btrfs: fix wrong reservation when doing delayed inode operations
[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 static 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         mm = check_mem_permission(task);
898         copied = PTR_ERR(mm);
899         if (IS_ERR(mm))
900                 goto out_task;
901
902         copied = -EIO;
903         if (file->private_data != (void *)((long)current->self_exec_id))
904                 goto out_mm;
905
906         copied = -ENOMEM;
907         page = (char *)__get_free_page(GFP_TEMPORARY);
908         if (!page)
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         free_page((unsigned long) page);
933 out_mm:
934         mmput(mm);
935 out_task:
936         put_task_struct(task);
937 out_no_task:
938         return copied;
939 }
940
941 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
942 {
943         switch (orig) {
944         case 0:
945                 file->f_pos = offset;
946                 break;
947         case 1:
948                 file->f_pos += offset;
949                 break;
950         default:
951                 return -EINVAL;
952         }
953         force_successful_syscall_return();
954         return file->f_pos;
955 }
956
957 static const struct file_operations proc_mem_operations = {
958         .llseek         = mem_lseek,
959         .read           = mem_read,
960         .write          = mem_write,
961         .open           = mem_open,
962 };
963
964 static ssize_t environ_read(struct file *file, char __user *buf,
965                         size_t count, loff_t *ppos)
966 {
967         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
968         char *page;
969         unsigned long src = *ppos;
970         int ret = -ESRCH;
971         struct mm_struct *mm;
972
973         if (!task)
974                 goto out_no_task;
975
976         ret = -ENOMEM;
977         page = (char *)__get_free_page(GFP_TEMPORARY);
978         if (!page)
979                 goto out;
980
981
982         mm = mm_for_maps(task);
983         ret = PTR_ERR(mm);
984         if (!mm || IS_ERR(mm))
985                 goto out_free;
986
987         ret = 0;
988         while (count > 0) {
989                 int this_len, retval, max_len;
990
991                 this_len = mm->env_end - (mm->env_start + src);
992
993                 if (this_len <= 0)
994                         break;
995
996                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
997                 this_len = (this_len > max_len) ? max_len : this_len;
998
999                 retval = access_process_vm(task, (mm->env_start + src),
1000                         page, this_len, 0);
1001
1002                 if (retval <= 0) {
1003                         ret = retval;
1004                         break;
1005                 }
1006
1007                 if (copy_to_user(buf, page, retval)) {
1008                         ret = -EFAULT;
1009                         break;
1010                 }
1011
1012                 ret += retval;
1013                 src += retval;
1014                 buf += retval;
1015                 count -= retval;
1016         }
1017         *ppos = src;
1018
1019         mmput(mm);
1020 out_free:
1021         free_page((unsigned long) page);
1022 out:
1023         put_task_struct(task);
1024 out_no_task:
1025         return ret;
1026 }
1027
1028 static const struct file_operations proc_environ_operations = {
1029         .read           = environ_read,
1030         .llseek         = generic_file_llseek,
1031 };
1032
1033 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1034                                 size_t count, loff_t *ppos)
1035 {
1036         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1037         char buffer[PROC_NUMBUF];
1038         size_t len;
1039         int oom_adjust = OOM_DISABLE;
1040         unsigned long flags;
1041
1042         if (!task)
1043                 return -ESRCH;
1044
1045         if (lock_task_sighand(task, &flags)) {
1046                 oom_adjust = task->signal->oom_adj;
1047                 unlock_task_sighand(task, &flags);
1048         }
1049
1050         put_task_struct(task);
1051
1052         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1053
1054         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1055 }
1056
1057 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1058                                 size_t count, loff_t *ppos)
1059 {
1060         struct task_struct *task;
1061         char buffer[PROC_NUMBUF];
1062         long oom_adjust;
1063         unsigned long flags;
1064         int err;
1065
1066         memset(buffer, 0, sizeof(buffer));
1067         if (count > sizeof(buffer) - 1)
1068                 count = sizeof(buffer) - 1;
1069         if (copy_from_user(buffer, buf, count)) {
1070                 err = -EFAULT;
1071                 goto out;
1072         }
1073
1074         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1075         if (err)
1076                 goto out;
1077         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1078              oom_adjust != OOM_DISABLE) {
1079                 err = -EINVAL;
1080                 goto out;
1081         }
1082
1083         task = get_proc_task(file->f_path.dentry->d_inode);
1084         if (!task) {
1085                 err = -ESRCH;
1086                 goto out;
1087         }
1088
1089         task_lock(task);
1090         if (!task->mm) {
1091                 err = -EINVAL;
1092                 goto err_task_lock;
1093         }
1094
1095         if (!lock_task_sighand(task, &flags)) {
1096                 err = -ESRCH;
1097                 goto err_task_lock;
1098         }
1099
1100         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1101                 err = -EACCES;
1102                 goto err_sighand;
1103         }
1104
1105         if (oom_adjust != task->signal->oom_adj) {
1106                 if (oom_adjust == OOM_DISABLE)
1107                         atomic_inc(&task->mm->oom_disable_count);
1108                 if (task->signal->oom_adj == OOM_DISABLE)
1109                         atomic_dec(&task->mm->oom_disable_count);
1110         }
1111
1112         /*
1113          * Warn that /proc/pid/oom_adj is deprecated, see
1114          * Documentation/feature-removal-schedule.txt.
1115          */
1116         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1117                         "please use /proc/%d/oom_score_adj instead.\n",
1118                         current->comm, task_pid_nr(current),
1119                         task_pid_nr(task), task_pid_nr(task));
1120         task->signal->oom_adj = oom_adjust;
1121         /*
1122          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1123          * value is always attainable.
1124          */
1125         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1126                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1127         else
1128                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1129                                                                 -OOM_DISABLE;
1130 err_sighand:
1131         unlock_task_sighand(task, &flags);
1132 err_task_lock:
1133         task_unlock(task);
1134         put_task_struct(task);
1135 out:
1136         return err < 0 ? err : count;
1137 }
1138
1139 static const struct file_operations proc_oom_adjust_operations = {
1140         .read           = oom_adjust_read,
1141         .write          = oom_adjust_write,
1142         .llseek         = generic_file_llseek,
1143 };
1144
1145 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1146                                         size_t count, loff_t *ppos)
1147 {
1148         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1149         char buffer[PROC_NUMBUF];
1150         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1151         unsigned long flags;
1152         size_t len;
1153
1154         if (!task)
1155                 return -ESRCH;
1156         if (lock_task_sighand(task, &flags)) {
1157                 oom_score_adj = task->signal->oom_score_adj;
1158                 unlock_task_sighand(task, &flags);
1159         }
1160         put_task_struct(task);
1161         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1162         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1163 }
1164
1165 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1166                                         size_t count, loff_t *ppos)
1167 {
1168         struct task_struct *task;
1169         char buffer[PROC_NUMBUF];
1170         unsigned long flags;
1171         long oom_score_adj;
1172         int err;
1173
1174         memset(buffer, 0, sizeof(buffer));
1175         if (count > sizeof(buffer) - 1)
1176                 count = sizeof(buffer) - 1;
1177         if (copy_from_user(buffer, buf, count)) {
1178                 err = -EFAULT;
1179                 goto out;
1180         }
1181
1182         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1183         if (err)
1184                 goto out;
1185         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1186                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1187                 err = -EINVAL;
1188                 goto out;
1189         }
1190
1191         task = get_proc_task(file->f_path.dentry->d_inode);
1192         if (!task) {
1193                 err = -ESRCH;
1194                 goto out;
1195         }
1196
1197         task_lock(task);
1198         if (!task->mm) {
1199                 err = -EINVAL;
1200                 goto err_task_lock;
1201         }
1202
1203         if (!lock_task_sighand(task, &flags)) {
1204                 err = -ESRCH;
1205                 goto err_task_lock;
1206         }
1207
1208         if (oom_score_adj < task->signal->oom_score_adj_min &&
1209                         !capable(CAP_SYS_RESOURCE)) {
1210                 err = -EACCES;
1211                 goto err_sighand;
1212         }
1213
1214         if (oom_score_adj != task->signal->oom_score_adj) {
1215                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1216                         atomic_inc(&task->mm->oom_disable_count);
1217                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1218                         atomic_dec(&task->mm->oom_disable_count);
1219         }
1220         task->signal->oom_score_adj = oom_score_adj;
1221         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1222                 task->signal->oom_score_adj_min = oom_score_adj;
1223         /*
1224          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1225          * always attainable.
1226          */
1227         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1228                 task->signal->oom_adj = OOM_DISABLE;
1229         else
1230                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1231                                                         OOM_SCORE_ADJ_MAX;
1232 err_sighand:
1233         unlock_task_sighand(task, &flags);
1234 err_task_lock:
1235         task_unlock(task);
1236         put_task_struct(task);
1237 out:
1238         return err < 0 ? err : count;
1239 }
1240
1241 static const struct file_operations proc_oom_score_adj_operations = {
1242         .read           = oom_score_adj_read,
1243         .write          = oom_score_adj_write,
1244         .llseek         = default_llseek,
1245 };
1246
1247 #ifdef CONFIG_AUDITSYSCALL
1248 #define TMPBUFLEN 21
1249 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1250                                   size_t count, loff_t *ppos)
1251 {
1252         struct inode * inode = file->f_path.dentry->d_inode;
1253         struct task_struct *task = get_proc_task(inode);
1254         ssize_t length;
1255         char tmpbuf[TMPBUFLEN];
1256
1257         if (!task)
1258                 return -ESRCH;
1259         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1260                                 audit_get_loginuid(task));
1261         put_task_struct(task);
1262         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1263 }
1264
1265 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1266                                    size_t count, loff_t *ppos)
1267 {
1268         struct inode * inode = file->f_path.dentry->d_inode;
1269         char *page, *tmp;
1270         ssize_t length;
1271         uid_t loginuid;
1272
1273         if (!capable(CAP_AUDIT_CONTROL))
1274                 return -EPERM;
1275
1276         rcu_read_lock();
1277         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1278                 rcu_read_unlock();
1279                 return -EPERM;
1280         }
1281         rcu_read_unlock();
1282
1283         if (count >= PAGE_SIZE)
1284                 count = PAGE_SIZE - 1;
1285
1286         if (*ppos != 0) {
1287                 /* No partial writes. */
1288                 return -EINVAL;
1289         }
1290         page = (char*)__get_free_page(GFP_TEMPORARY);
1291         if (!page)
1292                 return -ENOMEM;
1293         length = -EFAULT;
1294         if (copy_from_user(page, buf, count))
1295                 goto out_free_page;
1296
1297         page[count] = '\0';
1298         loginuid = simple_strtoul(page, &tmp, 10);
1299         if (tmp == page) {
1300                 length = -EINVAL;
1301                 goto out_free_page;
1302
1303         }
1304         length = audit_set_loginuid(current, loginuid);
1305         if (likely(length == 0))
1306                 length = count;
1307
1308 out_free_page:
1309         free_page((unsigned long) page);
1310         return length;
1311 }
1312
1313 static const struct file_operations proc_loginuid_operations = {
1314         .read           = proc_loginuid_read,
1315         .write          = proc_loginuid_write,
1316         .llseek         = generic_file_llseek,
1317 };
1318
1319 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1320                                   size_t count, loff_t *ppos)
1321 {
1322         struct inode * inode = file->f_path.dentry->d_inode;
1323         struct task_struct *task = get_proc_task(inode);
1324         ssize_t length;
1325         char tmpbuf[TMPBUFLEN];
1326
1327         if (!task)
1328                 return -ESRCH;
1329         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1330                                 audit_get_sessionid(task));
1331         put_task_struct(task);
1332         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1333 }
1334
1335 static const struct file_operations proc_sessionid_operations = {
1336         .read           = proc_sessionid_read,
1337         .llseek         = generic_file_llseek,
1338 };
1339 #endif
1340
1341 #ifdef CONFIG_FAULT_INJECTION
1342 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1343                                       size_t count, loff_t *ppos)
1344 {
1345         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1346         char buffer[PROC_NUMBUF];
1347         size_t len;
1348         int make_it_fail;
1349
1350         if (!task)
1351                 return -ESRCH;
1352         make_it_fail = task->make_it_fail;
1353         put_task_struct(task);
1354
1355         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1356
1357         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1358 }
1359
1360 static ssize_t proc_fault_inject_write(struct file * file,
1361                         const char __user * buf, size_t count, loff_t *ppos)
1362 {
1363         struct task_struct *task;
1364         char buffer[PROC_NUMBUF], *end;
1365         int make_it_fail;
1366
1367         if (!capable(CAP_SYS_RESOURCE))
1368                 return -EPERM;
1369         memset(buffer, 0, sizeof(buffer));
1370         if (count > sizeof(buffer) - 1)
1371                 count = sizeof(buffer) - 1;
1372         if (copy_from_user(buffer, buf, count))
1373                 return -EFAULT;
1374         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1375         if (*end)
1376                 return -EINVAL;
1377         task = get_proc_task(file->f_dentry->d_inode);
1378         if (!task)
1379                 return -ESRCH;
1380         task->make_it_fail = make_it_fail;
1381         put_task_struct(task);
1382
1383         return count;
1384 }
1385
1386 static const struct file_operations proc_fault_inject_operations = {
1387         .read           = proc_fault_inject_read,
1388         .write          = proc_fault_inject_write,
1389         .llseek         = generic_file_llseek,
1390 };
1391 #endif
1392
1393
1394 #ifdef CONFIG_SCHED_DEBUG
1395 /*
1396  * Print out various scheduling related per-task fields:
1397  */
1398 static int sched_show(struct seq_file *m, void *v)
1399 {
1400         struct inode *inode = m->private;
1401         struct task_struct *p;
1402
1403         p = get_proc_task(inode);
1404         if (!p)
1405                 return -ESRCH;
1406         proc_sched_show_task(p, m);
1407
1408         put_task_struct(p);
1409
1410         return 0;
1411 }
1412
1413 static ssize_t
1414 sched_write(struct file *file, const char __user *buf,
1415             size_t count, loff_t *offset)
1416 {
1417         struct inode *inode = file->f_path.dentry->d_inode;
1418         struct task_struct *p;
1419
1420         p = get_proc_task(inode);
1421         if (!p)
1422                 return -ESRCH;
1423         proc_sched_set_task(p);
1424
1425         put_task_struct(p);
1426
1427         return count;
1428 }
1429
1430 static int sched_open(struct inode *inode, struct file *filp)
1431 {
1432         return single_open(filp, sched_show, inode);
1433 }
1434
1435 static const struct file_operations proc_pid_sched_operations = {
1436         .open           = sched_open,
1437         .read           = seq_read,
1438         .write          = sched_write,
1439         .llseek         = seq_lseek,
1440         .release        = single_release,
1441 };
1442
1443 #endif
1444
1445 #ifdef CONFIG_SCHED_AUTOGROUP
1446 /*
1447  * Print out autogroup related information:
1448  */
1449 static int sched_autogroup_show(struct seq_file *m, void *v)
1450 {
1451         struct inode *inode = m->private;
1452         struct task_struct *p;
1453
1454         p = get_proc_task(inode);
1455         if (!p)
1456                 return -ESRCH;
1457         proc_sched_autogroup_show_task(p, m);
1458
1459         put_task_struct(p);
1460
1461         return 0;
1462 }
1463
1464 static ssize_t
1465 sched_autogroup_write(struct file *file, const char __user *buf,
1466             size_t count, loff_t *offset)
1467 {
1468         struct inode *inode = file->f_path.dentry->d_inode;
1469         struct task_struct *p;
1470         char buffer[PROC_NUMBUF];
1471         long nice;
1472         int err;
1473
1474         memset(buffer, 0, sizeof(buffer));
1475         if (count > sizeof(buffer) - 1)
1476                 count = sizeof(buffer) - 1;
1477         if (copy_from_user(buffer, buf, count))
1478                 return -EFAULT;
1479
1480         err = strict_strtol(strstrip(buffer), 0, &nice);
1481         if (err)
1482                 return -EINVAL;
1483
1484         p = get_proc_task(inode);
1485         if (!p)
1486                 return -ESRCH;
1487
1488         err = nice;
1489         err = proc_sched_autogroup_set_nice(p, &err);
1490         if (err)
1491                 count = err;
1492
1493         put_task_struct(p);
1494
1495         return count;
1496 }
1497
1498 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1499 {
1500         int ret;
1501
1502         ret = single_open(filp, sched_autogroup_show, NULL);
1503         if (!ret) {
1504                 struct seq_file *m = filp->private_data;
1505
1506                 m->private = inode;
1507         }
1508         return ret;
1509 }
1510
1511 static const struct file_operations proc_pid_sched_autogroup_operations = {
1512         .open           = sched_autogroup_open,
1513         .read           = seq_read,
1514         .write          = sched_autogroup_write,
1515         .llseek         = seq_lseek,
1516         .release        = single_release,
1517 };
1518
1519 #endif /* CONFIG_SCHED_AUTOGROUP */
1520
1521 static ssize_t comm_write(struct file *file, const char __user *buf,
1522                                 size_t count, loff_t *offset)
1523 {
1524         struct inode *inode = file->f_path.dentry->d_inode;
1525         struct task_struct *p;
1526         char buffer[TASK_COMM_LEN];
1527
1528         memset(buffer, 0, sizeof(buffer));
1529         if (count > sizeof(buffer) - 1)
1530                 count = sizeof(buffer) - 1;
1531         if (copy_from_user(buffer, buf, count))
1532                 return -EFAULT;
1533
1534         p = get_proc_task(inode);
1535         if (!p)
1536                 return -ESRCH;
1537
1538         if (same_thread_group(current, p))
1539                 set_task_comm(p, buffer);
1540         else
1541                 count = -EINVAL;
1542
1543         put_task_struct(p);
1544
1545         return count;
1546 }
1547
1548 static int comm_show(struct seq_file *m, void *v)
1549 {
1550         struct inode *inode = m->private;
1551         struct task_struct *p;
1552
1553         p = get_proc_task(inode);
1554         if (!p)
1555                 return -ESRCH;
1556
1557         task_lock(p);
1558         seq_printf(m, "%s\n", p->comm);
1559         task_unlock(p);
1560
1561         put_task_struct(p);
1562
1563         return 0;
1564 }
1565
1566 static int comm_open(struct inode *inode, struct file *filp)
1567 {
1568         return single_open(filp, comm_show, inode);
1569 }
1570
1571 static const struct file_operations proc_pid_set_comm_operations = {
1572         .open           = comm_open,
1573         .read           = seq_read,
1574         .write          = comm_write,
1575         .llseek         = seq_lseek,
1576         .release        = single_release,
1577 };
1578
1579 /*
1580  * We added or removed a vma mapping the executable. The vmas are only mapped
1581  * during exec and are not mapped with the mmap system call.
1582  * Callers must hold down_write() on the mm's mmap_sem for these
1583  */
1584 void added_exe_file_vma(struct mm_struct *mm)
1585 {
1586         mm->num_exe_file_vmas++;
1587 }
1588
1589 void removed_exe_file_vma(struct mm_struct *mm)
1590 {
1591         mm->num_exe_file_vmas--;
1592         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1593                 fput(mm->exe_file);
1594                 mm->exe_file = NULL;
1595         }
1596
1597 }
1598
1599 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1600 {
1601         if (new_exe_file)
1602                 get_file(new_exe_file);
1603         if (mm->exe_file)
1604                 fput(mm->exe_file);
1605         mm->exe_file = new_exe_file;
1606         mm->num_exe_file_vmas = 0;
1607 }
1608
1609 struct file *get_mm_exe_file(struct mm_struct *mm)
1610 {
1611         struct file *exe_file;
1612
1613         /* We need mmap_sem to protect against races with removal of
1614          * VM_EXECUTABLE vmas */
1615         down_read(&mm->mmap_sem);
1616         exe_file = mm->exe_file;
1617         if (exe_file)
1618                 get_file(exe_file);
1619         up_read(&mm->mmap_sem);
1620         return exe_file;
1621 }
1622
1623 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1624 {
1625         /* It's safe to write the exe_file pointer without exe_file_lock because
1626          * this is called during fork when the task is not yet in /proc */
1627         newmm->exe_file = get_mm_exe_file(oldmm);
1628 }
1629
1630 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1631 {
1632         struct task_struct *task;
1633         struct mm_struct *mm;
1634         struct file *exe_file;
1635
1636         task = get_proc_task(inode);
1637         if (!task)
1638                 return -ENOENT;
1639         mm = get_task_mm(task);
1640         put_task_struct(task);
1641         if (!mm)
1642                 return -ENOENT;
1643         exe_file = get_mm_exe_file(mm);
1644         mmput(mm);
1645         if (exe_file) {
1646                 *exe_path = exe_file->f_path;
1647                 path_get(&exe_file->f_path);
1648                 fput(exe_file);
1649                 return 0;
1650         } else
1651                 return -ENOENT;
1652 }
1653
1654 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1655 {
1656         struct inode *inode = dentry->d_inode;
1657         int error = -EACCES;
1658
1659         /* We don't need a base pointer in the /proc filesystem */
1660         path_put(&nd->path);
1661
1662         /* Are we allowed to snoop on the tasks file descriptors? */
1663         if (!proc_fd_access_allowed(inode))
1664                 goto out;
1665
1666         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1667 out:
1668         return ERR_PTR(error);
1669 }
1670
1671 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1672 {
1673         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1674         char *pathname;
1675         int len;
1676
1677         if (!tmp)
1678                 return -ENOMEM;
1679
1680         pathname = d_path(path, tmp, PAGE_SIZE);
1681         len = PTR_ERR(pathname);
1682         if (IS_ERR(pathname))
1683                 goto out;
1684         len = tmp + PAGE_SIZE - 1 - pathname;
1685
1686         if (len > buflen)
1687                 len = buflen;
1688         if (copy_to_user(buffer, pathname, len))
1689                 len = -EFAULT;
1690  out:
1691         free_page((unsigned long)tmp);
1692         return len;
1693 }
1694
1695 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1696 {
1697         int error = -EACCES;
1698         struct inode *inode = dentry->d_inode;
1699         struct path path;
1700
1701         /* Are we allowed to snoop on the tasks file descriptors? */
1702         if (!proc_fd_access_allowed(inode))
1703                 goto out;
1704
1705         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1706         if (error)
1707                 goto out;
1708
1709         error = do_proc_readlink(&path, buffer, buflen);
1710         path_put(&path);
1711 out:
1712         return error;
1713 }
1714
1715 static const struct inode_operations proc_pid_link_inode_operations = {
1716         .readlink       = proc_pid_readlink,
1717         .follow_link    = proc_pid_follow_link,
1718         .setattr        = proc_setattr,
1719 };
1720
1721
1722 /* building an inode */
1723
1724 static int task_dumpable(struct task_struct *task)
1725 {
1726         int dumpable = 0;
1727         struct mm_struct *mm;
1728
1729         task_lock(task);
1730         mm = task->mm;
1731         if (mm)
1732                 dumpable = get_dumpable(mm);
1733         task_unlock(task);
1734         if(dumpable == 1)
1735                 return 1;
1736         return 0;
1737 }
1738
1739
1740 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1741 {
1742         struct inode * inode;
1743         struct proc_inode *ei;
1744         const struct cred *cred;
1745
1746         /* We need a new inode */
1747
1748         inode = new_inode(sb);
1749         if (!inode)
1750                 goto out;
1751
1752         /* Common stuff */
1753         ei = PROC_I(inode);
1754         inode->i_ino = get_next_ino();
1755         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1756         inode->i_op = &proc_def_inode_operations;
1757
1758         /*
1759          * grab the reference to task.
1760          */
1761         ei->pid = get_task_pid(task, PIDTYPE_PID);
1762         if (!ei->pid)
1763                 goto out_unlock;
1764
1765         if (task_dumpable(task)) {
1766                 rcu_read_lock();
1767                 cred = __task_cred(task);
1768                 inode->i_uid = cred->euid;
1769                 inode->i_gid = cred->egid;
1770                 rcu_read_unlock();
1771         }
1772         security_task_to_inode(task, inode);
1773
1774 out:
1775         return inode;
1776
1777 out_unlock:
1778         iput(inode);
1779         return NULL;
1780 }
1781
1782 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1783 {
1784         struct inode *inode = dentry->d_inode;
1785         struct task_struct *task;
1786         const struct cred *cred;
1787
1788         generic_fillattr(inode, stat);
1789
1790         rcu_read_lock();
1791         stat->uid = 0;
1792         stat->gid = 0;
1793         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1794         if (task) {
1795                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1796                     task_dumpable(task)) {
1797                         cred = __task_cred(task);
1798                         stat->uid = cred->euid;
1799                         stat->gid = cred->egid;
1800                 }
1801         }
1802         rcu_read_unlock();
1803         return 0;
1804 }
1805
1806 /* dentry stuff */
1807
1808 /*
1809  *      Exceptional case: normally we are not allowed to unhash a busy
1810  * directory. In this case, however, we can do it - no aliasing problems
1811  * due to the way we treat inodes.
1812  *
1813  * Rewrite the inode's ownerships here because the owning task may have
1814  * performed a setuid(), etc.
1815  *
1816  * Before the /proc/pid/status file was created the only way to read
1817  * the effective uid of a /process was to stat /proc/pid.  Reading
1818  * /proc/pid/status is slow enough that procps and other packages
1819  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1820  * made this apply to all per process world readable and executable
1821  * directories.
1822  */
1823 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1824 {
1825         struct inode *inode;
1826         struct task_struct *task;
1827         const struct cred *cred;
1828
1829         if (nd && nd->flags & LOOKUP_RCU)
1830                 return -ECHILD;
1831
1832         inode = dentry->d_inode;
1833         task = get_proc_task(inode);
1834
1835         if (task) {
1836                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1837                     task_dumpable(task)) {
1838                         rcu_read_lock();
1839                         cred = __task_cred(task);
1840                         inode->i_uid = cred->euid;
1841                         inode->i_gid = cred->egid;
1842                         rcu_read_unlock();
1843                 } else {
1844                         inode->i_uid = 0;
1845                         inode->i_gid = 0;
1846                 }
1847                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1848                 security_task_to_inode(task, inode);
1849                 put_task_struct(task);
1850                 return 1;
1851         }
1852         d_drop(dentry);
1853         return 0;
1854 }
1855
1856 static int pid_delete_dentry(const struct dentry * dentry)
1857 {
1858         /* Is the task we represent dead?
1859          * If so, then don't put the dentry on the lru list,
1860          * kill it immediately.
1861          */
1862         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1863 }
1864
1865 static const struct dentry_operations pid_dentry_operations =
1866 {
1867         .d_revalidate   = pid_revalidate,
1868         .d_delete       = pid_delete_dentry,
1869 };
1870
1871 /* Lookups */
1872
1873 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1874                                 struct task_struct *, const void *);
1875
1876 /*
1877  * Fill a directory entry.
1878  *
1879  * If possible create the dcache entry and derive our inode number and
1880  * file type from dcache entry.
1881  *
1882  * Since all of the proc inode numbers are dynamically generated, the inode
1883  * numbers do not exist until the inode is cache.  This means creating the
1884  * the dcache entry in readdir is necessary to keep the inode numbers
1885  * reported by readdir in sync with the inode numbers reported
1886  * by stat.
1887  */
1888 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1889         char *name, int len,
1890         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1891 {
1892         struct dentry *child, *dir = filp->f_path.dentry;
1893         struct inode *inode;
1894         struct qstr qname;
1895         ino_t ino = 0;
1896         unsigned type = DT_UNKNOWN;
1897
1898         qname.name = name;
1899         qname.len  = len;
1900         qname.hash = full_name_hash(name, len);
1901
1902         child = d_lookup(dir, &qname);
1903         if (!child) {
1904                 struct dentry *new;
1905                 new = d_alloc(dir, &qname);
1906                 if (new) {
1907                         child = instantiate(dir->d_inode, new, task, ptr);
1908                         if (child)
1909                                 dput(new);
1910                         else
1911                                 child = new;
1912                 }
1913         }
1914         if (!child || IS_ERR(child) || !child->d_inode)
1915                 goto end_instantiate;
1916         inode = child->d_inode;
1917         if (inode) {
1918                 ino = inode->i_ino;
1919                 type = inode->i_mode >> 12;
1920         }
1921         dput(child);
1922 end_instantiate:
1923         if (!ino)
1924                 ino = find_inode_number(dir, &qname);
1925         if (!ino)
1926                 ino = 1;
1927         return filldir(dirent, name, len, filp->f_pos, ino, type);
1928 }
1929
1930 static unsigned name_to_int(struct dentry *dentry)
1931 {
1932         const char *name = dentry->d_name.name;
1933         int len = dentry->d_name.len;
1934         unsigned n = 0;
1935
1936         if (len > 1 && *name == '0')
1937                 goto out;
1938         while (len-- > 0) {
1939                 unsigned c = *name++ - '0';
1940                 if (c > 9)
1941                         goto out;
1942                 if (n >= (~0U-9)/10)
1943                         goto out;
1944                 n *= 10;
1945                 n += c;
1946         }
1947         return n;
1948 out:
1949         return ~0U;
1950 }
1951
1952 #define PROC_FDINFO_MAX 64
1953
1954 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1955 {
1956         struct task_struct *task = get_proc_task(inode);
1957         struct files_struct *files = NULL;
1958         struct file *file;
1959         int fd = proc_fd(inode);
1960
1961         if (task) {
1962                 files = get_files_struct(task);
1963                 put_task_struct(task);
1964         }
1965         if (files) {
1966                 /*
1967                  * We are not taking a ref to the file structure, so we must
1968                  * hold ->file_lock.
1969                  */
1970                 spin_lock(&files->file_lock);
1971                 file = fcheck_files(files, fd);
1972                 if (file) {
1973                         if (path) {
1974                                 *path = file->f_path;
1975                                 path_get(&file->f_path);
1976                         }
1977                         if (info)
1978                                 snprintf(info, PROC_FDINFO_MAX,
1979                                          "pos:\t%lli\n"
1980                                          "flags:\t0%o\n",
1981                                          (long long) file->f_pos,
1982                                          file->f_flags);
1983                         spin_unlock(&files->file_lock);
1984                         put_files_struct(files);
1985                         return 0;
1986                 }
1987                 spin_unlock(&files->file_lock);
1988                 put_files_struct(files);
1989         }
1990         return -ENOENT;
1991 }
1992
1993 static int proc_fd_link(struct inode *inode, struct path *path)
1994 {
1995         return proc_fd_info(inode, path, NULL);
1996 }
1997
1998 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1999 {
2000         struct inode *inode;
2001         struct task_struct *task;
2002         int fd;
2003         struct files_struct *files;
2004         const struct cred *cred;
2005
2006         if (nd && nd->flags & LOOKUP_RCU)
2007                 return -ECHILD;
2008
2009         inode = dentry->d_inode;
2010         task = get_proc_task(inode);
2011         fd = proc_fd(inode);
2012
2013         if (task) {
2014                 files = get_files_struct(task);
2015                 if (files) {
2016                         rcu_read_lock();
2017                         if (fcheck_files(files, fd)) {
2018                                 rcu_read_unlock();
2019                                 put_files_struct(files);
2020                                 if (task_dumpable(task)) {
2021                                         rcu_read_lock();
2022                                         cred = __task_cred(task);
2023                                         inode->i_uid = cred->euid;
2024                                         inode->i_gid = cred->egid;
2025                                         rcu_read_unlock();
2026                                 } else {
2027                                         inode->i_uid = 0;
2028                                         inode->i_gid = 0;
2029                                 }
2030                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
2031                                 security_task_to_inode(task, inode);
2032                                 put_task_struct(task);
2033                                 return 1;
2034                         }
2035                         rcu_read_unlock();
2036                         put_files_struct(files);
2037                 }
2038                 put_task_struct(task);
2039         }
2040         d_drop(dentry);
2041         return 0;
2042 }
2043
2044 static const struct dentry_operations tid_fd_dentry_operations =
2045 {
2046         .d_revalidate   = tid_fd_revalidate,
2047         .d_delete       = pid_delete_dentry,
2048 };
2049
2050 static struct dentry *proc_fd_instantiate(struct inode *dir,
2051         struct dentry *dentry, struct task_struct *task, const void *ptr)
2052 {
2053         unsigned fd = *(const unsigned *)ptr;
2054         struct file *file;
2055         struct files_struct *files;
2056         struct inode *inode;
2057         struct proc_inode *ei;
2058         struct dentry *error = ERR_PTR(-ENOENT);
2059
2060         inode = proc_pid_make_inode(dir->i_sb, task);
2061         if (!inode)
2062                 goto out;
2063         ei = PROC_I(inode);
2064         ei->fd = fd;
2065         files = get_files_struct(task);
2066         if (!files)
2067                 goto out_iput;
2068         inode->i_mode = S_IFLNK;
2069
2070         /*
2071          * We are not taking a ref to the file structure, so we must
2072          * hold ->file_lock.
2073          */
2074         spin_lock(&files->file_lock);
2075         file = fcheck_files(files, fd);
2076         if (!file)
2077                 goto out_unlock;
2078         if (file->f_mode & FMODE_READ)
2079                 inode->i_mode |= S_IRUSR | S_IXUSR;
2080         if (file->f_mode & FMODE_WRITE)
2081                 inode->i_mode |= S_IWUSR | S_IXUSR;
2082         spin_unlock(&files->file_lock);
2083         put_files_struct(files);
2084
2085         inode->i_op = &proc_pid_link_inode_operations;
2086         inode->i_size = 64;
2087         ei->op.proc_get_link = proc_fd_link;
2088         d_set_d_op(dentry, &tid_fd_dentry_operations);
2089         d_add(dentry, inode);
2090         /* Close the race of the process dying before we return the dentry */
2091         if (tid_fd_revalidate(dentry, NULL))
2092                 error = NULL;
2093
2094  out:
2095         return error;
2096 out_unlock:
2097         spin_unlock(&files->file_lock);
2098         put_files_struct(files);
2099 out_iput:
2100         iput(inode);
2101         goto out;
2102 }
2103
2104 static struct dentry *proc_lookupfd_common(struct inode *dir,
2105                                            struct dentry *dentry,
2106                                            instantiate_t instantiate)
2107 {
2108         struct task_struct *task = get_proc_task(dir);
2109         unsigned fd = name_to_int(dentry);
2110         struct dentry *result = ERR_PTR(-ENOENT);
2111
2112         if (!task)
2113                 goto out_no_task;
2114         if (fd == ~0U)
2115                 goto out;
2116
2117         result = instantiate(dir, dentry, task, &fd);
2118 out:
2119         put_task_struct(task);
2120 out_no_task:
2121         return result;
2122 }
2123
2124 static int proc_readfd_common(struct file * filp, void * dirent,
2125                               filldir_t filldir, instantiate_t instantiate)
2126 {
2127         struct dentry *dentry = filp->f_path.dentry;
2128         struct inode *inode = dentry->d_inode;
2129         struct task_struct *p = get_proc_task(inode);
2130         unsigned int fd, ino;
2131         int retval;
2132         struct files_struct * files;
2133
2134         retval = -ENOENT;
2135         if (!p)
2136                 goto out_no_task;
2137         retval = 0;
2138
2139         fd = filp->f_pos;
2140         switch (fd) {
2141                 case 0:
2142                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2143                                 goto out;
2144                         filp->f_pos++;
2145                 case 1:
2146                         ino = parent_ino(dentry);
2147                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2148                                 goto out;
2149                         filp->f_pos++;
2150                 default:
2151                         files = get_files_struct(p);
2152                         if (!files)
2153                                 goto out;
2154                         rcu_read_lock();
2155                         for (fd = filp->f_pos-2;
2156                              fd < files_fdtable(files)->max_fds;
2157                              fd++, filp->f_pos++) {
2158                                 char name[PROC_NUMBUF];
2159                                 int len;
2160
2161                                 if (!fcheck_files(files, fd))
2162                                         continue;
2163                                 rcu_read_unlock();
2164
2165                                 len = snprintf(name, sizeof(name), "%d", fd);
2166                                 if (proc_fill_cache(filp, dirent, filldir,
2167                                                     name, len, instantiate,
2168                                                     p, &fd) < 0) {
2169                                         rcu_read_lock();
2170                                         break;
2171                                 }
2172                                 rcu_read_lock();
2173                         }
2174                         rcu_read_unlock();
2175                         put_files_struct(files);
2176         }
2177 out:
2178         put_task_struct(p);
2179 out_no_task:
2180         return retval;
2181 }
2182
2183 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2184                                     struct nameidata *nd)
2185 {
2186         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2187 }
2188
2189 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2190 {
2191         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2192 }
2193
2194 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2195                                       size_t len, loff_t *ppos)
2196 {
2197         char tmp[PROC_FDINFO_MAX];
2198         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2199         if (!err)
2200                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2201         return err;
2202 }
2203
2204 static const struct file_operations proc_fdinfo_file_operations = {
2205         .open           = nonseekable_open,
2206         .read           = proc_fdinfo_read,
2207         .llseek         = no_llseek,
2208 };
2209
2210 static const struct file_operations proc_fd_operations = {
2211         .read           = generic_read_dir,
2212         .readdir        = proc_readfd,
2213         .llseek         = default_llseek,
2214 };
2215
2216 /*
2217  * /proc/pid/fd needs a special permission handler so that a process can still
2218  * access /proc/self/fd after it has executed a setuid().
2219  */
2220 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2221 {
2222         int rv;
2223
2224         if (flags & IPERM_FLAG_RCU)
2225                 return -ECHILD;
2226         rv = generic_permission(inode, mask, flags, NULL);
2227         if (rv == 0)
2228                 return 0;
2229         if (task_pid(current) == proc_pid(inode))
2230                 rv = 0;
2231         return rv;
2232 }
2233
2234 /*
2235  * proc directories can do almost nothing..
2236  */
2237 static const struct inode_operations proc_fd_inode_operations = {
2238         .lookup         = proc_lookupfd,
2239         .permission     = proc_fd_permission,
2240         .setattr        = proc_setattr,
2241 };
2242
2243 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2244         struct dentry *dentry, struct task_struct *task, const void *ptr)
2245 {
2246         unsigned fd = *(unsigned *)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         ei = PROC_I(inode);
2255         ei->fd = fd;
2256         inode->i_mode = S_IFREG | S_IRUSR;
2257         inode->i_fop = &proc_fdinfo_file_operations;
2258         d_set_d_op(dentry, &tid_fd_dentry_operations);
2259         d_add(dentry, inode);
2260         /* Close the race of the process dying before we return the dentry */
2261         if (tid_fd_revalidate(dentry, NULL))
2262                 error = NULL;
2263
2264  out:
2265         return error;
2266 }
2267
2268 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2269                                         struct dentry *dentry,
2270                                         struct nameidata *nd)
2271 {
2272         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2273 }
2274
2275 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2276 {
2277         return proc_readfd_common(filp, dirent, filldir,
2278                                   proc_fdinfo_instantiate);
2279 }
2280
2281 static const struct file_operations proc_fdinfo_operations = {
2282         .read           = generic_read_dir,
2283         .readdir        = proc_readfdinfo,
2284         .llseek         = default_llseek,
2285 };
2286
2287 /*
2288  * proc directories can do almost nothing..
2289  */
2290 static const struct inode_operations proc_fdinfo_inode_operations = {
2291         .lookup         = proc_lookupfdinfo,
2292         .setattr        = proc_setattr,
2293 };
2294
2295
2296 static struct dentry *proc_pident_instantiate(struct inode *dir,
2297         struct dentry *dentry, struct task_struct *task, const void *ptr)
2298 {
2299         const struct pid_entry *p = ptr;
2300         struct inode *inode;
2301         struct proc_inode *ei;
2302         struct dentry *error = ERR_PTR(-ENOENT);
2303
2304         inode = proc_pid_make_inode(dir->i_sb, task);
2305         if (!inode)
2306                 goto out;
2307
2308         ei = PROC_I(inode);
2309         inode->i_mode = p->mode;
2310         if (S_ISDIR(inode->i_mode))
2311                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2312         if (p->iop)
2313                 inode->i_op = p->iop;
2314         if (p->fop)
2315                 inode->i_fop = p->fop;
2316         ei->op = p->op;
2317         d_set_d_op(dentry, &pid_dentry_operations);
2318         d_add(dentry, inode);
2319         /* Close the race of the process dying before we return the dentry */
2320         if (pid_revalidate(dentry, NULL))
2321                 error = NULL;
2322 out:
2323         return error;
2324 }
2325
2326 static struct dentry *proc_pident_lookup(struct inode *dir, 
2327                                          struct dentry *dentry,
2328                                          const struct pid_entry *ents,
2329                                          unsigned int nents)
2330 {
2331         struct dentry *error;
2332         struct task_struct *task = get_proc_task(dir);
2333         const struct pid_entry *p, *last;
2334
2335         error = ERR_PTR(-ENOENT);
2336
2337         if (!task)
2338                 goto out_no_task;
2339
2340         /*
2341          * Yes, it does not scale. And it should not. Don't add
2342          * new entries into /proc/<tgid>/ without very good reasons.
2343          */
2344         last = &ents[nents - 1];
2345         for (p = ents; p <= last; p++) {
2346                 if (p->len != dentry->d_name.len)
2347                         continue;
2348                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2349                         break;
2350         }
2351         if (p > last)
2352                 goto out;
2353
2354         error = proc_pident_instantiate(dir, dentry, task, p);
2355 out:
2356         put_task_struct(task);
2357 out_no_task:
2358         return error;
2359 }
2360
2361 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2362         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2363 {
2364         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2365                                 proc_pident_instantiate, task, p);
2366 }
2367
2368 static int proc_pident_readdir(struct file *filp,
2369                 void *dirent, filldir_t filldir,
2370                 const struct pid_entry *ents, unsigned int nents)
2371 {
2372         int i;
2373         struct dentry *dentry = filp->f_path.dentry;
2374         struct inode *inode = dentry->d_inode;
2375         struct task_struct *task = get_proc_task(inode);
2376         const struct pid_entry *p, *last;
2377         ino_t ino;
2378         int ret;
2379
2380         ret = -ENOENT;
2381         if (!task)
2382                 goto out_no_task;
2383
2384         ret = 0;
2385         i = filp->f_pos;
2386         switch (i) {
2387         case 0:
2388                 ino = inode->i_ino;
2389                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2390                         goto out;
2391                 i++;
2392                 filp->f_pos++;
2393                 /* fall through */
2394         case 1:
2395                 ino = parent_ino(dentry);
2396                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2397                         goto out;
2398                 i++;
2399                 filp->f_pos++;
2400                 /* fall through */
2401         default:
2402                 i -= 2;
2403                 if (i >= nents) {
2404                         ret = 1;
2405                         goto out;
2406                 }
2407                 p = ents + i;
2408                 last = &ents[nents - 1];
2409                 while (p <= last) {
2410                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2411                                 goto out;
2412                         filp->f_pos++;
2413                         p++;
2414                 }
2415         }
2416
2417         ret = 1;
2418 out:
2419         put_task_struct(task);
2420 out_no_task:
2421         return ret;
2422 }
2423
2424 #ifdef CONFIG_SECURITY
2425 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2426                                   size_t count, loff_t *ppos)
2427 {
2428         struct inode * inode = file->f_path.dentry->d_inode;
2429         char *p = NULL;
2430         ssize_t length;
2431         struct task_struct *task = get_proc_task(inode);
2432
2433         if (!task)
2434                 return -ESRCH;
2435
2436         length = security_getprocattr(task,
2437                                       (char*)file->f_path.dentry->d_name.name,
2438                                       &p);
2439         put_task_struct(task);
2440         if (length > 0)
2441                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2442         kfree(p);
2443         return length;
2444 }
2445
2446 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2447                                    size_t count, loff_t *ppos)
2448 {
2449         struct inode * inode = file->f_path.dentry->d_inode;
2450         char *page;
2451         ssize_t length;
2452         struct task_struct *task = get_proc_task(inode);
2453
2454         length = -ESRCH;
2455         if (!task)
2456                 goto out_no_task;
2457         if (count > PAGE_SIZE)
2458                 count = PAGE_SIZE;
2459
2460         /* No partial writes. */
2461         length = -EINVAL;
2462         if (*ppos != 0)
2463                 goto out;
2464
2465         length = -ENOMEM;
2466         page = (char*)__get_free_page(GFP_TEMPORARY);
2467         if (!page)
2468                 goto out;
2469
2470         length = -EFAULT;
2471         if (copy_from_user(page, buf, count))
2472                 goto out_free;
2473
2474         /* Guard against adverse ptrace interaction */
2475         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2476         if (length < 0)
2477                 goto out_free;
2478
2479         length = security_setprocattr(task,
2480                                       (char*)file->f_path.dentry->d_name.name,
2481                                       (void*)page, count);
2482         mutex_unlock(&task->signal->cred_guard_mutex);
2483 out_free:
2484         free_page((unsigned long) page);
2485 out:
2486         put_task_struct(task);
2487 out_no_task:
2488         return length;
2489 }
2490
2491 static const struct file_operations proc_pid_attr_operations = {
2492         .read           = proc_pid_attr_read,
2493         .write          = proc_pid_attr_write,
2494         .llseek         = generic_file_llseek,
2495 };
2496
2497 static const struct pid_entry attr_dir_stuff[] = {
2498         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2499         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2500         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2501         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2502         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2503         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2504 };
2505
2506 static int proc_attr_dir_readdir(struct file * filp,
2507                              void * dirent, filldir_t filldir)
2508 {
2509         return proc_pident_readdir(filp,dirent,filldir,
2510                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2511 }
2512
2513 static const struct file_operations proc_attr_dir_operations = {
2514         .read           = generic_read_dir,
2515         .readdir        = proc_attr_dir_readdir,
2516         .llseek         = default_llseek,
2517 };
2518
2519 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2520                                 struct dentry *dentry, struct nameidata *nd)
2521 {
2522         return proc_pident_lookup(dir, dentry,
2523                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2524 }
2525
2526 static const struct inode_operations proc_attr_dir_inode_operations = {
2527         .lookup         = proc_attr_dir_lookup,
2528         .getattr        = pid_getattr,
2529         .setattr        = proc_setattr,
2530 };
2531
2532 #endif
2533
2534 #ifdef CONFIG_ELF_CORE
2535 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2536                                          size_t count, loff_t *ppos)
2537 {
2538         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2539         struct mm_struct *mm;
2540         char buffer[PROC_NUMBUF];
2541         size_t len;
2542         int ret;
2543
2544         if (!task)
2545                 return -ESRCH;
2546
2547         ret = 0;
2548         mm = get_task_mm(task);
2549         if (mm) {
2550                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2551                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2552                                 MMF_DUMP_FILTER_SHIFT));
2553                 mmput(mm);
2554                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2555         }
2556
2557         put_task_struct(task);
2558
2559         return ret;
2560 }
2561
2562 static ssize_t proc_coredump_filter_write(struct file *file,
2563                                           const char __user *buf,
2564                                           size_t count,
2565                                           loff_t *ppos)
2566 {
2567         struct task_struct *task;
2568         struct mm_struct *mm;
2569         char buffer[PROC_NUMBUF], *end;
2570         unsigned int val;
2571         int ret;
2572         int i;
2573         unsigned long mask;
2574
2575         ret = -EFAULT;
2576         memset(buffer, 0, sizeof(buffer));
2577         if (count > sizeof(buffer) - 1)
2578                 count = sizeof(buffer) - 1;
2579         if (copy_from_user(buffer, buf, count))
2580                 goto out_no_task;
2581
2582         ret = -EINVAL;
2583         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2584         if (*end == '\n')
2585                 end++;
2586         if (end - buffer == 0)
2587                 goto out_no_task;
2588
2589         ret = -ESRCH;
2590         task = get_proc_task(file->f_dentry->d_inode);
2591         if (!task)
2592                 goto out_no_task;
2593
2594         ret = end - buffer;
2595         mm = get_task_mm(task);
2596         if (!mm)
2597                 goto out_no_mm;
2598
2599         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2600                 if (val & mask)
2601                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2602                 else
2603                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2604         }
2605
2606         mmput(mm);
2607  out_no_mm:
2608         put_task_struct(task);
2609  out_no_task:
2610         return ret;
2611 }
2612
2613 static const struct file_operations proc_coredump_filter_operations = {
2614         .read           = proc_coredump_filter_read,
2615         .write          = proc_coredump_filter_write,
2616         .llseek         = generic_file_llseek,
2617 };
2618 #endif
2619
2620 /*
2621  * /proc/self:
2622  */
2623 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2624                               int buflen)
2625 {
2626         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2627         pid_t tgid = task_tgid_nr_ns(current, ns);
2628         char tmp[PROC_NUMBUF];
2629         if (!tgid)
2630                 return -ENOENT;
2631         sprintf(tmp, "%d", tgid);
2632         return vfs_readlink(dentry,buffer,buflen,tmp);
2633 }
2634
2635 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2636 {
2637         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2638         pid_t tgid = task_tgid_nr_ns(current, ns);
2639         char *name = ERR_PTR(-ENOENT);
2640         if (tgid) {
2641                 name = __getname();
2642                 if (!name)
2643                         name = ERR_PTR(-ENOMEM);
2644                 else
2645                         sprintf(name, "%d", tgid);
2646         }
2647         nd_set_link(nd, name);
2648         return NULL;
2649 }
2650
2651 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2652                                 void *cookie)
2653 {
2654         char *s = nd_get_link(nd);
2655         if (!IS_ERR(s))
2656                 __putname(s);
2657 }
2658
2659 static const struct inode_operations proc_self_inode_operations = {
2660         .readlink       = proc_self_readlink,
2661         .follow_link    = proc_self_follow_link,
2662         .put_link       = proc_self_put_link,
2663 };
2664
2665 /*
2666  * proc base
2667  *
2668  * These are the directory entries in the root directory of /proc
2669  * that properly belong to the /proc filesystem, as they describe
2670  * describe something that is process related.
2671  */
2672 static const struct pid_entry proc_base_stuff[] = {
2673         NOD("self", S_IFLNK|S_IRWXUGO,
2674                 &proc_self_inode_operations, NULL, {}),
2675 };
2676
2677 static struct dentry *proc_base_instantiate(struct inode *dir,
2678         struct dentry *dentry, struct task_struct *task, const void *ptr)
2679 {
2680         const struct pid_entry *p = ptr;
2681         struct inode *inode;
2682         struct proc_inode *ei;
2683         struct dentry *error;
2684
2685         /* Allocate the inode */
2686         error = ERR_PTR(-ENOMEM);
2687         inode = new_inode(dir->i_sb);
2688         if (!inode)
2689                 goto out;
2690
2691         /* Initialize the inode */
2692         ei = PROC_I(inode);
2693         inode->i_ino = get_next_ino();
2694         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2695
2696         /*
2697          * grab the reference to the task.
2698          */
2699         ei->pid = get_task_pid(task, PIDTYPE_PID);
2700         if (!ei->pid)
2701                 goto out_iput;
2702
2703         inode->i_mode = p->mode;
2704         if (S_ISDIR(inode->i_mode))
2705                 inode->i_nlink = 2;
2706         if (S_ISLNK(inode->i_mode))
2707                 inode->i_size = 64;
2708         if (p->iop)
2709                 inode->i_op = p->iop;
2710         if (p->fop)
2711                 inode->i_fop = p->fop;
2712         ei->op = p->op;
2713         d_add(dentry, inode);
2714         error = NULL;
2715 out:
2716         return error;
2717 out_iput:
2718         iput(inode);
2719         goto out;
2720 }
2721
2722 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2723 {
2724         struct dentry *error;
2725         struct task_struct *task = get_proc_task(dir);
2726         const struct pid_entry *p, *last;
2727
2728         error = ERR_PTR(-ENOENT);
2729
2730         if (!task)
2731                 goto out_no_task;
2732
2733         /* Lookup the directory entry */
2734         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2735         for (p = proc_base_stuff; p <= last; p++) {
2736                 if (p->len != dentry->d_name.len)
2737                         continue;
2738                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2739                         break;
2740         }
2741         if (p > last)
2742                 goto out;
2743
2744         error = proc_base_instantiate(dir, dentry, task, p);
2745
2746 out:
2747         put_task_struct(task);
2748 out_no_task:
2749         return error;
2750 }
2751
2752 static int proc_base_fill_cache(struct file *filp, void *dirent,
2753         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2754 {
2755         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2756                                 proc_base_instantiate, task, p);
2757 }
2758
2759 #ifdef CONFIG_TASK_IO_ACCOUNTING
2760 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2761 {
2762         struct task_io_accounting acct = task->ioac;
2763         unsigned long flags;
2764
2765         if (whole && lock_task_sighand(task, &flags)) {
2766                 struct task_struct *t = task;
2767
2768                 task_io_accounting_add(&acct, &task->signal->ioac);
2769                 while_each_thread(task, t)
2770                         task_io_accounting_add(&acct, &t->ioac);
2771
2772                 unlock_task_sighand(task, &flags);
2773         }
2774         return sprintf(buffer,
2775                         "rchar: %llu\n"
2776                         "wchar: %llu\n"
2777                         "syscr: %llu\n"
2778                         "syscw: %llu\n"
2779                         "read_bytes: %llu\n"
2780                         "write_bytes: %llu\n"
2781                         "cancelled_write_bytes: %llu\n",
2782                         (unsigned long long)acct.rchar,
2783                         (unsigned long long)acct.wchar,
2784                         (unsigned long long)acct.syscr,
2785                         (unsigned long long)acct.syscw,
2786                         (unsigned long long)acct.read_bytes,
2787                         (unsigned long long)acct.write_bytes,
2788                         (unsigned long long)acct.cancelled_write_bytes);
2789 }
2790
2791 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2792 {
2793         return do_io_accounting(task, buffer, 0);
2794 }
2795
2796 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2797 {
2798         return do_io_accounting(task, buffer, 1);
2799 }
2800 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2801
2802 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2803                                 struct pid *pid, struct task_struct *task)
2804 {
2805         int err = lock_trace(task);
2806         if (!err) {
2807                 seq_printf(m, "%08x\n", task->personality);
2808                 unlock_trace(task);
2809         }
2810         return err;
2811 }
2812
2813 /*
2814  * Thread groups
2815  */
2816 static const struct file_operations proc_task_operations;
2817 static const struct inode_operations proc_task_inode_operations;
2818
2819 static const struct pid_entry tgid_base_stuff[] = {
2820         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2821         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2822         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2823 #ifdef CONFIG_NET
2824         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2825 #endif
2826         REG("environ",    S_IRUSR, proc_environ_operations),
2827         INF("auxv",       S_IRUSR, proc_pid_auxv),
2828         ONE("status",     S_IRUGO, proc_pid_status),
2829         ONE("personality", S_IRUGO, proc_pid_personality),
2830         INF("limits",     S_IRUGO, proc_pid_limits),
2831 #ifdef CONFIG_SCHED_DEBUG
2832         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2833 #endif
2834 #ifdef CONFIG_SCHED_AUTOGROUP
2835         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2836 #endif
2837         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2838 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2839         INF("syscall",    S_IRUGO, proc_pid_syscall),
2840 #endif
2841         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2842         ONE("stat",       S_IRUGO, proc_tgid_stat),
2843         ONE("statm",      S_IRUGO, proc_pid_statm),
2844         REG("maps",       S_IRUGO, proc_maps_operations),
2845 #ifdef CONFIG_NUMA
2846         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2847 #endif
2848         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2849         LNK("cwd",        proc_cwd_link),
2850         LNK("root",       proc_root_link),
2851         LNK("exe",        proc_exe_link),
2852         REG("mounts",     S_IRUGO, proc_mounts_operations),
2853         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2854         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2855 #ifdef CONFIG_PROC_PAGE_MONITOR
2856         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2857         REG("smaps",      S_IRUGO, proc_smaps_operations),
2858         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2859 #endif
2860 #ifdef CONFIG_SECURITY
2861         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2862 #endif
2863 #ifdef CONFIG_KALLSYMS
2864         INF("wchan",      S_IRUGO, proc_pid_wchan),
2865 #endif
2866 #ifdef CONFIG_STACKTRACE
2867         ONE("stack",      S_IRUGO, proc_pid_stack),
2868 #endif
2869 #ifdef CONFIG_SCHEDSTATS
2870         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2871 #endif
2872 #ifdef CONFIG_LATENCYTOP
2873         REG("latency",  S_IRUGO, proc_lstats_operations),
2874 #endif
2875 #ifdef CONFIG_PROC_PID_CPUSET
2876         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2877 #endif
2878 #ifdef CONFIG_CGROUPS
2879         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2880 #endif
2881         INF("oom_score",  S_IRUGO, proc_oom_score),
2882         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2883         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2884 #ifdef CONFIG_AUDITSYSCALL
2885         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2886         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2887 #endif
2888 #ifdef CONFIG_FAULT_INJECTION
2889         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2890 #endif
2891 #ifdef CONFIG_ELF_CORE
2892         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2893 #endif
2894 #ifdef CONFIG_TASK_IO_ACCOUNTING
2895         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2896 #endif
2897 };
2898
2899 static int proc_tgid_base_readdir(struct file * filp,
2900                              void * dirent, filldir_t filldir)
2901 {
2902         return proc_pident_readdir(filp,dirent,filldir,
2903                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2904 }
2905
2906 static const struct file_operations proc_tgid_base_operations = {
2907         .read           = generic_read_dir,
2908         .readdir        = proc_tgid_base_readdir,
2909         .llseek         = default_llseek,
2910 };
2911
2912 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2913         return proc_pident_lookup(dir, dentry,
2914                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2915 }
2916
2917 static const struct inode_operations proc_tgid_base_inode_operations = {
2918         .lookup         = proc_tgid_base_lookup,
2919         .getattr        = pid_getattr,
2920         .setattr        = proc_setattr,
2921 };
2922
2923 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2924 {
2925         struct dentry *dentry, *leader, *dir;
2926         char buf[PROC_NUMBUF];
2927         struct qstr name;
2928
2929         name.name = buf;
2930         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2931         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2932         if (dentry) {
2933                 shrink_dcache_parent(dentry);
2934                 d_drop(dentry);
2935                 dput(dentry);
2936         }
2937
2938         name.name = buf;
2939         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2940         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2941         if (!leader)
2942                 goto out;
2943
2944         name.name = "task";
2945         name.len = strlen(name.name);
2946         dir = d_hash_and_lookup(leader, &name);
2947         if (!dir)
2948                 goto out_put_leader;
2949
2950         name.name = buf;
2951         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2952         dentry = d_hash_and_lookup(dir, &name);
2953         if (dentry) {
2954                 shrink_dcache_parent(dentry);
2955                 d_drop(dentry);
2956                 dput(dentry);
2957         }
2958
2959         dput(dir);
2960 out_put_leader:
2961         dput(leader);
2962 out:
2963         return;
2964 }
2965
2966 /**
2967  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2968  * @task: task that should be flushed.
2969  *
2970  * When flushing dentries from proc, one needs to flush them from global
2971  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2972  * in. This call is supposed to do all of this job.
2973  *
2974  * Looks in the dcache for
2975  * /proc/@pid
2976  * /proc/@tgid/task/@pid
2977  * if either directory is present flushes it and all of it'ts children
2978  * from the dcache.
2979  *
2980  * It is safe and reasonable to cache /proc entries for a task until
2981  * that task exits.  After that they just clog up the dcache with
2982  * useless entries, possibly causing useful dcache entries to be
2983  * flushed instead.  This routine is proved to flush those useless
2984  * dcache entries at process exit time.
2985  *
2986  * NOTE: This routine is just an optimization so it does not guarantee
2987  *       that no dcache entries will exist at process exit time it
2988  *       just makes it very unlikely that any will persist.
2989  */
2990
2991 void proc_flush_task(struct task_struct *task)
2992 {
2993         int i;
2994         struct pid *pid, *tgid;
2995         struct upid *upid;
2996
2997         pid = task_pid(task);
2998         tgid = task_tgid(task);
2999
3000         for (i = 0; i <= pid->level; i++) {
3001                 upid = &pid->numbers[i];
3002                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3003                                         tgid->numbers[i].nr);
3004         }
3005
3006         upid = &pid->numbers[pid->level];
3007         if (upid->nr == 1)
3008                 pid_ns_release_proc(upid->ns);
3009 }
3010
3011 static struct dentry *proc_pid_instantiate(struct inode *dir,
3012                                            struct dentry * dentry,
3013                                            struct task_struct *task, const void *ptr)
3014 {
3015         struct dentry *error = ERR_PTR(-ENOENT);
3016         struct inode *inode;
3017
3018         inode = proc_pid_make_inode(dir->i_sb, task);
3019         if (!inode)
3020                 goto out;
3021
3022         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3023         inode->i_op = &proc_tgid_base_inode_operations;
3024         inode->i_fop = &proc_tgid_base_operations;
3025         inode->i_flags|=S_IMMUTABLE;
3026
3027         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
3028                 ARRAY_SIZE(tgid_base_stuff));
3029
3030         d_set_d_op(dentry, &pid_dentry_operations);
3031
3032         d_add(dentry, inode);
3033         /* Close the race of the process dying before we return the dentry */
3034         if (pid_revalidate(dentry, NULL))
3035                 error = NULL;
3036 out:
3037         return error;
3038 }
3039
3040 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3041 {
3042         struct dentry *result;
3043         struct task_struct *task;
3044         unsigned tgid;
3045         struct pid_namespace *ns;
3046
3047         result = proc_base_lookup(dir, dentry);
3048         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3049                 goto out;
3050
3051         tgid = name_to_int(dentry);
3052         if (tgid == ~0U)
3053                 goto out;
3054
3055         ns = dentry->d_sb->s_fs_info;
3056         rcu_read_lock();
3057         task = find_task_by_pid_ns(tgid, ns);
3058         if (task)
3059                 get_task_struct(task);
3060         rcu_read_unlock();
3061         if (!task)
3062                 goto out;
3063
3064         result = proc_pid_instantiate(dir, dentry, task, NULL);
3065         put_task_struct(task);
3066 out:
3067         return result;
3068 }
3069
3070 /*
3071  * Find the first task with tgid >= tgid
3072  *
3073  */
3074 struct tgid_iter {
3075         unsigned int tgid;
3076         struct task_struct *task;
3077 };
3078 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3079 {
3080         struct pid *pid;
3081
3082         if (iter.task)
3083                 put_task_struct(iter.task);
3084         rcu_read_lock();
3085 retry:
3086         iter.task = NULL;
3087         pid = find_ge_pid(iter.tgid, ns);
3088         if (pid) {
3089                 iter.tgid = pid_nr_ns(pid, ns);
3090                 iter.task = pid_task(pid, PIDTYPE_PID);
3091                 /* What we to know is if the pid we have find is the
3092                  * pid of a thread_group_leader.  Testing for task
3093                  * being a thread_group_leader is the obvious thing
3094                  * todo but there is a window when it fails, due to
3095                  * the pid transfer logic in de_thread.
3096                  *
3097                  * So we perform the straight forward test of seeing
3098                  * if the pid we have found is the pid of a thread
3099                  * group leader, and don't worry if the task we have
3100                  * found doesn't happen to be a thread group leader.
3101                  * As we don't care in the case of readdir.
3102                  */
3103                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3104                         iter.tgid += 1;
3105                         goto retry;
3106                 }
3107                 get_task_struct(iter.task);
3108         }
3109         rcu_read_unlock();
3110         return iter;
3111 }
3112
3113 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3114
3115 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3116         struct tgid_iter iter)
3117 {
3118         char name[PROC_NUMBUF];
3119         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3120         return proc_fill_cache(filp, dirent, filldir, name, len,
3121                                 proc_pid_instantiate, iter.task, NULL);
3122 }
3123
3124 /* for the /proc/ directory itself, after non-process stuff has been done */
3125 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3126 {
3127         unsigned int nr;
3128         struct task_struct *reaper;
3129         struct tgid_iter iter;
3130         struct pid_namespace *ns;
3131
3132         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3133                 goto out_no_task;
3134         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3135
3136         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3137         if (!reaper)
3138                 goto out_no_task;
3139
3140         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3141                 const struct pid_entry *p = &proc_base_stuff[nr];
3142                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3143                         goto out;
3144         }
3145
3146         ns = filp->f_dentry->d_sb->s_fs_info;
3147         iter.task = NULL;
3148         iter.tgid = filp->f_pos - TGID_OFFSET;
3149         for (iter = next_tgid(ns, iter);
3150              iter.task;
3151              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3152                 filp->f_pos = iter.tgid + TGID_OFFSET;
3153                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3154                         put_task_struct(iter.task);
3155                         goto out;
3156                 }
3157         }
3158         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3159 out:
3160         put_task_struct(reaper);
3161 out_no_task:
3162         return 0;
3163 }
3164
3165 /*
3166  * Tasks
3167  */
3168 static const struct pid_entry tid_base_stuff[] = {
3169         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3170         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3171         REG("environ",   S_IRUSR, proc_environ_operations),
3172         INF("auxv",      S_IRUSR, proc_pid_auxv),
3173         ONE("status",    S_IRUGO, proc_pid_status),
3174         ONE("personality", S_IRUGO, proc_pid_personality),
3175         INF("limits",    S_IRUGO, proc_pid_limits),
3176 #ifdef CONFIG_SCHED_DEBUG
3177         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3178 #endif
3179         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3180 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3181         INF("syscall",   S_IRUGO, proc_pid_syscall),
3182 #endif
3183         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3184         ONE("stat",      S_IRUGO, proc_tid_stat),
3185         ONE("statm",     S_IRUGO, proc_pid_statm),
3186         REG("maps",      S_IRUGO, proc_maps_operations),
3187 #ifdef CONFIG_NUMA
3188         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3189 #endif
3190         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3191         LNK("cwd",       proc_cwd_link),
3192         LNK("root",      proc_root_link),
3193         LNK("exe",       proc_exe_link),
3194         REG("mounts",    S_IRUGO, proc_mounts_operations),
3195         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3196 #ifdef CONFIG_PROC_PAGE_MONITOR
3197         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3198         REG("smaps",     S_IRUGO, proc_smaps_operations),
3199         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3200 #endif
3201 #ifdef CONFIG_SECURITY
3202         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3203 #endif
3204 #ifdef CONFIG_KALLSYMS
3205         INF("wchan",     S_IRUGO, proc_pid_wchan),
3206 #endif
3207 #ifdef CONFIG_STACKTRACE
3208         ONE("stack",      S_IRUGO, proc_pid_stack),
3209 #endif
3210 #ifdef CONFIG_SCHEDSTATS
3211         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3212 #endif
3213 #ifdef CONFIG_LATENCYTOP
3214         REG("latency",  S_IRUGO, proc_lstats_operations),
3215 #endif
3216 #ifdef CONFIG_PROC_PID_CPUSET
3217         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3218 #endif
3219 #ifdef CONFIG_CGROUPS
3220         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3221 #endif
3222         INF("oom_score", S_IRUGO, proc_oom_score),
3223         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3224         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3225 #ifdef CONFIG_AUDITSYSCALL
3226         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3227         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3228 #endif
3229 #ifdef CONFIG_FAULT_INJECTION
3230         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3231 #endif
3232 #ifdef CONFIG_TASK_IO_ACCOUNTING
3233         INF("io",       S_IRUGO, proc_tid_io_accounting),
3234 #endif
3235 };
3236
3237 static int proc_tid_base_readdir(struct file * filp,
3238                              void * dirent, filldir_t filldir)
3239 {
3240         return proc_pident_readdir(filp,dirent,filldir,
3241                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3242 }
3243
3244 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3245         return proc_pident_lookup(dir, dentry,
3246                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3247 }
3248
3249 static const struct file_operations proc_tid_base_operations = {
3250         .read           = generic_read_dir,
3251         .readdir        = proc_tid_base_readdir,
3252         .llseek         = default_llseek,
3253 };
3254
3255 static const struct inode_operations proc_tid_base_inode_operations = {
3256         .lookup         = proc_tid_base_lookup,
3257         .getattr        = pid_getattr,
3258         .setattr        = proc_setattr,
3259 };
3260
3261 static struct dentry *proc_task_instantiate(struct inode *dir,
3262         struct dentry *dentry, struct task_struct *task, const void *ptr)
3263 {
3264         struct dentry *error = ERR_PTR(-ENOENT);
3265         struct inode *inode;
3266         inode = proc_pid_make_inode(dir->i_sb, task);
3267
3268         if (!inode)
3269                 goto out;
3270         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3271         inode->i_op = &proc_tid_base_inode_operations;
3272         inode->i_fop = &proc_tid_base_operations;
3273         inode->i_flags|=S_IMMUTABLE;
3274
3275         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3276                 ARRAY_SIZE(tid_base_stuff));
3277
3278         d_set_d_op(dentry, &pid_dentry_operations);
3279
3280         d_add(dentry, inode);
3281         /* Close the race of the process dying before we return the dentry */
3282         if (pid_revalidate(dentry, NULL))
3283                 error = NULL;
3284 out:
3285         return error;
3286 }
3287
3288 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3289 {
3290         struct dentry *result = ERR_PTR(-ENOENT);
3291         struct task_struct *task;
3292         struct task_struct *leader = get_proc_task(dir);
3293         unsigned tid;
3294         struct pid_namespace *ns;
3295
3296         if (!leader)
3297                 goto out_no_task;
3298
3299         tid = name_to_int(dentry);
3300         if (tid == ~0U)
3301                 goto out;
3302
3303         ns = dentry->d_sb->s_fs_info;
3304         rcu_read_lock();
3305         task = find_task_by_pid_ns(tid, ns);
3306         if (task)
3307                 get_task_struct(task);
3308         rcu_read_unlock();
3309         if (!task)
3310                 goto out;
3311         if (!same_thread_group(leader, task))
3312                 goto out_drop_task;
3313
3314         result = proc_task_instantiate(dir, dentry, task, NULL);
3315 out_drop_task:
3316         put_task_struct(task);
3317 out:
3318         put_task_struct(leader);
3319 out_no_task:
3320         return result;
3321 }
3322
3323 /*
3324  * Find the first tid of a thread group to return to user space.
3325  *
3326  * Usually this is just the thread group leader, but if the users
3327  * buffer was too small or there was a seek into the middle of the
3328  * directory we have more work todo.
3329  *
3330  * In the case of a short read we start with find_task_by_pid.
3331  *
3332  * In the case of a seek we start with the leader and walk nr
3333  * threads past it.
3334  */
3335 static struct task_struct *first_tid(struct task_struct *leader,
3336                 int tid, int nr, struct pid_namespace *ns)
3337 {
3338         struct task_struct *pos;
3339
3340         rcu_read_lock();
3341         /* Attempt to start with the pid of a thread */
3342         if (tid && (nr > 0)) {
3343                 pos = find_task_by_pid_ns(tid, ns);
3344                 if (pos && (pos->group_leader == leader))
3345                         goto found;
3346         }
3347
3348         /* If nr exceeds the number of threads there is nothing todo */
3349         pos = NULL;
3350         if (nr && nr >= get_nr_threads(leader))
3351                 goto out;
3352
3353         /* If we haven't found our starting place yet start
3354          * with the leader and walk nr threads forward.
3355          */
3356         for (pos = leader; nr > 0; --nr) {
3357                 pos = next_thread(pos);
3358                 if (pos == leader) {
3359                         pos = NULL;
3360                         goto out;
3361                 }
3362         }
3363 found:
3364         get_task_struct(pos);
3365 out:
3366         rcu_read_unlock();
3367         return pos;
3368 }
3369
3370 /*
3371  * Find the next thread in the thread list.
3372  * Return NULL if there is an error or no next thread.
3373  *
3374  * The reference to the input task_struct is released.
3375  */
3376 static struct task_struct *next_tid(struct task_struct *start)
3377 {
3378         struct task_struct *pos = NULL;
3379         rcu_read_lock();
3380         if (pid_alive(start)) {
3381                 pos = next_thread(start);
3382                 if (thread_group_leader(pos))
3383                         pos = NULL;
3384                 else
3385                         get_task_struct(pos);
3386         }
3387         rcu_read_unlock();
3388         put_task_struct(start);
3389         return pos;
3390 }
3391
3392 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3393         struct task_struct *task, int tid)
3394 {
3395         char name[PROC_NUMBUF];
3396         int len = snprintf(name, sizeof(name), "%d", tid);
3397         return proc_fill_cache(filp, dirent, filldir, name, len,
3398                                 proc_task_instantiate, task, NULL);
3399 }
3400
3401 /* for the /proc/TGID/task/ directories */
3402 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3403 {
3404         struct dentry *dentry = filp->f_path.dentry;
3405         struct inode *inode = dentry->d_inode;
3406         struct task_struct *leader = NULL;
3407         struct task_struct *task;
3408         int retval = -ENOENT;
3409         ino_t ino;
3410         int tid;
3411         struct pid_namespace *ns;
3412
3413         task = get_proc_task(inode);
3414         if (!task)
3415                 goto out_no_task;
3416         rcu_read_lock();
3417         if (pid_alive(task)) {
3418                 leader = task->group_leader;
3419                 get_task_struct(leader);
3420         }
3421         rcu_read_unlock();
3422         put_task_struct(task);
3423         if (!leader)
3424                 goto out_no_task;
3425         retval = 0;
3426
3427         switch ((unsigned long)filp->f_pos) {
3428         case 0:
3429                 ino = inode->i_ino;
3430                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3431                         goto out;
3432                 filp->f_pos++;
3433                 /* fall through */
3434         case 1:
3435                 ino = parent_ino(dentry);
3436                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3437                         goto out;
3438                 filp->f_pos++;
3439                 /* fall through */
3440         }
3441
3442         /* f_version caches the tgid value that the last readdir call couldn't
3443          * return. lseek aka telldir automagically resets f_version to 0.
3444          */
3445         ns = filp->f_dentry->d_sb->s_fs_info;
3446         tid = (int)filp->f_version;
3447         filp->f_version = 0;
3448         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3449              task;
3450              task = next_tid(task), filp->f_pos++) {
3451                 tid = task_pid_nr_ns(task, ns);
3452                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3453                         /* returning this tgid failed, save it as the first
3454                          * pid for the next readir call */
3455                         filp->f_version = (u64)tid;
3456                         put_task_struct(task);
3457                         break;
3458                 }
3459         }
3460 out:
3461         put_task_struct(leader);
3462 out_no_task:
3463         return retval;
3464 }
3465
3466 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3467 {
3468         struct inode *inode = dentry->d_inode;
3469         struct task_struct *p = get_proc_task(inode);
3470         generic_fillattr(inode, stat);
3471
3472         if (p) {
3473                 stat->nlink += get_nr_threads(p);
3474                 put_task_struct(p);
3475         }
3476
3477         return 0;
3478 }
3479
3480 static const struct inode_operations proc_task_inode_operations = {
3481         .lookup         = proc_task_lookup,
3482         .getattr        = proc_task_getattr,
3483         .setattr        = proc_setattr,
3484 };
3485
3486 static const struct file_operations proc_task_operations = {
3487         .read           = generic_read_dir,
3488         .readdir        = proc_task_readdir,
3489         .llseek         = default_llseek,
3490 };