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