fix autofs/afs/etc. magic mountpoint breakage
[linux-2.6.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/rcupdate.h>
67 #include <linux/kallsyms.h>
68 #include <linux/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;
238
239         if (mutex_lock_killable(&task->cred_guard_mutex))
240                 return NULL;
241
242         mm = get_task_mm(task);
243         if (mm && mm != current->mm &&
244                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
245                 mmput(mm);
246                 mm = NULL;
247         }
248         mutex_unlock(&task->cred_guard_mutex);
249
250         return mm;
251 }
252
253 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
254 {
255         int res = 0;
256         unsigned int len;
257         struct mm_struct *mm = get_task_mm(task);
258         if (!mm)
259                 goto out;
260         if (!mm->arg_end)
261                 goto out_mm;    /* Shh! No looking before we're done */
262
263         len = mm->arg_end - mm->arg_start;
264  
265         if (len > PAGE_SIZE)
266                 len = PAGE_SIZE;
267  
268         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
269
270         // If the nul at the end of args has been overwritten, then
271         // assume application is using setproctitle(3).
272         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
273                 len = strnlen(buffer, res);
274                 if (len < res) {
275                     res = len;
276                 } else {
277                         len = mm->env_end - mm->env_start;
278                         if (len > PAGE_SIZE - res)
279                                 len = PAGE_SIZE - res;
280                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
281                         res = strnlen(buffer, res);
282                 }
283         }
284 out_mm:
285         mmput(mm);
286 out:
287         return res;
288 }
289
290 static int proc_pid_auxv(struct task_struct *task, char *buffer)
291 {
292         int res = 0;
293         struct mm_struct *mm = get_task_mm(task);
294         if (mm) {
295                 unsigned int nwords = 0;
296                 do {
297                         nwords += 2;
298                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
299                 res = nwords * sizeof(mm->saved_auxv[0]);
300                 if (res > PAGE_SIZE)
301                         res = PAGE_SIZE;
302                 memcpy(buffer, mm->saved_auxv, res);
303                 mmput(mm);
304         }
305         return res;
306 }
307
308
309 #ifdef CONFIG_KALLSYMS
310 /*
311  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
312  * Returns the resolved symbol.  If that fails, simply return the address.
313  */
314 static int proc_pid_wchan(struct task_struct *task, char *buffer)
315 {
316         unsigned long wchan;
317         char symname[KSYM_NAME_LEN];
318
319         wchan = get_wchan(task);
320
321         if (lookup_symbol_name(wchan, symname) < 0)
322                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
323                         return 0;
324                 else
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->group_leader, 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", "seconds"},
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 = POLLIN | POLLRDNORM;
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 | POLLPRI;
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 = OOM_DISABLE;
1003         unsigned long flags;
1004
1005         if (!task)
1006                 return -ESRCH;
1007
1008         if (lock_task_sighand(task, &flags)) {
1009                 oom_adjust = task->signal->oom_adj;
1010                 unlock_task_sighand(task, &flags);
1011         }
1012
1013         put_task_struct(task);
1014
1015         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1016
1017         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1018 }
1019
1020 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1021                                 size_t count, loff_t *ppos)
1022 {
1023         struct task_struct *task;
1024         char buffer[PROC_NUMBUF];
1025         long oom_adjust;
1026         unsigned long flags;
1027         int err;
1028
1029         memset(buffer, 0, sizeof(buffer));
1030         if (count > sizeof(buffer) - 1)
1031                 count = sizeof(buffer) - 1;
1032         if (copy_from_user(buffer, buf, count))
1033                 return -EFAULT;
1034
1035         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1036         if (err)
1037                 return -EINVAL;
1038         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1039              oom_adjust != OOM_DISABLE)
1040                 return -EINVAL;
1041
1042         task = get_proc_task(file->f_path.dentry->d_inode);
1043         if (!task)
1044                 return -ESRCH;
1045         if (!lock_task_sighand(task, &flags)) {
1046                 put_task_struct(task);
1047                 return -ESRCH;
1048         }
1049
1050         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1051                 unlock_task_sighand(task, &flags);
1052                 put_task_struct(task);
1053                 return -EACCES;
1054         }
1055
1056         task->signal->oom_adj = oom_adjust;
1057
1058         unlock_task_sighand(task, &flags);
1059         put_task_struct(task);
1060
1061         return count;
1062 }
1063
1064 static const struct file_operations proc_oom_adjust_operations = {
1065         .read           = oom_adjust_read,
1066         .write          = oom_adjust_write,
1067 };
1068
1069 #ifdef CONFIG_AUDITSYSCALL
1070 #define TMPBUFLEN 21
1071 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1072                                   size_t count, loff_t *ppos)
1073 {
1074         struct inode * inode = file->f_path.dentry->d_inode;
1075         struct task_struct *task = get_proc_task(inode);
1076         ssize_t length;
1077         char tmpbuf[TMPBUFLEN];
1078
1079         if (!task)
1080                 return -ESRCH;
1081         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1082                                 audit_get_loginuid(task));
1083         put_task_struct(task);
1084         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1085 }
1086
1087 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1088                                    size_t count, loff_t *ppos)
1089 {
1090         struct inode * inode = file->f_path.dentry->d_inode;
1091         char *page, *tmp;
1092         ssize_t length;
1093         uid_t loginuid;
1094
1095         if (!capable(CAP_AUDIT_CONTROL))
1096                 return -EPERM;
1097
1098         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1099                 return -EPERM;
1100
1101         if (count >= PAGE_SIZE)
1102                 count = PAGE_SIZE - 1;
1103
1104         if (*ppos != 0) {
1105                 /* No partial writes. */
1106                 return -EINVAL;
1107         }
1108         page = (char*)__get_free_page(GFP_TEMPORARY);
1109         if (!page)
1110                 return -ENOMEM;
1111         length = -EFAULT;
1112         if (copy_from_user(page, buf, count))
1113                 goto out_free_page;
1114
1115         page[count] = '\0';
1116         loginuid = simple_strtoul(page, &tmp, 10);
1117         if (tmp == page) {
1118                 length = -EINVAL;
1119                 goto out_free_page;
1120
1121         }
1122         length = audit_set_loginuid(current, loginuid);
1123         if (likely(length == 0))
1124                 length = count;
1125
1126 out_free_page:
1127         free_page((unsigned long) page);
1128         return length;
1129 }
1130
1131 static const struct file_operations proc_loginuid_operations = {
1132         .read           = proc_loginuid_read,
1133         .write          = proc_loginuid_write,
1134 };
1135
1136 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1137                                   size_t count, loff_t *ppos)
1138 {
1139         struct inode * inode = file->f_path.dentry->d_inode;
1140         struct task_struct *task = get_proc_task(inode);
1141         ssize_t length;
1142         char tmpbuf[TMPBUFLEN];
1143
1144         if (!task)
1145                 return -ESRCH;
1146         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1147                                 audit_get_sessionid(task));
1148         put_task_struct(task);
1149         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1150 }
1151
1152 static const struct file_operations proc_sessionid_operations = {
1153         .read           = proc_sessionid_read,
1154 };
1155 #endif
1156
1157 #ifdef CONFIG_FAULT_INJECTION
1158 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1159                                       size_t count, loff_t *ppos)
1160 {
1161         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1162         char buffer[PROC_NUMBUF];
1163         size_t len;
1164         int make_it_fail;
1165
1166         if (!task)
1167                 return -ESRCH;
1168         make_it_fail = task->make_it_fail;
1169         put_task_struct(task);
1170
1171         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1172
1173         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1174 }
1175
1176 static ssize_t proc_fault_inject_write(struct file * file,
1177                         const char __user * buf, size_t count, loff_t *ppos)
1178 {
1179         struct task_struct *task;
1180         char buffer[PROC_NUMBUF], *end;
1181         int make_it_fail;
1182
1183         if (!capable(CAP_SYS_RESOURCE))
1184                 return -EPERM;
1185         memset(buffer, 0, sizeof(buffer));
1186         if (count > sizeof(buffer) - 1)
1187                 count = sizeof(buffer) - 1;
1188         if (copy_from_user(buffer, buf, count))
1189                 return -EFAULT;
1190         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1191         if (*end)
1192                 return -EINVAL;
1193         task = get_proc_task(file->f_dentry->d_inode);
1194         if (!task)
1195                 return -ESRCH;
1196         task->make_it_fail = make_it_fail;
1197         put_task_struct(task);
1198
1199         return count;
1200 }
1201
1202 static const struct file_operations proc_fault_inject_operations = {
1203         .read           = proc_fault_inject_read,
1204         .write          = proc_fault_inject_write,
1205 };
1206 #endif
1207
1208
1209 #ifdef CONFIG_SCHED_DEBUG
1210 /*
1211  * Print out various scheduling related per-task fields:
1212  */
1213 static int sched_show(struct seq_file *m, void *v)
1214 {
1215         struct inode *inode = m->private;
1216         struct task_struct *p;
1217
1218         p = get_proc_task(inode);
1219         if (!p)
1220                 return -ESRCH;
1221         proc_sched_show_task(p, m);
1222
1223         put_task_struct(p);
1224
1225         return 0;
1226 }
1227
1228 static ssize_t
1229 sched_write(struct file *file, const char __user *buf,
1230             size_t count, loff_t *offset)
1231 {
1232         struct inode *inode = file->f_path.dentry->d_inode;
1233         struct task_struct *p;
1234
1235         p = get_proc_task(inode);
1236         if (!p)
1237                 return -ESRCH;
1238         proc_sched_set_task(p);
1239
1240         put_task_struct(p);
1241
1242         return count;
1243 }
1244
1245 static int sched_open(struct inode *inode, struct file *filp)
1246 {
1247         int ret;
1248
1249         ret = single_open(filp, sched_show, NULL);
1250         if (!ret) {
1251                 struct seq_file *m = filp->private_data;
1252
1253                 m->private = inode;
1254         }
1255         return ret;
1256 }
1257
1258 static const struct file_operations proc_pid_sched_operations = {
1259         .open           = sched_open,
1260         .read           = seq_read,
1261         .write          = sched_write,
1262         .llseek         = seq_lseek,
1263         .release        = single_release,
1264 };
1265
1266 #endif
1267
1268 static ssize_t comm_write(struct file *file, const char __user *buf,
1269                                 size_t count, loff_t *offset)
1270 {
1271         struct inode *inode = file->f_path.dentry->d_inode;
1272         struct task_struct *p;
1273         char buffer[TASK_COMM_LEN];
1274
1275         memset(buffer, 0, sizeof(buffer));
1276         if (count > sizeof(buffer) - 1)
1277                 count = sizeof(buffer) - 1;
1278         if (copy_from_user(buffer, buf, count))
1279                 return -EFAULT;
1280
1281         p = get_proc_task(inode);
1282         if (!p)
1283                 return -ESRCH;
1284
1285         if (same_thread_group(current, p))
1286                 set_task_comm(p, buffer);
1287         else
1288                 count = -EINVAL;
1289
1290         put_task_struct(p);
1291
1292         return count;
1293 }
1294
1295 static int comm_show(struct seq_file *m, void *v)
1296 {
1297         struct inode *inode = m->private;
1298         struct task_struct *p;
1299
1300         p = get_proc_task(inode);
1301         if (!p)
1302                 return -ESRCH;
1303
1304         task_lock(p);
1305         seq_printf(m, "%s\n", p->comm);
1306         task_unlock(p);
1307
1308         put_task_struct(p);
1309
1310         return 0;
1311 }
1312
1313 static int comm_open(struct inode *inode, struct file *filp)
1314 {
1315         int ret;
1316
1317         ret = single_open(filp, comm_show, NULL);
1318         if (!ret) {
1319                 struct seq_file *m = filp->private_data;
1320
1321                 m->private = inode;
1322         }
1323         return ret;
1324 }
1325
1326 static const struct file_operations proc_pid_set_comm_operations = {
1327         .open           = comm_open,
1328         .read           = seq_read,
1329         .write          = comm_write,
1330         .llseek         = seq_lseek,
1331         .release        = single_release,
1332 };
1333
1334 /*
1335  * We added or removed a vma mapping the executable. The vmas are only mapped
1336  * during exec and are not mapped with the mmap system call.
1337  * Callers must hold down_write() on the mm's mmap_sem for these
1338  */
1339 void added_exe_file_vma(struct mm_struct *mm)
1340 {
1341         mm->num_exe_file_vmas++;
1342 }
1343
1344 void removed_exe_file_vma(struct mm_struct *mm)
1345 {
1346         mm->num_exe_file_vmas--;
1347         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1348                 fput(mm->exe_file);
1349                 mm->exe_file = NULL;
1350         }
1351
1352 }
1353
1354 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1355 {
1356         if (new_exe_file)
1357                 get_file(new_exe_file);
1358         if (mm->exe_file)
1359                 fput(mm->exe_file);
1360         mm->exe_file = new_exe_file;
1361         mm->num_exe_file_vmas = 0;
1362 }
1363
1364 struct file *get_mm_exe_file(struct mm_struct *mm)
1365 {
1366         struct file *exe_file;
1367
1368         /* We need mmap_sem to protect against races with removal of
1369          * VM_EXECUTABLE vmas */
1370         down_read(&mm->mmap_sem);
1371         exe_file = mm->exe_file;
1372         if (exe_file)
1373                 get_file(exe_file);
1374         up_read(&mm->mmap_sem);
1375         return exe_file;
1376 }
1377
1378 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1379 {
1380         /* It's safe to write the exe_file pointer without exe_file_lock because
1381          * this is called during fork when the task is not yet in /proc */
1382         newmm->exe_file = get_mm_exe_file(oldmm);
1383 }
1384
1385 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1386 {
1387         struct task_struct *task;
1388         struct mm_struct *mm;
1389         struct file *exe_file;
1390
1391         task = get_proc_task(inode);
1392         if (!task)
1393                 return -ENOENT;
1394         mm = get_task_mm(task);
1395         put_task_struct(task);
1396         if (!mm)
1397                 return -ENOENT;
1398         exe_file = get_mm_exe_file(mm);
1399         mmput(mm);
1400         if (exe_file) {
1401                 *exe_path = exe_file->f_path;
1402                 path_get(&exe_file->f_path);
1403                 fput(exe_file);
1404                 return 0;
1405         } else
1406                 return -ENOENT;
1407 }
1408
1409 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1410 {
1411         struct inode *inode = dentry->d_inode;
1412         int error = -EACCES;
1413
1414         /* We don't need a base pointer in the /proc filesystem */
1415         path_put(&nd->path);
1416
1417         /* Are we allowed to snoop on the tasks file descriptors? */
1418         if (!proc_fd_access_allowed(inode))
1419                 goto out;
1420
1421         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1422 out:
1423         return ERR_PTR(error);
1424 }
1425
1426 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1427 {
1428         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1429         char *pathname;
1430         int len;
1431
1432         if (!tmp)
1433                 return -ENOMEM;
1434
1435         pathname = d_path(path, tmp, PAGE_SIZE);
1436         len = PTR_ERR(pathname);
1437         if (IS_ERR(pathname))
1438                 goto out;
1439         len = tmp + PAGE_SIZE - 1 - pathname;
1440
1441         if (len > buflen)
1442                 len = buflen;
1443         if (copy_to_user(buffer, pathname, len))
1444                 len = -EFAULT;
1445  out:
1446         free_page((unsigned long)tmp);
1447         return len;
1448 }
1449
1450 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1451 {
1452         int error = -EACCES;
1453         struct inode *inode = dentry->d_inode;
1454         struct path path;
1455
1456         /* Are we allowed to snoop on the tasks file descriptors? */
1457         if (!proc_fd_access_allowed(inode))
1458                 goto out;
1459
1460         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1461         if (error)
1462                 goto out;
1463
1464         error = do_proc_readlink(&path, buffer, buflen);
1465         path_put(&path);
1466 out:
1467         return error;
1468 }
1469
1470 static const struct inode_operations proc_pid_link_inode_operations = {
1471         .readlink       = proc_pid_readlink,
1472         .follow_link    = proc_pid_follow_link,
1473         .setattr        = proc_setattr,
1474 };
1475
1476
1477 /* building an inode */
1478
1479 static int task_dumpable(struct task_struct *task)
1480 {
1481         int dumpable = 0;
1482         struct mm_struct *mm;
1483
1484         task_lock(task);
1485         mm = task->mm;
1486         if (mm)
1487                 dumpable = get_dumpable(mm);
1488         task_unlock(task);
1489         if(dumpable == 1)
1490                 return 1;
1491         return 0;
1492 }
1493
1494
1495 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1496 {
1497         struct inode * inode;
1498         struct proc_inode *ei;
1499         const struct cred *cred;
1500
1501         /* We need a new inode */
1502
1503         inode = new_inode(sb);
1504         if (!inode)
1505                 goto out;
1506
1507         /* Common stuff */
1508         ei = PROC_I(inode);
1509         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1510         inode->i_op = &proc_def_inode_operations;
1511
1512         /*
1513          * grab the reference to task.
1514          */
1515         ei->pid = get_task_pid(task, PIDTYPE_PID);
1516         if (!ei->pid)
1517                 goto out_unlock;
1518
1519         if (task_dumpable(task)) {
1520                 rcu_read_lock();
1521                 cred = __task_cred(task);
1522                 inode->i_uid = cred->euid;
1523                 inode->i_gid = cred->egid;
1524                 rcu_read_unlock();
1525         }
1526         security_task_to_inode(task, inode);
1527
1528 out:
1529         return inode;
1530
1531 out_unlock:
1532         iput(inode);
1533         return NULL;
1534 }
1535
1536 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1537 {
1538         struct inode *inode = dentry->d_inode;
1539         struct task_struct *task;
1540         const struct cred *cred;
1541
1542         generic_fillattr(inode, stat);
1543
1544         rcu_read_lock();
1545         stat->uid = 0;
1546         stat->gid = 0;
1547         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1548         if (task) {
1549                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1550                     task_dumpable(task)) {
1551                         cred = __task_cred(task);
1552                         stat->uid = cred->euid;
1553                         stat->gid = cred->egid;
1554                 }
1555         }
1556         rcu_read_unlock();
1557         return 0;
1558 }
1559
1560 /* dentry stuff */
1561
1562 /*
1563  *      Exceptional case: normally we are not allowed to unhash a busy
1564  * directory. In this case, however, we can do it - no aliasing problems
1565  * due to the way we treat inodes.
1566  *
1567  * Rewrite the inode's ownerships here because the owning task may have
1568  * performed a setuid(), etc.
1569  *
1570  * Before the /proc/pid/status file was created the only way to read
1571  * the effective uid of a /process was to stat /proc/pid.  Reading
1572  * /proc/pid/status is slow enough that procps and other packages
1573  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1574  * made this apply to all per process world readable and executable
1575  * directories.
1576  */
1577 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1578 {
1579         struct inode *inode = dentry->d_inode;
1580         struct task_struct *task = get_proc_task(inode);
1581         const struct cred *cred;
1582
1583         if (task) {
1584                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1585                     task_dumpable(task)) {
1586                         rcu_read_lock();
1587                         cred = __task_cred(task);
1588                         inode->i_uid = cred->euid;
1589                         inode->i_gid = cred->egid;
1590                         rcu_read_unlock();
1591                 } else {
1592                         inode->i_uid = 0;
1593                         inode->i_gid = 0;
1594                 }
1595                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1596                 security_task_to_inode(task, inode);
1597                 put_task_struct(task);
1598                 return 1;
1599         }
1600         d_drop(dentry);
1601         return 0;
1602 }
1603
1604 static int pid_delete_dentry(struct dentry * dentry)
1605 {
1606         /* Is the task we represent dead?
1607          * If so, then don't put the dentry on the lru list,
1608          * kill it immediately.
1609          */
1610         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1611 }
1612
1613 static const struct dentry_operations pid_dentry_operations =
1614 {
1615         .d_revalidate   = pid_revalidate,
1616         .d_delete       = pid_delete_dentry,
1617 };
1618
1619 /* Lookups */
1620
1621 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1622                                 struct task_struct *, const void *);
1623
1624 /*
1625  * Fill a directory entry.
1626  *
1627  * If possible create the dcache entry and derive our inode number and
1628  * file type from dcache entry.
1629  *
1630  * Since all of the proc inode numbers are dynamically generated, the inode
1631  * numbers do not exist until the inode is cache.  This means creating the
1632  * the dcache entry in readdir is necessary to keep the inode numbers
1633  * reported by readdir in sync with the inode numbers reported
1634  * by stat.
1635  */
1636 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1637         char *name, int len,
1638         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1639 {
1640         struct dentry *child, *dir = filp->f_path.dentry;
1641         struct inode *inode;
1642         struct qstr qname;
1643         ino_t ino = 0;
1644         unsigned type = DT_UNKNOWN;
1645
1646         qname.name = name;
1647         qname.len  = len;
1648         qname.hash = full_name_hash(name, len);
1649
1650         child = d_lookup(dir, &qname);
1651         if (!child) {
1652                 struct dentry *new;
1653                 new = d_alloc(dir, &qname);
1654                 if (new) {
1655                         child = instantiate(dir->d_inode, new, task, ptr);
1656                         if (child)
1657                                 dput(new);
1658                         else
1659                                 child = new;
1660                 }
1661         }
1662         if (!child || IS_ERR(child) || !child->d_inode)
1663                 goto end_instantiate;
1664         inode = child->d_inode;
1665         if (inode) {
1666                 ino = inode->i_ino;
1667                 type = inode->i_mode >> 12;
1668         }
1669         dput(child);
1670 end_instantiate:
1671         if (!ino)
1672                 ino = find_inode_number(dir, &qname);
1673         if (!ino)
1674                 ino = 1;
1675         return filldir(dirent, name, len, filp->f_pos, ino, type);
1676 }
1677
1678 static unsigned name_to_int(struct dentry *dentry)
1679 {
1680         const char *name = dentry->d_name.name;
1681         int len = dentry->d_name.len;
1682         unsigned n = 0;
1683
1684         if (len > 1 && *name == '0')
1685                 goto out;
1686         while (len-- > 0) {
1687                 unsigned c = *name++ - '0';
1688                 if (c > 9)
1689                         goto out;
1690                 if (n >= (~0U-9)/10)
1691                         goto out;
1692                 n *= 10;
1693                 n += c;
1694         }
1695         return n;
1696 out:
1697         return ~0U;
1698 }
1699
1700 #define PROC_FDINFO_MAX 64
1701
1702 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1703 {
1704         struct task_struct *task = get_proc_task(inode);
1705         struct files_struct *files = NULL;
1706         struct file *file;
1707         int fd = proc_fd(inode);
1708
1709         if (task) {
1710                 files = get_files_struct(task);
1711                 put_task_struct(task);
1712         }
1713         if (files) {
1714                 /*
1715                  * We are not taking a ref to the file structure, so we must
1716                  * hold ->file_lock.
1717                  */
1718                 spin_lock(&files->file_lock);
1719                 file = fcheck_files(files, fd);
1720                 if (file) {
1721                         if (path) {
1722                                 *path = file->f_path;
1723                                 path_get(&file->f_path);
1724                         }
1725                         if (info)
1726                                 snprintf(info, PROC_FDINFO_MAX,
1727                                          "pos:\t%lli\n"
1728                                          "flags:\t0%o\n",
1729                                          (long long) file->f_pos,
1730                                          file->f_flags);
1731                         spin_unlock(&files->file_lock);
1732                         put_files_struct(files);
1733                         return 0;
1734                 }
1735                 spin_unlock(&files->file_lock);
1736                 put_files_struct(files);
1737         }
1738         return -ENOENT;
1739 }
1740
1741 static int proc_fd_link(struct inode *inode, struct path *path)
1742 {
1743         return proc_fd_info(inode, path, NULL);
1744 }
1745
1746 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1747 {
1748         struct inode *inode = dentry->d_inode;
1749         struct task_struct *task = get_proc_task(inode);
1750         int fd = proc_fd(inode);
1751         struct files_struct *files;
1752         const struct cred *cred;
1753
1754         if (task) {
1755                 files = get_files_struct(task);
1756                 if (files) {
1757                         rcu_read_lock();
1758                         if (fcheck_files(files, fd)) {
1759                                 rcu_read_unlock();
1760                                 put_files_struct(files);
1761                                 if (task_dumpable(task)) {
1762                                         rcu_read_lock();
1763                                         cred = __task_cred(task);
1764                                         inode->i_uid = cred->euid;
1765                                         inode->i_gid = cred->egid;
1766                                         rcu_read_unlock();
1767                                 } else {
1768                                         inode->i_uid = 0;
1769                                         inode->i_gid = 0;
1770                                 }
1771                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1772                                 security_task_to_inode(task, inode);
1773                                 put_task_struct(task);
1774                                 return 1;
1775                         }
1776                         rcu_read_unlock();
1777                         put_files_struct(files);
1778                 }
1779                 put_task_struct(task);
1780         }
1781         d_drop(dentry);
1782         return 0;
1783 }
1784
1785 static const struct dentry_operations tid_fd_dentry_operations =
1786 {
1787         .d_revalidate   = tid_fd_revalidate,
1788         .d_delete       = pid_delete_dentry,
1789 };
1790
1791 static struct dentry *proc_fd_instantiate(struct inode *dir,
1792         struct dentry *dentry, struct task_struct *task, const void *ptr)
1793 {
1794         unsigned fd = *(const unsigned *)ptr;
1795         struct file *file;
1796         struct files_struct *files;
1797         struct inode *inode;
1798         struct proc_inode *ei;
1799         struct dentry *error = ERR_PTR(-ENOENT);
1800
1801         inode = proc_pid_make_inode(dir->i_sb, task);
1802         if (!inode)
1803                 goto out;
1804         ei = PROC_I(inode);
1805         ei->fd = fd;
1806         files = get_files_struct(task);
1807         if (!files)
1808                 goto out_iput;
1809         inode->i_mode = S_IFLNK;
1810
1811         /*
1812          * We are not taking a ref to the file structure, so we must
1813          * hold ->file_lock.
1814          */
1815         spin_lock(&files->file_lock);
1816         file = fcheck_files(files, fd);
1817         if (!file)
1818                 goto out_unlock;
1819         if (file->f_mode & FMODE_READ)
1820                 inode->i_mode |= S_IRUSR | S_IXUSR;
1821         if (file->f_mode & FMODE_WRITE)
1822                 inode->i_mode |= S_IWUSR | S_IXUSR;
1823         spin_unlock(&files->file_lock);
1824         put_files_struct(files);
1825
1826         inode->i_op = &proc_pid_link_inode_operations;
1827         inode->i_size = 64;
1828         ei->op.proc_get_link = proc_fd_link;
1829         dentry->d_op = &tid_fd_dentry_operations;
1830         d_add(dentry, inode);
1831         /* Close the race of the process dying before we return the dentry */
1832         if (tid_fd_revalidate(dentry, NULL))
1833                 error = NULL;
1834
1835  out:
1836         return error;
1837 out_unlock:
1838         spin_unlock(&files->file_lock);
1839         put_files_struct(files);
1840 out_iput:
1841         iput(inode);
1842         goto out;
1843 }
1844
1845 static struct dentry *proc_lookupfd_common(struct inode *dir,
1846                                            struct dentry *dentry,
1847                                            instantiate_t instantiate)
1848 {
1849         struct task_struct *task = get_proc_task(dir);
1850         unsigned fd = name_to_int(dentry);
1851         struct dentry *result = ERR_PTR(-ENOENT);
1852
1853         if (!task)
1854                 goto out_no_task;
1855         if (fd == ~0U)
1856                 goto out;
1857
1858         result = instantiate(dir, dentry, task, &fd);
1859 out:
1860         put_task_struct(task);
1861 out_no_task:
1862         return result;
1863 }
1864
1865 static int proc_readfd_common(struct file * filp, void * dirent,
1866                               filldir_t filldir, instantiate_t instantiate)
1867 {
1868         struct dentry *dentry = filp->f_path.dentry;
1869         struct inode *inode = dentry->d_inode;
1870         struct task_struct *p = get_proc_task(inode);
1871         unsigned int fd, ino;
1872         int retval;
1873         struct files_struct * files;
1874
1875         retval = -ENOENT;
1876         if (!p)
1877                 goto out_no_task;
1878         retval = 0;
1879
1880         fd = filp->f_pos;
1881         switch (fd) {
1882                 case 0:
1883                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1884                                 goto out;
1885                         filp->f_pos++;
1886                 case 1:
1887                         ino = parent_ino(dentry);
1888                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1889                                 goto out;
1890                         filp->f_pos++;
1891                 default:
1892                         files = get_files_struct(p);
1893                         if (!files)
1894                                 goto out;
1895                         rcu_read_lock();
1896                         for (fd = filp->f_pos-2;
1897                              fd < files_fdtable(files)->max_fds;
1898                              fd++, filp->f_pos++) {
1899                                 char name[PROC_NUMBUF];
1900                                 int len;
1901
1902                                 if (!fcheck_files(files, fd))
1903                                         continue;
1904                                 rcu_read_unlock();
1905
1906                                 len = snprintf(name, sizeof(name), "%d", fd);
1907                                 if (proc_fill_cache(filp, dirent, filldir,
1908                                                     name, len, instantiate,
1909                                                     p, &fd) < 0) {
1910                                         rcu_read_lock();
1911                                         break;
1912                                 }
1913                                 rcu_read_lock();
1914                         }
1915                         rcu_read_unlock();
1916                         put_files_struct(files);
1917         }
1918 out:
1919         put_task_struct(p);
1920 out_no_task:
1921         return retval;
1922 }
1923
1924 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1925                                     struct nameidata *nd)
1926 {
1927         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1928 }
1929
1930 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1931 {
1932         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1933 }
1934
1935 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1936                                       size_t len, loff_t *ppos)
1937 {
1938         char tmp[PROC_FDINFO_MAX];
1939         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1940         if (!err)
1941                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1942         return err;
1943 }
1944
1945 static const struct file_operations proc_fdinfo_file_operations = {
1946         .open           = nonseekable_open,
1947         .read           = proc_fdinfo_read,
1948 };
1949
1950 static const struct file_operations proc_fd_operations = {
1951         .read           = generic_read_dir,
1952         .readdir        = proc_readfd,
1953 };
1954
1955 /*
1956  * /proc/pid/fd needs a special permission handler so that a process can still
1957  * access /proc/self/fd after it has executed a setuid().
1958  */
1959 static int proc_fd_permission(struct inode *inode, int mask)
1960 {
1961         int rv;
1962
1963         rv = generic_permission(inode, mask, NULL);
1964         if (rv == 0)
1965                 return 0;
1966         if (task_pid(current) == proc_pid(inode))
1967                 rv = 0;
1968         return rv;
1969 }
1970
1971 /*
1972  * proc directories can do almost nothing..
1973  */
1974 static const struct inode_operations proc_fd_inode_operations = {
1975         .lookup         = proc_lookupfd,
1976         .permission     = proc_fd_permission,
1977         .setattr        = proc_setattr,
1978 };
1979
1980 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1981         struct dentry *dentry, struct task_struct *task, const void *ptr)
1982 {
1983         unsigned fd = *(unsigned *)ptr;
1984         struct inode *inode;
1985         struct proc_inode *ei;
1986         struct dentry *error = ERR_PTR(-ENOENT);
1987
1988         inode = proc_pid_make_inode(dir->i_sb, task);
1989         if (!inode)
1990                 goto out;
1991         ei = PROC_I(inode);
1992         ei->fd = fd;
1993         inode->i_mode = S_IFREG | S_IRUSR;
1994         inode->i_fop = &proc_fdinfo_file_operations;
1995         dentry->d_op = &tid_fd_dentry_operations;
1996         d_add(dentry, inode);
1997         /* Close the race of the process dying before we return the dentry */
1998         if (tid_fd_revalidate(dentry, NULL))
1999                 error = NULL;
2000
2001  out:
2002         return error;
2003 }
2004
2005 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2006                                         struct dentry *dentry,
2007                                         struct nameidata *nd)
2008 {
2009         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2010 }
2011
2012 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2013 {
2014         return proc_readfd_common(filp, dirent, filldir,
2015                                   proc_fdinfo_instantiate);
2016 }
2017
2018 static const struct file_operations proc_fdinfo_operations = {
2019         .read           = generic_read_dir,
2020         .readdir        = proc_readfdinfo,
2021 };
2022
2023 /*
2024  * proc directories can do almost nothing..
2025  */
2026 static const struct inode_operations proc_fdinfo_inode_operations = {
2027         .lookup         = proc_lookupfdinfo,
2028         .setattr        = proc_setattr,
2029 };
2030
2031
2032 static struct dentry *proc_pident_instantiate(struct inode *dir,
2033         struct dentry *dentry, struct task_struct *task, const void *ptr)
2034 {
2035         const struct pid_entry *p = ptr;
2036         struct inode *inode;
2037         struct proc_inode *ei;
2038         struct dentry *error = ERR_PTR(-ENOENT);
2039
2040         inode = proc_pid_make_inode(dir->i_sb, task);
2041         if (!inode)
2042                 goto out;
2043
2044         ei = PROC_I(inode);
2045         inode->i_mode = p->mode;
2046         if (S_ISDIR(inode->i_mode))
2047                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2048         if (p->iop)
2049                 inode->i_op = p->iop;
2050         if (p->fop)
2051                 inode->i_fop = p->fop;
2052         ei->op = p->op;
2053         dentry->d_op = &pid_dentry_operations;
2054         d_add(dentry, inode);
2055         /* Close the race of the process dying before we return the dentry */
2056         if (pid_revalidate(dentry, NULL))
2057                 error = NULL;
2058 out:
2059         return error;
2060 }
2061
2062 static struct dentry *proc_pident_lookup(struct inode *dir, 
2063                                          struct dentry *dentry,
2064                                          const struct pid_entry *ents,
2065                                          unsigned int nents)
2066 {
2067         struct dentry *error;
2068         struct task_struct *task = get_proc_task(dir);
2069         const struct pid_entry *p, *last;
2070
2071         error = ERR_PTR(-ENOENT);
2072
2073         if (!task)
2074                 goto out_no_task;
2075
2076         /*
2077          * Yes, it does not scale. And it should not. Don't add
2078          * new entries into /proc/<tgid>/ without very good reasons.
2079          */
2080         last = &ents[nents - 1];
2081         for (p = ents; p <= last; p++) {
2082                 if (p->len != dentry->d_name.len)
2083                         continue;
2084                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2085                         break;
2086         }
2087         if (p > last)
2088                 goto out;
2089
2090         error = proc_pident_instantiate(dir, dentry, task, p);
2091 out:
2092         put_task_struct(task);
2093 out_no_task:
2094         return error;
2095 }
2096
2097 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2098         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2099 {
2100         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2101                                 proc_pident_instantiate, task, p);
2102 }
2103
2104 static int proc_pident_readdir(struct file *filp,
2105                 void *dirent, filldir_t filldir,
2106                 const struct pid_entry *ents, unsigned int nents)
2107 {
2108         int i;
2109         struct dentry *dentry = filp->f_path.dentry;
2110         struct inode *inode = dentry->d_inode;
2111         struct task_struct *task = get_proc_task(inode);
2112         const struct pid_entry *p, *last;
2113         ino_t ino;
2114         int ret;
2115
2116         ret = -ENOENT;
2117         if (!task)
2118                 goto out_no_task;
2119
2120         ret = 0;
2121         i = filp->f_pos;
2122         switch (i) {
2123         case 0:
2124                 ino = inode->i_ino;
2125                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2126                         goto out;
2127                 i++;
2128                 filp->f_pos++;
2129                 /* fall through */
2130         case 1:
2131                 ino = parent_ino(dentry);
2132                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2133                         goto out;
2134                 i++;
2135                 filp->f_pos++;
2136                 /* fall through */
2137         default:
2138                 i -= 2;
2139                 if (i >= nents) {
2140                         ret = 1;
2141                         goto out;
2142                 }
2143                 p = ents + i;
2144                 last = &ents[nents - 1];
2145                 while (p <= last) {
2146                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2147                                 goto out;
2148                         filp->f_pos++;
2149                         p++;
2150                 }
2151         }
2152
2153         ret = 1;
2154 out:
2155         put_task_struct(task);
2156 out_no_task:
2157         return ret;
2158 }
2159
2160 #ifdef CONFIG_SECURITY
2161 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2162                                   size_t count, loff_t *ppos)
2163 {
2164         struct inode * inode = file->f_path.dentry->d_inode;
2165         char *p = NULL;
2166         ssize_t length;
2167         struct task_struct *task = get_proc_task(inode);
2168
2169         if (!task)
2170                 return -ESRCH;
2171
2172         length = security_getprocattr(task,
2173                                       (char*)file->f_path.dentry->d_name.name,
2174                                       &p);
2175         put_task_struct(task);
2176         if (length > 0)
2177                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2178         kfree(p);
2179         return length;
2180 }
2181
2182 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2183                                    size_t count, loff_t *ppos)
2184 {
2185         struct inode * inode = file->f_path.dentry->d_inode;
2186         char *page;
2187         ssize_t length;
2188         struct task_struct *task = get_proc_task(inode);
2189
2190         length = -ESRCH;
2191         if (!task)
2192                 goto out_no_task;
2193         if (count > PAGE_SIZE)
2194                 count = PAGE_SIZE;
2195
2196         /* No partial writes. */
2197         length = -EINVAL;
2198         if (*ppos != 0)
2199                 goto out;
2200
2201         length = -ENOMEM;
2202         page = (char*)__get_free_page(GFP_TEMPORARY);
2203         if (!page)
2204                 goto out;
2205
2206         length = -EFAULT;
2207         if (copy_from_user(page, buf, count))
2208                 goto out_free;
2209
2210         /* Guard against adverse ptrace interaction */
2211         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2212         if (length < 0)
2213                 goto out_free;
2214
2215         length = security_setprocattr(task,
2216                                       (char*)file->f_path.dentry->d_name.name,
2217                                       (void*)page, count);
2218         mutex_unlock(&task->cred_guard_mutex);
2219 out_free:
2220         free_page((unsigned long) page);
2221 out:
2222         put_task_struct(task);
2223 out_no_task:
2224         return length;
2225 }
2226
2227 static const struct file_operations proc_pid_attr_operations = {
2228         .read           = proc_pid_attr_read,
2229         .write          = proc_pid_attr_write,
2230 };
2231
2232 static const struct pid_entry attr_dir_stuff[] = {
2233         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2234         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2235         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2236         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2237         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2238         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2239 };
2240
2241 static int proc_attr_dir_readdir(struct file * filp,
2242                              void * dirent, filldir_t filldir)
2243 {
2244         return proc_pident_readdir(filp,dirent,filldir,
2245                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2246 }
2247
2248 static const struct file_operations proc_attr_dir_operations = {
2249         .read           = generic_read_dir,
2250         .readdir        = proc_attr_dir_readdir,
2251 };
2252
2253 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2254                                 struct dentry *dentry, struct nameidata *nd)
2255 {
2256         return proc_pident_lookup(dir, dentry,
2257                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2258 }
2259
2260 static const struct inode_operations proc_attr_dir_inode_operations = {
2261         .lookup         = proc_attr_dir_lookup,
2262         .getattr        = pid_getattr,
2263         .setattr        = proc_setattr,
2264 };
2265
2266 #endif
2267
2268 #ifdef CONFIG_ELF_CORE
2269 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2270                                          size_t count, loff_t *ppos)
2271 {
2272         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2273         struct mm_struct *mm;
2274         char buffer[PROC_NUMBUF];
2275         size_t len;
2276         int ret;
2277
2278         if (!task)
2279                 return -ESRCH;
2280
2281         ret = 0;
2282         mm = get_task_mm(task);
2283         if (mm) {
2284                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2285                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2286                                 MMF_DUMP_FILTER_SHIFT));
2287                 mmput(mm);
2288                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2289         }
2290
2291         put_task_struct(task);
2292
2293         return ret;
2294 }
2295
2296 static ssize_t proc_coredump_filter_write(struct file *file,
2297                                           const char __user *buf,
2298                                           size_t count,
2299                                           loff_t *ppos)
2300 {
2301         struct task_struct *task;
2302         struct mm_struct *mm;
2303         char buffer[PROC_NUMBUF], *end;
2304         unsigned int val;
2305         int ret;
2306         int i;
2307         unsigned long mask;
2308
2309         ret = -EFAULT;
2310         memset(buffer, 0, sizeof(buffer));
2311         if (count > sizeof(buffer) - 1)
2312                 count = sizeof(buffer) - 1;
2313         if (copy_from_user(buffer, buf, count))
2314                 goto out_no_task;
2315
2316         ret = -EINVAL;
2317         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2318         if (*end == '\n')
2319                 end++;
2320         if (end - buffer == 0)
2321                 goto out_no_task;
2322
2323         ret = -ESRCH;
2324         task = get_proc_task(file->f_dentry->d_inode);
2325         if (!task)
2326                 goto out_no_task;
2327
2328         ret = end - buffer;
2329         mm = get_task_mm(task);
2330         if (!mm)
2331                 goto out_no_mm;
2332
2333         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2334                 if (val & mask)
2335                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2336                 else
2337                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2338         }
2339
2340         mmput(mm);
2341  out_no_mm:
2342         put_task_struct(task);
2343  out_no_task:
2344         return ret;
2345 }
2346
2347 static const struct file_operations proc_coredump_filter_operations = {
2348         .read           = proc_coredump_filter_read,
2349         .write          = proc_coredump_filter_write,
2350 };
2351 #endif
2352
2353 /*
2354  * /proc/self:
2355  */
2356 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2357                               int buflen)
2358 {
2359         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2360         pid_t tgid = task_tgid_nr_ns(current, ns);
2361         char tmp[PROC_NUMBUF];
2362         if (!tgid)
2363                 return -ENOENT;
2364         sprintf(tmp, "%d", tgid);
2365         return vfs_readlink(dentry,buffer,buflen,tmp);
2366 }
2367
2368 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2369 {
2370         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2371         pid_t tgid = task_tgid_nr_ns(current, ns);
2372         char tmp[PROC_NUMBUF];
2373         if (!tgid)
2374                 return ERR_PTR(-ENOENT);
2375         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2376         return ERR_PTR(vfs_follow_link(nd,tmp));
2377 }
2378
2379 static const struct inode_operations proc_self_inode_operations = {
2380         .readlink       = proc_self_readlink,
2381         .follow_link    = proc_self_follow_link,
2382 };
2383
2384 /*
2385  * proc base
2386  *
2387  * These are the directory entries in the root directory of /proc
2388  * that properly belong to the /proc filesystem, as they describe
2389  * describe something that is process related.
2390  */
2391 static const struct pid_entry proc_base_stuff[] = {
2392         NOD("self", S_IFLNK|S_IRWXUGO,
2393                 &proc_self_inode_operations, NULL, {}),
2394 };
2395
2396 /*
2397  *      Exceptional case: normally we are not allowed to unhash a busy
2398  * directory. In this case, however, we can do it - no aliasing problems
2399  * due to the way we treat inodes.
2400  */
2401 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2402 {
2403         struct inode *inode = dentry->d_inode;
2404         struct task_struct *task = get_proc_task(inode);
2405         if (task) {
2406                 put_task_struct(task);
2407                 return 1;
2408         }
2409         d_drop(dentry);
2410         return 0;
2411 }
2412
2413 static const struct dentry_operations proc_base_dentry_operations =
2414 {
2415         .d_revalidate   = proc_base_revalidate,
2416         .d_delete       = pid_delete_dentry,
2417 };
2418
2419 static struct dentry *proc_base_instantiate(struct inode *dir,
2420         struct dentry *dentry, struct task_struct *task, const void *ptr)
2421 {
2422         const struct pid_entry *p = ptr;
2423         struct inode *inode;
2424         struct proc_inode *ei;
2425         struct dentry *error = ERR_PTR(-EINVAL);
2426
2427         /* Allocate the inode */
2428         error = ERR_PTR(-ENOMEM);
2429         inode = new_inode(dir->i_sb);
2430         if (!inode)
2431                 goto out;
2432
2433         /* Initialize the inode */
2434         ei = PROC_I(inode);
2435         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2436
2437         /*
2438          * grab the reference to the task.
2439          */
2440         ei->pid = get_task_pid(task, PIDTYPE_PID);
2441         if (!ei->pid)
2442                 goto out_iput;
2443
2444         inode->i_mode = p->mode;
2445         if (S_ISDIR(inode->i_mode))
2446                 inode->i_nlink = 2;
2447         if (S_ISLNK(inode->i_mode))
2448                 inode->i_size = 64;
2449         if (p->iop)
2450                 inode->i_op = p->iop;
2451         if (p->fop)
2452                 inode->i_fop = p->fop;
2453         ei->op = p->op;
2454         dentry->d_op = &proc_base_dentry_operations;
2455         d_add(dentry, inode);
2456         error = NULL;
2457 out:
2458         return error;
2459 out_iput:
2460         iput(inode);
2461         goto out;
2462 }
2463
2464 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2465 {
2466         struct dentry *error;
2467         struct task_struct *task = get_proc_task(dir);
2468         const struct pid_entry *p, *last;
2469
2470         error = ERR_PTR(-ENOENT);
2471
2472         if (!task)
2473                 goto out_no_task;
2474
2475         /* Lookup the directory entry */
2476         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2477         for (p = proc_base_stuff; p <= last; p++) {
2478                 if (p->len != dentry->d_name.len)
2479                         continue;
2480                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2481                         break;
2482         }
2483         if (p > last)
2484                 goto out;
2485
2486         error = proc_base_instantiate(dir, dentry, task, p);
2487
2488 out:
2489         put_task_struct(task);
2490 out_no_task:
2491         return error;
2492 }
2493
2494 static int proc_base_fill_cache(struct file *filp, void *dirent,
2495         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2496 {
2497         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2498                                 proc_base_instantiate, task, p);
2499 }
2500
2501 #ifdef CONFIG_TASK_IO_ACCOUNTING
2502 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2503 {
2504         struct task_io_accounting acct = task->ioac;
2505         unsigned long flags;
2506
2507         if (whole && lock_task_sighand(task, &flags)) {
2508                 struct task_struct *t = task;
2509
2510                 task_io_accounting_add(&acct, &task->signal->ioac);
2511                 while_each_thread(task, t)
2512                         task_io_accounting_add(&acct, &t->ioac);
2513
2514                 unlock_task_sighand(task, &flags);
2515         }
2516         return sprintf(buffer,
2517                         "rchar: %llu\n"
2518                         "wchar: %llu\n"
2519                         "syscr: %llu\n"
2520                         "syscw: %llu\n"
2521                         "read_bytes: %llu\n"
2522                         "write_bytes: %llu\n"
2523                         "cancelled_write_bytes: %llu\n",
2524                         (unsigned long long)acct.rchar,
2525                         (unsigned long long)acct.wchar,
2526                         (unsigned long long)acct.syscr,
2527                         (unsigned long long)acct.syscw,
2528                         (unsigned long long)acct.read_bytes,
2529                         (unsigned long long)acct.write_bytes,
2530                         (unsigned long long)acct.cancelled_write_bytes);
2531 }
2532
2533 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2534 {
2535         return do_io_accounting(task, buffer, 0);
2536 }
2537
2538 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2539 {
2540         return do_io_accounting(task, buffer, 1);
2541 }
2542 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2543
2544 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2545                                 struct pid *pid, struct task_struct *task)
2546 {
2547         seq_printf(m, "%08x\n", task->personality);
2548         return 0;
2549 }
2550
2551 /*
2552  * Thread groups
2553  */
2554 static const struct file_operations proc_task_operations;
2555 static const struct inode_operations proc_task_inode_operations;
2556
2557 static const struct pid_entry tgid_base_stuff[] = {
2558         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2559         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2560         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2561 #ifdef CONFIG_NET
2562         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2563 #endif
2564         REG("environ",    S_IRUSR, proc_environ_operations),
2565         INF("auxv",       S_IRUSR, proc_pid_auxv),
2566         ONE("status",     S_IRUGO, proc_pid_status),
2567         ONE("personality", S_IRUSR, proc_pid_personality),
2568         INF("limits",     S_IRUSR, proc_pid_limits),
2569 #ifdef CONFIG_SCHED_DEBUG
2570         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2571 #endif
2572         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2573 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2574         INF("syscall",    S_IRUSR, proc_pid_syscall),
2575 #endif
2576         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2577         ONE("stat",       S_IRUGO, proc_tgid_stat),
2578         ONE("statm",      S_IRUGO, proc_pid_statm),
2579         REG("maps",       S_IRUGO, proc_maps_operations),
2580 #ifdef CONFIG_NUMA
2581         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2582 #endif
2583         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2584         LNK("cwd",        proc_cwd_link),
2585         LNK("root",       proc_root_link),
2586         LNK("exe",        proc_exe_link),
2587         REG("mounts",     S_IRUGO, proc_mounts_operations),
2588         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2589         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2590 #ifdef CONFIG_PROC_PAGE_MONITOR
2591         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2592         REG("smaps",      S_IRUGO, proc_smaps_operations),
2593         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2594 #endif
2595 #ifdef CONFIG_SECURITY
2596         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2597 #endif
2598 #ifdef CONFIG_KALLSYMS
2599         INF("wchan",      S_IRUGO, proc_pid_wchan),
2600 #endif
2601 #ifdef CONFIG_STACKTRACE
2602         ONE("stack",      S_IRUSR, proc_pid_stack),
2603 #endif
2604 #ifdef CONFIG_SCHEDSTATS
2605         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2606 #endif
2607 #ifdef CONFIG_LATENCYTOP
2608         REG("latency",  S_IRUGO, proc_lstats_operations),
2609 #endif
2610 #ifdef CONFIG_PROC_PID_CPUSET
2611         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2612 #endif
2613 #ifdef CONFIG_CGROUPS
2614         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2615 #endif
2616         INF("oom_score",  S_IRUGO, proc_oom_score),
2617         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2618 #ifdef CONFIG_AUDITSYSCALL
2619         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2620         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2621 #endif
2622 #ifdef CONFIG_FAULT_INJECTION
2623         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2624 #endif
2625 #ifdef CONFIG_ELF_CORE
2626         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2627 #endif
2628 #ifdef CONFIG_TASK_IO_ACCOUNTING
2629         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2630 #endif
2631 };
2632
2633 static int proc_tgid_base_readdir(struct file * filp,
2634                              void * dirent, filldir_t filldir)
2635 {
2636         return proc_pident_readdir(filp,dirent,filldir,
2637                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2638 }
2639
2640 static const struct file_operations proc_tgid_base_operations = {
2641         .read           = generic_read_dir,
2642         .readdir        = proc_tgid_base_readdir,
2643 };
2644
2645 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2646         return proc_pident_lookup(dir, dentry,
2647                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2648 }
2649
2650 static const struct inode_operations proc_tgid_base_inode_operations = {
2651         .lookup         = proc_tgid_base_lookup,
2652         .getattr        = pid_getattr,
2653         .setattr        = proc_setattr,
2654 };
2655
2656 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2657 {
2658         struct dentry *dentry, *leader, *dir;
2659         char buf[PROC_NUMBUF];
2660         struct qstr name;
2661
2662         name.name = buf;
2663         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2664         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2665         if (dentry) {
2666                 shrink_dcache_parent(dentry);
2667                 d_drop(dentry);
2668                 dput(dentry);
2669         }
2670
2671         name.name = buf;
2672         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2673         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2674         if (!leader)
2675                 goto out;
2676
2677         name.name = "task";
2678         name.len = strlen(name.name);
2679         dir = d_hash_and_lookup(leader, &name);
2680         if (!dir)
2681                 goto out_put_leader;
2682
2683         name.name = buf;
2684         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2685         dentry = d_hash_and_lookup(dir, &name);
2686         if (dentry) {
2687                 shrink_dcache_parent(dentry);
2688                 d_drop(dentry);
2689                 dput(dentry);
2690         }
2691
2692         dput(dir);
2693 out_put_leader:
2694         dput(leader);
2695 out:
2696         return;
2697 }
2698
2699 /**
2700  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2701  * @task: task that should be flushed.
2702  *
2703  * When flushing dentries from proc, one needs to flush them from global
2704  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2705  * in. This call is supposed to do all of this job.
2706  *
2707  * Looks in the dcache for
2708  * /proc/@pid
2709  * /proc/@tgid/task/@pid
2710  * if either directory is present flushes it and all of it'ts children
2711  * from the dcache.
2712  *
2713  * It is safe and reasonable to cache /proc entries for a task until
2714  * that task exits.  After that they just clog up the dcache with
2715  * useless entries, possibly causing useful dcache entries to be
2716  * flushed instead.  This routine is proved to flush those useless
2717  * dcache entries at process exit time.
2718  *
2719  * NOTE: This routine is just an optimization so it does not guarantee
2720  *       that no dcache entries will exist at process exit time it
2721  *       just makes it very unlikely that any will persist.
2722  */
2723
2724 void proc_flush_task(struct task_struct *task)
2725 {
2726         int i;
2727         struct pid *pid, *tgid;
2728         struct upid *upid;
2729
2730         pid = task_pid(task);
2731         tgid = task_tgid(task);
2732
2733         for (i = 0; i <= pid->level; i++) {
2734                 upid = &pid->numbers[i];
2735                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2736                                         tgid->numbers[i].nr);
2737         }
2738
2739         upid = &pid->numbers[pid->level];
2740         if (upid->nr == 1)
2741                 pid_ns_release_proc(upid->ns);
2742 }
2743
2744 static struct dentry *proc_pid_instantiate(struct inode *dir,
2745                                            struct dentry * dentry,
2746                                            struct task_struct *task, const void *ptr)
2747 {
2748         struct dentry *error = ERR_PTR(-ENOENT);
2749         struct inode *inode;
2750
2751         inode = proc_pid_make_inode(dir->i_sb, task);
2752         if (!inode)
2753                 goto out;
2754
2755         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2756         inode->i_op = &proc_tgid_base_inode_operations;
2757         inode->i_fop = &proc_tgid_base_operations;
2758         inode->i_flags|=S_IMMUTABLE;
2759
2760         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2761                 ARRAY_SIZE(tgid_base_stuff));
2762
2763         dentry->d_op = &pid_dentry_operations;
2764
2765         d_add(dentry, inode);
2766         /* Close the race of the process dying before we return the dentry */
2767         if (pid_revalidate(dentry, NULL))
2768                 error = NULL;
2769 out:
2770         return error;
2771 }
2772
2773 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2774 {
2775         struct dentry *result = ERR_PTR(-ENOENT);
2776         struct task_struct *task;
2777         unsigned tgid;
2778         struct pid_namespace *ns;
2779
2780         result = proc_base_lookup(dir, dentry);
2781         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2782                 goto out;
2783
2784         tgid = name_to_int(dentry);
2785         if (tgid == ~0U)
2786                 goto out;
2787
2788         ns = dentry->d_sb->s_fs_info;
2789         rcu_read_lock();
2790         task = find_task_by_pid_ns(tgid, ns);
2791         if (task)
2792                 get_task_struct(task);
2793         rcu_read_unlock();
2794         if (!task)
2795                 goto out;
2796
2797         result = proc_pid_instantiate(dir, dentry, task, NULL);
2798         put_task_struct(task);
2799 out:
2800         return result;
2801 }
2802
2803 /*
2804  * Find the first task with tgid >= tgid
2805  *
2806  */
2807 struct tgid_iter {
2808         unsigned int tgid;
2809         struct task_struct *task;
2810 };
2811 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2812 {
2813         struct pid *pid;
2814
2815         if (iter.task)
2816                 put_task_struct(iter.task);
2817         rcu_read_lock();
2818 retry:
2819         iter.task = NULL;
2820         pid = find_ge_pid(iter.tgid, ns);
2821         if (pid) {
2822                 iter.tgid = pid_nr_ns(pid, ns);
2823                 iter.task = pid_task(pid, PIDTYPE_PID);
2824                 /* What we to know is if the pid we have find is the
2825                  * pid of a thread_group_leader.  Testing for task
2826                  * being a thread_group_leader is the obvious thing
2827                  * todo but there is a window when it fails, due to
2828                  * the pid transfer logic in de_thread.
2829                  *
2830                  * So we perform the straight forward test of seeing
2831                  * if the pid we have found is the pid of a thread
2832                  * group leader, and don't worry if the task we have
2833                  * found doesn't happen to be a thread group leader.
2834                  * As we don't care in the case of readdir.
2835                  */
2836                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2837                         iter.tgid += 1;
2838                         goto retry;
2839                 }
2840                 get_task_struct(iter.task);
2841         }
2842         rcu_read_unlock();
2843         return iter;
2844 }
2845
2846 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2847
2848 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2849         struct tgid_iter iter)
2850 {
2851         char name[PROC_NUMBUF];
2852         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2853         return proc_fill_cache(filp, dirent, filldir, name, len,
2854                                 proc_pid_instantiate, iter.task, NULL);
2855 }
2856
2857 /* for the /proc/ directory itself, after non-process stuff has been done */
2858 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2859 {
2860         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2861         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2862         struct tgid_iter iter;
2863         struct pid_namespace *ns;
2864
2865         if (!reaper)
2866                 goto out_no_task;
2867
2868         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2869                 const struct pid_entry *p = &proc_base_stuff[nr];
2870                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2871                         goto out;
2872         }
2873
2874         ns = filp->f_dentry->d_sb->s_fs_info;
2875         iter.task = NULL;
2876         iter.tgid = filp->f_pos - TGID_OFFSET;
2877         for (iter = next_tgid(ns, iter);
2878              iter.task;
2879              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2880                 filp->f_pos = iter.tgid + TGID_OFFSET;
2881                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2882                         put_task_struct(iter.task);
2883                         goto out;
2884                 }
2885         }
2886         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2887 out:
2888         put_task_struct(reaper);
2889 out_no_task:
2890         return 0;
2891 }
2892
2893 /*
2894  * Tasks
2895  */
2896 static const struct pid_entry tid_base_stuff[] = {
2897         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2898         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2899         REG("environ",   S_IRUSR, proc_environ_operations),
2900         INF("auxv",      S_IRUSR, proc_pid_auxv),
2901         ONE("status",    S_IRUGO, proc_pid_status),
2902         ONE("personality", S_IRUSR, proc_pid_personality),
2903         INF("limits",    S_IRUSR, proc_pid_limits),
2904 #ifdef CONFIG_SCHED_DEBUG
2905         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2906 #endif
2907         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2908 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2909         INF("syscall",   S_IRUSR, proc_pid_syscall),
2910 #endif
2911         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2912         ONE("stat",      S_IRUGO, proc_tid_stat),
2913         ONE("statm",     S_IRUGO, proc_pid_statm),
2914         REG("maps",      S_IRUGO, proc_maps_operations),
2915 #ifdef CONFIG_NUMA
2916         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2917 #endif
2918         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2919         LNK("cwd",       proc_cwd_link),
2920         LNK("root",      proc_root_link),
2921         LNK("exe",       proc_exe_link),
2922         REG("mounts",    S_IRUGO, proc_mounts_operations),
2923         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2924 #ifdef CONFIG_PROC_PAGE_MONITOR
2925         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2926         REG("smaps",     S_IRUGO, proc_smaps_operations),
2927         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2928 #endif
2929 #ifdef CONFIG_SECURITY
2930         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2931 #endif
2932 #ifdef CONFIG_KALLSYMS
2933         INF("wchan",     S_IRUGO, proc_pid_wchan),
2934 #endif
2935 #ifdef CONFIG_STACKTRACE
2936         ONE("stack",      S_IRUSR, proc_pid_stack),
2937 #endif
2938 #ifdef CONFIG_SCHEDSTATS
2939         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2940 #endif
2941 #ifdef CONFIG_LATENCYTOP
2942         REG("latency",  S_IRUGO, proc_lstats_operations),
2943 #endif
2944 #ifdef CONFIG_PROC_PID_CPUSET
2945         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2946 #endif
2947 #ifdef CONFIG_CGROUPS
2948         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2949 #endif
2950         INF("oom_score", S_IRUGO, proc_oom_score),
2951         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2952 #ifdef CONFIG_AUDITSYSCALL
2953         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2954         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2955 #endif
2956 #ifdef CONFIG_FAULT_INJECTION
2957         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2958 #endif
2959 #ifdef CONFIG_TASK_IO_ACCOUNTING
2960         INF("io",       S_IRUGO, proc_tid_io_accounting),
2961 #endif
2962 };
2963
2964 static int proc_tid_base_readdir(struct file * filp,
2965                              void * dirent, filldir_t filldir)
2966 {
2967         return proc_pident_readdir(filp,dirent,filldir,
2968                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2969 }
2970
2971 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2972         return proc_pident_lookup(dir, dentry,
2973                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2974 }
2975
2976 static const struct file_operations proc_tid_base_operations = {
2977         .read           = generic_read_dir,
2978         .readdir        = proc_tid_base_readdir,
2979 };
2980
2981 static const struct inode_operations proc_tid_base_inode_operations = {
2982         .lookup         = proc_tid_base_lookup,
2983         .getattr        = pid_getattr,
2984         .setattr        = proc_setattr,
2985 };
2986
2987 static struct dentry *proc_task_instantiate(struct inode *dir,
2988         struct dentry *dentry, struct task_struct *task, const void *ptr)
2989 {
2990         struct dentry *error = ERR_PTR(-ENOENT);
2991         struct inode *inode;
2992         inode = proc_pid_make_inode(dir->i_sb, task);
2993
2994         if (!inode)
2995                 goto out;
2996         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2997         inode->i_op = &proc_tid_base_inode_operations;
2998         inode->i_fop = &proc_tid_base_operations;
2999         inode->i_flags|=S_IMMUTABLE;
3000
3001         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3002                 ARRAY_SIZE(tid_base_stuff));
3003
3004         dentry->d_op = &pid_dentry_operations;
3005
3006         d_add(dentry, inode);
3007         /* Close the race of the process dying before we return the dentry */
3008         if (pid_revalidate(dentry, NULL))
3009                 error = NULL;
3010 out:
3011         return error;
3012 }
3013
3014 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3015 {
3016         struct dentry *result = ERR_PTR(-ENOENT);
3017         struct task_struct *task;
3018         struct task_struct *leader = get_proc_task(dir);
3019         unsigned tid;
3020         struct pid_namespace *ns;
3021
3022         if (!leader)
3023                 goto out_no_task;
3024
3025         tid = name_to_int(dentry);
3026         if (tid == ~0U)
3027                 goto out;
3028
3029         ns = dentry->d_sb->s_fs_info;
3030         rcu_read_lock();
3031         task = find_task_by_pid_ns(tid, ns);
3032         if (task)
3033                 get_task_struct(task);
3034         rcu_read_unlock();
3035         if (!task)
3036                 goto out;
3037         if (!same_thread_group(leader, task))
3038                 goto out_drop_task;
3039
3040         result = proc_task_instantiate(dir, dentry, task, NULL);
3041 out_drop_task:
3042         put_task_struct(task);
3043 out:
3044         put_task_struct(leader);
3045 out_no_task:
3046         return result;
3047 }
3048
3049 /*
3050  * Find the first tid of a thread group to return to user space.
3051  *
3052  * Usually this is just the thread group leader, but if the users
3053  * buffer was too small or there was a seek into the middle of the
3054  * directory we have more work todo.
3055  *
3056  * In the case of a short read we start with find_task_by_pid.
3057  *
3058  * In the case of a seek we start with the leader and walk nr
3059  * threads past it.
3060  */
3061 static struct task_struct *first_tid(struct task_struct *leader,
3062                 int tid, int nr, struct pid_namespace *ns)
3063 {
3064         struct task_struct *pos;
3065
3066         rcu_read_lock();
3067         /* Attempt to start with the pid of a thread */
3068         if (tid && (nr > 0)) {
3069                 pos = find_task_by_pid_ns(tid, ns);
3070                 if (pos && (pos->group_leader == leader))
3071                         goto found;
3072         }
3073
3074         /* If nr exceeds the number of threads there is nothing todo */
3075         pos = NULL;
3076         if (nr && nr >= get_nr_threads(leader))
3077                 goto out;
3078
3079         /* If we haven't found our starting place yet start
3080          * with the leader and walk nr threads forward.
3081          */
3082         for (pos = leader; nr > 0; --nr) {
3083                 pos = next_thread(pos);
3084                 if (pos == leader) {
3085                         pos = NULL;
3086                         goto out;
3087                 }
3088         }
3089 found:
3090         get_task_struct(pos);
3091 out:
3092         rcu_read_unlock();
3093         return pos;
3094 }
3095
3096 /*
3097  * Find the next thread in the thread list.
3098  * Return NULL if there is an error or no next thread.
3099  *
3100  * The reference to the input task_struct is released.
3101  */
3102 static struct task_struct *next_tid(struct task_struct *start)
3103 {
3104         struct task_struct *pos = NULL;
3105         rcu_read_lock();
3106         if (pid_alive(start)) {
3107                 pos = next_thread(start);
3108                 if (thread_group_leader(pos))
3109                         pos = NULL;
3110                 else
3111                         get_task_struct(pos);
3112         }
3113         rcu_read_unlock();
3114         put_task_struct(start);
3115         return pos;
3116 }
3117
3118 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3119         struct task_struct *task, int tid)
3120 {
3121         char name[PROC_NUMBUF];
3122         int len = snprintf(name, sizeof(name), "%d", tid);
3123         return proc_fill_cache(filp, dirent, filldir, name, len,
3124                                 proc_task_instantiate, task, NULL);
3125 }
3126
3127 /* for the /proc/TGID/task/ directories */
3128 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3129 {
3130         struct dentry *dentry = filp->f_path.dentry;
3131         struct inode *inode = dentry->d_inode;
3132         struct task_struct *leader = NULL;
3133         struct task_struct *task;
3134         int retval = -ENOENT;
3135         ino_t ino;
3136         int tid;
3137         struct pid_namespace *ns;
3138
3139         task = get_proc_task(inode);
3140         if (!task)
3141                 goto out_no_task;
3142         rcu_read_lock();
3143         if (pid_alive(task)) {
3144                 leader = task->group_leader;
3145                 get_task_struct(leader);
3146         }
3147         rcu_read_unlock();
3148         put_task_struct(task);
3149         if (!leader)
3150                 goto out_no_task;
3151         retval = 0;
3152
3153         switch ((unsigned long)filp->f_pos) {
3154         case 0:
3155                 ino = inode->i_ino;
3156                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3157                         goto out;
3158                 filp->f_pos++;
3159                 /* fall through */
3160         case 1:
3161                 ino = parent_ino(dentry);
3162                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3163                         goto out;
3164                 filp->f_pos++;
3165                 /* fall through */
3166         }
3167
3168         /* f_version caches the tgid value that the last readdir call couldn't
3169          * return. lseek aka telldir automagically resets f_version to 0.
3170          */
3171         ns = filp->f_dentry->d_sb->s_fs_info;
3172         tid = (int)filp->f_version;
3173         filp->f_version = 0;
3174         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3175              task;
3176              task = next_tid(task), filp->f_pos++) {
3177                 tid = task_pid_nr_ns(task, ns);
3178                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3179                         /* returning this tgid failed, save it as the first
3180                          * pid for the next readir call */
3181                         filp->f_version = (u64)tid;
3182                         put_task_struct(task);
3183                         break;
3184                 }
3185         }
3186 out:
3187         put_task_struct(leader);
3188 out_no_task:
3189         return retval;
3190 }
3191
3192 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3193 {
3194         struct inode *inode = dentry->d_inode;
3195         struct task_struct *p = get_proc_task(inode);
3196         generic_fillattr(inode, stat);
3197
3198         if (p) {
3199                 stat->nlink += get_nr_threads(p);
3200                 put_task_struct(p);
3201         }
3202
3203         return 0;
3204 }
3205
3206 static const struct inode_operations proc_task_inode_operations = {
3207         .lookup         = proc_task_lookup,
3208         .getattr        = proc_task_getattr,
3209         .setattr        = proc_setattr,
3210 };
3211
3212 static const struct file_operations proc_task_operations = {
3213         .read           = generic_read_dir,
3214         .readdir        = proc_task_readdir,
3215 };