.task);
3196         rcu_read_lock();
3197 retry:
3198         iter.task = NULL;
3199         pid = find_ge_pid(iter.tgid, ns);
3200         if (pid) {
3201                 iter.tgid = pid_nr_ns(pid, ns);
3202                 iter.task = pid_task(pid, PIDTYPE_PID);
3203                 /* What we to know is if the pid we have find is the
3204                  * pid of a thread_group_leader.  Testing for task
3205                  * being a thread_group_leader is the obvious thing
3206                  * todo but there is a window when it fails, due to
3207                  * the pid transfer logic in de_thread.
3208                  *
3209                  * So we perform the straight forward test of seeing
3210                  * if the pid we have found is the pid of a thread
3211                  * group leader, and don't worry if the task we have
3212                  * found doesn't happen to be a thread group leader.
3213                  * As we don't care in the case of readdir.
3214                  */
3215                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3216                         iter.tgid += 1;
3217                         goto retry;
3218                 }
3219                 get_task_struct(iter.task);
3220         }
3221         rcu_read_unlock();
3222         return iter;
3223 }
3224
3225 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3226
3227 /* for the /proc/ directory itself, after non-process stuff has been done */
3228 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3229 {
3230         struct tgid_iter iter;
3231         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3232         loff_t pos = ctx->pos;
3233
3234         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3235                 return 0;
3236
3237         if (pos == TGID_OFFSET - 2) {
3238                 struct inode *inode = d_inode(ns->proc_self);
3239                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3240                         return 0;
3241                 ctx->pos = pos = pos + 1;
3242         }
3243         if (pos == TGID_OFFSET - 1) {
3244                 struct inode *inode = d_inode(ns->proc_thread_self);
3245                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3246                         return 0;
3247                 ctx->pos = pos = pos + 1;
3248         }
3249         iter.tgid = pos - TGID_OFFSET;
3250         iter.task = NULL;
3251         for (iter = next_tgid(ns, iter);
3252              iter.task;
3253              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3254                 char name[10 + 1];
3255                 int len;
3256
3257                 cond_resched();
3258                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3259                         continue;
3260
3261                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3262                 ctx->pos = iter.tgid + TGID_OFFSET;
3263                 if (!proc_fill_cache(file, ctx, name, len,
3264                                      proc_pid_instantiate, iter.task, NULL)) {
3265                         put_task_struct(iter.task);
3266                         return 0;
3267                 }
3268         }
3269         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3270         return 0;
3271 }
3272
3273 /*
3274  * proc_tid_comm_permission is a special permission function exclusively
3275  * used for the node /proc/<pid>/task/<tid>/comm.
3276  * It bypasses generic permission checks in the case where a task of the same
3277  * task group attempts to access the node.
3278  * The rationale behind this is that glibc and bionic access this node for
3279  * cross thread naming (pthread_set/getname_np(!self)). However, if
3280  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3281  * which locks out the cross thread naming implementation.
3282  * This function makes sure that the node is always accessible for members of
3283  * same thread group.
3284  */
3285 static int proc_tid_comm_permission(struct inode *inode, int mask)
3286 {
3287         bool is_same_tgroup;
3288         struct task_struct *task;
3289
3290         task = get_proc_task(inode);
3291         if (!task)
3292                 return -ESRCH;
3293         is_same_tgroup = same_thread_group(current, task);
3294         put_task_struct(task);
3295
3296         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3297                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3298                  * read or written by the members of the corresponding
3299                  * thread group.
3300                  */
3301                 return 0;
3302         }
3303
3304         return generic_permission(inode, mask);
3305 }
3306
3307 static const struct inode_operations proc_tid_comm_inode_operations = {
3308                 .permission = proc_tid_comm_permission,
3309 };
3310
3311 /*
3312  * Tasks
3313  */
3314 static const struct pid_entry tid_base_stuff[] = {
3315         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3316         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3317         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3318 #ifdef CONFIG_NET
3319         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3320 #endif
3321         REG("environ",   S_IRUSR, proc_environ_operations),
3322         REG("auxv",      S_IRUSR, proc_auxv_operations),
3323         ONE("status",    S_IRUGO, proc_pid_status),
3324         ONE("personality", S_IRUSR, proc_pid_personality),
3325         ONE("limits",    S_IRUGO, proc_pid_limits),
3326 #ifdef CONFIG_SCHED_DEBUG
3327         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3328 #endif
3329         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3330                          &proc_tid_comm_inode_operations,
3331                          &proc_pid_set_comm_operations, {}),
3332 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3333         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3334 #endif
3335         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3336         ONE("stat",      S_IRUGO, proc_tid_stat),
3337         ONE("statm",     S_IRUGO, proc_pid_statm),
3338         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3339 #ifdef CONFIG_PROC_CHILDREN
3340         REG("children",  S_IRUGO, proc_tid_children_operations),
3341 #endif
3342 #ifdef CONFIG_NUMA
3343         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3344 #endif
3345         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3346         LNK("cwd",       proc_cwd_link),
3347         LNK("root",      proc_root_link),
3348         LNK("exe",       proc_exe_link),
3349         REG("mounts",    S_IRUGO, proc_mounts_operations),
3350         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3351 #ifdef CONFIG_PROC_PAGE_MONITOR
3352         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3353         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3354         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3355         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3356 #endif
3357 #ifdef CONFIG_SECURITY
3358         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3359 #endif
3360 #ifdef CONFIG_KALLSYMS
3361         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3362 #endif
3363 #ifdef CONFIG_STACKTRACE
3364         ONE("stack",      S_IRUSR, proc_pid_stack),
3365 #endif
3366 #ifdef CONFIG_SCHED_INFO
3367         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3368 #endif
3369 #ifdef CONFIG_LATENCYTOP
3370         REG("latency",  S_IRUGO, proc_lstats_operations),
3371 #endif
3372 #ifdef CONFIG_PROC_PID_CPUSET
3373         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3374 #endif
3375 #ifdef CONFIG_CGROUPS
3376         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3377 #endif
3378         ONE("oom_score", S_IRUGO, proc_oom_score),
3379         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3380         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3381 #ifdef CONFIG_AUDITSYSCALL
3382         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3383         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3384 #endif
3385 #ifdef CONFIG_FAULT_INJECTION
3386         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3387         REG("fail-nth", 0644, proc_fail_nth_operations),
3388 #endif
3389 #ifdef CONFIG_TASK_IO_ACCOUNTING
3390         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3391 #endif
3392 #ifdef CONFIG_USER_NS
3393         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3394         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3395         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3396         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3397 #endif
3398 #ifdef CONFIG_LIVEPATCH
3399         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3400 #endif
3401 };
3402
3403 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3404 {
3405         return proc_pident_readdir(file, ctx,
3406                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3407 }
3408
3409 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3410 {
3411         return proc_pident_lookup(dir, dentry,
3412                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3413 }
3414
3415 static const struct file_operations proc_tid_base_operations = {
3416         .read           = generic_read_dir,
3417         .iterate_shared = proc_tid_base_readdir,
3418         .llseek         = generic_file_llseek,
3419 };
3420
3421 static const struct inode_operations proc_tid_base_inode_operations = {
3422         .lookup         = proc_tid_base_lookup,
3423         .getattr        = pid_getattr,
3424         .setattr        = proc_setattr,
3425 };
3426
3427 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3428         struct task_struct *task, const void *ptr)
3429 {
3430         struct inode *inode;
3431         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3432         if (!inode)
3433                 return ERR_PTR(-ENOENT);
3434
3435         inode->i_op = &proc_tid_base_inode_operations;
3436         inode->i_fop = &proc_tid_base_operations;
3437         inode->i_flags |= S_IMMUTABLE;
3438
3439         set_nlink(inode, nlink_tid);
3440         pid_update_inode(task, inode);
3441
3442         d_set_d_op(dentry, &pid_dentry_operations);
3443         return d_splice_alias(inode, dentry);
3444 }
3445
3446 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3447 {
3448         struct task_struct *task;
3449         struct task_struct *leader = get_proc_task(dir);
3450         unsigned tid;
3451         struct pid_namespace *ns;
3452         struct dentry *result = ERR_PTR(-ENOENT);
3453
3454         if (!leader)
3455                 goto out_no_task;
3456
3457         tid = name_to_int(&dentry->d_name);
3458         if (tid == ~0U)
3459                 goto out;
3460
3461         ns = dentry->d_sb->s_fs_info;
3462         rcu_read_lock();
3463         task = find_task_by_pid_ns(tid, ns);
3464         if (task)
3465                 get_task_struct(task);
3466         rcu_read_unlock();
3467         if (!task)
3468                 goto out;
3469         if (!same_thread_group(leader, task))
3470                 goto out_drop_task;
3471
3472         result = proc_task_instantiate(dentry, task, NULL);
3473 out_drop_task:
3474         put_task_struct(task);
3475 out:
3476         put_task_struct(leader);
3477 out_no_task:
3478         return result;
3479 }
3480
3481 /*
3482  * Find the first tid of a thread group to return to user space.
3483  *
3484  * Usually this is just the thread group leader, but if the users
3485  * buffer was too small or there was a seek into the middle of the
3486  * directory we have more work todo.
3487  *
3488  * In the case of a short read we start with find_task_by_pid.
3489  *
3490  * In the case of a seek we start with the leader and walk nr
3491  * threads past it.
3492  */
3493 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3494                                         struct pid_namespace *ns)
3495 {
3496         struct task_struct *pos, *task;
3497         unsigned long nr = f_pos;
3498
3499         if (nr != f_pos)        /* 32bit overflow? */
3500                 return NULL;
3501
3502         rcu_read_lock();
3503         task = pid_task(pid, PIDTYPE_PID);
3504         if (!task)
3505                 goto fail;
3506
3507         /* Attempt to start with the tid of a thread */
3508         if (tid && nr) {
3509                 pos = find_task_by_pid_ns(tid, ns);
3510                 if (pos && same_thread_group(pos, task))
3511                         goto found;
3512         }
3513
3514         /* If nr exceeds the number of threads there is nothing todo */
3515         if (nr >= get_nr_threads(task))
3516                 goto fail;
3517
3518         /* If we haven't found our starting place yet start
3519          * with the leader and walk nr threads forward.
3520          */
3521         pos = task = task->group_leader;
3522         do {
3523                 if (!nr--)
3524                         goto found;
3525         } while_each_thread(task, pos);
3526 fail:
3527         pos = NULL;
3528         goto out;
3529 found:
3530         get_task_struct(pos);
3531 out:
3532         rcu_read_unlock();
3533         return pos;
3534 }
3535
3536 /*
3537  * Find the next thread in the thread list.
3538  * Return NULL if there is an error or no next thread.
3539  *
3540  * The reference to the input task_struct is released.
3541  */
3542 static struct task_struct *next_tid(struct task_struct *start)
3543 {
3544         struct task_struct *pos = NULL;
3545         rcu_read_lock();
3546         if (pid_alive(start)) {
3547                 pos = next_thread(start);
3548                 if (thread_group_leader(pos))
3549                         pos = NULL;
3550                 else
3551                         get_task_struct(pos);
3552         }
3553         rcu_read_unlock();
3554         put_task_struct(start);
3555         return pos;
3556 }
3557
3558 /* for the /proc/TGID/task/ directories */
3559 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3560 {
3561         struct inode *inode = file_inode(file);
3562         struct task_struct *task;
3563         struct pid_namespace *ns;
3564         int tid;
3565
3566         if (proc_inode_is_dead(inode))
3567                 return -ENOENT;
3568
3569         if (!dir_emit_dots(file, ctx))
3570                 return 0;
3571
3572         /* f_version caches the tgid value that the last readdir call couldn't
3573          * return. lseek aka telldir automagically resets f_version to 0.
3574          */
3575         ns = proc_pid_ns(inode);
3576         tid = (int)file->f_version;
3577         file->f_version = 0;
3578         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3579              task;
3580              task = next_tid(task), ctx->pos++) {
3581                 char name[10 + 1];
3582                 int len;
3583                 tid = task_pid_nr_ns(task, ns);
3584                 len = snprintf(name, sizeof(name), "%u", tid);
3585                 if (!proc_fill_cache(file, ctx, name, len,
3586                                 proc_task_instantiate, task, NULL)) {
3587                         /* returning this tgid failed, save it as the first
3588                          * pid for the next readir call */
3589                         file->f_version = (u64)tid;
3590                         put_task_struct(task);
3591                         break;
3592                 }
3593         }
3594
3595         return 0;
3596 }
3597
3598 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3599                              u32 request_mask, unsigned int query_flags)
3600 {
3601         struct inode *inode = d_inode(path->dentry);
3602         struct task_struct *p = get_proc_task(inode);
3603         generic_fillattr(inode, stat);
3604
3605         if (p) {
3606                 stat->nlink += get_nr_threads(p);
3607                 put_task_struct(p);
3608         }
3609
3610         return 0;
3611 }
3612
3613 static const struct inode_operations proc_task_inode_operations = {
3614         .lookup         = proc_task_lookup,
3615         .getattr        = proc_task_getattr,
3616         .setattr        = proc_setattr,
3617         .permission     = proc_pid_permission,
3618 };
3619
3620 static const struct file_operations proc_task_operations = {
3621         .read           = generic_read_dir,
3622         .iterate_shared = proc_task_readdir,
3623         .llseek         = generic_file_llseek,
3624 };
3625
3626 void __init set_proc_pid_nlink(void)
3627 {
3628         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3629         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3630 }