fault-inject: automatically detect the number base for fail-nth write interface
[muen/linux.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <linux/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/sched/autogroup.h>
89 #include <linux/sched/mm.h>
90 #include <linux/sched/coredump.h>
91 #include <linux/sched/debug.h>
92 #include <linux/sched/stat.h>
93 #include <linux/flex_array.h>
94 #include <linux/posix-timers.h>
95 #ifdef CONFIG_HARDWALL
96 #include <asm/hardwall.h>
97 #endif
98 #include <trace/events/oom.h>
99 #include "internal.h"
100 #include "fd.h"
101
102 /* NOTE:
103  *      Implementing inode permission operations in /proc is almost
104  *      certainly an error.  Permission checks need to happen during
105  *      each system call not at open time.  The reason is that most of
106  *      what we wish to check for permissions in /proc varies at runtime.
107  *
108  *      The classic example of a problem is opening file descriptors
109  *      in /proc for a task before it execs a suid executable.
110  */
111
112 static u8 nlink_tid;
113 static u8 nlink_tgid;
114
115 struct pid_entry {
116         const char *name;
117         unsigned int len;
118         umode_t mode;
119         const struct inode_operations *iop;
120         const struct file_operations *fop;
121         union proc_op op;
122 };
123
124 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
125         .name = (NAME),                                 \
126         .len  = sizeof(NAME) - 1,                       \
127         .mode = MODE,                                   \
128         .iop  = IOP,                                    \
129         .fop  = FOP,                                    \
130         .op   = OP,                                     \
131 }
132
133 #define DIR(NAME, MODE, iops, fops)     \
134         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
135 #define LNK(NAME, get_link)                                     \
136         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
137                 &proc_pid_link_inode_operations, NULL,          \
138                 { .proc_get_link = get_link } )
139 #define REG(NAME, MODE, fops)                           \
140         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
141 #define ONE(NAME, MODE, show)                           \
142         NOD(NAME, (S_IFREG|(MODE)),                     \
143                 NULL, &proc_single_file_operations,     \
144                 { .proc_show = show } )
145
146 /*
147  * Count the number of hardlinks for the pid_entry table, excluding the .
148  * and .. links.
149  */
150 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
151         unsigned int n)
152 {
153         unsigned int i;
154         unsigned int count;
155
156         count = 2;
157         for (i = 0; i < n; ++i) {
158                 if (S_ISDIR(entries[i].mode))
159                         ++count;
160         }
161
162         return count;
163 }
164
165 static int get_task_root(struct task_struct *task, struct path *root)
166 {
167         int result = -ENOENT;
168
169         task_lock(task);
170         if (task->fs) {
171                 get_fs_root(task->fs, root);
172                 result = 0;
173         }
174         task_unlock(task);
175         return result;
176 }
177
178 static int proc_cwd_link(struct dentry *dentry, struct path *path)
179 {
180         struct task_struct *task = get_proc_task(d_inode(dentry));
181         int result = -ENOENT;
182
183         if (task) {
184                 task_lock(task);
185                 if (task->fs) {
186                         get_fs_pwd(task->fs, path);
187                         result = 0;
188                 }
189                 task_unlock(task);
190                 put_task_struct(task);
191         }
192         return result;
193 }
194
195 static int proc_root_link(struct dentry *dentry, struct path *path)
196 {
197         struct task_struct *task = get_proc_task(d_inode(dentry));
198         int result = -ENOENT;
199
200         if (task) {
201                 result = get_task_root(task, path);
202                 put_task_struct(task);
203         }
204         return result;
205 }
206
207 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
208                                      size_t _count, loff_t *pos)
209 {
210         struct task_struct *tsk;
211         struct mm_struct *mm;
212         char *page;
213         unsigned long count = _count;
214         unsigned long arg_start, arg_end, env_start, env_end;
215         unsigned long len1, len2, len;
216         unsigned long p;
217         char c;
218         ssize_t rv;
219
220         BUG_ON(*pos < 0);
221
222         tsk = get_proc_task(file_inode(file));
223         if (!tsk)
224                 return -ESRCH;
225         mm = get_task_mm(tsk);
226         put_task_struct(tsk);
227         if (!mm)
228                 return 0;
229         /* Check if process spawned far enough to have cmdline. */
230         if (!mm->env_end) {
231                 rv = 0;
232                 goto out_mmput;
233         }
234
235         page = (char *)__get_free_page(GFP_TEMPORARY);
236         if (!page) {
237                 rv = -ENOMEM;
238                 goto out_mmput;
239         }
240
241         down_read(&mm->mmap_sem);
242         arg_start = mm->arg_start;
243         arg_end = mm->arg_end;
244         env_start = mm->env_start;
245         env_end = mm->env_end;
246         up_read(&mm->mmap_sem);
247
248         BUG_ON(arg_start > arg_end);
249         BUG_ON(env_start > env_end);
250
251         len1 = arg_end - arg_start;
252         len2 = env_end - env_start;
253
254         /* Empty ARGV. */
255         if (len1 == 0) {
256                 rv = 0;
257                 goto out_free_page;
258         }
259         /*
260          * Inherently racy -- command line shares address space
261          * with code and data.
262          */
263         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
264         if (rv <= 0)
265                 goto out_free_page;
266
267         rv = 0;
268
269         if (c == '\0') {
270                 /* Command line (set of strings) occupies whole ARGV. */
271                 if (len1 <= *pos)
272                         goto out_free_page;
273
274                 p = arg_start + *pos;
275                 len = len1 - *pos;
276                 while (count > 0 && len > 0) {
277                         unsigned int _count;
278                         int nr_read;
279
280                         _count = min3(count, len, PAGE_SIZE);
281                         nr_read = access_remote_vm(mm, p, page, _count, 0);
282                         if (nr_read < 0)
283                                 rv = nr_read;
284                         if (nr_read <= 0)
285                                 goto out_free_page;
286
287                         if (copy_to_user(buf, page, nr_read)) {
288                                 rv = -EFAULT;
289                                 goto out_free_page;
290                         }
291
292                         p       += nr_read;
293                         len     -= nr_read;
294                         buf     += nr_read;
295                         count   -= nr_read;
296                         rv      += nr_read;
297                 }
298         } else {
299                 /*
300                  * Command line (1 string) occupies ARGV and
301                  * extends into ENVP.
302                  */
303                 struct {
304                         unsigned long p;
305                         unsigned long len;
306                 } cmdline[2] = {
307                         { .p = arg_start, .len = len1 },
308                         { .p = env_start, .len = len2 },
309                 };
310                 loff_t pos1 = *pos;
311                 unsigned int i;
312
313                 i = 0;
314                 while (i < 2 && pos1 >= cmdline[i].len) {
315                         pos1 -= cmdline[i].len;
316                         i++;
317                 }
318                 while (i < 2) {
319                         p = cmdline[i].p + pos1;
320                         len = cmdline[i].len - pos1;
321                         while (count > 0 && len > 0) {
322                                 unsigned int _count, l;
323                                 int nr_read;
324                                 bool final;
325
326                                 _count = min3(count, len, PAGE_SIZE);
327                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
328                                 if (nr_read < 0)
329                                         rv = nr_read;
330                                 if (nr_read <= 0)
331                                         goto out_free_page;
332
333                                 /*
334                                  * Command line can be shorter than whole ARGV
335                                  * even if last "marker" byte says it is not.
336                                  */
337                                 final = false;
338                                 l = strnlen(page, nr_read);
339                                 if (l < nr_read) {
340                                         nr_read = l;
341                                         final = true;
342                                 }
343
344                                 if (copy_to_user(buf, page, nr_read)) {
345                                         rv = -EFAULT;
346                                         goto out_free_page;
347                                 }
348
349                                 p       += nr_read;
350                                 len     -= nr_read;
351                                 buf     += nr_read;
352                                 count   -= nr_read;
353                                 rv      += nr_read;
354
355                                 if (final)
356                                         goto out_free_page;
357                         }
358
359                         /* Only first chunk can be read partially. */
360                         pos1 = 0;
361                         i++;
362                 }
363         }
364
365 out_free_page:
366         free_page((unsigned long)page);
367 out_mmput:
368         mmput(mm);
369         if (rv > 0)
370                 *pos += rv;
371         return rv;
372 }
373
374 static const struct file_operations proc_pid_cmdline_ops = {
375         .read   = proc_pid_cmdline_read,
376         .llseek = generic_file_llseek,
377 };
378
379 #ifdef CONFIG_KALLSYMS
380 /*
381  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
382  * Returns the resolved symbol.  If that fails, simply return the address.
383  */
384 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
385                           struct pid *pid, struct task_struct *task)
386 {
387         unsigned long wchan;
388         char symname[KSYM_NAME_LEN];
389
390         wchan = get_wchan(task);
391
392         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
393                         && !lookup_symbol_name(wchan, symname))
394                 seq_printf(m, "%s", symname);
395         else
396                 seq_putc(m, '0');
397
398         return 0;
399 }
400 #endif /* CONFIG_KALLSYMS */
401
402 static int lock_trace(struct task_struct *task)
403 {
404         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
405         if (err)
406                 return err;
407         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
408                 mutex_unlock(&task->signal->cred_guard_mutex);
409                 return -EPERM;
410         }
411         return 0;
412 }
413
414 static void unlock_trace(struct task_struct *task)
415 {
416         mutex_unlock(&task->signal->cred_guard_mutex);
417 }
418
419 #ifdef CONFIG_STACKTRACE
420
421 #define MAX_STACK_TRACE_DEPTH   64
422
423 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
424                           struct pid *pid, struct task_struct *task)
425 {
426         struct stack_trace trace;
427         unsigned long *entries;
428         int err;
429         int i;
430
431         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
432         if (!entries)
433                 return -ENOMEM;
434
435         trace.nr_entries        = 0;
436         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
437         trace.entries           = entries;
438         trace.skip              = 0;
439
440         err = lock_trace(task);
441         if (!err) {
442                 save_stack_trace_tsk(task, &trace);
443
444                 for (i = 0; i < trace.nr_entries; i++) {
445                         seq_printf(m, "[<%pK>] %pB\n",
446                                    (void *)entries[i], (void *)entries[i]);
447                 }
448                 unlock_trace(task);
449         }
450         kfree(entries);
451
452         return err;
453 }
454 #endif
455
456 #ifdef CONFIG_SCHED_INFO
457 /*
458  * Provides /proc/PID/schedstat
459  */
460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
461                               struct pid *pid, struct task_struct *task)
462 {
463         if (unlikely(!sched_info_on()))
464                 seq_printf(m, "0 0 0\n");
465         else
466                 seq_printf(m, "%llu %llu %lu\n",
467                    (unsigned long long)task->se.sum_exec_runtime,
468                    (unsigned long long)task->sched_info.run_delay,
469                    task->sched_info.pcount);
470
471         return 0;
472 }
473 #endif
474
475 #ifdef CONFIG_LATENCYTOP
476 static int lstats_show_proc(struct seq_file *m, void *v)
477 {
478         int i;
479         struct inode *inode = m->private;
480         struct task_struct *task = get_proc_task(inode);
481
482         if (!task)
483                 return -ESRCH;
484         seq_puts(m, "Latency Top version : v0.1\n");
485         for (i = 0; i < 32; i++) {
486                 struct latency_record *lr = &task->latency_record[i];
487                 if (lr->backtrace[0]) {
488                         int q;
489                         seq_printf(m, "%i %li %li",
490                                    lr->count, lr->time, lr->max);
491                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
492                                 unsigned long bt = lr->backtrace[q];
493                                 if (!bt)
494                                         break;
495                                 if (bt == ULONG_MAX)
496                                         break;
497                                 seq_printf(m, " %ps", (void *)bt);
498                         }
499                         seq_putc(m, '\n');
500                 }
501
502         }
503         put_task_struct(task);
504         return 0;
505 }
506
507 static int lstats_open(struct inode *inode, struct file *file)
508 {
509         return single_open(file, lstats_show_proc, inode);
510 }
511
512 static ssize_t lstats_write(struct file *file, const char __user *buf,
513                             size_t count, loff_t *offs)
514 {
515         struct task_struct *task = get_proc_task(file_inode(file));
516
517         if (!task)
518                 return -ESRCH;
519         clear_all_latency_tracing(task);
520         put_task_struct(task);
521
522         return count;
523 }
524
525 static const struct file_operations proc_lstats_operations = {
526         .open           = lstats_open,
527         .read           = seq_read,
528         .write          = lstats_write,
529         .llseek         = seq_lseek,
530         .release        = single_release,
531 };
532
533 #endif
534
535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
536                           struct pid *pid, struct task_struct *task)
537 {
538         unsigned long totalpages = totalram_pages + total_swap_pages;
539         unsigned long points = 0;
540
541         points = oom_badness(task, NULL, NULL, totalpages) *
542                                         1000 / totalpages;
543         seq_printf(m, "%lu\n", points);
544
545         return 0;
546 }
547
548 struct limit_names {
549         const char *name;
550         const char *unit;
551 };
552
553 static const struct limit_names lnames[RLIM_NLIMITS] = {
554         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
555         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
556         [RLIMIT_DATA] = {"Max data size", "bytes"},
557         [RLIMIT_STACK] = {"Max stack size", "bytes"},
558         [RLIMIT_CORE] = {"Max core file size", "bytes"},
559         [RLIMIT_RSS] = {"Max resident set", "bytes"},
560         [RLIMIT_NPROC] = {"Max processes", "processes"},
561         [RLIMIT_NOFILE] = {"Max open files", "files"},
562         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
563         [RLIMIT_AS] = {"Max address space", "bytes"},
564         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
565         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
566         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
567         [RLIMIT_NICE] = {"Max nice priority", NULL},
568         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
569         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
570 };
571
572 /* Display limits for a process */
573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
574                            struct pid *pid, struct task_struct *task)
575 {
576         unsigned int i;
577         unsigned long flags;
578
579         struct rlimit rlim[RLIM_NLIMITS];
580
581         if (!lock_task_sighand(task, &flags))
582                 return 0;
583         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
584         unlock_task_sighand(task, &flags);
585
586         /*
587          * print the file header
588          */
589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
590                   "Limit", "Soft Limit", "Hard Limit", "Units");
591
592         for (i = 0; i < RLIM_NLIMITS; i++) {
593                 if (rlim[i].rlim_cur == RLIM_INFINITY)
594                         seq_printf(m, "%-25s %-20s ",
595                                    lnames[i].name, "unlimited");
596                 else
597                         seq_printf(m, "%-25s %-20lu ",
598                                    lnames[i].name, rlim[i].rlim_cur);
599
600                 if (rlim[i].rlim_max == RLIM_INFINITY)
601                         seq_printf(m, "%-20s ", "unlimited");
602                 else
603                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
604
605                 if (lnames[i].unit)
606                         seq_printf(m, "%-10s\n", lnames[i].unit);
607                 else
608                         seq_putc(m, '\n');
609         }
610
611         return 0;
612 }
613
614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
616                             struct pid *pid, struct task_struct *task)
617 {
618         long nr;
619         unsigned long args[6], sp, pc;
620         int res;
621
622         res = lock_trace(task);
623         if (res)
624                 return res;
625
626         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
627                 seq_puts(m, "running\n");
628         else if (nr < 0)
629                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
630         else
631                 seq_printf(m,
632                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
633                        nr,
634                        args[0], args[1], args[2], args[3], args[4], args[5],
635                        sp, pc);
636         unlock_trace(task);
637
638         return 0;
639 }
640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
641
642 /************************************************************************/
643 /*                       Here the fs part begins                        */
644 /************************************************************************/
645
646 /* permission checks */
647 static int proc_fd_access_allowed(struct inode *inode)
648 {
649         struct task_struct *task;
650         int allowed = 0;
651         /* Allow access to a task's file descriptors if it is us or we
652          * may use ptrace attach to the process and find out that
653          * information.
654          */
655         task = get_proc_task(inode);
656         if (task) {
657                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
658                 put_task_struct(task);
659         }
660         return allowed;
661 }
662
663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
664 {
665         int error;
666         struct inode *inode = d_inode(dentry);
667
668         if (attr->ia_valid & ATTR_MODE)
669                 return -EPERM;
670
671         error = setattr_prepare(dentry, attr);
672         if (error)
673                 return error;
674
675         setattr_copy(inode, attr);
676         mark_inode_dirty(inode);
677         return 0;
678 }
679
680 /*
681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
682  * or euid/egid (for hide_pid_min=2)?
683  */
684 static bool has_pid_permissions(struct pid_namespace *pid,
685                                  struct task_struct *task,
686                                  int hide_pid_min)
687 {
688         if (pid->hide_pid < hide_pid_min)
689                 return true;
690         if (in_group_p(pid->pid_gid))
691                 return true;
692         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
693 }
694
695
696 static int proc_pid_permission(struct inode *inode, int mask)
697 {
698         struct pid_namespace *pid = inode->i_sb->s_fs_info;
699         struct task_struct *task;
700         bool has_perms;
701
702         task = get_proc_task(inode);
703         if (!task)
704                 return -ESRCH;
705         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
706         put_task_struct(task);
707
708         if (!has_perms) {
709                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
710                         /*
711                          * Let's make getdents(), stat(), and open()
712                          * consistent with each other.  If a process
713                          * may not stat() a file, it shouldn't be seen
714                          * in procfs at all.
715                          */
716                         return -ENOENT;
717                 }
718
719                 return -EPERM;
720         }
721         return generic_permission(inode, mask);
722 }
723
724
725
726 static const struct inode_operations proc_def_inode_operations = {
727         .setattr        = proc_setattr,
728 };
729
730 static int proc_single_show(struct seq_file *m, void *v)
731 {
732         struct inode *inode = m->private;
733         struct pid_namespace *ns;
734         struct pid *pid;
735         struct task_struct *task;
736         int ret;
737
738         ns = inode->i_sb->s_fs_info;
739         pid = proc_pid(inode);
740         task = get_pid_task(pid, PIDTYPE_PID);
741         if (!task)
742                 return -ESRCH;
743
744         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
745
746         put_task_struct(task);
747         return ret;
748 }
749
750 static int proc_single_open(struct inode *inode, struct file *filp)
751 {
752         return single_open(filp, proc_single_show, inode);
753 }
754
755 static const struct file_operations proc_single_file_operations = {
756         .open           = proc_single_open,
757         .read           = seq_read,
758         .llseek         = seq_lseek,
759         .release        = single_release,
760 };
761
762
763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
764 {
765         struct task_struct *task = get_proc_task(inode);
766         struct mm_struct *mm = ERR_PTR(-ESRCH);
767
768         if (task) {
769                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
770                 put_task_struct(task);
771
772                 if (!IS_ERR_OR_NULL(mm)) {
773                         /* ensure this mm_struct can't be freed */
774                         mmgrab(mm);
775                         /* but do not pin its memory */
776                         mmput(mm);
777                 }
778         }
779
780         return mm;
781 }
782
783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
784 {
785         struct mm_struct *mm = proc_mem_open(inode, mode);
786
787         if (IS_ERR(mm))
788                 return PTR_ERR(mm);
789
790         file->private_data = mm;
791         return 0;
792 }
793
794 static int mem_open(struct inode *inode, struct file *file)
795 {
796         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
797
798         /* OK to pass negative loff_t, we can catch out-of-range */
799         file->f_mode |= FMODE_UNSIGNED_OFFSET;
800
801         return ret;
802 }
803
804 static ssize_t mem_rw(struct file *file, char __user *buf,
805                         size_t count, loff_t *ppos, int write)
806 {
807         struct mm_struct *mm = file->private_data;
808         unsigned long addr = *ppos;
809         ssize_t copied;
810         char *page;
811         unsigned int flags;
812
813         if (!mm)
814                 return 0;
815
816         page = (char *)__get_free_page(GFP_TEMPORARY);
817         if (!page)
818                 return -ENOMEM;
819
820         copied = 0;
821         if (!mmget_not_zero(mm))
822                 goto free;
823
824         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
825
826         while (count > 0) {
827                 int this_len = min_t(int, count, PAGE_SIZE);
828
829                 if (write && copy_from_user(page, buf, this_len)) {
830                         copied = -EFAULT;
831                         break;
832                 }
833
834                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
835                 if (!this_len) {
836                         if (!copied)
837                                 copied = -EIO;
838                         break;
839                 }
840
841                 if (!write && copy_to_user(buf, page, this_len)) {
842                         copied = -EFAULT;
843                         break;
844                 }
845
846                 buf += this_len;
847                 addr += this_len;
848                 copied += this_len;
849                 count -= this_len;
850         }
851         *ppos = addr;
852
853         mmput(mm);
854 free:
855         free_page((unsigned long) page);
856         return copied;
857 }
858
859 static ssize_t mem_read(struct file *file, char __user *buf,
860                         size_t count, loff_t *ppos)
861 {
862         return mem_rw(file, buf, count, ppos, 0);
863 }
864
865 static ssize_t mem_write(struct file *file, const char __user *buf,
866                          size_t count, loff_t *ppos)
867 {
868         return mem_rw(file, (char __user*)buf, count, ppos, 1);
869 }
870
871 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
872 {
873         switch (orig) {
874         case 0:
875                 file->f_pos = offset;
876                 break;
877         case 1:
878                 file->f_pos += offset;
879                 break;
880         default:
881                 return -EINVAL;
882         }
883         force_successful_syscall_return();
884         return file->f_pos;
885 }
886
887 static int mem_release(struct inode *inode, struct file *file)
888 {
889         struct mm_struct *mm = file->private_data;
890         if (mm)
891                 mmdrop(mm);
892         return 0;
893 }
894
895 static const struct file_operations proc_mem_operations = {
896         .llseek         = mem_lseek,
897         .read           = mem_read,
898         .write          = mem_write,
899         .open           = mem_open,
900         .release        = mem_release,
901 };
902
903 static int environ_open(struct inode *inode, struct file *file)
904 {
905         return __mem_open(inode, file, PTRACE_MODE_READ);
906 }
907
908 static ssize_t environ_read(struct file *file, char __user *buf,
909                         size_t count, loff_t *ppos)
910 {
911         char *page;
912         unsigned long src = *ppos;
913         int ret = 0;
914         struct mm_struct *mm = file->private_data;
915         unsigned long env_start, env_end;
916
917         /* Ensure the process spawned far enough to have an environment. */
918         if (!mm || !mm->env_end)
919                 return 0;
920
921         page = (char *)__get_free_page(GFP_TEMPORARY);
922         if (!page)
923                 return -ENOMEM;
924
925         ret = 0;
926         if (!mmget_not_zero(mm))
927                 goto free;
928
929         down_read(&mm->mmap_sem);
930         env_start = mm->env_start;
931         env_end = mm->env_end;
932         up_read(&mm->mmap_sem);
933
934         while (count > 0) {
935                 size_t this_len, max_len;
936                 int retval;
937
938                 if (src >= (env_end - env_start))
939                         break;
940
941                 this_len = env_end - (env_start + src);
942
943                 max_len = min_t(size_t, PAGE_SIZE, count);
944                 this_len = min(max_len, this_len);
945
946                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
947
948                 if (retval <= 0) {
949                         ret = retval;
950                         break;
951                 }
952
953                 if (copy_to_user(buf, page, retval)) {
954                         ret = -EFAULT;
955                         break;
956                 }
957
958                 ret += retval;
959                 src += retval;
960                 buf += retval;
961                 count -= retval;
962         }
963         *ppos = src;
964         mmput(mm);
965
966 free:
967         free_page((unsigned long) page);
968         return ret;
969 }
970
971 static const struct file_operations proc_environ_operations = {
972         .open           = environ_open,
973         .read           = environ_read,
974         .llseek         = generic_file_llseek,
975         .release        = mem_release,
976 };
977
978 static int auxv_open(struct inode *inode, struct file *file)
979 {
980         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
981 }
982
983 static ssize_t auxv_read(struct file *file, char __user *buf,
984                         size_t count, loff_t *ppos)
985 {
986         struct mm_struct *mm = file->private_data;
987         unsigned int nwords = 0;
988
989         if (!mm)
990                 return 0;
991         do {
992                 nwords += 2;
993         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
994         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
995                                        nwords * sizeof(mm->saved_auxv[0]));
996 }
997
998 static const struct file_operations proc_auxv_operations = {
999         .open           = auxv_open,
1000         .read           = auxv_read,
1001         .llseek         = generic_file_llseek,
1002         .release        = mem_release,
1003 };
1004
1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006                             loff_t *ppos)
1007 {
1008         struct task_struct *task = get_proc_task(file_inode(file));
1009         char buffer[PROC_NUMBUF];
1010         int oom_adj = OOM_ADJUST_MIN;
1011         size_t len;
1012
1013         if (!task)
1014                 return -ESRCH;
1015         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1016                 oom_adj = OOM_ADJUST_MAX;
1017         else
1018                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1019                           OOM_SCORE_ADJ_MAX;
1020         put_task_struct(task);
1021         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1022         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 }
1024
1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1026 {
1027         static DEFINE_MUTEX(oom_adj_mutex);
1028         struct mm_struct *mm = NULL;
1029         struct task_struct *task;
1030         int err = 0;
1031
1032         task = get_proc_task(file_inode(file));
1033         if (!task)
1034                 return -ESRCH;
1035
1036         mutex_lock(&oom_adj_mutex);
1037         if (legacy) {
1038                 if (oom_adj < task->signal->oom_score_adj &&
1039                                 !capable(CAP_SYS_RESOURCE)) {
1040                         err = -EACCES;
1041                         goto err_unlock;
1042                 }
1043                 /*
1044                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1045                  * /proc/pid/oom_score_adj instead.
1046                  */
1047                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048                           current->comm, task_pid_nr(current), task_pid_nr(task),
1049                           task_pid_nr(task));
1050         } else {
1051                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1052                                 !capable(CAP_SYS_RESOURCE)) {
1053                         err = -EACCES;
1054                         goto err_unlock;
1055                 }
1056         }
1057
1058         /*
1059          * Make sure we will check other processes sharing the mm if this is
1060          * not vfrok which wants its own oom_score_adj.
1061          * pin the mm so it doesn't go away and get reused after task_unlock
1062          */
1063         if (!task->vfork_done) {
1064                 struct task_struct *p = find_lock_task_mm(task);
1065
1066                 if (p) {
1067                         if (atomic_read(&p->mm->mm_users) > 1) {
1068                                 mm = p->mm;
1069                                 mmgrab(mm);
1070                         }
1071                         task_unlock(p);
1072                 }
1073         }
1074
1075         task->signal->oom_score_adj = oom_adj;
1076         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1077                 task->signal->oom_score_adj_min = (short)oom_adj;
1078         trace_oom_score_adj_update(task);
1079
1080         if (mm) {
1081                 struct task_struct *p;
1082
1083                 rcu_read_lock();
1084                 for_each_process(p) {
1085                         if (same_thread_group(task, p))
1086                                 continue;
1087
1088                         /* do not touch kernel threads or the global init */
1089                         if (p->flags & PF_KTHREAD || is_global_init(p))
1090                                 continue;
1091
1092                         task_lock(p);
1093                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1094                                 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",
1095                                                 task_pid_nr(p), p->comm,
1096                                                 p->signal->oom_score_adj, oom_adj,
1097                                                 task_pid_nr(task), task->comm);
1098                                 p->signal->oom_score_adj = oom_adj;
1099                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1100                                         p->signal->oom_score_adj_min = (short)oom_adj;
1101                         }
1102                         task_unlock(p);
1103                 }
1104                 rcu_read_unlock();
1105                 mmdrop(mm);
1106         }
1107 err_unlock:
1108         mutex_unlock(&oom_adj_mutex);
1109         put_task_struct(task);
1110         return err;
1111 }
1112
1113 /*
1114  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1115  * kernels.  The effective policy is defined by oom_score_adj, which has a
1116  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1117  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1118  * Processes that become oom disabled via oom_adj will still be oom disabled
1119  * with this implementation.
1120  *
1121  * oom_adj cannot be removed since existing userspace binaries use it.
1122  */
1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1124                              size_t count, loff_t *ppos)
1125 {
1126         char buffer[PROC_NUMBUF];
1127         int oom_adj;
1128         int err;
1129
1130         memset(buffer, 0, sizeof(buffer));
1131         if (count > sizeof(buffer) - 1)
1132                 count = sizeof(buffer) - 1;
1133         if (copy_from_user(buffer, buf, count)) {
1134                 err = -EFAULT;
1135                 goto out;
1136         }
1137
1138         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139         if (err)
1140                 goto out;
1141         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1142              oom_adj != OOM_DISABLE) {
1143                 err = -EINVAL;
1144                 goto out;
1145         }
1146
1147         /*
1148          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1149          * value is always attainable.
1150          */
1151         if (oom_adj == OOM_ADJUST_MAX)
1152                 oom_adj = OOM_SCORE_ADJ_MAX;
1153         else
1154                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1155
1156         err = __set_oom_adj(file, oom_adj, true);
1157 out:
1158         return err < 0 ? err : count;
1159 }
1160
1161 static const struct file_operations proc_oom_adj_operations = {
1162         .read           = oom_adj_read,
1163         .write          = oom_adj_write,
1164         .llseek         = generic_file_llseek,
1165 };
1166
1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1168                                         size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file_inode(file));
1171         char buffer[PROC_NUMBUF];
1172         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1173         size_t len;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         oom_score_adj = task->signal->oom_score_adj;
1178         put_task_struct(task);
1179         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1180         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182
1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1184                                         size_t count, loff_t *ppos)
1185 {
1186         char buffer[PROC_NUMBUF];
1187         int oom_score_adj;
1188         int err;
1189
1190         memset(buffer, 0, sizeof(buffer));
1191         if (count > sizeof(buffer) - 1)
1192                 count = sizeof(buffer) - 1;
1193         if (copy_from_user(buffer, buf, count)) {
1194                 err = -EFAULT;
1195                 goto out;
1196         }
1197
1198         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199         if (err)
1200                 goto out;
1201         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1202                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1203                 err = -EINVAL;
1204                 goto out;
1205         }
1206
1207         err = __set_oom_adj(file, oom_score_adj, false);
1208 out:
1209         return err < 0 ? err : count;
1210 }
1211
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213         .read           = oom_score_adj_read,
1214         .write          = oom_score_adj_write,
1215         .llseek         = default_llseek,
1216 };
1217
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 11
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221                                   size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file_inode(file);
1224         struct task_struct *task = get_proc_task(inode);
1225         ssize_t length;
1226         char tmpbuf[TMPBUFLEN];
1227
1228         if (!task)
1229                 return -ESRCH;
1230         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231                            from_kuid(file->f_cred->user_ns,
1232                                      audit_get_loginuid(task)));
1233         put_task_struct(task);
1234         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238                                    size_t count, loff_t *ppos)
1239 {
1240         struct inode * inode = file_inode(file);
1241         uid_t loginuid;
1242         kuid_t kloginuid;
1243         int rv;
1244
1245         rcu_read_lock();
1246         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247                 rcu_read_unlock();
1248                 return -EPERM;
1249         }
1250         rcu_read_unlock();
1251
1252         if (*ppos != 0) {
1253                 /* No partial writes. */
1254                 return -EINVAL;
1255         }
1256
1257         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258         if (rv < 0)
1259                 return rv;
1260
1261         /* is userspace tring to explicitly UNSET the loginuid? */
1262         if (loginuid == AUDIT_UID_UNSET) {
1263                 kloginuid = INVALID_UID;
1264         } else {
1265                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266                 if (!uid_valid(kloginuid))
1267                         return -EINVAL;
1268         }
1269
1270         rv = audit_set_loginuid(kloginuid);
1271         if (rv < 0)
1272                 return rv;
1273         return count;
1274 }
1275
1276 static const struct file_operations proc_loginuid_operations = {
1277         .read           = proc_loginuid_read,
1278         .write          = proc_loginuid_write,
1279         .llseek         = generic_file_llseek,
1280 };
1281
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283                                   size_t count, loff_t *ppos)
1284 {
1285         struct inode * inode = file_inode(file);
1286         struct task_struct *task = get_proc_task(inode);
1287         ssize_t length;
1288         char tmpbuf[TMPBUFLEN];
1289
1290         if (!task)
1291                 return -ESRCH;
1292         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293                                 audit_get_sessionid(task));
1294         put_task_struct(task);
1295         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297
1298 static const struct file_operations proc_sessionid_operations = {
1299         .read           = proc_sessionid_read,
1300         .llseek         = generic_file_llseek,
1301 };
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306                                       size_t count, loff_t *ppos)
1307 {
1308         struct task_struct *task = get_proc_task(file_inode(file));
1309         char buffer[PROC_NUMBUF];
1310         size_t len;
1311         int make_it_fail;
1312
1313         if (!task)
1314                 return -ESRCH;
1315         make_it_fail = task->make_it_fail;
1316         put_task_struct(task);
1317
1318         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319
1320         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324                         const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326         struct task_struct *task;
1327         char buffer[PROC_NUMBUF];
1328         int make_it_fail;
1329         int rv;
1330
1331         if (!capable(CAP_SYS_RESOURCE))
1332                 return -EPERM;
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339         if (rv < 0)
1340                 return rv;
1341         if (make_it_fail < 0 || make_it_fail > 1)
1342                 return -EINVAL;
1343
1344         task = get_proc_task(file_inode(file));
1345         if (!task)
1346                 return -ESRCH;
1347         task->make_it_fail = make_it_fail;
1348         put_task_struct(task);
1349
1350         return count;
1351 }
1352
1353 static const struct file_operations proc_fault_inject_operations = {
1354         .read           = proc_fault_inject_read,
1355         .write          = proc_fault_inject_write,
1356         .llseek         = generic_file_llseek,
1357 };
1358
1359 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1360                                    size_t count, loff_t *ppos)
1361 {
1362         struct task_struct *task;
1363         int err, n;
1364
1365         task = get_proc_task(file_inode(file));
1366         if (!task)
1367                 return -ESRCH;
1368         put_task_struct(task);
1369         if (task != current)
1370                 return -EPERM;
1371         err = kstrtoint_from_user(buf, count, 0, &n);
1372         if (err)
1373                 return err;
1374         if (n < 0 || n == INT_MAX)
1375                 return -EINVAL;
1376         current->fail_nth = n + 1;
1377         return count;
1378 }
1379
1380 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1381                                   size_t count, loff_t *ppos)
1382 {
1383         struct task_struct *task;
1384         int err;
1385
1386         task = get_proc_task(file_inode(file));
1387         if (!task)
1388                 return -ESRCH;
1389         put_task_struct(task);
1390         if (task != current)
1391                 return -EPERM;
1392         if (count < 1)
1393                 return -EINVAL;
1394         err = put_user((char)(current->fail_nth ? 'N' : 'Y'), buf);
1395         if (err)
1396                 return err;
1397         current->fail_nth = 0;
1398         return 1;
1399 }
1400
1401 static const struct file_operations proc_fail_nth_operations = {
1402         .read           = proc_fail_nth_read,
1403         .write          = proc_fail_nth_write,
1404 };
1405 #endif
1406
1407
1408 #ifdef CONFIG_SCHED_DEBUG
1409 /*
1410  * Print out various scheduling related per-task fields:
1411  */
1412 static int sched_show(struct seq_file *m, void *v)
1413 {
1414         struct inode *inode = m->private;
1415         struct task_struct *p;
1416
1417         p = get_proc_task(inode);
1418         if (!p)
1419                 return -ESRCH;
1420         proc_sched_show_task(p, m);
1421
1422         put_task_struct(p);
1423
1424         return 0;
1425 }
1426
1427 static ssize_t
1428 sched_write(struct file *file, const char __user *buf,
1429             size_t count, loff_t *offset)
1430 {
1431         struct inode *inode = file_inode(file);
1432         struct task_struct *p;
1433
1434         p = get_proc_task(inode);
1435         if (!p)
1436                 return -ESRCH;
1437         proc_sched_set_task(p);
1438
1439         put_task_struct(p);
1440
1441         return count;
1442 }
1443
1444 static int sched_open(struct inode *inode, struct file *filp)
1445 {
1446         return single_open(filp, sched_show, inode);
1447 }
1448
1449 static const struct file_operations proc_pid_sched_operations = {
1450         .open           = sched_open,
1451         .read           = seq_read,
1452         .write          = sched_write,
1453         .llseek         = seq_lseek,
1454         .release        = single_release,
1455 };
1456
1457 #endif
1458
1459 #ifdef CONFIG_SCHED_AUTOGROUP
1460 /*
1461  * Print out autogroup related information:
1462  */
1463 static int sched_autogroup_show(struct seq_file *m, void *v)
1464 {
1465         struct inode *inode = m->private;
1466         struct task_struct *p;
1467
1468         p = get_proc_task(inode);
1469         if (!p)
1470                 return -ESRCH;
1471         proc_sched_autogroup_show_task(p, m);
1472
1473         put_task_struct(p);
1474
1475         return 0;
1476 }
1477
1478 static ssize_t
1479 sched_autogroup_write(struct file *file, const char __user *buf,
1480             size_t count, loff_t *offset)
1481 {
1482         struct inode *inode = file_inode(file);
1483         struct task_struct *p;
1484         char buffer[PROC_NUMBUF];
1485         int nice;
1486         int err;
1487
1488         memset(buffer, 0, sizeof(buffer));
1489         if (count > sizeof(buffer) - 1)
1490                 count = sizeof(buffer) - 1;
1491         if (copy_from_user(buffer, buf, count))
1492                 return -EFAULT;
1493
1494         err = kstrtoint(strstrip(buffer), 0, &nice);
1495         if (err < 0)
1496                 return err;
1497
1498         p = get_proc_task(inode);
1499         if (!p)
1500                 return -ESRCH;
1501
1502         err = proc_sched_autogroup_set_nice(p, nice);
1503         if (err)
1504                 count = err;
1505
1506         put_task_struct(p);
1507
1508         return count;
1509 }
1510
1511 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1512 {
1513         int ret;
1514
1515         ret = single_open(filp, sched_autogroup_show, NULL);
1516         if (!ret) {
1517                 struct seq_file *m = filp->private_data;
1518
1519                 m->private = inode;
1520         }
1521         return ret;
1522 }
1523
1524 static const struct file_operations proc_pid_sched_autogroup_operations = {
1525         .open           = sched_autogroup_open,
1526         .read           = seq_read,
1527         .write          = sched_autogroup_write,
1528         .llseek         = seq_lseek,
1529         .release        = single_release,
1530 };
1531
1532 #endif /* CONFIG_SCHED_AUTOGROUP */
1533
1534 static ssize_t comm_write(struct file *file, const char __user *buf,
1535                                 size_t count, loff_t *offset)
1536 {
1537         struct inode *inode = file_inode(file);
1538         struct task_struct *p;
1539         char buffer[TASK_COMM_LEN];
1540         const size_t maxlen = sizeof(buffer) - 1;
1541
1542         memset(buffer, 0, sizeof(buffer));
1543         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1544                 return -EFAULT;
1545
1546         p = get_proc_task(inode);
1547         if (!p)
1548                 return -ESRCH;
1549
1550         if (same_thread_group(current, p))
1551                 set_task_comm(p, buffer);
1552         else
1553                 count = -EINVAL;
1554
1555         put_task_struct(p);
1556
1557         return count;
1558 }
1559
1560 static int comm_show(struct seq_file *m, void *v)
1561 {
1562         struct inode *inode = m->private;
1563         struct task_struct *p;
1564
1565         p = get_proc_task(inode);
1566         if (!p)
1567                 return -ESRCH;
1568
1569         task_lock(p);
1570         seq_printf(m, "%s\n", p->comm);
1571         task_unlock(p);
1572
1573         put_task_struct(p);
1574
1575         return 0;
1576 }
1577
1578 static int comm_open(struct inode *inode, struct file *filp)
1579 {
1580         return single_open(filp, comm_show, inode);
1581 }
1582
1583 static const struct file_operations proc_pid_set_comm_operations = {
1584         .open           = comm_open,
1585         .read           = seq_read,
1586         .write          = comm_write,
1587         .llseek         = seq_lseek,
1588         .release        = single_release,
1589 };
1590
1591 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1592 {
1593         struct task_struct *task;
1594         struct file *exe_file;
1595
1596         task = get_proc_task(d_inode(dentry));
1597         if (!task)
1598                 return -ENOENT;
1599         exe_file = get_task_exe_file(task);
1600         put_task_struct(task);
1601         if (exe_file) {
1602                 *exe_path = exe_file->f_path;
1603                 path_get(&exe_file->f_path);
1604                 fput(exe_file);
1605                 return 0;
1606         } else
1607                 return -ENOENT;
1608 }
1609
1610 static const char *proc_pid_get_link(struct dentry *dentry,
1611                                      struct inode *inode,
1612                                      struct delayed_call *done)
1613 {
1614         struct path path;
1615         int error = -EACCES;
1616
1617         if (!dentry)
1618                 return ERR_PTR(-ECHILD);
1619
1620         /* Are we allowed to snoop on the tasks file descriptors? */
1621         if (!proc_fd_access_allowed(inode))
1622                 goto out;
1623
1624         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1625         if (error)
1626                 goto out;
1627
1628         nd_jump_link(&path);
1629         return NULL;
1630 out:
1631         return ERR_PTR(error);
1632 }
1633
1634 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1635 {
1636         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1637         char *pathname;
1638         int len;
1639
1640         if (!tmp)
1641                 return -ENOMEM;
1642
1643         pathname = d_path(path, tmp, PAGE_SIZE);
1644         len = PTR_ERR(pathname);
1645         if (IS_ERR(pathname))
1646                 goto out;
1647         len = tmp + PAGE_SIZE - 1 - pathname;
1648
1649         if (len > buflen)
1650                 len = buflen;
1651         if (copy_to_user(buffer, pathname, len))
1652                 len = -EFAULT;
1653  out:
1654         free_page((unsigned long)tmp);
1655         return len;
1656 }
1657
1658 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1659 {
1660         int error = -EACCES;
1661         struct inode *inode = d_inode(dentry);
1662         struct path path;
1663
1664         /* Are we allowed to snoop on the tasks file descriptors? */
1665         if (!proc_fd_access_allowed(inode))
1666                 goto out;
1667
1668         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1669         if (error)
1670                 goto out;
1671
1672         error = do_proc_readlink(&path, buffer, buflen);
1673         path_put(&path);
1674 out:
1675         return error;
1676 }
1677
1678 const struct inode_operations proc_pid_link_inode_operations = {
1679         .readlink       = proc_pid_readlink,
1680         .get_link       = proc_pid_get_link,
1681         .setattr        = proc_setattr,
1682 };
1683
1684
1685 /* building an inode */
1686
1687 void task_dump_owner(struct task_struct *task, mode_t mode,
1688                      kuid_t *ruid, kgid_t *rgid)
1689 {
1690         /* Depending on the state of dumpable compute who should own a
1691          * proc file for a task.
1692          */
1693         const struct cred *cred;
1694         kuid_t uid;
1695         kgid_t gid;
1696
1697         /* Default to the tasks effective ownership */
1698         rcu_read_lock();
1699         cred = __task_cred(task);
1700         uid = cred->euid;
1701         gid = cred->egid;
1702         rcu_read_unlock();
1703
1704         /*
1705          * Before the /proc/pid/status file was created the only way to read
1706          * the effective uid of a /process was to stat /proc/pid.  Reading
1707          * /proc/pid/status is slow enough that procps and other packages
1708          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1709          * made this apply to all per process world readable and executable
1710          * directories.
1711          */
1712         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1713                 struct mm_struct *mm;
1714                 task_lock(task);
1715                 mm = task->mm;
1716                 /* Make non-dumpable tasks owned by some root */
1717                 if (mm) {
1718                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1719                                 struct user_namespace *user_ns = mm->user_ns;
1720
1721                                 uid = make_kuid(user_ns, 0);
1722                                 if (!uid_valid(uid))
1723                                         uid = GLOBAL_ROOT_UID;
1724
1725                                 gid = make_kgid(user_ns, 0);
1726                                 if (!gid_valid(gid))
1727                                         gid = GLOBAL_ROOT_GID;
1728                         }
1729                 } else {
1730                         uid = GLOBAL_ROOT_UID;
1731                         gid = GLOBAL_ROOT_GID;
1732                 }
1733                 task_unlock(task);
1734         }
1735         *ruid = uid;
1736         *rgid = gid;
1737 }
1738
1739 struct inode *proc_pid_make_inode(struct super_block * sb,
1740                                   struct task_struct *task, umode_t mode)
1741 {
1742         struct inode * inode;
1743         struct proc_inode *ei;
1744
1745         /* We need a new inode */
1746
1747         inode = new_inode(sb);
1748         if (!inode)
1749                 goto out;
1750
1751         /* Common stuff */
1752         ei = PROC_I(inode);
1753         inode->i_mode = mode;
1754         inode->i_ino = get_next_ino();
1755         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1756         inode->i_op = &proc_def_inode_operations;
1757
1758         /*
1759          * grab the reference to task.
1760          */
1761         ei->pid = get_task_pid(task, PIDTYPE_PID);
1762         if (!ei->pid)
1763                 goto out_unlock;
1764
1765         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1766         security_task_to_inode(task, inode);
1767
1768 out:
1769         return inode;
1770
1771 out_unlock:
1772         iput(inode);
1773         return NULL;
1774 }
1775
1776 int pid_getattr(const struct path *path, struct kstat *stat,
1777                 u32 request_mask, unsigned int query_flags)
1778 {
1779         struct inode *inode = d_inode(path->dentry);
1780         struct task_struct *task;
1781         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1782
1783         generic_fillattr(inode, stat);
1784
1785         rcu_read_lock();
1786         stat->uid = GLOBAL_ROOT_UID;
1787         stat->gid = GLOBAL_ROOT_GID;
1788         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1789         if (task) {
1790                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1791                         rcu_read_unlock();
1792                         /*
1793                          * This doesn't prevent learning whether PID exists,
1794                          * it only makes getattr() consistent with readdir().
1795                          */
1796                         return -ENOENT;
1797                 }
1798                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1799         }
1800         rcu_read_unlock();
1801         return 0;
1802 }
1803
1804 /* dentry stuff */
1805
1806 /*
1807  *      Exceptional case: normally we are not allowed to unhash a busy
1808  * directory. In this case, however, we can do it - no aliasing problems
1809  * due to the way we treat inodes.
1810  *
1811  * Rewrite the inode's ownerships here because the owning task may have
1812  * performed a setuid(), etc.
1813  *
1814  */
1815 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1816 {
1817         struct inode *inode;
1818         struct task_struct *task;
1819
1820         if (flags & LOOKUP_RCU)
1821                 return -ECHILD;
1822
1823         inode = d_inode(dentry);
1824         task = get_proc_task(inode);
1825
1826         if (task) {
1827                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1828
1829                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1830                 security_task_to_inode(task, inode);
1831                 put_task_struct(task);
1832                 return 1;
1833         }
1834         return 0;
1835 }
1836
1837 static inline bool proc_inode_is_dead(struct inode *inode)
1838 {
1839         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1840 }
1841
1842 int pid_delete_dentry(const struct dentry *dentry)
1843 {
1844         /* Is the task we represent dead?
1845          * If so, then don't put the dentry on the lru list,
1846          * kill it immediately.
1847          */
1848         return proc_inode_is_dead(d_inode(dentry));
1849 }
1850
1851 const struct dentry_operations pid_dentry_operations =
1852 {
1853         .d_revalidate   = pid_revalidate,
1854         .d_delete       = pid_delete_dentry,
1855 };
1856
1857 /* Lookups */
1858
1859 /*
1860  * Fill a directory entry.
1861  *
1862  * If possible create the dcache entry and derive our inode number and
1863  * file type from dcache entry.
1864  *
1865  * Since all of the proc inode numbers are dynamically generated, the inode
1866  * numbers do not exist until the inode is cache.  This means creating the
1867  * the dcache entry in readdir is necessary to keep the inode numbers
1868  * reported by readdir in sync with the inode numbers reported
1869  * by stat.
1870  */
1871 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1872         const char *name, int len,
1873         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1874 {
1875         struct dentry *child, *dir = file->f_path.dentry;
1876         struct qstr qname = QSTR_INIT(name, len);
1877         struct inode *inode;
1878         unsigned type;
1879         ino_t ino;
1880
1881         child = d_hash_and_lookup(dir, &qname);
1882         if (!child) {
1883                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1884                 child = d_alloc_parallel(dir, &qname, &wq);
1885                 if (IS_ERR(child))
1886                         goto end_instantiate;
1887                 if (d_in_lookup(child)) {
1888                         int err = instantiate(d_inode(dir), child, task, ptr);
1889                         d_lookup_done(child);
1890                         if (err < 0) {
1891                                 dput(child);
1892                                 goto end_instantiate;
1893                         }
1894                 }
1895         }
1896         inode = d_inode(child);
1897         ino = inode->i_ino;
1898         type = inode->i_mode >> 12;
1899         dput(child);
1900         return dir_emit(ctx, name, len, ino, type);
1901
1902 end_instantiate:
1903         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1904 }
1905
1906 /*
1907  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1908  * which represent vma start and end addresses.
1909  */
1910 static int dname_to_vma_addr(struct dentry *dentry,
1911                              unsigned long *start, unsigned long *end)
1912 {
1913         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1914                 return -EINVAL;
1915
1916         return 0;
1917 }
1918
1919 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1920 {
1921         unsigned long vm_start, vm_end;
1922         bool exact_vma_exists = false;
1923         struct mm_struct *mm = NULL;
1924         struct task_struct *task;
1925         struct inode *inode;
1926         int status = 0;
1927
1928         if (flags & LOOKUP_RCU)
1929                 return -ECHILD;
1930
1931         inode = d_inode(dentry);
1932         task = get_proc_task(inode);
1933         if (!task)
1934                 goto out_notask;
1935
1936         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1937         if (IS_ERR_OR_NULL(mm))
1938                 goto out;
1939
1940         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1941                 down_read(&mm->mmap_sem);
1942                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1943                 up_read(&mm->mmap_sem);
1944         }
1945
1946         mmput(mm);
1947
1948         if (exact_vma_exists) {
1949                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1950
1951                 security_task_to_inode(task, inode);
1952                 status = 1;
1953         }
1954
1955 out:
1956         put_task_struct(task);
1957
1958 out_notask:
1959         return status;
1960 }
1961
1962 static const struct dentry_operations tid_map_files_dentry_operations = {
1963         .d_revalidate   = map_files_d_revalidate,
1964         .d_delete       = pid_delete_dentry,
1965 };
1966
1967 static int map_files_get_link(struct dentry *dentry, struct path *path)
1968 {
1969         unsigned long vm_start, vm_end;
1970         struct vm_area_struct *vma;
1971         struct task_struct *task;
1972         struct mm_struct *mm;
1973         int rc;
1974
1975         rc = -ENOENT;
1976         task = get_proc_task(d_inode(dentry));
1977         if (!task)
1978                 goto out;
1979
1980         mm = get_task_mm(task);
1981         put_task_struct(task);
1982         if (!mm)
1983                 goto out;
1984
1985         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1986         if (rc)
1987                 goto out_mmput;
1988
1989         rc = -ENOENT;
1990         down_read(&mm->mmap_sem);
1991         vma = find_exact_vma(mm, vm_start, vm_end);
1992         if (vma && vma->vm_file) {
1993                 *path = vma->vm_file->f_path;
1994                 path_get(path);
1995                 rc = 0;
1996         }
1997         up_read(&mm->mmap_sem);
1998
1999 out_mmput:
2000         mmput(mm);
2001 out:
2002         return rc;
2003 }
2004
2005 struct map_files_info {
2006         fmode_t         mode;
2007         unsigned int    len;
2008         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2009 };
2010
2011 /*
2012  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2013  * symlinks may be used to bypass permissions on ancestor directories in the
2014  * path to the file in question.
2015  */
2016 static const char *
2017 proc_map_files_get_link(struct dentry *dentry,
2018                         struct inode *inode,
2019                         struct delayed_call *done)
2020 {
2021         if (!capable(CAP_SYS_ADMIN))
2022                 return ERR_PTR(-EPERM);
2023
2024         return proc_pid_get_link(dentry, inode, done);
2025 }
2026
2027 /*
2028  * Identical to proc_pid_link_inode_operations except for get_link()
2029  */
2030 static const struct inode_operations proc_map_files_link_inode_operations = {
2031         .readlink       = proc_pid_readlink,
2032         .get_link       = proc_map_files_get_link,
2033         .setattr        = proc_setattr,
2034 };
2035
2036 static int
2037 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2038                            struct task_struct *task, const void *ptr)
2039 {
2040         fmode_t mode = (fmode_t)(unsigned long)ptr;
2041         struct proc_inode *ei;
2042         struct inode *inode;
2043
2044         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2045                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2046                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2047         if (!inode)
2048                 return -ENOENT;
2049
2050         ei = PROC_I(inode);
2051         ei->op.proc_get_link = map_files_get_link;
2052
2053         inode->i_op = &proc_map_files_link_inode_operations;
2054         inode->i_size = 64;
2055
2056         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2057         d_add(dentry, inode);
2058
2059         return 0;
2060 }
2061
2062 static struct dentry *proc_map_files_lookup(struct inode *dir,
2063                 struct dentry *dentry, unsigned int flags)
2064 {
2065         unsigned long vm_start, vm_end;
2066         struct vm_area_struct *vma;
2067         struct task_struct *task;
2068         int result;
2069         struct mm_struct *mm;
2070
2071         result = -ENOENT;
2072         task = get_proc_task(dir);
2073         if (!task)
2074                 goto out;
2075
2076         result = -EACCES;
2077         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2078                 goto out_put_task;
2079
2080         result = -ENOENT;
2081         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2082                 goto out_put_task;
2083
2084         mm = get_task_mm(task);
2085         if (!mm)
2086                 goto out_put_task;
2087
2088         down_read(&mm->mmap_sem);
2089         vma = find_exact_vma(mm, vm_start, vm_end);
2090         if (!vma)
2091                 goto out_no_vma;
2092
2093         if (vma->vm_file)
2094                 result = proc_map_files_instantiate(dir, dentry, task,
2095                                 (void *)(unsigned long)vma->vm_file->f_mode);
2096
2097 out_no_vma:
2098         up_read(&mm->mmap_sem);
2099         mmput(mm);
2100 out_put_task:
2101         put_task_struct(task);
2102 out:
2103         return ERR_PTR(result);
2104 }
2105
2106 static const struct inode_operations proc_map_files_inode_operations = {
2107         .lookup         = proc_map_files_lookup,
2108         .permission     = proc_fd_permission,
2109         .setattr        = proc_setattr,
2110 };
2111
2112 static int
2113 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2114 {
2115         struct vm_area_struct *vma;
2116         struct task_struct *task;
2117         struct mm_struct *mm;
2118         unsigned long nr_files, pos, i;
2119         struct flex_array *fa = NULL;
2120         struct map_files_info info;
2121         struct map_files_info *p;
2122         int ret;
2123
2124         ret = -ENOENT;
2125         task = get_proc_task(file_inode(file));
2126         if (!task)
2127                 goto out;
2128
2129         ret = -EACCES;
2130         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2131                 goto out_put_task;
2132
2133         ret = 0;
2134         if (!dir_emit_dots(file, ctx))
2135                 goto out_put_task;
2136
2137         mm = get_task_mm(task);
2138         if (!mm)
2139                 goto out_put_task;
2140         down_read(&mm->mmap_sem);
2141
2142         nr_files = 0;
2143
2144         /*
2145          * We need two passes here:
2146          *
2147          *  1) Collect vmas of mapped files with mmap_sem taken
2148          *  2) Release mmap_sem and instantiate entries
2149          *
2150          * otherwise we get lockdep complained, since filldir()
2151          * routine might require mmap_sem taken in might_fault().
2152          */
2153
2154         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2155                 if (vma->vm_file && ++pos > ctx->pos)
2156                         nr_files++;
2157         }
2158
2159         if (nr_files) {
2160                 fa = flex_array_alloc(sizeof(info), nr_files,
2161                                         GFP_KERNEL);
2162                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2163                                                 GFP_KERNEL)) {
2164                         ret = -ENOMEM;
2165                         if (fa)
2166                                 flex_array_free(fa);
2167                         up_read(&mm->mmap_sem);
2168                         mmput(mm);
2169                         goto out_put_task;
2170                 }
2171                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2172                                 vma = vma->vm_next) {
2173                         if (!vma->vm_file)
2174                                 continue;
2175                         if (++pos <= ctx->pos)
2176                                 continue;
2177
2178                         info.mode = vma->vm_file->f_mode;
2179                         info.len = snprintf(info.name,
2180                                         sizeof(info.name), "%lx-%lx",
2181                                         vma->vm_start, vma->vm_end);
2182                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2183                                 BUG();
2184                 }
2185         }
2186         up_read(&mm->mmap_sem);
2187
2188         for (i = 0; i < nr_files; i++) {
2189                 p = flex_array_get(fa, i);
2190                 if (!proc_fill_cache(file, ctx,
2191                                       p->name, p->len,
2192                                       proc_map_files_instantiate,
2193                                       task,
2194                                       (void *)(unsigned long)p->mode))
2195                         break;
2196                 ctx->pos++;
2197         }
2198         if (fa)
2199                 flex_array_free(fa);
2200         mmput(mm);
2201
2202 out_put_task:
2203         put_task_struct(task);
2204 out:
2205         return ret;
2206 }
2207
2208 static const struct file_operations proc_map_files_operations = {
2209         .read           = generic_read_dir,
2210         .iterate_shared = proc_map_files_readdir,
2211         .llseek         = generic_file_llseek,
2212 };
2213
2214 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2215 struct timers_private {
2216         struct pid *pid;
2217         struct task_struct *task;
2218         struct sighand_struct *sighand;
2219         struct pid_namespace *ns;
2220         unsigned long flags;
2221 };
2222
2223 static void *timers_start(struct seq_file *m, loff_t *pos)
2224 {
2225         struct timers_private *tp = m->private;
2226
2227         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2228         if (!tp->task)
2229                 return ERR_PTR(-ESRCH);
2230
2231         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2232         if (!tp->sighand)
2233                 return ERR_PTR(-ESRCH);
2234
2235         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2236 }
2237
2238 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2239 {
2240         struct timers_private *tp = m->private;
2241         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2242 }
2243
2244 static void timers_stop(struct seq_file *m, void *v)
2245 {
2246         struct timers_private *tp = m->private;
2247
2248         if (tp->sighand) {
2249                 unlock_task_sighand(tp->task, &tp->flags);
2250                 tp->sighand = NULL;
2251         }
2252
2253         if (tp->task) {
2254                 put_task_struct(tp->task);
2255                 tp->task = NULL;
2256         }
2257 }
2258
2259 static int show_timer(struct seq_file *m, void *v)
2260 {
2261         struct k_itimer *timer;
2262         struct timers_private *tp = m->private;
2263         int notify;
2264         static const char * const nstr[] = {
2265                 [SIGEV_SIGNAL] = "signal",
2266                 [SIGEV_NONE] = "none",
2267                 [SIGEV_THREAD] = "thread",
2268         };
2269
2270         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2271         notify = timer->it_sigev_notify;
2272
2273         seq_printf(m, "ID: %d\n", timer->it_id);
2274         seq_printf(m, "signal: %d/%p\n",
2275                    timer->sigq->info.si_signo,
2276                    timer->sigq->info.si_value.sival_ptr);
2277         seq_printf(m, "notify: %s/%s.%d\n",
2278                    nstr[notify & ~SIGEV_THREAD_ID],
2279                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2280                    pid_nr_ns(timer->it_pid, tp->ns));
2281         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2282
2283         return 0;
2284 }
2285
2286 static const struct seq_operations proc_timers_seq_ops = {
2287         .start  = timers_start,
2288         .next   = timers_next,
2289         .stop   = timers_stop,
2290         .show   = show_timer,
2291 };
2292
2293 static int proc_timers_open(struct inode *inode, struct file *file)
2294 {
2295         struct timers_private *tp;
2296
2297         tp = __seq_open_private(file, &proc_timers_seq_ops,
2298                         sizeof(struct timers_private));
2299         if (!tp)
2300                 return -ENOMEM;
2301
2302         tp->pid = proc_pid(inode);
2303         tp->ns = inode->i_sb->s_fs_info;
2304         return 0;
2305 }
2306
2307 static const struct file_operations proc_timers_operations = {
2308         .open           = proc_timers_open,
2309         .read           = seq_read,
2310         .llseek         = seq_lseek,
2311         .release        = seq_release_private,
2312 };
2313 #endif
2314
2315 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2316                                         size_t count, loff_t *offset)
2317 {
2318         struct inode *inode = file_inode(file);
2319         struct task_struct *p;
2320         u64 slack_ns;
2321         int err;
2322
2323         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2324         if (err < 0)
2325                 return err;
2326
2327         p = get_proc_task(inode);
2328         if (!p)
2329                 return -ESRCH;
2330
2331         if (p != current) {
2332                 if (!capable(CAP_SYS_NICE)) {
2333                         count = -EPERM;
2334                         goto out;
2335                 }
2336
2337                 err = security_task_setscheduler(p);
2338                 if (err) {
2339                         count = err;
2340                         goto out;
2341                 }
2342         }
2343
2344         task_lock(p);
2345         if (slack_ns == 0)
2346                 p->timer_slack_ns = p->default_timer_slack_ns;
2347         else
2348                 p->timer_slack_ns = slack_ns;
2349         task_unlock(p);
2350
2351 out:
2352         put_task_struct(p);
2353
2354         return count;
2355 }
2356
2357 static int timerslack_ns_show(struct seq_file *m, void *v)
2358 {
2359         struct inode *inode = m->private;
2360         struct task_struct *p;
2361         int err = 0;
2362
2363         p = get_proc_task(inode);
2364         if (!p)
2365                 return -ESRCH;
2366
2367         if (p != current) {
2368
2369                 if (!capable(CAP_SYS_NICE)) {
2370                         err = -EPERM;
2371                         goto out;
2372                 }
2373                 err = security_task_getscheduler(p);
2374                 if (err)
2375                         goto out;
2376         }
2377
2378         task_lock(p);
2379         seq_printf(m, "%llu\n", p->timer_slack_ns);
2380         task_unlock(p);
2381
2382 out:
2383         put_task_struct(p);
2384
2385         return err;
2386 }
2387
2388 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2389 {
2390         return single_open(filp, timerslack_ns_show, inode);
2391 }
2392
2393 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2394         .open           = timerslack_ns_open,
2395         .read           = seq_read,
2396         .write          = timerslack_ns_write,
2397         .llseek         = seq_lseek,
2398         .release        = single_release,
2399 };
2400
2401 static int proc_pident_instantiate(struct inode *dir,
2402         struct dentry *dentry, struct task_struct *task, const void *ptr)
2403 {
2404         const struct pid_entry *p = ptr;
2405         struct inode *inode;
2406         struct proc_inode *ei;
2407
2408         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2409         if (!inode)
2410                 goto out;
2411
2412         ei = PROC_I(inode);
2413         if (S_ISDIR(inode->i_mode))
2414                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2415         if (p->iop)
2416                 inode->i_op = p->iop;
2417         if (p->fop)
2418                 inode->i_fop = p->fop;
2419         ei->op = p->op;
2420         d_set_d_op(dentry, &pid_dentry_operations);
2421         d_add(dentry, inode);
2422         /* Close the race of the process dying before we return the dentry */
2423         if (pid_revalidate(dentry, 0))
2424                 return 0;
2425 out:
2426         return -ENOENT;
2427 }
2428
2429 static struct dentry *proc_pident_lookup(struct inode *dir, 
2430                                          struct dentry *dentry,
2431                                          const struct pid_entry *ents,
2432                                          unsigned int nents)
2433 {
2434         int error;
2435         struct task_struct *task = get_proc_task(dir);
2436         const struct pid_entry *p, *last;
2437
2438         error = -ENOENT;
2439
2440         if (!task)
2441                 goto out_no_task;
2442
2443         /*
2444          * Yes, it does not scale. And it should not. Don't add
2445          * new entries into /proc/<tgid>/ without very good reasons.
2446          */
2447         last = &ents[nents];
2448         for (p = ents; p < last; p++) {
2449                 if (p->len != dentry->d_name.len)
2450                         continue;
2451                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2452                         break;
2453         }
2454         if (p >= last)
2455                 goto out;
2456
2457         error = proc_pident_instantiate(dir, dentry, task, p);
2458 out:
2459         put_task_struct(task);
2460 out_no_task:
2461         return ERR_PTR(error);
2462 }
2463
2464 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2465                 const struct pid_entry *ents, unsigned int nents)
2466 {
2467         struct task_struct *task = get_proc_task(file_inode(file));
2468         const struct pid_entry *p;
2469
2470         if (!task)
2471                 return -ENOENT;
2472
2473         if (!dir_emit_dots(file, ctx))
2474                 goto out;
2475
2476         if (ctx->pos >= nents + 2)
2477                 goto out;
2478
2479         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2480                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2481                                 proc_pident_instantiate, task, p))
2482                         break;
2483                 ctx->pos++;
2484         }
2485 out:
2486         put_task_struct(task);
2487         return 0;
2488 }
2489
2490 #ifdef CONFIG_SECURITY
2491 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2492                                   size_t count, loff_t *ppos)
2493 {
2494         struct inode * inode = file_inode(file);
2495         char *p = NULL;
2496         ssize_t length;
2497         struct task_struct *task = get_proc_task(inode);
2498
2499         if (!task)
2500                 return -ESRCH;
2501
2502         length = security_getprocattr(task,
2503                                       (char*)file->f_path.dentry->d_name.name,
2504                                       &p);
2505         put_task_struct(task);
2506         if (length > 0)
2507                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2508         kfree(p);
2509         return length;
2510 }
2511
2512 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2513                                    size_t count, loff_t *ppos)
2514 {
2515         struct inode * inode = file_inode(file);
2516         void *page;
2517         ssize_t length;
2518         struct task_struct *task = get_proc_task(inode);
2519
2520         length = -ESRCH;
2521         if (!task)
2522                 goto out_no_task;
2523
2524         /* A task may only write its own attributes. */
2525         length = -EACCES;
2526         if (current != task)
2527                 goto out;
2528
2529         if (count > PAGE_SIZE)
2530                 count = PAGE_SIZE;
2531
2532         /* No partial writes. */
2533         length = -EINVAL;
2534         if (*ppos != 0)
2535                 goto out;
2536
2537         page = memdup_user(buf, count);
2538         if (IS_ERR(page)) {
2539                 length = PTR_ERR(page);
2540                 goto out;
2541         }
2542
2543         /* Guard against adverse ptrace interaction */
2544         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2545         if (length < 0)
2546                 goto out_free;
2547
2548         length = security_setprocattr(file->f_path.dentry->d_name.name,
2549                                       page, count);
2550         mutex_unlock(&current->signal->cred_guard_mutex);
2551 out_free:
2552         kfree(page);
2553 out:
2554         put_task_struct(task);
2555 out_no_task:
2556         return length;
2557 }
2558
2559 static const struct file_operations proc_pid_attr_operations = {
2560         .read           = proc_pid_attr_read,
2561         .write          = proc_pid_attr_write,
2562         .llseek         = generic_file_llseek,
2563 };
2564
2565 static const struct pid_entry attr_dir_stuff[] = {
2566         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2567         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2568         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2569         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2570         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2571         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2572 };
2573
2574 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2575 {
2576         return proc_pident_readdir(file, ctx, 
2577                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2578 }
2579
2580 static const struct file_operations proc_attr_dir_operations = {
2581         .read           = generic_read_dir,
2582         .iterate_shared = proc_attr_dir_readdir,
2583         .llseek         = generic_file_llseek,
2584 };
2585
2586 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2587                                 struct dentry *dentry, unsigned int flags)
2588 {
2589         return proc_pident_lookup(dir, dentry,
2590                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2591 }
2592
2593 static const struct inode_operations proc_attr_dir_inode_operations = {
2594         .lookup         = proc_attr_dir_lookup,
2595         .getattr        = pid_getattr,
2596         .setattr        = proc_setattr,
2597 };
2598
2599 #endif
2600
2601 #ifdef CONFIG_ELF_CORE
2602 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2603                                          size_t count, loff_t *ppos)
2604 {
2605         struct task_struct *task = get_proc_task(file_inode(file));
2606         struct mm_struct *mm;
2607         char buffer[PROC_NUMBUF];
2608         size_t len;
2609         int ret;
2610
2611         if (!task)
2612                 return -ESRCH;
2613
2614         ret = 0;
2615         mm = get_task_mm(task);
2616         if (mm) {
2617                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2618                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2619                                 MMF_DUMP_FILTER_SHIFT));
2620                 mmput(mm);
2621                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2622         }
2623
2624         put_task_struct(task);
2625
2626         return ret;
2627 }
2628
2629 static ssize_t proc_coredump_filter_write(struct file *file,
2630                                           const char __user *buf,
2631                                           size_t count,
2632                                           loff_t *ppos)
2633 {
2634         struct task_struct *task;
2635         struct mm_struct *mm;
2636         unsigned int val;
2637         int ret;
2638         int i;
2639         unsigned long mask;
2640
2641         ret = kstrtouint_from_user(buf, count, 0, &val);
2642         if (ret < 0)
2643                 return ret;
2644
2645         ret = -ESRCH;
2646         task = get_proc_task(file_inode(file));
2647         if (!task)
2648                 goto out_no_task;
2649
2650         mm = get_task_mm(task);
2651         if (!mm)
2652                 goto out_no_mm;
2653         ret = 0;
2654
2655         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2656                 if (val & mask)
2657                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2658                 else
2659                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2660         }
2661
2662         mmput(mm);
2663  out_no_mm:
2664         put_task_struct(task);
2665  out_no_task:
2666         if (ret < 0)
2667                 return ret;
2668         return count;
2669 }
2670
2671 static const struct file_operations proc_coredump_filter_operations = {
2672         .read           = proc_coredump_filter_read,
2673         .write          = proc_coredump_filter_write,
2674         .llseek         = generic_file_llseek,
2675 };
2676 #endif
2677
2678 #ifdef CONFIG_TASK_IO_ACCOUNTING
2679 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2680 {
2681         struct task_io_accounting acct = task->ioac;
2682         unsigned long flags;
2683         int result;
2684
2685         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2686         if (result)
2687                 return result;
2688
2689         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2690                 result = -EACCES;
2691                 goto out_unlock;
2692         }
2693
2694         if (whole && lock_task_sighand(task, &flags)) {
2695                 struct task_struct *t = task;
2696
2697                 task_io_accounting_add(&acct, &task->signal->ioac);
2698                 while_each_thread(task, t)
2699                         task_io_accounting_add(&acct, &t->ioac);
2700
2701                 unlock_task_sighand(task, &flags);
2702         }
2703         seq_printf(m,
2704                    "rchar: %llu\n"
2705                    "wchar: %llu\n"
2706                    "syscr: %llu\n"
2707                    "syscw: %llu\n"
2708                    "read_bytes: %llu\n"
2709                    "write_bytes: %llu\n"
2710                    "cancelled_write_bytes: %llu\n",
2711                    (unsigned long long)acct.rchar,
2712                    (unsigned long long)acct.wchar,
2713                    (unsigned long long)acct.syscr,
2714                    (unsigned long long)acct.syscw,
2715                    (unsigned long long)acct.read_bytes,
2716                    (unsigned long long)acct.write_bytes,
2717                    (unsigned long long)acct.cancelled_write_bytes);
2718         result = 0;
2719
2720 out_unlock:
2721         mutex_unlock(&task->signal->cred_guard_mutex);
2722         return result;
2723 }
2724
2725 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2726                                   struct pid *pid, struct task_struct *task)
2727 {
2728         return do_io_accounting(task, m, 0);
2729 }
2730
2731 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2732                                    struct pid *pid, struct task_struct *task)
2733 {
2734         return do_io_accounting(task, m, 1);
2735 }
2736 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2737
2738 #ifdef CONFIG_USER_NS
2739 static int proc_id_map_open(struct inode *inode, struct file *file,
2740         const struct seq_operations *seq_ops)
2741 {
2742         struct user_namespace *ns = NULL;
2743         struct task_struct *task;
2744         struct seq_file *seq;
2745         int ret = -EINVAL;
2746
2747         task = get_proc_task(inode);
2748         if (task) {
2749                 rcu_read_lock();
2750                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2751                 rcu_read_unlock();
2752                 put_task_struct(task);
2753         }
2754         if (!ns)
2755                 goto err;
2756
2757         ret = seq_open(file, seq_ops);
2758         if (ret)
2759                 goto err_put_ns;
2760
2761         seq = file->private_data;
2762         seq->private = ns;
2763
2764         return 0;
2765 err_put_ns:
2766         put_user_ns(ns);
2767 err:
2768         return ret;
2769 }
2770
2771 static int proc_id_map_release(struct inode *inode, struct file *file)
2772 {
2773         struct seq_file *seq = file->private_data;
2774         struct user_namespace *ns = seq->private;
2775         put_user_ns(ns);
2776         return seq_release(inode, file);
2777 }
2778
2779 static int proc_uid_map_open(struct inode *inode, struct file *file)
2780 {
2781         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2782 }
2783
2784 static int proc_gid_map_open(struct inode *inode, struct file *file)
2785 {
2786         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2787 }
2788
2789 static int proc_projid_map_open(struct inode *inode, struct file *file)
2790 {
2791         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2792 }
2793
2794 static const struct file_operations proc_uid_map_operations = {
2795         .open           = proc_uid_map_open,
2796         .write          = proc_uid_map_write,
2797         .read           = seq_read,
2798         .llseek         = seq_lseek,
2799         .release        = proc_id_map_release,
2800 };
2801
2802 static const struct file_operations proc_gid_map_operations = {
2803         .open           = proc_gid_map_open,
2804         .write          = proc_gid_map_write,
2805         .read           = seq_read,
2806         .llseek         = seq_lseek,
2807         .release        = proc_id_map_release,
2808 };
2809
2810 static const struct file_operations proc_projid_map_operations = {
2811         .open           = proc_projid_map_open,
2812         .write          = proc_projid_map_write,
2813         .read           = seq_read,
2814         .llseek         = seq_lseek,
2815         .release        = proc_id_map_release,
2816 };
2817
2818 static int proc_setgroups_open(struct inode *inode, struct file *file)
2819 {
2820         struct user_namespace *ns = NULL;
2821         struct task_struct *task;
2822         int ret;
2823
2824         ret = -ESRCH;
2825         task = get_proc_task(inode);
2826         if (task) {
2827                 rcu_read_lock();
2828                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2829                 rcu_read_unlock();
2830                 put_task_struct(task);
2831         }
2832         if (!ns)
2833                 goto err;
2834
2835         if (file->f_mode & FMODE_WRITE) {
2836                 ret = -EACCES;
2837                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2838                         goto err_put_ns;
2839         }
2840
2841         ret = single_open(file, &proc_setgroups_show, ns);
2842         if (ret)
2843                 goto err_put_ns;
2844
2845         return 0;
2846 err_put_ns:
2847         put_user_ns(ns);
2848 err:
2849         return ret;
2850 }
2851
2852 static int proc_setgroups_release(struct inode *inode, struct file *file)
2853 {
2854         struct seq_file *seq = file->private_data;
2855         struct user_namespace *ns = seq->private;
2856         int ret = single_release(inode, file);
2857         put_user_ns(ns);
2858         return ret;
2859 }
2860
2861 static const struct file_operations proc_setgroups_operations = {
2862         .open           = proc_setgroups_open,
2863         .write          = proc_setgroups_write,
2864         .read           = seq_read,
2865         .llseek         = seq_lseek,
2866         .release        = proc_setgroups_release,
2867 };
2868 #endif /* CONFIG_USER_NS */
2869
2870 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2871                                 struct pid *pid, struct task_struct *task)
2872 {
2873         int err = lock_trace(task);
2874         if (!err) {
2875                 seq_printf(m, "%08x\n", task->personality);
2876                 unlock_trace(task);
2877         }
2878         return err;
2879 }
2880
2881 #ifdef CONFIG_LIVEPATCH
2882 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2883                                 struct pid *pid, struct task_struct *task)
2884 {
2885         seq_printf(m, "%d\n", task->patch_state);
2886         return 0;
2887 }
2888 #endif /* CONFIG_LIVEPATCH */
2889
2890 /*
2891  * Thread groups
2892  */
2893 static const struct file_operations proc_task_operations;
2894 static const struct inode_operations proc_task_inode_operations;
2895
2896 static const struct pid_entry tgid_base_stuff[] = {
2897         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2898         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2899         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2900         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2901         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2902 #ifdef CONFIG_NET
2903         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2904 #endif
2905         REG("environ",    S_IRUSR, proc_environ_operations),
2906         REG("auxv",       S_IRUSR, proc_auxv_operations),
2907         ONE("status",     S_IRUGO, proc_pid_status),
2908         ONE("personality", S_IRUSR, proc_pid_personality),
2909         ONE("limits",     S_IRUGO, proc_pid_limits),
2910 #ifdef CONFIG_SCHED_DEBUG
2911         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2912 #endif
2913 #ifdef CONFIG_SCHED_AUTOGROUP
2914         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2915 #endif
2916         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2917 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2918         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2919 #endif
2920         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2921         ONE("stat",       S_IRUGO, proc_tgid_stat),
2922         ONE("statm",      S_IRUGO, proc_pid_statm),
2923         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2924 #ifdef CONFIG_NUMA
2925         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2926 #endif
2927         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2928         LNK("cwd",        proc_cwd_link),
2929         LNK("root",       proc_root_link),
2930         LNK("exe",        proc_exe_link),
2931         REG("mounts",     S_IRUGO, proc_mounts_operations),
2932         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2933         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2934 #ifdef CONFIG_PROC_PAGE_MONITOR
2935         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2936         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2937         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2938 #endif
2939 #ifdef CONFIG_SECURITY
2940         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2941 #endif
2942 #ifdef CONFIG_KALLSYMS
2943         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2944 #endif
2945 #ifdef CONFIG_STACKTRACE
2946         ONE("stack",      S_IRUSR, proc_pid_stack),
2947 #endif
2948 #ifdef CONFIG_SCHED_INFO
2949         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2950 #endif
2951 #ifdef CONFIG_LATENCYTOP
2952         REG("latency",  S_IRUGO, proc_lstats_operations),
2953 #endif
2954 #ifdef CONFIG_PROC_PID_CPUSET
2955         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2956 #endif
2957 #ifdef CONFIG_CGROUPS
2958         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2959 #endif
2960         ONE("oom_score",  S_IRUGO, proc_oom_score),
2961         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2962         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2963 #ifdef CONFIG_AUDITSYSCALL
2964         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2965         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2966 #endif
2967 #ifdef CONFIG_FAULT_INJECTION
2968         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2969 #endif
2970 #ifdef CONFIG_ELF_CORE
2971         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2972 #endif
2973 #ifdef CONFIG_TASK_IO_ACCOUNTING
2974         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2975 #endif
2976 #ifdef CONFIG_HARDWALL
2977         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2978 #endif
2979 #ifdef CONFIG_USER_NS
2980         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2981         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2982         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2983         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2984 #endif
2985 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2986         REG("timers",     S_IRUGO, proc_timers_operations),
2987 #endif
2988         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2989 #ifdef CONFIG_LIVEPATCH
2990         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2991 #endif
2992 };
2993
2994 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2995 {
2996         return proc_pident_readdir(file, ctx,
2997                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2998 }
2999
3000 static const struct file_operations proc_tgid_base_operations = {
3001         .read           = generic_read_dir,
3002         .iterate_shared = proc_tgid_base_readdir,
3003         .llseek         = generic_file_llseek,
3004 };
3005
3006 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3007 {
3008         return proc_pident_lookup(dir, dentry,
3009                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3010 }
3011
3012 static const struct inode_operations proc_tgid_base_inode_operations = {
3013         .lookup         = proc_tgid_base_lookup,
3014         .getattr        = pid_getattr,
3015         .setattr        = proc_setattr,
3016         .permission     = proc_pid_permission,
3017 };
3018
3019 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3020 {
3021         struct dentry *dentry, *leader, *dir;
3022         char buf[PROC_NUMBUF];
3023         struct qstr name;
3024
3025         name.name = buf;
3026         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3027         /* no ->d_hash() rejects on procfs */
3028         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3029         if (dentry) {
3030                 d_invalidate(dentry);
3031                 dput(dentry);
3032         }
3033
3034         if (pid == tgid)
3035                 return;
3036
3037         name.name = buf;
3038         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3039         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3040         if (!leader)
3041                 goto out;
3042
3043         name.name = "task";
3044         name.len = strlen(name.name);
3045         dir = d_hash_and_lookup(leader, &name);
3046         if (!dir)
3047                 goto out_put_leader;
3048
3049         name.name = buf;
3050         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3051         dentry = d_hash_and_lookup(dir, &name);
3052         if (dentry) {
3053                 d_invalidate(dentry);
3054                 dput(dentry);
3055         }
3056
3057         dput(dir);
3058 out_put_leader:
3059         dput(leader);
3060 out:
3061         return;
3062 }
3063
3064 /**
3065  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3066  * @task: task that should be flushed.
3067  *
3068  * When flushing dentries from proc, one needs to flush them from global
3069  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3070  * in. This call is supposed to do all of this job.
3071  *
3072  * Looks in the dcache for
3073  * /proc/@pid
3074  * /proc/@tgid/task/@pid
3075  * if either directory is present flushes it and all of it'ts children
3076  * from the dcache.
3077  *
3078  * It is safe and reasonable to cache /proc entries for a task until
3079  * that task exits.  After that they just clog up the dcache with
3080  * useless entries, possibly causing useful dcache entries to be
3081  * flushed instead.  This routine is proved to flush those useless
3082  * dcache entries at process exit time.
3083  *
3084  * NOTE: This routine is just an optimization so it does not guarantee
3085  *       that no dcache entries will exist at process exit time it
3086  *       just makes it very unlikely that any will persist.
3087  */
3088
3089 void proc_flush_task(struct task_struct *task)
3090 {
3091         int i;
3092         struct pid *pid, *tgid;
3093         struct upid *upid;
3094
3095         pid = task_pid(task);
3096         tgid = task_tgid(task);
3097
3098         for (i = 0; i <= pid->level; i++) {
3099                 upid = &pid->numbers[i];
3100                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3101                                         tgid->numbers[i].nr);
3102         }
3103 }
3104
3105 static int proc_pid_instantiate(struct inode *dir,
3106                                    struct dentry * dentry,
3107                                    struct task_struct *task, const void *ptr)
3108 {
3109         struct inode *inode;
3110
3111         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3112         if (!inode)
3113                 goto out;
3114
3115         inode->i_op = &proc_tgid_base_inode_operations;
3116         inode->i_fop = &proc_tgid_base_operations;
3117         inode->i_flags|=S_IMMUTABLE;
3118
3119         set_nlink(inode, nlink_tgid);
3120
3121         d_set_d_op(dentry, &pid_dentry_operations);
3122
3123         d_add(dentry, inode);
3124         /* Close the race of the process dying before we return the dentry */
3125         if (pid_revalidate(dentry, 0))
3126                 return 0;
3127 out:
3128         return -ENOENT;
3129 }
3130
3131 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3132 {
3133         int result = -ENOENT;
3134         struct task_struct *task;
3135         unsigned tgid;
3136         struct pid_namespace *ns;
3137
3138         tgid = name_to_int(&dentry->d_name);
3139         if (tgid == ~0U)
3140                 goto out;
3141
3142         ns = dentry->d_sb->s_fs_info;
3143         rcu_read_lock();
3144         task = find_task_by_pid_ns(tgid, ns);
3145         if (task)
3146                 get_task_struct(task);
3147         rcu_read_unlock();
3148         if (!task)
3149                 goto out;
3150
3151         result = proc_pid_instantiate(dir, dentry, task, NULL);
3152         put_task_struct(task);
3153 out:
3154         return ERR_PTR(result);
3155 }
3156
3157 /*
3158  * Find the first task with tgid >= tgid
3159  *
3160  */
3161 struct tgid_iter {
3162         unsigned int tgid;
3163         struct task_struct *task;
3164 };
3165 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3166 {
3167         struct pid *pid;
3168
3169         if (iter.task)
3170                 put_task_struct(iter.task);
3171         rcu_read_lock();
3172 retry:
3173         iter.task = NULL;
3174         pid = find_ge_pid(iter.tgid, ns);
3175         if (pid) {
3176                 iter.tgid = pid_nr_ns(pid, ns);
3177                 iter.task = pid_task(pid, PIDTYPE_PID);
3178                 /* What we to know is if the pid we have find is the
3179                  * pid of a thread_group_leader.  Testing for task
3180                  * being a thread_group_leader is the obvious thing
3181                  * todo but there is a window when it fails, due to
3182                  * the pid transfer logic in de_thread.
3183                  *
3184                  * So we perform the straight forward test of seeing
3185                  * if the pid we have found is the pid of a thread
3186                  * group leader, and don't worry if the task we have
3187                  * found doesn't happen to be a thread group leader.
3188                  * As we don't care in the case of readdir.
3189                  */
3190                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3191                         iter.tgid += 1;
3192                         goto retry;
3193                 }
3194                 get_task_struct(iter.task);
3195         }
3196         rcu_read_unlock();
3197         return iter;
3198 }
3199
3200 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3201
3202 /* for the /proc/ directory itself, after non-process stuff has been done */
3203 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3204 {
3205         struct tgid_iter iter;
3206         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3207         loff_t pos = ctx->pos;
3208
3209         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3210                 return 0;
3211
3212         if (pos == TGID_OFFSET - 2) {
3213                 struct inode *inode = d_inode(ns->proc_self);
3214                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3215                         return 0;
3216                 ctx->pos = pos = pos + 1;
3217         }
3218         if (pos == TGID_OFFSET - 1) {
3219                 struct inode *inode = d_inode(ns->proc_thread_self);
3220                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3221                         return 0;
3222                 ctx->pos = pos = pos + 1;
3223         }
3224         iter.tgid = pos - TGID_OFFSET;
3225         iter.task = NULL;
3226         for (iter = next_tgid(ns, iter);
3227              iter.task;
3228              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3229                 char name[PROC_NUMBUF];
3230                 int len;
3231
3232                 cond_resched();
3233                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3234                         continue;
3235
3236                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3237                 ctx->pos = iter.tgid + TGID_OFFSET;
3238                 if (!proc_fill_cache(file, ctx, name, len,
3239                                      proc_pid_instantiate, iter.task, NULL)) {
3240                         put_task_struct(iter.task);
3241                         return 0;
3242                 }
3243         }
3244         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3245         return 0;
3246 }
3247
3248 /*
3249  * proc_tid_comm_permission is a special permission function exclusively
3250  * used for the node /proc/<pid>/task/<tid>/comm.
3251  * It bypasses generic permission checks in the case where a task of the same
3252  * task group attempts to access the node.
3253  * The rationale behind this is that glibc and bionic access this node for
3254  * cross thread naming (pthread_set/getname_np(!self)). However, if
3255  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3256  * which locks out the cross thread naming implementation.
3257  * This function makes sure that the node is always accessible for members of
3258  * same thread group.
3259  */
3260 static int proc_tid_comm_permission(struct inode *inode, int mask)
3261 {
3262         bool is_same_tgroup;
3263         struct task_struct *task;
3264
3265         task = get_proc_task(inode);
3266         if (!task)
3267                 return -ESRCH;
3268         is_same_tgroup = same_thread_group(current, task);
3269         put_task_struct(task);
3270
3271         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3272                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3273                  * read or written by the members of the corresponding
3274                  * thread group.
3275                  */
3276                 return 0;
3277         }
3278
3279         return generic_permission(inode, mask);
3280 }
3281
3282 static const struct inode_operations proc_tid_comm_inode_operations = {
3283                 .permission = proc_tid_comm_permission,
3284 };
3285
3286 /*
3287  * Tasks
3288  */
3289 static const struct pid_entry tid_base_stuff[] = {
3290         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3291         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3292         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3293 #ifdef CONFIG_NET
3294         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3295 #endif
3296         REG("environ",   S_IRUSR, proc_environ_operations),
3297         REG("auxv",      S_IRUSR, proc_auxv_operations),
3298         ONE("status",    S_IRUGO, proc_pid_status),
3299         ONE("personality", S_IRUSR, proc_pid_personality),
3300         ONE("limits",    S_IRUGO, proc_pid_limits),
3301 #ifdef CONFIG_SCHED_DEBUG
3302         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3303 #endif
3304         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3305                          &proc_tid_comm_inode_operations,
3306                          &proc_pid_set_comm_operations, {}),
3307 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3308         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3309 #endif
3310         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3311         ONE("stat",      S_IRUGO, proc_tid_stat),
3312         ONE("statm",     S_IRUGO, proc_pid_statm),
3313         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3314 #ifdef CONFIG_PROC_CHILDREN
3315         REG("children",  S_IRUGO, proc_tid_children_operations),
3316 #endif
3317 #ifdef CONFIG_NUMA
3318         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3319 #endif
3320         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3321         LNK("cwd",       proc_cwd_link),
3322         LNK("root",      proc_root_link),
3323         LNK("exe",       proc_exe_link),
3324         REG("mounts",    S_IRUGO, proc_mounts_operations),
3325         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3326 #ifdef CONFIG_PROC_PAGE_MONITOR
3327         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3328         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3329         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3330 #endif
3331 #ifdef CONFIG_SECURITY
3332         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3333 #endif
3334 #ifdef CONFIG_KALLSYMS
3335         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3336 #endif
3337 #ifdef CONFIG_STACKTRACE
3338         ONE("stack",      S_IRUSR, proc_pid_stack),
3339 #endif
3340 #ifdef CONFIG_SCHED_INFO
3341         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3342 #endif
3343 #ifdef CONFIG_LATENCYTOP
3344         REG("latency",  S_IRUGO, proc_lstats_operations),
3345 #endif
3346 #ifdef CONFIG_PROC_PID_CPUSET
3347         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3348 #endif
3349 #ifdef CONFIG_CGROUPS
3350         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3351 #endif
3352         ONE("oom_score", S_IRUGO, proc_oom_score),
3353         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3354         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3355 #ifdef CONFIG_AUDITSYSCALL
3356         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3357         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3358 #endif
3359 #ifdef CONFIG_FAULT_INJECTION
3360         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3361         /*
3362          * Operations on the file check that the task is current,
3363          * so we create it with 0666 to support testing under unprivileged user.
3364          */
3365         REG("fail-nth", 0666, proc_fail_nth_operations),
3366 #endif
3367 #ifdef CONFIG_TASK_IO_ACCOUNTING
3368         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3369 #endif
3370 #ifdef CONFIG_HARDWALL
3371         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3372 #endif
3373 #ifdef CONFIG_USER_NS
3374         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3375         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3376         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3377         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3378 #endif
3379 #ifdef CONFIG_LIVEPATCH
3380         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3381 #endif
3382 };
3383
3384 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3385 {
3386         return proc_pident_readdir(file, ctx,
3387                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3388 }
3389
3390 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3391 {
3392         return proc_pident_lookup(dir, dentry,
3393                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3394 }
3395
3396 static const struct file_operations proc_tid_base_operations = {
3397         .read           = generic_read_dir,
3398         .iterate_shared = proc_tid_base_readdir,
3399         .llseek         = generic_file_llseek,
3400 };
3401
3402 static const struct inode_operations proc_tid_base_inode_operations = {
3403         .lookup         = proc_tid_base_lookup,
3404         .getattr        = pid_getattr,
3405         .setattr        = proc_setattr,
3406 };
3407
3408 static int proc_task_instantiate(struct inode *dir,
3409         struct dentry *dentry, struct task_struct *task, const void *ptr)
3410 {
3411         struct inode *inode;
3412         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3413
3414         if (!inode)
3415                 goto out;
3416         inode->i_op = &proc_tid_base_inode_operations;
3417         inode->i_fop = &proc_tid_base_operations;
3418         inode->i_flags|=S_IMMUTABLE;
3419
3420         set_nlink(inode, nlink_tid);
3421
3422         d_set_d_op(dentry, &pid_dentry_operations);
3423
3424         d_add(dentry, inode);
3425         /* Close the race of the process dying before we return the dentry */
3426         if (pid_revalidate(dentry, 0))
3427                 return 0;
3428 out:
3429         return -ENOENT;
3430 }
3431
3432 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3433 {
3434         int result = -ENOENT;
3435         struct task_struct *task;
3436         struct task_struct *leader = get_proc_task(dir);
3437         unsigned tid;
3438         struct pid_namespace *ns;
3439
3440         if (!leader)
3441                 goto out_no_task;
3442
3443         tid = name_to_int(&dentry->d_name);
3444         if (tid == ~0U)
3445                 goto out;
3446
3447         ns = dentry->d_sb->s_fs_info;
3448         rcu_read_lock();
3449         task = find_task_by_pid_ns(tid, ns);
3450         if (task)
3451                 get_task_struct(task);
3452         rcu_read_unlock();
3453         if (!task)
3454                 goto out;
3455         if (!same_thread_group(leader, task))
3456                 goto out_drop_task;
3457
3458         result = proc_task_instantiate(dir, dentry, task, NULL);
3459 out_drop_task:
3460         put_task_struct(task);
3461 out:
3462         put_task_struct(leader);
3463 out_no_task:
3464         return ERR_PTR(result);
3465 }
3466
3467 /*
3468  * Find the first tid of a thread group to return to user space.
3469  *
3470  * Usually this is just the thread group leader, but if the users
3471  * buffer was too small or there was a seek into the middle of the
3472  * directory we have more work todo.
3473  *
3474  * In the case of a short read we start with find_task_by_pid.
3475  *
3476  * In the case of a seek we start with the leader and walk nr
3477  * threads past it.
3478  */
3479 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3480                                         struct pid_namespace *ns)
3481 {
3482         struct task_struct *pos, *task;
3483         unsigned long nr = f_pos;
3484
3485         if (nr != f_pos)        /* 32bit overflow? */
3486                 return NULL;
3487
3488         rcu_read_lock();
3489         task = pid_task(pid, PIDTYPE_PID);
3490         if (!task)
3491                 goto fail;
3492
3493         /* Attempt to start with the tid of a thread */
3494         if (tid && nr) {
3495                 pos = find_task_by_pid_ns(tid, ns);
3496                 if (pos && same_thread_group(pos, task))
3497                         goto found;
3498         }
3499
3500         /* If nr exceeds the number of threads there is nothing todo */
3501         if (nr >= get_nr_threads(task))
3502                 goto fail;
3503
3504         /* If we haven't found our starting place yet start
3505          * with the leader and walk nr threads forward.
3506          */
3507         pos = task = task->group_leader;
3508         do {
3509                 if (!nr--)
3510                         goto found;
3511         } while_each_thread(task, pos);
3512 fail:
3513         pos = NULL;
3514         goto out;
3515 found:
3516         get_task_struct(pos);
3517 out:
3518         rcu_read_unlock();
3519         return pos;
3520 }
3521
3522 /*
3523  * Find the next thread in the thread list.
3524  * Return NULL if there is an error or no next thread.
3525  *
3526  * The reference to the input task_struct is released.
3527  */
3528 static struct task_struct *next_tid(struct task_struct *start)
3529 {
3530         struct task_struct *pos = NULL;
3531         rcu_read_lock();
3532         if (pid_alive(start)) {
3533                 pos = next_thread(start);
3534                 if (thread_group_leader(pos))
3535                         pos = NULL;
3536                 else
3537                         get_task_struct(pos);
3538         }
3539         rcu_read_unlock();
3540         put_task_struct(start);
3541         return pos;
3542 }
3543
3544 /* for the /proc/TGID/task/ directories */
3545 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3546 {
3547         struct inode *inode = file_inode(file);
3548         struct task_struct *task;
3549         struct pid_namespace *ns;
3550         int tid;
3551
3552         if (proc_inode_is_dead(inode))
3553                 return -ENOENT;
3554
3555         if (!dir_emit_dots(file, ctx))
3556                 return 0;
3557
3558         /* f_version caches the tgid value that the last readdir call couldn't
3559          * return. lseek aka telldir automagically resets f_version to 0.
3560          */
3561         ns = inode->i_sb->s_fs_info;
3562         tid = (int)file->f_version;
3563         file->f_version = 0;
3564         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3565              task;
3566              task = next_tid(task), ctx->pos++) {
3567                 char name[PROC_NUMBUF];
3568                 int len;
3569                 tid = task_pid_nr_ns(task, ns);
3570                 len = snprintf(name, sizeof(name), "%d", tid);
3571                 if (!proc_fill_cache(file, ctx, name, len,
3572                                 proc_task_instantiate, task, NULL)) {
3573                         /* returning this tgid failed, save it as the first
3574                          * pid for the next readir call */
3575                         file->f_version = (u64)tid;
3576                         put_task_struct(task);
3577                         break;
3578                 }
3579         }
3580
3581         return 0;
3582 }
3583
3584 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3585                              u32 request_mask, unsigned int query_flags)
3586 {
3587         struct inode *inode = d_inode(path->dentry);
3588         struct task_struct *p = get_proc_task(inode);
3589         generic_fillattr(inode, stat);
3590
3591         if (p) {
3592                 stat->nlink += get_nr_threads(p);
3593                 put_task_struct(p);
3594         }
3595
3596         return 0;
3597 }
3598
3599 static const struct inode_operations proc_task_inode_operations = {
3600         .lookup         = proc_task_lookup,
3601         .getattr        = proc_task_getattr,
3602         .setattr        = proc_setattr,
3603         .permission     = proc_pid_permission,
3604 };
3605
3606 static const struct file_operations proc_task_operations = {
3607         .read           = generic_read_dir,
3608         .iterate_shared = proc_task_readdir,
3609         .llseek         = generic_file_llseek,
3610 };
3611
3612 void __init set_proc_pid_nlink(void)
3613 {
3614         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3615         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3616 }