maps4: move clear_refs code to task_mmu.c
[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/init.h>
57 #include <linux/capability.h>
58 #include <linux/file.h>
59 #include <linux/string.h>
60 #include <linux/seq_file.h>
61 #include <linux/namei.h>
62 #include <linux/mnt_namespace.h>
63 #include <linux/mm.h>
64 #include <linux/rcupdate.h>
65 #include <linux/kallsyms.h>
66 #include <linux/resource.h>
67 #include <linux/module.h>
68 #include <linux/mount.h>
69 #include <linux/security.h>
70 #include <linux/ptrace.h>
71 #include <linux/cgroup.h>
72 #include <linux/cpuset.h>
73 #include <linux/audit.h>
74 #include <linux/poll.h>
75 #include <linux/nsproxy.h>
76 #include <linux/oom.h>
77 #include <linux/elf.h>
78 #include <linux/pid_namespace.h>
79 #include "internal.h"
80
81 /* NOTE:
82  *      Implementing inode permission operations in /proc is almost
83  *      certainly an error.  Permission checks need to happen during
84  *      each system call not at open time.  The reason is that most of
85  *      what we wish to check for permissions in /proc varies at runtime.
86  *
87  *      The classic example of a problem is opening file descriptors
88  *      in /proc for a task before it execs a suid executable.
89  */
90
91 struct pid_entry {
92         char *name;
93         int len;
94         mode_t mode;
95         const struct inode_operations *iop;
96         const struct file_operations *fop;
97         union proc_op op;
98 };
99
100 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
101         .name = (NAME),                                 \
102         .len  = sizeof(NAME) - 1,                       \
103         .mode = MODE,                                   \
104         .iop  = IOP,                                    \
105         .fop  = FOP,                                    \
106         .op   = OP,                                     \
107 }
108
109 #define DIR(NAME, MODE, OTYPE)                                                  \
110         NOD(NAME, (S_IFDIR|(MODE)),                                             \
111                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
112                 {} )
113 #define LNK(NAME, OTYPE)                                        \
114         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
115                 &proc_pid_link_inode_operations, NULL,          \
116                 { .proc_get_link = &proc_##OTYPE##_link } )
117 #define REG(NAME, MODE, OTYPE)                          \
118         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
119                 &proc_##OTYPE##_operations, {})
120 #define INF(NAME, MODE, OTYPE)                          \
121         NOD(NAME, (S_IFREG|(MODE)),                     \
122                 NULL, &proc_info_file_operations,       \
123                 { .proc_read = &proc_##OTYPE } )
124
125 int maps_protect;
126 EXPORT_SYMBOL(maps_protect);
127
128 static struct fs_struct *get_fs_struct(struct task_struct *task)
129 {
130         struct fs_struct *fs;
131         task_lock(task);
132         fs = task->fs;
133         if(fs)
134                 atomic_inc(&fs->count);
135         task_unlock(task);
136         return fs;
137 }
138
139 static int get_nr_threads(struct task_struct *tsk)
140 {
141         /* Must be called with the rcu_read_lock held */
142         unsigned long flags;
143         int count = 0;
144
145         if (lock_task_sighand(tsk, &flags)) {
146                 count = atomic_read(&tsk->signal->count);
147                 unlock_task_sighand(tsk, &flags);
148         }
149         return count;
150 }
151
152 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
153 {
154         struct task_struct *task = get_proc_task(inode);
155         struct fs_struct *fs = NULL;
156         int result = -ENOENT;
157
158         if (task) {
159                 fs = get_fs_struct(task);
160                 put_task_struct(task);
161         }
162         if (fs) {
163                 read_lock(&fs->lock);
164                 *mnt = mntget(fs->pwdmnt);
165                 *dentry = dget(fs->pwd);
166                 read_unlock(&fs->lock);
167                 result = 0;
168                 put_fs_struct(fs);
169         }
170         return result;
171 }
172
173 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
174 {
175         struct task_struct *task = get_proc_task(inode);
176         struct fs_struct *fs = NULL;
177         int result = -ENOENT;
178
179         if (task) {
180                 fs = get_fs_struct(task);
181                 put_task_struct(task);
182         }
183         if (fs) {
184                 read_lock(&fs->lock);
185                 *mnt = mntget(fs->rootmnt);
186                 *dentry = dget(fs->root);
187                 read_unlock(&fs->lock);
188                 result = 0;
189                 put_fs_struct(fs);
190         }
191         return result;
192 }
193
194 #define MAY_PTRACE(task) \
195         (task == current || \
196         (task->parent == current && \
197         (task->ptrace & PT_PTRACED) && \
198          (task_is_stopped_or_traced(task)) && \
199          security_ptrace(current,task) == 0))
200
201 struct mm_struct *mm_for_maps(struct task_struct *task)
202 {
203         struct mm_struct *mm = get_task_mm(task);
204         if (!mm)
205                 return NULL;
206         down_read(&mm->mmap_sem);
207         task_lock(task);
208         if (task->mm != mm)
209                 goto out;
210         if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
211                 goto out;
212         task_unlock(task);
213         return mm;
214 out:
215         task_unlock(task);
216         up_read(&mm->mmap_sem);
217         mmput(mm);
218         return NULL;
219 }
220
221 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
222 {
223         int res = 0;
224         unsigned int len;
225         struct mm_struct *mm = get_task_mm(task);
226         if (!mm)
227                 goto out;
228         if (!mm->arg_end)
229                 goto out_mm;    /* Shh! No looking before we're done */
230
231         len = mm->arg_end - mm->arg_start;
232  
233         if (len > PAGE_SIZE)
234                 len = PAGE_SIZE;
235  
236         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
237
238         // If the nul at the end of args has been overwritten, then
239         // assume application is using setproctitle(3).
240         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
241                 len = strnlen(buffer, res);
242                 if (len < res) {
243                     res = len;
244                 } else {
245                         len = mm->env_end - mm->env_start;
246                         if (len > PAGE_SIZE - res)
247                                 len = PAGE_SIZE - res;
248                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
249                         res = strnlen(buffer, res);
250                 }
251         }
252 out_mm:
253         mmput(mm);
254 out:
255         return res;
256 }
257
258 static int proc_pid_auxv(struct task_struct *task, char *buffer)
259 {
260         int res = 0;
261         struct mm_struct *mm = get_task_mm(task);
262         if (mm) {
263                 unsigned int nwords = 0;
264                 do
265                         nwords += 2;
266                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
267                 res = nwords * sizeof(mm->saved_auxv[0]);
268                 if (res > PAGE_SIZE)
269                         res = PAGE_SIZE;
270                 memcpy(buffer, mm->saved_auxv, res);
271                 mmput(mm);
272         }
273         return res;
274 }
275
276
277 #ifdef CONFIG_KALLSYMS
278 /*
279  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
280  * Returns the resolved symbol.  If that fails, simply return the address.
281  */
282 static int proc_pid_wchan(struct task_struct *task, char *buffer)
283 {
284         unsigned long wchan;
285         char symname[KSYM_NAME_LEN];
286
287         wchan = get_wchan(task);
288
289         if (lookup_symbol_name(wchan, symname) < 0)
290                 return sprintf(buffer, "%lu", wchan);
291         else
292                 return sprintf(buffer, "%s", symname);
293 }
294 #endif /* CONFIG_KALLSYMS */
295
296 #ifdef CONFIG_SCHEDSTATS
297 /*
298  * Provides /proc/PID/schedstat
299  */
300 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
301 {
302         return sprintf(buffer, "%llu %llu %lu\n",
303                         task->sched_info.cpu_time,
304                         task->sched_info.run_delay,
305                         task->sched_info.pcount);
306 }
307 #endif
308
309 #ifdef CONFIG_LATENCYTOP
310 static int lstats_show_proc(struct seq_file *m, void *v)
311 {
312         int i;
313         struct task_struct *task = m->private;
314         seq_puts(m, "Latency Top version : v0.1\n");
315
316         for (i = 0; i < 32; i++) {
317                 if (task->latency_record[i].backtrace[0]) {
318                         int q;
319                         seq_printf(m, "%i %li %li ",
320                                 task->latency_record[i].count,
321                                 task->latency_record[i].time,
322                                 task->latency_record[i].max);
323                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
324                                 char sym[KSYM_NAME_LEN];
325                                 char *c;
326                                 if (!task->latency_record[i].backtrace[q])
327                                         break;
328                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
329                                         break;
330                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
331                                 c = strchr(sym, '+');
332                                 if (c)
333                                         *c = 0;
334                                 seq_printf(m, "%s ", sym);
335                         }
336                         seq_printf(m, "\n");
337                 }
338
339         }
340         return 0;
341 }
342
343 static int lstats_open(struct inode *inode, struct file *file)
344 {
345         int ret;
346         struct seq_file *m;
347         struct task_struct *task = get_proc_task(inode);
348
349         ret = single_open(file, lstats_show_proc, NULL);
350         if (!ret) {
351                 m = file->private_data;
352                 m->private = task;
353         }
354         return ret;
355 }
356
357 static ssize_t lstats_write(struct file *file, const char __user *buf,
358                             size_t count, loff_t *offs)
359 {
360         struct seq_file *m;
361         struct task_struct *task;
362
363         m = file->private_data;
364         task = m->private;
365         clear_all_latency_tracing(task);
366
367         return count;
368 }
369
370 static const struct file_operations proc_lstats_operations = {
371         .open           = lstats_open,
372         .read           = seq_read,
373         .write          = lstats_write,
374         .llseek         = seq_lseek,
375         .release        = single_release,
376 };
377
378 #endif
379
380 /* The badness from the OOM killer */
381 unsigned long badness(struct task_struct *p, unsigned long uptime);
382 static int proc_oom_score(struct task_struct *task, char *buffer)
383 {
384         unsigned long points;
385         struct timespec uptime;
386
387         do_posix_clock_monotonic_gettime(&uptime);
388         read_lock(&tasklist_lock);
389         points = badness(task, uptime.tv_sec);
390         read_unlock(&tasklist_lock);
391         return sprintf(buffer, "%lu\n", points);
392 }
393
394 struct limit_names {
395         char *name;
396         char *unit;
397 };
398
399 static const struct limit_names lnames[RLIM_NLIMITS] = {
400         [RLIMIT_CPU] = {"Max cpu time", "ms"},
401         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
402         [RLIMIT_DATA] = {"Max data size", "bytes"},
403         [RLIMIT_STACK] = {"Max stack size", "bytes"},
404         [RLIMIT_CORE] = {"Max core file size", "bytes"},
405         [RLIMIT_RSS] = {"Max resident set", "bytes"},
406         [RLIMIT_NPROC] = {"Max processes", "processes"},
407         [RLIMIT_NOFILE] = {"Max open files", "files"},
408         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
409         [RLIMIT_AS] = {"Max address space", "bytes"},
410         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
411         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
412         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
413         [RLIMIT_NICE] = {"Max nice priority", NULL},
414         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
415 };
416
417 /* Display limits for a process */
418 static int proc_pid_limits(struct task_struct *task, char *buffer)
419 {
420         unsigned int i;
421         int count = 0;
422         unsigned long flags;
423         char *bufptr = buffer;
424
425         struct rlimit rlim[RLIM_NLIMITS];
426
427         rcu_read_lock();
428         if (!lock_task_sighand(task,&flags)) {
429                 rcu_read_unlock();
430                 return 0;
431         }
432         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
433         unlock_task_sighand(task, &flags);
434         rcu_read_unlock();
435
436         /*
437          * print the file header
438          */
439         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
440                         "Limit", "Soft Limit", "Hard Limit", "Units");
441
442         for (i = 0; i < RLIM_NLIMITS; i++) {
443                 if (rlim[i].rlim_cur == RLIM_INFINITY)
444                         count += sprintf(&bufptr[count], "%-25s %-20s ",
445                                          lnames[i].name, "unlimited");
446                 else
447                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
448                                          lnames[i].name, rlim[i].rlim_cur);
449
450                 if (rlim[i].rlim_max == RLIM_INFINITY)
451                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
452                 else
453                         count += sprintf(&bufptr[count], "%-20lu ",
454                                          rlim[i].rlim_max);
455
456                 if (lnames[i].unit)
457                         count += sprintf(&bufptr[count], "%-10s\n",
458                                          lnames[i].unit);
459                 else
460                         count += sprintf(&bufptr[count], "\n");
461         }
462
463         return count;
464 }
465
466 /************************************************************************/
467 /*                       Here the fs part begins                        */
468 /************************************************************************/
469
470 /* permission checks */
471 static int proc_fd_access_allowed(struct inode *inode)
472 {
473         struct task_struct *task;
474         int allowed = 0;
475         /* Allow access to a task's file descriptors if it is us or we
476          * may use ptrace attach to the process and find out that
477          * information.
478          */
479         task = get_proc_task(inode);
480         if (task) {
481                 allowed = ptrace_may_attach(task);
482                 put_task_struct(task);
483         }
484         return allowed;
485 }
486
487 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
488 {
489         int error;
490         struct inode *inode = dentry->d_inode;
491
492         if (attr->ia_valid & ATTR_MODE)
493                 return -EPERM;
494
495         error = inode_change_ok(inode, attr);
496         if (!error)
497                 error = inode_setattr(inode, attr);
498         return error;
499 }
500
501 static const struct inode_operations proc_def_inode_operations = {
502         .setattr        = proc_setattr,
503 };
504
505 extern struct seq_operations mounts_op;
506 struct proc_mounts {
507         struct seq_file m;
508         int event;
509 };
510
511 static int mounts_open(struct inode *inode, struct file *file)
512 {
513         struct task_struct *task = get_proc_task(inode);
514         struct nsproxy *nsp;
515         struct mnt_namespace *ns = NULL;
516         struct proc_mounts *p;
517         int ret = -EINVAL;
518
519         if (task) {
520                 rcu_read_lock();
521                 nsp = task_nsproxy(task);
522                 if (nsp) {
523                         ns = nsp->mnt_ns;
524                         if (ns)
525                                 get_mnt_ns(ns);
526                 }
527                 rcu_read_unlock();
528
529                 put_task_struct(task);
530         }
531
532         if (ns) {
533                 ret = -ENOMEM;
534                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
535                 if (p) {
536                         file->private_data = &p->m;
537                         ret = seq_open(file, &mounts_op);
538                         if (!ret) {
539                                 p->m.private = ns;
540                                 p->event = ns->event;
541                                 return 0;
542                         }
543                         kfree(p);
544                 }
545                 put_mnt_ns(ns);
546         }
547         return ret;
548 }
549
550 static int mounts_release(struct inode *inode, struct file *file)
551 {
552         struct seq_file *m = file->private_data;
553         struct mnt_namespace *ns = m->private;
554         put_mnt_ns(ns);
555         return seq_release(inode, file);
556 }
557
558 static unsigned mounts_poll(struct file *file, poll_table *wait)
559 {
560         struct proc_mounts *p = file->private_data;
561         struct mnt_namespace *ns = p->m.private;
562         unsigned res = 0;
563
564         poll_wait(file, &ns->poll, wait);
565
566         spin_lock(&vfsmount_lock);
567         if (p->event != ns->event) {
568                 p->event = ns->event;
569                 res = POLLERR;
570         }
571         spin_unlock(&vfsmount_lock);
572
573         return res;
574 }
575
576 static const struct file_operations proc_mounts_operations = {
577         .open           = mounts_open,
578         .read           = seq_read,
579         .llseek         = seq_lseek,
580         .release        = mounts_release,
581         .poll           = mounts_poll,
582 };
583
584 extern struct seq_operations mountstats_op;
585 static int mountstats_open(struct inode *inode, struct file *file)
586 {
587         int ret = seq_open(file, &mountstats_op);
588
589         if (!ret) {
590                 struct seq_file *m = file->private_data;
591                 struct nsproxy *nsp;
592                 struct mnt_namespace *mnt_ns = NULL;
593                 struct task_struct *task = get_proc_task(inode);
594
595                 if (task) {
596                         rcu_read_lock();
597                         nsp = task_nsproxy(task);
598                         if (nsp) {
599                                 mnt_ns = nsp->mnt_ns;
600                                 if (mnt_ns)
601                                         get_mnt_ns(mnt_ns);
602                         }
603                         rcu_read_unlock();
604
605                         put_task_struct(task);
606                 }
607
608                 if (mnt_ns)
609                         m->private = mnt_ns;
610                 else {
611                         seq_release(inode, file);
612                         ret = -EINVAL;
613                 }
614         }
615         return ret;
616 }
617
618 static const struct file_operations proc_mountstats_operations = {
619         .open           = mountstats_open,
620         .read           = seq_read,
621         .llseek         = seq_lseek,
622         .release        = mounts_release,
623 };
624
625 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
626
627 static ssize_t proc_info_read(struct file * file, char __user * buf,
628                           size_t count, loff_t *ppos)
629 {
630         struct inode * inode = file->f_path.dentry->d_inode;
631         unsigned long page;
632         ssize_t length;
633         struct task_struct *task = get_proc_task(inode);
634
635         length = -ESRCH;
636         if (!task)
637                 goto out_no_task;
638
639         if (count > PROC_BLOCK_SIZE)
640                 count = PROC_BLOCK_SIZE;
641
642         length = -ENOMEM;
643         if (!(page = __get_free_page(GFP_TEMPORARY)))
644                 goto out;
645
646         length = PROC_I(inode)->op.proc_read(task, (char*)page);
647
648         if (length >= 0)
649                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
650         free_page(page);
651 out:
652         put_task_struct(task);
653 out_no_task:
654         return length;
655 }
656
657 static const struct file_operations proc_info_file_operations = {
658         .read           = proc_info_read,
659 };
660
661 static int mem_open(struct inode* inode, struct file* file)
662 {
663         file->private_data = (void*)((long)current->self_exec_id);
664         return 0;
665 }
666
667 static ssize_t mem_read(struct file * file, char __user * buf,
668                         size_t count, loff_t *ppos)
669 {
670         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
671         char *page;
672         unsigned long src = *ppos;
673         int ret = -ESRCH;
674         struct mm_struct *mm;
675
676         if (!task)
677                 goto out_no_task;
678
679         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
680                 goto out;
681
682         ret = -ENOMEM;
683         page = (char *)__get_free_page(GFP_TEMPORARY);
684         if (!page)
685                 goto out;
686
687         ret = 0;
688  
689         mm = get_task_mm(task);
690         if (!mm)
691                 goto out_free;
692
693         ret = -EIO;
694  
695         if (file->private_data != (void*)((long)current->self_exec_id))
696                 goto out_put;
697
698         ret = 0;
699  
700         while (count > 0) {
701                 int this_len, retval;
702
703                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
704                 retval = access_process_vm(task, src, page, this_len, 0);
705                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
706                         if (!ret)
707                                 ret = -EIO;
708                         break;
709                 }
710
711                 if (copy_to_user(buf, page, retval)) {
712                         ret = -EFAULT;
713                         break;
714                 }
715  
716                 ret += retval;
717                 src += retval;
718                 buf += retval;
719                 count -= retval;
720         }
721         *ppos = src;
722
723 out_put:
724         mmput(mm);
725 out_free:
726         free_page((unsigned long) page);
727 out:
728         put_task_struct(task);
729 out_no_task:
730         return ret;
731 }
732
733 #define mem_write NULL
734
735 #ifndef mem_write
736 /* This is a security hazard */
737 static ssize_t mem_write(struct file * file, const char __user *buf,
738                          size_t count, loff_t *ppos)
739 {
740         int copied;
741         char *page;
742         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
743         unsigned long dst = *ppos;
744
745         copied = -ESRCH;
746         if (!task)
747                 goto out_no_task;
748
749         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
750                 goto out;
751
752         copied = -ENOMEM;
753         page = (char *)__get_free_page(GFP_TEMPORARY);
754         if (!page)
755                 goto out;
756
757         copied = 0;
758         while (count > 0) {
759                 int this_len, retval;
760
761                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
762                 if (copy_from_user(page, buf, this_len)) {
763                         copied = -EFAULT;
764                         break;
765                 }
766                 retval = access_process_vm(task, dst, page, this_len, 1);
767                 if (!retval) {
768                         if (!copied)
769                                 copied = -EIO;
770                         break;
771                 }
772                 copied += retval;
773                 buf += retval;
774                 dst += retval;
775                 count -= retval;                        
776         }
777         *ppos = dst;
778         free_page((unsigned long) page);
779 out:
780         put_task_struct(task);
781 out_no_task:
782         return copied;
783 }
784 #endif
785
786 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
787 {
788         switch (orig) {
789         case 0:
790                 file->f_pos = offset;
791                 break;
792         case 1:
793                 file->f_pos += offset;
794                 break;
795         default:
796                 return -EINVAL;
797         }
798         force_successful_syscall_return();
799         return file->f_pos;
800 }
801
802 static const struct file_operations proc_mem_operations = {
803         .llseek         = mem_lseek,
804         .read           = mem_read,
805         .write          = mem_write,
806         .open           = mem_open,
807 };
808
809 static ssize_t environ_read(struct file *file, char __user *buf,
810                         size_t count, loff_t *ppos)
811 {
812         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
813         char *page;
814         unsigned long src = *ppos;
815         int ret = -ESRCH;
816         struct mm_struct *mm;
817
818         if (!task)
819                 goto out_no_task;
820
821         if (!ptrace_may_attach(task))
822                 goto out;
823
824         ret = -ENOMEM;
825         page = (char *)__get_free_page(GFP_TEMPORARY);
826         if (!page)
827                 goto out;
828
829         ret = 0;
830
831         mm = get_task_mm(task);
832         if (!mm)
833                 goto out_free;
834
835         while (count > 0) {
836                 int this_len, retval, max_len;
837
838                 this_len = mm->env_end - (mm->env_start + src);
839
840                 if (this_len <= 0)
841                         break;
842
843                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
844                 this_len = (this_len > max_len) ? max_len : this_len;
845
846                 retval = access_process_vm(task, (mm->env_start + src),
847                         page, this_len, 0);
848
849                 if (retval <= 0) {
850                         ret = retval;
851                         break;
852                 }
853
854                 if (copy_to_user(buf, page, retval)) {
855                         ret = -EFAULT;
856                         break;
857                 }
858
859                 ret += retval;
860                 src += retval;
861                 buf += retval;
862                 count -= retval;
863         }
864         *ppos = src;
865
866         mmput(mm);
867 out_free:
868         free_page((unsigned long) page);
869 out:
870         put_task_struct(task);
871 out_no_task:
872         return ret;
873 }
874
875 static const struct file_operations proc_environ_operations = {
876         .read           = environ_read,
877 };
878
879 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
880                                 size_t count, loff_t *ppos)
881 {
882         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
883         char buffer[PROC_NUMBUF];
884         size_t len;
885         int oom_adjust;
886
887         if (!task)
888                 return -ESRCH;
889         oom_adjust = task->oomkilladj;
890         put_task_struct(task);
891
892         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
893
894         return simple_read_from_buffer(buf, count, ppos, buffer, len);
895 }
896
897 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
898                                 size_t count, loff_t *ppos)
899 {
900         struct task_struct *task;
901         char buffer[PROC_NUMBUF], *end;
902         int oom_adjust;
903
904         memset(buffer, 0, sizeof(buffer));
905         if (count > sizeof(buffer) - 1)
906                 count = sizeof(buffer) - 1;
907         if (copy_from_user(buffer, buf, count))
908                 return -EFAULT;
909         oom_adjust = simple_strtol(buffer, &end, 0);
910         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
911              oom_adjust != OOM_DISABLE)
912                 return -EINVAL;
913         if (*end == '\n')
914                 end++;
915         task = get_proc_task(file->f_path.dentry->d_inode);
916         if (!task)
917                 return -ESRCH;
918         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
919                 put_task_struct(task);
920                 return -EACCES;
921         }
922         task->oomkilladj = oom_adjust;
923         put_task_struct(task);
924         if (end - buffer == 0)
925                 return -EIO;
926         return end - buffer;
927 }
928
929 static const struct file_operations proc_oom_adjust_operations = {
930         .read           = oom_adjust_read,
931         .write          = oom_adjust_write,
932 };
933
934 #ifdef CONFIG_AUDITSYSCALL
935 #define TMPBUFLEN 21
936 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
937                                   size_t count, loff_t *ppos)
938 {
939         struct inode * inode = file->f_path.dentry->d_inode;
940         struct task_struct *task = get_proc_task(inode);
941         ssize_t length;
942         char tmpbuf[TMPBUFLEN];
943
944         if (!task)
945                 return -ESRCH;
946         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
947                                 audit_get_loginuid(task));
948         put_task_struct(task);
949         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
950 }
951
952 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
953                                    size_t count, loff_t *ppos)
954 {
955         struct inode * inode = file->f_path.dentry->d_inode;
956         char *page, *tmp;
957         ssize_t length;
958         uid_t loginuid;
959
960         if (!capable(CAP_AUDIT_CONTROL))
961                 return -EPERM;
962
963         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
964                 return -EPERM;
965
966         if (count >= PAGE_SIZE)
967                 count = PAGE_SIZE - 1;
968
969         if (*ppos != 0) {
970                 /* No partial writes. */
971                 return -EINVAL;
972         }
973         page = (char*)__get_free_page(GFP_TEMPORARY);
974         if (!page)
975                 return -ENOMEM;
976         length = -EFAULT;
977         if (copy_from_user(page, buf, count))
978                 goto out_free_page;
979
980         page[count] = '\0';
981         loginuid = simple_strtoul(page, &tmp, 10);
982         if (tmp == page) {
983                 length = -EINVAL;
984                 goto out_free_page;
985
986         }
987         length = audit_set_loginuid(current, loginuid);
988         if (likely(length == 0))
989                 length = count;
990
991 out_free_page:
992         free_page((unsigned long) page);
993         return length;
994 }
995
996 static const struct file_operations proc_loginuid_operations = {
997         .read           = proc_loginuid_read,
998         .write          = proc_loginuid_write,
999 };
1000 #endif
1001
1002 #ifdef CONFIG_FAULT_INJECTION
1003 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1004                                       size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1007         char buffer[PROC_NUMBUF];
1008         size_t len;
1009         int make_it_fail;
1010
1011         if (!task)
1012                 return -ESRCH;
1013         make_it_fail = task->make_it_fail;
1014         put_task_struct(task);
1015
1016         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1017
1018         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1019 }
1020
1021 static ssize_t proc_fault_inject_write(struct file * file,
1022                         const char __user * buf, size_t count, loff_t *ppos)
1023 {
1024         struct task_struct *task;
1025         char buffer[PROC_NUMBUF], *end;
1026         int make_it_fail;
1027
1028         if (!capable(CAP_SYS_RESOURCE))
1029                 return -EPERM;
1030         memset(buffer, 0, sizeof(buffer));
1031         if (count > sizeof(buffer) - 1)
1032                 count = sizeof(buffer) - 1;
1033         if (copy_from_user(buffer, buf, count))
1034                 return -EFAULT;
1035         make_it_fail = simple_strtol(buffer, &end, 0);
1036         if (*end == '\n')
1037                 end++;
1038         task = get_proc_task(file->f_dentry->d_inode);
1039         if (!task)
1040                 return -ESRCH;
1041         task->make_it_fail = make_it_fail;
1042         put_task_struct(task);
1043         if (end - buffer == 0)
1044                 return -EIO;
1045         return end - buffer;
1046 }
1047
1048 static const struct file_operations proc_fault_inject_operations = {
1049         .read           = proc_fault_inject_read,
1050         .write          = proc_fault_inject_write,
1051 };
1052 #endif
1053
1054
1055 #ifdef CONFIG_SCHED_DEBUG
1056 /*
1057  * Print out various scheduling related per-task fields:
1058  */
1059 static int sched_show(struct seq_file *m, void *v)
1060 {
1061         struct inode *inode = m->private;
1062         struct task_struct *p;
1063
1064         WARN_ON(!inode);
1065
1066         p = get_proc_task(inode);
1067         if (!p)
1068                 return -ESRCH;
1069         proc_sched_show_task(p, m);
1070
1071         put_task_struct(p);
1072
1073         return 0;
1074 }
1075
1076 static ssize_t
1077 sched_write(struct file *file, const char __user *buf,
1078             size_t count, loff_t *offset)
1079 {
1080         struct inode *inode = file->f_path.dentry->d_inode;
1081         struct task_struct *p;
1082
1083         WARN_ON(!inode);
1084
1085         p = get_proc_task(inode);
1086         if (!p)
1087                 return -ESRCH;
1088         proc_sched_set_task(p);
1089
1090         put_task_struct(p);
1091
1092         return count;
1093 }
1094
1095 static int sched_open(struct inode *inode, struct file *filp)
1096 {
1097         int ret;
1098
1099         ret = single_open(filp, sched_show, NULL);
1100         if (!ret) {
1101                 struct seq_file *m = filp->private_data;
1102
1103                 m->private = inode;
1104         }
1105         return ret;
1106 }
1107
1108 static const struct file_operations proc_pid_sched_operations = {
1109         .open           = sched_open,
1110         .read           = seq_read,
1111         .write          = sched_write,
1112         .llseek         = seq_lseek,
1113         .release        = single_release,
1114 };
1115
1116 #endif
1117
1118 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1119 {
1120         struct inode *inode = dentry->d_inode;
1121         int error = -EACCES;
1122
1123         /* We don't need a base pointer in the /proc filesystem */
1124         path_release(nd);
1125
1126         /* Are we allowed to snoop on the tasks file descriptors? */
1127         if (!proc_fd_access_allowed(inode))
1128                 goto out;
1129
1130         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1131         nd->last_type = LAST_BIND;
1132 out:
1133         return ERR_PTR(error);
1134 }
1135
1136 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1137                             char __user *buffer, int buflen)
1138 {
1139         struct inode * inode;
1140         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1141         char *path;
1142         int len;
1143
1144         if (!tmp)
1145                 return -ENOMEM;
1146
1147         inode = dentry->d_inode;
1148         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1149         len = PTR_ERR(path);
1150         if (IS_ERR(path))
1151                 goto out;
1152         len = tmp + PAGE_SIZE - 1 - path;
1153
1154         if (len > buflen)
1155                 len = buflen;
1156         if (copy_to_user(buffer, path, len))
1157                 len = -EFAULT;
1158  out:
1159         free_page((unsigned long)tmp);
1160         return len;
1161 }
1162
1163 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1164 {
1165         int error = -EACCES;
1166         struct inode *inode = dentry->d_inode;
1167         struct dentry *de;
1168         struct vfsmount *mnt = NULL;
1169
1170         /* Are we allowed to snoop on the tasks file descriptors? */
1171         if (!proc_fd_access_allowed(inode))
1172                 goto out;
1173
1174         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1175         if (error)
1176                 goto out;
1177
1178         error = do_proc_readlink(de, mnt, buffer, buflen);
1179         dput(de);
1180         mntput(mnt);
1181 out:
1182         return error;
1183 }
1184
1185 static const struct inode_operations proc_pid_link_inode_operations = {
1186         .readlink       = proc_pid_readlink,
1187         .follow_link    = proc_pid_follow_link,
1188         .setattr        = proc_setattr,
1189 };
1190
1191
1192 /* building an inode */
1193
1194 static int task_dumpable(struct task_struct *task)
1195 {
1196         int dumpable = 0;
1197         struct mm_struct *mm;
1198
1199         task_lock(task);
1200         mm = task->mm;
1201         if (mm)
1202                 dumpable = get_dumpable(mm);
1203         task_unlock(task);
1204         if(dumpable == 1)
1205                 return 1;
1206         return 0;
1207 }
1208
1209
1210 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1211 {
1212         struct inode * inode;
1213         struct proc_inode *ei;
1214
1215         /* We need a new inode */
1216
1217         inode = new_inode(sb);
1218         if (!inode)
1219                 goto out;
1220
1221         /* Common stuff */
1222         ei = PROC_I(inode);
1223         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1224         inode->i_op = &proc_def_inode_operations;
1225
1226         /*
1227          * grab the reference to task.
1228          */
1229         ei->pid = get_task_pid(task, PIDTYPE_PID);
1230         if (!ei->pid)
1231                 goto out_unlock;
1232
1233         inode->i_uid = 0;
1234         inode->i_gid = 0;
1235         if (task_dumpable(task)) {
1236                 inode->i_uid = task->euid;
1237                 inode->i_gid = task->egid;
1238         }
1239         security_task_to_inode(task, inode);
1240
1241 out:
1242         return inode;
1243
1244 out_unlock:
1245         iput(inode);
1246         return NULL;
1247 }
1248
1249 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1250 {
1251         struct inode *inode = dentry->d_inode;
1252         struct task_struct *task;
1253         generic_fillattr(inode, stat);
1254
1255         rcu_read_lock();
1256         stat->uid = 0;
1257         stat->gid = 0;
1258         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1259         if (task) {
1260                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1261                     task_dumpable(task)) {
1262                         stat->uid = task->euid;
1263                         stat->gid = task->egid;
1264                 }
1265         }
1266         rcu_read_unlock();
1267         return 0;
1268 }
1269
1270 /* dentry stuff */
1271
1272 /*
1273  *      Exceptional case: normally we are not allowed to unhash a busy
1274  * directory. In this case, however, we can do it - no aliasing problems
1275  * due to the way we treat inodes.
1276  *
1277  * Rewrite the inode's ownerships here because the owning task may have
1278  * performed a setuid(), etc.
1279  *
1280  * Before the /proc/pid/status file was created the only way to read
1281  * the effective uid of a /process was to stat /proc/pid.  Reading
1282  * /proc/pid/status is slow enough that procps and other packages
1283  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1284  * made this apply to all per process world readable and executable
1285  * directories.
1286  */
1287 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1288 {
1289         struct inode *inode = dentry->d_inode;
1290         struct task_struct *task = get_proc_task(inode);
1291         if (task) {
1292                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1293                     task_dumpable(task)) {
1294                         inode->i_uid = task->euid;
1295                         inode->i_gid = task->egid;
1296                 } else {
1297                         inode->i_uid = 0;
1298                         inode->i_gid = 0;
1299                 }
1300                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1301                 security_task_to_inode(task, inode);
1302                 put_task_struct(task);
1303                 return 1;
1304         }
1305         d_drop(dentry);
1306         return 0;
1307 }
1308
1309 static int pid_delete_dentry(struct dentry * dentry)
1310 {
1311         /* Is the task we represent dead?
1312          * If so, then don't put the dentry on the lru list,
1313          * kill it immediately.
1314          */
1315         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1316 }
1317
1318 static struct dentry_operations pid_dentry_operations =
1319 {
1320         .d_revalidate   = pid_revalidate,
1321         .d_delete       = pid_delete_dentry,
1322 };
1323
1324 /* Lookups */
1325
1326 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1327                                 struct task_struct *, const void *);
1328
1329 /*
1330  * Fill a directory entry.
1331  *
1332  * If possible create the dcache entry and derive our inode number and
1333  * file type from dcache entry.
1334  *
1335  * Since all of the proc inode numbers are dynamically generated, the inode
1336  * numbers do not exist until the inode is cache.  This means creating the
1337  * the dcache entry in readdir is necessary to keep the inode numbers
1338  * reported by readdir in sync with the inode numbers reported
1339  * by stat.
1340  */
1341 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1342         char *name, int len,
1343         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1344 {
1345         struct dentry *child, *dir = filp->f_path.dentry;
1346         struct inode *inode;
1347         struct qstr qname;
1348         ino_t ino = 0;
1349         unsigned type = DT_UNKNOWN;
1350
1351         qname.name = name;
1352         qname.len  = len;
1353         qname.hash = full_name_hash(name, len);
1354
1355         child = d_lookup(dir, &qname);
1356         if (!child) {
1357                 struct dentry *new;
1358                 new = d_alloc(dir, &qname);
1359                 if (new) {
1360                         child = instantiate(dir->d_inode, new, task, ptr);
1361                         if (child)
1362                                 dput(new);
1363                         else
1364                                 child = new;
1365                 }
1366         }
1367         if (!child || IS_ERR(child) || !child->d_inode)
1368                 goto end_instantiate;
1369         inode = child->d_inode;
1370         if (inode) {
1371                 ino = inode->i_ino;
1372                 type = inode->i_mode >> 12;
1373         }
1374         dput(child);
1375 end_instantiate:
1376         if (!ino)
1377                 ino = find_inode_number(dir, &qname);
1378         if (!ino)
1379                 ino = 1;
1380         return filldir(dirent, name, len, filp->f_pos, ino, type);
1381 }
1382
1383 static unsigned name_to_int(struct dentry *dentry)
1384 {
1385         const char *name = dentry->d_name.name;
1386         int len = dentry->d_name.len;
1387         unsigned n = 0;
1388
1389         if (len > 1 && *name == '0')
1390                 goto out;
1391         while (len-- > 0) {
1392                 unsigned c = *name++ - '0';
1393                 if (c > 9)
1394                         goto out;
1395                 if (n >= (~0U-9)/10)
1396                         goto out;
1397                 n *= 10;
1398                 n += c;
1399         }
1400         return n;
1401 out:
1402         return ~0U;
1403 }
1404
1405 #define PROC_FDINFO_MAX 64
1406
1407 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
1408                         struct vfsmount **mnt, char *info)
1409 {
1410         struct task_struct *task = get_proc_task(inode);
1411         struct files_struct *files = NULL;
1412         struct file *file;
1413         int fd = proc_fd(inode);
1414
1415         if (task) {
1416                 files = get_files_struct(task);
1417                 put_task_struct(task);
1418         }
1419         if (files) {
1420                 /*
1421                  * We are not taking a ref to the file structure, so we must
1422                  * hold ->file_lock.
1423                  */
1424                 spin_lock(&files->file_lock);
1425                 file = fcheck_files(files, fd);
1426                 if (file) {
1427                         if (mnt)
1428                                 *mnt = mntget(file->f_path.mnt);
1429                         if (dentry)
1430                                 *dentry = dget(file->f_path.dentry);
1431                         if (info)
1432                                 snprintf(info, PROC_FDINFO_MAX,
1433                                          "pos:\t%lli\n"
1434                                          "flags:\t0%o\n",
1435                                          (long long) file->f_pos,
1436                                          file->f_flags);
1437                         spin_unlock(&files->file_lock);
1438                         put_files_struct(files);
1439                         return 0;
1440                 }
1441                 spin_unlock(&files->file_lock);
1442                 put_files_struct(files);
1443         }
1444         return -ENOENT;
1445 }
1446
1447 static int proc_fd_link(struct inode *inode, struct dentry **dentry,
1448                         struct vfsmount **mnt)
1449 {
1450         return proc_fd_info(inode, dentry, mnt, NULL);
1451 }
1452
1453 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1454 {
1455         struct inode *inode = dentry->d_inode;
1456         struct task_struct *task = get_proc_task(inode);
1457         int fd = proc_fd(inode);
1458         struct files_struct *files;
1459
1460         if (task) {
1461                 files = get_files_struct(task);
1462                 if (files) {
1463                         rcu_read_lock();
1464                         if (fcheck_files(files, fd)) {
1465                                 rcu_read_unlock();
1466                                 put_files_struct(files);
1467                                 if (task_dumpable(task)) {
1468                                         inode->i_uid = task->euid;
1469                                         inode->i_gid = task->egid;
1470                                 } else {
1471                                         inode->i_uid = 0;
1472                                         inode->i_gid = 0;
1473                                 }
1474                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1475                                 security_task_to_inode(task, inode);
1476                                 put_task_struct(task);
1477                                 return 1;
1478                         }
1479                         rcu_read_unlock();
1480                         put_files_struct(files);
1481                 }
1482                 put_task_struct(task);
1483         }
1484         d_drop(dentry);
1485         return 0;
1486 }
1487
1488 static struct dentry_operations tid_fd_dentry_operations =
1489 {
1490         .d_revalidate   = tid_fd_revalidate,
1491         .d_delete       = pid_delete_dentry,
1492 };
1493
1494 static struct dentry *proc_fd_instantiate(struct inode *dir,
1495         struct dentry *dentry, struct task_struct *task, const void *ptr)
1496 {
1497         unsigned fd = *(const unsigned *)ptr;
1498         struct file *file;
1499         struct files_struct *files;
1500         struct inode *inode;
1501         struct proc_inode *ei;
1502         struct dentry *error = ERR_PTR(-ENOENT);
1503
1504         inode = proc_pid_make_inode(dir->i_sb, task);
1505         if (!inode)
1506                 goto out;
1507         ei = PROC_I(inode);
1508         ei->fd = fd;
1509         files = get_files_struct(task);
1510         if (!files)
1511                 goto out_iput;
1512         inode->i_mode = S_IFLNK;
1513
1514         /*
1515          * We are not taking a ref to the file structure, so we must
1516          * hold ->file_lock.
1517          */
1518         spin_lock(&files->file_lock);
1519         file = fcheck_files(files, fd);
1520         if (!file)
1521                 goto out_unlock;
1522         if (file->f_mode & 1)
1523                 inode->i_mode |= S_IRUSR | S_IXUSR;
1524         if (file->f_mode & 2)
1525                 inode->i_mode |= S_IWUSR | S_IXUSR;
1526         spin_unlock(&files->file_lock);
1527         put_files_struct(files);
1528
1529         inode->i_op = &proc_pid_link_inode_operations;
1530         inode->i_size = 64;
1531         ei->op.proc_get_link = proc_fd_link;
1532         dentry->d_op = &tid_fd_dentry_operations;
1533         d_add(dentry, inode);
1534         /* Close the race of the process dying before we return the dentry */
1535         if (tid_fd_revalidate(dentry, NULL))
1536                 error = NULL;
1537
1538  out:
1539         return error;
1540 out_unlock:
1541         spin_unlock(&files->file_lock);
1542         put_files_struct(files);
1543 out_iput:
1544         iput(inode);
1545         goto out;
1546 }
1547
1548 static struct dentry *proc_lookupfd_common(struct inode *dir,
1549                                            struct dentry *dentry,
1550                                            instantiate_t instantiate)
1551 {
1552         struct task_struct *task = get_proc_task(dir);
1553         unsigned fd = name_to_int(dentry);
1554         struct dentry *result = ERR_PTR(-ENOENT);
1555
1556         if (!task)
1557                 goto out_no_task;
1558         if (fd == ~0U)
1559                 goto out;
1560
1561         result = instantiate(dir, dentry, task, &fd);
1562 out:
1563         put_task_struct(task);
1564 out_no_task:
1565         return result;
1566 }
1567
1568 static int proc_readfd_common(struct file * filp, void * dirent,
1569                               filldir_t filldir, instantiate_t instantiate)
1570 {
1571         struct dentry *dentry = filp->f_path.dentry;
1572         struct inode *inode = dentry->d_inode;
1573         struct task_struct *p = get_proc_task(inode);
1574         unsigned int fd, ino;
1575         int retval;
1576         struct files_struct * files;
1577         struct fdtable *fdt;
1578
1579         retval = -ENOENT;
1580         if (!p)
1581                 goto out_no_task;
1582         retval = 0;
1583
1584         fd = filp->f_pos;
1585         switch (fd) {
1586                 case 0:
1587                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1588                                 goto out;
1589                         filp->f_pos++;
1590                 case 1:
1591                         ino = parent_ino(dentry);
1592                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1593                                 goto out;
1594                         filp->f_pos++;
1595                 default:
1596                         files = get_files_struct(p);
1597                         if (!files)
1598                                 goto out;
1599                         rcu_read_lock();
1600                         fdt = files_fdtable(files);
1601                         for (fd = filp->f_pos-2;
1602                              fd < fdt->max_fds;
1603                              fd++, filp->f_pos++) {
1604                                 char name[PROC_NUMBUF];
1605                                 int len;
1606
1607                                 if (!fcheck_files(files, fd))
1608                                         continue;
1609                                 rcu_read_unlock();
1610
1611                                 len = snprintf(name, sizeof(name), "%d", fd);
1612                                 if (proc_fill_cache(filp, dirent, filldir,
1613                                                     name, len, instantiate,
1614                                                     p, &fd) < 0) {
1615                                         rcu_read_lock();
1616                                         break;
1617                                 }
1618                                 rcu_read_lock();
1619                         }
1620                         rcu_read_unlock();
1621                         put_files_struct(files);
1622         }
1623 out:
1624         put_task_struct(p);
1625 out_no_task:
1626         return retval;
1627 }
1628
1629 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1630                                     struct nameidata *nd)
1631 {
1632         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1633 }
1634
1635 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1636 {
1637         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1638 }
1639
1640 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1641                                       size_t len, loff_t *ppos)
1642 {
1643         char tmp[PROC_FDINFO_MAX];
1644         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp);
1645         if (!err)
1646                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1647         return err;
1648 }
1649
1650 static const struct file_operations proc_fdinfo_file_operations = {
1651         .open           = nonseekable_open,
1652         .read           = proc_fdinfo_read,
1653 };
1654
1655 static const struct file_operations proc_fd_operations = {
1656         .read           = generic_read_dir,
1657         .readdir        = proc_readfd,
1658 };
1659
1660 /*
1661  * /proc/pid/fd needs a special permission handler so that a process can still
1662  * access /proc/self/fd after it has executed a setuid().
1663  */
1664 static int proc_fd_permission(struct inode *inode, int mask,
1665                                 struct nameidata *nd)
1666 {
1667         int rv;
1668
1669         rv = generic_permission(inode, mask, NULL);
1670         if (rv == 0)
1671                 return 0;
1672         if (task_pid(current) == proc_pid(inode))
1673                 rv = 0;
1674         return rv;
1675 }
1676
1677 /*
1678  * proc directories can do almost nothing..
1679  */
1680 static const struct inode_operations proc_fd_inode_operations = {
1681         .lookup         = proc_lookupfd,
1682         .permission     = proc_fd_permission,
1683         .setattr        = proc_setattr,
1684 };
1685
1686 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1687         struct dentry *dentry, struct task_struct *task, const void *ptr)
1688 {
1689         unsigned fd = *(unsigned *)ptr;
1690         struct inode *inode;
1691         struct proc_inode *ei;
1692         struct dentry *error = ERR_PTR(-ENOENT);
1693
1694         inode = proc_pid_make_inode(dir->i_sb, task);
1695         if (!inode)
1696                 goto out;
1697         ei = PROC_I(inode);
1698         ei->fd = fd;
1699         inode->i_mode = S_IFREG | S_IRUSR;
1700         inode->i_fop = &proc_fdinfo_file_operations;
1701         dentry->d_op = &tid_fd_dentry_operations;
1702         d_add(dentry, inode);
1703         /* Close the race of the process dying before we return the dentry */
1704         if (tid_fd_revalidate(dentry, NULL))
1705                 error = NULL;
1706
1707  out:
1708         return error;
1709 }
1710
1711 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1712                                         struct dentry *dentry,
1713                                         struct nameidata *nd)
1714 {
1715         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1716 }
1717
1718 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1719 {
1720         return proc_readfd_common(filp, dirent, filldir,
1721                                   proc_fdinfo_instantiate);
1722 }
1723
1724 static const struct file_operations proc_fdinfo_operations = {
1725         .read           = generic_read_dir,
1726         .readdir        = proc_readfdinfo,
1727 };
1728
1729 /*
1730  * proc directories can do almost nothing..
1731  */
1732 static const struct inode_operations proc_fdinfo_inode_operations = {
1733         .lookup         = proc_lookupfdinfo,
1734         .setattr        = proc_setattr,
1735 };
1736
1737
1738 static struct dentry *proc_pident_instantiate(struct inode *dir,
1739         struct dentry *dentry, struct task_struct *task, const void *ptr)
1740 {
1741         const struct pid_entry *p = ptr;
1742         struct inode *inode;
1743         struct proc_inode *ei;
1744         struct dentry *error = ERR_PTR(-EINVAL);
1745
1746         inode = proc_pid_make_inode(dir->i_sb, task);
1747         if (!inode)
1748                 goto out;
1749
1750         ei = PROC_I(inode);
1751         inode->i_mode = p->mode;
1752         if (S_ISDIR(inode->i_mode))
1753                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1754         if (p->iop)
1755                 inode->i_op = p->iop;
1756         if (p->fop)
1757                 inode->i_fop = p->fop;
1758         ei->op = p->op;
1759         dentry->d_op = &pid_dentry_operations;
1760         d_add(dentry, inode);
1761         /* Close the race of the process dying before we return the dentry */
1762         if (pid_revalidate(dentry, NULL))
1763                 error = NULL;
1764 out:
1765         return error;
1766 }
1767
1768 static struct dentry *proc_pident_lookup(struct inode *dir, 
1769                                          struct dentry *dentry,
1770                                          const struct pid_entry *ents,
1771                                          unsigned int nents)
1772 {
1773         struct inode *inode;
1774         struct dentry *error;
1775         struct task_struct *task = get_proc_task(dir);
1776         const struct pid_entry *p, *last;
1777
1778         error = ERR_PTR(-ENOENT);
1779         inode = NULL;
1780
1781         if (!task)
1782                 goto out_no_task;
1783
1784         /*
1785          * Yes, it does not scale. And it should not. Don't add
1786          * new entries into /proc/<tgid>/ without very good reasons.
1787          */
1788         last = &ents[nents - 1];
1789         for (p = ents; p <= last; p++) {
1790                 if (p->len != dentry->d_name.len)
1791                         continue;
1792                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1793                         break;
1794         }
1795         if (p > last)
1796                 goto out;
1797
1798         error = proc_pident_instantiate(dir, dentry, task, p);
1799 out:
1800         put_task_struct(task);
1801 out_no_task:
1802         return error;
1803 }
1804
1805 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1806         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1807 {
1808         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1809                                 proc_pident_instantiate, task, p);
1810 }
1811
1812 static int proc_pident_readdir(struct file *filp,
1813                 void *dirent, filldir_t filldir,
1814                 const struct pid_entry *ents, unsigned int nents)
1815 {
1816         int i;
1817         struct dentry *dentry = filp->f_path.dentry;
1818         struct inode *inode = dentry->d_inode;
1819         struct task_struct *task = get_proc_task(inode);
1820         const struct pid_entry *p, *last;
1821         ino_t ino;
1822         int ret;
1823
1824         ret = -ENOENT;
1825         if (!task)
1826                 goto out_no_task;
1827
1828         ret = 0;
1829         i = filp->f_pos;
1830         switch (i) {
1831         case 0:
1832                 ino = inode->i_ino;
1833                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1834                         goto out;
1835                 i++;
1836                 filp->f_pos++;
1837                 /* fall through */
1838         case 1:
1839                 ino = parent_ino(dentry);
1840                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1841                         goto out;
1842                 i++;
1843                 filp->f_pos++;
1844                 /* fall through */
1845         default:
1846                 i -= 2;
1847                 if (i >= nents) {
1848                         ret = 1;
1849                         goto out;
1850                 }
1851                 p = ents + i;
1852                 last = &ents[nents - 1];
1853                 while (p <= last) {
1854                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1855                                 goto out;
1856                         filp->f_pos++;
1857                         p++;
1858                 }
1859         }
1860
1861         ret = 1;
1862 out:
1863         put_task_struct(task);
1864 out_no_task:
1865         return ret;
1866 }
1867
1868 #ifdef CONFIG_SECURITY
1869 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1870                                   size_t count, loff_t *ppos)
1871 {
1872         struct inode * inode = file->f_path.dentry->d_inode;
1873         char *p = NULL;
1874         ssize_t length;
1875         struct task_struct *task = get_proc_task(inode);
1876
1877         if (!task)
1878                 return -ESRCH;
1879
1880         length = security_getprocattr(task,
1881                                       (char*)file->f_path.dentry->d_name.name,
1882                                       &p);
1883         put_task_struct(task);
1884         if (length > 0)
1885                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1886         kfree(p);
1887         return length;
1888 }
1889
1890 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1891                                    size_t count, loff_t *ppos)
1892 {
1893         struct inode * inode = file->f_path.dentry->d_inode;
1894         char *page;
1895         ssize_t length;
1896         struct task_struct *task = get_proc_task(inode);
1897
1898         length = -ESRCH;
1899         if (!task)
1900                 goto out_no_task;
1901         if (count > PAGE_SIZE)
1902                 count = PAGE_SIZE;
1903
1904         /* No partial writes. */
1905         length = -EINVAL;
1906         if (*ppos != 0)
1907                 goto out;
1908
1909         length = -ENOMEM;
1910         page = (char*)__get_free_page(GFP_TEMPORARY);
1911         if (!page)
1912                 goto out;
1913
1914         length = -EFAULT;
1915         if (copy_from_user(page, buf, count))
1916                 goto out_free;
1917
1918         length = security_setprocattr(task,
1919                                       (char*)file->f_path.dentry->d_name.name,
1920                                       (void*)page, count);
1921 out_free:
1922         free_page((unsigned long) page);
1923 out:
1924         put_task_struct(task);
1925 out_no_task:
1926         return length;
1927 }
1928
1929 static const struct file_operations proc_pid_attr_operations = {
1930         .read           = proc_pid_attr_read,
1931         .write          = proc_pid_attr_write,
1932 };
1933
1934 static const struct pid_entry attr_dir_stuff[] = {
1935         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1936         REG("prev",       S_IRUGO,         pid_attr),
1937         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1938         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1939         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1940         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1941 };
1942
1943 static int proc_attr_dir_readdir(struct file * filp,
1944                              void * dirent, filldir_t filldir)
1945 {
1946         return proc_pident_readdir(filp,dirent,filldir,
1947                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1948 }
1949
1950 static const struct file_operations proc_attr_dir_operations = {
1951         .read           = generic_read_dir,
1952         .readdir        = proc_attr_dir_readdir,
1953 };
1954
1955 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1956                                 struct dentry *dentry, struct nameidata *nd)
1957 {
1958         return proc_pident_lookup(dir, dentry,
1959                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
1960 }
1961
1962 static const struct inode_operations proc_attr_dir_inode_operations = {
1963         .lookup         = proc_attr_dir_lookup,
1964         .getattr        = pid_getattr,
1965         .setattr        = proc_setattr,
1966 };
1967
1968 #endif
1969
1970 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1971 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
1972                                          size_t count, loff_t *ppos)
1973 {
1974         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1975         struct mm_struct *mm;
1976         char buffer[PROC_NUMBUF];
1977         size_t len;
1978         int ret;
1979
1980         if (!task)
1981                 return -ESRCH;
1982
1983         ret = 0;
1984         mm = get_task_mm(task);
1985         if (mm) {
1986                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
1987                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
1988                                 MMF_DUMP_FILTER_SHIFT));
1989                 mmput(mm);
1990                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
1991         }
1992
1993         put_task_struct(task);
1994
1995         return ret;
1996 }
1997
1998 static ssize_t proc_coredump_filter_write(struct file *file,
1999                                           const char __user *buf,
2000                                           size_t count,
2001                                           loff_t *ppos)
2002 {
2003         struct task_struct *task;
2004         struct mm_struct *mm;
2005         char buffer[PROC_NUMBUF], *end;
2006         unsigned int val;
2007         int ret;
2008         int i;
2009         unsigned long mask;
2010
2011         ret = -EFAULT;
2012         memset(buffer, 0, sizeof(buffer));
2013         if (count > sizeof(buffer) - 1)
2014                 count = sizeof(buffer) - 1;
2015         if (copy_from_user(buffer, buf, count))
2016                 goto out_no_task;
2017
2018         ret = -EINVAL;
2019         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2020         if (*end == '\n')
2021                 end++;
2022         if (end - buffer == 0)
2023                 goto out_no_task;
2024
2025         ret = -ESRCH;
2026         task = get_proc_task(file->f_dentry->d_inode);
2027         if (!task)
2028                 goto out_no_task;
2029
2030         ret = end - buffer;
2031         mm = get_task_mm(task);
2032         if (!mm)
2033                 goto out_no_mm;
2034
2035         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2036                 if (val & mask)
2037                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2038                 else
2039                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2040         }
2041
2042         mmput(mm);
2043  out_no_mm:
2044         put_task_struct(task);
2045  out_no_task:
2046         return ret;
2047 }
2048
2049 static const struct file_operations proc_coredump_filter_operations = {
2050         .read           = proc_coredump_filter_read,
2051         .write          = proc_coredump_filter_write,
2052 };
2053 #endif
2054
2055 /*
2056  * /proc/self:
2057  */
2058 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2059                               int buflen)
2060 {
2061         char tmp[PROC_NUMBUF];
2062         sprintf(tmp, "%d", task_tgid_vnr(current));
2063         return vfs_readlink(dentry,buffer,buflen,tmp);
2064 }
2065
2066 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2067 {
2068         char tmp[PROC_NUMBUF];
2069         sprintf(tmp, "%d", task_tgid_vnr(current));
2070         return ERR_PTR(vfs_follow_link(nd,tmp));
2071 }
2072
2073 static const struct inode_operations proc_self_inode_operations = {
2074         .readlink       = proc_self_readlink,
2075         .follow_link    = proc_self_follow_link,
2076 };
2077
2078 /*
2079  * proc base
2080  *
2081  * These are the directory entries in the root directory of /proc
2082  * that properly belong to the /proc filesystem, as they describe
2083  * describe something that is process related.
2084  */
2085 static const struct pid_entry proc_base_stuff[] = {
2086         NOD("self", S_IFLNK|S_IRWXUGO,
2087                 &proc_self_inode_operations, NULL, {}),
2088 };
2089
2090 /*
2091  *      Exceptional case: normally we are not allowed to unhash a busy
2092  * directory. In this case, however, we can do it - no aliasing problems
2093  * due to the way we treat inodes.
2094  */
2095 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2096 {
2097         struct inode *inode = dentry->d_inode;
2098         struct task_struct *task = get_proc_task(inode);
2099         if (task) {
2100                 put_task_struct(task);
2101                 return 1;
2102         }
2103         d_drop(dentry);
2104         return 0;
2105 }
2106
2107 static struct dentry_operations proc_base_dentry_operations =
2108 {
2109         .d_revalidate   = proc_base_revalidate,
2110         .d_delete       = pid_delete_dentry,
2111 };
2112
2113 static struct dentry *proc_base_instantiate(struct inode *dir,
2114         struct dentry *dentry, struct task_struct *task, const void *ptr)
2115 {
2116         const struct pid_entry *p = ptr;
2117         struct inode *inode;
2118         struct proc_inode *ei;
2119         struct dentry *error = ERR_PTR(-EINVAL);
2120
2121         /* Allocate the inode */
2122         error = ERR_PTR(-ENOMEM);
2123         inode = new_inode(dir->i_sb);
2124         if (!inode)
2125                 goto out;
2126
2127         /* Initialize the inode */
2128         ei = PROC_I(inode);
2129         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2130
2131         /*
2132          * grab the reference to the task.
2133          */
2134         ei->pid = get_task_pid(task, PIDTYPE_PID);
2135         if (!ei->pid)
2136                 goto out_iput;
2137
2138         inode->i_uid = 0;
2139         inode->i_gid = 0;
2140         inode->i_mode = p->mode;
2141         if (S_ISDIR(inode->i_mode))
2142                 inode->i_nlink = 2;
2143         if (S_ISLNK(inode->i_mode))
2144                 inode->i_size = 64;
2145         if (p->iop)
2146                 inode->i_op = p->iop;
2147         if (p->fop)
2148                 inode->i_fop = p->fop;
2149         ei->op = p->op;
2150         dentry->d_op = &proc_base_dentry_operations;
2151         d_add(dentry, inode);
2152         error = NULL;
2153 out:
2154         return error;
2155 out_iput:
2156         iput(inode);
2157         goto out;
2158 }
2159
2160 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2161 {
2162         struct dentry *error;
2163         struct task_struct *task = get_proc_task(dir);
2164         const struct pid_entry *p, *last;
2165
2166         error = ERR_PTR(-ENOENT);
2167
2168         if (!task)
2169                 goto out_no_task;
2170
2171         /* Lookup the directory entry */
2172         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2173         for (p = proc_base_stuff; p <= last; p++) {
2174                 if (p->len != dentry->d_name.len)
2175                         continue;
2176                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2177                         break;
2178         }
2179         if (p > last)
2180                 goto out;
2181
2182         error = proc_base_instantiate(dir, dentry, task, p);
2183
2184 out:
2185         put_task_struct(task);
2186 out_no_task:
2187         return error;
2188 }
2189
2190 static int proc_base_fill_cache(struct file *filp, void *dirent,
2191         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2192 {
2193         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2194                                 proc_base_instantiate, task, p);
2195 }
2196
2197 #ifdef CONFIG_TASK_IO_ACCOUNTING
2198 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2199 {
2200         return sprintf(buffer,
2201 #ifdef CONFIG_TASK_XACCT
2202                         "rchar: %llu\n"
2203                         "wchar: %llu\n"
2204                         "syscr: %llu\n"
2205                         "syscw: %llu\n"
2206 #endif
2207                         "read_bytes: %llu\n"
2208                         "write_bytes: %llu\n"
2209                         "cancelled_write_bytes: %llu\n",
2210 #ifdef CONFIG_TASK_XACCT
2211                         (unsigned long long)task->rchar,
2212                         (unsigned long long)task->wchar,
2213                         (unsigned long long)task->syscr,
2214                         (unsigned long long)task->syscw,
2215 #endif
2216                         (unsigned long long)task->ioac.read_bytes,
2217                         (unsigned long long)task->ioac.write_bytes,
2218                         (unsigned long long)task->ioac.cancelled_write_bytes);
2219 }
2220 #endif
2221
2222 /*
2223  * Thread groups
2224  */
2225 static const struct file_operations proc_task_operations;
2226 static const struct inode_operations proc_task_inode_operations;
2227
2228 static const struct pid_entry tgid_base_stuff[] = {
2229         DIR("task",       S_IRUGO|S_IXUGO, task),
2230         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2231         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2232         REG("environ",    S_IRUSR, environ),
2233         INF("auxv",       S_IRUSR, pid_auxv),
2234         INF("status",     S_IRUGO, pid_status),
2235         INF("limits",     S_IRUSR, pid_limits),
2236 #ifdef CONFIG_SCHED_DEBUG
2237         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2238 #endif
2239         INF("cmdline",    S_IRUGO, pid_cmdline),
2240         INF("stat",       S_IRUGO, tgid_stat),
2241         INF("statm",      S_IRUGO, pid_statm),
2242         REG("maps",       S_IRUGO, maps),
2243 #ifdef CONFIG_NUMA
2244         REG("numa_maps",  S_IRUGO, numa_maps),
2245 #endif
2246         REG("mem",        S_IRUSR|S_IWUSR, mem),
2247         LNK("cwd",        cwd),
2248         LNK("root",       root),
2249         LNK("exe",        exe),
2250         REG("mounts",     S_IRUGO, mounts),
2251         REG("mountstats", S_IRUSR, mountstats),
2252 #ifdef CONFIG_MMU
2253         REG("clear_refs", S_IWUSR, clear_refs),
2254         REG("smaps",      S_IRUGO, smaps),
2255 #endif
2256 #ifdef CONFIG_SECURITY
2257         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2258 #endif
2259 #ifdef CONFIG_KALLSYMS
2260         INF("wchan",      S_IRUGO, pid_wchan),
2261 #endif
2262 #ifdef CONFIG_SCHEDSTATS
2263         INF("schedstat",  S_IRUGO, pid_schedstat),
2264 #endif
2265 #ifdef CONFIG_LATENCYTOP
2266         REG("latency",  S_IRUGO, lstats),
2267 #endif
2268 #ifdef CONFIG_PROC_PID_CPUSET
2269         REG("cpuset",     S_IRUGO, cpuset),
2270 #endif
2271 #ifdef CONFIG_CGROUPS
2272         REG("cgroup",  S_IRUGO, cgroup),
2273 #endif
2274         INF("oom_score",  S_IRUGO, oom_score),
2275         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2276 #ifdef CONFIG_AUDITSYSCALL
2277         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2278 #endif
2279 #ifdef CONFIG_FAULT_INJECTION
2280         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2281 #endif
2282 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2283         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2284 #endif
2285 #ifdef CONFIG_TASK_IO_ACCOUNTING
2286         INF("io",       S_IRUGO, pid_io_accounting),
2287 #endif
2288 };
2289
2290 static int proc_tgid_base_readdir(struct file * filp,
2291                              void * dirent, filldir_t filldir)
2292 {
2293         return proc_pident_readdir(filp,dirent,filldir,
2294                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2295 }
2296
2297 static const struct file_operations proc_tgid_base_operations = {
2298         .read           = generic_read_dir,
2299         .readdir        = proc_tgid_base_readdir,
2300 };
2301
2302 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2303         return proc_pident_lookup(dir, dentry,
2304                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2305 }
2306
2307 static const struct inode_operations proc_tgid_base_inode_operations = {
2308         .lookup         = proc_tgid_base_lookup,
2309         .getattr        = pid_getattr,
2310         .setattr        = proc_setattr,
2311 };
2312
2313 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2314 {
2315         struct dentry *dentry, *leader, *dir;
2316         char buf[PROC_NUMBUF];
2317         struct qstr name;
2318
2319         name.name = buf;
2320         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2321         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2322         if (dentry) {
2323                 shrink_dcache_parent(dentry);
2324                 d_drop(dentry);
2325                 dput(dentry);
2326         }
2327
2328         if (tgid == 0)
2329                 goto out;
2330
2331         name.name = buf;
2332         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2333         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2334         if (!leader)
2335                 goto out;
2336
2337         name.name = "task";
2338         name.len = strlen(name.name);
2339         dir = d_hash_and_lookup(leader, &name);
2340         if (!dir)
2341                 goto out_put_leader;
2342
2343         name.name = buf;
2344         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2345         dentry = d_hash_and_lookup(dir, &name);
2346         if (dentry) {
2347                 shrink_dcache_parent(dentry);
2348                 d_drop(dentry);
2349                 dput(dentry);
2350         }
2351
2352         dput(dir);
2353 out_put_leader:
2354         dput(leader);
2355 out:
2356         return;
2357 }
2358
2359 /**
2360  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2361  * @task: task that should be flushed.
2362  *
2363  * When flushing dentries from proc, one needs to flush them from global
2364  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2365  * in. This call is supposed to do all of this job.
2366  *
2367  * Looks in the dcache for
2368  * /proc/@pid
2369  * /proc/@tgid/task/@pid
2370  * if either directory is present flushes it and all of it'ts children
2371  * from the dcache.
2372  *
2373  * It is safe and reasonable to cache /proc entries for a task until
2374  * that task exits.  After that they just clog up the dcache with
2375  * useless entries, possibly causing useful dcache entries to be
2376  * flushed instead.  This routine is proved to flush those useless
2377  * dcache entries at process exit time.
2378  *
2379  * NOTE: This routine is just an optimization so it does not guarantee
2380  *       that no dcache entries will exist at process exit time it
2381  *       just makes it very unlikely that any will persist.
2382  */
2383
2384 void proc_flush_task(struct task_struct *task)
2385 {
2386         int i;
2387         struct pid *pid, *tgid = NULL;
2388         struct upid *upid;
2389
2390         pid = task_pid(task);
2391         if (thread_group_leader(task))
2392                 tgid = task_tgid(task);
2393
2394         for (i = 0; i <= pid->level; i++) {
2395                 upid = &pid->numbers[i];
2396                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2397                         tgid ? tgid->numbers[i].nr : 0);
2398         }
2399
2400         upid = &pid->numbers[pid->level];
2401         if (upid->nr == 1)
2402                 pid_ns_release_proc(upid->ns);
2403 }
2404
2405 static struct dentry *proc_pid_instantiate(struct inode *dir,
2406                                            struct dentry * dentry,
2407                                            struct task_struct *task, const void *ptr)
2408 {
2409         struct dentry *error = ERR_PTR(-ENOENT);
2410         struct inode *inode;
2411
2412         inode = proc_pid_make_inode(dir->i_sb, task);
2413         if (!inode)
2414                 goto out;
2415
2416         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2417         inode->i_op = &proc_tgid_base_inode_operations;
2418         inode->i_fop = &proc_tgid_base_operations;
2419         inode->i_flags|=S_IMMUTABLE;
2420         inode->i_nlink = 5;
2421 #ifdef CONFIG_SECURITY
2422         inode->i_nlink += 1;
2423 #endif
2424
2425         dentry->d_op = &pid_dentry_operations;
2426
2427         d_add(dentry, inode);
2428         /* Close the race of the process dying before we return the dentry */
2429         if (pid_revalidate(dentry, NULL))
2430                 error = NULL;
2431 out:
2432         return error;
2433 }
2434
2435 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2436 {
2437         struct dentry *result = ERR_PTR(-ENOENT);
2438         struct task_struct *task;
2439         unsigned tgid;
2440         struct pid_namespace *ns;
2441
2442         result = proc_base_lookup(dir, dentry);
2443         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2444                 goto out;
2445
2446         tgid = name_to_int(dentry);
2447         if (tgid == ~0U)
2448                 goto out;
2449
2450         ns = dentry->d_sb->s_fs_info;
2451         rcu_read_lock();
2452         task = find_task_by_pid_ns(tgid, ns);
2453         if (task)
2454                 get_task_struct(task);
2455         rcu_read_unlock();
2456         if (!task)
2457                 goto out;
2458
2459         result = proc_pid_instantiate(dir, dentry, task, NULL);
2460         put_task_struct(task);
2461 out:
2462         return result;
2463 }
2464
2465 /*
2466  * Find the first task with tgid >= tgid
2467  *
2468  */
2469 struct tgid_iter {
2470         unsigned int tgid;
2471         struct task_struct *task;
2472 };
2473 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2474 {
2475         struct pid *pid;
2476
2477         if (iter.task)
2478                 put_task_struct(iter.task);
2479         rcu_read_lock();
2480 retry:
2481         iter.task = NULL;
2482         pid = find_ge_pid(iter.tgid, ns);
2483         if (pid) {
2484                 iter.tgid = pid_nr_ns(pid, ns);
2485                 iter.task = pid_task(pid, PIDTYPE_PID);
2486                 /* What we to know is if the pid we have find is the
2487                  * pid of a thread_group_leader.  Testing for task
2488                  * being a thread_group_leader is the obvious thing
2489                  * todo but there is a window when it fails, due to
2490                  * the pid transfer logic in de_thread.
2491                  *
2492                  * So we perform the straight forward test of seeing
2493                  * if the pid we have found is the pid of a thread
2494                  * group leader, and don't worry if the task we have
2495                  * found doesn't happen to be a thread group leader.
2496                  * As we don't care in the case of readdir.
2497                  */
2498                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2499                         iter.tgid += 1;
2500                         goto retry;
2501                 }
2502                 get_task_struct(iter.task);
2503         }
2504         rcu_read_unlock();
2505         return iter;
2506 }
2507
2508 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2509
2510 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2511         struct tgid_iter iter)
2512 {
2513         char name[PROC_NUMBUF];
2514         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2515         return proc_fill_cache(filp, dirent, filldir, name, len,
2516                                 proc_pid_instantiate, iter.task, NULL);
2517 }
2518
2519 /* for the /proc/ directory itself, after non-process stuff has been done */
2520 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2521 {
2522         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2523         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2524         struct tgid_iter iter;
2525         struct pid_namespace *ns;
2526
2527         if (!reaper)
2528                 goto out_no_task;
2529
2530         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2531                 const struct pid_entry *p = &proc_base_stuff[nr];
2532                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2533                         goto out;
2534         }
2535
2536         ns = filp->f_dentry->d_sb->s_fs_info;
2537         iter.task = NULL;
2538         iter.tgid = filp->f_pos - TGID_OFFSET;
2539         for (iter = next_tgid(ns, iter);
2540              iter.task;
2541              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2542                 filp->f_pos = iter.tgid + TGID_OFFSET;
2543                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2544                         put_task_struct(iter.task);
2545                         goto out;
2546                 }
2547         }
2548         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2549 out:
2550         put_task_struct(reaper);
2551 out_no_task:
2552         return 0;
2553 }
2554
2555 /*
2556  * Tasks
2557  */
2558 static const struct pid_entry tid_base_stuff[] = {
2559         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2560         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2561         REG("environ",   S_IRUSR, environ),
2562         INF("auxv",      S_IRUSR, pid_auxv),
2563         INF("status",    S_IRUGO, pid_status),
2564         INF("limits",    S_IRUSR, pid_limits),
2565 #ifdef CONFIG_SCHED_DEBUG
2566         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2567 #endif
2568         INF("cmdline",   S_IRUGO, pid_cmdline),
2569         INF("stat",      S_IRUGO, tid_stat),
2570         INF("statm",     S_IRUGO, pid_statm),
2571         REG("maps",      S_IRUGO, maps),
2572 #ifdef CONFIG_NUMA
2573         REG("numa_maps", S_IRUGO, numa_maps),
2574 #endif
2575         REG("mem",       S_IRUSR|S_IWUSR, mem),
2576         LNK("cwd",       cwd),
2577         LNK("root",      root),
2578         LNK("exe",       exe),
2579         REG("mounts",    S_IRUGO, mounts),
2580 #ifdef CONFIG_MMU
2581         REG("clear_refs", S_IWUSR, clear_refs),
2582         REG("smaps",     S_IRUGO, smaps),
2583 #endif
2584 #ifdef CONFIG_SECURITY
2585         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2586 #endif
2587 #ifdef CONFIG_KALLSYMS
2588         INF("wchan",     S_IRUGO, pid_wchan),
2589 #endif
2590 #ifdef CONFIG_SCHEDSTATS
2591         INF("schedstat", S_IRUGO, pid_schedstat),
2592 #endif
2593 #ifdef CONFIG_LATENCYTOP
2594         REG("latency",  S_IRUGO, lstats),
2595 #endif
2596 #ifdef CONFIG_PROC_PID_CPUSET
2597         REG("cpuset",    S_IRUGO, cpuset),
2598 #endif
2599 #ifdef CONFIG_CGROUPS
2600         REG("cgroup",  S_IRUGO, cgroup),
2601 #endif
2602         INF("oom_score", S_IRUGO, oom_score),
2603         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2604 #ifdef CONFIG_AUDITSYSCALL
2605         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2606 #endif
2607 #ifdef CONFIG_FAULT_INJECTION
2608         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2609 #endif
2610 };
2611
2612 static int proc_tid_base_readdir(struct file * filp,
2613                              void * dirent, filldir_t filldir)
2614 {
2615         return proc_pident_readdir(filp,dirent,filldir,
2616                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2617 }
2618
2619 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2620         return proc_pident_lookup(dir, dentry,
2621                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2622 }
2623
2624 static const struct file_operations proc_tid_base_operations = {
2625         .read           = generic_read_dir,
2626         .readdir        = proc_tid_base_readdir,
2627 };
2628
2629 static const struct inode_operations proc_tid_base_inode_operations = {
2630         .lookup         = proc_tid_base_lookup,
2631         .getattr        = pid_getattr,
2632         .setattr        = proc_setattr,
2633 };
2634
2635 static struct dentry *proc_task_instantiate(struct inode *dir,
2636         struct dentry *dentry, struct task_struct *task, const void *ptr)
2637 {
2638         struct dentry *error = ERR_PTR(-ENOENT);
2639         struct inode *inode;
2640         inode = proc_pid_make_inode(dir->i_sb, task);
2641
2642         if (!inode)
2643                 goto out;
2644         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2645         inode->i_op = &proc_tid_base_inode_operations;
2646         inode->i_fop = &proc_tid_base_operations;
2647         inode->i_flags|=S_IMMUTABLE;
2648         inode->i_nlink = 4;
2649 #ifdef CONFIG_SECURITY
2650         inode->i_nlink += 1;
2651 #endif
2652
2653         dentry->d_op = &pid_dentry_operations;
2654
2655         d_add(dentry, inode);
2656         /* Close the race of the process dying before we return the dentry */
2657         if (pid_revalidate(dentry, NULL))
2658                 error = NULL;
2659 out:
2660         return error;
2661 }
2662
2663 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2664 {
2665         struct dentry *result = ERR_PTR(-ENOENT);
2666         struct task_struct *task;
2667         struct task_struct *leader = get_proc_task(dir);
2668         unsigned tid;
2669         struct pid_namespace *ns;
2670
2671         if (!leader)
2672                 goto out_no_task;
2673
2674         tid = name_to_int(dentry);
2675         if (tid == ~0U)
2676                 goto out;
2677
2678         ns = dentry->d_sb->s_fs_info;
2679         rcu_read_lock();
2680         task = find_task_by_pid_ns(tid, ns);
2681         if (task)
2682                 get_task_struct(task);
2683         rcu_read_unlock();
2684         if (!task)
2685                 goto out;
2686         if (!same_thread_group(leader, task))
2687                 goto out_drop_task;
2688
2689         result = proc_task_instantiate(dir, dentry, task, NULL);
2690 out_drop_task:
2691         put_task_struct(task);
2692 out:
2693         put_task_struct(leader);
2694 out_no_task:
2695         return result;
2696 }
2697
2698 /*
2699  * Find the first tid of a thread group to return to user space.
2700  *
2701  * Usually this is just the thread group leader, but if the users
2702  * buffer was too small or there was a seek into the middle of the
2703  * directory we have more work todo.
2704  *
2705  * In the case of a short read we start with find_task_by_pid.
2706  *
2707  * In the case of a seek we start with the leader and walk nr
2708  * threads past it.
2709  */
2710 static struct task_struct *first_tid(struct task_struct *leader,
2711                 int tid, int nr, struct pid_namespace *ns)
2712 {
2713         struct task_struct *pos;
2714
2715         rcu_read_lock();
2716         /* Attempt to start with the pid of a thread */
2717         if (tid && (nr > 0)) {
2718                 pos = find_task_by_pid_ns(tid, ns);
2719                 if (pos && (pos->group_leader == leader))
2720                         goto found;
2721         }
2722
2723         /* If nr exceeds the number of threads there is nothing todo */
2724         pos = NULL;
2725         if (nr && nr >= get_nr_threads(leader))
2726                 goto out;
2727
2728         /* If we haven't found our starting place yet start
2729          * with the leader and walk nr threads forward.
2730          */
2731         for (pos = leader; nr > 0; --nr) {
2732                 pos = next_thread(pos);
2733                 if (pos == leader) {
2734                         pos = NULL;
2735                         goto out;
2736                 }
2737         }
2738 found:
2739         get_task_struct(pos);
2740 out:
2741         rcu_read_unlock();
2742         return pos;
2743 }
2744
2745 /*
2746  * Find the next thread in the thread list.
2747  * Return NULL if there is an error or no next thread.
2748  *
2749  * The reference to the input task_struct is released.
2750  */
2751 static struct task_struct *next_tid(struct task_struct *start)
2752 {
2753         struct task_struct *pos = NULL;
2754         rcu_read_lock();
2755         if (pid_alive(start)) {
2756                 pos = next_thread(start);
2757                 if (thread_group_leader(pos))
2758                         pos = NULL;
2759                 else
2760                         get_task_struct(pos);
2761         }
2762         rcu_read_unlock();
2763         put_task_struct(start);
2764         return pos;
2765 }
2766
2767 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2768         struct task_struct *task, int tid)
2769 {
2770         char name[PROC_NUMBUF];
2771         int len = snprintf(name, sizeof(name), "%d", tid);
2772         return proc_fill_cache(filp, dirent, filldir, name, len,
2773                                 proc_task_instantiate, task, NULL);
2774 }
2775
2776 /* for the /proc/TGID/task/ directories */
2777 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2778 {
2779         struct dentry *dentry = filp->f_path.dentry;
2780         struct inode *inode = dentry->d_inode;
2781         struct task_struct *leader = NULL;
2782         struct task_struct *task;
2783         int retval = -ENOENT;
2784         ino_t ino;
2785         int tid;
2786         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2787         struct pid_namespace *ns;
2788
2789         task = get_proc_task(inode);
2790         if (!task)
2791                 goto out_no_task;
2792         rcu_read_lock();
2793         if (pid_alive(task)) {
2794                 leader = task->group_leader;
2795                 get_task_struct(leader);
2796         }
2797         rcu_read_unlock();
2798         put_task_struct(task);
2799         if (!leader)
2800                 goto out_no_task;
2801         retval = 0;
2802
2803         switch (pos) {
2804         case 0:
2805                 ino = inode->i_ino;
2806                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2807                         goto out;
2808                 pos++;
2809                 /* fall through */
2810         case 1:
2811                 ino = parent_ino(dentry);
2812                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2813                         goto out;
2814                 pos++;
2815                 /* fall through */
2816         }
2817
2818         /* f_version caches the tgid value that the last readdir call couldn't
2819          * return. lseek aka telldir automagically resets f_version to 0.
2820          */
2821         ns = filp->f_dentry->d_sb->s_fs_info;
2822         tid = (int)filp->f_version;
2823         filp->f_version = 0;
2824         for (task = first_tid(leader, tid, pos - 2, ns);
2825              task;
2826              task = next_tid(task), pos++) {
2827                 tid = task_pid_nr_ns(task, ns);
2828                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2829                         /* returning this tgid failed, save it as the first
2830                          * pid for the next readir call */
2831                         filp->f_version = (u64)tid;
2832                         put_task_struct(task);
2833                         break;
2834                 }
2835         }
2836 out:
2837         filp->f_pos = pos;
2838         put_task_struct(leader);
2839 out_no_task:
2840         return retval;
2841 }
2842
2843 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2844 {
2845         struct inode *inode = dentry->d_inode;
2846         struct task_struct *p = get_proc_task(inode);
2847         generic_fillattr(inode, stat);
2848
2849         if (p) {
2850                 rcu_read_lock();
2851                 stat->nlink += get_nr_threads(p);
2852                 rcu_read_unlock();
2853                 put_task_struct(p);
2854         }
2855
2856         return 0;
2857 }
2858
2859 static const struct inode_operations proc_task_inode_operations = {
2860         .lookup         = proc_task_lookup,
2861         .getattr        = proc_task_getattr,
2862         .setattr        = proc_setattr,
2863 };
2864
2865 static const struct file_operations proc_task_operations = {
2866         .read           = generic_read_dir,
2867         .readdir        = proc_task_readdir,
2868 };