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