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