Merge branch 'ext3-latency-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6] / 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/stacktrace.h>
69 #include <linux/resource.h>
70 #include <linux/module.h>
71 #include <linux/mount.h>
72 #include <linux/security.h>
73 #include <linux/ptrace.h>
74 #include <linux/tracehook.h>
75 #include <linux/cgroup.h>
76 #include <linux/cpuset.h>
77 #include <linux/audit.h>
78 #include <linux/poll.h>
79 #include <linux/nsproxy.h>
80 #include <linux/oom.h>
81 #include <linux/elf.h>
82 #include <linux/pid_namespace.h>
83 #include <linux/fs_struct.h>
84 #include "internal.h"
85
86 /* NOTE:
87  *      Implementing inode permission operations in /proc is almost
88  *      certainly an error.  Permission checks need to happen during
89  *      each system call not at open time.  The reason is that most of
90  *      what we wish to check for permissions in /proc varies at runtime.
91  *
92  *      The classic example of a problem is opening file descriptors
93  *      in /proc for a task before it execs a suid executable.
94  */
95
96 struct pid_entry {
97         char *name;
98         int len;
99         mode_t mode;
100         const struct inode_operations *iop;
101         const struct file_operations *fop;
102         union proc_op op;
103 };
104
105 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
106         .name = (NAME),                                 \
107         .len  = sizeof(NAME) - 1,                       \
108         .mode = MODE,                                   \
109         .iop  = IOP,                                    \
110         .fop  = FOP,                                    \
111         .op   = OP,                                     \
112 }
113
114 #define DIR(NAME, MODE, iops, fops)     \
115         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
116 #define LNK(NAME, get_link)                                     \
117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
118                 &proc_pid_link_inode_operations, NULL,          \
119                 { .proc_get_link = get_link } )
120 #define REG(NAME, MODE, fops)                           \
121         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
122 #define INF(NAME, MODE, read)                           \
123         NOD(NAME, (S_IFREG|(MODE)),                     \
124                 NULL, &proc_info_file_operations,       \
125                 { .proc_read = read } )
126 #define ONE(NAME, MODE, show)                           \
127         NOD(NAME, (S_IFREG|(MODE)),                     \
128                 NULL, &proc_single_file_operations,     \
129                 { .proc_show = show } )
130
131 /*
132  * Count the number of hardlinks for the pid_entry table, excluding the .
133  * and .. links.
134  */
135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
136         unsigned int n)
137 {
138         unsigned int i;
139         unsigned int count;
140
141         count = 0;
142         for (i = 0; i < n; ++i) {
143                 if (S_ISDIR(entries[i].mode))
144                         ++count;
145         }
146
147         return count;
148 }
149
150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
151 {
152         struct fs_struct *fs;
153         int result = -ENOENT;
154
155         task_lock(task);
156         fs = task->fs;
157         if (fs) {
158                 read_lock(&fs->lock);
159                 *path = root ? fs->root : fs->pwd;
160                 path_get(path);
161                 read_unlock(&fs->lock);
162                 result = 0;
163         }
164         task_unlock(task);
165         return result;
166 }
167
168 static int get_nr_threads(struct task_struct *tsk)
169 {
170         unsigned long flags;
171         int count = 0;
172
173         if (lock_task_sighand(tsk, &flags)) {
174                 count = atomic_read(&tsk->signal->count);
175                 unlock_task_sighand(tsk, &flags);
176         }
177         return count;
178 }
179
180 static int proc_cwd_link(struct inode *inode, struct path *path)
181 {
182         struct task_struct *task = get_proc_task(inode);
183         int result = -ENOENT;
184
185         if (task) {
186                 result = get_fs_path(task, path, 0);
187                 put_task_struct(task);
188         }
189         return result;
190 }
191
192 static int proc_root_link(struct inode *inode, struct path *path)
193 {
194         struct task_struct *task = get_proc_task(inode);
195         int result = -ENOENT;
196
197         if (task) {
198                 result = get_fs_path(task, path, 1);
199                 put_task_struct(task);
200         }
201         return result;
202 }
203
204 /*
205  * Return zero if current may access user memory in @task, -error if not.
206  */
207 static int check_mem_permission(struct task_struct *task)
208 {
209         /*
210          * A task can always look at itself, in case it chooses
211          * to use system calls instead of load instructions.
212          */
213         if (task == current)
214                 return 0;
215
216         /*
217          * If current is actively ptrace'ing, and would also be
218          * permitted to freshly attach with ptrace now, permit it.
219          */
220         if (task_is_stopped_or_traced(task)) {
221                 int match;
222                 rcu_read_lock();
223                 match = (tracehook_tracer_task(task) == current);
224                 rcu_read_unlock();
225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
226                         return 0;
227         }
228
229         /*
230          * Noone else is allowed.
231          */
232         return -EPERM;
233 }
234
235 struct mm_struct *mm_for_maps(struct task_struct *task)
236 {
237         struct mm_struct *mm = get_task_mm(task);
238         if (!mm)
239                 return NULL;
240         down_read(&mm->mmap_sem);
241         task_lock(task);
242         if (task->mm != mm)
243                 goto out;
244         if (task->mm != current->mm &&
245             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
246                 goto out;
247         task_unlock(task);
248         return mm;
249 out:
250         task_unlock(task);
251         up_read(&mm->mmap_sem);
252         mmput(mm);
253         return NULL;
254 }
255
256 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
257 {
258         int res = 0;
259         unsigned int len;
260         struct mm_struct *mm = get_task_mm(task);
261         if (!mm)
262                 goto out;
263         if (!mm->arg_end)
264                 goto out_mm;    /* Shh! No looking before we're done */
265
266         len = mm->arg_end - mm->arg_start;
267  
268         if (len > PAGE_SIZE)
269                 len = PAGE_SIZE;
270  
271         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
272
273         // If the nul at the end of args has been overwritten, then
274         // assume application is using setproctitle(3).
275         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
276                 len = strnlen(buffer, res);
277                 if (len < res) {
278                     res = len;
279                 } else {
280                         len = mm->env_end - mm->env_start;
281                         if (len > PAGE_SIZE - res)
282                                 len = PAGE_SIZE - res;
283                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
284                         res = strnlen(buffer, res);
285                 }
286         }
287 out_mm:
288         mmput(mm);
289 out:
290         return res;
291 }
292
293 static int proc_pid_auxv(struct task_struct *task, char *buffer)
294 {
295         int res = 0;
296         struct mm_struct *mm = get_task_mm(task);
297         if (mm) {
298                 unsigned int nwords = 0;
299                 do {
300                         nwords += 2;
301                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
302                 res = nwords * sizeof(mm->saved_auxv[0]);
303                 if (res > PAGE_SIZE)
304                         res = PAGE_SIZE;
305                 memcpy(buffer, mm->saved_auxv, res);
306                 mmput(mm);
307         }
308         return res;
309 }
310
311
312 #ifdef CONFIG_KALLSYMS
313 /*
314  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
315  * Returns the resolved symbol.  If that fails, simply return the address.
316  */
317 static int proc_pid_wchan(struct task_struct *task, char *buffer)
318 {
319         unsigned long wchan;
320         char symname[KSYM_NAME_LEN];
321
322         wchan = get_wchan(task);
323
324         if (lookup_symbol_name(wchan, symname) < 0)
325                 return sprintf(buffer, "%lu", wchan);
326         else
327                 return sprintf(buffer, "%s", symname);
328 }
329 #endif /* CONFIG_KALLSYMS */
330
331 #ifdef CONFIG_STACKTRACE
332
333 #define MAX_STACK_TRACE_DEPTH   64
334
335 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
336                           struct pid *pid, struct task_struct *task)
337 {
338         struct stack_trace trace;
339         unsigned long *entries;
340         int i;
341
342         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
343         if (!entries)
344                 return -ENOMEM;
345
346         trace.nr_entries        = 0;
347         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
348         trace.entries           = entries;
349         trace.skip              = 0;
350         save_stack_trace_tsk(task, &trace);
351
352         for (i = 0; i < trace.nr_entries; i++) {
353                 seq_printf(m, "[<%p>] %pS\n",
354                            (void *)entries[i], (void *)entries[i]);
355         }
356         kfree(entries);
357
358         return 0;
359 }
360 #endif
361
362 #ifdef CONFIG_SCHEDSTATS
363 /*
364  * Provides /proc/PID/schedstat
365  */
366 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
367 {
368         return sprintf(buffer, "%llu %llu %lu\n",
369                         (unsigned long long)task->se.sum_exec_runtime,
370                         (unsigned long long)task->sched_info.run_delay,
371                         task->sched_info.pcount);
372 }
373 #endif
374
375 #ifdef CONFIG_LATENCYTOP
376 static int lstats_show_proc(struct seq_file *m, void *v)
377 {
378         int i;
379         struct inode *inode = m->private;
380         struct task_struct *task = get_proc_task(inode);
381
382         if (!task)
383                 return -ESRCH;
384         seq_puts(m, "Latency Top version : v0.1\n");
385         for (i = 0; i < 32; i++) {
386                 if (task->latency_record[i].backtrace[0]) {
387                         int q;
388                         seq_printf(m, "%i %li %li ",
389                                 task->latency_record[i].count,
390                                 task->latency_record[i].time,
391                                 task->latency_record[i].max);
392                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
393                                 char sym[KSYM_SYMBOL_LEN];
394                                 char *c;
395                                 if (!task->latency_record[i].backtrace[q])
396                                         break;
397                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
398                                         break;
399                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
400                                 c = strchr(sym, '+');
401                                 if (c)
402                                         *c = 0;
403                                 seq_printf(m, "%s ", sym);
404                         }
405                         seq_printf(m, "\n");
406                 }
407
408         }
409         put_task_struct(task);
410         return 0;
411 }
412
413 static int lstats_open(struct inode *inode, struct file *file)
414 {
415         return single_open(file, lstats_show_proc, inode);
416 }
417
418 static ssize_t lstats_write(struct file *file, const char __user *buf,
419                             size_t count, loff_t *offs)
420 {
421         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
422
423         if (!task)
424                 return -ESRCH;
425         clear_all_latency_tracing(task);
426         put_task_struct(task);
427
428         return count;
429 }
430
431 static const struct file_operations proc_lstats_operations = {
432         .open           = lstats_open,
433         .read           = seq_read,
434         .write          = lstats_write,
435         .llseek         = seq_lseek,
436         .release        = single_release,
437 };
438
439 #endif
440
441 /* The badness from the OOM killer */
442 unsigned long badness(struct task_struct *p, unsigned long uptime);
443 static int proc_oom_score(struct task_struct *task, char *buffer)
444 {
445         unsigned long points;
446         struct timespec uptime;
447
448         do_posix_clock_monotonic_gettime(&uptime);
449         read_lock(&tasklist_lock);
450         points = badness(task, uptime.tv_sec);
451         read_unlock(&tasklist_lock);
452         return sprintf(buffer, "%lu\n", points);
453 }
454
455 struct limit_names {
456         char *name;
457         char *unit;
458 };
459
460 static const struct limit_names lnames[RLIM_NLIMITS] = {
461         [RLIMIT_CPU] = {"Max cpu time", "ms"},
462         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
463         [RLIMIT_DATA] = {"Max data size", "bytes"},
464         [RLIMIT_STACK] = {"Max stack size", "bytes"},
465         [RLIMIT_CORE] = {"Max core file size", "bytes"},
466         [RLIMIT_RSS] = {"Max resident set", "bytes"},
467         [RLIMIT_NPROC] = {"Max processes", "processes"},
468         [RLIMIT_NOFILE] = {"Max open files", "files"},
469         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
470         [RLIMIT_AS] = {"Max address space", "bytes"},
471         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
472         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
473         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
474         [RLIMIT_NICE] = {"Max nice priority", NULL},
475         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
476         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
477 };
478
479 /* Display limits for a process */
480 static int proc_pid_limits(struct task_struct *task, char *buffer)
481 {
482         unsigned int i;
483         int count = 0;
484         unsigned long flags;
485         char *bufptr = buffer;
486
487         struct rlimit rlim[RLIM_NLIMITS];
488
489         if (!lock_task_sighand(task, &flags))
490                 return 0;
491         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
492         unlock_task_sighand(task, &flags);
493
494         /*
495          * print the file header
496          */
497         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
498                         "Limit", "Soft Limit", "Hard Limit", "Units");
499
500         for (i = 0; i < RLIM_NLIMITS; i++) {
501                 if (rlim[i].rlim_cur == RLIM_INFINITY)
502                         count += sprintf(&bufptr[count], "%-25s %-20s ",
503                                          lnames[i].name, "unlimited");
504                 else
505                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
506                                          lnames[i].name, rlim[i].rlim_cur);
507
508                 if (rlim[i].rlim_max == RLIM_INFINITY)
509                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
510                 else
511                         count += sprintf(&bufptr[count], "%-20lu ",
512                                          rlim[i].rlim_max);
513
514                 if (lnames[i].unit)
515                         count += sprintf(&bufptr[count], "%-10s\n",
516                                          lnames[i].unit);
517                 else
518                         count += sprintf(&bufptr[count], "\n");
519         }
520
521         return count;
522 }
523
524 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
525 static int proc_pid_syscall(struct task_struct *task, char *buffer)
526 {
527         long nr;
528         unsigned long args[6], sp, pc;
529
530         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
531                 return sprintf(buffer, "running\n");
532
533         if (nr < 0)
534                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
535
536         return sprintf(buffer,
537                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
538                        nr,
539                        args[0], args[1], args[2], args[3], args[4], args[5],
540                        sp, pc);
541 }
542 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
543
544 /************************************************************************/
545 /*                       Here the fs part begins                        */
546 /************************************************************************/
547
548 /* permission checks */
549 static int proc_fd_access_allowed(struct inode *inode)
550 {
551         struct task_struct *task;
552         int allowed = 0;
553         /* Allow access to a task's file descriptors if it is us or we
554          * may use ptrace attach to the process and find out that
555          * information.
556          */
557         task = get_proc_task(inode);
558         if (task) {
559                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
560                 put_task_struct(task);
561         }
562         return allowed;
563 }
564
565 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
566 {
567         int error;
568         struct inode *inode = dentry->d_inode;
569
570         if (attr->ia_valid & ATTR_MODE)
571                 return -EPERM;
572
573         error = inode_change_ok(inode, attr);
574         if (!error)
575                 error = inode_setattr(inode, attr);
576         return error;
577 }
578
579 static const struct inode_operations proc_def_inode_operations = {
580         .setattr        = proc_setattr,
581 };
582
583 static int mounts_open_common(struct inode *inode, struct file *file,
584                               const struct seq_operations *op)
585 {
586         struct task_struct *task = get_proc_task(inode);
587         struct nsproxy *nsp;
588         struct mnt_namespace *ns = NULL;
589         struct path root;
590         struct proc_mounts *p;
591         int ret = -EINVAL;
592
593         if (task) {
594                 rcu_read_lock();
595                 nsp = task_nsproxy(task);
596                 if (nsp) {
597                         ns = nsp->mnt_ns;
598                         if (ns)
599                                 get_mnt_ns(ns);
600                 }
601                 rcu_read_unlock();
602                 if (ns && get_fs_path(task, &root, 1) == 0)
603                         ret = 0;
604                 put_task_struct(task);
605         }
606
607         if (!ns)
608                 goto err;
609         if (ret)
610                 goto err_put_ns;
611
612         ret = -ENOMEM;
613         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
614         if (!p)
615                 goto err_put_path;
616
617         file->private_data = &p->m;
618         ret = seq_open(file, op);
619         if (ret)
620                 goto err_free;
621
622         p->m.private = p;
623         p->ns = ns;
624         p->root = root;
625         p->event = ns->event;
626
627         return 0;
628
629  err_free:
630         kfree(p);
631  err_put_path:
632         path_put(&root);
633  err_put_ns:
634         put_mnt_ns(ns);
635  err:
636         return ret;
637 }
638
639 static int mounts_release(struct inode *inode, struct file *file)
640 {
641         struct proc_mounts *p = file->private_data;
642         path_put(&p->root);
643         put_mnt_ns(p->ns);
644         return seq_release(inode, file);
645 }
646
647 static unsigned mounts_poll(struct file *file, poll_table *wait)
648 {
649         struct proc_mounts *p = file->private_data;
650         struct mnt_namespace *ns = p->ns;
651         unsigned res = 0;
652
653         poll_wait(file, &ns->poll, wait);
654
655         spin_lock(&vfsmount_lock);
656         if (p->event != ns->event) {
657                 p->event = ns->event;
658                 res = POLLERR;
659         }
660         spin_unlock(&vfsmount_lock);
661
662         return res;
663 }
664
665 static int mounts_open(struct inode *inode, struct file *file)
666 {
667         return mounts_open_common(inode, file, &mounts_op);
668 }
669
670 static const struct file_operations proc_mounts_operations = {
671         .open           = mounts_open,
672         .read           = seq_read,
673         .llseek         = seq_lseek,
674         .release        = mounts_release,
675         .poll           = mounts_poll,
676 };
677
678 static int mountinfo_open(struct inode *inode, struct file *file)
679 {
680         return mounts_open_common(inode, file, &mountinfo_op);
681 }
682
683 static const struct file_operations proc_mountinfo_operations = {
684         .open           = mountinfo_open,
685         .read           = seq_read,
686         .llseek         = seq_lseek,
687         .release        = mounts_release,
688         .poll           = mounts_poll,
689 };
690
691 static int mountstats_open(struct inode *inode, struct file *file)
692 {
693         return mounts_open_common(inode, file, &mountstats_op);
694 }
695
696 static const struct file_operations proc_mountstats_operations = {
697         .open           = mountstats_open,
698         .read           = seq_read,
699         .llseek         = seq_lseek,
700         .release        = mounts_release,
701 };
702
703 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
704
705 static ssize_t proc_info_read(struct file * file, char __user * buf,
706                           size_t count, loff_t *ppos)
707 {
708         struct inode * inode = file->f_path.dentry->d_inode;
709         unsigned long page;
710         ssize_t length;
711         struct task_struct *task = get_proc_task(inode);
712
713         length = -ESRCH;
714         if (!task)
715                 goto out_no_task;
716
717         if (count > PROC_BLOCK_SIZE)
718                 count = PROC_BLOCK_SIZE;
719
720         length = -ENOMEM;
721         if (!(page = __get_free_page(GFP_TEMPORARY)))
722                 goto out;
723
724         length = PROC_I(inode)->op.proc_read(task, (char*)page);
725
726         if (length >= 0)
727                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
728         free_page(page);
729 out:
730         put_task_struct(task);
731 out_no_task:
732         return length;
733 }
734
735 static const struct file_operations proc_info_file_operations = {
736         .read           = proc_info_read,
737 };
738
739 static int proc_single_show(struct seq_file *m, void *v)
740 {
741         struct inode *inode = m->private;
742         struct pid_namespace *ns;
743         struct pid *pid;
744         struct task_struct *task;
745         int ret;
746
747         ns = inode->i_sb->s_fs_info;
748         pid = proc_pid(inode);
749         task = get_pid_task(pid, PIDTYPE_PID);
750         if (!task)
751                 return -ESRCH;
752
753         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
754
755         put_task_struct(task);
756         return ret;
757 }
758
759 static int proc_single_open(struct inode *inode, struct file *filp)
760 {
761         int ret;
762         ret = single_open(filp, proc_single_show, NULL);
763         if (!ret) {
764                 struct seq_file *m = filp->private_data;
765
766                 m->private = inode;
767         }
768         return ret;
769 }
770
771 static const struct file_operations proc_single_file_operations = {
772         .open           = proc_single_open,
773         .read           = seq_read,
774         .llseek         = seq_lseek,
775         .release        = single_release,
776 };
777
778 static int mem_open(struct inode* inode, struct file* file)
779 {
780         file->private_data = (void*)((long)current->self_exec_id);
781         return 0;
782 }
783
784 static ssize_t mem_read(struct file * file, char __user * buf,
785                         size_t count, loff_t *ppos)
786 {
787         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
788         char *page;
789         unsigned long src = *ppos;
790         int ret = -ESRCH;
791         struct mm_struct *mm;
792
793         if (!task)
794                 goto out_no_task;
795
796         if (check_mem_permission(task))
797                 goto out;
798
799         ret = -ENOMEM;
800         page = (char *)__get_free_page(GFP_TEMPORARY);
801         if (!page)
802                 goto out;
803
804         ret = 0;
805  
806         mm = get_task_mm(task);
807         if (!mm)
808                 goto out_free;
809
810         ret = -EIO;
811  
812         if (file->private_data != (void*)((long)current->self_exec_id))
813                 goto out_put;
814
815         ret = 0;
816  
817         while (count > 0) {
818                 int this_len, retval;
819
820                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
821                 retval = access_process_vm(task, src, page, this_len, 0);
822                 if (!retval || check_mem_permission(task)) {
823                         if (!ret)
824                                 ret = -EIO;
825                         break;
826                 }
827
828                 if (copy_to_user(buf, page, retval)) {
829                         ret = -EFAULT;
830                         break;
831                 }
832  
833                 ret += retval;
834                 src += retval;
835                 buf += retval;
836                 count -= retval;
837         }
838         *ppos = src;
839
840 out_put:
841         mmput(mm);
842 out_free:
843         free_page((unsigned long) page);
844 out:
845         put_task_struct(task);
846 out_no_task:
847         return ret;
848 }
849
850 #define mem_write NULL
851
852 #ifndef mem_write
853 /* This is a security hazard */
854 static ssize_t mem_write(struct file * file, const char __user *buf,
855                          size_t count, loff_t *ppos)
856 {
857         int copied;
858         char *page;
859         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
860         unsigned long dst = *ppos;
861
862         copied = -ESRCH;
863         if (!task)
864                 goto out_no_task;
865
866         if (check_mem_permission(task))
867                 goto out;
868
869         copied = -ENOMEM;
870         page = (char *)__get_free_page(GFP_TEMPORARY);
871         if (!page)
872                 goto out;
873
874         copied = 0;
875         while (count > 0) {
876                 int this_len, retval;
877
878                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
879                 if (copy_from_user(page, buf, this_len)) {
880                         copied = -EFAULT;
881                         break;
882                 }
883                 retval = access_process_vm(task, dst, page, this_len, 1);
884                 if (!retval) {
885                         if (!copied)
886                                 copied = -EIO;
887                         break;
888                 }
889                 copied += retval;
890                 buf += retval;
891                 dst += retval;
892                 count -= retval;                        
893         }
894         *ppos = dst;
895         free_page((unsigned long) page);
896 out:
897         put_task_struct(task);
898 out_no_task:
899         return copied;
900 }
901 #endif
902
903 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
904 {
905         switch (orig) {
906         case 0:
907                 file->f_pos = offset;
908                 break;
909         case 1:
910                 file->f_pos += offset;
911                 break;
912         default:
913                 return -EINVAL;
914         }
915         force_successful_syscall_return();
916         return file->f_pos;
917 }
918
919 static const struct file_operations proc_mem_operations = {
920         .llseek         = mem_lseek,
921         .read           = mem_read,
922         .write          = mem_write,
923         .open           = mem_open,
924 };
925
926 static ssize_t environ_read(struct file *file, char __user *buf,
927                         size_t count, loff_t *ppos)
928 {
929         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
930         char *page;
931         unsigned long src = *ppos;
932         int ret = -ESRCH;
933         struct mm_struct *mm;
934
935         if (!task)
936                 goto out_no_task;
937
938         if (!ptrace_may_access(task, PTRACE_MODE_READ))
939                 goto out;
940
941         ret = -ENOMEM;
942         page = (char *)__get_free_page(GFP_TEMPORARY);
943         if (!page)
944                 goto out;
945
946         ret = 0;
947
948         mm = get_task_mm(task);
949         if (!mm)
950                 goto out_free;
951
952         while (count > 0) {
953                 int this_len, retval, max_len;
954
955                 this_len = mm->env_end - (mm->env_start + src);
956
957                 if (this_len <= 0)
958                         break;
959
960                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
961                 this_len = (this_len > max_len) ? max_len : this_len;
962
963                 retval = access_process_vm(task, (mm->env_start + src),
964                         page, this_len, 0);
965
966                 if (retval <= 0) {
967                         ret = retval;
968                         break;
969                 }
970
971                 if (copy_to_user(buf, page, retval)) {
972                         ret = -EFAULT;
973                         break;
974                 }
975
976                 ret += retval;
977                 src += retval;
978                 buf += retval;
979                 count -= retval;
980         }
981         *ppos = src;
982
983         mmput(mm);
984 out_free:
985         free_page((unsigned long) page);
986 out:
987         put_task_struct(task);
988 out_no_task:
989         return ret;
990 }
991
992 static const struct file_operations proc_environ_operations = {
993         .read           = environ_read,
994 };
995
996 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
997                                 size_t count, loff_t *ppos)
998 {
999         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1000         char buffer[PROC_NUMBUF];
1001         size_t len;
1002         int oom_adjust;
1003
1004         if (!task)
1005                 return -ESRCH;
1006         oom_adjust = task->oomkilladj;
1007         put_task_struct(task);
1008
1009         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1010
1011         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1012 }
1013
1014 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1015                                 size_t count, loff_t *ppos)
1016 {
1017         struct task_struct *task;
1018         char buffer[PROC_NUMBUF], *end;
1019         int oom_adjust;
1020
1021         memset(buffer, 0, sizeof(buffer));
1022         if (count > sizeof(buffer) - 1)
1023                 count = sizeof(buffer) - 1;
1024         if (copy_from_user(buffer, buf, count))
1025                 return -EFAULT;
1026         oom_adjust = simple_strtol(buffer, &end, 0);
1027         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1028              oom_adjust != OOM_DISABLE)
1029                 return -EINVAL;
1030         if (*end == '\n')
1031                 end++;
1032         task = get_proc_task(file->f_path.dentry->d_inode);
1033         if (!task)
1034                 return -ESRCH;
1035         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1036                 put_task_struct(task);
1037                 return -EACCES;
1038         }
1039         task->oomkilladj = oom_adjust;
1040         put_task_struct(task);
1041         if (end - buffer == 0)
1042                 return -EIO;
1043         return end - buffer;
1044 }
1045
1046 static const struct file_operations proc_oom_adjust_operations = {
1047         .read           = oom_adjust_read,
1048         .write          = oom_adjust_write,
1049 };
1050
1051 #ifdef CONFIG_AUDITSYSCALL
1052 #define TMPBUFLEN 21
1053 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1054                                   size_t count, loff_t *ppos)
1055 {
1056         struct inode * inode = file->f_path.dentry->d_inode;
1057         struct task_struct *task = get_proc_task(inode);
1058         ssize_t length;
1059         char tmpbuf[TMPBUFLEN];
1060
1061         if (!task)
1062                 return -ESRCH;
1063         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1064                                 audit_get_loginuid(task));
1065         put_task_struct(task);
1066         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1067 }
1068
1069 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1070                                    size_t count, loff_t *ppos)
1071 {
1072         struct inode * inode = file->f_path.dentry->d_inode;
1073         char *page, *tmp;
1074         ssize_t length;
1075         uid_t loginuid;
1076
1077         if (!capable(CAP_AUDIT_CONTROL))
1078                 return -EPERM;
1079
1080         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1081                 return -EPERM;
1082
1083         if (count >= PAGE_SIZE)
1084                 count = PAGE_SIZE - 1;
1085
1086         if (*ppos != 0) {
1087                 /* No partial writes. */
1088                 return -EINVAL;
1089         }
1090         page = (char*)__get_free_page(GFP_TEMPORARY);
1091         if (!page)
1092                 return -ENOMEM;
1093         length = -EFAULT;
1094         if (copy_from_user(page, buf, count))
1095                 goto out_free_page;
1096
1097         page[count] = '\0';
1098         loginuid = simple_strtoul(page, &tmp, 10);
1099         if (tmp == page) {
1100                 length = -EINVAL;
1101                 goto out_free_page;
1102
1103         }
1104         length = audit_set_loginuid(current, loginuid);
1105         if (likely(length == 0))
1106                 length = count;
1107
1108 out_free_page:
1109         free_page((unsigned long) page);
1110         return length;
1111 }
1112
1113 static const struct file_operations proc_loginuid_operations = {
1114         .read           = proc_loginuid_read,
1115         .write          = proc_loginuid_write,
1116 };
1117
1118 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1119                                   size_t count, loff_t *ppos)
1120 {
1121         struct inode * inode = file->f_path.dentry->d_inode;
1122         struct task_struct *task = get_proc_task(inode);
1123         ssize_t length;
1124         char tmpbuf[TMPBUFLEN];
1125
1126         if (!task)
1127                 return -ESRCH;
1128         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1129                                 audit_get_sessionid(task));
1130         put_task_struct(task);
1131         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1132 }
1133
1134 static const struct file_operations proc_sessionid_operations = {
1135         .read           = proc_sessionid_read,
1136 };
1137 #endif
1138
1139 #ifdef CONFIG_FAULT_INJECTION
1140 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1141                                       size_t count, loff_t *ppos)
1142 {
1143         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1144         char buffer[PROC_NUMBUF];
1145         size_t len;
1146         int make_it_fail;
1147
1148         if (!task)
1149                 return -ESRCH;
1150         make_it_fail = task->make_it_fail;
1151         put_task_struct(task);
1152
1153         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1154
1155         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1156 }
1157
1158 static ssize_t proc_fault_inject_write(struct file * file,
1159                         const char __user * buf, size_t count, loff_t *ppos)
1160 {
1161         struct task_struct *task;
1162         char buffer[PROC_NUMBUF], *end;
1163         int make_it_fail;
1164
1165         if (!capable(CAP_SYS_RESOURCE))
1166                 return -EPERM;
1167         memset(buffer, 0, sizeof(buffer));
1168         if (count > sizeof(buffer) - 1)
1169                 count = sizeof(buffer) - 1;
1170         if (copy_from_user(buffer, buf, count))
1171                 return -EFAULT;
1172         make_it_fail = simple_strtol(buffer, &end, 0);
1173         if (*end == '\n')
1174                 end++;
1175         task = get_proc_task(file->f_dentry->d_inode);
1176         if (!task)
1177                 return -ESRCH;
1178         task->make_it_fail = make_it_fail;
1179         put_task_struct(task);
1180         if (end - buffer == 0)
1181                 return -EIO;
1182         return end - buffer;
1183 }
1184
1185 static const struct file_operations proc_fault_inject_operations = {
1186         .read           = proc_fault_inject_read,
1187         .write          = proc_fault_inject_write,
1188 };
1189 #endif
1190
1191
1192 #ifdef CONFIG_SCHED_DEBUG
1193 /*
1194  * Print out various scheduling related per-task fields:
1195  */
1196 static int sched_show(struct seq_file *m, void *v)
1197 {
1198         struct inode *inode = m->private;
1199         struct task_struct *p;
1200
1201         p = get_proc_task(inode);
1202         if (!p)
1203                 return -ESRCH;
1204         proc_sched_show_task(p, m);
1205
1206         put_task_struct(p);
1207
1208         return 0;
1209 }
1210
1211 static ssize_t
1212 sched_write(struct file *file, const char __user *buf,
1213             size_t count, loff_t *offset)
1214 {
1215         struct inode *inode = file->f_path.dentry->d_inode;
1216         struct task_struct *p;
1217
1218         p = get_proc_task(inode);
1219         if (!p)
1220                 return -ESRCH;
1221         proc_sched_set_task(p);
1222
1223         put_task_struct(p);
1224
1225         return count;
1226 }
1227
1228 static int sched_open(struct inode *inode, struct file *filp)
1229 {
1230         int ret;
1231
1232         ret = single_open(filp, sched_show, NULL);
1233         if (!ret) {
1234                 struct seq_file *m = filp->private_data;
1235
1236                 m->private = inode;
1237         }
1238         return ret;
1239 }
1240
1241 static const struct file_operations proc_pid_sched_operations = {
1242         .open           = sched_open,
1243         .read           = seq_read,
1244         .write          = sched_write,
1245         .llseek         = seq_lseek,
1246         .release        = single_release,
1247 };
1248
1249 #endif
1250
1251 /*
1252  * We added or removed a vma mapping the executable. The vmas are only mapped
1253  * during exec and are not mapped with the mmap system call.
1254  * Callers must hold down_write() on the mm's mmap_sem for these
1255  */
1256 void added_exe_file_vma(struct mm_struct *mm)
1257 {
1258         mm->num_exe_file_vmas++;
1259 }
1260
1261 void removed_exe_file_vma(struct mm_struct *mm)
1262 {
1263         mm->num_exe_file_vmas--;
1264         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1265                 fput(mm->exe_file);
1266                 mm->exe_file = NULL;
1267         }
1268
1269 }
1270
1271 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1272 {
1273         if (new_exe_file)
1274                 get_file(new_exe_file);
1275         if (mm->exe_file)
1276                 fput(mm->exe_file);
1277         mm->exe_file = new_exe_file;
1278         mm->num_exe_file_vmas = 0;
1279 }
1280
1281 struct file *get_mm_exe_file(struct mm_struct *mm)
1282 {
1283         struct file *exe_file;
1284
1285         /* We need mmap_sem to protect against races with removal of
1286          * VM_EXECUTABLE vmas */
1287         down_read(&mm->mmap_sem);
1288         exe_file = mm->exe_file;
1289         if (exe_file)
1290                 get_file(exe_file);
1291         up_read(&mm->mmap_sem);
1292         return exe_file;
1293 }
1294
1295 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1296 {
1297         /* It's safe to write the exe_file pointer without exe_file_lock because
1298          * this is called during fork when the task is not yet in /proc */
1299         newmm->exe_file = get_mm_exe_file(oldmm);
1300 }
1301
1302 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1303 {
1304         struct task_struct *task;
1305         struct mm_struct *mm;
1306         struct file *exe_file;
1307
1308         task = get_proc_task(inode);
1309         if (!task)
1310                 return -ENOENT;
1311         mm = get_task_mm(task);
1312         put_task_struct(task);
1313         if (!mm)
1314                 return -ENOENT;
1315         exe_file = get_mm_exe_file(mm);
1316         mmput(mm);
1317         if (exe_file) {
1318                 *exe_path = exe_file->f_path;
1319                 path_get(&exe_file->f_path);
1320                 fput(exe_file);
1321                 return 0;
1322         } else
1323                 return -ENOENT;
1324 }
1325
1326 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1327 {
1328         struct inode *inode = dentry->d_inode;
1329         int error = -EACCES;
1330
1331         /* We don't need a base pointer in the /proc filesystem */
1332         path_put(&nd->path);
1333
1334         /* Are we allowed to snoop on the tasks file descriptors? */
1335         if (!proc_fd_access_allowed(inode))
1336                 goto out;
1337
1338         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1339         nd->last_type = LAST_BIND;
1340 out:
1341         return ERR_PTR(error);
1342 }
1343
1344 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1345 {
1346         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1347         char *pathname;
1348         int len;
1349
1350         if (!tmp)
1351                 return -ENOMEM;
1352
1353         pathname = d_path(path, tmp, PAGE_SIZE);
1354         len = PTR_ERR(pathname);
1355         if (IS_ERR(pathname))
1356                 goto out;
1357         len = tmp + PAGE_SIZE - 1 - pathname;
1358
1359         if (len > buflen)
1360                 len = buflen;
1361         if (copy_to_user(buffer, pathname, len))
1362                 len = -EFAULT;
1363  out:
1364         free_page((unsigned long)tmp);
1365         return len;
1366 }
1367
1368 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1369 {
1370         int error = -EACCES;
1371         struct inode *inode = dentry->d_inode;
1372         struct path path;
1373
1374         /* Are we allowed to snoop on the tasks file descriptors? */
1375         if (!proc_fd_access_allowed(inode))
1376                 goto out;
1377
1378         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1379         if (error)
1380                 goto out;
1381
1382         error = do_proc_readlink(&path, buffer, buflen);
1383         path_put(&path);
1384 out:
1385         return error;
1386 }
1387
1388 static const struct inode_operations proc_pid_link_inode_operations = {
1389         .readlink       = proc_pid_readlink,
1390         .follow_link    = proc_pid_follow_link,
1391         .setattr        = proc_setattr,
1392 };
1393
1394
1395 /* building an inode */
1396
1397 static int task_dumpable(struct task_struct *task)
1398 {
1399         int dumpable = 0;
1400         struct mm_struct *mm;
1401
1402         task_lock(task);
1403         mm = task->mm;
1404         if (mm)
1405                 dumpable = get_dumpable(mm);
1406         task_unlock(task);
1407         if(dumpable == 1)
1408                 return 1;
1409         return 0;
1410 }
1411
1412
1413 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1414 {
1415         struct inode * inode;
1416         struct proc_inode *ei;
1417         const struct cred *cred;
1418
1419         /* We need a new inode */
1420
1421         inode = new_inode(sb);
1422         if (!inode)
1423                 goto out;
1424
1425         /* Common stuff */
1426         ei = PROC_I(inode);
1427         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1428         inode->i_op = &proc_def_inode_operations;
1429
1430         /*
1431          * grab the reference to task.
1432          */
1433         ei->pid = get_task_pid(task, PIDTYPE_PID);
1434         if (!ei->pid)
1435                 goto out_unlock;
1436
1437         if (task_dumpable(task)) {
1438                 rcu_read_lock();
1439                 cred = __task_cred(task);
1440                 inode->i_uid = cred->euid;
1441                 inode->i_gid = cred->egid;
1442                 rcu_read_unlock();
1443         }
1444         security_task_to_inode(task, inode);
1445
1446 out:
1447         return inode;
1448
1449 out_unlock:
1450         iput(inode);
1451         return NULL;
1452 }
1453
1454 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1455 {
1456         struct inode *inode = dentry->d_inode;
1457         struct task_struct *task;
1458         const struct cred *cred;
1459
1460         generic_fillattr(inode, stat);
1461
1462         rcu_read_lock();
1463         stat->uid = 0;
1464         stat->gid = 0;
1465         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1466         if (task) {
1467                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1468                     task_dumpable(task)) {
1469                         cred = __task_cred(task);
1470                         stat->uid = cred->euid;
1471                         stat->gid = cred->egid;
1472                 }
1473         }
1474         rcu_read_unlock();
1475         return 0;
1476 }
1477
1478 /* dentry stuff */
1479
1480 /*
1481  *      Exceptional case: normally we are not allowed to unhash a busy
1482  * directory. In this case, however, we can do it - no aliasing problems
1483  * due to the way we treat inodes.
1484  *
1485  * Rewrite the inode's ownerships here because the owning task may have
1486  * performed a setuid(), etc.
1487  *
1488  * Before the /proc/pid/status file was created the only way to read
1489  * the effective uid of a /process was to stat /proc/pid.  Reading
1490  * /proc/pid/status is slow enough that procps and other packages
1491  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1492  * made this apply to all per process world readable and executable
1493  * directories.
1494  */
1495 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1496 {
1497         struct inode *inode = dentry->d_inode;
1498         struct task_struct *task = get_proc_task(inode);
1499         const struct cred *cred;
1500
1501         if (task) {
1502                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1503                     task_dumpable(task)) {
1504                         rcu_read_lock();
1505                         cred = __task_cred(task);
1506                         inode->i_uid = cred->euid;
1507                         inode->i_gid = cred->egid;
1508                         rcu_read_unlock();
1509                 } else {
1510                         inode->i_uid = 0;
1511                         inode->i_gid = 0;
1512                 }
1513                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1514                 security_task_to_inode(task, inode);
1515                 put_task_struct(task);
1516                 return 1;
1517         }
1518         d_drop(dentry);
1519         return 0;
1520 }
1521
1522 static int pid_delete_dentry(struct dentry * dentry)
1523 {
1524         /* Is the task we represent dead?
1525          * If so, then don't put the dentry on the lru list,
1526          * kill it immediately.
1527          */
1528         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1529 }
1530
1531 static const struct dentry_operations pid_dentry_operations =
1532 {
1533         .d_revalidate   = pid_revalidate,
1534         .d_delete       = pid_delete_dentry,
1535 };
1536
1537 /* Lookups */
1538
1539 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1540                                 struct task_struct *, const void *);
1541
1542 /*
1543  * Fill a directory entry.
1544  *
1545  * If possible create the dcache entry and derive our inode number and
1546  * file type from dcache entry.
1547  *
1548  * Since all of the proc inode numbers are dynamically generated, the inode
1549  * numbers do not exist until the inode is cache.  This means creating the
1550  * the dcache entry in readdir is necessary to keep the inode numbers
1551  * reported by readdir in sync with the inode numbers reported
1552  * by stat.
1553  */
1554 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1555         char *name, int len,
1556         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1557 {
1558         struct dentry *child, *dir = filp->f_path.dentry;
1559         struct inode *inode;
1560         struct qstr qname;
1561         ino_t ino = 0;
1562         unsigned type = DT_UNKNOWN;
1563
1564         qname.name = name;
1565         qname.len  = len;
1566         qname.hash = full_name_hash(name, len);
1567
1568         child = d_lookup(dir, &qname);
1569         if (!child) {
1570                 struct dentry *new;
1571                 new = d_alloc(dir, &qname);
1572                 if (new) {
1573                         child = instantiate(dir->d_inode, new, task, ptr);
1574                         if (child)
1575                                 dput(new);
1576                         else
1577                                 child = new;
1578                 }
1579         }
1580         if (!child || IS_ERR(child) || !child->d_inode)
1581                 goto end_instantiate;
1582         inode = child->d_inode;
1583         if (inode) {
1584                 ino = inode->i_ino;
1585                 type = inode->i_mode >> 12;
1586         }
1587         dput(child);
1588 end_instantiate:
1589         if (!ino)
1590                 ino = find_inode_number(dir, &qname);
1591         if (!ino)
1592                 ino = 1;
1593         return filldir(dirent, name, len, filp->f_pos, ino, type);
1594 }
1595
1596 static unsigned name_to_int(struct dentry *dentry)
1597 {
1598         const char *name = dentry->d_name.name;
1599         int len = dentry->d_name.len;
1600         unsigned n = 0;
1601
1602         if (len > 1 && *name == '0')
1603                 goto out;
1604         while (len-- > 0) {
1605                 unsigned c = *name++ - '0';
1606                 if (c > 9)
1607                         goto out;
1608                 if (n >= (~0U-9)/10)
1609                         goto out;
1610                 n *= 10;
1611                 n += c;
1612         }
1613         return n;
1614 out:
1615         return ~0U;
1616 }
1617
1618 #define PROC_FDINFO_MAX 64
1619
1620 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1621 {
1622         struct task_struct *task = get_proc_task(inode);
1623         struct files_struct *files = NULL;
1624         struct file *file;
1625         int fd = proc_fd(inode);
1626
1627         if (task) {
1628                 files = get_files_struct(task);
1629                 put_task_struct(task);
1630         }
1631         if (files) {
1632                 /*
1633                  * We are not taking a ref to the file structure, so we must
1634                  * hold ->file_lock.
1635                  */
1636                 spin_lock(&files->file_lock);
1637                 file = fcheck_files(files, fd);
1638                 if (file) {
1639                         if (path) {
1640                                 *path = file->f_path;
1641                                 path_get(&file->f_path);
1642                         }
1643                         if (info)
1644                                 snprintf(info, PROC_FDINFO_MAX,
1645                                          "pos:\t%lli\n"
1646                                          "flags:\t0%o\n",
1647                                          (long long) file->f_pos,
1648                                          file->f_flags);
1649                         spin_unlock(&files->file_lock);
1650                         put_files_struct(files);
1651                         return 0;
1652                 }
1653                 spin_unlock(&files->file_lock);
1654                 put_files_struct(files);
1655         }
1656         return -ENOENT;
1657 }
1658
1659 static int proc_fd_link(struct inode *inode, struct path *path)
1660 {
1661         return proc_fd_info(inode, path, NULL);
1662 }
1663
1664 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1665 {
1666         struct inode *inode = dentry->d_inode;
1667         struct task_struct *task = get_proc_task(inode);
1668         int fd = proc_fd(inode);
1669         struct files_struct *files;
1670         const struct cred *cred;
1671
1672         if (task) {
1673                 files = get_files_struct(task);
1674                 if (files) {
1675                         rcu_read_lock();
1676                         if (fcheck_files(files, fd)) {
1677                                 rcu_read_unlock();
1678                                 put_files_struct(files);
1679                                 if (task_dumpable(task)) {
1680                                         rcu_read_lock();
1681                                         cred = __task_cred(task);
1682                                         inode->i_uid = cred->euid;
1683                                         inode->i_gid = cred->egid;
1684                                         rcu_read_unlock();
1685                                 } else {
1686                                         inode->i_uid = 0;
1687                                         inode->i_gid = 0;
1688                                 }
1689                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1690                                 security_task_to_inode(task, inode);
1691                                 put_task_struct(task);
1692                                 return 1;
1693                         }
1694                         rcu_read_unlock();
1695                         put_files_struct(files);
1696                 }
1697                 put_task_struct(task);
1698         }
1699         d_drop(dentry);
1700         return 0;
1701 }
1702
1703 static const struct dentry_operations tid_fd_dentry_operations =
1704 {
1705         .d_revalidate   = tid_fd_revalidate,
1706         .d_delete       = pid_delete_dentry,
1707 };
1708
1709 static struct dentry *proc_fd_instantiate(struct inode *dir,
1710         struct dentry *dentry, struct task_struct *task, const void *ptr)
1711 {
1712         unsigned fd = *(const unsigned *)ptr;
1713         struct file *file;
1714         struct files_struct *files;
1715         struct inode *inode;
1716         struct proc_inode *ei;
1717         struct dentry *error = ERR_PTR(-ENOENT);
1718
1719         inode = proc_pid_make_inode(dir->i_sb, task);
1720         if (!inode)
1721                 goto out;
1722         ei = PROC_I(inode);
1723         ei->fd = fd;
1724         files = get_files_struct(task);
1725         if (!files)
1726                 goto out_iput;
1727         inode->i_mode = S_IFLNK;
1728
1729         /*
1730          * We are not taking a ref to the file structure, so we must
1731          * hold ->file_lock.
1732          */
1733         spin_lock(&files->file_lock);
1734         file = fcheck_files(files, fd);
1735         if (!file)
1736                 goto out_unlock;
1737         if (file->f_mode & FMODE_READ)
1738                 inode->i_mode |= S_IRUSR | S_IXUSR;
1739         if (file->f_mode & FMODE_WRITE)
1740                 inode->i_mode |= S_IWUSR | S_IXUSR;
1741         spin_unlock(&files->file_lock);
1742         put_files_struct(files);
1743
1744         inode->i_op = &proc_pid_link_inode_operations;
1745         inode->i_size = 64;
1746         ei->op.proc_get_link = proc_fd_link;
1747         dentry->d_op = &tid_fd_dentry_operations;
1748         d_add(dentry, inode);
1749         /* Close the race of the process dying before we return the dentry */
1750         if (tid_fd_revalidate(dentry, NULL))
1751                 error = NULL;
1752
1753  out:
1754         return error;
1755 out_unlock:
1756         spin_unlock(&files->file_lock);
1757         put_files_struct(files);
1758 out_iput:
1759         iput(inode);
1760         goto out;
1761 }
1762
1763 static struct dentry *proc_lookupfd_common(struct inode *dir,
1764                                            struct dentry *dentry,
1765                                            instantiate_t instantiate)
1766 {
1767         struct task_struct *task = get_proc_task(dir);
1768         unsigned fd = name_to_int(dentry);
1769         struct dentry *result = ERR_PTR(-ENOENT);
1770
1771         if (!task)
1772                 goto out_no_task;
1773         if (fd == ~0U)
1774                 goto out;
1775
1776         result = instantiate(dir, dentry, task, &fd);
1777 out:
1778         put_task_struct(task);
1779 out_no_task:
1780         return result;
1781 }
1782
1783 static int proc_readfd_common(struct file * filp, void * dirent,
1784                               filldir_t filldir, instantiate_t instantiate)
1785 {
1786         struct dentry *dentry = filp->f_path.dentry;
1787         struct inode *inode = dentry->d_inode;
1788         struct task_struct *p = get_proc_task(inode);
1789         unsigned int fd, ino;
1790         int retval;
1791         struct files_struct * files;
1792
1793         retval = -ENOENT;
1794         if (!p)
1795                 goto out_no_task;
1796         retval = 0;
1797
1798         fd = filp->f_pos;
1799         switch (fd) {
1800                 case 0:
1801                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1802                                 goto out;
1803                         filp->f_pos++;
1804                 case 1:
1805                         ino = parent_ino(dentry);
1806                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1807                                 goto out;
1808                         filp->f_pos++;
1809                 default:
1810                         files = get_files_struct(p);
1811                         if (!files)
1812                                 goto out;
1813                         rcu_read_lock();
1814                         for (fd = filp->f_pos-2;
1815                              fd < files_fdtable(files)->max_fds;
1816                              fd++, filp->f_pos++) {
1817                                 char name[PROC_NUMBUF];
1818                                 int len;
1819
1820                                 if (!fcheck_files(files, fd))
1821                                         continue;
1822                                 rcu_read_unlock();
1823
1824                                 len = snprintf(name, sizeof(name), "%d", fd);
1825                                 if (proc_fill_cache(filp, dirent, filldir,
1826                                                     name, len, instantiate,
1827                                                     p, &fd) < 0) {
1828                                         rcu_read_lock();
1829                                         break;
1830                                 }
1831                                 rcu_read_lock();
1832                         }
1833                         rcu_read_unlock();
1834                         put_files_struct(files);
1835         }
1836 out:
1837         put_task_struct(p);
1838 out_no_task:
1839         return retval;
1840 }
1841
1842 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1843                                     struct nameidata *nd)
1844 {
1845         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1846 }
1847
1848 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1849 {
1850         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1851 }
1852
1853 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1854                                       size_t len, loff_t *ppos)
1855 {
1856         char tmp[PROC_FDINFO_MAX];
1857         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1858         if (!err)
1859                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1860         return err;
1861 }
1862
1863 static const struct file_operations proc_fdinfo_file_operations = {
1864         .open           = nonseekable_open,
1865         .read           = proc_fdinfo_read,
1866 };
1867
1868 static const struct file_operations proc_fd_operations = {
1869         .read           = generic_read_dir,
1870         .readdir        = proc_readfd,
1871 };
1872
1873 /*
1874  * /proc/pid/fd needs a special permission handler so that a process can still
1875  * access /proc/self/fd after it has executed a setuid().
1876  */
1877 static int proc_fd_permission(struct inode *inode, int mask)
1878 {
1879         int rv;
1880
1881         rv = generic_permission(inode, mask, NULL);
1882         if (rv == 0)
1883                 return 0;
1884         if (task_pid(current) == proc_pid(inode))
1885                 rv = 0;
1886         return rv;
1887 }
1888
1889 /*
1890  * proc directories can do almost nothing..
1891  */
1892 static const struct inode_operations proc_fd_inode_operations = {
1893         .lookup         = proc_lookupfd,
1894         .permission     = proc_fd_permission,
1895         .setattr        = proc_setattr,
1896 };
1897
1898 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1899         struct dentry *dentry, struct task_struct *task, const void *ptr)
1900 {
1901         unsigned fd = *(unsigned *)ptr;
1902         struct inode *inode;
1903         struct proc_inode *ei;
1904         struct dentry *error = ERR_PTR(-ENOENT);
1905
1906         inode = proc_pid_make_inode(dir->i_sb, task);
1907         if (!inode)
1908                 goto out;
1909         ei = PROC_I(inode);
1910         ei->fd = fd;
1911         inode->i_mode = S_IFREG | S_IRUSR;
1912         inode->i_fop = &proc_fdinfo_file_operations;
1913         dentry->d_op = &tid_fd_dentry_operations;
1914         d_add(dentry, inode);
1915         /* Close the race of the process dying before we return the dentry */
1916         if (tid_fd_revalidate(dentry, NULL))
1917                 error = NULL;
1918
1919  out:
1920         return error;
1921 }
1922
1923 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1924                                         struct dentry *dentry,
1925                                         struct nameidata *nd)
1926 {
1927         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1928 }
1929
1930 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1931 {
1932         return proc_readfd_common(filp, dirent, filldir,
1933                                   proc_fdinfo_instantiate);
1934 }
1935
1936 static const struct file_operations proc_fdinfo_operations = {
1937         .read           = generic_read_dir,
1938         .readdir        = proc_readfdinfo,
1939 };
1940
1941 /*
1942  * proc directories can do almost nothing..
1943  */
1944 static const struct inode_operations proc_fdinfo_inode_operations = {
1945         .lookup         = proc_lookupfdinfo,
1946         .setattr        = proc_setattr,
1947 };
1948
1949
1950 static struct dentry *proc_pident_instantiate(struct inode *dir,
1951         struct dentry *dentry, struct task_struct *task, const void *ptr)
1952 {
1953         const struct pid_entry *p = ptr;
1954         struct inode *inode;
1955         struct proc_inode *ei;
1956         struct dentry *error = ERR_PTR(-EINVAL);
1957
1958         inode = proc_pid_make_inode(dir->i_sb, task);
1959         if (!inode)
1960                 goto out;
1961
1962         ei = PROC_I(inode);
1963         inode->i_mode = p->mode;
1964         if (S_ISDIR(inode->i_mode))
1965                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1966         if (p->iop)
1967                 inode->i_op = p->iop;
1968         if (p->fop)
1969                 inode->i_fop = p->fop;
1970         ei->op = p->op;
1971         dentry->d_op = &pid_dentry_operations;
1972         d_add(dentry, inode);
1973         /* Close the race of the process dying before we return the dentry */
1974         if (pid_revalidate(dentry, NULL))
1975                 error = NULL;
1976 out:
1977         return error;
1978 }
1979
1980 static struct dentry *proc_pident_lookup(struct inode *dir, 
1981                                          struct dentry *dentry,
1982                                          const struct pid_entry *ents,
1983                                          unsigned int nents)
1984 {
1985         struct dentry *error;
1986         struct task_struct *task = get_proc_task(dir);
1987         const struct pid_entry *p, *last;
1988
1989         error = ERR_PTR(-ENOENT);
1990
1991         if (!task)
1992                 goto out_no_task;
1993
1994         /*
1995          * Yes, it does not scale. And it should not. Don't add
1996          * new entries into /proc/<tgid>/ without very good reasons.
1997          */
1998         last = &ents[nents - 1];
1999         for (p = ents; p <= last; p++) {
2000                 if (p->len != dentry->d_name.len)
2001                         continue;
2002                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2003                         break;
2004         }
2005         if (p > last)
2006                 goto out;
2007
2008         error = proc_pident_instantiate(dir, dentry, task, p);
2009 out:
2010         put_task_struct(task);
2011 out_no_task:
2012         return error;
2013 }
2014
2015 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2016         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2017 {
2018         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2019                                 proc_pident_instantiate, task, p);
2020 }
2021
2022 static int proc_pident_readdir(struct file *filp,
2023                 void *dirent, filldir_t filldir,
2024                 const struct pid_entry *ents, unsigned int nents)
2025 {
2026         int i;
2027         struct dentry *dentry = filp->f_path.dentry;
2028         struct inode *inode = dentry->d_inode;
2029         struct task_struct *task = get_proc_task(inode);
2030         const struct pid_entry *p, *last;
2031         ino_t ino;
2032         int ret;
2033
2034         ret = -ENOENT;
2035         if (!task)
2036                 goto out_no_task;
2037
2038         ret = 0;
2039         i = filp->f_pos;
2040         switch (i) {
2041         case 0:
2042                 ino = inode->i_ino;
2043                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2044                         goto out;
2045                 i++;
2046                 filp->f_pos++;
2047                 /* fall through */
2048         case 1:
2049                 ino = parent_ino(dentry);
2050                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2051                         goto out;
2052                 i++;
2053                 filp->f_pos++;
2054                 /* fall through */
2055         default:
2056                 i -= 2;
2057                 if (i >= nents) {
2058                         ret = 1;
2059                         goto out;
2060                 }
2061                 p = ents + i;
2062                 last = &ents[nents - 1];
2063                 while (p <= last) {
2064                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2065                                 goto out;
2066                         filp->f_pos++;
2067                         p++;
2068                 }
2069         }
2070
2071         ret = 1;
2072 out:
2073         put_task_struct(task);
2074 out_no_task:
2075         return ret;
2076 }
2077
2078 #ifdef CONFIG_SECURITY
2079 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2080                                   size_t count, loff_t *ppos)
2081 {
2082         struct inode * inode = file->f_path.dentry->d_inode;
2083         char *p = NULL;
2084         ssize_t length;
2085         struct task_struct *task = get_proc_task(inode);
2086
2087         if (!task)
2088                 return -ESRCH;
2089
2090         length = security_getprocattr(task,
2091                                       (char*)file->f_path.dentry->d_name.name,
2092                                       &p);
2093         put_task_struct(task);
2094         if (length > 0)
2095                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2096         kfree(p);
2097         return length;
2098 }
2099
2100 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2101                                    size_t count, loff_t *ppos)
2102 {
2103         struct inode * inode = file->f_path.dentry->d_inode;
2104         char *page;
2105         ssize_t length;
2106         struct task_struct *task = get_proc_task(inode);
2107
2108         length = -ESRCH;
2109         if (!task)
2110                 goto out_no_task;
2111         if (count > PAGE_SIZE)
2112                 count = PAGE_SIZE;
2113
2114         /* No partial writes. */
2115         length = -EINVAL;
2116         if (*ppos != 0)
2117                 goto out;
2118
2119         length = -ENOMEM;
2120         page = (char*)__get_free_page(GFP_TEMPORARY);
2121         if (!page)
2122                 goto out;
2123
2124         length = -EFAULT;
2125         if (copy_from_user(page, buf, count))
2126                 goto out_free;
2127
2128         length = security_setprocattr(task,
2129                                       (char*)file->f_path.dentry->d_name.name,
2130                                       (void*)page, count);
2131 out_free:
2132         free_page((unsigned long) page);
2133 out:
2134         put_task_struct(task);
2135 out_no_task:
2136         return length;
2137 }
2138
2139 static const struct file_operations proc_pid_attr_operations = {
2140         .read           = proc_pid_attr_read,
2141         .write          = proc_pid_attr_write,
2142 };
2143
2144 static const struct pid_entry attr_dir_stuff[] = {
2145         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2146         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2147         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2148         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2149         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2150         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2151 };
2152
2153 static int proc_attr_dir_readdir(struct file * filp,
2154                              void * dirent, filldir_t filldir)
2155 {
2156         return proc_pident_readdir(filp,dirent,filldir,
2157                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2158 }
2159
2160 static const struct file_operations proc_attr_dir_operations = {
2161         .read           = generic_read_dir,
2162         .readdir        = proc_attr_dir_readdir,
2163 };
2164
2165 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2166                                 struct dentry *dentry, struct nameidata *nd)
2167 {
2168         return proc_pident_lookup(dir, dentry,
2169                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2170 }
2171
2172 static const struct inode_operations proc_attr_dir_inode_operations = {
2173         .lookup         = proc_attr_dir_lookup,
2174         .getattr        = pid_getattr,
2175         .setattr        = proc_setattr,
2176 };
2177
2178 #endif
2179
2180 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2181 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2182                                          size_t count, loff_t *ppos)
2183 {
2184         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2185         struct mm_struct *mm;
2186         char buffer[PROC_NUMBUF];
2187         size_t len;
2188         int ret;
2189
2190         if (!task)
2191                 return -ESRCH;
2192
2193         ret = 0;
2194         mm = get_task_mm(task);
2195         if (mm) {
2196                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2197                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2198                                 MMF_DUMP_FILTER_SHIFT));
2199                 mmput(mm);
2200                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2201         }
2202
2203         put_task_struct(task);
2204
2205         return ret;
2206 }
2207
2208 static ssize_t proc_coredump_filter_write(struct file *file,
2209                                           const char __user *buf,
2210                                           size_t count,
2211                                           loff_t *ppos)
2212 {
2213         struct task_struct *task;
2214         struct mm_struct *mm;
2215         char buffer[PROC_NUMBUF], *end;
2216         unsigned int val;
2217         int ret;
2218         int i;
2219         unsigned long mask;
2220
2221         ret = -EFAULT;
2222         memset(buffer, 0, sizeof(buffer));
2223         if (count > sizeof(buffer) - 1)
2224                 count = sizeof(buffer) - 1;
2225         if (copy_from_user(buffer, buf, count))
2226                 goto out_no_task;
2227
2228         ret = -EINVAL;
2229         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2230         if (*end == '\n')
2231                 end++;
2232         if (end - buffer == 0)
2233                 goto out_no_task;
2234
2235         ret = -ESRCH;
2236         task = get_proc_task(file->f_dentry->d_inode);
2237         if (!task)
2238                 goto out_no_task;
2239
2240         ret = end - buffer;
2241         mm = get_task_mm(task);
2242         if (!mm)
2243                 goto out_no_mm;
2244
2245         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2246                 if (val & mask)
2247                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2248                 else
2249                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2250         }
2251
2252         mmput(mm);
2253  out_no_mm:
2254         put_task_struct(task);
2255  out_no_task:
2256         return ret;
2257 }
2258
2259 static const struct file_operations proc_coredump_filter_operations = {
2260         .read           = proc_coredump_filter_read,
2261         .write          = proc_coredump_filter_write,
2262 };
2263 #endif
2264
2265 /*
2266  * /proc/self:
2267  */
2268 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2269                               int buflen)
2270 {
2271         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2272         pid_t tgid = task_tgid_nr_ns(current, ns);
2273         char tmp[PROC_NUMBUF];
2274         if (!tgid)
2275                 return -ENOENT;
2276         sprintf(tmp, "%d", tgid);
2277         return vfs_readlink(dentry,buffer,buflen,tmp);
2278 }
2279
2280 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2281 {
2282         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2283         pid_t tgid = task_tgid_nr_ns(current, ns);
2284         char tmp[PROC_NUMBUF];
2285         if (!tgid)
2286                 return ERR_PTR(-ENOENT);
2287         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2288         return ERR_PTR(vfs_follow_link(nd,tmp));
2289 }
2290
2291 static const struct inode_operations proc_self_inode_operations = {
2292         .readlink       = proc_self_readlink,
2293         .follow_link    = proc_self_follow_link,
2294 };
2295
2296 /*
2297  * proc base
2298  *
2299  * These are the directory entries in the root directory of /proc
2300  * that properly belong to the /proc filesystem, as they describe
2301  * describe something that is process related.
2302  */
2303 static const struct pid_entry proc_base_stuff[] = {
2304         NOD("self", S_IFLNK|S_IRWXUGO,
2305                 &proc_self_inode_operations, NULL, {}),
2306 };
2307
2308 /*
2309  *      Exceptional case: normally we are not allowed to unhash a busy
2310  * directory. In this case, however, we can do it - no aliasing problems
2311  * due to the way we treat inodes.
2312  */
2313 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2314 {
2315         struct inode *inode = dentry->d_inode;
2316         struct task_struct *task = get_proc_task(inode);
2317         if (task) {
2318                 put_task_struct(task);
2319                 return 1;
2320         }
2321         d_drop(dentry);
2322         return 0;
2323 }
2324
2325 static const struct dentry_operations proc_base_dentry_operations =
2326 {
2327         .d_revalidate   = proc_base_revalidate,
2328         .d_delete       = pid_delete_dentry,
2329 };
2330
2331 static struct dentry *proc_base_instantiate(struct inode *dir,
2332         struct dentry *dentry, struct task_struct *task, const void *ptr)
2333 {
2334         const struct pid_entry *p = ptr;
2335         struct inode *inode;
2336         struct proc_inode *ei;
2337         struct dentry *error = ERR_PTR(-EINVAL);
2338
2339         /* Allocate the inode */
2340         error = ERR_PTR(-ENOMEM);
2341         inode = new_inode(dir->i_sb);
2342         if (!inode)
2343                 goto out;
2344
2345         /* Initialize the inode */
2346         ei = PROC_I(inode);
2347         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2348
2349         /*
2350          * grab the reference to the task.
2351          */
2352         ei->pid = get_task_pid(task, PIDTYPE_PID);
2353         if (!ei->pid)
2354                 goto out_iput;
2355
2356         inode->i_mode = p->mode;
2357         if (S_ISDIR(inode->i_mode))
2358                 inode->i_nlink = 2;
2359         if (S_ISLNK(inode->i_mode))
2360                 inode->i_size = 64;
2361         if (p->iop)
2362                 inode->i_op = p->iop;
2363         if (p->fop)
2364                 inode->i_fop = p->fop;
2365         ei->op = p->op;
2366         dentry->d_op = &proc_base_dentry_operations;
2367         d_add(dentry, inode);
2368         error = NULL;
2369 out:
2370         return error;
2371 out_iput:
2372         iput(inode);
2373         goto out;
2374 }
2375
2376 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2377 {
2378         struct dentry *error;
2379         struct task_struct *task = get_proc_task(dir);
2380         const struct pid_entry *p, *last;
2381
2382         error = ERR_PTR(-ENOENT);
2383
2384         if (!task)
2385                 goto out_no_task;
2386
2387         /* Lookup the directory entry */
2388         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2389         for (p = proc_base_stuff; p <= last; p++) {
2390                 if (p->len != dentry->d_name.len)
2391                         continue;
2392                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2393                         break;
2394         }
2395         if (p > last)
2396                 goto out;
2397
2398         error = proc_base_instantiate(dir, dentry, task, p);
2399
2400 out:
2401         put_task_struct(task);
2402 out_no_task:
2403         return error;
2404 }
2405
2406 static int proc_base_fill_cache(struct file *filp, void *dirent,
2407         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2408 {
2409         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2410                                 proc_base_instantiate, task, p);
2411 }
2412
2413 #ifdef CONFIG_TASK_IO_ACCOUNTING
2414 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2415 {
2416         struct task_io_accounting acct = task->ioac;
2417         unsigned long flags;
2418
2419         if (whole && lock_task_sighand(task, &flags)) {
2420                 struct task_struct *t = task;
2421
2422                 task_io_accounting_add(&acct, &task->signal->ioac);
2423                 while_each_thread(task, t)
2424                         task_io_accounting_add(&acct, &t->ioac);
2425
2426                 unlock_task_sighand(task, &flags);
2427         }
2428         return sprintf(buffer,
2429                         "rchar: %llu\n"
2430                         "wchar: %llu\n"
2431                         "syscr: %llu\n"
2432                         "syscw: %llu\n"
2433                         "read_bytes: %llu\n"
2434                         "write_bytes: %llu\n"
2435                         "cancelled_write_bytes: %llu\n",
2436                         (unsigned long long)acct.rchar,
2437                         (unsigned long long)acct.wchar,
2438                         (unsigned long long)acct.syscr,
2439                         (unsigned long long)acct.syscw,
2440                         (unsigned long long)acct.read_bytes,
2441                         (unsigned long long)acct.write_bytes,
2442                         (unsigned long long)acct.cancelled_write_bytes);
2443 }
2444
2445 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2446 {
2447         return do_io_accounting(task, buffer, 0);
2448 }
2449
2450 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2451 {
2452         return do_io_accounting(task, buffer, 1);
2453 }
2454 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2455
2456 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2457                                 struct pid *pid, struct task_struct *task)
2458 {
2459         seq_printf(m, "%08x\n", task->personality);
2460         return 0;
2461 }
2462
2463 /*
2464  * Thread groups
2465  */
2466 static const struct file_operations proc_task_operations;
2467 static const struct inode_operations proc_task_inode_operations;
2468
2469 static const struct pid_entry tgid_base_stuff[] = {
2470         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2471         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2472         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2473 #ifdef CONFIG_NET
2474         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2475 #endif
2476         REG("environ",    S_IRUSR, proc_environ_operations),
2477         INF("auxv",       S_IRUSR, proc_pid_auxv),
2478         ONE("status",     S_IRUGO, proc_pid_status),
2479         ONE("personality", S_IRUSR, proc_pid_personality),
2480         INF("limits",     S_IRUSR, proc_pid_limits),
2481 #ifdef CONFIG_SCHED_DEBUG
2482         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2483 #endif
2484 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2485         INF("syscall",    S_IRUSR, proc_pid_syscall),
2486 #endif
2487         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2488         ONE("stat",       S_IRUGO, proc_tgid_stat),
2489         ONE("statm",      S_IRUGO, proc_pid_statm),
2490         REG("maps",       S_IRUGO, proc_maps_operations),
2491 #ifdef CONFIG_NUMA
2492         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2493 #endif
2494         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2495         LNK("cwd",        proc_cwd_link),
2496         LNK("root",       proc_root_link),
2497         LNK("exe",        proc_exe_link),
2498         REG("mounts",     S_IRUGO, proc_mounts_operations),
2499         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2500         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2501 #ifdef CONFIG_PROC_PAGE_MONITOR
2502         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2503         REG("smaps",      S_IRUGO, proc_smaps_operations),
2504         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2505 #endif
2506 #ifdef CONFIG_SECURITY
2507         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2508 #endif
2509 #ifdef CONFIG_KALLSYMS
2510         INF("wchan",      S_IRUGO, proc_pid_wchan),
2511 #endif
2512 #ifdef CONFIG_STACKTRACE
2513         ONE("stack",      S_IRUSR, proc_pid_stack),
2514 #endif
2515 #ifdef CONFIG_SCHEDSTATS
2516         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2517 #endif
2518 #ifdef CONFIG_LATENCYTOP
2519         REG("latency",  S_IRUGO, proc_lstats_operations),
2520 #endif
2521 #ifdef CONFIG_PROC_PID_CPUSET
2522         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2523 #endif
2524 #ifdef CONFIG_CGROUPS
2525         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2526 #endif
2527         INF("oom_score",  S_IRUGO, proc_oom_score),
2528         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2529 #ifdef CONFIG_AUDITSYSCALL
2530         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2531         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2532 #endif
2533 #ifdef CONFIG_FAULT_INJECTION
2534         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2535 #endif
2536 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2537         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2538 #endif
2539 #ifdef CONFIG_TASK_IO_ACCOUNTING
2540         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2541 #endif
2542 };
2543
2544 static int proc_tgid_base_readdir(struct file * filp,
2545                              void * dirent, filldir_t filldir)
2546 {
2547         return proc_pident_readdir(filp,dirent,filldir,
2548                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2549 }
2550
2551 static const struct file_operations proc_tgid_base_operations = {
2552         .read           = generic_read_dir,
2553         .readdir        = proc_tgid_base_readdir,
2554 };
2555
2556 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2557         return proc_pident_lookup(dir, dentry,
2558                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2559 }
2560
2561 static const struct inode_operations proc_tgid_base_inode_operations = {
2562         .lookup         = proc_tgid_base_lookup,
2563         .getattr        = pid_getattr,
2564         .setattr        = proc_setattr,
2565 };
2566
2567 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2568 {
2569         struct dentry *dentry, *leader, *dir;
2570         char buf[PROC_NUMBUF];
2571         struct qstr name;
2572
2573         name.name = buf;
2574         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2575         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2576         if (dentry) {
2577                 if (!(current->flags & PF_EXITING))
2578                         shrink_dcache_parent(dentry);
2579                 d_drop(dentry);
2580                 dput(dentry);
2581         }
2582
2583         if (tgid == 0)
2584                 goto out;
2585
2586         name.name = buf;
2587         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2588         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2589         if (!leader)
2590                 goto out;
2591
2592         name.name = "task";
2593         name.len = strlen(name.name);
2594         dir = d_hash_and_lookup(leader, &name);
2595         if (!dir)
2596                 goto out_put_leader;
2597
2598         name.name = buf;
2599         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2600         dentry = d_hash_and_lookup(dir, &name);
2601         if (dentry) {
2602                 shrink_dcache_parent(dentry);
2603                 d_drop(dentry);
2604                 dput(dentry);
2605         }
2606
2607         dput(dir);
2608 out_put_leader:
2609         dput(leader);
2610 out:
2611         return;
2612 }
2613
2614 /**
2615  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2616  * @task: task that should be flushed.
2617  *
2618  * When flushing dentries from proc, one needs to flush them from global
2619  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2620  * in. This call is supposed to do all of this job.
2621  *
2622  * Looks in the dcache for
2623  * /proc/@pid
2624  * /proc/@tgid/task/@pid
2625  * if either directory is present flushes it and all of it'ts children
2626  * from the dcache.
2627  *
2628  * It is safe and reasonable to cache /proc entries for a task until
2629  * that task exits.  After that they just clog up the dcache with
2630  * useless entries, possibly causing useful dcache entries to be
2631  * flushed instead.  This routine is proved to flush those useless
2632  * dcache entries at process exit time.
2633  *
2634  * NOTE: This routine is just an optimization so it does not guarantee
2635  *       that no dcache entries will exist at process exit time it
2636  *       just makes it very unlikely that any will persist.
2637  */
2638
2639 void proc_flush_task(struct task_struct *task)
2640 {
2641         int i;
2642         struct pid *pid, *tgid = NULL;
2643         struct upid *upid;
2644
2645         pid = task_pid(task);
2646         if (thread_group_leader(task))
2647                 tgid = task_tgid(task);
2648
2649         for (i = 0; i <= pid->level; i++) {
2650                 upid = &pid->numbers[i];
2651                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2652                         tgid ? tgid->numbers[i].nr : 0);
2653         }
2654
2655         upid = &pid->numbers[pid->level];
2656         if (upid->nr == 1)
2657                 pid_ns_release_proc(upid->ns);
2658 }
2659
2660 static struct dentry *proc_pid_instantiate(struct inode *dir,
2661                                            struct dentry * dentry,
2662                                            struct task_struct *task, const void *ptr)
2663 {
2664         struct dentry *error = ERR_PTR(-ENOENT);
2665         struct inode *inode;
2666
2667         inode = proc_pid_make_inode(dir->i_sb, task);
2668         if (!inode)
2669                 goto out;
2670
2671         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2672         inode->i_op = &proc_tgid_base_inode_operations;
2673         inode->i_fop = &proc_tgid_base_operations;
2674         inode->i_flags|=S_IMMUTABLE;
2675
2676         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2677                 ARRAY_SIZE(tgid_base_stuff));
2678
2679         dentry->d_op = &pid_dentry_operations;
2680
2681         d_add(dentry, inode);
2682         /* Close the race of the process dying before we return the dentry */
2683         if (pid_revalidate(dentry, NULL))
2684                 error = NULL;
2685 out:
2686         return error;
2687 }
2688
2689 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2690 {
2691         struct dentry *result = ERR_PTR(-ENOENT);
2692         struct task_struct *task;
2693         unsigned tgid;
2694         struct pid_namespace *ns;
2695
2696         result = proc_base_lookup(dir, dentry);
2697         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2698                 goto out;
2699
2700         tgid = name_to_int(dentry);
2701         if (tgid == ~0U)
2702                 goto out;
2703
2704         ns = dentry->d_sb->s_fs_info;
2705         rcu_read_lock();
2706         task = find_task_by_pid_ns(tgid, ns);
2707         if (task)
2708                 get_task_struct(task);
2709         rcu_read_unlock();
2710         if (!task)
2711                 goto out;
2712
2713         result = proc_pid_instantiate(dir, dentry, task, NULL);
2714         put_task_struct(task);
2715 out:
2716         return result;
2717 }
2718
2719 /*
2720  * Find the first task with tgid >= tgid
2721  *
2722  */
2723 struct tgid_iter {
2724         unsigned int tgid;
2725         struct task_struct *task;
2726 };
2727 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2728 {
2729         struct pid *pid;
2730
2731         if (iter.task)
2732                 put_task_struct(iter.task);
2733         rcu_read_lock();
2734 retry:
2735         iter.task = NULL;
2736         pid = find_ge_pid(iter.tgid, ns);
2737         if (pid) {
2738                 iter.tgid = pid_nr_ns(pid, ns);
2739                 iter.task = pid_task(pid, PIDTYPE_PID);
2740                 /* What we to know is if the pid we have find is the
2741                  * pid of a thread_group_leader.  Testing for task
2742                  * being a thread_group_leader is the obvious thing
2743                  * todo but there is a window when it fails, due to
2744                  * the pid transfer logic in de_thread.
2745                  *
2746                  * So we perform the straight forward test of seeing
2747                  * if the pid we have found is the pid of a thread
2748                  * group leader, and don't worry if the task we have
2749                  * found doesn't happen to be a thread group leader.
2750                  * As we don't care in the case of readdir.
2751                  */
2752                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2753                         iter.tgid += 1;
2754                         goto retry;
2755                 }
2756                 get_task_struct(iter.task);
2757         }
2758         rcu_read_unlock();
2759         return iter;
2760 }
2761
2762 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2763
2764 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2765         struct tgid_iter iter)
2766 {
2767         char name[PROC_NUMBUF];
2768         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2769         return proc_fill_cache(filp, dirent, filldir, name, len,
2770                                 proc_pid_instantiate, iter.task, NULL);
2771 }
2772
2773 /* for the /proc/ directory itself, after non-process stuff has been done */
2774 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2775 {
2776         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2777         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2778         struct tgid_iter iter;
2779         struct pid_namespace *ns;
2780
2781         if (!reaper)
2782                 goto out_no_task;
2783
2784         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2785                 const struct pid_entry *p = &proc_base_stuff[nr];
2786                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2787                         goto out;
2788         }
2789
2790         ns = filp->f_dentry->d_sb->s_fs_info;
2791         iter.task = NULL;
2792         iter.tgid = filp->f_pos - TGID_OFFSET;
2793         for (iter = next_tgid(ns, iter);
2794              iter.task;
2795              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2796                 filp->f_pos = iter.tgid + TGID_OFFSET;
2797                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2798                         put_task_struct(iter.task);
2799                         goto out;
2800                 }
2801         }
2802         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2803 out:
2804         put_task_struct(reaper);
2805 out_no_task:
2806         return 0;
2807 }
2808
2809 /*
2810  * Tasks
2811  */
2812 static const struct pid_entry tid_base_stuff[] = {
2813         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2814         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2815         REG("environ",   S_IRUSR, proc_environ_operations),
2816         INF("auxv",      S_IRUSR, proc_pid_auxv),
2817         ONE("status",    S_IRUGO, proc_pid_status),
2818         ONE("personality", S_IRUSR, proc_pid_personality),
2819         INF("limits",    S_IRUSR, proc_pid_limits),
2820 #ifdef CONFIG_SCHED_DEBUG
2821         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2822 #endif
2823 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2824         INF("syscall",   S_IRUSR, proc_pid_syscall),
2825 #endif
2826         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2827         ONE("stat",      S_IRUGO, proc_tid_stat),
2828         ONE("statm",     S_IRUGO, proc_pid_statm),
2829         REG("maps",      S_IRUGO, proc_maps_operations),
2830 #ifdef CONFIG_NUMA
2831         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2832 #endif
2833         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2834         LNK("cwd",       proc_cwd_link),
2835         LNK("root",      proc_root_link),
2836         LNK("exe",       proc_exe_link),
2837         REG("mounts",    S_IRUGO, proc_mounts_operations),
2838         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2839 #ifdef CONFIG_PROC_PAGE_MONITOR
2840         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2841         REG("smaps",     S_IRUGO, proc_smaps_operations),
2842         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2843 #endif
2844 #ifdef CONFIG_SECURITY
2845         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2846 #endif
2847 #ifdef CONFIG_KALLSYMS
2848         INF("wchan",     S_IRUGO, proc_pid_wchan),
2849 #endif
2850 #ifdef CONFIG_STACKTRACE
2851         ONE("stack",      S_IRUSR, proc_pid_stack),
2852 #endif
2853 #ifdef CONFIG_SCHEDSTATS
2854         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2855 #endif
2856 #ifdef CONFIG_LATENCYTOP
2857         REG("latency",  S_IRUGO, proc_lstats_operations),
2858 #endif
2859 #ifdef CONFIG_PROC_PID_CPUSET
2860         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2861 #endif
2862 #ifdef CONFIG_CGROUPS
2863         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2864 #endif
2865         INF("oom_score", S_IRUGO, proc_oom_score),
2866         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2867 #ifdef CONFIG_AUDITSYSCALL
2868         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2869         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2870 #endif
2871 #ifdef CONFIG_FAULT_INJECTION
2872         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2873 #endif
2874 #ifdef CONFIG_TASK_IO_ACCOUNTING
2875         INF("io",       S_IRUGO, proc_tid_io_accounting),
2876 #endif
2877 };
2878
2879 static int proc_tid_base_readdir(struct file * filp,
2880                              void * dirent, filldir_t filldir)
2881 {
2882         return proc_pident_readdir(filp,dirent,filldir,
2883                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2884 }
2885
2886 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2887         return proc_pident_lookup(dir, dentry,
2888                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2889 }
2890
2891 static const struct file_operations proc_tid_base_operations = {
2892         .read           = generic_read_dir,
2893         .readdir        = proc_tid_base_readdir,
2894 };
2895
2896 static const struct inode_operations proc_tid_base_inode_operations = {
2897         .lookup         = proc_tid_base_lookup,
2898         .getattr        = pid_getattr,
2899         .setattr        = proc_setattr,
2900 };
2901
2902 static struct dentry *proc_task_instantiate(struct inode *dir,
2903         struct dentry *dentry, struct task_struct *task, const void *ptr)
2904 {
2905         struct dentry *error = ERR_PTR(-ENOENT);
2906         struct inode *inode;
2907         inode = proc_pid_make_inode(dir->i_sb, task);
2908
2909         if (!inode)
2910                 goto out;
2911         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2912         inode->i_op = &proc_tid_base_inode_operations;
2913         inode->i_fop = &proc_tid_base_operations;
2914         inode->i_flags|=S_IMMUTABLE;
2915
2916         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2917                 ARRAY_SIZE(tid_base_stuff));
2918
2919         dentry->d_op = &pid_dentry_operations;
2920
2921         d_add(dentry, inode);
2922         /* Close the race of the process dying before we return the dentry */
2923         if (pid_revalidate(dentry, NULL))
2924                 error = NULL;
2925 out:
2926         return error;
2927 }
2928
2929 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2930 {
2931         struct dentry *result = ERR_PTR(-ENOENT);
2932         struct task_struct *task;
2933         struct task_struct *leader = get_proc_task(dir);
2934         unsigned tid;
2935         struct pid_namespace *ns;
2936
2937         if (!leader)
2938                 goto out_no_task;
2939
2940         tid = name_to_int(dentry);
2941         if (tid == ~0U)
2942                 goto out;
2943
2944         ns = dentry->d_sb->s_fs_info;
2945         rcu_read_lock();
2946         task = find_task_by_pid_ns(tid, ns);
2947         if (task)
2948                 get_task_struct(task);
2949         rcu_read_unlock();
2950         if (!task)
2951                 goto out;
2952         if (!same_thread_group(leader, task))
2953                 goto out_drop_task;
2954
2955         result = proc_task_instantiate(dir, dentry, task, NULL);
2956 out_drop_task:
2957         put_task_struct(task);
2958 out:
2959         put_task_struct(leader);
2960 out_no_task:
2961         return result;
2962 }
2963
2964 /*
2965  * Find the first tid of a thread group to return to user space.
2966  *
2967  * Usually this is just the thread group leader, but if the users
2968  * buffer was too small or there was a seek into the middle of the
2969  * directory we have more work todo.
2970  *
2971  * In the case of a short read we start with find_task_by_pid.
2972  *
2973  * In the case of a seek we start with the leader and walk nr
2974  * threads past it.
2975  */
2976 static struct task_struct *first_tid(struct task_struct *leader,
2977                 int tid, int nr, struct pid_namespace *ns)
2978 {
2979         struct task_struct *pos;
2980
2981         rcu_read_lock();
2982         /* Attempt to start with the pid of a thread */
2983         if (tid && (nr > 0)) {
2984                 pos = find_task_by_pid_ns(tid, ns);
2985                 if (pos && (pos->group_leader == leader))
2986                         goto found;
2987         }
2988
2989         /* If nr exceeds the number of threads there is nothing todo */
2990         pos = NULL;
2991         if (nr && nr >= get_nr_threads(leader))
2992                 goto out;
2993
2994         /* If we haven't found our starting place yet start
2995          * with the leader and walk nr threads forward.
2996          */
2997         for (pos = leader; nr > 0; --nr) {
2998                 pos = next_thread(pos);
2999                 if (pos == leader) {
3000                         pos = NULL;
3001                         goto out;
3002                 }
3003         }
3004 found:
3005         get_task_struct(pos);
3006 out:
3007         rcu_read_unlock();
3008         return pos;
3009 }
3010
3011 /*
3012  * Find the next thread in the thread list.
3013  * Return NULL if there is an error or no next thread.
3014  *
3015  * The reference to the input task_struct is released.
3016  */
3017 static struct task_struct *next_tid(struct task_struct *start)
3018 {
3019         struct task_struct *pos = NULL;
3020         rcu_read_lock();
3021         if (pid_alive(start)) {
3022                 pos = next_thread(start);
3023                 if (thread_group_leader(pos))
3024                         pos = NULL;
3025                 else
3026                         get_task_struct(pos);
3027         }
3028         rcu_read_unlock();
3029         put_task_struct(start);
3030         return pos;
3031 }
3032
3033 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3034         struct task_struct *task, int tid)
3035 {
3036         char name[PROC_NUMBUF];
3037         int len = snprintf(name, sizeof(name), "%d", tid);
3038         return proc_fill_cache(filp, dirent, filldir, name, len,
3039                                 proc_task_instantiate, task, NULL);
3040 }
3041
3042 /* for the /proc/TGID/task/ directories */
3043 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3044 {
3045         struct dentry *dentry = filp->f_path.dentry;
3046         struct inode *inode = dentry->d_inode;
3047         struct task_struct *leader = NULL;
3048         struct task_struct *task;
3049         int retval = -ENOENT;
3050         ino_t ino;
3051         int tid;
3052         struct pid_namespace *ns;
3053
3054         task = get_proc_task(inode);
3055         if (!task)
3056                 goto out_no_task;
3057         rcu_read_lock();
3058         if (pid_alive(task)) {
3059                 leader = task->group_leader;
3060                 get_task_struct(leader);
3061         }
3062         rcu_read_unlock();
3063         put_task_struct(task);
3064         if (!leader)
3065                 goto out_no_task;
3066         retval = 0;
3067
3068         switch ((unsigned long)filp->f_pos) {
3069         case 0:
3070                 ino = inode->i_ino;
3071                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3072                         goto out;
3073                 filp->f_pos++;
3074                 /* fall through */
3075         case 1:
3076                 ino = parent_ino(dentry);
3077                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3078                         goto out;
3079                 filp->f_pos++;
3080                 /* fall through */
3081         }
3082
3083         /* f_version caches the tgid value that the last readdir call couldn't
3084          * return. lseek aka telldir automagically resets f_version to 0.
3085          */
3086         ns = filp->f_dentry->d_sb->s_fs_info;
3087         tid = (int)filp->f_version;
3088         filp->f_version = 0;
3089         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3090              task;
3091              task = next_tid(task), filp->f_pos++) {
3092                 tid = task_pid_nr_ns(task, ns);
3093                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3094                         /* returning this tgid failed, save it as the first
3095                          * pid for the next readir call */
3096                         filp->f_version = (u64)tid;
3097                         put_task_struct(task);
3098                         break;
3099                 }
3100         }
3101 out:
3102         put_task_struct(leader);
3103 out_no_task:
3104         return retval;
3105 }
3106
3107 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3108 {
3109         struct inode *inode = dentry->d_inode;
3110         struct task_struct *p = get_proc_task(inode);
3111         generic_fillattr(inode, stat);
3112
3113         if (p) {
3114                 stat->nlink += get_nr_threads(p);
3115                 put_task_struct(p);
3116         }
3117
3118         return 0;
3119 }
3120
3121 static const struct inode_operations proc_task_inode_operations = {
3122         .lookup         = proc_task_lookup,
3123         .getattr        = proc_task_getattr,
3124         .setattr        = proc_setattr,
3125 };
3126
3127 static const struct file_operations proc_task_operations = {
3128         .read           = generic_read_dir,
3129         .readdir        = proc_task_readdir,
3130 };