workqueue: move function definitions within CONFIG_SMP block
[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, 0);
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, 0);
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, 0);
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 = inode->i_sb->s_fs_info;
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;
737         struct pid *pid;
738         struct task_struct *task;
739         int ret;
740
741         ns = inode->i_sb->s_fs_info;
742         pid = proc_pid(inode);
743         task = get_pid_task(pid, PIDTYPE_PID);
744         if (!task)
745                 return -ESRCH;
746
747         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
748
749         put_task_struct(task);
750         return ret;
751 }
752
753 static int proc_single_open(struct inode *inode, struct file *filp)
754 {
755         return single_open(filp, proc_single_show, inode);
756 }
757
758 static const struct file_operations proc_single_file_operations = {
759         .open           = proc_single_open,
760         .read           = seq_read,
761         .llseek         = seq_lseek,
762         .release        = single_release,
763 };
764
765
766 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
767 {
768         struct task_struct *task = get_proc_task(inode);
769         struct mm_struct *mm = ERR_PTR(-ESRCH);
770
771         if (task) {
772                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
773                 put_task_struct(task);
774
775                 if (!IS_ERR_OR_NULL(mm)) {
776                         /* ensure this mm_struct can't be freed */
777                         mmgrab(mm);
778                         /* but do not pin its memory */
779                         mmput(mm);
780                 }
781         }
782
783         return mm;
784 }
785
786 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
787 {
788         struct mm_struct *mm = proc_mem_open(inode, mode);
789
790         if (IS_ERR(mm))
791                 return PTR_ERR(mm);
792
793         file->private_data = mm;
794         return 0;
795 }
796
797 static int mem_open(struct inode *inode, struct file *file)
798 {
799         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
800
801         /* OK to pass negative loff_t, we can catch out-of-range */
802         file->f_mode |= FMODE_UNSIGNED_OFFSET;
803
804         return ret;
805 }
806
807 static ssize_t mem_rw(struct file *file, char __user *buf,
808                         size_t count, loff_t *ppos, int write)
809 {
810         struct mm_struct *mm = file->private_data;
811         unsigned long addr = *ppos;
812         ssize_t copied;
813         char *page;
814         unsigned int flags;
815
816         if (!mm)
817                 return 0;
818
819         page = (char *)__get_free_page(GFP_KERNEL);
820         if (!page)
821                 return -ENOMEM;
822
823         copied = 0;
824         if (!mmget_not_zero(mm))
825                 goto free;
826
827         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
828
829         while (count > 0) {
830                 int this_len = min_t(int, count, PAGE_SIZE);
831
832                 if (write && copy_from_user(page, buf, this_len)) {
833                         copied = -EFAULT;
834                         break;
835                 }
836
837                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
838                 if (!this_len) {
839                         if (!copied)
840                                 copied = -EIO;
841                         break;
842                 }
843
844                 if (!write && copy_to_user(buf, page, this_len)) {
845                         copied = -EFAULT;
846                         break;
847                 }
848
849                 buf += this_len;
850                 addr += this_len;
851                 copied += this_len;
852                 count -= this_len;
853         }
854         *ppos = addr;
855
856         mmput(mm);
857 free:
858         free_page((unsigned long) page);
859         return copied;
860 }
861
862 static ssize_t mem_read(struct file *file, char __user *buf,
863                         size_t count, loff_t *ppos)
864 {
865         return mem_rw(file, buf, count, ppos, 0);
866 }
867
868 static ssize_t mem_write(struct file *file, const char __user *buf,
869                          size_t count, loff_t *ppos)
870 {
871         return mem_rw(file, (char __user*)buf, count, ppos, 1);
872 }
873
874 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
875 {
876         switch (orig) {
877         case 0:
878                 file->f_pos = offset;
879                 break;
880         case 1:
881                 file->f_pos += offset;
882                 break;
883         default:
884                 return -EINVAL;
885         }
886         force_successful_syscall_return();
887         return file->f_pos;
888 }
889
890 static int mem_release(struct inode *inode, struct file *file)
891 {
892         struct mm_struct *mm = file->private_data;
893         if (mm)
894                 mmdrop(mm);
895         return 0;
896 }
897
898 static const struct file_operations proc_mem_operations = {
899         .llseek         = mem_lseek,
900         .read           = mem_read,
901         .write          = mem_write,
902         .open           = mem_open,
903         .release        = mem_release,
904 };
905
906 static int environ_open(struct inode *inode, struct file *file)
907 {
908         return __mem_open(inode, file, PTRACE_MODE_READ);
909 }
910
911 static ssize_t environ_read(struct file *file, char __user *buf,
912                         size_t count, loff_t *ppos)
913 {
914         char *page;
915         unsigned long src = *ppos;
916         int ret = 0;
917         struct mm_struct *mm = file->private_data;
918         unsigned long env_start, env_end;
919
920         /* Ensure the process spawned far enough to have an environment. */
921         if (!mm || !mm->env_end)
922                 return 0;
923
924         page = (char *)__get_free_page(GFP_KERNEL);
925         if (!page)
926                 return -ENOMEM;
927
928         ret = 0;
929         if (!mmget_not_zero(mm))
930                 goto free;
931
932         down_read(&mm->mmap_sem);
933         env_start = mm->env_start;
934         env_end = mm->env_end;
935         up_read(&mm->mmap_sem);
936
937         while (count > 0) {
938                 size_t this_len, max_len;
939                 int retval;
940
941                 if (src >= (env_end - env_start))
942                         break;
943
944                 this_len = env_end - (env_start + src);
945
946                 max_len = min_t(size_t, PAGE_SIZE, count);
947                 this_len = min(max_len, this_len);
948
949                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
950
951                 if (retval <= 0) {
952                         ret = retval;
953                         break;
954                 }
955
956                 if (copy_to_user(buf, page, retval)) {
957                         ret = -EFAULT;
958                         break;
959                 }
960
961                 ret += retval;
962                 src += retval;
963                 buf += retval;
964                 count -= retval;
965         }
966         *ppos = src;
967         mmput(mm);
968
969 free:
970         free_page((unsigned long) page);
971         return ret;
972 }
973
974 static const struct file_operations proc_environ_operations = {
975         .open           = environ_open,
976         .read           = environ_read,
977         .llseek         = generic_file_llseek,
978         .release        = mem_release,
979 };
980
981 static int auxv_open(struct inode *inode, struct file *file)
982 {
983         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
984 }
985
986 static ssize_t auxv_read(struct file *file, char __user *buf,
987                         size_t count, loff_t *ppos)
988 {
989         struct mm_struct *mm = file->private_data;
990         unsigned int nwords = 0;
991
992         if (!mm)
993                 return 0;
994         do {
995                 nwords += 2;
996         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
997         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
998                                        nwords * sizeof(mm->saved_auxv[0]));
999 }
1000
1001 static const struct file_operations proc_auxv_operations = {
1002         .open           = auxv_open,
1003         .read           = auxv_read,
1004         .llseek         = generic_file_llseek,
1005         .release        = mem_release,
1006 };
1007
1008 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1009                             loff_t *ppos)
1010 {
1011         struct task_struct *task = get_proc_task(file_inode(file));
1012         char buffer[PROC_NUMBUF];
1013         int oom_adj = OOM_ADJUST_MIN;
1014         size_t len;
1015
1016         if (!task)
1017                 return -ESRCH;
1018         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1019                 oom_adj = OOM_ADJUST_MAX;
1020         else
1021                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1022                           OOM_SCORE_ADJ_MAX;
1023         put_task_struct(task);
1024         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1025         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1026 }
1027
1028 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1029 {
1030         static DEFINE_MUTEX(oom_adj_mutex);
1031         struct mm_struct *mm = NULL;
1032         struct task_struct *task;
1033         int err = 0;
1034
1035         task = get_proc_task(file_inode(file));
1036         if (!task)
1037                 return -ESRCH;
1038
1039         mutex_lock(&oom_adj_mutex);
1040         if (legacy) {
1041                 if (oom_adj < task->signal->oom_score_adj &&
1042                                 !capable(CAP_SYS_RESOURCE)) {
1043                         err = -EACCES;
1044                         goto err_unlock;
1045                 }
1046                 /*
1047                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1048                  * /proc/pid/oom_score_adj instead.
1049                  */
1050                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1051                           current->comm, task_pid_nr(current), task_pid_nr(task),
1052                           task_pid_nr(task));
1053         } else {
1054                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1055                                 !capable(CAP_SYS_RESOURCE)) {
1056                         err = -EACCES;
1057                         goto err_unlock;
1058                 }
1059         }
1060
1061         /*
1062          * Make sure we will check other processes sharing the mm if this is
1063          * not vfrok which wants its own oom_score_adj.
1064          * pin the mm so it doesn't go away and get reused after task_unlock
1065          */
1066         if (!task->vfork_done) {
1067                 struct task_struct *p = find_lock_task_mm(task);
1068
1069                 if (p) {
1070                         if (atomic_read(&p->mm->mm_users) > 1) {
1071                                 mm = p->mm;
1072                                 mmgrab(mm);
1073                         }
1074                         task_unlock(p);
1075                 }
1076         }
1077
1078         task->signal->oom_score_adj = oom_adj;
1079         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1080                 task->signal->oom_score_adj_min = (short)oom_adj;
1081         trace_oom_score_adj_update(task);
1082
1083         if (mm) {
1084                 struct task_struct *p;
1085
1086                 rcu_read_lock();
1087                 for_each_process(p) {
1088                         if (same_thread_group(task, p))
1089                                 continue;
1090
1091                         /* do not touch kernel threads or the global init */
1092                         if (p->flags & PF_KTHREAD || is_global_init(p))
1093                                 continue;
1094
1095                         task_lock(p);
1096                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1097                                 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",
1098                                                 task_pid_nr(p), p->comm,
1099                                                 p->signal->oom_score_adj, oom_adj,
1100                                                 task_pid_nr(task), task->comm);
1101                                 p->signal->oom_score_adj = oom_adj;
1102                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1103                                         p->signal->oom_score_adj_min = (short)oom_adj;
1104                         }
1105                         task_unlock(p);
1106                 }
1107                 rcu_read_unlock();
1108                 mmdrop(mm);
1109         }
1110 err_unlock:
1111         mutex_unlock(&oom_adj_mutex);
1112         put_task_struct(task);
1113         return err;
1114 }
1115
1116 /*
1117  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1118  * kernels.  The effective policy is defined by oom_score_adj, which has a
1119  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1120  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1121  * Processes that become oom disabled via oom_adj will still be oom disabled
1122  * with this implementation.
1123  *
1124  * oom_adj cannot be removed since existing userspace binaries use it.
1125  */
1126 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1127                              size_t count, loff_t *ppos)
1128 {
1129         char buffer[PROC_NUMBUF];
1130         int oom_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1142         if (err)
1143                 goto out;
1144         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1145              oom_adj != OOM_DISABLE) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         /*
1151          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1152          * value is always attainable.
1153          */
1154         if (oom_adj == OOM_ADJUST_MAX)
1155                 oom_adj = OOM_SCORE_ADJ_MAX;
1156         else
1157                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1158
1159         err = __set_oom_adj(file, oom_adj, true);
1160 out:
1161         return err < 0 ? err : count;
1162 }
1163
1164 static const struct file_operations proc_oom_adj_operations = {
1165         .read           = oom_adj_read,
1166         .write          = oom_adj_write,
1167         .llseek         = generic_file_llseek,
1168 };
1169
1170 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1171                                         size_t count, loff_t *ppos)
1172 {
1173         struct task_struct *task = get_proc_task(file_inode(file));
1174         char buffer[PROC_NUMBUF];
1175         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1176         size_t len;
1177
1178         if (!task)
1179                 return -ESRCH;
1180         oom_score_adj = task->signal->oom_score_adj;
1181         put_task_struct(task);
1182         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1183         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1184 }
1185
1186 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1187                                         size_t count, loff_t *ppos)
1188 {
1189         char buffer[PROC_NUMBUF];
1190         int oom_score_adj;
1191         int err;
1192
1193         memset(buffer, 0, sizeof(buffer));
1194         if (count > sizeof(buffer) - 1)
1195                 count = sizeof(buffer) - 1;
1196         if (copy_from_user(buffer, buf, count)) {
1197                 err = -EFAULT;
1198                 goto out;
1199         }
1200
1201         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1202         if (err)
1203                 goto out;
1204         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1205                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1206                 err = -EINVAL;
1207                 goto out;
1208         }
1209
1210         err = __set_oom_adj(file, oom_score_adj, false);
1211 out:
1212         return err < 0 ? err : count;
1213 }
1214
1215 static const struct file_operations proc_oom_score_adj_operations = {
1216         .read           = oom_score_adj_read,
1217         .write          = oom_score_adj_write,
1218         .llseek         = default_llseek,
1219 };
1220
1221 #ifdef CONFIG_AUDITSYSCALL
1222 #define TMPBUFLEN 11
1223 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1224                                   size_t count, loff_t *ppos)
1225 {
1226         struct inode * inode = file_inode(file);
1227         struct task_struct *task = get_proc_task(inode);
1228         ssize_t length;
1229         char tmpbuf[TMPBUFLEN];
1230
1231         if (!task)
1232                 return -ESRCH;
1233         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1234                            from_kuid(file->f_cred->user_ns,
1235                                      audit_get_loginuid(task)));
1236         put_task_struct(task);
1237         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1238 }
1239
1240 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1241                                    size_t count, loff_t *ppos)
1242 {
1243         struct inode * inode = file_inode(file);
1244         uid_t loginuid;
1245         kuid_t kloginuid;
1246         int rv;
1247
1248         rcu_read_lock();
1249         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1250                 rcu_read_unlock();
1251                 return -EPERM;
1252         }
1253         rcu_read_unlock();
1254
1255         if (*ppos != 0) {
1256                 /* No partial writes. */
1257                 return -EINVAL;
1258         }
1259
1260         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1261         if (rv < 0)
1262                 return rv;
1263
1264         /* is userspace tring to explicitly UNSET the loginuid? */
1265         if (loginuid == AUDIT_UID_UNSET) {
1266                 kloginuid = INVALID_UID;
1267         } else {
1268                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1269                 if (!uid_valid(kloginuid))
1270                         return -EINVAL;
1271         }
1272
1273         rv = audit_set_loginuid(kloginuid);
1274         if (rv < 0)
1275                 return rv;
1276         return count;
1277 }
1278
1279 static const struct file_operations proc_loginuid_operations = {
1280         .read           = proc_loginuid_read,
1281         .write          = proc_loginuid_write,
1282         .llseek         = generic_file_llseek,
1283 };
1284
1285 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1286                                   size_t count, loff_t *ppos)
1287 {
1288         struct inode * inode = file_inode(file);
1289         struct task_struct *task = get_proc_task(inode);
1290         ssize_t length;
1291         char tmpbuf[TMPBUFLEN];
1292
1293         if (!task)
1294                 return -ESRCH;
1295         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1296                                 audit_get_sessionid(task));
1297         put_task_struct(task);
1298         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1299 }
1300
1301 static const struct file_operations proc_sessionid_operations = {
1302         .read           = proc_sessionid_read,
1303         .llseek         = generic_file_llseek,
1304 };
1305 #endif
1306
1307 #ifdef CONFIG_FAULT_INJECTION
1308 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1309                                       size_t count, loff_t *ppos)
1310 {
1311         struct task_struct *task = get_proc_task(file_inode(file));
1312         char buffer[PROC_NUMBUF];
1313         size_t len;
1314         int make_it_fail;
1315
1316         if (!task)
1317                 return -ESRCH;
1318         make_it_fail = task->make_it_fail;
1319         put_task_struct(task);
1320
1321         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1322
1323         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1324 }
1325
1326 static ssize_t proc_fault_inject_write(struct file * file,
1327                         const char __user * buf, size_t count, loff_t *ppos)
1328 {
1329         struct task_struct *task;
1330         char buffer[PROC_NUMBUF];
1331         int make_it_fail;
1332         int rv;
1333
1334         if (!capable(CAP_SYS_RESOURCE))
1335                 return -EPERM;
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1342         if (rv < 0)
1343                 return rv;
1344         if (make_it_fail < 0 || make_it_fail > 1)
1345                 return -EINVAL;
1346
1347         task = get_proc_task(file_inode(file));
1348         if (!task)
1349                 return -ESRCH;
1350         task->make_it_fail = make_it_fail;
1351         put_task_struct(task);
1352
1353         return count;
1354 }
1355
1356 static const struct file_operations proc_fault_inject_operations = {
1357         .read           = proc_fault_inject_read,
1358         .write          = proc_fault_inject_write,
1359         .llseek         = generic_file_llseek,
1360 };
1361
1362 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1363                                    size_t count, loff_t *ppos)
1364 {
1365         struct task_struct *task;
1366         int err;
1367         unsigned int n;
1368
1369         err = kstrtouint_from_user(buf, count, 0, &n);
1370         if (err)
1371                 return err;
1372
1373         task = get_proc_task(file_inode(file));
1374         if (!task)
1375                 return -ESRCH;
1376         task->fail_nth = n;
1377         put_task_struct(task);
1378
1379         return count;
1380 }
1381
1382 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1383                                   size_t count, loff_t *ppos)
1384 {
1385         struct task_struct *task;
1386         char numbuf[PROC_NUMBUF];
1387         ssize_t len;
1388
1389         task = get_proc_task(file_inode(file));
1390         if (!task)
1391                 return -ESRCH;
1392         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1393         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1394         put_task_struct(task);
1395
1396         return len;
1397 }
1398
1399 static const struct file_operations proc_fail_nth_operations = {
1400         .read           = proc_fail_nth_read,
1401         .write          = proc_fail_nth_write,
1402 };
1403 #endif
1404
1405
1406 #ifdef CONFIG_SCHED_DEBUG
1407 /*
1408  * Print out various scheduling related per-task fields:
1409  */
1410 static int sched_show(struct seq_file *m, void *v)
1411 {
1412         struct inode *inode = m->private;
1413         struct pid_namespace *ns = inode->i_sb->s_fs_info;
1414         struct task_struct *p;
1415
1416         p = get_proc_task(inode);
1417         if (!p)
1418                 return -ESRCH;
1419         proc_sched_show_task(p, ns, m);
1420
1421         put_task_struct(p);
1422
1423         return 0;
1424 }
1425
1426 static ssize_t
1427 sched_write(struct file *file, const char __user *buf,
1428             size_t count, loff_t *offset)
1429 {
1430         struct inode *inode = file_inode(file);
1431         struct task_struct *p;
1432
1433         p = get_proc_task(inode);
1434         if (!p)
1435                 return -ESRCH;
1436         proc_sched_set_task(p);
1437
1438         put_task_struct(p);
1439
1440         return count;
1441 }
1442
1443 static int sched_open(struct inode *inode, struct file *filp)
1444 {
1445         return single_open(filp, sched_show, inode);
1446 }
1447
1448 static const struct file_operations proc_pid_sched_operations = {
1449         .open           = sched_open,
1450         .read           = seq_read,
1451         .write          = sched_write,
1452         .llseek         = seq_lseek,
1453         .release        = single_release,
1454 };
1455
1456 #endif
1457
1458 #ifdef CONFIG_SCHED_AUTOGROUP
1459 /*
1460  * Print out autogroup related information:
1461  */
1462 static int sched_autogroup_show(struct seq_file *m, void *v)
1463 {
1464         struct inode *inode = m->private;
1465         struct task_struct *p;
1466
1467         p = get_proc_task(inode);
1468         if (!p)
1469                 return -ESRCH;
1470         proc_sched_autogroup_show_task(p, m);
1471
1472         put_task_struct(p);
1473
1474         return 0;
1475 }
1476
1477 static ssize_t
1478 sched_autogroup_write(struct file *file, const char __user *buf,
1479             size_t count, loff_t *offset)
1480 {
1481         struct inode *inode = file_inode(file);
1482         struct task_struct *p;
1483         char buffer[PROC_NUMBUF];
1484         int nice;
1485         int err;
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (count > sizeof(buffer) - 1)
1489                 count = sizeof(buffer) - 1;
1490         if (copy_from_user(buffer, buf, count))
1491                 return -EFAULT;
1492
1493         err = kstrtoint(strstrip(buffer), 0, &nice);
1494         if (err < 0)
1495                 return err;
1496
1497         p = get_proc_task(inode);
1498         if (!p)
1499                 return -ESRCH;
1500
1501         err = proc_sched_autogroup_set_nice(p, nice);
1502         if (err)
1503                 count = err;
1504
1505         put_task_struct(p);
1506
1507         return count;
1508 }
1509
1510 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1511 {
1512         int ret;
1513
1514         ret = single_open(filp, sched_autogroup_show, NULL);
1515         if (!ret) {
1516                 struct seq_file *m = filp->private_data;
1517
1518                 m->private = inode;
1519         }
1520         return ret;
1521 }
1522
1523 static const struct file_operations proc_pid_sched_autogroup_operations = {
1524         .open           = sched_autogroup_open,
1525         .read           = seq_read,
1526         .write          = sched_autogroup_write,
1527         .llseek         = seq_lseek,
1528         .release        = single_release,
1529 };
1530
1531 #endif /* CONFIG_SCHED_AUTOGROUP */
1532
1533 static ssize_t comm_write(struct file *file, const char __user *buf,
1534                                 size_t count, loff_t *offset)
1535 {
1536         struct inode *inode = file_inode(file);
1537         struct task_struct *p;
1538         char buffer[TASK_COMM_LEN];
1539         const size_t maxlen = sizeof(buffer) - 1;
1540
1541         memset(buffer, 0, sizeof(buffer));
1542         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1543                 return -EFAULT;
1544
1545         p = get_proc_task(inode);
1546         if (!p)
1547                 return -ESRCH;
1548
1549         if (same_thread_group(current, p))
1550                 set_task_comm(p, buffer);
1551         else
1552                 count = -EINVAL;
1553
1554         put_task_struct(p);
1555
1556         return count;
1557 }
1558
1559 static int comm_show(struct seq_file *m, void *v)
1560 {
1561         struct inode *inode = m->private;
1562         struct task_struct *p;
1563
1564         p = get_proc_task(inode);
1565         if (!p)
1566                 return -ESRCH;
1567
1568         proc_task_name(m, p, false);
1569         seq_putc(m, '\n');
1570
1571         put_task_struct(p);
1572
1573         return 0;
1574 }
1575
1576 static int comm_open(struct inode *inode, struct file *filp)
1577 {
1578         return single_open(filp, comm_show, inode);
1579 }
1580
1581 static const struct file_operations proc_pid_set_comm_operations = {
1582         .open           = comm_open,
1583         .read           = seq_read,
1584         .write          = comm_write,
1585         .llseek         = seq_lseek,
1586         .release        = single_release,
1587 };
1588
1589 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1590 {
1591         struct task_struct *task;
1592         struct file *exe_file;
1593
1594         task = get_proc_task(d_inode(dentry));
1595         if (!task)
1596                 return -ENOENT;
1597         exe_file = get_task_exe_file(task);
1598         put_task_struct(task);
1599         if (exe_file) {
1600                 *exe_path = exe_file->f_path;
1601                 path_get(&exe_file->f_path);
1602                 fput(exe_file);
1603                 return 0;
1604         } else
1605                 return -ENOENT;
1606 }
1607
1608 static const char *proc_pid_get_link(struct dentry *dentry,
1609                                      struct inode *inode,
1610                                      struct delayed_call *done)
1611 {
1612         struct path path;
1613         int error = -EACCES;
1614
1615         if (!dentry)
1616                 return ERR_PTR(-ECHILD);
1617
1618         /* Are we allowed to snoop on the tasks file descriptors? */
1619         if (!proc_fd_access_allowed(inode))
1620                 goto out;
1621
1622         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1623         if (error)
1624                 goto out;
1625
1626         nd_jump_link(&path);
1627         return NULL;
1628 out:
1629         return ERR_PTR(error);
1630 }
1631
1632 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1633 {
1634         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1635         char *pathname;
1636         int len;
1637
1638         if (!tmp)
1639                 return -ENOMEM;
1640
1641         pathname = d_path(path, tmp, PAGE_SIZE);
1642         len = PTR_ERR(pathname);
1643         if (IS_ERR(pathname))
1644                 goto out;
1645         len = tmp + PAGE_SIZE - 1 - pathname;
1646
1647         if (len > buflen)
1648                 len = buflen;
1649         if (copy_to_user(buffer, pathname, len))
1650                 len = -EFAULT;
1651  out:
1652         free_page((unsigned long)tmp);
1653         return len;
1654 }
1655
1656 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1657 {
1658         int error = -EACCES;
1659         struct inode *inode = d_inode(dentry);
1660         struct path path;
1661
1662         /* Are we allowed to snoop on the tasks file descriptors? */
1663         if (!proc_fd_access_allowed(inode))
1664                 goto out;
1665
1666         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1667         if (error)
1668                 goto out;
1669
1670         error = do_proc_readlink(&path, buffer, buflen);
1671         path_put(&path);
1672 out:
1673         return error;
1674 }
1675
1676 const struct inode_operations proc_pid_link_inode_operations = {
1677         .readlink       = proc_pid_readlink,
1678         .get_link       = proc_pid_get_link,
1679         .setattr        = proc_setattr,
1680 };
1681
1682
1683 /* building an inode */
1684
1685 void task_dump_owner(struct task_struct *task, umode_t mode,
1686                      kuid_t *ruid, kgid_t *rgid)
1687 {
1688         /* Depending on the state of dumpable compute who should own a
1689          * proc file for a task.
1690          */
1691         const struct cred *cred;
1692         kuid_t uid;
1693         kgid_t gid;
1694
1695         if (unlikely(task->flags & PF_KTHREAD)) {
1696                 *ruid = GLOBAL_ROOT_UID;
1697                 *rgid = GLOBAL_ROOT_GID;
1698                 return;
1699         }
1700
1701         /* Default to the tasks effective ownership */
1702         rcu_read_lock();
1703         cred = __task_cred(task);
1704         uid = cred->euid;
1705         gid = cred->egid;
1706         rcu_read_unlock();
1707
1708         /*
1709          * Before the /proc/pid/status file was created the only way to read
1710          * the effective uid of a /process was to stat /proc/pid.  Reading
1711          * /proc/pid/status is slow enough that procps and other packages
1712          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1713          * made this apply to all per process world readable and executable
1714          * directories.
1715          */
1716         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1717                 struct mm_struct *mm;
1718                 task_lock(task);
1719                 mm = task->mm;
1720                 /* Make non-dumpable tasks owned by some root */
1721                 if (mm) {
1722                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1723                                 struct user_namespace *user_ns = mm->user_ns;
1724
1725                                 uid = make_kuid(user_ns, 0);
1726                                 if (!uid_valid(uid))
1727                                         uid = GLOBAL_ROOT_UID;
1728
1729                                 gid = make_kgid(user_ns, 0);
1730                                 if (!gid_valid(gid))
1731                                         gid = GLOBAL_ROOT_GID;
1732                         }
1733                 } else {
1734                         uid = GLOBAL_ROOT_UID;
1735                         gid = GLOBAL_ROOT_GID;
1736                 }
1737                 task_unlock(task);
1738         }
1739         *ruid = uid;
1740         *rgid = gid;
1741 }
1742
1743 struct inode *proc_pid_make_inode(struct super_block * sb,
1744                                   struct task_struct *task, umode_t mode)
1745 {
1746         struct inode * inode;
1747         struct proc_inode *ei;
1748
1749         /* We need a new inode */
1750
1751         inode = new_inode(sb);
1752         if (!inode)
1753                 goto out;
1754
1755         /* Common stuff */
1756         ei = PROC_I(inode);
1757         inode->i_mode = mode;
1758         inode->i_ino = get_next_ino();
1759         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1760         inode->i_op = &proc_def_inode_operations;
1761
1762         /*
1763          * grab the reference to task.
1764          */
1765         ei->pid = get_task_pid(task, PIDTYPE_PID);
1766         if (!ei->pid)
1767                 goto out_unlock;
1768
1769         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1770         security_task_to_inode(task, inode);
1771
1772 out:
1773         return inode;
1774
1775 out_unlock:
1776         iput(inode);
1777         return NULL;
1778 }
1779
1780 int pid_getattr(const struct path *path, struct kstat *stat,
1781                 u32 request_mask, unsigned int query_flags)
1782 {
1783         struct inode *inode = d_inode(path->dentry);
1784         struct task_struct *task;
1785         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1786
1787         generic_fillattr(inode, stat);
1788
1789         rcu_read_lock();
1790         stat->uid = GLOBAL_ROOT_UID;
1791         stat->gid = GLOBAL_ROOT_GID;
1792         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1793         if (task) {
1794                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1795                         rcu_read_unlock();
1796                         /*
1797                          * This doesn't prevent learning whether PID exists,
1798                          * it only makes getattr() consistent with readdir().
1799                          */
1800                         return -ENOENT;
1801                 }
1802                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1803         }
1804         rcu_read_unlock();
1805         return 0;
1806 }
1807
1808 /* dentry stuff */
1809
1810 /*
1811  *      Exceptional case: normally we are not allowed to unhash a busy
1812  * directory. In this case, however, we can do it - no aliasing problems
1813  * due to the way we treat inodes.
1814  *
1815  * Rewrite the inode's ownerships here because the owning task may have
1816  * performed a setuid(), etc.
1817  *
1818  */
1819 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1820 {
1821         struct inode *inode;
1822         struct task_struct *task;
1823
1824         if (flags & LOOKUP_RCU)
1825                 return -ECHILD;
1826
1827         inode = d_inode(dentry);
1828         task = get_proc_task(inode);
1829
1830         if (task) {
1831                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1832
1833                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1834                 security_task_to_inode(task, inode);
1835                 put_task_struct(task);
1836                 return 1;
1837         }
1838         return 0;
1839 }
1840
1841 static inline bool proc_inode_is_dead(struct inode *inode)
1842 {
1843         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1844 }
1845
1846 int pid_delete_dentry(const struct dentry *dentry)
1847 {
1848         /* Is the task we represent dead?
1849          * If so, then don't put the dentry on the lru list,
1850          * kill it immediately.
1851          */
1852         return proc_inode_is_dead(d_inode(dentry));
1853 }
1854
1855 const struct dentry_operations pid_dentry_operations =
1856 {
1857         .d_revalidate   = pid_revalidate,
1858         .d_delete       = pid_delete_dentry,
1859 };
1860
1861 /* Lookups */
1862
1863 /*
1864  * Fill a directory entry.
1865  *
1866  * If possible create the dcache entry and derive our inode number and
1867  * file type from dcache entry.
1868  *
1869  * Since all of the proc inode numbers are dynamically generated, the inode
1870  * numbers do not exist until the inode is cache.  This means creating the
1871  * the dcache entry in readdir is necessary to keep the inode numbers
1872  * reported by readdir in sync with the inode numbers reported
1873  * by stat.
1874  */
1875 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1876         const char *name, int len,
1877         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1878 {
1879         struct dentry *child, *dir = file->f_path.dentry;
1880         struct qstr qname = QSTR_INIT(name, len);
1881         struct inode *inode;
1882         unsigned type;
1883         ino_t ino;
1884
1885         child = d_hash_and_lookup(dir, &qname);
1886         if (!child) {
1887                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1888                 child = d_alloc_parallel(dir, &qname, &wq);
1889                 if (IS_ERR(child))
1890                         goto end_instantiate;
1891                 if (d_in_lookup(child)) {
1892                         int err = instantiate(d_inode(dir), child, task, ptr);
1893                         d_lookup_done(child);
1894                         if (err < 0) {
1895                                 dput(child);
1896                                 goto end_instantiate;
1897                         }
1898                 }
1899         }
1900         inode = d_inode(child);
1901         ino = inode->i_ino;
1902         type = inode->i_mode >> 12;
1903         dput(child);
1904         return dir_emit(ctx, name, len, ino, type);
1905
1906 end_instantiate:
1907         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1908 }
1909
1910 /*
1911  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1912  * which represent vma start and end addresses.
1913  */
1914 static int dname_to_vma_addr(struct dentry *dentry,
1915                              unsigned long *start, unsigned long *end)
1916 {
1917         const char *str = dentry->d_name.name;
1918         unsigned long long sval, eval;
1919         unsigned int len;
1920
1921         if (str[0] == '0' && str[1] != '-')
1922                 return -EINVAL;
1923         len = _parse_integer(str, 16, &sval);
1924         if (len & KSTRTOX_OVERFLOW)
1925                 return -EINVAL;
1926         if (sval != (unsigned long)sval)
1927                 return -EINVAL;
1928         str += len;
1929
1930         if (*str != '-')
1931                 return -EINVAL;
1932         str++;
1933
1934         if (str[0] == '0' && str[1])
1935                 return -EINVAL;
1936         len = _parse_integer(str, 16, &eval);
1937         if (len & KSTRTOX_OVERFLOW)
1938                 return -EINVAL;
1939         if (eval != (unsigned long)eval)
1940                 return -EINVAL;
1941         str += len;
1942
1943         if (*str != '\0')
1944                 return -EINVAL;
1945
1946         *start = sval;
1947         *end = eval;
1948
1949         return 0;
1950 }
1951
1952 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1953 {
1954         unsigned long vm_start, vm_end;
1955         bool exact_vma_exists = false;
1956         struct mm_struct *mm = NULL;
1957         struct task_struct *task;
1958         struct inode *inode;
1959         int status = 0;
1960
1961         if (flags & LOOKUP_RCU)
1962                 return -ECHILD;
1963
1964         inode = d_inode(dentry);
1965         task = get_proc_task(inode);
1966         if (!task)
1967                 goto out_notask;
1968
1969         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1970         if (IS_ERR_OR_NULL(mm))
1971                 goto out;
1972
1973         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1974                 down_read(&mm->mmap_sem);
1975                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1976                 up_read(&mm->mmap_sem);
1977         }
1978
1979         mmput(mm);
1980
1981         if (exact_vma_exists) {
1982                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1983
1984                 security_task_to_inode(task, inode);
1985                 status = 1;
1986         }
1987
1988 out:
1989         put_task_struct(task);
1990
1991 out_notask:
1992         return status;
1993 }
1994
1995 static const struct dentry_operations tid_map_files_dentry_operations = {
1996         .d_revalidate   = map_files_d_revalidate,
1997         .d_delete       = pid_delete_dentry,
1998 };
1999
2000 static int map_files_get_link(struct dentry *dentry, struct path *path)
2001 {
2002         unsigned long vm_start, vm_end;
2003         struct vm_area_struct *vma;
2004         struct task_struct *task;
2005         struct mm_struct *mm;
2006         int rc;
2007
2008         rc = -ENOENT;
2009         task = get_proc_task(d_inode(dentry));
2010         if (!task)
2011                 goto out;
2012
2013         mm = get_task_mm(task);
2014         put_task_struct(task);
2015         if (!mm)
2016                 goto out;
2017
2018         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2019         if (rc)
2020                 goto out_mmput;
2021
2022         rc = -ENOENT;
2023         down_read(&mm->mmap_sem);
2024         vma = find_exact_vma(mm, vm_start, vm_end);
2025         if (vma && vma->vm_file) {
2026                 *path = vma->vm_file->f_path;
2027                 path_get(path);
2028                 rc = 0;
2029         }
2030         up_read(&mm->mmap_sem);
2031
2032 out_mmput:
2033         mmput(mm);
2034 out:
2035         return rc;
2036 }
2037
2038 struct map_files_info {
2039         unsigned long   start;
2040         unsigned long   end;
2041         fmode_t         mode;
2042 };
2043
2044 /*
2045  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2046  * symlinks may be used to bypass permissions on ancestor directories in the
2047  * path to the file in question.
2048  */
2049 static const char *
2050 proc_map_files_get_link(struct dentry *dentry,
2051                         struct inode *inode,
2052                         struct delayed_call *done)
2053 {
2054         if (!capable(CAP_SYS_ADMIN))
2055                 return ERR_PTR(-EPERM);
2056
2057         return proc_pid_get_link(dentry, inode, done);
2058 }
2059
2060 /*
2061  * Identical to proc_pid_link_inode_operations except for get_link()
2062  */
2063 static const struct inode_operations proc_map_files_link_inode_operations = {
2064         .readlink       = proc_pid_readlink,
2065         .get_link       = proc_map_files_get_link,
2066         .setattr        = proc_setattr,
2067 };
2068
2069 static int
2070 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2071                            struct task_struct *task, const void *ptr)
2072 {
2073         fmode_t mode = (fmode_t)(unsigned long)ptr;
2074         struct proc_inode *ei;
2075         struct inode *inode;
2076
2077         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2078                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2079                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2080         if (!inode)
2081                 return -ENOENT;
2082
2083         ei = PROC_I(inode);
2084         ei->op.proc_get_link = map_files_get_link;
2085
2086         inode->i_op = &proc_map_files_link_inode_operations;
2087         inode->i_size = 64;
2088
2089         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2090         d_add(dentry, inode);
2091
2092         return 0;
2093 }
2094
2095 static struct dentry *proc_map_files_lookup(struct inode *dir,
2096                 struct dentry *dentry, unsigned int flags)
2097 {
2098         unsigned long vm_start, vm_end;
2099         struct vm_area_struct *vma;
2100         struct task_struct *task;
2101         int result;
2102         struct mm_struct *mm;
2103
2104         result = -ENOENT;
2105         task = get_proc_task(dir);
2106         if (!task)
2107                 goto out;
2108
2109         result = -EACCES;
2110         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2111                 goto out_put_task;
2112
2113         result = -ENOENT;
2114         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2115                 goto out_put_task;
2116
2117         mm = get_task_mm(task);
2118         if (!mm)
2119                 goto out_put_task;
2120
2121         down_read(&mm->mmap_sem);
2122         vma = find_exact_vma(mm, vm_start, vm_end);
2123         if (!vma)
2124                 goto out_no_vma;
2125
2126         if (vma->vm_file)
2127                 result = proc_map_files_instantiate(dir, dentry, task,
2128                                 (void *)(unsigned long)vma->vm_file->f_mode);
2129
2130 out_no_vma:
2131         up_read(&mm->mmap_sem);
2132         mmput(mm);
2133 out_put_task:
2134         put_task_struct(task);
2135 out:
2136         return ERR_PTR(result);
2137 }
2138
2139 static const struct inode_operations proc_map_files_inode_operations = {
2140         .lookup         = proc_map_files_lookup,
2141         .permission     = proc_fd_permission,
2142         .setattr        = proc_setattr,
2143 };
2144
2145 static int
2146 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2147 {
2148         struct vm_area_struct *vma;
2149         struct task_struct *task;
2150         struct mm_struct *mm;
2151         unsigned long nr_files, pos, i;
2152         struct flex_array *fa = NULL;
2153         struct map_files_info info;
2154         struct map_files_info *p;
2155         int ret;
2156
2157         ret = -ENOENT;
2158         task = get_proc_task(file_inode(file));
2159         if (!task)
2160                 goto out;
2161
2162         ret = -EACCES;
2163         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2164                 goto out_put_task;
2165
2166         ret = 0;
2167         if (!dir_emit_dots(file, ctx))
2168                 goto out_put_task;
2169
2170         mm = get_task_mm(task);
2171         if (!mm)
2172                 goto out_put_task;
2173         down_read(&mm->mmap_sem);
2174
2175         nr_files = 0;
2176
2177         /*
2178          * We need two passes here:
2179          *
2180          *  1) Collect vmas of mapped files with mmap_sem taken
2181          *  2) Release mmap_sem and instantiate entries
2182          *
2183          * otherwise we get lockdep complained, since filldir()
2184          * routine might require mmap_sem taken in might_fault().
2185          */
2186
2187         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2188                 if (vma->vm_file && ++pos > ctx->pos)
2189                         nr_files++;
2190         }
2191
2192         if (nr_files) {
2193                 fa = flex_array_alloc(sizeof(info), nr_files,
2194                                         GFP_KERNEL);
2195                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2196                                                 GFP_KERNEL)) {
2197                         ret = -ENOMEM;
2198                         if (fa)
2199                                 flex_array_free(fa);
2200                         up_read(&mm->mmap_sem);
2201                         mmput(mm);
2202                         goto out_put_task;
2203                 }
2204                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2205                                 vma = vma->vm_next) {
2206                         if (!vma->vm_file)
2207                                 continue;
2208                         if (++pos <= ctx->pos)
2209                                 continue;
2210
2211                         info.start = vma->vm_start;
2212                         info.end = vma->vm_end;
2213                         info.mode = vma->vm_file->f_mode;
2214                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2215                                 BUG();
2216                 }
2217         }
2218         up_read(&mm->mmap_sem);
2219         mmput(mm);
2220
2221         for (i = 0; i < nr_files; i++) {
2222                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2223                 unsigned int len;
2224
2225                 p = flex_array_get(fa, i);
2226                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2227                 if (!proc_fill_cache(file, ctx,
2228                                       buf, len,
2229                                       proc_map_files_instantiate,
2230                                       task,
2231                                       (void *)(unsigned long)p->mode))
2232                         break;
2233                 ctx->pos++;
2234         }
2235         if (fa)
2236                 flex_array_free(fa);
2237
2238 out_put_task:
2239         put_task_struct(task);
2240 out:
2241         return ret;
2242 }
2243
2244 static const struct file_operations proc_map_files_operations = {
2245         .read           = generic_read_dir,
2246         .iterate_shared = proc_map_files_readdir,
2247         .llseek         = generic_file_llseek,
2248 };
2249
2250 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2251 struct timers_private {
2252         struct pid *pid;
2253         struct task_struct *task;
2254         struct sighand_struct *sighand;
2255         struct pid_namespace *ns;
2256         unsigned long flags;
2257 };
2258
2259 static void *timers_start(struct seq_file *m, loff_t *pos)
2260 {
2261         struct timers_private *tp = m->private;
2262
2263         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2264         if (!tp->task)
2265                 return ERR_PTR(-ESRCH);
2266
2267         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2268         if (!tp->sighand)
2269                 return ERR_PTR(-ESRCH);
2270
2271         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2272 }
2273
2274 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2275 {
2276         struct timers_private *tp = m->private;
2277         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2278 }
2279
2280 static void timers_stop(struct seq_file *m, void *v)
2281 {
2282         struct timers_private *tp = m->private;
2283
2284         if (tp->sighand) {
2285                 unlock_task_sighand(tp->task, &tp->flags);
2286                 tp->sighand = NULL;
2287         }
2288
2289         if (tp->task) {
2290                 put_task_struct(tp->task);
2291                 tp->task = NULL;
2292         }
2293 }
2294
2295 static int show_timer(struct seq_file *m, void *v)
2296 {
2297         struct k_itimer *timer;
2298         struct timers_private *tp = m->private;
2299         int notify;
2300         static const char * const nstr[] = {
2301                 [SIGEV_SIGNAL] = "signal",
2302                 [SIGEV_NONE] = "none",
2303                 [SIGEV_THREAD] = "thread",
2304         };
2305
2306         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2307         notify = timer->it_sigev_notify;
2308
2309         seq_printf(m, "ID: %d\n", timer->it_id);
2310         seq_printf(m, "signal: %d/%px\n",
2311                    timer->sigq->info.si_signo,
2312                    timer->sigq->info.si_value.sival_ptr);
2313         seq_printf(m, "notify: %s/%s.%d\n",
2314                    nstr[notify & ~SIGEV_THREAD_ID],
2315                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2316                    pid_nr_ns(timer->it_pid, tp->ns));
2317         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2318
2319         return 0;
2320 }
2321
2322 static const struct seq_operations proc_timers_seq_ops = {
2323         .start  = timers_start,
2324         .next   = timers_next,
2325         .stop   = timers_stop,
2326         .show   = show_timer,
2327 };
2328
2329 static int proc_timers_open(struct inode *inode, struct file *file)
2330 {
2331         struct timers_private *tp;
2332
2333         tp = __seq_open_private(file, &proc_timers_seq_ops,
2334                         sizeof(struct timers_private));
2335         if (!tp)
2336                 return -ENOMEM;
2337
2338         tp->pid = proc_pid(inode);
2339         tp->ns = inode->i_sb->s_fs_info;
2340         return 0;
2341 }
2342
2343 static const struct file_operations proc_timers_operations = {
2344         .open           = proc_timers_open,
2345         .read           = seq_read,
2346         .llseek         = seq_lseek,
2347         .release        = seq_release_private,
2348 };
2349 #endif
2350
2351 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2352                                         size_t count, loff_t *offset)
2353 {
2354         struct inode *inode = file_inode(file);
2355         struct task_struct *p;
2356         u64 slack_ns;
2357         int err;
2358
2359         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2360         if (err < 0)
2361                 return err;
2362
2363         p = get_proc_task(inode);
2364         if (!p)
2365                 return -ESRCH;
2366
2367         if (p != current) {
2368                 if (!capable(CAP_SYS_NICE)) {
2369                         count = -EPERM;
2370                         goto out;
2371                 }
2372
2373                 err = security_task_setscheduler(p);
2374                 if (err) {
2375                         count = err;
2376                         goto out;
2377                 }
2378         }
2379
2380         task_lock(p);
2381         if (slack_ns == 0)
2382                 p->timer_slack_ns = p->default_timer_slack_ns;
2383         else
2384                 p->timer_slack_ns = slack_ns;
2385         task_unlock(p);
2386
2387 out:
2388         put_task_struct(p);
2389
2390         return count;
2391 }
2392
2393 static int timerslack_ns_show(struct seq_file *m, void *v)
2394 {
2395         struct inode *inode = m->private;
2396         struct task_struct *p;
2397         int err = 0;
2398
2399         p = get_proc_task(inode);
2400         if (!p)
2401                 return -ESRCH;
2402
2403         if (p != current) {
2404
2405                 if (!capable(CAP_SYS_NICE)) {
2406                         err = -EPERM;
2407                         goto out;
2408                 }
2409                 err = security_task_getscheduler(p);
2410                 if (err)
2411                         goto out;
2412         }
2413
2414         task_lock(p);
2415         seq_printf(m, "%llu\n", p->timer_slack_ns);
2416         task_unlock(p);
2417
2418 out:
2419         put_task_struct(p);
2420
2421         return err;
2422 }
2423
2424 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2425 {
2426         return single_open(filp, timerslack_ns_show, inode);
2427 }
2428
2429 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2430         .open           = timerslack_ns_open,
2431         .read           = seq_read,
2432         .write          = timerslack_ns_write,
2433         .llseek         = seq_lseek,
2434         .release        = single_release,
2435 };
2436
2437 static int proc_pident_instantiate(struct inode *dir,
2438         struct dentry *dentry, struct task_struct *task, const void *ptr)
2439 {
2440         const struct pid_entry *p = ptr;
2441         struct inode *inode;
2442         struct proc_inode *ei;
2443
2444         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2445         if (!inode)
2446                 goto out;
2447
2448         ei = PROC_I(inode);
2449         if (S_ISDIR(inode->i_mode))
2450                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2451         if (p->iop)
2452                 inode->i_op = p->iop;
2453         if (p->fop)
2454                 inode->i_fop = p->fop;
2455         ei->op = p->op;
2456         d_set_d_op(dentry, &pid_dentry_operations);
2457         d_add(dentry, inode);
2458         /* Close the race of the process dying before we return the dentry */
2459         if (pid_revalidate(dentry, 0))
2460                 return 0;
2461 out:
2462         return -ENOENT;
2463 }
2464
2465 static struct dentry *proc_pident_lookup(struct inode *dir, 
2466                                          struct dentry *dentry,
2467                                          const struct pid_entry *ents,
2468                                          unsigned int nents)
2469 {
2470         int error;
2471         struct task_struct *task = get_proc_task(dir);
2472         const struct pid_entry *p, *last;
2473
2474         error = -ENOENT;
2475
2476         if (!task)
2477                 goto out_no_task;
2478
2479         /*
2480          * Yes, it does not scale. And it should not. Don't add
2481          * new entries into /proc/<tgid>/ without very good reasons.
2482          */
2483         last = &ents[nents];
2484         for (p = ents; p < last; p++) {
2485                 if (p->len != dentry->d_name.len)
2486                         continue;
2487                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2488                         break;
2489         }
2490         if (p >= last)
2491                 goto out;
2492
2493         error = proc_pident_instantiate(dir, dentry, task, p);
2494 out:
2495         put_task_struct(task);
2496 out_no_task:
2497         return ERR_PTR(error);
2498 }
2499
2500 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2501                 const struct pid_entry *ents, unsigned int nents)
2502 {
2503         struct task_struct *task = get_proc_task(file_inode(file));
2504         const struct pid_entry *p;
2505
2506         if (!task)
2507                 return -ENOENT;
2508
2509         if (!dir_emit_dots(file, ctx))
2510                 goto out;
2511
2512         if (ctx->pos >= nents + 2)
2513                 goto out;
2514
2515         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2516                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2517                                 proc_pident_instantiate, task, p))
2518                         break;
2519                 ctx->pos++;
2520         }
2521 out:
2522         put_task_struct(task);
2523         return 0;
2524 }
2525
2526 #ifdef CONFIG_SECURITY
2527 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2528                                   size_t count, loff_t *ppos)
2529 {
2530         struct inode * inode = file_inode(file);
2531         char *p = NULL;
2532         ssize_t length;
2533         struct task_struct *task = get_proc_task(inode);
2534
2535         if (!task)
2536                 return -ESRCH;
2537
2538         length = security_getprocattr(task,
2539                                       (char*)file->f_path.dentry->d_name.name,
2540                                       &p);
2541         put_task_struct(task);
2542         if (length > 0)
2543                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2544         kfree(p);
2545         return length;
2546 }
2547
2548 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2549                                    size_t count, loff_t *ppos)
2550 {
2551         struct inode * inode = file_inode(file);
2552         void *page;
2553         ssize_t length;
2554         struct task_struct *task = get_proc_task(inode);
2555
2556         length = -ESRCH;
2557         if (!task)
2558                 goto out_no_task;
2559
2560         /* A task may only write its own attributes. */
2561         length = -EACCES;
2562         if (current != task)
2563                 goto out;
2564
2565         if (count > PAGE_SIZE)
2566                 count = PAGE_SIZE;
2567
2568         /* No partial writes. */
2569         length = -EINVAL;
2570         if (*ppos != 0)
2571                 goto out;
2572
2573         page = memdup_user(buf, count);
2574         if (IS_ERR(page)) {
2575                 length = PTR_ERR(page);
2576                 goto out;
2577         }
2578
2579         /* Guard against adverse ptrace interaction */
2580         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2581         if (length < 0)
2582                 goto out_free;
2583
2584         length = security_setprocattr(file->f_path.dentry->d_name.name,
2585                                       page, count);
2586         mutex_unlock(&current->signal->cred_guard_mutex);
2587 out_free:
2588         kfree(page);
2589 out:
2590         put_task_struct(task);
2591 out_no_task:
2592         return length;
2593 }
2594
2595 static const struct file_operations proc_pid_attr_operations = {
2596         .read           = proc_pid_attr_read,
2597         .write          = proc_pid_attr_write,
2598         .llseek         = generic_file_llseek,
2599 };
2600
2601 static const struct pid_entry attr_dir_stuff[] = {
2602         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2603         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2604         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2605         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2606         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2607         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2608 };
2609
2610 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2611 {
2612         return proc_pident_readdir(file, ctx, 
2613                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2614 }
2615
2616 static const struct file_operations proc_attr_dir_operations = {
2617         .read           = generic_read_dir,
2618         .iterate_shared = proc_attr_dir_readdir,
2619         .llseek         = generic_file_llseek,
2620 };
2621
2622 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2623                                 struct dentry *dentry, unsigned int flags)
2624 {
2625         return proc_pident_lookup(dir, dentry,
2626                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2627 }
2628
2629 static const struct inode_operations proc_attr_dir_inode_operations = {
2630         .lookup         = proc_attr_dir_lookup,
2631         .getattr        = pid_getattr,
2632         .setattr        = proc_setattr,
2633 };
2634
2635 #endif
2636
2637 #ifdef CONFIG_ELF_CORE
2638 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2639                                          size_t count, loff_t *ppos)
2640 {
2641         struct task_struct *task = get_proc_task(file_inode(file));
2642         struct mm_struct *mm;
2643         char buffer[PROC_NUMBUF];
2644         size_t len;
2645         int ret;
2646
2647         if (!task)
2648                 return -ESRCH;
2649
2650         ret = 0;
2651         mm = get_task_mm(task);
2652         if (mm) {
2653                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2654                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2655                                 MMF_DUMP_FILTER_SHIFT));
2656                 mmput(mm);
2657                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2658         }
2659
2660         put_task_struct(task);
2661
2662         return ret;
2663 }
2664
2665 static ssize_t proc_coredump_filter_write(struct file *file,
2666                                           const char __user *buf,
2667                                           size_t count,
2668                                           loff_t *ppos)
2669 {
2670         struct task_struct *task;
2671         struct mm_struct *mm;
2672         unsigned int val;
2673         int ret;
2674         int i;
2675         unsigned long mask;
2676
2677         ret = kstrtouint_from_user(buf, count, 0, &val);
2678         if (ret < 0)
2679                 return ret;
2680
2681         ret = -ESRCH;
2682         task = get_proc_task(file_inode(file));
2683         if (!task)
2684                 goto out_no_task;
2685
2686         mm = get_task_mm(task);
2687         if (!mm)
2688                 goto out_no_mm;
2689         ret = 0;
2690
2691         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2692                 if (val & mask)
2693                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2694                 else
2695                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2696         }
2697
2698         mmput(mm);
2699  out_no_mm:
2700         put_task_struct(task);
2701  out_no_task:
2702         if (ret < 0)
2703                 return ret;
2704         return count;
2705 }
2706
2707 static const struct file_operations proc_coredump_filter_operations = {
2708         .read           = proc_coredump_filter_read,
2709         .write          = proc_coredump_filter_write,
2710         .llseek         = generic_file_llseek,
2711 };
2712 #endif
2713
2714 #ifdef CONFIG_TASK_IO_ACCOUNTING
2715 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2716 {
2717         struct task_io_accounting acct = task->ioac;
2718         unsigned long flags;
2719         int result;
2720
2721         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2722         if (result)
2723                 return result;
2724
2725         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2726                 result = -EACCES;
2727                 goto out_unlock;
2728         }
2729
2730         if (whole && lock_task_sighand(task, &flags)) {
2731                 struct task_struct *t = task;
2732
2733                 task_io_accounting_add(&acct, &task->signal->ioac);
2734                 while_each_thread(task, t)
2735                         task_io_accounting_add(&acct, &t->ioac);
2736
2737                 unlock_task_sighand(task, &flags);
2738         }
2739         seq_printf(m,
2740                    "rchar: %llu\n"
2741                    "wchar: %llu\n"
2742                    "syscr: %llu\n"
2743                    "syscw: %llu\n"
2744                    "read_bytes: %llu\n"
2745                    "write_bytes: %llu\n"
2746                    "cancelled_write_bytes: %llu\n",
2747                    (unsigned long long)acct.rchar,
2748                    (unsigned long long)acct.wchar,
2749                    (unsigned long long)acct.syscr,
2750                    (unsigned long long)acct.syscw,
2751                    (unsigned long long)acct.read_bytes,
2752                    (unsigned long long)acct.write_bytes,
2753                    (unsigned long long)acct.cancelled_write_bytes);
2754         result = 0;
2755
2756 out_unlock:
2757         mutex_unlock(&task->signal->cred_guard_mutex);
2758         return result;
2759 }
2760
2761 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2762                                   struct pid *pid, struct task_struct *task)
2763 {
2764         return do_io_accounting(task, m, 0);
2765 }
2766
2767 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2768                                    struct pid *pid, struct task_struct *task)
2769 {
2770         return do_io_accounting(task, m, 1);
2771 }
2772 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2773
2774 #ifdef CONFIG_USER_NS
2775 static int proc_id_map_open(struct inode *inode, struct file *file,
2776         const struct seq_operations *seq_ops)
2777 {
2778         struct user_namespace *ns = NULL;
2779         struct task_struct *task;
2780         struct seq_file *seq;
2781         int ret = -EINVAL;
2782
2783         task = get_proc_task(inode);
2784         if (task) {
2785                 rcu_read_lock();
2786                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2787                 rcu_read_unlock();
2788                 put_task_struct(task);
2789         }
2790         if (!ns)
2791                 goto err;
2792
2793         ret = seq_open(file, seq_ops);
2794         if (ret)
2795                 goto err_put_ns;
2796
2797         seq = file->private_data;
2798         seq->private = ns;
2799
2800         return 0;
2801 err_put_ns:
2802         put_user_ns(ns);
2803 err:
2804         return ret;
2805 }
2806
2807 static int proc_id_map_release(struct inode *inode, struct file *file)
2808 {
2809         struct seq_file *seq = file->private_data;
2810         struct user_namespace *ns = seq->private;
2811         put_user_ns(ns);
2812         return seq_release(inode, file);
2813 }
2814
2815 static int proc_uid_map_open(struct inode *inode, struct file *file)
2816 {
2817         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2818 }
2819
2820 static int proc_gid_map_open(struct inode *inode, struct file *file)
2821 {
2822         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2823 }
2824
2825 static int proc_projid_map_open(struct inode *inode, struct file *file)
2826 {
2827         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2828 }
2829
2830 static const struct file_operations proc_uid_map_operations = {
2831         .open           = proc_uid_map_open,
2832         .write          = proc_uid_map_write,
2833         .read           = seq_read,
2834         .llseek         = seq_lseek,
2835         .release        = proc_id_map_release,
2836 };
2837
2838 static const struct file_operations proc_gid_map_operations = {
2839         .open           = proc_gid_map_open,
2840         .write          = proc_gid_map_write,
2841         .read           = seq_read,
2842         .llseek         = seq_lseek,
2843         .release        = proc_id_map_release,
2844 };
2845
2846 static const struct file_operations proc_projid_map_operations = {
2847         .open           = proc_projid_map_open,
2848         .write          = proc_projid_map_write,
2849         .read           = seq_read,
2850         .llseek         = seq_lseek,
2851         .release        = proc_id_map_release,
2852 };
2853
2854 static int proc_setgroups_open(struct inode *inode, struct file *file)
2855 {
2856         struct user_namespace *ns = NULL;
2857         struct task_struct *task;
2858         int ret;
2859
2860         ret = -ESRCH;
2861         task = get_proc_task(inode);
2862         if (task) {
2863                 rcu_read_lock();
2864                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2865                 rcu_read_unlock();
2866                 put_task_struct(task);
2867         }
2868         if (!ns)
2869                 goto err;
2870
2871         if (file->f_mode & FMODE_WRITE) {
2872                 ret = -EACCES;
2873                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2874                         goto err_put_ns;
2875         }
2876
2877         ret = single_open(file, &proc_setgroups_show, ns);
2878         if (ret)
2879                 goto err_put_ns;
2880
2881         return 0;
2882 err_put_ns:
2883         put_user_ns(ns);
2884 err:
2885         return ret;
2886 }
2887
2888 static int proc_setgroups_release(struct inode *inode, struct file *file)
2889 {
2890         struct seq_file *seq = file->private_data;
2891         struct user_namespace *ns = seq->private;
2892         int ret = single_release(inode, file);
2893         put_user_ns(ns);
2894         return ret;
2895 }
2896
2897 static const struct file_operations proc_setgroups_operations = {
2898         .open           = proc_setgroups_open,
2899         .write          = proc_setgroups_write,
2900         .read           = seq_read,
2901         .llseek         = seq_lseek,
2902         .release        = proc_setgroups_release,
2903 };
2904 #endif /* CONFIG_USER_NS */
2905
2906 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2907                                 struct pid *pid, struct task_struct *task)
2908 {
2909         int err = lock_trace(task);
2910         if (!err) {
2911                 seq_printf(m, "%08x\n", task->personality);
2912                 unlock_trace(task);
2913         }
2914         return err;
2915 }
2916
2917 #ifdef CONFIG_LIVEPATCH
2918 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2919                                 struct pid *pid, struct task_struct *task)
2920 {
2921         seq_printf(m, "%d\n", task->patch_state);
2922         return 0;
2923 }
2924 #endif /* CONFIG_LIVEPATCH */
2925
2926 /*
2927  * Thread groups
2928  */
2929 static const struct file_operations proc_task_operations;
2930 static const struct inode_operations proc_task_inode_operations;
2931
2932 static const struct pid_entry tgid_base_stuff[] = {
2933         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2934         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2935         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2936         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2937         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2938 #ifdef CONFIG_NET
2939         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2940 #endif
2941         REG("environ",    S_IRUSR, proc_environ_operations),
2942         REG("auxv",       S_IRUSR, proc_auxv_operations),
2943         ONE("status",     S_IRUGO, proc_pid_status),
2944         ONE("personality", S_IRUSR, proc_pid_personality),
2945         ONE("limits",     S_IRUGO, proc_pid_limits),
2946 #ifdef CONFIG_SCHED_DEBUG
2947         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2948 #endif
2949 #ifdef CONFIG_SCHED_AUTOGROUP
2950         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2951 #endif
2952         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2953 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2954         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2955 #endif
2956         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2957         ONE("stat",       S_IRUGO, proc_tgid_stat),
2958         ONE("statm",      S_IRUGO, proc_pid_statm),
2959         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2960 #ifdef CONFIG_NUMA
2961         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2962 #endif
2963         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2964         LNK("cwd",        proc_cwd_link),
2965         LNK("root",       proc_root_link),
2966         LNK("exe",        proc_exe_link),
2967         REG("mounts",     S_IRUGO, proc_mounts_operations),
2968         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2969         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2970 #ifdef CONFIG_PROC_PAGE_MONITOR
2971         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2972         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2973         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2974         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2975 #endif
2976 #ifdef CONFIG_SECURITY
2977         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2978 #endif
2979 #ifdef CONFIG_KALLSYMS
2980         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2981 #endif
2982 #ifdef CONFIG_STACKTRACE
2983         ONE("stack",      S_IRUSR, proc_pid_stack),
2984 #endif
2985 #ifdef CONFIG_SCHED_INFO
2986         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2987 #endif
2988 #ifdef CONFIG_LATENCYTOP
2989         REG("latency",  S_IRUGO, proc_lstats_operations),
2990 #endif
2991 #ifdef CONFIG_PROC_PID_CPUSET
2992         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2993 #endif
2994 #ifdef CONFIG_CGROUPS
2995         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2996 #endif
2997         ONE("oom_score",  S_IRUGO, proc_oom_score),
2998         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2999         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3000 #ifdef CONFIG_AUDITSYSCALL
3001         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3002         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3003 #endif
3004 #ifdef CONFIG_FAULT_INJECTION
3005         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3006         REG("fail-nth", 0644, proc_fail_nth_operations),
3007 #endif
3008 #ifdef CONFIG_ELF_CORE
3009         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3010 #endif
3011 #ifdef CONFIG_TASK_IO_ACCOUNTING
3012         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3013 #endif
3014 #ifdef CONFIG_USER_NS
3015         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3016         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3017         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3018         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3019 #endif
3020 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3021         REG("timers",     S_IRUGO, proc_timers_operations),
3022 #endif
3023         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3024 #ifdef CONFIG_LIVEPATCH
3025         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3026 #endif
3027 };
3028
3029 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3030 {
3031         return proc_pident_readdir(file, ctx,
3032                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3033 }
3034
3035 static const struct file_operations proc_tgid_base_operations = {
3036         .read           = generic_read_dir,
3037         .iterate_shared = proc_tgid_base_readdir,
3038         .llseek         = generic_file_llseek,
3039 };
3040
3041 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3042 {
3043         return proc_pident_lookup(dir, dentry,
3044                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3045 }
3046
3047 static const struct inode_operations proc_tgid_base_inode_operations = {
3048         .lookup         = proc_tgid_base_lookup,
3049         .getattr        = pid_getattr,
3050         .setattr        = proc_setattr,
3051         .permission     = proc_pid_permission,
3052 };
3053
3054 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3055 {
3056         struct dentry *dentry, *leader, *dir;
3057         char buf[10 + 1];
3058         struct qstr name;
3059
3060         name.name = buf;
3061         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3062         /* no ->d_hash() rejects on procfs */
3063         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3064         if (dentry) {
3065                 d_invalidate(dentry);
3066                 dput(dentry);
3067         }
3068
3069         if (pid == tgid)
3070                 return;
3071
3072         name.name = buf;
3073         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3074         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3075         if (!leader)
3076                 goto out;
3077
3078         name.name = "task";
3079         name.len = strlen(name.name);
3080         dir = d_hash_and_lookup(leader, &name);
3081         if (!dir)
3082                 goto out_put_leader;
3083
3084         name.name = buf;
3085         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3086         dentry = d_hash_and_lookup(dir, &name);
3087         if (dentry) {
3088                 d_invalidate(dentry);
3089                 dput(dentry);
3090         }
3091
3092         dput(dir);
3093 out_put_leader:
3094         dput(leader);
3095 out:
3096         return;
3097 }
3098
3099 /**
3100  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3101  * @task: task that should be flushed.
3102  *
3103  * When flushing dentries from proc, one needs to flush them from global
3104  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3105  * in. This call is supposed to do all of this job.
3106  *
3107  * Looks in the dcache for
3108  * /proc/@pid
3109  * /proc/@tgid/task/@pid
3110  * if either directory is present flushes it and all of it'ts children
3111  * from the dcache.
3112  *
3113  * It is safe and reasonable to cache /proc entries for a task until
3114  * that task exits.  After that they just clog up the dcache with
3115  * useless entries, possibly causing useful dcache entries to be
3116  * flushed instead.  This routine is proved to flush those useless
3117  * dcache entries at process exit time.
3118  *
3119  * NOTE: This routine is just an optimization so it does not guarantee
3120  *       that no dcache entries will exist at process exit time it
3121  *       just makes it very unlikely that any will persist.
3122  */
3123
3124 void proc_flush_task(struct task_struct *task)
3125 {
3126         int i;
3127         struct pid *pid, *tgid;
3128         struct upid *upid;
3129
3130         pid = task_pid(task);
3131         tgid = task_tgid(task);
3132
3133         for (i = 0; i <= pid->level; i++) {
3134                 upid = &pid->numbers[i];
3135                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3136                                         tgid->numbers[i].nr);
3137         }
3138 }
3139
3140 static int proc_pid_instantiate(struct inode *dir,
3141                                    struct dentry * dentry,
3142                                    struct task_struct *task, const void *ptr)
3143 {
3144         struct inode *inode;
3145
3146         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3147         if (!inode)
3148                 goto out;
3149
3150         inode->i_op = &proc_tgid_base_inode_operations;
3151         inode->i_fop = &proc_tgid_base_operations;
3152         inode->i_flags|=S_IMMUTABLE;
3153
3154         set_nlink(inode, nlink_tgid);
3155
3156         d_set_d_op(dentry, &pid_dentry_operations);
3157
3158         d_add(dentry, inode);
3159         /* Close the race of the process dying before we return the dentry */
3160         if (pid_revalidate(dentry, 0))
3161                 return 0;
3162 out:
3163         return -ENOENT;
3164 }
3165
3166 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3167 {
3168         int result = -ENOENT;
3169         struct task_struct *task;
3170         unsigned tgid;
3171         struct pid_namespace *ns;
3172
3173         tgid = name_to_int(&dentry->d_name);
3174         if (tgid == ~0U)
3175                 goto out;
3176
3177         ns = dentry->d_sb->s_fs_info;
3178         rcu_read_lock();
3179         task = find_task_by_pid_ns(tgid, ns);
3180         if (task)
3181                 get_task_struct(task);
3182         rcu_read_unlock();
3183         if (!task)
3184                 goto out;
3185
3186         result = proc_pid_instantiate(dir, dentry, task, NULL);
3187         put_task_struct(task);
3188 out:
3189         return ERR_PTR(result);
3190 }
3191
3192 /*
3193  * Find the first task with tgid >= tgid
3194  *
3195  */
3196 struct tgid_iter {
3197         unsigned int tgid;
3198         struct task_struct *task;
3199 };
3200 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3201 {
3202         struct pid *pid;
3203
3204         if (iter.task)
3205                 put_task_struct(iter.task);
3206         rcu_read_lock();
3207 retry:
3208         iter.task = NULL;
3209         pid = find_ge_pid(iter.tgid, ns);
3210         if (pid) {
3211                 iter.tgid = pid_nr_ns(pid, ns);
3212                 iter.task = pid_task(pid, PIDTYPE_PID);
3213                 /* What we to know is if the pid we have find is the
3214                  * pid of a thread_group_leader.  Testing for task
3215                  * being a thread_group_leader is the obvious thing
3216                  * todo but there is a window when it fails, due to
3217                  * the pid transfer logic in de_thread.
3218                  *
3219                  * So we perform the straight forward test of seeing
3220                  * if the pid we have found is the pid of a thread
3221                  * group leader, and don't worry if the task we have
3222                  * found doesn't happen to be a thread group leader.
3223                  * As we don't care in the case of readdir.
3224                  */
3225                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3226                         iter.tgid += 1;
3227                         goto retry;
3228                 }
3229                 get_task_struct(iter.task);
3230         }
3231         rcu_read_unlock();
3232         return iter;
3233 }
3234
3235 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3236
3237 /* for the /proc/ directory itself, after non-process stuff has been done */
3238 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3239 {
3240         struct tgid_iter iter;
3241         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3242         loff_t pos = ctx->pos;
3243
3244         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3245                 return 0;
3246
3247         if (pos == TGID_OFFSET - 2) {
3248                 struct inode *inode = d_inode(ns->proc_self);
3249                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3250                         return 0;
3251                 ctx->pos = pos = pos + 1;
3252         }
3253         if (pos == TGID_OFFSET - 1) {
3254                 struct inode *inode = d_inode(ns->proc_thread_self);
3255                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3256                         return 0;
3257                 ctx->pos = pos = pos + 1;
3258         }
3259         iter.tgid = pos - TGID_OFFSET;
3260         iter.task = NULL;
3261         for (iter = next_tgid(ns, iter);
3262              iter.task;
3263              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3264                 char name[10 + 1];
3265                 int len;
3266
3267                 cond_resched();
3268                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3269                         continue;
3270
3271                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3272                 ctx->pos = iter.tgid + TGID_OFFSET;
3273                 if (!proc_fill_cache(file, ctx, name, len,
3274                                      proc_pid_instantiate, iter.task, NULL)) {
3275                         put_task_struct(iter.task);
3276                         return 0;
3277                 }
3278         }
3279         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3280         return 0;
3281 }
3282
3283 /*
3284  * proc_tid_comm_permission is a special permission function exclusively
3285  * used for the node /proc/<pid>/task/<tid>/comm.
3286  * It bypasses generic permission checks in the case where a task of the same
3287  * task group attempts to access the node.
3288  * The rationale behind this is that glibc and bionic access this node for
3289  * cross thread naming (pthread_set/getname_np(!self)). However, if
3290  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3291  * which locks out the cross thread naming implementation.
3292  * This function makes sure that the node is always accessible for members of
3293  * same thread group.
3294  */
3295 static int proc_tid_comm_permission(struct inode *inode, int mask)
3296 {
3297         bool is_same_tgroup;
3298         struct task_struct *task;
3299
3300         task = get_proc_task(inode);
3301         if (!task)
3302                 return -ESRCH;
3303         is_same_tgroup = same_thread_group(current, task);
3304         put_task_struct(task);
3305
3306         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3307                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3308                  * read or written by the members of the corresponding
3309                  * thread group.
3310                  */
3311                 return 0;
3312         }
3313
3314         return generic_permission(inode, mask);
3315 }
3316
3317 static const struct inode_operations proc_tid_comm_inode_operations = {
3318                 .permission = proc_tid_comm_permission,
3319 };
3320
3321 /*
3322  * Tasks
3323  */
3324 static const struct pid_entry tid_base_stuff[] = {
3325         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3326         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3327         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3328 #ifdef CONFIG_NET
3329         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3330 #endif
3331         REG("environ",   S_IRUSR, proc_environ_operations),
3332         REG("auxv",      S_IRUSR, proc_auxv_operations),
3333         ONE("status",    S_IRUGO, proc_pid_status),
3334         ONE("personality", S_IRUSR, proc_pid_personality),
3335         ONE("limits",    S_IRUGO, proc_pid_limits),
3336 #ifdef CONFIG_SCHED_DEBUG
3337         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3338 #endif
3339         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3340                          &proc_tid_comm_inode_operations,
3341                          &proc_pid_set_comm_operations, {}),
3342 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3343         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3344 #endif
3345         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3346         ONE("stat",      S_IRUGO, proc_tid_stat),
3347         ONE("statm",     S_IRUGO, proc_pid_statm),
3348         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3349 #ifdef CONFIG_PROC_CHILDREN
3350         REG("children",  S_IRUGO, proc_tid_children_operations),
3351 #endif
3352 #ifdef CONFIG_NUMA
3353         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3354 #endif
3355         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3356         LNK("cwd",       proc_cwd_link),
3357         LNK("root",      proc_root_link),
3358         LNK("exe",       proc_exe_link),
3359         REG("mounts",    S_IRUGO, proc_mounts_operations),
3360         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3361 #ifdef CONFIG_PROC_PAGE_MONITOR
3362         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3363         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3364         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3365         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3366 #endif
3367 #ifdef CONFIG_SECURITY
3368         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3369 #endif
3370 #ifdef CONFIG_KALLSYMS
3371         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3372 #endif
3373 #ifdef CONFIG_STACKTRACE
3374         ONE("stack",      S_IRUSR, proc_pid_stack),
3375 #endif
3376 #ifdef CONFIG_SCHED_INFO
3377         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3378 #endif
3379 #ifdef CONFIG_LATENCYTOP
3380         REG("latency",  S_IRUGO, proc_lstats_operations),
3381 #endif
3382 #ifdef CONFIG_PROC_PID_CPUSET
3383         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3384 #endif
3385 #ifdef CONFIG_CGROUPS
3386         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3387 #endif
3388         ONE("oom_score", S_IRUGO, proc_oom_score),
3389         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3390         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3391 #ifdef CONFIG_AUDITSYSCALL
3392         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3393         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3394 #endif
3395 #ifdef CONFIG_FAULT_INJECTION
3396         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3397         REG("fail-nth", 0644, proc_fail_nth_operations),
3398 #endif
3399 #ifdef CONFIG_TASK_IO_ACCOUNTING
3400         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3401 #endif
3402 #ifdef CONFIG_USER_NS
3403         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3404         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3405         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3406         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3407 #endif
3408 #ifdef CONFIG_LIVEPATCH
3409         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3410 #endif
3411 };
3412
3413 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3414 {
3415         return proc_pident_readdir(file, ctx,
3416                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3417 }
3418
3419 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3420 {
3421         return proc_pident_lookup(dir, dentry,
3422                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3423 }
3424
3425 static const struct file_operations proc_tid_base_operations = {
3426         .read           = generic_read_dir,
3427         .iterate_shared = proc_tid_base_readdir,
3428         .llseek         = generic_file_llseek,
3429 };
3430
3431 static const struct inode_operations proc_tid_base_inode_operations = {
3432         .lookup         = proc_tid_base_lookup,
3433         .getattr        = pid_getattr,
3434         .setattr        = proc_setattr,
3435 };
3436
3437 static int proc_task_instantiate(struct inode *dir,
3438         struct dentry *dentry, struct task_struct *task, const void *ptr)
3439 {
3440         struct inode *inode;
3441         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3442
3443         if (!inode)
3444                 goto out;
3445         inode->i_op = &proc_tid_base_inode_operations;
3446         inode->i_fop = &proc_tid_base_operations;
3447         inode->i_flags|=S_IMMUTABLE;
3448
3449         set_nlink(inode, nlink_tid);
3450
3451         d_set_d_op(dentry, &pid_dentry_operations);
3452
3453         d_add(dentry, inode);
3454         /* Close the race of the process dying before we return the dentry */
3455         if (pid_revalidate(dentry, 0))
3456                 return 0;
3457 out:
3458         return -ENOENT;
3459 }
3460
3461 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3462 {
3463         int result = -ENOENT;
3464         struct task_struct *task;
3465         struct task_struct *leader = get_proc_task(dir);
3466         unsigned tid;
3467         struct pid_namespace *ns;
3468
3469         if (!leader)
3470                 goto out_no_task;
3471
3472         tid = name_to_int(&dentry->d_name);
3473         if (tid == ~0U)
3474                 goto out;
3475
3476         ns = dentry->d_sb->s_fs_info;
3477         rcu_read_lock();
3478         task = find_task_by_pid_ns(tid, ns);
3479         if (task)
3480                 get_task_struct(task);
3481         rcu_read_unlock();
3482         if (!task)
3483                 goto out;
3484         if (!same_thread_group(leader, task))
3485                 goto out_drop_task;
3486
3487         result = proc_task_instantiate(dir, dentry, task, NULL);
3488 out_drop_task:
3489         put_task_struct(task);
3490 out:
3491         put_task_struct(leader);
3492 out_no_task:
3493         return ERR_PTR(result);
3494 }
3495
3496 /*
3497  * Find the first tid of a thread group to return to user space.
3498  *
3499  * Usually this is just the thread group leader, but if the users
3500  * buffer was too small or there was a seek into the middle of the
3501  * directory we have more work todo.
3502  *
3503  * In the case of a short read we start with find_task_by_pid.
3504  *
3505  * In the case of a seek we start with the leader and walk nr
3506  * threads past it.
3507  */
3508 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3509                                         struct pid_namespace *ns)
3510 {
3511         struct task_struct *pos, *task;
3512         unsigned long nr = f_pos;
3513
3514         if (nr != f_pos)        /* 32bit overflow? */
3515                 return NULL;
3516
3517         rcu_read_lock();
3518         task = pid_task(pid, PIDTYPE_PID);
3519         if (!task)
3520                 goto fail;
3521
3522         /* Attempt to start with the tid of a thread */
3523         if (tid && nr) {
3524                 pos = find_task_by_pid_ns(tid, ns);
3525                 if (pos && same_thread_group(pos, task))
3526                         goto found;
3527         }
3528
3529         /* If nr exceeds the number of threads there is nothing todo */
3530         if (nr >= get_nr_threads(task))
3531                 goto fail;
3532
3533         /* If we haven't found our starting place yet start
3534          * with the leader and walk nr threads forward.
3535          */
3536         pos = task = task->group_leader;
3537         do {
3538                 if (!nr--)
3539                         goto found;
3540         } while_each_thread(task, pos);
3541 fail:
3542         pos = NULL;
3543         goto out;
3544 found:
3545         get_task_struct(pos);
3546 out:
3547         rcu_read_unlock();
3548         return pos;
3549 }
3550
3551 /*
3552  * Find the next thread in the thread list.
3553  * Return NULL if there is an error or no next thread.
3554  *
3555  * The reference to the input task_struct is released.
3556  */
3557 static struct task_struct *next_tid(struct task_struct *start)
3558 {
3559         struct task_struct *pos = NULL;
3560         rcu_read_lock();
3561         if (pid_alive(start)) {
3562                 pos = next_thread(start);
3563                 if (thread_group_leader(pos))
3564                         pos = NULL;
3565                 else
3566                         get_task_struct(pos);
3567         }
3568         rcu_read_unlock();
3569         put_task_struct(start);
3570         return pos;
3571 }
3572
3573 /* for the /proc/TGID/task/ directories */
3574 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3575 {
3576         struct inode *inode = file_inode(file);
3577         struct task_struct *task;
3578         struct pid_namespace *ns;
3579         int tid;
3580
3581         if (proc_inode_is_dead(inode))
3582                 return -ENOENT;
3583
3584         if (!dir_emit_dots(file, ctx))
3585                 return 0;
3586
3587         /* f_version caches the tgid value that the last readdir call couldn't
3588          * return. lseek aka telldir automagically resets f_version to 0.
3589          */
3590         ns = inode->i_sb->s_fs_info;
3591         tid = (int)file->f_version;
3592         file->f_version = 0;
3593         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3594              task;
3595              task = next_tid(task), ctx->pos++) {
3596                 char name[10 + 1];
3597                 int len;
3598                 tid = task_pid_nr_ns(task, ns);
3599                 len = snprintf(name, sizeof(name), "%u", tid);
3600                 if (!proc_fill_cache(file, ctx, name, len,
3601                                 proc_task_instantiate, task, NULL)) {
3602                         /* returning this tgid failed, save it as the first
3603                          * pid for the next readir call */
3604                         file->f_version = (u64)tid;
3605                         put_task_struct(task);
3606                         break;
3607                 }
3608         }
3609
3610         return 0;
3611 }
3612
3613 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3614                              u32 request_mask, unsigned int query_flags)
3615 {
3616         struct inode *inode = d_inode(path->dentry);
3617         struct task_struct *p = get_proc_task(inode);
3618         generic_fillattr(inode, stat);
3619
3620         if (p) {
3621                 stat->nlink += get_nr_threads(p);
3622                 put_task_struct(p);
3623         }
3624
3625         return 0;
3626 }
3627
3628 static const struct inode_operations proc_task_inode_operations = {
3629         .lookup         = proc_task_lookup,
3630         .getattr        = proc_task_getattr,
3631         .setattr        = proc_setattr,
3632         .permission     = proc_pid_permission,
3633 };