proc: remove kernel.maps_protect
[linux-2.6.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/rcupdate.h>
67 #include <linux/kallsyms.h>
68 #include <linux/resource.h>
69 #include <linux/module.h>
70 #include <linux/mount.h>
71 #include <linux/security.h>
72 #include <linux/ptrace.h>
73 #include <linux/tracehook.h>
74 #include <linux/cgroup.h>
75 #include <linux/cpuset.h>
76 #include <linux/audit.h>
77 #include <linux/poll.h>
78 #include <linux/nsproxy.h>
79 #include <linux/oom.h>
80 #include <linux/elf.h>
81 #include <linux/pid_namespace.h>
82 #include "internal.h"
83
84 /* NOTE:
85  *      Implementing inode permission operations in /proc is almost
86  *      certainly an error.  Permission checks need to happen during
87  *      each system call not at open time.  The reason is that most of
88  *      what we wish to check for permissions in /proc varies at runtime.
89  *
90  *      The classic example of a problem is opening file descriptors
91  *      in /proc for a task before it execs a suid executable.
92  */
93
94 struct pid_entry {
95         char *name;
96         int len;
97         mode_t mode;
98         const struct inode_operations *iop;
99         const struct file_operations *fop;
100         union proc_op op;
101 };
102
103 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
104         .name = (NAME),                                 \
105         .len  = sizeof(NAME) - 1,                       \
106         .mode = MODE,                                   \
107         .iop  = IOP,                                    \
108         .fop  = FOP,                                    \
109         .op   = OP,                                     \
110 }
111
112 #define DIR(NAME, MODE, OTYPE)                                                  \
113         NOD(NAME, (S_IFDIR|(MODE)),                                             \
114                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
115                 {} )
116 #define LNK(NAME, OTYPE)                                        \
117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
118                 &proc_pid_link_inode_operations, NULL,          \
119                 { .proc_get_link = &proc_##OTYPE##_link } )
120 #define REG(NAME, MODE, OTYPE)                          \
121         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
122                 &proc_##OTYPE##_operations, {})
123 #define INF(NAME, MODE, OTYPE)                          \
124         NOD(NAME, (S_IFREG|(MODE)),                     \
125                 NULL, &proc_info_file_operations,       \
126                 { .proc_read = &proc_##OTYPE } )
127 #define ONE(NAME, MODE, OTYPE)                          \
128         NOD(NAME, (S_IFREG|(MODE)),                     \
129                 NULL, &proc_single_file_operations,     \
130                 { .proc_show = &proc_##OTYPE } )
131
132 /*
133  * Count the number of hardlinks for the pid_entry table, excluding the .
134  * and .. links.
135  */
136 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
137         unsigned int n)
138 {
139         unsigned int i;
140         unsigned int count;
141
142         count = 0;
143         for (i = 0; i < n; ++i) {
144                 if (S_ISDIR(entries[i].mode))
145                         ++count;
146         }
147
148         return count;
149 }
150
151 static struct fs_struct *get_fs_struct(struct task_struct *task)
152 {
153         struct fs_struct *fs;
154         task_lock(task);
155         fs = task->fs;
156         if(fs)
157                 atomic_inc(&fs->count);
158         task_unlock(task);
159         return fs;
160 }
161
162 static int get_nr_threads(struct task_struct *tsk)
163 {
164         unsigned long flags;
165         int count = 0;
166
167         if (lock_task_sighand(tsk, &flags)) {
168                 count = atomic_read(&tsk->signal->count);
169                 unlock_task_sighand(tsk, &flags);
170         }
171         return count;
172 }
173
174 static int proc_cwd_link(struct inode *inode, struct path *path)
175 {
176         struct task_struct *task = get_proc_task(inode);
177         struct fs_struct *fs = NULL;
178         int result = -ENOENT;
179
180         if (task) {
181                 fs = get_fs_struct(task);
182                 put_task_struct(task);
183         }
184         if (fs) {
185                 read_lock(&fs->lock);
186                 *path = fs->pwd;
187                 path_get(&fs->pwd);
188                 read_unlock(&fs->lock);
189                 result = 0;
190                 put_fs_struct(fs);
191         }
192         return result;
193 }
194
195 static int proc_root_link(struct inode *inode, struct path *path)
196 {
197         struct task_struct *task = get_proc_task(inode);
198         struct fs_struct *fs = NULL;
199         int result = -ENOENT;
200
201         if (task) {
202                 fs = get_fs_struct(task);
203                 put_task_struct(task);
204         }
205         if (fs) {
206                 read_lock(&fs->lock);
207                 *path = fs->root;
208                 path_get(&fs->root);
209                 read_unlock(&fs->lock);
210                 result = 0;
211                 put_fs_struct(fs);
212         }
213         return result;
214 }
215
216 /*
217  * Return zero if current may access user memory in @task, -error if not.
218  */
219 static int check_mem_permission(struct task_struct *task)
220 {
221         /*
222          * A task can always look at itself, in case it chooses
223          * to use system calls instead of load instructions.
224          */
225         if (task == current)
226                 return 0;
227
228         /*
229          * If current is actively ptrace'ing, and would also be
230          * permitted to freshly attach with ptrace now, permit it.
231          */
232         if (task_is_stopped_or_traced(task)) {
233                 int match;
234                 rcu_read_lock();
235                 match = (tracehook_tracer_task(task) == current);
236                 rcu_read_unlock();
237                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
238                         return 0;
239         }
240
241         /*
242          * Noone else is allowed.
243          */
244         return -EPERM;
245 }
246
247 struct mm_struct *mm_for_maps(struct task_struct *task)
248 {
249         struct mm_struct *mm = get_task_mm(task);
250         if (!mm)
251                 return NULL;
252         down_read(&mm->mmap_sem);
253         task_lock(task);
254         if (task->mm != mm)
255                 goto out;
256         if (task->mm != current->mm &&
257             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
258                 goto out;
259         task_unlock(task);
260         return mm;
261 out:
262         task_unlock(task);
263         up_read(&mm->mmap_sem);
264         mmput(mm);
265         return NULL;
266 }
267
268 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
269 {
270         int res = 0;
271         unsigned int len;
272         struct mm_struct *mm = get_task_mm(task);
273         if (!mm)
274                 goto out;
275         if (!mm->arg_end)
276                 goto out_mm;    /* Shh! No looking before we're done */
277
278         len = mm->arg_end - mm->arg_start;
279  
280         if (len > PAGE_SIZE)
281                 len = PAGE_SIZE;
282  
283         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
284
285         // If the nul at the end of args has been overwritten, then
286         // assume application is using setproctitle(3).
287         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
288                 len = strnlen(buffer, res);
289                 if (len < res) {
290                     res = len;
291                 } else {
292                         len = mm->env_end - mm->env_start;
293                         if (len > PAGE_SIZE - res)
294                                 len = PAGE_SIZE - res;
295                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
296                         res = strnlen(buffer, res);
297                 }
298         }
299 out_mm:
300         mmput(mm);
301 out:
302         return res;
303 }
304
305 static int proc_pid_auxv(struct task_struct *task, char *buffer)
306 {
307         int res = 0;
308         struct mm_struct *mm = get_task_mm(task);
309         if (mm) {
310                 unsigned int nwords = 0;
311                 do
312                         nwords += 2;
313                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
314                 res = nwords * sizeof(mm->saved_auxv[0]);
315                 if (res > PAGE_SIZE)
316                         res = PAGE_SIZE;
317                 memcpy(buffer, mm->saved_auxv, res);
318                 mmput(mm);
319         }
320         return res;
321 }
322
323
324 #ifdef CONFIG_KALLSYMS
325 /*
326  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
327  * Returns the resolved symbol.  If that fails, simply return the address.
328  */
329 static int proc_pid_wchan(struct task_struct *task, char *buffer)
330 {
331         unsigned long wchan;
332         char symname[KSYM_NAME_LEN];
333
334         wchan = get_wchan(task);
335
336         if (lookup_symbol_name(wchan, symname) < 0)
337                 return sprintf(buffer, "%lu", wchan);
338         else
339                 return sprintf(buffer, "%s", symname);
340 }
341 #endif /* CONFIG_KALLSYMS */
342
343 #ifdef CONFIG_SCHEDSTATS
344 /*
345  * Provides /proc/PID/schedstat
346  */
347 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
348 {
349         return sprintf(buffer, "%llu %llu %lu\n",
350                         task->sched_info.cpu_time,
351                         task->sched_info.run_delay,
352                         task->sched_info.pcount);
353 }
354 #endif
355
356 #ifdef CONFIG_LATENCYTOP
357 static int lstats_show_proc(struct seq_file *m, void *v)
358 {
359         int i;
360         struct inode *inode = m->private;
361         struct task_struct *task = get_proc_task(inode);
362
363         if (!task)
364                 return -ESRCH;
365         seq_puts(m, "Latency Top version : v0.1\n");
366         for (i = 0; i < 32; i++) {
367                 if (task->latency_record[i].backtrace[0]) {
368                         int q;
369                         seq_printf(m, "%i %li %li ",
370                                 task->latency_record[i].count,
371                                 task->latency_record[i].time,
372                                 task->latency_record[i].max);
373                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
374                                 char sym[KSYM_NAME_LEN];
375                                 char *c;
376                                 if (!task->latency_record[i].backtrace[q])
377                                         break;
378                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
379                                         break;
380                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
381                                 c = strchr(sym, '+');
382                                 if (c)
383                                         *c = 0;
384                                 seq_printf(m, "%s ", sym);
385                         }
386                         seq_printf(m, "\n");
387                 }
388
389         }
390         put_task_struct(task);
391         return 0;
392 }
393
394 static int lstats_open(struct inode *inode, struct file *file)
395 {
396         return single_open(file, lstats_show_proc, inode);
397 }
398
399 static ssize_t lstats_write(struct file *file, const char __user *buf,
400                             size_t count, loff_t *offs)
401 {
402         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
403
404         if (!task)
405                 return -ESRCH;
406         clear_all_latency_tracing(task);
407         put_task_struct(task);
408
409         return count;
410 }
411
412 static const struct file_operations proc_lstats_operations = {
413         .open           = lstats_open,
414         .read           = seq_read,
415         .write          = lstats_write,
416         .llseek         = seq_lseek,
417         .release        = single_release,
418 };
419
420 #endif
421
422 /* The badness from the OOM killer */
423 unsigned long badness(struct task_struct *p, unsigned long uptime);
424 static int proc_oom_score(struct task_struct *task, char *buffer)
425 {
426         unsigned long points;
427         struct timespec uptime;
428
429         do_posix_clock_monotonic_gettime(&uptime);
430         read_lock(&tasklist_lock);
431         points = badness(task, uptime.tv_sec);
432         read_unlock(&tasklist_lock);
433         return sprintf(buffer, "%lu\n", points);
434 }
435
436 struct limit_names {
437         char *name;
438         char *unit;
439 };
440
441 static const struct limit_names lnames[RLIM_NLIMITS] = {
442         [RLIMIT_CPU] = {"Max cpu time", "ms"},
443         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
444         [RLIMIT_DATA] = {"Max data size", "bytes"},
445         [RLIMIT_STACK] = {"Max stack size", "bytes"},
446         [RLIMIT_CORE] = {"Max core file size", "bytes"},
447         [RLIMIT_RSS] = {"Max resident set", "bytes"},
448         [RLIMIT_NPROC] = {"Max processes", "processes"},
449         [RLIMIT_NOFILE] = {"Max open files", "files"},
450         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
451         [RLIMIT_AS] = {"Max address space", "bytes"},
452         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
453         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
454         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
455         [RLIMIT_NICE] = {"Max nice priority", NULL},
456         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
457         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
458 };
459
460 /* Display limits for a process */
461 static int proc_pid_limits(struct task_struct *task, char *buffer)
462 {
463         unsigned int i;
464         int count = 0;
465         unsigned long flags;
466         char *bufptr = buffer;
467
468         struct rlimit rlim[RLIM_NLIMITS];
469
470         if (!lock_task_sighand(task, &flags))
471                 return 0;
472         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
473         unlock_task_sighand(task, &flags);
474
475         /*
476          * print the file header
477          */
478         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
479                         "Limit", "Soft Limit", "Hard Limit", "Units");
480
481         for (i = 0; i < RLIM_NLIMITS; i++) {
482                 if (rlim[i].rlim_cur == RLIM_INFINITY)
483                         count += sprintf(&bufptr[count], "%-25s %-20s ",
484                                          lnames[i].name, "unlimited");
485                 else
486                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
487                                          lnames[i].name, rlim[i].rlim_cur);
488
489                 if (rlim[i].rlim_max == RLIM_INFINITY)
490                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
491                 else
492                         count += sprintf(&bufptr[count], "%-20lu ",
493                                          rlim[i].rlim_max);
494
495                 if (lnames[i].unit)
496                         count += sprintf(&bufptr[count], "%-10s\n",
497                                          lnames[i].unit);
498                 else
499                         count += sprintf(&bufptr[count], "\n");
500         }
501
502         return count;
503 }
504
505 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
506 static int proc_pid_syscall(struct task_struct *task, char *buffer)
507 {
508         long nr;
509         unsigned long args[6], sp, pc;
510
511         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
512                 return sprintf(buffer, "running\n");
513
514         if (nr < 0)
515                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
516
517         return sprintf(buffer,
518                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
519                        nr,
520                        args[0], args[1], args[2], args[3], args[4], args[5],
521                        sp, pc);
522 }
523 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
524
525 /************************************************************************/
526 /*                       Here the fs part begins                        */
527 /************************************************************************/
528
529 /* permission checks */
530 static int proc_fd_access_allowed(struct inode *inode)
531 {
532         struct task_struct *task;
533         int allowed = 0;
534         /* Allow access to a task's file descriptors if it is us or we
535          * may use ptrace attach to the process and find out that
536          * information.
537          */
538         task = get_proc_task(inode);
539         if (task) {
540                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
541                 put_task_struct(task);
542         }
543         return allowed;
544 }
545
546 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
547 {
548         int error;
549         struct inode *inode = dentry->d_inode;
550
551         if (attr->ia_valid & ATTR_MODE)
552                 return -EPERM;
553
554         error = inode_change_ok(inode, attr);
555         if (!error)
556                 error = inode_setattr(inode, attr);
557         return error;
558 }
559
560 static const struct inode_operations proc_def_inode_operations = {
561         .setattr        = proc_setattr,
562 };
563
564 static int mounts_open_common(struct inode *inode, struct file *file,
565                               const struct seq_operations *op)
566 {
567         struct task_struct *task = get_proc_task(inode);
568         struct nsproxy *nsp;
569         struct mnt_namespace *ns = NULL;
570         struct fs_struct *fs = NULL;
571         struct path root;
572         struct proc_mounts *p;
573         int ret = -EINVAL;
574
575         if (task) {
576                 rcu_read_lock();
577                 nsp = task_nsproxy(task);
578                 if (nsp) {
579                         ns = nsp->mnt_ns;
580                         if (ns)
581                                 get_mnt_ns(ns);
582                 }
583                 rcu_read_unlock();
584                 if (ns)
585                         fs = get_fs_struct(task);
586                 put_task_struct(task);
587         }
588
589         if (!ns)
590                 goto err;
591         if (!fs)
592                 goto err_put_ns;
593
594         read_lock(&fs->lock);
595         root = fs->root;
596         path_get(&root);
597         read_unlock(&fs->lock);
598         put_fs_struct(fs);
599
600         ret = -ENOMEM;
601         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
602         if (!p)
603                 goto err_put_path;
604
605         file->private_data = &p->m;
606         ret = seq_open(file, op);
607         if (ret)
608                 goto err_free;
609
610         p->m.private = p;
611         p->ns = ns;
612         p->root = root;
613         p->event = ns->event;
614
615         return 0;
616
617  err_free:
618         kfree(p);
619  err_put_path:
620         path_put(&root);
621  err_put_ns:
622         put_mnt_ns(ns);
623  err:
624         return ret;
625 }
626
627 static int mounts_release(struct inode *inode, struct file *file)
628 {
629         struct proc_mounts *p = file->private_data;
630         path_put(&p->root);
631         put_mnt_ns(p->ns);
632         return seq_release(inode, file);
633 }
634
635 static unsigned mounts_poll(struct file *file, poll_table *wait)
636 {
637         struct proc_mounts *p = file->private_data;
638         struct mnt_namespace *ns = p->ns;
639         unsigned res = 0;
640
641         poll_wait(file, &ns->poll, wait);
642
643         spin_lock(&vfsmount_lock);
644         if (p->event != ns->event) {
645                 p->event = ns->event;
646                 res = POLLERR;
647         }
648         spin_unlock(&vfsmount_lock);
649
650         return res;
651 }
652
653 static int mounts_open(struct inode *inode, struct file *file)
654 {
655         return mounts_open_common(inode, file, &mounts_op);
656 }
657
658 static const struct file_operations proc_mounts_operations = {
659         .open           = mounts_open,
660         .read           = seq_read,
661         .llseek         = seq_lseek,
662         .release        = mounts_release,
663         .poll           = mounts_poll,
664 };
665
666 static int mountinfo_open(struct inode *inode, struct file *file)
667 {
668         return mounts_open_common(inode, file, &mountinfo_op);
669 }
670
671 static const struct file_operations proc_mountinfo_operations = {
672         .open           = mountinfo_open,
673         .read           = seq_read,
674         .llseek         = seq_lseek,
675         .release        = mounts_release,
676         .poll           = mounts_poll,
677 };
678
679 static int mountstats_open(struct inode *inode, struct file *file)
680 {
681         return mounts_open_common(inode, file, &mountstats_op);
682 }
683
684 static const struct file_operations proc_mountstats_operations = {
685         .open           = mountstats_open,
686         .read           = seq_read,
687         .llseek         = seq_lseek,
688         .release        = mounts_release,
689 };
690
691 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
692
693 static ssize_t proc_info_read(struct file * file, char __user * buf,
694                           size_t count, loff_t *ppos)
695 {
696         struct inode * inode = file->f_path.dentry->d_inode;
697         unsigned long page;
698         ssize_t length;
699         struct task_struct *task = get_proc_task(inode);
700
701         length = -ESRCH;
702         if (!task)
703                 goto out_no_task;
704
705         if (count > PROC_BLOCK_SIZE)
706                 count = PROC_BLOCK_SIZE;
707
708         length = -ENOMEM;
709         if (!(page = __get_free_page(GFP_TEMPORARY)))
710                 goto out;
711
712         length = PROC_I(inode)->op.proc_read(task, (char*)page);
713
714         if (length >= 0)
715                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
716         free_page(page);
717 out:
718         put_task_struct(task);
719 out_no_task:
720         return length;
721 }
722
723 static const struct file_operations proc_info_file_operations = {
724         .read           = proc_info_read,
725 };
726
727 static int proc_single_show(struct seq_file *m, void *v)
728 {
729         struct inode *inode = m->private;
730         struct pid_namespace *ns;
731         struct pid *pid;
732         struct task_struct *task;
733         int ret;
734
735         ns = inode->i_sb->s_fs_info;
736         pid = proc_pid(inode);
737         task = get_pid_task(pid, PIDTYPE_PID);
738         if (!task)
739                 return -ESRCH;
740
741         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
742
743         put_task_struct(task);
744         return ret;
745 }
746
747 static int proc_single_open(struct inode *inode, struct file *filp)
748 {
749         int ret;
750         ret = single_open(filp, proc_single_show, NULL);
751         if (!ret) {
752                 struct seq_file *m = filp->private_data;
753
754                 m->private = inode;
755         }
756         return ret;
757 }
758
759 static const struct file_operations proc_single_file_operations = {
760         .open           = proc_single_open,
761         .read           = seq_read,
762         .llseek         = seq_lseek,
763         .release        = single_release,
764 };
765
766 static int mem_open(struct inode* inode, struct file* file)
767 {
768         file->private_data = (void*)((long)current->self_exec_id);
769         return 0;
770 }
771
772 static ssize_t mem_read(struct file * file, char __user * buf,
773                         size_t count, loff_t *ppos)
774 {
775         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
776         char *page;
777         unsigned long src = *ppos;
778         int ret = -ESRCH;
779         struct mm_struct *mm;
780
781         if (!task)
782                 goto out_no_task;
783
784         if (check_mem_permission(task))
785                 goto out;
786
787         ret = -ENOMEM;
788         page = (char *)__get_free_page(GFP_TEMPORARY);
789         if (!page)
790                 goto out;
791
792         ret = 0;
793  
794         mm = get_task_mm(task);
795         if (!mm)
796                 goto out_free;
797
798         ret = -EIO;
799  
800         if (file->private_data != (void*)((long)current->self_exec_id))
801                 goto out_put;
802
803         ret = 0;
804  
805         while (count > 0) {
806                 int this_len, retval;
807
808                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
809                 retval = access_process_vm(task, src, page, this_len, 0);
810                 if (!retval || check_mem_permission(task)) {
811                         if (!ret)
812                                 ret = -EIO;
813                         break;
814                 }
815
816                 if (copy_to_user(buf, page, retval)) {
817                         ret = -EFAULT;
818                         break;
819                 }
820  
821                 ret += retval;
822                 src += retval;
823                 buf += retval;
824                 count -= retval;
825         }
826         *ppos = src;
827
828 out_put:
829         mmput(mm);
830 out_free:
831         free_page((unsigned long) page);
832 out:
833         put_task_struct(task);
834 out_no_task:
835         return ret;
836 }
837
838 #define mem_write NULL
839
840 #ifndef mem_write
841 /* This is a security hazard */
842 static ssize_t mem_write(struct file * file, const char __user *buf,
843                          size_t count, loff_t *ppos)
844 {
845         int copied;
846         char *page;
847         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
848         unsigned long dst = *ppos;
849
850         copied = -ESRCH;
851         if (!task)
852                 goto out_no_task;
853
854         if (check_mem_permission(task))
855                 goto out;
856
857         copied = -ENOMEM;
858         page = (char *)__get_free_page(GFP_TEMPORARY);
859         if (!page)
860                 goto out;
861
862         copied = 0;
863         while (count > 0) {
864                 int this_len, retval;
865
866                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
867                 if (copy_from_user(page, buf, this_len)) {
868                         copied = -EFAULT;
869                         break;
870                 }
871                 retval = access_process_vm(task, dst, page, this_len, 1);
872                 if (!retval) {
873                         if (!copied)
874                                 copied = -EIO;
875                         break;
876                 }
877                 copied += retval;
878                 buf += retval;
879                 dst += retval;
880                 count -= retval;                        
881         }
882         *ppos = dst;
883         free_page((unsigned long) page);
884 out:
885         put_task_struct(task);
886 out_no_task:
887         return copied;
888 }
889 #endif
890
891 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
892 {
893         switch (orig) {
894         case 0:
895                 file->f_pos = offset;
896                 break;
897         case 1:
898                 file->f_pos += offset;
899                 break;
900         default:
901                 return -EINVAL;
902         }
903         force_successful_syscall_return();
904         return file->f_pos;
905 }
906
907 static const struct file_operations proc_mem_operations = {
908         .llseek         = mem_lseek,
909         .read           = mem_read,
910         .write          = mem_write,
911         .open           = mem_open,
912 };
913
914 static ssize_t environ_read(struct file *file, char __user *buf,
915                         size_t count, loff_t *ppos)
916 {
917         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
918         char *page;
919         unsigned long src = *ppos;
920         int ret = -ESRCH;
921         struct mm_struct *mm;
922
923         if (!task)
924                 goto out_no_task;
925
926         if (!ptrace_may_access(task, PTRACE_MODE_READ))
927                 goto out;
928
929         ret = -ENOMEM;
930         page = (char *)__get_free_page(GFP_TEMPORARY);
931         if (!page)
932                 goto out;
933
934         ret = 0;
935
936         mm = get_task_mm(task);
937         if (!mm)
938                 goto out_free;
939
940         while (count > 0) {
941                 int this_len, retval, max_len;
942
943                 this_len = mm->env_end - (mm->env_start + src);
944
945                 if (this_len <= 0)
946                         break;
947
948                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
949                 this_len = (this_len > max_len) ? max_len : this_len;
950
951                 retval = access_process_vm(task, (mm->env_start + src),
952                         page, this_len, 0);
953
954                 if (retval <= 0) {
955                         ret = retval;
956                         break;
957                 }
958
959                 if (copy_to_user(buf, page, retval)) {
960                         ret = -EFAULT;
961                         break;
962                 }
963
964                 ret += retval;
965                 src += retval;
966                 buf += retval;
967                 count -= retval;
968         }
969         *ppos = src;
970
971         mmput(mm);
972 out_free:
973         free_page((unsigned long) page);
974 out:
975         put_task_struct(task);
976 out_no_task:
977         return ret;
978 }
979
980 static const struct file_operations proc_environ_operations = {
981         .read           = environ_read,
982 };
983
984 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
985                                 size_t count, loff_t *ppos)
986 {
987         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
988         char buffer[PROC_NUMBUF];
989         size_t len;
990         int oom_adjust;
991
992         if (!task)
993                 return -ESRCH;
994         oom_adjust = task->oomkilladj;
995         put_task_struct(task);
996
997         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
998
999         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1000 }
1001
1002 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1003                                 size_t count, loff_t *ppos)
1004 {
1005         struct task_struct *task;
1006         char buffer[PROC_NUMBUF], *end;
1007         int oom_adjust;
1008
1009         memset(buffer, 0, sizeof(buffer));
1010         if (count > sizeof(buffer) - 1)
1011                 count = sizeof(buffer) - 1;
1012         if (copy_from_user(buffer, buf, count))
1013                 return -EFAULT;
1014         oom_adjust = simple_strtol(buffer, &end, 0);
1015         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1016              oom_adjust != OOM_DISABLE)
1017                 return -EINVAL;
1018         if (*end == '\n')
1019                 end++;
1020         task = get_proc_task(file->f_path.dentry->d_inode);
1021         if (!task)
1022                 return -ESRCH;
1023         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1024                 put_task_struct(task);
1025                 return -EACCES;
1026         }
1027         task->oomkilladj = oom_adjust;
1028         put_task_struct(task);
1029         if (end - buffer == 0)
1030                 return -EIO;
1031         return end - buffer;
1032 }
1033
1034 static const struct file_operations proc_oom_adjust_operations = {
1035         .read           = oom_adjust_read,
1036         .write          = oom_adjust_write,
1037 };
1038
1039 #ifdef CONFIG_AUDITSYSCALL
1040 #define TMPBUFLEN 21
1041 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1042                                   size_t count, loff_t *ppos)
1043 {
1044         struct inode * inode = file->f_path.dentry->d_inode;
1045         struct task_struct *task = get_proc_task(inode);
1046         ssize_t length;
1047         char tmpbuf[TMPBUFLEN];
1048
1049         if (!task)
1050                 return -ESRCH;
1051         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1052                                 audit_get_loginuid(task));
1053         put_task_struct(task);
1054         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1055 }
1056
1057 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1058                                    size_t count, loff_t *ppos)
1059 {
1060         struct inode * inode = file->f_path.dentry->d_inode;
1061         char *page, *tmp;
1062         ssize_t length;
1063         uid_t loginuid;
1064
1065         if (!capable(CAP_AUDIT_CONTROL))
1066                 return -EPERM;
1067
1068         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1069                 return -EPERM;
1070
1071         if (count >= PAGE_SIZE)
1072                 count = PAGE_SIZE - 1;
1073
1074         if (*ppos != 0) {
1075                 /* No partial writes. */
1076                 return -EINVAL;
1077         }
1078         page = (char*)__get_free_page(GFP_TEMPORARY);
1079         if (!page)
1080                 return -ENOMEM;
1081         length = -EFAULT;
1082         if (copy_from_user(page, buf, count))
1083                 goto out_free_page;
1084
1085         page[count] = '\0';
1086         loginuid = simple_strtoul(page, &tmp, 10);
1087         if (tmp == page) {
1088                 length = -EINVAL;
1089                 goto out_free_page;
1090
1091         }
1092         length = audit_set_loginuid(current, loginuid);
1093         if (likely(length == 0))
1094                 length = count;
1095
1096 out_free_page:
1097         free_page((unsigned long) page);
1098         return length;
1099 }
1100
1101 static const struct file_operations proc_loginuid_operations = {
1102         .read           = proc_loginuid_read,
1103         .write          = proc_loginuid_write,
1104 };
1105
1106 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1107                                   size_t count, loff_t *ppos)
1108 {
1109         struct inode * inode = file->f_path.dentry->d_inode;
1110         struct task_struct *task = get_proc_task(inode);
1111         ssize_t length;
1112         char tmpbuf[TMPBUFLEN];
1113
1114         if (!task)
1115                 return -ESRCH;
1116         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1117                                 audit_get_sessionid(task));
1118         put_task_struct(task);
1119         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1120 }
1121
1122 static const struct file_operations proc_sessionid_operations = {
1123         .read           = proc_sessionid_read,
1124 };
1125 #endif
1126
1127 #ifdef CONFIG_FAULT_INJECTION
1128 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1129                                       size_t count, loff_t *ppos)
1130 {
1131         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1132         char buffer[PROC_NUMBUF];
1133         size_t len;
1134         int make_it_fail;
1135
1136         if (!task)
1137                 return -ESRCH;
1138         make_it_fail = task->make_it_fail;
1139         put_task_struct(task);
1140
1141         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1142
1143         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1144 }
1145
1146 static ssize_t proc_fault_inject_write(struct file * file,
1147                         const char __user * buf, size_t count, loff_t *ppos)
1148 {
1149         struct task_struct *task;
1150         char buffer[PROC_NUMBUF], *end;
1151         int make_it_fail;
1152
1153         if (!capable(CAP_SYS_RESOURCE))
1154                 return -EPERM;
1155         memset(buffer, 0, sizeof(buffer));
1156         if (count > sizeof(buffer) - 1)
1157                 count = sizeof(buffer) - 1;
1158         if (copy_from_user(buffer, buf, count))
1159                 return -EFAULT;
1160         make_it_fail = simple_strtol(buffer, &end, 0);
1161         if (*end == '\n')
1162                 end++;
1163         task = get_proc_task(file->f_dentry->d_inode);
1164         if (!task)
1165                 return -ESRCH;
1166         task->make_it_fail = make_it_fail;
1167         put_task_struct(task);
1168         if (end - buffer == 0)
1169                 return -EIO;
1170         return end - buffer;
1171 }
1172
1173 static const struct file_operations proc_fault_inject_operations = {
1174         .read           = proc_fault_inject_read,
1175         .write          = proc_fault_inject_write,
1176 };
1177 #endif
1178
1179
1180 #ifdef CONFIG_SCHED_DEBUG
1181 /*
1182  * Print out various scheduling related per-task fields:
1183  */
1184 static int sched_show(struct seq_file *m, void *v)
1185 {
1186         struct inode *inode = m->private;
1187         struct task_struct *p;
1188
1189         WARN_ON(!inode);
1190
1191         p = get_proc_task(inode);
1192         if (!p)
1193                 return -ESRCH;
1194         proc_sched_show_task(p, m);
1195
1196         put_task_struct(p);
1197
1198         return 0;
1199 }
1200
1201 static ssize_t
1202 sched_write(struct file *file, const char __user *buf,
1203             size_t count, loff_t *offset)
1204 {
1205         struct inode *inode = file->f_path.dentry->d_inode;
1206         struct task_struct *p;
1207
1208         WARN_ON(!inode);
1209
1210         p = get_proc_task(inode);
1211         if (!p)
1212                 return -ESRCH;
1213         proc_sched_set_task(p);
1214
1215         put_task_struct(p);
1216
1217         return count;
1218 }
1219
1220 static int sched_open(struct inode *inode, struct file *filp)
1221 {
1222         int ret;
1223
1224         ret = single_open(filp, sched_show, NULL);
1225         if (!ret) {
1226                 struct seq_file *m = filp->private_data;
1227
1228                 m->private = inode;
1229         }
1230         return ret;
1231 }
1232
1233 static const struct file_operations proc_pid_sched_operations = {
1234         .open           = sched_open,
1235         .read           = seq_read,
1236         .write          = sched_write,
1237         .llseek         = seq_lseek,
1238         .release        = single_release,
1239 };
1240
1241 #endif
1242
1243 /*
1244  * We added or removed a vma mapping the executable. The vmas are only mapped
1245  * during exec and are not mapped with the mmap system call.
1246  * Callers must hold down_write() on the mm's mmap_sem for these
1247  */
1248 void added_exe_file_vma(struct mm_struct *mm)
1249 {
1250         mm->num_exe_file_vmas++;
1251 }
1252
1253 void removed_exe_file_vma(struct mm_struct *mm)
1254 {
1255         mm->num_exe_file_vmas--;
1256         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1257                 fput(mm->exe_file);
1258                 mm->exe_file = NULL;
1259         }
1260
1261 }
1262
1263 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1264 {
1265         if (new_exe_file)
1266                 get_file(new_exe_file);
1267         if (mm->exe_file)
1268                 fput(mm->exe_file);
1269         mm->exe_file = new_exe_file;
1270         mm->num_exe_file_vmas = 0;
1271 }
1272
1273 struct file *get_mm_exe_file(struct mm_struct *mm)
1274 {
1275         struct file *exe_file;
1276
1277         /* We need mmap_sem to protect against races with removal of
1278          * VM_EXECUTABLE vmas */
1279         down_read(&mm->mmap_sem);
1280         exe_file = mm->exe_file;
1281         if (exe_file)
1282                 get_file(exe_file);
1283         up_read(&mm->mmap_sem);
1284         return exe_file;
1285 }
1286
1287 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1288 {
1289         /* It's safe to write the exe_file pointer without exe_file_lock because
1290          * this is called during fork when the task is not yet in /proc */
1291         newmm->exe_file = get_mm_exe_file(oldmm);
1292 }
1293
1294 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1295 {
1296         struct task_struct *task;
1297         struct mm_struct *mm;
1298         struct file *exe_file;
1299
1300         task = get_proc_task(inode);
1301         if (!task)
1302                 return -ENOENT;
1303         mm = get_task_mm(task);
1304         put_task_struct(task);
1305         if (!mm)
1306                 return -ENOENT;
1307         exe_file = get_mm_exe_file(mm);
1308         mmput(mm);
1309         if (exe_file) {
1310                 *exe_path = exe_file->f_path;
1311                 path_get(&exe_file->f_path);
1312                 fput(exe_file);
1313                 return 0;
1314         } else
1315                 return -ENOENT;
1316 }
1317
1318 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1319 {
1320         struct inode *inode = dentry->d_inode;
1321         int error = -EACCES;
1322
1323         /* We don't need a base pointer in the /proc filesystem */
1324         path_put(&nd->path);
1325
1326         /* Are we allowed to snoop on the tasks file descriptors? */
1327         if (!proc_fd_access_allowed(inode))
1328                 goto out;
1329
1330         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1331         nd->last_type = LAST_BIND;
1332 out:
1333         return ERR_PTR(error);
1334 }
1335
1336 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1337 {
1338         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1339         char *pathname;
1340         int len;
1341
1342         if (!tmp)
1343                 return -ENOMEM;
1344
1345         pathname = d_path(path, tmp, PAGE_SIZE);
1346         len = PTR_ERR(pathname);
1347         if (IS_ERR(pathname))
1348                 goto out;
1349         len = tmp + PAGE_SIZE - 1 - pathname;
1350
1351         if (len > buflen)
1352                 len = buflen;
1353         if (copy_to_user(buffer, pathname, len))
1354                 len = -EFAULT;
1355  out:
1356         free_page((unsigned long)tmp);
1357         return len;
1358 }
1359
1360 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1361 {
1362         int error = -EACCES;
1363         struct inode *inode = dentry->d_inode;
1364         struct path path;
1365
1366         /* Are we allowed to snoop on the tasks file descriptors? */
1367         if (!proc_fd_access_allowed(inode))
1368                 goto out;
1369
1370         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1371         if (error)
1372                 goto out;
1373
1374         error = do_proc_readlink(&path, buffer, buflen);
1375         path_put(&path);
1376 out:
1377         return error;
1378 }
1379
1380 static const struct inode_operations proc_pid_link_inode_operations = {
1381         .readlink       = proc_pid_readlink,
1382         .follow_link    = proc_pid_follow_link,
1383         .setattr        = proc_setattr,
1384 };
1385
1386
1387 /* building an inode */
1388
1389 static int task_dumpable(struct task_struct *task)
1390 {
1391         int dumpable = 0;
1392         struct mm_struct *mm;
1393
1394         task_lock(task);
1395         mm = task->mm;
1396         if (mm)
1397                 dumpable = get_dumpable(mm);
1398         task_unlock(task);
1399         if(dumpable == 1)
1400                 return 1;
1401         return 0;
1402 }
1403
1404
1405 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1406 {
1407         struct inode * inode;
1408         struct proc_inode *ei;
1409
1410         /* We need a new inode */
1411
1412         inode = new_inode(sb);
1413         if (!inode)
1414                 goto out;
1415
1416         /* Common stuff */
1417         ei = PROC_I(inode);
1418         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1419         inode->i_op = &proc_def_inode_operations;
1420
1421         /*
1422          * grab the reference to task.
1423          */
1424         ei->pid = get_task_pid(task, PIDTYPE_PID);
1425         if (!ei->pid)
1426                 goto out_unlock;
1427
1428         inode->i_uid = 0;
1429         inode->i_gid = 0;
1430         if (task_dumpable(task)) {
1431                 inode->i_uid = task->euid;
1432                 inode->i_gid = task->egid;
1433         }
1434         security_task_to_inode(task, inode);
1435
1436 out:
1437         return inode;
1438
1439 out_unlock:
1440         iput(inode);
1441         return NULL;
1442 }
1443
1444 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1445 {
1446         struct inode *inode = dentry->d_inode;
1447         struct task_struct *task;
1448         generic_fillattr(inode, stat);
1449
1450         rcu_read_lock();
1451         stat->uid = 0;
1452         stat->gid = 0;
1453         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1454         if (task) {
1455                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1456                     task_dumpable(task)) {
1457                         stat->uid = task->euid;
1458                         stat->gid = task->egid;
1459                 }
1460         }
1461         rcu_read_unlock();
1462         return 0;
1463 }
1464
1465 /* dentry stuff */
1466
1467 /*
1468  *      Exceptional case: normally we are not allowed to unhash a busy
1469  * directory. In this case, however, we can do it - no aliasing problems
1470  * due to the way we treat inodes.
1471  *
1472  * Rewrite the inode's ownerships here because the owning task may have
1473  * performed a setuid(), etc.
1474  *
1475  * Before the /proc/pid/status file was created the only way to read
1476  * the effective uid of a /process was to stat /proc/pid.  Reading
1477  * /proc/pid/status is slow enough that procps and other packages
1478  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1479  * made this apply to all per process world readable and executable
1480  * directories.
1481  */
1482 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1483 {
1484         struct inode *inode = dentry->d_inode;
1485         struct task_struct *task = get_proc_task(inode);
1486         if (task) {
1487                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1488                     task_dumpable(task)) {
1489                         inode->i_uid = task->euid;
1490                         inode->i_gid = task->egid;
1491                 } else {
1492                         inode->i_uid = 0;
1493                         inode->i_gid = 0;
1494                 }
1495                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1496                 security_task_to_inode(task, inode);
1497                 put_task_struct(task);
1498                 return 1;
1499         }
1500         d_drop(dentry);
1501         return 0;
1502 }
1503
1504 static int pid_delete_dentry(struct dentry * dentry)
1505 {
1506         /* Is the task we represent dead?
1507          * If so, then don't put the dentry on the lru list,
1508          * kill it immediately.
1509          */
1510         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1511 }
1512
1513 static struct dentry_operations pid_dentry_operations =
1514 {
1515         .d_revalidate   = pid_revalidate,
1516         .d_delete       = pid_delete_dentry,
1517 };
1518
1519 /* Lookups */
1520
1521 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1522                                 struct task_struct *, const void *);
1523
1524 /*
1525  * Fill a directory entry.
1526  *
1527  * If possible create the dcache entry and derive our inode number and
1528  * file type from dcache entry.
1529  *
1530  * Since all of the proc inode numbers are dynamically generated, the inode
1531  * numbers do not exist until the inode is cache.  This means creating the
1532  * the dcache entry in readdir is necessary to keep the inode numbers
1533  * reported by readdir in sync with the inode numbers reported
1534  * by stat.
1535  */
1536 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1537         char *name, int len,
1538         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1539 {
1540         struct dentry *child, *dir = filp->f_path.dentry;
1541         struct inode *inode;
1542         struct qstr qname;
1543         ino_t ino = 0;
1544         unsigned type = DT_UNKNOWN;
1545
1546         qname.name = name;
1547         qname.len  = len;
1548         qname.hash = full_name_hash(name, len);
1549
1550         child = d_lookup(dir, &qname);
1551         if (!child) {
1552                 struct dentry *new;
1553                 new = d_alloc(dir, &qname);
1554                 if (new) {
1555                         child = instantiate(dir->d_inode, new, task, ptr);
1556                         if (child)
1557                                 dput(new);
1558                         else
1559                                 child = new;
1560                 }
1561         }
1562         if (!child || IS_ERR(child) || !child->d_inode)
1563                 goto end_instantiate;
1564         inode = child->d_inode;
1565         if (inode) {
1566                 ino = inode->i_ino;
1567                 type = inode->i_mode >> 12;
1568         }
1569         dput(child);
1570 end_instantiate:
1571         if (!ino)
1572                 ino = find_inode_number(dir, &qname);
1573         if (!ino)
1574                 ino = 1;
1575         return filldir(dirent, name, len, filp->f_pos, ino, type);
1576 }
1577
1578 static unsigned name_to_int(struct dentry *dentry)
1579 {
1580         const char *name = dentry->d_name.name;
1581         int len = dentry->d_name.len;
1582         unsigned n = 0;
1583
1584         if (len > 1 && *name == '0')
1585                 goto out;
1586         while (len-- > 0) {
1587                 unsigned c = *name++ - '0';
1588                 if (c > 9)
1589                         goto out;
1590                 if (n >= (~0U-9)/10)
1591                         goto out;
1592                 n *= 10;
1593                 n += c;
1594         }
1595         return n;
1596 out:
1597         return ~0U;
1598 }
1599
1600 #define PROC_FDINFO_MAX 64
1601
1602 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1603 {
1604         struct task_struct *task = get_proc_task(inode);
1605         struct files_struct *files = NULL;
1606         struct file *file;
1607         int fd = proc_fd(inode);
1608
1609         if (task) {
1610                 files = get_files_struct(task);
1611                 put_task_struct(task);
1612         }
1613         if (files) {
1614                 /*
1615                  * We are not taking a ref to the file structure, so we must
1616                  * hold ->file_lock.
1617                  */
1618                 spin_lock(&files->file_lock);
1619                 file = fcheck_files(files, fd);
1620                 if (file) {
1621                         if (path) {
1622                                 *path = file->f_path;
1623                                 path_get(&file->f_path);
1624                         }
1625                         if (info)
1626                                 snprintf(info, PROC_FDINFO_MAX,
1627                                          "pos:\t%lli\n"
1628                                          "flags:\t0%o\n",
1629                                          (long long) file->f_pos,
1630                                          file->f_flags);
1631                         spin_unlock(&files->file_lock);
1632                         put_files_struct(files);
1633                         return 0;
1634                 }
1635                 spin_unlock(&files->file_lock);
1636                 put_files_struct(files);
1637         }
1638         return -ENOENT;
1639 }
1640
1641 static int proc_fd_link(struct inode *inode, struct path *path)
1642 {
1643         return proc_fd_info(inode, path, NULL);
1644 }
1645
1646 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1647 {
1648         struct inode *inode = dentry->d_inode;
1649         struct task_struct *task = get_proc_task(inode);
1650         int fd = proc_fd(inode);
1651         struct files_struct *files;
1652
1653         if (task) {
1654                 files = get_files_struct(task);
1655                 if (files) {
1656                         rcu_read_lock();
1657                         if (fcheck_files(files, fd)) {
1658                                 rcu_read_unlock();
1659                                 put_files_struct(files);
1660                                 if (task_dumpable(task)) {
1661                                         inode->i_uid = task->euid;
1662                                         inode->i_gid = task->egid;
1663                                 } else {
1664                                         inode->i_uid = 0;
1665                                         inode->i_gid = 0;
1666                                 }
1667                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1668                                 security_task_to_inode(task, inode);
1669                                 put_task_struct(task);
1670                                 return 1;
1671                         }
1672                         rcu_read_unlock();
1673                         put_files_struct(files);
1674                 }
1675                 put_task_struct(task);
1676         }
1677         d_drop(dentry);
1678         return 0;
1679 }
1680
1681 static struct dentry_operations tid_fd_dentry_operations =
1682 {
1683         .d_revalidate   = tid_fd_revalidate,
1684         .d_delete       = pid_delete_dentry,
1685 };
1686
1687 static struct dentry *proc_fd_instantiate(struct inode *dir,
1688         struct dentry *dentry, struct task_struct *task, const void *ptr)
1689 {
1690         unsigned fd = *(const unsigned *)ptr;
1691         struct file *file;
1692         struct files_struct *files;
1693         struct inode *inode;
1694         struct proc_inode *ei;
1695         struct dentry *error = ERR_PTR(-ENOENT);
1696
1697         inode = proc_pid_make_inode(dir->i_sb, task);
1698         if (!inode)
1699                 goto out;
1700         ei = PROC_I(inode);
1701         ei->fd = fd;
1702         files = get_files_struct(task);
1703         if (!files)
1704                 goto out_iput;
1705         inode->i_mode = S_IFLNK;
1706
1707         /*
1708          * We are not taking a ref to the file structure, so we must
1709          * hold ->file_lock.
1710          */
1711         spin_lock(&files->file_lock);
1712         file = fcheck_files(files, fd);
1713         if (!file)
1714                 goto out_unlock;
1715         if (file->f_mode & 1)
1716                 inode->i_mode |= S_IRUSR | S_IXUSR;
1717         if (file->f_mode & 2)
1718                 inode->i_mode |= S_IWUSR | S_IXUSR;
1719         spin_unlock(&files->file_lock);
1720         put_files_struct(files);
1721
1722         inode->i_op = &proc_pid_link_inode_operations;
1723         inode->i_size = 64;
1724         ei->op.proc_get_link = proc_fd_link;
1725         dentry->d_op = &tid_fd_dentry_operations;
1726         d_add(dentry, inode);
1727         /* Close the race of the process dying before we return the dentry */
1728         if (tid_fd_revalidate(dentry, NULL))
1729                 error = NULL;
1730
1731  out:
1732         return error;
1733 out_unlock:
1734         spin_unlock(&files->file_lock);
1735         put_files_struct(files);
1736 out_iput:
1737         iput(inode);
1738         goto out;
1739 }
1740
1741 static struct dentry *proc_lookupfd_common(struct inode *dir,
1742                                            struct dentry *dentry,
1743                                            instantiate_t instantiate)
1744 {
1745         struct task_struct *task = get_proc_task(dir);
1746         unsigned fd = name_to_int(dentry);
1747         struct dentry *result = ERR_PTR(-ENOENT);
1748
1749         if (!task)
1750                 goto out_no_task;
1751         if (fd == ~0U)
1752                 goto out;
1753
1754         result = instantiate(dir, dentry, task, &fd);
1755 out:
1756         put_task_struct(task);
1757 out_no_task:
1758         return result;
1759 }
1760
1761 static int proc_readfd_common(struct file * filp, void * dirent,
1762                               filldir_t filldir, instantiate_t instantiate)
1763 {
1764         struct dentry *dentry = filp->f_path.dentry;
1765         struct inode *inode = dentry->d_inode;
1766         struct task_struct *p = get_proc_task(inode);
1767         unsigned int fd, ino;
1768         int retval;
1769         struct files_struct * files;
1770
1771         retval = -ENOENT;
1772         if (!p)
1773                 goto out_no_task;
1774         retval = 0;
1775
1776         fd = filp->f_pos;
1777         switch (fd) {
1778                 case 0:
1779                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1780                                 goto out;
1781                         filp->f_pos++;
1782                 case 1:
1783                         ino = parent_ino(dentry);
1784                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1785                                 goto out;
1786                         filp->f_pos++;
1787                 default:
1788                         files = get_files_struct(p);
1789                         if (!files)
1790                                 goto out;
1791                         rcu_read_lock();
1792                         for (fd = filp->f_pos-2;
1793                              fd < files_fdtable(files)->max_fds;
1794                              fd++, filp->f_pos++) {
1795                                 char name[PROC_NUMBUF];
1796                                 int len;
1797
1798                                 if (!fcheck_files(files, fd))
1799                                         continue;
1800                                 rcu_read_unlock();
1801
1802                                 len = snprintf(name, sizeof(name), "%d", fd);
1803                                 if (proc_fill_cache(filp, dirent, filldir,
1804                                                     name, len, instantiate,
1805                                                     p, &fd) < 0) {
1806                                         rcu_read_lock();
1807                                         break;
1808                                 }
1809                                 rcu_read_lock();
1810                         }
1811                         rcu_read_unlock();
1812                         put_files_struct(files);
1813         }
1814 out:
1815         put_task_struct(p);
1816 out_no_task:
1817         return retval;
1818 }
1819
1820 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1821                                     struct nameidata *nd)
1822 {
1823         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1824 }
1825
1826 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1827 {
1828         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1829 }
1830
1831 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1832                                       size_t len, loff_t *ppos)
1833 {
1834         char tmp[PROC_FDINFO_MAX];
1835         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1836         if (!err)
1837                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1838         return err;
1839 }
1840
1841 static const struct file_operations proc_fdinfo_file_operations = {
1842         .open           = nonseekable_open,
1843         .read           = proc_fdinfo_read,
1844 };
1845
1846 static const struct file_operations proc_fd_operations = {
1847         .read           = generic_read_dir,
1848         .readdir        = proc_readfd,
1849 };
1850
1851 /*
1852  * /proc/pid/fd needs a special permission handler so that a process can still
1853  * access /proc/self/fd after it has executed a setuid().
1854  */
1855 static int proc_fd_permission(struct inode *inode, int mask)
1856 {
1857         int rv;
1858
1859         rv = generic_permission(inode, mask, NULL);
1860         if (rv == 0)
1861                 return 0;
1862         if (task_pid(current) == proc_pid(inode))
1863                 rv = 0;
1864         return rv;
1865 }
1866
1867 /*
1868  * proc directories can do almost nothing..
1869  */
1870 static const struct inode_operations proc_fd_inode_operations = {
1871         .lookup         = proc_lookupfd,
1872         .permission     = proc_fd_permission,
1873         .setattr        = proc_setattr,
1874 };
1875
1876 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1877         struct dentry *dentry, struct task_struct *task, const void *ptr)
1878 {
1879         unsigned fd = *(unsigned *)ptr;
1880         struct inode *inode;
1881         struct proc_inode *ei;
1882         struct dentry *error = ERR_PTR(-ENOENT);
1883
1884         inode = proc_pid_make_inode(dir->i_sb, task);
1885         if (!inode)
1886                 goto out;
1887         ei = PROC_I(inode);
1888         ei->fd = fd;
1889         inode->i_mode = S_IFREG | S_IRUSR;
1890         inode->i_fop = &proc_fdinfo_file_operations;
1891         dentry->d_op = &tid_fd_dentry_operations;
1892         d_add(dentry, inode);
1893         /* Close the race of the process dying before we return the dentry */
1894         if (tid_fd_revalidate(dentry, NULL))
1895                 error = NULL;
1896
1897  out:
1898         return error;
1899 }
1900
1901 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1902                                         struct dentry *dentry,
1903                                         struct nameidata *nd)
1904 {
1905         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1906 }
1907
1908 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1909 {
1910         return proc_readfd_common(filp, dirent, filldir,
1911                                   proc_fdinfo_instantiate);
1912 }
1913
1914 static const struct file_operations proc_fdinfo_operations = {
1915         .read           = generic_read_dir,
1916         .readdir        = proc_readfdinfo,
1917 };
1918
1919 /*
1920  * proc directories can do almost nothing..
1921  */
1922 static const struct inode_operations proc_fdinfo_inode_operations = {
1923         .lookup         = proc_lookupfdinfo,
1924         .setattr        = proc_setattr,
1925 };
1926
1927
1928 static struct dentry *proc_pident_instantiate(struct inode *dir,
1929         struct dentry *dentry, struct task_struct *task, const void *ptr)
1930 {
1931         const struct pid_entry *p = ptr;
1932         struct inode *inode;
1933         struct proc_inode *ei;
1934         struct dentry *error = ERR_PTR(-EINVAL);
1935
1936         inode = proc_pid_make_inode(dir->i_sb, task);
1937         if (!inode)
1938                 goto out;
1939
1940         ei = PROC_I(inode);
1941         inode->i_mode = p->mode;
1942         if (S_ISDIR(inode->i_mode))
1943                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1944         if (p->iop)
1945                 inode->i_op = p->iop;
1946         if (p->fop)
1947                 inode->i_fop = p->fop;
1948         ei->op = p->op;
1949         dentry->d_op = &pid_dentry_operations;
1950         d_add(dentry, inode);
1951         /* Close the race of the process dying before we return the dentry */
1952         if (pid_revalidate(dentry, NULL))
1953                 error = NULL;
1954 out:
1955         return error;
1956 }
1957
1958 static struct dentry *proc_pident_lookup(struct inode *dir, 
1959                                          struct dentry *dentry,
1960                                          const struct pid_entry *ents,
1961                                          unsigned int nents)
1962 {
1963         struct inode *inode;
1964         struct dentry *error;
1965         struct task_struct *task = get_proc_task(dir);
1966         const struct pid_entry *p, *last;
1967
1968         error = ERR_PTR(-ENOENT);
1969         inode = NULL;
1970
1971         if (!task)
1972                 goto out_no_task;
1973
1974         /*
1975          * Yes, it does not scale. And it should not. Don't add
1976          * new entries into /proc/<tgid>/ without very good reasons.
1977          */
1978         last = &ents[nents - 1];
1979         for (p = ents; p <= last; p++) {
1980                 if (p->len != dentry->d_name.len)
1981                         continue;
1982                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1983                         break;
1984         }
1985         if (p > last)
1986                 goto out;
1987
1988         error = proc_pident_instantiate(dir, dentry, task, p);
1989 out:
1990         put_task_struct(task);
1991 out_no_task:
1992         return error;
1993 }
1994
1995 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1996         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1997 {
1998         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1999                                 proc_pident_instantiate, task, p);
2000 }
2001
2002 static int proc_pident_readdir(struct file *filp,
2003                 void *dirent, filldir_t filldir,
2004                 const struct pid_entry *ents, unsigned int nents)
2005 {
2006         int i;
2007         struct dentry *dentry = filp->f_path.dentry;
2008         struct inode *inode = dentry->d_inode;
2009         struct task_struct *task = get_proc_task(inode);
2010         const struct pid_entry *p, *last;
2011         ino_t ino;
2012         int ret;
2013
2014         ret = -ENOENT;
2015         if (!task)
2016                 goto out_no_task;
2017
2018         ret = 0;
2019         i = filp->f_pos;
2020         switch (i) {
2021         case 0:
2022                 ino = inode->i_ino;
2023                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2024                         goto out;
2025                 i++;
2026                 filp->f_pos++;
2027                 /* fall through */
2028         case 1:
2029                 ino = parent_ino(dentry);
2030                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2031                         goto out;
2032                 i++;
2033                 filp->f_pos++;
2034                 /* fall through */
2035         default:
2036                 i -= 2;
2037                 if (i >= nents) {
2038                         ret = 1;
2039                         goto out;
2040                 }
2041                 p = ents + i;
2042                 last = &ents[nents - 1];
2043                 while (p <= last) {
2044                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2045                                 goto out;
2046                         filp->f_pos++;
2047                         p++;
2048                 }
2049         }
2050
2051         ret = 1;
2052 out:
2053         put_task_struct(task);
2054 out_no_task:
2055         return ret;
2056 }
2057
2058 #ifdef CONFIG_SECURITY
2059 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2060                                   size_t count, loff_t *ppos)
2061 {
2062         struct inode * inode = file->f_path.dentry->d_inode;
2063         char *p = NULL;
2064         ssize_t length;
2065         struct task_struct *task = get_proc_task(inode);
2066
2067         if (!task)
2068                 return -ESRCH;
2069
2070         length = security_getprocattr(task,
2071                                       (char*)file->f_path.dentry->d_name.name,
2072                                       &p);
2073         put_task_struct(task);
2074         if (length > 0)
2075                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2076         kfree(p);
2077         return length;
2078 }
2079
2080 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2081                                    size_t count, loff_t *ppos)
2082 {
2083         struct inode * inode = file->f_path.dentry->d_inode;
2084         char *page;
2085         ssize_t length;
2086         struct task_struct *task = get_proc_task(inode);
2087
2088         length = -ESRCH;
2089         if (!task)
2090                 goto out_no_task;
2091         if (count > PAGE_SIZE)
2092                 count = PAGE_SIZE;
2093
2094         /* No partial writes. */
2095         length = -EINVAL;
2096         if (*ppos != 0)
2097                 goto out;
2098
2099         length = -ENOMEM;
2100         page = (char*)__get_free_page(GFP_TEMPORARY);
2101         if (!page)
2102                 goto out;
2103
2104         length = -EFAULT;
2105         if (copy_from_user(page, buf, count))
2106                 goto out_free;
2107
2108         length = security_setprocattr(task,
2109                                       (char*)file->f_path.dentry->d_name.name,
2110                                       (void*)page, count);
2111 out_free:
2112         free_page((unsigned long) page);
2113 out:
2114         put_task_struct(task);
2115 out_no_task:
2116         return length;
2117 }
2118
2119 static const struct file_operations proc_pid_attr_operations = {
2120         .read           = proc_pid_attr_read,
2121         .write          = proc_pid_attr_write,
2122 };
2123
2124 static const struct pid_entry attr_dir_stuff[] = {
2125         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
2126         REG("prev",       S_IRUGO,         pid_attr),
2127         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
2128         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
2129         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
2130         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
2131 };
2132
2133 static int proc_attr_dir_readdir(struct file * filp,
2134                              void * dirent, filldir_t filldir)
2135 {
2136         return proc_pident_readdir(filp,dirent,filldir,
2137                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2138 }
2139
2140 static const struct file_operations proc_attr_dir_operations = {
2141         .read           = generic_read_dir,
2142         .readdir        = proc_attr_dir_readdir,
2143 };
2144
2145 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2146                                 struct dentry *dentry, struct nameidata *nd)
2147 {
2148         return proc_pident_lookup(dir, dentry,
2149                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2150 }
2151
2152 static const struct inode_operations proc_attr_dir_inode_operations = {
2153         .lookup         = proc_attr_dir_lookup,
2154         .getattr        = pid_getattr,
2155         .setattr        = proc_setattr,
2156 };
2157
2158 #endif
2159
2160 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2161 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2162                                          size_t count, loff_t *ppos)
2163 {
2164         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2165         struct mm_struct *mm;
2166         char buffer[PROC_NUMBUF];
2167         size_t len;
2168         int ret;
2169
2170         if (!task)
2171                 return -ESRCH;
2172
2173         ret = 0;
2174         mm = get_task_mm(task);
2175         if (mm) {
2176                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2177                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2178                                 MMF_DUMP_FILTER_SHIFT));
2179                 mmput(mm);
2180                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2181         }
2182
2183         put_task_struct(task);
2184
2185         return ret;
2186 }
2187
2188 static ssize_t proc_coredump_filter_write(struct file *file,
2189                                           const char __user *buf,
2190                                           size_t count,
2191                                           loff_t *ppos)
2192 {
2193         struct task_struct *task;
2194         struct mm_struct *mm;
2195         char buffer[PROC_NUMBUF], *end;
2196         unsigned int val;
2197         int ret;
2198         int i;
2199         unsigned long mask;
2200
2201         ret = -EFAULT;
2202         memset(buffer, 0, sizeof(buffer));
2203         if (count > sizeof(buffer) - 1)
2204                 count = sizeof(buffer) - 1;
2205         if (copy_from_user(buffer, buf, count))
2206                 goto out_no_task;
2207
2208         ret = -EINVAL;
2209         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2210         if (*end == '\n')
2211                 end++;
2212         if (end - buffer == 0)
2213                 goto out_no_task;
2214
2215         ret = -ESRCH;
2216         task = get_proc_task(file->f_dentry->d_inode);
2217         if (!task)
2218                 goto out_no_task;
2219
2220         ret = end - buffer;
2221         mm = get_task_mm(task);
2222         if (!mm)
2223                 goto out_no_mm;
2224
2225         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2226                 if (val & mask)
2227                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2228                 else
2229                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2230         }
2231
2232         mmput(mm);
2233  out_no_mm:
2234         put_task_struct(task);
2235  out_no_task:
2236         return ret;
2237 }
2238
2239 static const struct file_operations proc_coredump_filter_operations = {
2240         .read           = proc_coredump_filter_read,
2241         .write          = proc_coredump_filter_write,
2242 };
2243 #endif
2244
2245 /*
2246  * /proc/self:
2247  */
2248 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2249                               int buflen)
2250 {
2251         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2252         pid_t tgid = task_tgid_nr_ns(current, ns);
2253         char tmp[PROC_NUMBUF];
2254         if (!tgid)
2255                 return -ENOENT;
2256         sprintf(tmp, "%d", tgid);
2257         return vfs_readlink(dentry,buffer,buflen,tmp);
2258 }
2259
2260 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2261 {
2262         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2263         pid_t tgid = task_tgid_nr_ns(current, ns);
2264         char tmp[PROC_NUMBUF];
2265         if (!tgid)
2266                 return ERR_PTR(-ENOENT);
2267         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2268         return ERR_PTR(vfs_follow_link(nd,tmp));
2269 }
2270
2271 static const struct inode_operations proc_self_inode_operations = {
2272         .readlink       = proc_self_readlink,
2273         .follow_link    = proc_self_follow_link,
2274 };
2275
2276 /*
2277  * proc base
2278  *
2279  * These are the directory entries in the root directory of /proc
2280  * that properly belong to the /proc filesystem, as they describe
2281  * describe something that is process related.
2282  */
2283 static const struct pid_entry proc_base_stuff[] = {
2284         NOD("self", S_IFLNK|S_IRWXUGO,
2285                 &proc_self_inode_operations, NULL, {}),
2286 };
2287
2288 /*
2289  *      Exceptional case: normally we are not allowed to unhash a busy
2290  * directory. In this case, however, we can do it - no aliasing problems
2291  * due to the way we treat inodes.
2292  */
2293 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2294 {
2295         struct inode *inode = dentry->d_inode;
2296         struct task_struct *task = get_proc_task(inode);
2297         if (task) {
2298                 put_task_struct(task);
2299                 return 1;
2300         }
2301         d_drop(dentry);
2302         return 0;
2303 }
2304
2305 static struct dentry_operations proc_base_dentry_operations =
2306 {
2307         .d_revalidate   = proc_base_revalidate,
2308         .d_delete       = pid_delete_dentry,
2309 };
2310
2311 static struct dentry *proc_base_instantiate(struct inode *dir,
2312         struct dentry *dentry, struct task_struct *task, const void *ptr)
2313 {
2314         const struct pid_entry *p = ptr;
2315         struct inode *inode;
2316         struct proc_inode *ei;
2317         struct dentry *error = ERR_PTR(-EINVAL);
2318
2319         /* Allocate the inode */
2320         error = ERR_PTR(-ENOMEM);
2321         inode = new_inode(dir->i_sb);
2322         if (!inode)
2323                 goto out;
2324
2325         /* Initialize the inode */
2326         ei = PROC_I(inode);
2327         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2328
2329         /*
2330          * grab the reference to the task.
2331          */
2332         ei->pid = get_task_pid(task, PIDTYPE_PID);
2333         if (!ei->pid)
2334                 goto out_iput;
2335
2336         inode->i_uid = 0;
2337         inode->i_gid = 0;
2338         inode->i_mode = p->mode;
2339         if (S_ISDIR(inode->i_mode))
2340                 inode->i_nlink = 2;
2341         if (S_ISLNK(inode->i_mode))
2342                 inode->i_size = 64;
2343         if (p->iop)
2344                 inode->i_op = p->iop;
2345         if (p->fop)
2346                 inode->i_fop = p->fop;
2347         ei->op = p->op;
2348         dentry->d_op = &proc_base_dentry_operations;
2349         d_add(dentry, inode);
2350         error = NULL;
2351 out:
2352         return error;
2353 out_iput:
2354         iput(inode);
2355         goto out;
2356 }
2357
2358 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2359 {
2360         struct dentry *error;
2361         struct task_struct *task = get_proc_task(dir);
2362         const struct pid_entry *p, *last;
2363
2364         error = ERR_PTR(-ENOENT);
2365
2366         if (!task)
2367                 goto out_no_task;
2368
2369         /* Lookup the directory entry */
2370         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2371         for (p = proc_base_stuff; p <= last; p++) {
2372                 if (p->len != dentry->d_name.len)
2373                         continue;
2374                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2375                         break;
2376         }
2377         if (p > last)
2378                 goto out;
2379
2380         error = proc_base_instantiate(dir, dentry, task, p);
2381
2382 out:
2383         put_task_struct(task);
2384 out_no_task:
2385         return error;
2386 }
2387
2388 static int proc_base_fill_cache(struct file *filp, void *dirent,
2389         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2390 {
2391         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2392                                 proc_base_instantiate, task, p);
2393 }
2394
2395 #ifdef CONFIG_TASK_IO_ACCOUNTING
2396 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2397 {
2398         struct task_io_accounting acct = task->ioac;
2399         unsigned long flags;
2400
2401         if (whole && lock_task_sighand(task, &flags)) {
2402                 struct task_struct *t = task;
2403
2404                 task_io_accounting_add(&acct, &task->signal->ioac);
2405                 while_each_thread(task, t)
2406                         task_io_accounting_add(&acct, &t->ioac);
2407
2408                 unlock_task_sighand(task, &flags);
2409         }
2410         return sprintf(buffer,
2411                         "rchar: %llu\n"
2412                         "wchar: %llu\n"
2413                         "syscr: %llu\n"
2414                         "syscw: %llu\n"
2415                         "read_bytes: %llu\n"
2416                         "write_bytes: %llu\n"
2417                         "cancelled_write_bytes: %llu\n",
2418                         (unsigned long long)acct.rchar,
2419                         (unsigned long long)acct.wchar,
2420                         (unsigned long long)acct.syscr,
2421                         (unsigned long long)acct.syscw,
2422                         (unsigned long long)acct.read_bytes,
2423                         (unsigned long long)acct.write_bytes,
2424                         (unsigned long long)acct.cancelled_write_bytes);
2425 }
2426
2427 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2428 {
2429         return do_io_accounting(task, buffer, 0);
2430 }
2431
2432 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2433 {
2434         return do_io_accounting(task, buffer, 1);
2435 }
2436 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2437
2438 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2439                                 struct pid *pid, struct task_struct *task)
2440 {
2441         seq_printf(m, "%08x\n", task->personality);
2442         return 0;
2443 }
2444
2445 /*
2446  * Thread groups
2447  */
2448 static const struct file_operations proc_task_operations;
2449 static const struct inode_operations proc_task_inode_operations;
2450
2451 static const struct pid_entry tgid_base_stuff[] = {
2452         DIR("task",       S_IRUGO|S_IXUGO, task),
2453         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2454         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2455 #ifdef CONFIG_NET
2456         DIR("net",        S_IRUGO|S_IXUGO, net),
2457 #endif
2458         REG("environ",    S_IRUSR, environ),
2459         INF("auxv",       S_IRUSR, pid_auxv),
2460         ONE("status",     S_IRUGO, pid_status),
2461         ONE("personality", S_IRUSR, pid_personality),
2462         INF("limits",     S_IRUSR, pid_limits),
2463 #ifdef CONFIG_SCHED_DEBUG
2464         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2465 #endif
2466 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2467         INF("syscall",    S_IRUSR, pid_syscall),
2468 #endif
2469         INF("cmdline",    S_IRUGO, pid_cmdline),
2470         ONE("stat",       S_IRUGO, tgid_stat),
2471         ONE("statm",      S_IRUGO, pid_statm),
2472         REG("maps",       S_IRUGO, maps),
2473 #ifdef CONFIG_NUMA
2474         REG("numa_maps",  S_IRUGO, numa_maps),
2475 #endif
2476         REG("mem",        S_IRUSR|S_IWUSR, mem),
2477         LNK("cwd",        cwd),
2478         LNK("root",       root),
2479         LNK("exe",        exe),
2480         REG("mounts",     S_IRUGO, mounts),
2481         REG("mountinfo",  S_IRUGO, mountinfo),
2482         REG("mountstats", S_IRUSR, mountstats),
2483 #ifdef CONFIG_PROC_PAGE_MONITOR
2484         REG("clear_refs", S_IWUSR, clear_refs),
2485         REG("smaps",      S_IRUGO, smaps),
2486         REG("pagemap",    S_IRUSR, pagemap),
2487 #endif
2488 #ifdef CONFIG_SECURITY
2489         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2490 #endif
2491 #ifdef CONFIG_KALLSYMS
2492         INF("wchan",      S_IRUGO, pid_wchan),
2493 #endif
2494 #ifdef CONFIG_SCHEDSTATS
2495         INF("schedstat",  S_IRUGO, pid_schedstat),
2496 #endif
2497 #ifdef CONFIG_LATENCYTOP
2498         REG("latency",  S_IRUGO, lstats),
2499 #endif
2500 #ifdef CONFIG_PROC_PID_CPUSET
2501         REG("cpuset",     S_IRUGO, cpuset),
2502 #endif
2503 #ifdef CONFIG_CGROUPS
2504         REG("cgroup",  S_IRUGO, cgroup),
2505 #endif
2506         INF("oom_score",  S_IRUGO, oom_score),
2507         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2508 #ifdef CONFIG_AUDITSYSCALL
2509         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2510         REG("sessionid",  S_IRUGO, sessionid),
2511 #endif
2512 #ifdef CONFIG_FAULT_INJECTION
2513         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2514 #endif
2515 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2516         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2517 #endif
2518 #ifdef CONFIG_TASK_IO_ACCOUNTING
2519         INF("io",       S_IRUGO, tgid_io_accounting),
2520 #endif
2521 };
2522
2523 static int proc_tgid_base_readdir(struct file * filp,
2524                              void * dirent, filldir_t filldir)
2525 {
2526         return proc_pident_readdir(filp,dirent,filldir,
2527                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2528 }
2529
2530 static const struct file_operations proc_tgid_base_operations = {
2531         .read           = generic_read_dir,
2532         .readdir        = proc_tgid_base_readdir,
2533 };
2534
2535 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2536         return proc_pident_lookup(dir, dentry,
2537                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2538 }
2539
2540 static const struct inode_operations proc_tgid_base_inode_operations = {
2541         .lookup         = proc_tgid_base_lookup,
2542         .getattr        = pid_getattr,
2543         .setattr        = proc_setattr,
2544 };
2545
2546 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2547 {
2548         struct dentry *dentry, *leader, *dir;
2549         char buf[PROC_NUMBUF];
2550         struct qstr name;
2551
2552         name.name = buf;
2553         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2554         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2555         if (dentry) {
2556                 if (!(current->flags & PF_EXITING))
2557                         shrink_dcache_parent(dentry);
2558                 d_drop(dentry);
2559                 dput(dentry);
2560         }
2561
2562         if (tgid == 0)
2563                 goto out;
2564
2565         name.name = buf;
2566         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2567         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2568         if (!leader)
2569                 goto out;
2570
2571         name.name = "task";
2572         name.len = strlen(name.name);
2573         dir = d_hash_and_lookup(leader, &name);
2574         if (!dir)
2575                 goto out_put_leader;
2576
2577         name.name = buf;
2578         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2579         dentry = d_hash_and_lookup(dir, &name);
2580         if (dentry) {
2581                 shrink_dcache_parent(dentry);
2582                 d_drop(dentry);
2583                 dput(dentry);
2584         }
2585
2586         dput(dir);
2587 out_put_leader:
2588         dput(leader);
2589 out:
2590         return;
2591 }
2592
2593 /**
2594  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2595  * @task: task that should be flushed.
2596  *
2597  * When flushing dentries from proc, one needs to flush them from global
2598  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2599  * in. This call is supposed to do all of this job.
2600  *
2601  * Looks in the dcache for
2602  * /proc/@pid
2603  * /proc/@tgid/task/@pid
2604  * if either directory is present flushes it and all of it'ts children
2605  * from the dcache.
2606  *
2607  * It is safe and reasonable to cache /proc entries for a task until
2608  * that task exits.  After that they just clog up the dcache with
2609  * useless entries, possibly causing useful dcache entries to be
2610  * flushed instead.  This routine is proved to flush those useless
2611  * dcache entries at process exit time.
2612  *
2613  * NOTE: This routine is just an optimization so it does not guarantee
2614  *       that no dcache entries will exist at process exit time it
2615  *       just makes it very unlikely that any will persist.
2616  */
2617
2618 void proc_flush_task(struct task_struct *task)
2619 {
2620         int i;
2621         struct pid *pid, *tgid = NULL;
2622         struct upid *upid;
2623
2624         pid = task_pid(task);
2625         if (thread_group_leader(task))
2626                 tgid = task_tgid(task);
2627
2628         for (i = 0; i <= pid->level; i++) {
2629                 upid = &pid->numbers[i];
2630                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2631                         tgid ? tgid->numbers[i].nr : 0);
2632         }
2633
2634         upid = &pid->numbers[pid->level];
2635         if (upid->nr == 1)
2636                 pid_ns_release_proc(upid->ns);
2637 }
2638
2639 static struct dentry *proc_pid_instantiate(struct inode *dir,
2640                                            struct dentry * dentry,
2641                                            struct task_struct *task, const void *ptr)
2642 {
2643         struct dentry *error = ERR_PTR(-ENOENT);
2644         struct inode *inode;
2645
2646         inode = proc_pid_make_inode(dir->i_sb, task);
2647         if (!inode)
2648                 goto out;
2649
2650         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2651         inode->i_op = &proc_tgid_base_inode_operations;
2652         inode->i_fop = &proc_tgid_base_operations;
2653         inode->i_flags|=S_IMMUTABLE;
2654
2655         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2656                 ARRAY_SIZE(tgid_base_stuff));
2657
2658         dentry->d_op = &pid_dentry_operations;
2659
2660         d_add(dentry, inode);
2661         /* Close the race of the process dying before we return the dentry */
2662         if (pid_revalidate(dentry, NULL))
2663                 error = NULL;
2664 out:
2665         return error;
2666 }
2667
2668 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2669 {
2670         struct dentry *result = ERR_PTR(-ENOENT);
2671         struct task_struct *task;
2672         unsigned tgid;
2673         struct pid_namespace *ns;
2674
2675         result = proc_base_lookup(dir, dentry);
2676         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2677                 goto out;
2678
2679         tgid = name_to_int(dentry);
2680         if (tgid == ~0U)
2681                 goto out;
2682
2683         ns = dentry->d_sb->s_fs_info;
2684         rcu_read_lock();
2685         task = find_task_by_pid_ns(tgid, ns);
2686         if (task)
2687                 get_task_struct(task);
2688         rcu_read_unlock();
2689         if (!task)
2690                 goto out;
2691
2692         result = proc_pid_instantiate(dir, dentry, task, NULL);
2693         put_task_struct(task);
2694 out:
2695         return result;
2696 }
2697
2698 /*
2699  * Find the first task with tgid >= tgid
2700  *
2701  */
2702 struct tgid_iter {
2703         unsigned int tgid;
2704         struct task_struct *task;
2705 };
2706 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2707 {
2708         struct pid *pid;
2709
2710         if (iter.task)
2711                 put_task_struct(iter.task);
2712         rcu_read_lock();
2713 retry:
2714         iter.task = NULL;
2715         pid = find_ge_pid(iter.tgid, ns);
2716         if (pid) {
2717                 iter.tgid = pid_nr_ns(pid, ns);
2718                 iter.task = pid_task(pid, PIDTYPE_PID);
2719                 /* What we to know is if the pid we have find is the
2720                  * pid of a thread_group_leader.  Testing for task
2721                  * being a thread_group_leader is the obvious thing
2722                  * todo but there is a window when it fails, due to
2723                  * the pid transfer logic in de_thread.
2724                  *
2725                  * So we perform the straight forward test of seeing
2726                  * if the pid we have found is the pid of a thread
2727                  * group leader, and don't worry if the task we have
2728                  * found doesn't happen to be a thread group leader.
2729                  * As we don't care in the case of readdir.
2730                  */
2731                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2732                         iter.tgid += 1;
2733                         goto retry;
2734                 }
2735                 get_task_struct(iter.task);
2736         }
2737         rcu_read_unlock();
2738         return iter;
2739 }
2740
2741 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2742
2743 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2744         struct tgid_iter iter)
2745 {
2746         char name[PROC_NUMBUF];
2747         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2748         return proc_fill_cache(filp, dirent, filldir, name, len,
2749                                 proc_pid_instantiate, iter.task, NULL);
2750 }
2751
2752 /* for the /proc/ directory itself, after non-process stuff has been done */
2753 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2754 {
2755         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2756         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2757         struct tgid_iter iter;
2758         struct pid_namespace *ns;
2759
2760         if (!reaper)
2761                 goto out_no_task;
2762
2763         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2764                 const struct pid_entry *p = &proc_base_stuff[nr];
2765                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2766                         goto out;
2767         }
2768
2769         ns = filp->f_dentry->d_sb->s_fs_info;
2770         iter.task = NULL;
2771         iter.tgid = filp->f_pos - TGID_OFFSET;
2772         for (iter = next_tgid(ns, iter);
2773              iter.task;
2774              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2775                 filp->f_pos = iter.tgid + TGID_OFFSET;
2776                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2777                         put_task_struct(iter.task);
2778                         goto out;
2779                 }
2780         }
2781         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2782 out:
2783         put_task_struct(reaper);
2784 out_no_task:
2785         return 0;
2786 }
2787
2788 /*
2789  * Tasks
2790  */
2791 static const struct pid_entry tid_base_stuff[] = {
2792         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2793         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2794         REG("environ",   S_IRUSR, environ),
2795         INF("auxv",      S_IRUSR, pid_auxv),
2796         ONE("status",    S_IRUGO, pid_status),
2797         ONE("personality", S_IRUSR, pid_personality),
2798         INF("limits",    S_IRUSR, pid_limits),
2799 #ifdef CONFIG_SCHED_DEBUG
2800         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2801 #endif
2802 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2803         INF("syscall",   S_IRUSR, pid_syscall),
2804 #endif
2805         INF("cmdline",   S_IRUGO, pid_cmdline),
2806         ONE("stat",      S_IRUGO, tid_stat),
2807         ONE("statm",     S_IRUGO, pid_statm),
2808         REG("maps",      S_IRUGO, maps),
2809 #ifdef CONFIG_NUMA
2810         REG("numa_maps", S_IRUGO, numa_maps),
2811 #endif
2812         REG("mem",       S_IRUSR|S_IWUSR, mem),
2813         LNK("cwd",       cwd),
2814         LNK("root",      root),
2815         LNK("exe",       exe),
2816         REG("mounts",    S_IRUGO, mounts),
2817         REG("mountinfo",  S_IRUGO, mountinfo),
2818 #ifdef CONFIG_PROC_PAGE_MONITOR
2819         REG("clear_refs", S_IWUSR, clear_refs),
2820         REG("smaps",     S_IRUGO, smaps),
2821         REG("pagemap",    S_IRUSR, pagemap),
2822 #endif
2823 #ifdef CONFIG_SECURITY
2824         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2825 #endif
2826 #ifdef CONFIG_KALLSYMS
2827         INF("wchan",     S_IRUGO, pid_wchan),
2828 #endif
2829 #ifdef CONFIG_SCHEDSTATS
2830         INF("schedstat", S_IRUGO, pid_schedstat),
2831 #endif
2832 #ifdef CONFIG_LATENCYTOP
2833         REG("latency",  S_IRUGO, lstats),
2834 #endif
2835 #ifdef CONFIG_PROC_PID_CPUSET
2836         REG("cpuset",    S_IRUGO, cpuset),
2837 #endif
2838 #ifdef CONFIG_CGROUPS
2839         REG("cgroup",  S_IRUGO, cgroup),
2840 #endif
2841         INF("oom_score", S_IRUGO, oom_score),
2842         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2843 #ifdef CONFIG_AUDITSYSCALL
2844         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2845         REG("sessionid",  S_IRUSR, sessionid),
2846 #endif
2847 #ifdef CONFIG_FAULT_INJECTION
2848         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2849 #endif
2850 #ifdef CONFIG_TASK_IO_ACCOUNTING
2851         INF("io",       S_IRUGO, tid_io_accounting),
2852 #endif
2853 };
2854
2855 static int proc_tid_base_readdir(struct file * filp,
2856                              void * dirent, filldir_t filldir)
2857 {
2858         return proc_pident_readdir(filp,dirent,filldir,
2859                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2860 }
2861
2862 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2863         return proc_pident_lookup(dir, dentry,
2864                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2865 }
2866
2867 static const struct file_operations proc_tid_base_operations = {
2868         .read           = generic_read_dir,
2869         .readdir        = proc_tid_base_readdir,
2870 };
2871
2872 static const struct inode_operations proc_tid_base_inode_operations = {
2873         .lookup         = proc_tid_base_lookup,
2874         .getattr        = pid_getattr,
2875         .setattr        = proc_setattr,
2876 };
2877
2878 static struct dentry *proc_task_instantiate(struct inode *dir,
2879         struct dentry *dentry, struct task_struct *task, const void *ptr)
2880 {
2881         struct dentry *error = ERR_PTR(-ENOENT);
2882         struct inode *inode;
2883         inode = proc_pid_make_inode(dir->i_sb, task);
2884
2885         if (!inode)
2886                 goto out;
2887         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2888         inode->i_op = &proc_tid_base_inode_operations;
2889         inode->i_fop = &proc_tid_base_operations;
2890         inode->i_flags|=S_IMMUTABLE;
2891
2892         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2893                 ARRAY_SIZE(tid_base_stuff));
2894
2895         dentry->d_op = &pid_dentry_operations;
2896
2897         d_add(dentry, inode);
2898         /* Close the race of the process dying before we return the dentry */
2899         if (pid_revalidate(dentry, NULL))
2900                 error = NULL;
2901 out:
2902         return error;
2903 }
2904
2905 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2906 {
2907         struct dentry *result = ERR_PTR(-ENOENT);
2908         struct task_struct *task;
2909         struct task_struct *leader = get_proc_task(dir);
2910         unsigned tid;
2911         struct pid_namespace *ns;
2912
2913         if (!leader)
2914                 goto out_no_task;
2915
2916         tid = name_to_int(dentry);
2917         if (tid == ~0U)
2918                 goto out;
2919
2920         ns = dentry->d_sb->s_fs_info;
2921         rcu_read_lock();
2922         task = find_task_by_pid_ns(tid, ns);
2923         if (task)
2924                 get_task_struct(task);
2925         rcu_read_unlock();
2926         if (!task)
2927                 goto out;
2928         if (!same_thread_group(leader, task))
2929                 goto out_drop_task;
2930
2931         result = proc_task_instantiate(dir, dentry, task, NULL);
2932 out_drop_task:
2933         put_task_struct(task);
2934 out:
2935         put_task_struct(leader);
2936 out_no_task:
2937         return result;
2938 }
2939
2940 /*
2941  * Find the first tid of a thread group to return to user space.
2942  *
2943  * Usually this is just the thread group leader, but if the users
2944  * buffer was too small or there was a seek into the middle of the
2945  * directory we have more work todo.
2946  *
2947  * In the case of a short read we start with find_task_by_pid.
2948  *
2949  * In the case of a seek we start with the leader and walk nr
2950  * threads past it.
2951  */
2952 static struct task_struct *first_tid(struct task_struct *leader,
2953                 int tid, int nr, struct pid_namespace *ns)
2954 {
2955         struct task_struct *pos;
2956
2957         rcu_read_lock();
2958         /* Attempt to start with the pid of a thread */
2959         if (tid && (nr > 0)) {
2960                 pos = find_task_by_pid_ns(tid, ns);
2961                 if (pos && (pos->group_leader == leader))
2962                         goto found;
2963         }
2964
2965         /* If nr exceeds the number of threads there is nothing todo */
2966         pos = NULL;
2967         if (nr && nr >= get_nr_threads(leader))
2968                 goto out;
2969
2970         /* If we haven't found our starting place yet start
2971          * with the leader and walk nr threads forward.
2972          */
2973         for (pos = leader; nr > 0; --nr) {
2974                 pos = next_thread(pos);
2975                 if (pos == leader) {
2976                         pos = NULL;
2977                         goto out;
2978                 }
2979         }
2980 found:
2981         get_task_struct(pos);
2982 out:
2983         rcu_read_unlock();
2984         return pos;
2985 }
2986
2987 /*
2988  * Find the next thread in the thread list.
2989  * Return NULL if there is an error or no next thread.
2990  *
2991  * The reference to the input task_struct is released.
2992  */
2993 static struct task_struct *next_tid(struct task_struct *start)
2994 {
2995         struct task_struct *pos = NULL;
2996         rcu_read_lock();
2997         if (pid_alive(start)) {
2998                 pos = next_thread(start);
2999                 if (thread_group_leader(pos))
3000                         pos = NULL;
3001                 else
3002                         get_task_struct(pos);
3003         }
3004         rcu_read_unlock();
3005         put_task_struct(start);
3006         return pos;
3007 }
3008
3009 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3010         struct task_struct *task, int tid)
3011 {
3012         char name[PROC_NUMBUF];
3013         int len = snprintf(name, sizeof(name), "%d", tid);
3014         return proc_fill_cache(filp, dirent, filldir, name, len,
3015                                 proc_task_instantiate, task, NULL);
3016 }
3017
3018 /* for the /proc/TGID/task/ directories */
3019 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3020 {
3021         struct dentry *dentry = filp->f_path.dentry;
3022         struct inode *inode = dentry->d_inode;
3023         struct task_struct *leader = NULL;
3024         struct task_struct *task;
3025         int retval = -ENOENT;
3026         ino_t ino;
3027         int tid;
3028         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
3029         struct pid_namespace *ns;
3030
3031         task = get_proc_task(inode);
3032         if (!task)
3033                 goto out_no_task;
3034         rcu_read_lock();
3035         if (pid_alive(task)) {
3036                 leader = task->group_leader;
3037                 get_task_struct(leader);
3038         }
3039         rcu_read_unlock();
3040         put_task_struct(task);
3041         if (!leader)
3042                 goto out_no_task;
3043         retval = 0;
3044
3045         switch (pos) {
3046         case 0:
3047                 ino = inode->i_ino;
3048                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
3049                         goto out;
3050                 pos++;
3051                 /* fall through */
3052         case 1:
3053                 ino = parent_ino(dentry);
3054                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
3055                         goto out;
3056                 pos++;
3057                 /* fall through */
3058         }
3059
3060         /* f_version caches the tgid value that the last readdir call couldn't
3061          * return. lseek aka telldir automagically resets f_version to 0.
3062          */
3063         ns = filp->f_dentry->d_sb->s_fs_info;
3064         tid = (int)filp->f_version;
3065         filp->f_version = 0;
3066         for (task = first_tid(leader, tid, pos - 2, ns);
3067              task;
3068              task = next_tid(task), pos++) {
3069                 tid = task_pid_nr_ns(task, ns);
3070                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3071                         /* returning this tgid failed, save it as the first
3072                          * pid for the next readir call */
3073                         filp->f_version = (u64)tid;
3074                         put_task_struct(task);
3075                         break;
3076                 }
3077         }
3078 out:
3079         filp->f_pos = pos;
3080         put_task_struct(leader);
3081 out_no_task:
3082         return retval;
3083 }
3084
3085 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3086 {
3087         struct inode *inode = dentry->d_inode;
3088         struct task_struct *p = get_proc_task(inode);
3089         generic_fillattr(inode, stat);
3090
3091         if (p) {
3092                 stat->nlink += get_nr_threads(p);
3093                 put_task_struct(p);
3094         }
3095
3096         return 0;
3097 }
3098
3099 static const struct inode_operations proc_task_inode_operations = {
3100         .lookup         = proc_task_lookup,
3101         .getattr        = proc_task_getattr,
3102         .setattr        = proc_setattr,
3103 };
3104
3105 static const struct file_operations proc_task_operations = {
3106         .read           = generic_read_dir,
3107         .readdir        = proc_task_readdir,
3108 };