audit: remove task argument to audit_set_loginuid
[linux-2.6.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 <asm/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/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/fs_struct.h>
85 #include <linux/slab.h>
86 #include <linux/flex_array.h>
87 #ifdef CONFIG_HARDWALL
88 #include <asm/hardwall.h>
89 #endif
90 #include <trace/events/oom.h>
91 #include "internal.h"
92
93 /* NOTE:
94  *      Implementing inode permission operations in /proc is almost
95  *      certainly an error.  Permission checks need to happen during
96  *      each system call not at open time.  The reason is that most of
97  *      what we wish to check for permissions in /proc varies at runtime.
98  *
99  *      The classic example of a problem is opening file descriptors
100  *      in /proc for a task before it execs a suid executable.
101  */
102
103 struct pid_entry {
104         char *name;
105         int len;
106         umode_t mode;
107         const struct inode_operations *iop;
108         const struct file_operations *fop;
109         union proc_op op;
110 };
111
112 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
113         .name = (NAME),                                 \
114         .len  = sizeof(NAME) - 1,                       \
115         .mode = MODE,                                   \
116         .iop  = IOP,                                    \
117         .fop  = FOP,                                    \
118         .op   = OP,                                     \
119 }
120
121 #define DIR(NAME, MODE, iops, fops)     \
122         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
123 #define LNK(NAME, get_link)                                     \
124         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
125                 &proc_pid_link_inode_operations, NULL,          \
126                 { .proc_get_link = get_link } )
127 #define REG(NAME, MODE, fops)                           \
128         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
129 #define INF(NAME, MODE, read)                           \
130         NOD(NAME, (S_IFREG|(MODE)),                     \
131                 NULL, &proc_info_file_operations,       \
132                 { .proc_read = read } )
133 #define ONE(NAME, MODE, show)                           \
134         NOD(NAME, (S_IFREG|(MODE)),                     \
135                 NULL, &proc_single_file_operations,     \
136                 { .proc_show = show } )
137
138 static int proc_fd_permission(struct inode *inode, int mask);
139
140 /*
141  * Count the number of hardlinks for the pid_entry table, excluding the .
142  * and .. links.
143  */
144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
145         unsigned int n)
146 {
147         unsigned int i;
148         unsigned int count;
149
150         count = 0;
151         for (i = 0; i < n; ++i) {
152                 if (S_ISDIR(entries[i].mode))
153                         ++count;
154         }
155
156         return count;
157 }
158
159 static int get_task_root(struct task_struct *task, struct path *root)
160 {
161         int result = -ENOENT;
162
163         task_lock(task);
164         if (task->fs) {
165                 get_fs_root(task->fs, root);
166                 result = 0;
167         }
168         task_unlock(task);
169         return result;
170 }
171
172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
173 {
174         struct task_struct *task = get_proc_task(dentry->d_inode);
175         int result = -ENOENT;
176
177         if (task) {
178                 task_lock(task);
179                 if (task->fs) {
180                         get_fs_pwd(task->fs, path);
181                         result = 0;
182                 }
183                 task_unlock(task);
184                 put_task_struct(task);
185         }
186         return result;
187 }
188
189 static int proc_root_link(struct dentry *dentry, struct path *path)
190 {
191         struct task_struct *task = get_proc_task(dentry->d_inode);
192         int result = -ENOENT;
193
194         if (task) {
195                 result = get_task_root(task, path);
196                 put_task_struct(task);
197         }
198         return result;
199 }
200
201 static struct mm_struct *__check_mem_permission(struct task_struct *task)
202 {
203         struct mm_struct *mm;
204
205         mm = get_task_mm(task);
206         if (!mm)
207                 return ERR_PTR(-EINVAL);
208
209         /*
210          * A task can always look at itself, in case it chooses
211          * to use system calls instead of load instructions.
212          */
213         if (task == current)
214                 return mm;
215
216         /*
217          * If current is actively ptrace'ing, and would also be
218          * permitted to freshly attach with ptrace now, permit it.
219          */
220         if (task_is_stopped_or_traced(task)) {
221                 int match;
222                 rcu_read_lock();
223                 match = (ptrace_parent(task) == current);
224                 rcu_read_unlock();
225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
226                         return mm;
227         }
228
229         /*
230          * No one else is allowed.
231          */
232         mmput(mm);
233         return ERR_PTR(-EPERM);
234 }
235
236 /*
237  * If current may access user memory in @task return a reference to the
238  * corresponding mm, otherwise ERR_PTR.
239  */
240 static struct mm_struct *check_mem_permission(struct task_struct *task)
241 {
242         struct mm_struct *mm;
243         int err;
244
245         /*
246          * Avoid racing if task exec's as we might get a new mm but validate
247          * against old credentials.
248          */
249         err = mutex_lock_killable(&task->signal->cred_guard_mutex);
250         if (err)
251                 return ERR_PTR(err);
252
253         mm = __check_mem_permission(task);
254         mutex_unlock(&task->signal->cred_guard_mutex);
255
256         return mm;
257 }
258
259 struct mm_struct *mm_for_maps(struct task_struct *task)
260 {
261         struct mm_struct *mm;
262         int err;
263
264         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
265         if (err)
266                 return ERR_PTR(err);
267
268         mm = get_task_mm(task);
269         if (mm && mm != current->mm &&
270                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
271                 mmput(mm);
272                 mm = ERR_PTR(-EACCES);
273         }
274         mutex_unlock(&task->signal->cred_guard_mutex);
275
276         return mm;
277 }
278
279 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
280 {
281         int res = 0;
282         unsigned int len;
283         struct mm_struct *mm = get_task_mm(task);
284         if (!mm)
285                 goto out;
286         if (!mm->arg_end)
287                 goto out_mm;    /* Shh! No looking before we're done */
288
289         len = mm->arg_end - mm->arg_start;
290  
291         if (len > PAGE_SIZE)
292                 len = PAGE_SIZE;
293  
294         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
295
296         // If the nul at the end of args has been overwritten, then
297         // assume application is using setproctitle(3).
298         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
299                 len = strnlen(buffer, res);
300                 if (len < res) {
301                     res = len;
302                 } else {
303                         len = mm->env_end - mm->env_start;
304                         if (len > PAGE_SIZE - res)
305                                 len = PAGE_SIZE - res;
306                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
307                         res = strnlen(buffer, res);
308                 }
309         }
310 out_mm:
311         mmput(mm);
312 out:
313         return res;
314 }
315
316 static int proc_pid_auxv(struct task_struct *task, char *buffer)
317 {
318         struct mm_struct *mm = mm_for_maps(task);
319         int res = PTR_ERR(mm);
320         if (mm && !IS_ERR(mm)) {
321                 unsigned int nwords = 0;
322                 do {
323                         nwords += 2;
324                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
325                 res = nwords * sizeof(mm->saved_auxv[0]);
326                 if (res > PAGE_SIZE)
327                         res = PAGE_SIZE;
328                 memcpy(buffer, mm->saved_auxv, res);
329                 mmput(mm);
330         }
331         return res;
332 }
333
334
335 #ifdef CONFIG_KALLSYMS
336 /*
337  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
338  * Returns the resolved symbol.  If that fails, simply return the address.
339  */
340 static int proc_pid_wchan(struct task_struct *task, char *buffer)
341 {
342         unsigned long wchan;
343         char symname[KSYM_NAME_LEN];
344
345         wchan = get_wchan(task);
346
347         if (lookup_symbol_name(wchan, symname) < 0)
348                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
349                         return 0;
350                 else
351                         return sprintf(buffer, "%lu", wchan);
352         else
353                 return sprintf(buffer, "%s", symname);
354 }
355 #endif /* CONFIG_KALLSYMS */
356
357 static int lock_trace(struct task_struct *task)
358 {
359         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
360         if (err)
361                 return err;
362         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
363                 mutex_unlock(&task->signal->cred_guard_mutex);
364                 return -EPERM;
365         }
366         return 0;
367 }
368
369 static void unlock_trace(struct task_struct *task)
370 {
371         mutex_unlock(&task->signal->cred_guard_mutex);
372 }
373
374 #ifdef CONFIG_STACKTRACE
375
376 #define MAX_STACK_TRACE_DEPTH   64
377
378 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
379                           struct pid *pid, struct task_struct *task)
380 {
381         struct stack_trace trace;
382         unsigned long *entries;
383         int err;
384         int i;
385
386         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
387         if (!entries)
388                 return -ENOMEM;
389
390         trace.nr_entries        = 0;
391         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
392         trace.entries           = entries;
393         trace.skip              = 0;
394
395         err = lock_trace(task);
396         if (!err) {
397                 save_stack_trace_tsk(task, &trace);
398
399                 for (i = 0; i < trace.nr_entries; i++) {
400                         seq_printf(m, "[<%pK>] %pS\n",
401                                    (void *)entries[i], (void *)entries[i]);
402                 }
403                 unlock_trace(task);
404         }
405         kfree(entries);
406
407         return err;
408 }
409 #endif
410
411 #ifdef CONFIG_SCHEDSTATS
412 /*
413  * Provides /proc/PID/schedstat
414  */
415 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
416 {
417         return sprintf(buffer, "%llu %llu %lu\n",
418                         (unsigned long long)task->se.sum_exec_runtime,
419                         (unsigned long long)task->sched_info.run_delay,
420                         task->sched_info.pcount);
421 }
422 #endif
423
424 #ifdef CONFIG_LATENCYTOP
425 static int lstats_show_proc(struct seq_file *m, void *v)
426 {
427         int i;
428         struct inode *inode = m->private;
429         struct task_struct *task = get_proc_task(inode);
430
431         if (!task)
432                 return -ESRCH;
433         seq_puts(m, "Latency Top version : v0.1\n");
434         for (i = 0; i < 32; i++) {
435                 struct latency_record *lr = &task->latency_record[i];
436                 if (lr->backtrace[0]) {
437                         int q;
438                         seq_printf(m, "%i %li %li",
439                                    lr->count, lr->time, lr->max);
440                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
441                                 unsigned long bt = lr->backtrace[q];
442                                 if (!bt)
443                                         break;
444                                 if (bt == ULONG_MAX)
445                                         break;
446                                 seq_printf(m, " %ps", (void *)bt);
447                         }
448                         seq_putc(m, '\n');
449                 }
450
451         }
452         put_task_struct(task);
453         return 0;
454 }
455
456 static int lstats_open(struct inode *inode, struct file *file)
457 {
458         return single_open(file, lstats_show_proc, inode);
459 }
460
461 static ssize_t lstats_write(struct file *file, const char __user *buf,
462                             size_t count, loff_t *offs)
463 {
464         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
465
466         if (!task)
467                 return -ESRCH;
468         clear_all_latency_tracing(task);
469         put_task_struct(task);
470
471         return count;
472 }
473
474 static const struct file_operations proc_lstats_operations = {
475         .open           = lstats_open,
476         .read           = seq_read,
477         .write          = lstats_write,
478         .llseek         = seq_lseek,
479         .release        = single_release,
480 };
481
482 #endif
483
484 static int proc_oom_score(struct task_struct *task, char *buffer)
485 {
486         unsigned long points = 0;
487
488         read_lock(&tasklist_lock);
489         if (pid_alive(task))
490                 points = oom_badness(task, NULL, NULL,
491                                         totalram_pages + total_swap_pages);
492         read_unlock(&tasklist_lock);
493         return sprintf(buffer, "%lu\n", points);
494 }
495
496 struct limit_names {
497         char *name;
498         char *unit;
499 };
500
501 static const struct limit_names lnames[RLIM_NLIMITS] = {
502         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
503         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
504         [RLIMIT_DATA] = {"Max data size", "bytes"},
505         [RLIMIT_STACK] = {"Max stack size", "bytes"},
506         [RLIMIT_CORE] = {"Max core file size", "bytes"},
507         [RLIMIT_RSS] = {"Max resident set", "bytes"},
508         [RLIMIT_NPROC] = {"Max processes", "processes"},
509         [RLIMIT_NOFILE] = {"Max open files", "files"},
510         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
511         [RLIMIT_AS] = {"Max address space", "bytes"},
512         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
513         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
514         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
515         [RLIMIT_NICE] = {"Max nice priority", NULL},
516         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
517         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
518 };
519
520 /* Display limits for a process */
521 static int proc_pid_limits(struct task_struct *task, char *buffer)
522 {
523         unsigned int i;
524         int count = 0;
525         unsigned long flags;
526         char *bufptr = buffer;
527
528         struct rlimit rlim[RLIM_NLIMITS];
529
530         if (!lock_task_sighand(task, &flags))
531                 return 0;
532         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
533         unlock_task_sighand(task, &flags);
534
535         /*
536          * print the file header
537          */
538         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
539                         "Limit", "Soft Limit", "Hard Limit", "Units");
540
541         for (i = 0; i < RLIM_NLIMITS; i++) {
542                 if (rlim[i].rlim_cur == RLIM_INFINITY)
543                         count += sprintf(&bufptr[count], "%-25s %-20s ",
544                                          lnames[i].name, "unlimited");
545                 else
546                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
547                                          lnames[i].name, rlim[i].rlim_cur);
548
549                 if (rlim[i].rlim_max == RLIM_INFINITY)
550                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
551                 else
552                         count += sprintf(&bufptr[count], "%-20lu ",
553                                          rlim[i].rlim_max);
554
555                 if (lnames[i].unit)
556                         count += sprintf(&bufptr[count], "%-10s\n",
557                                          lnames[i].unit);
558                 else
559                         count += sprintf(&bufptr[count], "\n");
560         }
561
562         return count;
563 }
564
565 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
566 static int proc_pid_syscall(struct task_struct *task, char *buffer)
567 {
568         long nr;
569         unsigned long args[6], sp, pc;
570         int res = lock_trace(task);
571         if (res)
572                 return res;
573
574         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
575                 res = sprintf(buffer, "running\n");
576         else if (nr < 0)
577                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
578         else
579                 res = sprintf(buffer,
580                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
581                        nr,
582                        args[0], args[1], args[2], args[3], args[4], args[5],
583                        sp, pc);
584         unlock_trace(task);
585         return res;
586 }
587 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
588
589 /************************************************************************/
590 /*                       Here the fs part begins                        */
591 /************************************************************************/
592
593 /* permission checks */
594 static int proc_fd_access_allowed(struct inode *inode)
595 {
596         struct task_struct *task;
597         int allowed = 0;
598         /* Allow access to a task's file descriptors if it is us or we
599          * may use ptrace attach to the process and find out that
600          * information.
601          */
602         task = get_proc_task(inode);
603         if (task) {
604                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
605                 put_task_struct(task);
606         }
607         return allowed;
608 }
609
610 int proc_setattr(struct dentry *dentry, struct iattr *attr)
611 {
612         int error;
613         struct inode *inode = dentry->d_inode;
614
615         if (attr->ia_valid & ATTR_MODE)
616                 return -EPERM;
617
618         error = inode_change_ok(inode, attr);
619         if (error)
620                 return error;
621
622         if ((attr->ia_valid & ATTR_SIZE) &&
623             attr->ia_size != i_size_read(inode)) {
624                 error = vmtruncate(inode, attr->ia_size);
625                 if (error)
626                         return error;
627         }
628
629         setattr_copy(inode, attr);
630         mark_inode_dirty(inode);
631         return 0;
632 }
633
634 /*
635  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
636  * or euid/egid (for hide_pid_min=2)?
637  */
638 static bool has_pid_permissions(struct pid_namespace *pid,
639                                  struct task_struct *task,
640                                  int hide_pid_min)
641 {
642         if (pid->hide_pid < hide_pid_min)
643                 return true;
644         if (in_group_p(pid->pid_gid))
645                 return true;
646         return ptrace_may_access(task, PTRACE_MODE_READ);
647 }
648
649
650 static int proc_pid_permission(struct inode *inode, int mask)
651 {
652         struct pid_namespace *pid = inode->i_sb->s_fs_info;
653         struct task_struct *task;
654         bool has_perms;
655
656         task = get_proc_task(inode);
657         has_perms = has_pid_permissions(pid, task, 1);
658         put_task_struct(task);
659
660         if (!has_perms) {
661                 if (pid->hide_pid == 2) {
662                         /*
663                          * Let's make getdents(), stat(), and open()
664                          * consistent with each other.  If a process
665                          * may not stat() a file, it shouldn't be seen
666                          * in procfs at all.
667                          */
668                         return -ENOENT;
669                 }
670
671                 return -EPERM;
672         }
673         return generic_permission(inode, mask);
674 }
675
676
677
678 static const struct inode_operations proc_def_inode_operations = {
679         .setattr        = proc_setattr,
680 };
681
682 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
683
684 static ssize_t proc_info_read(struct file * file, char __user * buf,
685                           size_t count, loff_t *ppos)
686 {
687         struct inode * inode = file->f_path.dentry->d_inode;
688         unsigned long page;
689         ssize_t length;
690         struct task_struct *task = get_proc_task(inode);
691
692         length = -ESRCH;
693         if (!task)
694                 goto out_no_task;
695
696         if (count > PROC_BLOCK_SIZE)
697                 count = PROC_BLOCK_SIZE;
698
699         length = -ENOMEM;
700         if (!(page = __get_free_page(GFP_TEMPORARY)))
701                 goto out;
702
703         length = PROC_I(inode)->op.proc_read(task, (char*)page);
704
705         if (length >= 0)
706                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
707         free_page(page);
708 out:
709         put_task_struct(task);
710 out_no_task:
711         return length;
712 }
713
714 static const struct file_operations proc_info_file_operations = {
715         .read           = proc_info_read,
716         .llseek         = generic_file_llseek,
717 };
718
719 static int proc_single_show(struct seq_file *m, void *v)
720 {
721         struct inode *inode = m->private;
722         struct pid_namespace *ns;
723         struct pid *pid;
724         struct task_struct *task;
725         int ret;
726
727         ns = inode->i_sb->s_fs_info;
728         pid = proc_pid(inode);
729         task = get_pid_task(pid, PIDTYPE_PID);
730         if (!task)
731                 return -ESRCH;
732
733         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
734
735         put_task_struct(task);
736         return ret;
737 }
738
739 static int proc_single_open(struct inode *inode, struct file *filp)
740 {
741         return single_open(filp, proc_single_show, inode);
742 }
743
744 static const struct file_operations proc_single_file_operations = {
745         .open           = proc_single_open,
746         .read           = seq_read,
747         .llseek         = seq_lseek,
748         .release        = single_release,
749 };
750
751 static int mem_open(struct inode* inode, struct file* file)
752 {
753         file->private_data = (void*)((long)current->self_exec_id);
754         /* OK to pass negative loff_t, we can catch out-of-range */
755         file->f_mode |= FMODE_UNSIGNED_OFFSET;
756         return 0;
757 }
758
759 static ssize_t mem_read(struct file * file, char __user * buf,
760                         size_t count, loff_t *ppos)
761 {
762         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
763         char *page;
764         unsigned long src = *ppos;
765         int ret = -ESRCH;
766         struct mm_struct *mm;
767
768         if (!task)
769                 goto out_no_task;
770
771         ret = -ENOMEM;
772         page = (char *)__get_free_page(GFP_TEMPORARY);
773         if (!page)
774                 goto out;
775
776         mm = check_mem_permission(task);
777         ret = PTR_ERR(mm);
778         if (IS_ERR(mm))
779                 goto out_free;
780
781         ret = -EIO;
782  
783         if (file->private_data != (void*)((long)current->self_exec_id))
784                 goto out_put;
785
786         ret = 0;
787  
788         while (count > 0) {
789                 int this_len, retval;
790
791                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
792                 retval = access_remote_vm(mm, src, page, this_len, 0);
793                 if (!retval) {
794                         if (!ret)
795                                 ret = -EIO;
796                         break;
797                 }
798
799                 if (copy_to_user(buf, page, retval)) {
800                         ret = -EFAULT;
801                         break;
802                 }
803  
804                 ret += retval;
805                 src += retval;
806                 buf += retval;
807                 count -= retval;
808         }
809         *ppos = src;
810
811 out_put:
812         mmput(mm);
813 out_free:
814         free_page((unsigned long) page);
815 out:
816         put_task_struct(task);
817 out_no_task:
818         return ret;
819 }
820
821 static ssize_t mem_write(struct file * file, const char __user *buf,
822                          size_t count, loff_t *ppos)
823 {
824         int copied;
825         char *page;
826         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
827         unsigned long dst = *ppos;
828         struct mm_struct *mm;
829
830         copied = -ESRCH;
831         if (!task)
832                 goto out_no_task;
833
834         copied = -ENOMEM;
835         page = (char *)__get_free_page(GFP_TEMPORARY);
836         if (!page)
837                 goto out_task;
838
839         mm = check_mem_permission(task);
840         copied = PTR_ERR(mm);
841         if (IS_ERR(mm))
842                 goto out_free;
843
844         copied = -EIO;
845         if (file->private_data != (void *)((long)current->self_exec_id))
846                 goto out_mm;
847
848         copied = 0;
849         while (count > 0) {
850                 int this_len, retval;
851
852                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
853                 if (copy_from_user(page, buf, this_len)) {
854                         copied = -EFAULT;
855                         break;
856                 }
857                 retval = access_remote_vm(mm, dst, page, this_len, 1);
858                 if (!retval) {
859                         if (!copied)
860                                 copied = -EIO;
861                         break;
862                 }
863                 copied += retval;
864                 buf += retval;
865                 dst += retval;
866                 count -= retval;                        
867         }
868         *ppos = dst;
869
870 out_mm:
871         mmput(mm);
872 out_free:
873         free_page((unsigned long) page);
874 out_task:
875         put_task_struct(task);
876 out_no_task:
877         return copied;
878 }
879
880 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
881 {
882         switch (orig) {
883         case 0:
884                 file->f_pos = offset;
885                 break;
886         case 1:
887                 file->f_pos += offset;
888                 break;
889         default:
890                 return -EINVAL;
891         }
892         force_successful_syscall_return();
893         return file->f_pos;
894 }
895
896 static const struct file_operations proc_mem_operations = {
897         .llseek         = mem_lseek,
898         .read           = mem_read,
899         .write          = mem_write,
900         .open           = mem_open,
901 };
902
903 static ssize_t environ_read(struct file *file, char __user *buf,
904                         size_t count, loff_t *ppos)
905 {
906         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
907         char *page;
908         unsigned long src = *ppos;
909         int ret = -ESRCH;
910         struct mm_struct *mm;
911
912         if (!task)
913                 goto out_no_task;
914
915         ret = -ENOMEM;
916         page = (char *)__get_free_page(GFP_TEMPORARY);
917         if (!page)
918                 goto out;
919
920
921         mm = mm_for_maps(task);
922         ret = PTR_ERR(mm);
923         if (!mm || IS_ERR(mm))
924                 goto out_free;
925
926         ret = 0;
927         while (count > 0) {
928                 int this_len, retval, max_len;
929
930                 this_len = mm->env_end - (mm->env_start + src);
931
932                 if (this_len <= 0)
933                         break;
934
935                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
936                 this_len = (this_len > max_len) ? max_len : this_len;
937
938                 retval = access_process_vm(task, (mm->env_start + src),
939                         page, this_len, 0);
940
941                 if (retval <= 0) {
942                         ret = retval;
943                         break;
944                 }
945
946                 if (copy_to_user(buf, page, retval)) {
947                         ret = -EFAULT;
948                         break;
949                 }
950
951                 ret += retval;
952                 src += retval;
953                 buf += retval;
954                 count -= retval;
955         }
956         *ppos = src;
957
958         mmput(mm);
959 out_free:
960         free_page((unsigned long) page);
961 out:
962         put_task_struct(task);
963 out_no_task:
964         return ret;
965 }
966
967 static const struct file_operations proc_environ_operations = {
968         .read           = environ_read,
969         .llseek         = generic_file_llseek,
970 };
971
972 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
973                                 size_t count, loff_t *ppos)
974 {
975         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
976         char buffer[PROC_NUMBUF];
977         size_t len;
978         int oom_adjust = OOM_DISABLE;
979         unsigned long flags;
980
981         if (!task)
982                 return -ESRCH;
983
984         if (lock_task_sighand(task, &flags)) {
985                 oom_adjust = task->signal->oom_adj;
986                 unlock_task_sighand(task, &flags);
987         }
988
989         put_task_struct(task);
990
991         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
992
993         return simple_read_from_buffer(buf, count, ppos, buffer, len);
994 }
995
996 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
997                                 size_t count, loff_t *ppos)
998 {
999         struct task_struct *task;
1000         char buffer[PROC_NUMBUF];
1001         int oom_adjust;
1002         unsigned long flags;
1003         int err;
1004
1005         memset(buffer, 0, sizeof(buffer));
1006         if (count > sizeof(buffer) - 1)
1007                 count = sizeof(buffer) - 1;
1008         if (copy_from_user(buffer, buf, count)) {
1009                 err = -EFAULT;
1010                 goto out;
1011         }
1012
1013         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
1014         if (err)
1015                 goto out;
1016         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1017              oom_adjust != OOM_DISABLE) {
1018                 err = -EINVAL;
1019                 goto out;
1020         }
1021
1022         task = get_proc_task(file->f_path.dentry->d_inode);
1023         if (!task) {
1024                 err = -ESRCH;
1025                 goto out;
1026         }
1027
1028         task_lock(task);
1029         if (!task->mm) {
1030                 err = -EINVAL;
1031                 goto err_task_lock;
1032         }
1033
1034         if (!lock_task_sighand(task, &flags)) {
1035                 err = -ESRCH;
1036                 goto err_task_lock;
1037         }
1038
1039         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1040                 err = -EACCES;
1041                 goto err_sighand;
1042         }
1043
1044         /*
1045          * Warn that /proc/pid/oom_adj is deprecated, see
1046          * Documentation/feature-removal-schedule.txt.
1047          */
1048         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1049                   current->comm, task_pid_nr(current), task_pid_nr(task),
1050                   task_pid_nr(task));
1051         task->signal->oom_adj = oom_adjust;
1052         /*
1053          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1054          * value is always attainable.
1055          */
1056         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1057                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1058         else
1059                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1060                                                                 -OOM_DISABLE;
1061         trace_oom_score_adj_update(task);
1062 err_sighand:
1063         unlock_task_sighand(task, &flags);
1064 err_task_lock:
1065         task_unlock(task);
1066         put_task_struct(task);
1067 out:
1068         return err < 0 ? err : count;
1069 }
1070
1071 static const struct file_operations proc_oom_adjust_operations = {
1072         .read           = oom_adjust_read,
1073         .write          = oom_adjust_write,
1074         .llseek         = generic_file_llseek,
1075 };
1076
1077 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1078                                         size_t count, loff_t *ppos)
1079 {
1080         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1081         char buffer[PROC_NUMBUF];
1082         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1083         unsigned long flags;
1084         size_t len;
1085
1086         if (!task)
1087                 return -ESRCH;
1088         if (lock_task_sighand(task, &flags)) {
1089                 oom_score_adj = task->signal->oom_score_adj;
1090                 unlock_task_sighand(task, &flags);
1091         }
1092         put_task_struct(task);
1093         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1094         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1095 }
1096
1097 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1098                                         size_t count, loff_t *ppos)
1099 {
1100         struct task_struct *task;
1101         char buffer[PROC_NUMBUF];
1102         unsigned long flags;
1103         int oom_score_adj;
1104         int err;
1105
1106         memset(buffer, 0, sizeof(buffer));
1107         if (count > sizeof(buffer) - 1)
1108                 count = sizeof(buffer) - 1;
1109         if (copy_from_user(buffer, buf, count)) {
1110                 err = -EFAULT;
1111                 goto out;
1112         }
1113
1114         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1115         if (err)
1116                 goto out;
1117         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1118                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1119                 err = -EINVAL;
1120                 goto out;
1121         }
1122
1123         task = get_proc_task(file->f_path.dentry->d_inode);
1124         if (!task) {
1125                 err = -ESRCH;
1126                 goto out;
1127         }
1128
1129         task_lock(task);
1130         if (!task->mm) {
1131                 err = -EINVAL;
1132                 goto err_task_lock;
1133         }
1134
1135         if (!lock_task_sighand(task, &flags)) {
1136                 err = -ESRCH;
1137                 goto err_task_lock;
1138         }
1139
1140         if (oom_score_adj < task->signal->oom_score_adj_min &&
1141                         !capable(CAP_SYS_RESOURCE)) {
1142                 err = -EACCES;
1143                 goto err_sighand;
1144         }
1145
1146         task->signal->oom_score_adj = oom_score_adj;
1147         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1148                 task->signal->oom_score_adj_min = oom_score_adj;
1149         trace_oom_score_adj_update(task);
1150         /*
1151          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1152          * always attainable.
1153          */
1154         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1155                 task->signal->oom_adj = OOM_DISABLE;
1156         else
1157                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1158                                                         OOM_SCORE_ADJ_MAX;
1159 err_sighand:
1160         unlock_task_sighand(task, &flags);
1161 err_task_lock:
1162         task_unlock(task);
1163         put_task_struct(task);
1164 out:
1165         return err < 0 ? err : count;
1166 }
1167
1168 static const struct file_operations proc_oom_score_adj_operations = {
1169         .read           = oom_score_adj_read,
1170         .write          = oom_score_adj_write,
1171         .llseek         = default_llseek,
1172 };
1173
1174 #ifdef CONFIG_AUDITSYSCALL
1175 #define TMPBUFLEN 21
1176 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1177                                   size_t count, loff_t *ppos)
1178 {
1179         struct inode * inode = file->f_path.dentry->d_inode;
1180         struct task_struct *task = get_proc_task(inode);
1181         ssize_t length;
1182         char tmpbuf[TMPBUFLEN];
1183
1184         if (!task)
1185                 return -ESRCH;
1186         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1187                                 audit_get_loginuid(task));
1188         put_task_struct(task);
1189         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1190 }
1191
1192 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1193                                    size_t count, loff_t *ppos)
1194 {
1195         struct inode * inode = file->f_path.dentry->d_inode;
1196         char *page, *tmp;
1197         ssize_t length;
1198         uid_t loginuid;
1199
1200         if (!capable(CAP_AUDIT_CONTROL))
1201                 return -EPERM;
1202
1203         rcu_read_lock();
1204         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1205                 rcu_read_unlock();
1206                 return -EPERM;
1207         }
1208         rcu_read_unlock();
1209
1210         if (count >= PAGE_SIZE)
1211                 count = PAGE_SIZE - 1;
1212
1213         if (*ppos != 0) {
1214                 /* No partial writes. */
1215                 return -EINVAL;
1216         }
1217         page = (char*)__get_free_page(GFP_TEMPORARY);
1218         if (!page)
1219                 return -ENOMEM;
1220         length = -EFAULT;
1221         if (copy_from_user(page, buf, count))
1222                 goto out_free_page;
1223
1224         page[count] = '\0';
1225         loginuid = simple_strtoul(page, &tmp, 10);
1226         if (tmp == page) {
1227                 length = -EINVAL;
1228                 goto out_free_page;
1229
1230         }
1231         length = audit_set_loginuid(loginuid);
1232         if (likely(length == 0))
1233                 length = count;
1234
1235 out_free_page:
1236         free_page((unsigned long) page);
1237         return length;
1238 }
1239
1240 static const struct file_operations proc_loginuid_operations = {
1241         .read           = proc_loginuid_read,
1242         .write          = proc_loginuid_write,
1243         .llseek         = generic_file_llseek,
1244 };
1245
1246 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1247                                   size_t count, loff_t *ppos)
1248 {
1249         struct inode * inode = file->f_path.dentry->d_inode;
1250         struct task_struct *task = get_proc_task(inode);
1251         ssize_t length;
1252         char tmpbuf[TMPBUFLEN];
1253
1254         if (!task)
1255                 return -ESRCH;
1256         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1257                                 audit_get_sessionid(task));
1258         put_task_struct(task);
1259         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1260 }
1261
1262 static const struct file_operations proc_sessionid_operations = {
1263         .read           = proc_sessionid_read,
1264         .llseek         = generic_file_llseek,
1265 };
1266 #endif
1267
1268 #ifdef CONFIG_FAULT_INJECTION
1269 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1270                                       size_t count, loff_t *ppos)
1271 {
1272         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1273         char buffer[PROC_NUMBUF];
1274         size_t len;
1275         int make_it_fail;
1276
1277         if (!task)
1278                 return -ESRCH;
1279         make_it_fail = task->make_it_fail;
1280         put_task_struct(task);
1281
1282         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1283
1284         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1285 }
1286
1287 static ssize_t proc_fault_inject_write(struct file * file,
1288                         const char __user * buf, size_t count, loff_t *ppos)
1289 {
1290         struct task_struct *task;
1291         char buffer[PROC_NUMBUF], *end;
1292         int make_it_fail;
1293
1294         if (!capable(CAP_SYS_RESOURCE))
1295                 return -EPERM;
1296         memset(buffer, 0, sizeof(buffer));
1297         if (count > sizeof(buffer) - 1)
1298                 count = sizeof(buffer) - 1;
1299         if (copy_from_user(buffer, buf, count))
1300                 return -EFAULT;
1301         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1302         if (*end)
1303                 return -EINVAL;
1304         task = get_proc_task(file->f_dentry->d_inode);
1305         if (!task)
1306                 return -ESRCH;
1307         task->make_it_fail = make_it_fail;
1308         put_task_struct(task);
1309
1310         return count;
1311 }
1312
1313 static const struct file_operations proc_fault_inject_operations = {
1314         .read           = proc_fault_inject_read,
1315         .write          = proc_fault_inject_write,
1316         .llseek         = generic_file_llseek,
1317 };
1318 #endif
1319
1320
1321 #ifdef CONFIG_SCHED_DEBUG
1322 /*
1323  * Print out various scheduling related per-task fields:
1324  */
1325 static int sched_show(struct seq_file *m, void *v)
1326 {
1327         struct inode *inode = m->private;
1328         struct task_struct *p;
1329
1330         p = get_proc_task(inode);
1331         if (!p)
1332                 return -ESRCH;
1333         proc_sched_show_task(p, m);
1334
1335         put_task_struct(p);
1336
1337         return 0;
1338 }
1339
1340 static ssize_t
1341 sched_write(struct file *file, const char __user *buf,
1342             size_t count, loff_t *offset)
1343 {
1344         struct inode *inode = file->f_path.dentry->d_inode;
1345         struct task_struct *p;
1346
1347         p = get_proc_task(inode);
1348         if (!p)
1349                 return -ESRCH;
1350         proc_sched_set_task(p);
1351
1352         put_task_struct(p);
1353
1354         return count;
1355 }
1356
1357 static int sched_open(struct inode *inode, struct file *filp)
1358 {
1359         return single_open(filp, sched_show, inode);
1360 }
1361
1362 static const struct file_operations proc_pid_sched_operations = {
1363         .open           = sched_open,
1364         .read           = seq_read,
1365         .write          = sched_write,
1366         .llseek         = seq_lseek,
1367         .release        = single_release,
1368 };
1369
1370 #endif
1371
1372 #ifdef CONFIG_SCHED_AUTOGROUP
1373 /*
1374  * Print out autogroup related information:
1375  */
1376 static int sched_autogroup_show(struct seq_file *m, void *v)
1377 {
1378         struct inode *inode = m->private;
1379         struct task_struct *p;
1380
1381         p = get_proc_task(inode);
1382         if (!p)
1383                 return -ESRCH;
1384         proc_sched_autogroup_show_task(p, m);
1385
1386         put_task_struct(p);
1387
1388         return 0;
1389 }
1390
1391 static ssize_t
1392 sched_autogroup_write(struct file *file, const char __user *buf,
1393             size_t count, loff_t *offset)
1394 {
1395         struct inode *inode = file->f_path.dentry->d_inode;
1396         struct task_struct *p;
1397         char buffer[PROC_NUMBUF];
1398         int nice;
1399         int err;
1400
1401         memset(buffer, 0, sizeof(buffer));
1402         if (count > sizeof(buffer) - 1)
1403                 count = sizeof(buffer) - 1;
1404         if (copy_from_user(buffer, buf, count))
1405                 return -EFAULT;
1406
1407         err = kstrtoint(strstrip(buffer), 0, &nice);
1408         if (err < 0)
1409                 return err;
1410
1411         p = get_proc_task(inode);
1412         if (!p)
1413                 return -ESRCH;
1414
1415         err = nice;
1416         err = proc_sched_autogroup_set_nice(p, &err);
1417         if (err)
1418                 count = err;
1419
1420         put_task_struct(p);
1421
1422         return count;
1423 }
1424
1425 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1426 {
1427         int ret;
1428
1429         ret = single_open(filp, sched_autogroup_show, NULL);
1430         if (!ret) {
1431                 struct seq_file *m = filp->private_data;
1432
1433                 m->private = inode;
1434         }
1435         return ret;
1436 }
1437
1438 static const struct file_operations proc_pid_sched_autogroup_operations = {
1439         .open           = sched_autogroup_open,
1440         .read           = seq_read,
1441         .write          = sched_autogroup_write,
1442         .llseek         = seq_lseek,
1443         .release        = single_release,
1444 };
1445
1446 #endif /* CONFIG_SCHED_AUTOGROUP */
1447
1448 static ssize_t comm_write(struct file *file, const char __user *buf,
1449                                 size_t count, loff_t *offset)
1450 {
1451         struct inode *inode = file->f_path.dentry->d_inode;
1452         struct task_struct *p;
1453         char buffer[TASK_COMM_LEN];
1454
1455         memset(buffer, 0, sizeof(buffer));
1456         if (count > sizeof(buffer) - 1)
1457                 count = sizeof(buffer) - 1;
1458         if (copy_from_user(buffer, buf, count))
1459                 return -EFAULT;
1460
1461         p = get_proc_task(inode);
1462         if (!p)
1463                 return -ESRCH;
1464
1465         if (same_thread_group(current, p))
1466                 set_task_comm(p, buffer);
1467         else
1468                 count = -EINVAL;
1469
1470         put_task_struct(p);
1471
1472         return count;
1473 }
1474
1475 static int comm_show(struct seq_file *m, void *v)
1476 {
1477         struct inode *inode = m->private;
1478         struct task_struct *p;
1479
1480         p = get_proc_task(inode);
1481         if (!p)
1482                 return -ESRCH;
1483
1484         task_lock(p);
1485         seq_printf(m, "%s\n", p->comm);
1486         task_unlock(p);
1487
1488         put_task_struct(p);
1489
1490         return 0;
1491 }
1492
1493 static int comm_open(struct inode *inode, struct file *filp)
1494 {
1495         return single_open(filp, comm_show, inode);
1496 }
1497
1498 static const struct file_operations proc_pid_set_comm_operations = {
1499         .open           = comm_open,
1500         .read           = seq_read,
1501         .write          = comm_write,
1502         .llseek         = seq_lseek,
1503         .release        = single_release,
1504 };
1505
1506 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1507 {
1508         struct task_struct *task;
1509         struct mm_struct *mm;
1510         struct file *exe_file;
1511
1512         task = get_proc_task(dentry->d_inode);
1513         if (!task)
1514                 return -ENOENT;
1515         mm = get_task_mm(task);
1516         put_task_struct(task);
1517         if (!mm)
1518                 return -ENOENT;
1519         exe_file = get_mm_exe_file(mm);
1520         mmput(mm);
1521         if (exe_file) {
1522                 *exe_path = exe_file->f_path;
1523                 path_get(&exe_file->f_path);
1524                 fput(exe_file);
1525                 return 0;
1526         } else
1527                 return -ENOENT;
1528 }
1529
1530 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1531 {
1532         struct inode *inode = dentry->d_inode;
1533         int error = -EACCES;
1534
1535         /* We don't need a base pointer in the /proc filesystem */
1536         path_put(&nd->path);
1537
1538         /* Are we allowed to snoop on the tasks file descriptors? */
1539         if (!proc_fd_access_allowed(inode))
1540                 goto out;
1541
1542         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1543 out:
1544         return ERR_PTR(error);
1545 }
1546
1547 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1548 {
1549         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1550         char *pathname;
1551         int len;
1552
1553         if (!tmp)
1554                 return -ENOMEM;
1555
1556         pathname = d_path(path, tmp, PAGE_SIZE);
1557         len = PTR_ERR(pathname);
1558         if (IS_ERR(pathname))
1559                 goto out;
1560         len = tmp + PAGE_SIZE - 1 - pathname;
1561
1562         if (len > buflen)
1563                 len = buflen;
1564         if (copy_to_user(buffer, pathname, len))
1565                 len = -EFAULT;
1566  out:
1567         free_page((unsigned long)tmp);
1568         return len;
1569 }
1570
1571 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1572 {
1573         int error = -EACCES;
1574         struct inode *inode = dentry->d_inode;
1575         struct path path;
1576
1577         /* Are we allowed to snoop on the tasks file descriptors? */
1578         if (!proc_fd_access_allowed(inode))
1579                 goto out;
1580
1581         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1582         if (error)
1583                 goto out;
1584
1585         error = do_proc_readlink(&path, buffer, buflen);
1586         path_put(&path);
1587 out:
1588         return error;
1589 }
1590
1591 static const struct inode_operations proc_pid_link_inode_operations = {
1592         .readlink       = proc_pid_readlink,
1593         .follow_link    = proc_pid_follow_link,
1594         .setattr        = proc_setattr,
1595 };
1596
1597
1598 /* building an inode */
1599
1600 static int task_dumpable(struct task_struct *task)
1601 {
1602         int dumpable = 0;
1603         struct mm_struct *mm;
1604
1605         task_lock(task);
1606         mm = task->mm;
1607         if (mm)
1608                 dumpable = get_dumpable(mm);
1609         task_unlock(task);
1610         if(dumpable == 1)
1611                 return 1;
1612         return 0;
1613 }
1614
1615 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1616 {
1617         struct inode * inode;
1618         struct proc_inode *ei;
1619         const struct cred *cred;
1620
1621         /* We need a new inode */
1622
1623         inode = new_inode(sb);
1624         if (!inode)
1625                 goto out;
1626
1627         /* Common stuff */
1628         ei = PROC_I(inode);
1629         inode->i_ino = get_next_ino();
1630         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1631         inode->i_op = &proc_def_inode_operations;
1632
1633         /*
1634          * grab the reference to task.
1635          */
1636         ei->pid = get_task_pid(task, PIDTYPE_PID);
1637         if (!ei->pid)
1638                 goto out_unlock;
1639
1640         if (task_dumpable(task)) {
1641                 rcu_read_lock();
1642                 cred = __task_cred(task);
1643                 inode->i_uid = cred->euid;
1644                 inode->i_gid = cred->egid;
1645                 rcu_read_unlock();
1646         }
1647         security_task_to_inode(task, inode);
1648
1649 out:
1650         return inode;
1651
1652 out_unlock:
1653         iput(inode);
1654         return NULL;
1655 }
1656
1657 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1658 {
1659         struct inode *inode = dentry->d_inode;
1660         struct task_struct *task;
1661         const struct cred *cred;
1662         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1663
1664         generic_fillattr(inode, stat);
1665
1666         rcu_read_lock();
1667         stat->uid = 0;
1668         stat->gid = 0;
1669         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1670         if (task) {
1671                 if (!has_pid_permissions(pid, task, 2)) {
1672                         rcu_read_unlock();
1673                         /*
1674                          * This doesn't prevent learning whether PID exists,
1675                          * it only makes getattr() consistent with readdir().
1676                          */
1677                         return -ENOENT;
1678                 }
1679                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1680                     task_dumpable(task)) {
1681                         cred = __task_cred(task);
1682                         stat->uid = cred->euid;
1683                         stat->gid = cred->egid;
1684                 }
1685         }
1686         rcu_read_unlock();
1687         return 0;
1688 }
1689
1690 /* dentry stuff */
1691
1692 /*
1693  *      Exceptional case: normally we are not allowed to unhash a busy
1694  * directory. In this case, however, we can do it - no aliasing problems
1695  * due to the way we treat inodes.
1696  *
1697  * Rewrite the inode's ownerships here because the owning task may have
1698  * performed a setuid(), etc.
1699  *
1700  * Before the /proc/pid/status file was created the only way to read
1701  * the effective uid of a /process was to stat /proc/pid.  Reading
1702  * /proc/pid/status is slow enough that procps and other packages
1703  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1704  * made this apply to all per process world readable and executable
1705  * directories.
1706  */
1707 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1708 {
1709         struct inode *inode;
1710         struct task_struct *task;
1711         const struct cred *cred;
1712
1713         if (nd && nd->flags & LOOKUP_RCU)
1714                 return -ECHILD;
1715
1716         inode = dentry->d_inode;
1717         task = get_proc_task(inode);
1718
1719         if (task) {
1720                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1721                     task_dumpable(task)) {
1722                         rcu_read_lock();
1723                         cred = __task_cred(task);
1724                         inode->i_uid = cred->euid;
1725                         inode->i_gid = cred->egid;
1726                         rcu_read_unlock();
1727                 } else {
1728                         inode->i_uid = 0;
1729                         inode->i_gid = 0;
1730                 }
1731                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1732                 security_task_to_inode(task, inode);
1733                 put_task_struct(task);
1734                 return 1;
1735         }
1736         d_drop(dentry);
1737         return 0;
1738 }
1739
1740 static int pid_delete_dentry(const struct dentry * dentry)
1741 {
1742         /* Is the task we represent dead?
1743          * If so, then don't put the dentry on the lru list,
1744          * kill it immediately.
1745          */
1746         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1747 }
1748
1749 const struct dentry_operations pid_dentry_operations =
1750 {
1751         .d_revalidate   = pid_revalidate,
1752         .d_delete       = pid_delete_dentry,
1753 };
1754
1755 /* Lookups */
1756
1757 /*
1758  * Fill a directory entry.
1759  *
1760  * If possible create the dcache entry and derive our inode number and
1761  * file type from dcache entry.
1762  *
1763  * Since all of the proc inode numbers are dynamically generated, the inode
1764  * numbers do not exist until the inode is cache.  This means creating the
1765  * the dcache entry in readdir is necessary to keep the inode numbers
1766  * reported by readdir in sync with the inode numbers reported
1767  * by stat.
1768  */
1769 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1770         const char *name, int len,
1771         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1772 {
1773         struct dentry *child, *dir = filp->f_path.dentry;
1774         struct inode *inode;
1775         struct qstr qname;
1776         ino_t ino = 0;
1777         unsigned type = DT_UNKNOWN;
1778
1779         qname.name = name;
1780         qname.len  = len;
1781         qname.hash = full_name_hash(name, len);
1782
1783         child = d_lookup(dir, &qname);
1784         if (!child) {
1785                 struct dentry *new;
1786                 new = d_alloc(dir, &qname);
1787                 if (new) {
1788                         child = instantiate(dir->d_inode, new, task, ptr);
1789                         if (child)
1790                                 dput(new);
1791                         else
1792                                 child = new;
1793                 }
1794         }
1795         if (!child || IS_ERR(child) || !child->d_inode)
1796                 goto end_instantiate;
1797         inode = child->d_inode;
1798         if (inode) {
1799                 ino = inode->i_ino;
1800                 type = inode->i_mode >> 12;
1801         }
1802         dput(child);
1803 end_instantiate:
1804         if (!ino)
1805                 ino = find_inode_number(dir, &qname);
1806         if (!ino)
1807                 ino = 1;
1808         return filldir(dirent, name, len, filp->f_pos, ino, type);
1809 }
1810
1811 static unsigned name_to_int(struct dentry *dentry)
1812 {
1813         const char *name = dentry->d_name.name;
1814         int len = dentry->d_name.len;
1815         unsigned n = 0;
1816
1817         if (len > 1 && *name == '0')
1818                 goto out;
1819         while (len-- > 0) {
1820                 unsigned c = *name++ - '0';
1821                 if (c > 9)
1822                         goto out;
1823                 if (n >= (~0U-9)/10)
1824                         goto out;
1825                 n *= 10;
1826                 n += c;
1827         }
1828         return n;
1829 out:
1830         return ~0U;
1831 }
1832
1833 #define PROC_FDINFO_MAX 64
1834
1835 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1836 {
1837         struct task_struct *task = get_proc_task(inode);
1838         struct files_struct *files = NULL;
1839         struct file *file;
1840         int fd = proc_fd(inode);
1841
1842         if (task) {
1843                 files = get_files_struct(task);
1844                 put_task_struct(task);
1845         }
1846         if (files) {
1847                 /*
1848                  * We are not taking a ref to the file structure, so we must
1849                  * hold ->file_lock.
1850                  */
1851                 spin_lock(&files->file_lock);
1852                 file = fcheck_files(files, fd);
1853                 if (file) {
1854                         unsigned int f_flags;
1855                         struct fdtable *fdt;
1856
1857                         fdt = files_fdtable(files);
1858                         f_flags = file->f_flags & ~O_CLOEXEC;
1859                         if (FD_ISSET(fd, fdt->close_on_exec))
1860                                 f_flags |= O_CLOEXEC;
1861
1862                         if (path) {
1863                                 *path = file->f_path;
1864                                 path_get(&file->f_path);
1865                         }
1866                         if (info)
1867                                 snprintf(info, PROC_FDINFO_MAX,
1868                                          "pos:\t%lli\n"
1869                                          "flags:\t0%o\n",
1870                                          (long long) file->f_pos,
1871                                          f_flags);
1872                         spin_unlock(&files->file_lock);
1873                         put_files_struct(files);
1874                         return 0;
1875                 }
1876                 spin_unlock(&files->file_lock);
1877                 put_files_struct(files);
1878         }
1879         return -ENOENT;
1880 }
1881
1882 static int proc_fd_link(struct dentry *dentry, struct path *path)
1883 {
1884         return proc_fd_info(dentry->d_inode, path, NULL);
1885 }
1886
1887 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1888 {
1889         struct inode *inode;
1890         struct task_struct *task;
1891         int fd;
1892         struct files_struct *files;
1893         const struct cred *cred;
1894
1895         if (nd && nd->flags & LOOKUP_RCU)
1896                 return -ECHILD;
1897
1898         inode = dentry->d_inode;
1899         task = get_proc_task(inode);
1900         fd = proc_fd(inode);
1901
1902         if (task) {
1903                 files = get_files_struct(task);
1904                 if (files) {
1905                         rcu_read_lock();
1906                         if (fcheck_files(files, fd)) {
1907                                 rcu_read_unlock();
1908                                 put_files_struct(files);
1909                                 if (task_dumpable(task)) {
1910                                         rcu_read_lock();
1911                                         cred = __task_cred(task);
1912                                         inode->i_uid = cred->euid;
1913                                         inode->i_gid = cred->egid;
1914                                         rcu_read_unlock();
1915                                 } else {
1916                                         inode->i_uid = 0;
1917                                         inode->i_gid = 0;
1918                                 }
1919                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1920                                 security_task_to_inode(task, inode);
1921                                 put_task_struct(task);
1922                                 return 1;
1923                         }
1924                         rcu_read_unlock();
1925                         put_files_struct(files);
1926                 }
1927                 put_task_struct(task);
1928         }
1929         d_drop(dentry);
1930         return 0;
1931 }
1932
1933 static const struct dentry_operations tid_fd_dentry_operations =
1934 {
1935         .d_revalidate   = tid_fd_revalidate,
1936         .d_delete       = pid_delete_dentry,
1937 };
1938
1939 static struct dentry *proc_fd_instantiate(struct inode *dir,
1940         struct dentry *dentry, struct task_struct *task, const void *ptr)
1941 {
1942         unsigned fd = *(const unsigned *)ptr;
1943         struct file *file;
1944         struct files_struct *files;
1945         struct inode *inode;
1946         struct proc_inode *ei;
1947         struct dentry *error = ERR_PTR(-ENOENT);
1948
1949         inode = proc_pid_make_inode(dir->i_sb, task);
1950         if (!inode)
1951                 goto out;
1952         ei = PROC_I(inode);
1953         ei->fd = fd;
1954         files = get_files_struct(task);
1955         if (!files)
1956                 goto out_iput;
1957         inode->i_mode = S_IFLNK;
1958
1959         /*
1960          * We are not taking a ref to the file structure, so we must
1961          * hold ->file_lock.
1962          */
1963         spin_lock(&files->file_lock);
1964         file = fcheck_files(files, fd);
1965         if (!file)
1966                 goto out_unlock;
1967         if (file->f_mode & FMODE_READ)
1968                 inode->i_mode |= S_IRUSR | S_IXUSR;
1969         if (file->f_mode & FMODE_WRITE)
1970                 inode->i_mode |= S_IWUSR | S_IXUSR;
1971         spin_unlock(&files->file_lock);
1972         put_files_struct(files);
1973
1974         inode->i_op = &proc_pid_link_inode_operations;
1975         inode->i_size = 64;
1976         ei->op.proc_get_link = proc_fd_link;
1977         d_set_d_op(dentry, &tid_fd_dentry_operations);
1978         d_add(dentry, inode);
1979         /* Close the race of the process dying before we return the dentry */
1980         if (tid_fd_revalidate(dentry, NULL))
1981                 error = NULL;
1982
1983  out:
1984         return error;
1985 out_unlock:
1986         spin_unlock(&files->file_lock);
1987         put_files_struct(files);
1988 out_iput:
1989         iput(inode);
1990         goto out;
1991 }
1992
1993 static struct dentry *proc_lookupfd_common(struct inode *dir,
1994                                            struct dentry *dentry,
1995                                            instantiate_t instantiate)
1996 {
1997         struct task_struct *task = get_proc_task(dir);
1998         unsigned fd = name_to_int(dentry);
1999         struct dentry *result = ERR_PTR(-ENOENT);
2000
2001         if (!task)
2002                 goto out_no_task;
2003         if (fd == ~0U)
2004                 goto out;
2005
2006         result = instantiate(dir, dentry, task, &fd);
2007 out:
2008         put_task_struct(task);
2009 out_no_task:
2010         return result;
2011 }
2012
2013 static int proc_readfd_common(struct file * filp, void * dirent,
2014                               filldir_t filldir, instantiate_t instantiate)
2015 {
2016         struct dentry *dentry = filp->f_path.dentry;
2017         struct inode *inode = dentry->d_inode;
2018         struct task_struct *p = get_proc_task(inode);
2019         unsigned int fd, ino;
2020         int retval;
2021         struct files_struct * files;
2022
2023         retval = -ENOENT;
2024         if (!p)
2025                 goto out_no_task;
2026         retval = 0;
2027
2028         fd = filp->f_pos;
2029         switch (fd) {
2030                 case 0:
2031                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2032                                 goto out;
2033                         filp->f_pos++;
2034                 case 1:
2035                         ino = parent_ino(dentry);
2036                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2037                                 goto out;
2038                         filp->f_pos++;
2039                 default:
2040                         files = get_files_struct(p);
2041                         if (!files)
2042                                 goto out;
2043                         rcu_read_lock();
2044                         for (fd = filp->f_pos-2;
2045                              fd < files_fdtable(files)->max_fds;
2046                              fd++, filp->f_pos++) {
2047                                 char name[PROC_NUMBUF];
2048                                 int len;
2049
2050                                 if (!fcheck_files(files, fd))
2051                                         continue;
2052                                 rcu_read_unlock();
2053
2054                                 len = snprintf(name, sizeof(name), "%d", fd);
2055                                 if (proc_fill_cache(filp, dirent, filldir,
2056                                                     name, len, instantiate,
2057                                                     p, &fd) < 0) {
2058                                         rcu_read_lock();
2059                                         break;
2060                                 }
2061                                 rcu_read_lock();
2062                         }
2063                         rcu_read_unlock();
2064                         put_files_struct(files);
2065         }
2066 out:
2067         put_task_struct(p);
2068 out_no_task:
2069         return retval;
2070 }
2071
2072 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2073                                     struct nameidata *nd)
2074 {
2075         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2076 }
2077
2078 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2079 {
2080         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2081 }
2082
2083 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2084                                       size_t len, loff_t *ppos)
2085 {
2086         char tmp[PROC_FDINFO_MAX];
2087         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2088         if (!err)
2089                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2090         return err;
2091 }
2092
2093 static const struct file_operations proc_fdinfo_file_operations = {
2094         .open           = nonseekable_open,
2095         .read           = proc_fdinfo_read,
2096         .llseek         = no_llseek,
2097 };
2098
2099 static const struct file_operations proc_fd_operations = {
2100         .read           = generic_read_dir,
2101         .readdir        = proc_readfd,
2102         .llseek         = default_llseek,
2103 };
2104
2105 #ifdef CONFIG_CHECKPOINT_RESTORE
2106
2107 /*
2108  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2109  * which represent vma start and end addresses.
2110  */
2111 static int dname_to_vma_addr(struct dentry *dentry,
2112                              unsigned long *start, unsigned long *end)
2113 {
2114         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2115                 return -EINVAL;
2116
2117         return 0;
2118 }
2119
2120 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2121 {
2122         unsigned long vm_start, vm_end;
2123         bool exact_vma_exists = false;
2124         struct mm_struct *mm = NULL;
2125         struct task_struct *task;
2126         const struct cred *cred;
2127         struct inode *inode;
2128         int status = 0;
2129
2130         if (nd && nd->flags & LOOKUP_RCU)
2131                 return -ECHILD;
2132
2133         if (!capable(CAP_SYS_ADMIN)) {
2134                 status = -EACCES;
2135                 goto out_notask;
2136         }
2137
2138         inode = dentry->d_inode;
2139         task = get_proc_task(inode);
2140         if (!task)
2141                 goto out_notask;
2142
2143         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2144                 goto out;
2145
2146         mm = get_task_mm(task);
2147         if (!mm)
2148                 goto out;
2149
2150         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2151                 down_read(&mm->mmap_sem);
2152                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2153                 up_read(&mm->mmap_sem);
2154         }
2155
2156         mmput(mm);
2157
2158         if (exact_vma_exists) {
2159                 if (task_dumpable(task)) {
2160                         rcu_read_lock();
2161                         cred = __task_cred(task);
2162                         inode->i_uid = cred->euid;
2163                         inode->i_gid = cred->egid;
2164                         rcu_read_unlock();
2165                 } else {
2166                         inode->i_uid = 0;
2167                         inode->i_gid = 0;
2168                 }
2169                 security_task_to_inode(task, inode);
2170                 status = 1;
2171         }
2172
2173 out:
2174         put_task_struct(task);
2175
2176 out_notask:
2177         if (status <= 0)
2178                 d_drop(dentry);
2179
2180         return status;
2181 }
2182
2183 static const struct dentry_operations tid_map_files_dentry_operations = {
2184         .d_revalidate   = map_files_d_revalidate,
2185         .d_delete       = pid_delete_dentry,
2186 };
2187
2188 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2189 {
2190         unsigned long vm_start, vm_end;
2191         struct vm_area_struct *vma;
2192         struct task_struct *task;
2193         struct mm_struct *mm;
2194         int rc;
2195
2196         rc = -ENOENT;
2197         task = get_proc_task(dentry->d_inode);
2198         if (!task)
2199                 goto out;
2200
2201         mm = get_task_mm(task);
2202         put_task_struct(task);
2203         if (!mm)
2204                 goto out;
2205
2206         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2207         if (rc)
2208                 goto out_mmput;
2209
2210         down_read(&mm->mmap_sem);
2211         vma = find_exact_vma(mm, vm_start, vm_end);
2212         if (vma && vma->vm_file) {
2213                 *path = vma->vm_file->f_path;
2214                 path_get(path);
2215                 rc = 0;
2216         }
2217         up_read(&mm->mmap_sem);
2218
2219 out_mmput:
2220         mmput(mm);
2221 out:
2222         return rc;
2223 }
2224
2225 struct map_files_info {
2226         struct file     *file;
2227         unsigned long   len;
2228         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2229 };
2230
2231 static struct dentry *
2232 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2233                            struct task_struct *task, const void *ptr)
2234 {
2235         const struct file *file = ptr;
2236         struct proc_inode *ei;
2237         struct inode *inode;
2238
2239         if (!file)
2240                 return ERR_PTR(-ENOENT);
2241
2242         inode = proc_pid_make_inode(dir->i_sb, task);
2243         if (!inode)
2244                 return ERR_PTR(-ENOENT);
2245
2246         ei = PROC_I(inode);
2247         ei->op.proc_get_link = proc_map_files_get_link;
2248
2249         inode->i_op = &proc_pid_link_inode_operations;
2250         inode->i_size = 64;
2251         inode->i_mode = S_IFLNK;
2252
2253         if (file->f_mode & FMODE_READ)
2254                 inode->i_mode |= S_IRUSR;
2255         if (file->f_mode & FMODE_WRITE)
2256                 inode->i_mode |= S_IWUSR;
2257
2258         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2259         d_add(dentry, inode);
2260
2261         return NULL;
2262 }
2263
2264 static struct dentry *proc_map_files_lookup(struct inode *dir,
2265                 struct dentry *dentry, struct nameidata *nd)
2266 {
2267         unsigned long vm_start, vm_end;
2268         struct vm_area_struct *vma;
2269         struct task_struct *task;
2270         struct dentry *result;
2271         struct mm_struct *mm;
2272
2273         result = ERR_PTR(-EACCES);
2274         if (!capable(CAP_SYS_ADMIN))
2275                 goto out;
2276
2277         result = ERR_PTR(-ENOENT);
2278         task = get_proc_task(dir);
2279         if (!task)
2280                 goto out;
2281
2282         result = ERR_PTR(-EACCES);
2283         if (lock_trace(task))
2284                 goto out_put_task;
2285
2286         result = ERR_PTR(-ENOENT);
2287         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2288                 goto out_unlock;
2289
2290         mm = get_task_mm(task);
2291         if (!mm)
2292                 goto out_unlock;
2293
2294         down_read(&mm->mmap_sem);
2295         vma = find_exact_vma(mm, vm_start, vm_end);
2296         if (!vma)
2297                 goto out_no_vma;
2298
2299         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2300
2301 out_no_vma:
2302         up_read(&mm->mmap_sem);
2303         mmput(mm);
2304 out_unlock:
2305         unlock_trace(task);
2306 out_put_task:
2307         put_task_struct(task);
2308 out:
2309         return result;
2310 }
2311
2312 static const struct inode_operations proc_map_files_inode_operations = {
2313         .lookup         = proc_map_files_lookup,
2314         .permission     = proc_fd_permission,
2315         .setattr        = proc_setattr,
2316 };
2317
2318 static int
2319 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2320 {
2321         struct dentry *dentry = filp->f_path.dentry;
2322         struct inode *inode = dentry->d_inode;
2323         struct vm_area_struct *vma;
2324         struct task_struct *task;
2325         struct mm_struct *mm;
2326         ino_t ino;
2327         int ret;
2328
2329         ret = -EACCES;
2330         if (!capable(CAP_SYS_ADMIN))
2331                 goto out;
2332
2333         ret = -ENOENT;
2334         task = get_proc_task(inode);
2335         if (!task)
2336                 goto out;
2337
2338         ret = -EACCES;
2339         if (lock_trace(task))
2340                 goto out_put_task;
2341
2342         ret = 0;
2343         switch (filp->f_pos) {
2344         case 0:
2345                 ino = inode->i_ino;
2346                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2347                         goto out_unlock;
2348                 filp->f_pos++;
2349         case 1:
2350                 ino = parent_ino(dentry);
2351                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2352                         goto out_unlock;
2353                 filp->f_pos++;
2354         default:
2355         {
2356                 unsigned long nr_files, pos, i;
2357                 struct flex_array *fa = NULL;
2358                 struct map_files_info info;
2359                 struct map_files_info *p;
2360
2361                 mm = get_task_mm(task);
2362                 if (!mm)
2363                         goto out_unlock;
2364                 down_read(&mm->mmap_sem);
2365
2366                 nr_files = 0;
2367
2368                 /*
2369                  * We need two passes here:
2370                  *
2371                  *  1) Collect vmas of mapped files with mmap_sem taken
2372                  *  2) Release mmap_sem and instantiate entries
2373                  *
2374                  * otherwise we get lockdep complained, since filldir()
2375                  * routine might require mmap_sem taken in might_fault().
2376                  */
2377
2378                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2379                         if (vma->vm_file && ++pos > filp->f_pos)
2380                                 nr_files++;
2381                 }
2382
2383                 if (nr_files) {
2384                         fa = flex_array_alloc(sizeof(info), nr_files,
2385                                                 GFP_KERNEL);
2386                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2387                                                         GFP_KERNEL)) {
2388                                 ret = -ENOMEM;
2389                                 if (fa)
2390                                         flex_array_free(fa);
2391                                 up_read(&mm->mmap_sem);
2392                                 mmput(mm);
2393                                 goto out_unlock;
2394                         }
2395                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2396                                         vma = vma->vm_next) {
2397                                 if (!vma->vm_file)
2398                                         continue;
2399                                 if (++pos <= filp->f_pos)
2400                                         continue;
2401
2402                                 get_file(vma->vm_file);
2403                                 info.file = vma->vm_file;
2404                                 info.len = snprintf(info.name,
2405                                                 sizeof(info.name), "%lx-%lx",
2406                                                 vma->vm_start, vma->vm_end);
2407                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2408                                         BUG();
2409                         }
2410                 }
2411                 up_read(&mm->mmap_sem);
2412
2413                 for (i = 0; i < nr_files; i++) {
2414                         p = flex_array_get(fa, i);
2415                         ret = proc_fill_cache(filp, dirent, filldir,
2416                                               p->name, p->len,
2417                                               proc_map_files_instantiate,
2418                                               task, p->file);
2419                         if (ret)
2420                                 break;
2421                         filp->f_pos++;
2422                         fput(p->file);
2423                 }
2424                 for (; i < nr_files; i++) {
2425                         /*
2426                          * In case of error don't forget
2427                          * to put rest of file refs.
2428                          */
2429                         p = flex_array_get(fa, i);
2430                         fput(p->file);
2431                 }
2432                 if (fa)
2433                         flex_array_free(fa);
2434                 mmput(mm);
2435         }
2436         }
2437
2438 out_unlock:
2439         unlock_trace(task);
2440 out_put_task:
2441         put_task_struct(task);
2442 out:
2443         return ret;
2444 }
2445
2446 static const struct file_operations proc_map_files_operations = {
2447         .read           = generic_read_dir,
2448         .readdir        = proc_map_files_readdir,
2449         .llseek         = default_llseek,
2450 };
2451
2452 #endif /* CONFIG_CHECKPOINT_RESTORE */
2453
2454 /*
2455  * /proc/pid/fd needs a special permission handler so that a process can still
2456  * access /proc/self/fd after it has executed a setuid().
2457  */
2458 static int proc_fd_permission(struct inode *inode, int mask)
2459 {
2460         int rv = generic_permission(inode, mask);
2461         if (rv == 0)
2462                 return 0;
2463         if (task_pid(current) == proc_pid(inode))
2464                 rv = 0;
2465         return rv;
2466 }
2467
2468 /*
2469  * proc directories can do almost nothing..
2470  */
2471 static const struct inode_operations proc_fd_inode_operations = {
2472         .lookup         = proc_lookupfd,
2473         .permission     = proc_fd_permission,
2474         .setattr        = proc_setattr,
2475 };
2476
2477 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2478         struct dentry *dentry, struct task_struct *task, const void *ptr)
2479 {
2480         unsigned fd = *(unsigned *)ptr;
2481         struct inode *inode;
2482         struct proc_inode *ei;
2483         struct dentry *error = ERR_PTR(-ENOENT);
2484
2485         inode = proc_pid_make_inode(dir->i_sb, task);
2486         if (!inode)
2487                 goto out;
2488         ei = PROC_I(inode);
2489         ei->fd = fd;
2490         inode->i_mode = S_IFREG | S_IRUSR;
2491         inode->i_fop = &proc_fdinfo_file_operations;
2492         d_set_d_op(dentry, &tid_fd_dentry_operations);
2493         d_add(dentry, inode);
2494         /* Close the race of the process dying before we return the dentry */
2495         if (tid_fd_revalidate(dentry, NULL))
2496                 error = NULL;
2497
2498  out:
2499         return error;
2500 }
2501
2502 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2503                                         struct dentry *dentry,
2504                                         struct nameidata *nd)
2505 {
2506         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2507 }
2508
2509 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2510 {
2511         return proc_readfd_common(filp, dirent, filldir,
2512                                   proc_fdinfo_instantiate);
2513 }
2514
2515 static const struct file_operations proc_fdinfo_operations = {
2516         .read           = generic_read_dir,
2517         .readdir        = proc_readfdinfo,
2518         .llseek         = default_llseek,
2519 };
2520
2521 /*
2522  * proc directories can do almost nothing..
2523  */
2524 static const struct inode_operations proc_fdinfo_inode_operations = {
2525         .lookup         = proc_lookupfdinfo,
2526         .setattr        = proc_setattr,
2527 };
2528
2529
2530 static struct dentry *proc_pident_instantiate(struct inode *dir,
2531         struct dentry *dentry, struct task_struct *task, const void *ptr)
2532 {
2533         const struct pid_entry *p = ptr;
2534         struct inode *inode;
2535         struct proc_inode *ei;
2536         struct dentry *error = ERR_PTR(-ENOENT);
2537
2538         inode = proc_pid_make_inode(dir->i_sb, task);
2539         if (!inode)
2540                 goto out;
2541
2542         ei = PROC_I(inode);
2543         inode->i_mode = p->mode;
2544         if (S_ISDIR(inode->i_mode))
2545                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2546         if (p->iop)
2547                 inode->i_op = p->iop;
2548         if (p->fop)
2549                 inode->i_fop = p->fop;
2550         ei->op = p->op;
2551         d_set_d_op(dentry, &pid_dentry_operations);
2552         d_add(dentry, inode);
2553         /* Close the race of the process dying before we return the dentry */
2554         if (pid_revalidate(dentry, NULL))
2555                 error = NULL;
2556 out:
2557         return error;
2558 }
2559
2560 static struct dentry *proc_pident_lookup(struct inode *dir, 
2561                                          struct dentry *dentry,
2562                                          const struct pid_entry *ents,
2563                                          unsigned int nents)
2564 {
2565         struct dentry *error;
2566         struct task_struct *task = get_proc_task(dir);
2567         const struct pid_entry *p, *last;
2568
2569         error = ERR_PTR(-ENOENT);
2570
2571         if (!task)
2572                 goto out_no_task;
2573
2574         /*
2575          * Yes, it does not scale. And it should not. Don't add
2576          * new entries into /proc/<tgid>/ without very good reasons.
2577          */
2578         last = &ents[nents - 1];
2579         for (p = ents; p <= last; p++) {
2580                 if (p->len != dentry->d_name.len)
2581                         continue;
2582                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2583                         break;
2584         }
2585         if (p > last)
2586                 goto out;
2587
2588         error = proc_pident_instantiate(dir, dentry, task, p);
2589 out:
2590         put_task_struct(task);
2591 out_no_task:
2592         return error;
2593 }
2594
2595 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2596         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2597 {
2598         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2599                                 proc_pident_instantiate, task, p);
2600 }
2601
2602 static int proc_pident_readdir(struct file *filp,
2603                 void *dirent, filldir_t filldir,
2604                 const struct pid_entry *ents, unsigned int nents)
2605 {
2606         int i;
2607         struct dentry *dentry = filp->f_path.dentry;
2608         struct inode *inode = dentry->d_inode;
2609         struct task_struct *task = get_proc_task(inode);
2610         const struct pid_entry *p, *last;
2611         ino_t ino;
2612         int ret;
2613
2614         ret = -ENOENT;
2615         if (!task)
2616                 goto out_no_task;
2617
2618         ret = 0;
2619         i = filp->f_pos;
2620         switch (i) {
2621         case 0:
2622                 ino = inode->i_ino;
2623                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2624                         goto out;
2625                 i++;
2626                 filp->f_pos++;
2627                 /* fall through */
2628         case 1:
2629                 ino = parent_ino(dentry);
2630                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2631                         goto out;
2632                 i++;
2633                 filp->f_pos++;
2634                 /* fall through */
2635         default:
2636                 i -= 2;
2637                 if (i >= nents) {
2638                         ret = 1;
2639                         goto out;
2640                 }
2641                 p = ents + i;
2642                 last = &ents[nents - 1];
2643                 while (p <= last) {
2644                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2645                                 goto out;
2646                         filp->f_pos++;
2647                         p++;
2648                 }
2649         }
2650
2651         ret = 1;
2652 out:
2653         put_task_struct(task);
2654 out_no_task:
2655         return ret;
2656 }
2657
2658 #ifdef CONFIG_SECURITY
2659 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2660                                   size_t count, loff_t *ppos)
2661 {
2662         struct inode * inode = file->f_path.dentry->d_inode;
2663         char *p = NULL;
2664         ssize_t length;
2665         struct task_struct *task = get_proc_task(inode);
2666
2667         if (!task)
2668                 return -ESRCH;
2669
2670         length = security_getprocattr(task,
2671                                       (char*)file->f_path.dentry->d_name.name,
2672                                       &p);
2673         put_task_struct(task);
2674         if (length > 0)
2675                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2676         kfree(p);
2677         return length;
2678 }
2679
2680 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2681                                    size_t count, loff_t *ppos)
2682 {
2683         struct inode * inode = file->f_path.dentry->d_inode;
2684         char *page;
2685         ssize_t length;
2686         struct task_struct *task = get_proc_task(inode);
2687
2688         length = -ESRCH;
2689         if (!task)
2690                 goto out_no_task;
2691         if (count > PAGE_SIZE)
2692                 count = PAGE_SIZE;
2693
2694         /* No partial writes. */
2695         length = -EINVAL;
2696         if (*ppos != 0)
2697                 goto out;
2698
2699         length = -ENOMEM;
2700         page = (char*)__get_free_page(GFP_TEMPORARY);
2701         if (!page)
2702                 goto out;
2703
2704         length = -EFAULT;
2705         if (copy_from_user(page, buf, count))
2706                 goto out_free;
2707
2708         /* Guard against adverse ptrace interaction */
2709         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2710         if (length < 0)
2711                 goto out_free;
2712
2713         length = security_setprocattr(task,
2714                                       (char*)file->f_path.dentry->d_name.name,
2715                                       (void*)page, count);
2716         mutex_unlock(&task->signal->cred_guard_mutex);
2717 out_free:
2718         free_page((unsigned long) page);
2719 out:
2720         put_task_struct(task);
2721 out_no_task:
2722         return length;
2723 }
2724
2725 static const struct file_operations proc_pid_attr_operations = {
2726         .read           = proc_pid_attr_read,
2727         .write          = proc_pid_attr_write,
2728         .llseek         = generic_file_llseek,
2729 };
2730
2731 static const struct pid_entry attr_dir_stuff[] = {
2732         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2733         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2734         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2735         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2736         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2737         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2738 };
2739
2740 static int proc_attr_dir_readdir(struct file * filp,
2741                              void * dirent, filldir_t filldir)
2742 {
2743         return proc_pident_readdir(filp,dirent,filldir,
2744                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2745 }
2746
2747 static const struct file_operations proc_attr_dir_operations = {
2748         .read           = generic_read_dir,
2749         .readdir        = proc_attr_dir_readdir,
2750         .llseek         = default_llseek,
2751 };
2752
2753 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2754                                 struct dentry *dentry, struct nameidata *nd)
2755 {
2756         return proc_pident_lookup(dir, dentry,
2757                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2758 }
2759
2760 static const struct inode_operations proc_attr_dir_inode_operations = {
2761         .lookup         = proc_attr_dir_lookup,
2762         .getattr        = pid_getattr,
2763         .setattr        = proc_setattr,
2764 };
2765
2766 #endif
2767
2768 #ifdef CONFIG_ELF_CORE
2769 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2770                                          size_t count, loff_t *ppos)
2771 {
2772         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2773         struct mm_struct *mm;
2774         char buffer[PROC_NUMBUF];
2775         size_t len;
2776         int ret;
2777
2778         if (!task)
2779                 return -ESRCH;
2780
2781         ret = 0;
2782         mm = get_task_mm(task);
2783         if (mm) {
2784                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2785                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2786                                 MMF_DUMP_FILTER_SHIFT));
2787                 mmput(mm);
2788                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2789         }
2790
2791         put_task_struct(task);
2792
2793         return ret;
2794 }
2795
2796 static ssize_t proc_coredump_filter_write(struct file *file,
2797                                           const char __user *buf,
2798                                           size_t count,
2799                                           loff_t *ppos)
2800 {
2801         struct task_struct *task;
2802         struct mm_struct *mm;
2803         char buffer[PROC_NUMBUF], *end;
2804         unsigned int val;
2805         int ret;
2806         int i;
2807         unsigned long mask;
2808
2809         ret = -EFAULT;
2810         memset(buffer, 0, sizeof(buffer));
2811         if (count > sizeof(buffer) - 1)
2812                 count = sizeof(buffer) - 1;
2813         if (copy_from_user(buffer, buf, count))
2814                 goto out_no_task;
2815
2816         ret = -EINVAL;
2817         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2818         if (*end == '\n')
2819                 end++;
2820         if (end - buffer == 0)
2821                 goto out_no_task;
2822
2823         ret = -ESRCH;
2824         task = get_proc_task(file->f_dentry->d_inode);
2825         if (!task)
2826                 goto out_no_task;
2827
2828         ret = end - buffer;
2829         mm = get_task_mm(task);
2830         if (!mm)
2831                 goto out_no_mm;
2832
2833         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2834                 if (val & mask)
2835                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2836                 else
2837                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2838         }
2839
2840         mmput(mm);
2841  out_no_mm:
2842         put_task_struct(task);
2843  out_no_task:
2844         return ret;
2845 }
2846
2847 static const struct file_operations proc_coredump_filter_operations = {
2848         .read           = proc_coredump_filter_read,
2849         .write          = proc_coredump_filter_write,
2850         .llseek         = generic_file_llseek,
2851 };
2852 #endif
2853
2854 /*
2855  * /proc/self:
2856  */
2857 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2858                               int buflen)
2859 {
2860         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2861         pid_t tgid = task_tgid_nr_ns(current, ns);
2862         char tmp[PROC_NUMBUF];
2863         if (!tgid)
2864                 return -ENOENT;
2865         sprintf(tmp, "%d", tgid);
2866         return vfs_readlink(dentry,buffer,buflen,tmp);
2867 }
2868
2869 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2870 {
2871         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2872         pid_t tgid = task_tgid_nr_ns(current, ns);
2873         char *name = ERR_PTR(-ENOENT);
2874         if (tgid) {
2875                 name = __getname();
2876                 if (!name)
2877                         name = ERR_PTR(-ENOMEM);
2878                 else
2879                         sprintf(name, "%d", tgid);
2880         }
2881         nd_set_link(nd, name);
2882         return NULL;
2883 }
2884
2885 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2886                                 void *cookie)
2887 {
2888         char *s = nd_get_link(nd);
2889         if (!IS_ERR(s))
2890                 __putname(s);
2891 }
2892
2893 static const struct inode_operations proc_self_inode_operations = {
2894         .readlink       = proc_self_readlink,
2895         .follow_link    = proc_self_follow_link,
2896         .put_link       = proc_self_put_link,
2897 };
2898
2899 /*
2900  * proc base
2901  *
2902  * These are the directory entries in the root directory of /proc
2903  * that properly belong to the /proc filesystem, as they describe
2904  * describe something that is process related.
2905  */
2906 static const struct pid_entry proc_base_stuff[] = {
2907         NOD("self", S_IFLNK|S_IRWXUGO,
2908                 &proc_self_inode_operations, NULL, {}),
2909 };
2910
2911 static struct dentry *proc_base_instantiate(struct inode *dir,
2912         struct dentry *dentry, struct task_struct *task, const void *ptr)
2913 {
2914         const struct pid_entry *p = ptr;
2915         struct inode *inode;
2916         struct proc_inode *ei;
2917         struct dentry *error;
2918
2919         /* Allocate the inode */
2920         error = ERR_PTR(-ENOMEM);
2921         inode = new_inode(dir->i_sb);
2922         if (!inode)
2923                 goto out;
2924
2925         /* Initialize the inode */
2926         ei = PROC_I(inode);
2927         inode->i_ino = get_next_ino();
2928         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2929
2930         /*
2931          * grab the reference to the task.
2932          */
2933         ei->pid = get_task_pid(task, PIDTYPE_PID);
2934         if (!ei->pid)
2935                 goto out_iput;
2936
2937         inode->i_mode = p->mode;
2938         if (S_ISDIR(inode->i_mode))
2939                 set_nlink(inode, 2);
2940         if (S_ISLNK(inode->i_mode))
2941                 inode->i_size = 64;
2942         if (p->iop)
2943                 inode->i_op = p->iop;
2944         if (p->fop)
2945                 inode->i_fop = p->fop;
2946         ei->op = p->op;
2947         d_add(dentry, inode);
2948         error = NULL;
2949 out:
2950         return error;
2951 out_iput:
2952         iput(inode);
2953         goto out;
2954 }
2955
2956 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2957 {
2958         struct dentry *error;
2959         struct task_struct *task = get_proc_task(dir);
2960         const struct pid_entry *p, *last;
2961
2962         error = ERR_PTR(-ENOENT);
2963
2964         if (!task)
2965                 goto out_no_task;
2966
2967         /* Lookup the directory entry */
2968         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2969         for (p = proc_base_stuff; p <= last; p++) {
2970                 if (p->len != dentry->d_name.len)
2971                         continue;
2972                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2973                         break;
2974         }
2975         if (p > last)
2976                 goto out;
2977
2978         error = proc_base_instantiate(dir, dentry, task, p);
2979
2980 out:
2981         put_task_struct(task);
2982 out_no_task:
2983         return error;
2984 }
2985
2986 static int proc_base_fill_cache(struct file *filp, void *dirent,
2987         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2988 {
2989         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2990                                 proc_base_instantiate, task, p);
2991 }
2992
2993 #ifdef CONFIG_TASK_IO_ACCOUNTING
2994 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2995 {
2996         struct task_io_accounting acct = task->ioac;
2997         unsigned long flags;
2998         int result;
2999
3000         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
3001         if (result)
3002                 return result;
3003
3004         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
3005                 result = -EACCES;
3006                 goto out_unlock;
3007         }
3008
3009         if (whole && lock_task_sighand(task, &flags)) {
3010                 struct task_struct *t = task;
3011
3012                 task_io_accounting_add(&acct, &task->signal->ioac);
3013                 while_each_thread(task, t)
3014                         task_io_accounting_add(&acct, &t->ioac);
3015
3016                 unlock_task_sighand(task, &flags);
3017         }
3018         result = sprintf(buffer,
3019                         "rchar: %llu\n"
3020                         "wchar: %llu\n"
3021                         "syscr: %llu\n"
3022                         "syscw: %llu\n"
3023                         "read_bytes: %llu\n"
3024                         "write_bytes: %llu\n"
3025                         "cancelled_write_bytes: %llu\n",
3026                         (unsigned long long)acct.rchar,
3027                         (unsigned long long)acct.wchar,
3028                         (unsigned long long)acct.syscr,
3029                         (unsigned long long)acct.syscw,
3030                         (unsigned long long)acct.read_bytes,
3031                         (unsigned long long)acct.write_bytes,
3032                         (unsigned long long)acct.cancelled_write_bytes);
3033 out_unlock:
3034         mutex_unlock(&task->signal->cred_guard_mutex);
3035         return result;
3036 }
3037
3038 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
3039 {
3040         return do_io_accounting(task, buffer, 0);
3041 }
3042
3043 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
3044 {
3045         return do_io_accounting(task, buffer, 1);
3046 }
3047 #endif /* CONFIG_TASK_IO_ACCOUNTING */
3048
3049 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3050                                 struct pid *pid, struct task_struct *task)
3051 {
3052         int err = lock_trace(task);
3053         if (!err) {
3054                 seq_printf(m, "%08x\n", task->personality);
3055                 unlock_trace(task);
3056         }
3057         return err;
3058 }
3059
3060 /*
3061  * Thread groups
3062  */
3063 static const struct file_operations proc_task_operations;
3064 static const struct inode_operations proc_task_inode_operations;
3065
3066 static const struct pid_entry tgid_base_stuff[] = {
3067         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3068         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3069 #ifdef CONFIG_CHECKPOINT_RESTORE
3070         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3071 #endif
3072         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3073         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3074 #ifdef CONFIG_NET
3075         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3076 #endif
3077         REG("environ",    S_IRUSR, proc_environ_operations),
3078         INF("auxv",       S_IRUSR, proc_pid_auxv),
3079         ONE("status",     S_IRUGO, proc_pid_status),
3080         ONE("personality", S_IRUGO, proc_pid_personality),
3081         INF("limits",     S_IRUGO, proc_pid_limits),
3082 #ifdef CONFIG_SCHED_DEBUG
3083         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3084 #endif
3085 #ifdef CONFIG_SCHED_AUTOGROUP
3086         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3087 #endif
3088         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3089 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3090         INF("syscall",    S_IRUGO, proc_pid_syscall),
3091 #endif
3092         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3093         ONE("stat",       S_IRUGO, proc_tgid_stat),
3094         ONE("statm",      S_IRUGO, proc_pid_statm),
3095         REG("maps",       S_IRUGO, proc_maps_operations),
3096 #ifdef CONFIG_NUMA
3097         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
3098 #endif
3099         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3100         LNK("cwd",        proc_cwd_link),
3101         LNK("root",       proc_root_link),
3102         LNK("exe",        proc_exe_link),
3103         REG("mounts",     S_IRUGO, proc_mounts_operations),
3104         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3105         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3106 #ifdef CONFIG_PROC_PAGE_MONITOR
3107         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3108         REG("smaps",      S_IRUGO, proc_smaps_operations),
3109         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3110 #endif
3111 #ifdef CONFIG_SECURITY
3112         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3113 #endif
3114 #ifdef CONFIG_KALLSYMS
3115         INF("wchan",      S_IRUGO, proc_pid_wchan),
3116 #endif
3117 #ifdef CONFIG_STACKTRACE
3118         ONE("stack",      S_IRUGO, proc_pid_stack),
3119 #endif
3120 #ifdef CONFIG_SCHEDSTATS
3121         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3122 #endif
3123 #ifdef CONFIG_LATENCYTOP
3124         REG("latency",  S_IRUGO, proc_lstats_operations),
3125 #endif
3126 #ifdef CONFIG_PROC_PID_CPUSET
3127         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3128 #endif
3129 #ifdef CONFIG_CGROUPS
3130         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3131 #endif
3132         INF("oom_score",  S_IRUGO, proc_oom_score),
3133         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3134         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3135 #ifdef CONFIG_AUDITSYSCALL
3136         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3137         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3138 #endif
3139 #ifdef CONFIG_FAULT_INJECTION
3140         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3141 #endif
3142 #ifdef CONFIG_ELF_CORE
3143         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3144 #endif
3145 #ifdef CONFIG_TASK_IO_ACCOUNTING
3146         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3147 #endif
3148 #ifdef CONFIG_HARDWALL
3149         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3150 #endif
3151 };
3152
3153 static int proc_tgid_base_readdir(struct file * filp,
3154                              void * dirent, filldir_t filldir)
3155 {
3156         return proc_pident_readdir(filp,dirent,filldir,
3157                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3158 }
3159
3160 static const struct file_operations proc_tgid_base_operations = {
3161         .read           = generic_read_dir,
3162         .readdir        = proc_tgid_base_readdir,
3163         .llseek         = default_llseek,
3164 };
3165
3166 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3167         return proc_pident_lookup(dir, dentry,
3168                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3169 }
3170
3171 static const struct inode_operations proc_tgid_base_inode_operations = {
3172         .lookup         = proc_tgid_base_lookup,
3173         .getattr        = pid_getattr,
3174         .setattr        = proc_setattr,
3175         .permission     = proc_pid_permission,
3176 };
3177
3178 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3179 {
3180         struct dentry *dentry, *leader, *dir;
3181         char buf[PROC_NUMBUF];
3182         struct qstr name;
3183
3184         name.name = buf;
3185         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3186         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3187         if (dentry) {
3188                 shrink_dcache_parent(dentry);
3189                 d_drop(dentry);
3190                 dput(dentry);
3191         }
3192
3193         name.name = buf;
3194         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3195         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3196         if (!leader)
3197                 goto out;
3198
3199         name.name = "task";
3200         name.len = strlen(name.name);
3201         dir = d_hash_and_lookup(leader, &name);
3202         if (!dir)
3203                 goto out_put_leader;
3204
3205         name.name = buf;
3206         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3207         dentry = d_hash_and_lookup(dir, &name);
3208         if (dentry) {
3209                 shrink_dcache_parent(dentry);
3210                 d_drop(dentry);
3211                 dput(dentry);
3212         }
3213
3214         dput(dir);
3215 out_put_leader:
3216         dput(leader);
3217 out:
3218         return;
3219 }
3220
3221 /**
3222  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3223  * @task: task that should be flushed.
3224  *
3225  * When flushing dentries from proc, one needs to flush them from global
3226  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3227  * in. This call is supposed to do all of this job.
3228  *
3229  * Looks in the dcache for
3230  * /proc/@pid
3231  * /proc/@tgid/task/@pid
3232  * if either directory is present flushes it and all of it'ts children
3233  * from the dcache.
3234  *
3235  * It is safe and reasonable to cache /proc entries for a task until
3236  * that task exits.  After that they just clog up the dcache with
3237  * useless entries, possibly causing useful dcache entries to be
3238  * flushed instead.  This routine is proved to flush those useless
3239  * dcache entries at process exit time.
3240  *
3241  * NOTE: This routine is just an optimization so it does not guarantee
3242  *       that no dcache entries will exist at process exit time it
3243  *       just makes it very unlikely that any will persist.
3244  */
3245
3246 void proc_flush_task(struct task_struct *task)
3247 {
3248         int i;
3249         struct pid *pid, *tgid;
3250         struct upid *upid;
3251
3252         pid = task_pid(task);
3253         tgid = task_tgid(task);
3254
3255         for (i = 0; i <= pid->level; i++) {
3256                 upid = &pid->numbers[i];
3257                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3258                                         tgid->numbers[i].nr);
3259         }
3260
3261         upid = &pid->numbers[pid->level];
3262         if (upid->nr == 1)
3263                 pid_ns_release_proc(upid->ns);
3264 }
3265
3266 static struct dentry *proc_pid_instantiate(struct inode *dir,
3267                                            struct dentry * dentry,
3268                                            struct task_struct *task, const void *ptr)
3269 {
3270         struct dentry *error = ERR_PTR(-ENOENT);
3271         struct inode *inode;
3272
3273         inode = proc_pid_make_inode(dir->i_sb, task);
3274         if (!inode)
3275                 goto out;
3276
3277         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3278         inode->i_op = &proc_tgid_base_inode_operations;
3279         inode->i_fop = &proc_tgid_base_operations;
3280         inode->i_flags|=S_IMMUTABLE;
3281
3282         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3283                                                   ARRAY_SIZE(tgid_base_stuff)));
3284
3285         d_set_d_op(dentry, &pid_dentry_operations);
3286
3287         d_add(dentry, inode);
3288         /* Close the race of the process dying before we return the dentry */
3289         if (pid_revalidate(dentry, NULL))
3290                 error = NULL;
3291 out:
3292         return error;
3293 }
3294
3295 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3296 {
3297         struct dentry *result;
3298         struct task_struct *task;
3299         unsigned tgid;
3300         struct pid_namespace *ns;
3301
3302         result = proc_base_lookup(dir, dentry);
3303         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3304                 goto out;
3305
3306         tgid = name_to_int(dentry);
3307         if (tgid == ~0U)
3308                 goto out;
3309
3310         ns = dentry->d_sb->s_fs_info;
3311         rcu_read_lock();
3312         task = find_task_by_pid_ns(tgid, ns);
3313         if (task)
3314                 get_task_struct(task);
3315         rcu_read_unlock();
3316         if (!task)
3317                 goto out;
3318
3319         result = proc_pid_instantiate(dir, dentry, task, NULL);
3320         put_task_struct(task);
3321 out:
3322         return result;
3323 }
3324
3325 /*
3326  * Find the first task with tgid >= tgid
3327  *
3328  */
3329 struct tgid_iter {
3330         unsigned int tgid;
3331         struct task_struct *task;
3332 };
3333 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3334 {
3335         struct pid *pid;
3336
3337         if (iter.task)
3338                 put_task_struct(iter.task);
3339         rcu_read_lock();
3340 retry:
3341         iter.task = NULL;
3342         pid = find_ge_pid(iter.tgid, ns);
3343         if (pid) {
3344                 iter.tgid = pid_nr_ns(pid, ns);
3345                 iter.task = pid_task(pid, PIDTYPE_PID);
3346                 /* What we to know is if the pid we have find is the
3347                  * pid of a thread_group_leader.  Testing for task
3348                  * being a thread_group_leader is the obvious thing
3349                  * todo but there is a window when it fails, due to
3350                  * the pid transfer logic in de_thread.
3351                  *
3352                  * So we perform the straight forward test of seeing
3353                  * if the pid we have found is the pid of a thread
3354                  * group leader, and don't worry if the task we have
3355                  * found doesn't happen to be a thread group leader.
3356                  * As we don't care in the case of readdir.
3357                  */
3358                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3359                         iter.tgid += 1;
3360                         goto retry;
3361                 }
3362                 get_task_struct(iter.task);
3363         }
3364         rcu_read_unlock();
3365         return iter;
3366 }
3367
3368 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3369
3370 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3371         struct tgid_iter iter)
3372 {
3373         char name[PROC_NUMBUF];
3374         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3375         return proc_fill_cache(filp, dirent, filldir, name, len,
3376                                 proc_pid_instantiate, iter.task, NULL);
3377 }
3378
3379 static int fake_filldir(void *buf, const char *name, int namelen,
3380                         loff_t offset, u64 ino, unsigned d_type)
3381 {
3382         return 0;
3383 }
3384
3385 /* for the /proc/ directory itself, after non-process stuff has been done */
3386 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3387 {
3388         unsigned int nr;
3389         struct task_struct *reaper;
3390         struct tgid_iter iter;
3391         struct pid_namespace *ns;
3392         filldir_t __filldir;
3393
3394         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3395                 goto out_no_task;
3396         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3397
3398         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3399         if (!reaper)
3400                 goto out_no_task;
3401
3402         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3403                 const struct pid_entry *p = &proc_base_stuff[nr];
3404                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3405                         goto out;
3406         }
3407
3408         ns = filp->f_dentry->d_sb->s_fs_info;
3409         iter.task = NULL;
3410         iter.tgid = filp->f_pos - TGID_OFFSET;
3411         for (iter = next_tgid(ns, iter);
3412              iter.task;
3413              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3414                 if (has_pid_permissions(ns, iter.task, 2))
3415                         __filldir = filldir;
3416                 else
3417                         __filldir = fake_filldir;
3418
3419                 filp->f_pos = iter.tgid + TGID_OFFSET;
3420                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3421                         put_task_struct(iter.task);
3422                         goto out;
3423                 }
3424         }
3425         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3426 out:
3427         put_task_struct(reaper);
3428 out_no_task:
3429         return 0;
3430 }
3431
3432 /*
3433  * Tasks
3434  */
3435 static const struct pid_entry tid_base_stuff[] = {
3436         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3437         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3438         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3439         REG("environ",   S_IRUSR, proc_environ_operations),
3440         INF("auxv",      S_IRUSR, proc_pid_auxv),
3441         ONE("status",    S_IRUGO, proc_pid_status),
3442         ONE("personality", S_IRUGO, proc_pid_personality),
3443         INF("limits",    S_IRUGO, proc_pid_limits),
3444 #ifdef CONFIG_SCHED_DEBUG
3445         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3446 #endif
3447         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3448 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3449         INF("syscall",   S_IRUGO, proc_pid_syscall),
3450 #endif
3451         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3452         ONE("stat",      S_IRUGO, proc_tid_stat),
3453         ONE("statm",     S_IRUGO, proc_pid_statm),
3454         REG("maps",      S_IRUGO, proc_maps_operations),
3455 #ifdef CONFIG_NUMA
3456         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3457 #endif
3458         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3459         LNK("cwd",       proc_cwd_link),
3460         LNK("root",      proc_root_link),
3461         LNK("exe",       proc_exe_link),
3462         REG("mounts",    S_IRUGO, proc_mounts_operations),
3463         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3464 #ifdef CONFIG_PROC_PAGE_MONITOR
3465         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3466         REG("smaps",     S_IRUGO, proc_smaps_operations),
3467         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3468 #endif
3469 #ifdef CONFIG_SECURITY
3470         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3471 #endif
3472 #ifdef CONFIG_KALLSYMS
3473         INF("wchan",     S_IRUGO, proc_pid_wchan),
3474 #endif
3475 #ifdef CONFIG_STACKTRACE
3476         ONE("stack",      S_IRUGO, proc_pid_stack),
3477 #endif
3478 #ifdef CONFIG_SCHEDSTATS
3479         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3480 #endif
3481 #ifdef CONFIG_LATENCYTOP
3482         REG("latency",  S_IRUGO, proc_lstats_operations),
3483 #endif
3484 #ifdef CONFIG_PROC_PID_CPUSET
3485         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3486 #endif
3487 #ifdef CONFIG_CGROUPS
3488         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3489 #endif
3490         INF("oom_score", S_IRUGO, proc_oom_score),
3491         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3492         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3493 #ifdef CONFIG_AUDITSYSCALL
3494         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3495         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3496 #endif
3497 #ifdef CONFIG_FAULT_INJECTION
3498         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3499 #endif
3500 #ifdef CONFIG_TASK_IO_ACCOUNTING
3501         INF("io",       S_IRUSR, proc_tid_io_accounting),
3502 #endif
3503 #ifdef CONFIG_HARDWALL
3504         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3505 #endif
3506 };
3507
3508 static int proc_tid_base_readdir(struct file * filp,
3509                              void * dirent, filldir_t filldir)
3510 {
3511         return proc_pident_readdir(filp,dirent,filldir,
3512                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3513 }
3514
3515 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3516         return proc_pident_lookup(dir, dentry,
3517                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3518 }
3519
3520 static const struct file_operations proc_tid_base_operations = {
3521         .read           = generic_read_dir,
3522         .readdir        = proc_tid_base_readdir,
3523         .llseek         = default_llseek,
3524 };
3525
3526 static const struct inode_operations proc_tid_base_inode_operations = {
3527         .lookup         = proc_tid_base_lookup,
3528         .getattr        = pid_getattr,
3529         .setattr        = proc_setattr,
3530 };
3531
3532 static struct dentry *proc_task_instantiate(struct inode *dir,
3533         struct dentry *dentry, struct task_struct *task, const void *ptr)
3534 {
3535         struct dentry *error = ERR_PTR(-ENOENT);
3536         struct inode *inode;
3537         inode = proc_pid_make_inode(dir->i_sb, task);
3538
3539         if (!inode)
3540                 goto out;
3541         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3542         inode->i_op = &proc_tid_base_inode_operations;
3543         inode->i_fop = &proc_tid_base_operations;
3544         inode->i_flags|=S_IMMUTABLE;
3545
3546         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3547                                                   ARRAY_SIZE(tid_base_stuff)));
3548
3549         d_set_d_op(dentry, &pid_dentry_operations);
3550
3551         d_add(dentry, inode);
3552         /* Close the race of the process dying before we return the dentry */
3553         if (pid_revalidate(dentry, NULL))
3554                 error = NULL;
3555 out:
3556         return error;
3557 }
3558
3559 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3560 {
3561         struct dentry *result = ERR_PTR(-ENOENT);
3562         struct task_struct *task;
3563         struct task_struct *leader = get_proc_task(dir);
3564         unsigned tid;
3565         struct pid_namespace *ns;
3566
3567         if (!leader)
3568                 goto out_no_task;
3569
3570         tid = name_to_int(dentry);
3571         if (tid == ~0U)
3572                 goto out;
3573
3574         ns = dentry->d_sb->s_fs_info;
3575         rcu_read_lock();
3576         task = find_task_by_pid_ns(tid, ns);
3577         if (task)
3578                 get_task_struct(task);
3579         rcu_read_unlock();
3580         if (!task)
3581                 goto out;
3582         if (!same_thread_group(leader, task))
3583                 goto out_drop_task;
3584
3585         result = proc_task_instantiate(dir, dentry, task, NULL);
3586 out_drop_task:
3587         put_task_struct(task);
3588 out:
3589         put_task_struct(leader);
3590 out_no_task:
3591         return result;
3592 }
3593
3594 /*
3595  * Find the first tid of a thread group to return to user space.
3596  *
3597  * Usually this is just the thread group leader, but if the users
3598  * buffer was too small or there was a seek into the middle of the
3599  * directory we have more work todo.
3600  *
3601  * In the case of a short read we start with find_task_by_pid.
3602  *
3603  * In the case of a seek we start with the leader and walk nr
3604  * threads past it.
3605  */
3606 static struct task_struct *first_tid(struct task_struct *leader,
3607                 int tid, int nr, struct pid_namespace *ns)
3608 {
3609         struct task_struct *pos;
3610
3611         rcu_read_lock();
3612         /* Attempt to start with the pid of a thread */
3613         if (tid && (nr > 0)) {
3614                 pos = find_task_by_pid_ns(tid, ns);
3615                 if (pos && (pos->group_leader == leader))
3616                         goto found;
3617         }
3618
3619         /* If nr exceeds the number of threads there is nothing todo */
3620         pos = NULL;
3621         if (nr && nr >= get_nr_threads(leader))
3622                 goto out;
3623
3624         /* If we haven't found our starting place yet start
3625          * with the leader and walk nr threads forward.
3626          */
3627         for (pos = leader; nr > 0; --nr) {
3628                 pos = next_thread(pos);
3629                 if (pos == leader) {
3630                         pos = NULL;
3631                         goto out;
3632                 }
3633         }
3634 found:
3635         get_task_struct(pos);
3636 out:
3637         rcu_read_unlock();
3638         return pos;
3639 }
3640
3641 /*
3642  * Find the next thread in the thread list.
3643  * Return NULL if there is an error or no next thread.
3644  *
3645  * The reference to the input task_struct is released.
3646  */
3647 static struct task_struct *next_tid(struct task_struct *start)
3648 {
3649         struct task_struct *pos = NULL;
3650         rcu_read_lock();
3651         if (pid_alive(start)) {
3652                 pos = next_thread(start);
3653                 if (thread_group_leader(pos))
3654                         pos = NULL;
3655                 else
3656                         get_task_struct(pos);
3657         }
3658         rcu_read_unlock();
3659         put_task_struct(start);
3660         return pos;
3661 }
3662
3663 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3664         struct task_struct *task, int tid)
3665 {
3666         char name[PROC_NUMBUF];
3667         int len = snprintf(name, sizeof(name), "%d", tid);
3668         return proc_fill_cache(filp, dirent, filldir, name, len,
3669                                 proc_task_instantiate, task, NULL);
3670 }
3671
3672 /* for the /proc/TGID/task/ directories */
3673 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3674 {
3675         struct dentry *dentry = filp->f_path.dentry;
3676         struct inode *inode = dentry->d_inode;
3677         struct task_struct *leader = NULL;
3678         struct task_struct *task;
3679         int retval = -ENOENT;
3680         ino_t ino;
3681         int tid;
3682         struct pid_namespace *ns;
3683
3684         task = get_proc_task(inode);
3685         if (!task)
3686                 goto out_no_task;
3687         rcu_read_lock();
3688         if (pid_alive(task)) {
3689                 leader = task->group_leader;
3690                 get_task_struct(leader);
3691         }
3692         rcu_read_unlock();
3693         put_task_struct(task);
3694         if (!leader)
3695                 goto out_no_task;
3696         retval = 0;
3697
3698         switch ((unsigned long)filp->f_pos) {
3699         case 0:
3700                 ino = inode->i_ino;
3701                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3702                         goto out;
3703                 filp->f_pos++;
3704                 /* fall through */
3705         case 1:
3706                 ino = parent_ino(dentry);
3707                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3708                         goto out;
3709                 filp->f_pos++;
3710                 /* fall through */
3711         }
3712
3713         /* f_version caches the tgid value that the last readdir call couldn't
3714          * return. lseek aka telldir automagically resets f_version to 0.
3715          */
3716         ns = filp->f_dentry->d_sb->s_fs_info;
3717         tid = (int)filp->f_version;
3718         filp->f_version = 0;
3719         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3720              task;
3721              task = next_tid(task), filp->f_pos++) {
3722                 tid = task_pid_nr_ns(task, ns);
3723                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3724                         /* returning this tgid failed, save it as the first
3725                          * pid for the next readir call */
3726                         filp->f_version = (u64)tid;
3727                         put_task_struct(task);
3728                         break;
3729                 }
3730         }
3731 out:
3732         put_task_struct(leader);
3733 out_no_task:
3734         return retval;
3735 }
3736
3737 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3738 {
3739         struct inode *inode = dentry->d_inode;
3740         struct task_struct *p = get_proc_task(inode);
3741         generic_fillattr(inode, stat);
3742
3743         if (p) {
3744                 stat->nlink += get_nr_threads(p);
3745                 put_task_struct(p);
3746         }
3747
3748         return 0;
3749 }
3750
3751 static const struct inode_operations proc_task_inode_operations = {
3752         .lookup         = proc_task_lookup,
3753         .getattr        = proc_task_getattr,
3754         .setattr        = proc_setattr,
3755         .permission     = proc_pid_permission,
3756 };
3757
3758 static const struct file_operations proc_task_operations = {
3759         .read           = generic_read_dir,
3760         .readdir        = proc_task_readdir,
3761         .llseek         = default_llseek,
3762 };