Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/init.h>
57 #include <linux/capability.h>
58 #include <linux/file.h>
59 #include <linux/fdtable.h>
60 #include <linux/string.h>
61 #include <linux/seq_file.h>
62 #include <linux/namei.h>
63 #include <linux/mnt_namespace.h>
64 #include <linux/mm.h>
65 #include <linux/rcupdate.h>
66 #include <linux/kallsyms.h>
67 #include <linux/resource.h>
68 #include <linux/module.h>
69 #include <linux/mount.h>
70 #include <linux/security.h>
71 #include <linux/ptrace.h>
72 #include <linux/cgroup.h>
73 #include <linux/cpuset.h>
74 #include <linux/audit.h>
75 #include <linux/poll.h>
76 #include <linux/nsproxy.h>
77 #include <linux/oom.h>
78 #include <linux/elf.h>
79 #include <linux/pid_namespace.h>
80 #include "internal.h"
81
82 /* NOTE:
83  *      Implementing inode permission operations in /proc is almost
84  *      certainly an error.  Permission checks need to happen during
85  *      each system call not at open time.  The reason is that most of
86  *      what we wish to check for permissions in /proc varies at runtime.
87  *
88  *      The classic example of a problem is opening file descriptors
89  *      in /proc for a task before it execs a suid executable.
90  */
91
92 struct pid_entry {
93         char *name;
94         int len;
95         mode_t mode;
96         const struct inode_operations *iop;
97         const struct file_operations *fop;
98         union proc_op op;
99 };
100
101 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
102         .name = (NAME),                                 \
103         .len  = sizeof(NAME) - 1,                       \
104         .mode = MODE,                                   \
105         .iop  = IOP,                                    \
106         .fop  = FOP,                                    \
107         .op   = OP,                                     \
108 }
109
110 #define DIR(NAME, MODE, OTYPE)                                                  \
111         NOD(NAME, (S_IFDIR|(MODE)),                                             \
112                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
113                 {} )
114 #define LNK(NAME, OTYPE)                                        \
115         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
116                 &proc_pid_link_inode_operations, NULL,          \
117                 { .proc_get_link = &proc_##OTYPE##_link } )
118 #define REG(NAME, MODE, OTYPE)                          \
119         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
120                 &proc_##OTYPE##_operations, {})
121 #define INF(NAME, MODE, OTYPE)                          \
122         NOD(NAME, (S_IFREG|(MODE)),                     \
123                 NULL, &proc_info_file_operations,       \
124                 { .proc_read = &proc_##OTYPE } )
125 #define ONE(NAME, MODE, OTYPE)                          \
126         NOD(NAME, (S_IFREG|(MODE)),                     \
127                 NULL, &proc_single_file_operations,     \
128                 { .proc_show = &proc_##OTYPE } )
129
130 /*
131  * Count the number of hardlinks for the pid_entry table, excluding the .
132  * and .. links.
133  */
134 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
135         unsigned int n)
136 {
137         unsigned int i;
138         unsigned int count;
139
140         count = 0;
141         for (i = 0; i < n; ++i) {
142                 if (S_ISDIR(entries[i].mode))
143                         ++count;
144         }
145
146         return count;
147 }
148
149 int maps_protect;
150 EXPORT_SYMBOL(maps_protect);
151
152 static struct fs_struct *get_fs_struct(struct task_struct *task)
153 {
154         struct fs_struct *fs;
155         task_lock(task);
156         fs = task->fs;
157         if(fs)
158                 atomic_inc(&fs->count);
159         task_unlock(task);
160         return fs;
161 }
162
163 static int get_nr_threads(struct task_struct *tsk)
164 {
165         /* Must be called with the rcu_read_lock held */
166         unsigned long flags;
167         int count = 0;
168
169         if (lock_task_sighand(tsk, &flags)) {
170                 count = atomic_read(&tsk->signal->count);
171                 unlock_task_sighand(tsk, &flags);
172         }
173         return count;
174 }
175
176 static int proc_cwd_link(struct inode *inode, struct path *path)
177 {
178         struct task_struct *task = get_proc_task(inode);
179         struct fs_struct *fs = NULL;
180         int result = -ENOENT;
181
182         if (task) {
183                 fs = get_fs_struct(task);
184                 put_task_struct(task);
185         }
186         if (fs) {
187                 read_lock(&fs->lock);
188                 *path = fs->pwd;
189                 path_get(&fs->pwd);
190                 read_unlock(&fs->lock);
191                 result = 0;
192                 put_fs_struct(fs);
193         }
194         return result;
195 }
196
197 static int proc_root_link(struct inode *inode, struct path *path)
198 {
199         struct task_struct *task = get_proc_task(inode);
200         struct fs_struct *fs = NULL;
201         int result = -ENOENT;
202
203         if (task) {
204                 fs = get_fs_struct(task);
205                 put_task_struct(task);
206         }
207         if (fs) {
208                 read_lock(&fs->lock);
209                 *path = fs->root;
210                 path_get(&fs->root);
211                 read_unlock(&fs->lock);
212                 result = 0;
213                 put_fs_struct(fs);
214         }
215         return result;
216 }
217
218 /*
219  * Return zero if current may access user memory in @task, -error if not.
220  */
221 static int check_mem_permission(struct task_struct *task)
222 {
223         /*
224          * A task can always look at itself, in case it chooses
225          * to use system calls instead of load instructions.
226          */
227         if (task == current)
228                 return 0;
229
230         /*
231          * If current is actively ptrace'ing, and would also be
232          * permitted to freshly attach with ptrace now, permit it.
233          */
234         if (task->parent == current && (task->ptrace & PT_PTRACED) &&
235             task_is_stopped_or_traced(task) &&
236             ptrace_may_access(task, PTRACE_MODE_ATTACH))
237                 return 0;
238
239         /*
240          * Noone else is allowed.
241          */
242         return -EPERM;
243 }
244
245 struct mm_struct *mm_for_maps(struct task_struct *task)
246 {
247         struct mm_struct *mm = get_task_mm(task);
248         if (!mm)
249                 return NULL;
250         down_read(&mm->mmap_sem);
251         task_lock(task);
252         if (task->mm != mm)
253                 goto out;
254         if (task->mm != current->mm &&
255             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
256                 goto out;
257         task_unlock(task);
258         return mm;
259 out:
260         task_unlock(task);
261         up_read(&mm->mmap_sem);
262         mmput(mm);
263         return NULL;
264 }
265
266 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
267 {
268         int res = 0;
269         unsigned int len;
270         struct mm_struct *mm = get_task_mm(task);
271         if (!mm)
272                 goto out;
273         if (!mm->arg_end)
274                 goto out_mm;    /* Shh! No looking before we're done */
275
276         len = mm->arg_end - mm->arg_start;
277  
278         if (len > PAGE_SIZE)
279                 len = PAGE_SIZE;
280  
281         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
282
283         // If the nul at the end of args has been overwritten, then
284         // assume application is using setproctitle(3).
285         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
286                 len = strnlen(buffer, res);
287                 if (len < res) {
288                     res = len;
289                 } else {
290                         len = mm->env_end - mm->env_start;
291                         if (len > PAGE_SIZE - res)
292                                 len = PAGE_SIZE - res;
293                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
294                         res = strnlen(buffer, res);
295                 }
296         }
297 out_mm:
298         mmput(mm);
299 out:
300         return res;
301 }
302
303 static int proc_pid_auxv(struct task_struct *task, char *buffer)
304 {
305         int res = 0;
306         struct mm_struct *mm = get_task_mm(task);
307         if (mm) {
308                 unsigned int nwords = 0;
309                 do
310                         nwords += 2;
311                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
312                 res = nwords * sizeof(mm->saved_auxv[0]);
313                 if (res > PAGE_SIZE)
314                         res = PAGE_SIZE;
315                 memcpy(buffer, mm->saved_auxv, res);
316                 mmput(mm);
317         }
318         return res;
319 }
320
321
322 #ifdef CONFIG_KALLSYMS
323 /*
324  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
325  * Returns the resolved symbol.  If that fails, simply return the address.
326  */
327 static int proc_pid_wchan(struct task_struct *task, char *buffer)
328 {
329         unsigned long wchan;
330         char symname[KSYM_NAME_LEN];
331
332         wchan = get_wchan(task);
333
334         if (lookup_symbol_name(wchan, symname) < 0)
335                 return sprintf(buffer, "%lu", wchan);
336         else
337                 return sprintf(buffer, "%s", symname);
338 }
339 #endif /* CONFIG_KALLSYMS */
340
341 #ifdef CONFIG_SCHEDSTATS
342 /*
343  * Provides /proc/PID/schedstat
344  */
345 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
346 {
347         return sprintf(buffer, "%llu %llu %lu\n",
348                         task->sched_info.cpu_time,
349                         task->sched_info.run_delay,
350                         task->sched_info.pcount);
351 }
352 #endif
353
354 #ifdef CONFIG_LATENCYTOP
355 static int lstats_show_proc(struct seq_file *m, void *v)
356 {
357         int i;
358         struct inode *inode = m->private;
359         struct task_struct *task = get_proc_task(inode);
360
361         if (!task)
362                 return -ESRCH;
363         seq_puts(m, "Latency Top version : v0.1\n");
364         for (i = 0; i < 32; i++) {
365                 if (task->latency_record[i].backtrace[0]) {
366                         int q;
367                         seq_printf(m, "%i %li %li ",
368                                 task->latency_record[i].count,
369                                 task->latency_record[i].time,
370                                 task->latency_record[i].max);
371                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
372                                 char sym[KSYM_NAME_LEN];
373                                 char *c;
374                                 if (!task->latency_record[i].backtrace[q])
375                                         break;
376                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
377                                         break;
378                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
379                                 c = strchr(sym, '+');
380                                 if (c)
381                                         *c = 0;
382                                 seq_printf(m, "%s ", sym);
383                         }
384                         seq_printf(m, "\n");
385                 }
386
387         }
388         put_task_struct(task);
389         return 0;
390 }
391
392 static int lstats_open(struct inode *inode, struct file *file)
393 {
394         return single_open(file, lstats_show_proc, inode);
395 }
396
397 static ssize_t lstats_write(struct file *file, const char __user *buf,
398                             size_t count, loff_t *offs)
399 {
400         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
401
402         if (!task)
403                 return -ESRCH;
404         clear_all_latency_tracing(task);
405         put_task_struct(task);
406
407         return count;
408 }
409
410 static const struct file_operations proc_lstats_operations = {
411         .open           = lstats_open,
412         .read           = seq_read,
413         .write          = lstats_write,
414         .llseek         = seq_lseek,
415         .release        = single_release,
416 };
417
418 #endif
419
420 /* The badness from the OOM killer */
421 unsigned long badness(struct task_struct *p, unsigned long uptime);
422 static int proc_oom_score(struct task_struct *task, char *buffer)
423 {
424         unsigned long points;
425         struct timespec uptime;
426
427         do_posix_clock_monotonic_gettime(&uptime);
428         read_lock(&tasklist_lock);
429         points = badness(task, uptime.tv_sec);
430         read_unlock(&tasklist_lock);
431         return sprintf(buffer, "%lu\n", points);
432 }
433
434 struct limit_names {
435         char *name;
436         char *unit;
437 };
438
439 static const struct limit_names lnames[RLIM_NLIMITS] = {
440         [RLIMIT_CPU] = {"Max cpu time", "ms"},
441         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
442         [RLIMIT_DATA] = {"Max data size", "bytes"},
443         [RLIMIT_STACK] = {"Max stack size", "bytes"},
444         [RLIMIT_CORE] = {"Max core file size", "bytes"},
445         [RLIMIT_RSS] = {"Max resident set", "bytes"},
446         [RLIMIT_NPROC] = {"Max processes", "processes"},
447         [RLIMIT_NOFILE] = {"Max open files", "files"},
448         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
449         [RLIMIT_AS] = {"Max address space", "bytes"},
450         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
451         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
452         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
453         [RLIMIT_NICE] = {"Max nice priority", NULL},
454         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
455         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
456 };
457
458 /* Display limits for a process */
459 static int proc_pid_limits(struct task_struct *task, char *buffer)
460 {
461         unsigned int i;
462         int count = 0;
463         unsigned long flags;
464         char *bufptr = buffer;
465
466         struct rlimit rlim[RLIM_NLIMITS];
467
468         rcu_read_lock();
469         if (!lock_task_sighand(task,&flags)) {
470                 rcu_read_unlock();
471                 return 0;
472         }
473         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
474         unlock_task_sighand(task, &flags);
475         rcu_read_unlock();
476
477         /*
478          * print the file header
479          */
480         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
481                         "Limit", "Soft Limit", "Hard Limit", "Units");
482
483         for (i = 0; i < RLIM_NLIMITS; i++) {
484                 if (rlim[i].rlim_cur == RLIM_INFINITY)
485                         count += sprintf(&bufptr[count], "%-25s %-20s ",
486                                          lnames[i].name, "unlimited");
487                 else
488                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
489                                          lnames[i].name, rlim[i].rlim_cur);
490
491                 if (rlim[i].rlim_max == RLIM_INFINITY)
492                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
493                 else
494                         count += sprintf(&bufptr[count], "%-20lu ",
495                                          rlim[i].rlim_max);
496
497                 if (lnames[i].unit)
498                         count += sprintf(&bufptr[count], "%-10s\n",
499                                          lnames[i].unit);
500                 else
501                         count += sprintf(&bufptr[count], "\n");
502         }
503
504         return count;
505 }
506
507 /************************************************************************/
508 /*                       Here the fs part begins                        */
509 /************************************************************************/
510
511 /* permission checks */
512 static int proc_fd_access_allowed(struct inode *inode)
513 {
514         struct task_struct *task;
515         int allowed = 0;
516         /* Allow access to a task's file descriptors if it is us or we
517          * may use ptrace attach to the process and find out that
518          * information.
519          */
520         task = get_proc_task(inode);
521         if (task) {
522                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
523                 put_task_struct(task);
524         }
525         return allowed;
526 }
527
528 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
529 {
530         int error;
531         struct inode *inode = dentry->d_inode;
532
533         if (attr->ia_valid & ATTR_MODE)
534                 return -EPERM;
535
536         error = inode_change_ok(inode, attr);
537         if (!error)
538                 error = inode_setattr(inode, attr);
539         return error;
540 }
541
542 static const struct inode_operations proc_def_inode_operations = {
543         .setattr        = proc_setattr,
544 };
545
546 static int mounts_open_common(struct inode *inode, struct file *file,
547                               const struct seq_operations *op)
548 {
549         struct task_struct *task = get_proc_task(inode);
550         struct nsproxy *nsp;
551         struct mnt_namespace *ns = NULL;
552         struct fs_struct *fs = NULL;
553         struct path root;
554         struct proc_mounts *p;
555         int ret = -EINVAL;
556
557         if (task) {
558                 rcu_read_lock();
559                 nsp = task_nsproxy(task);
560                 if (nsp) {
561                         ns = nsp->mnt_ns;
562                         if (ns)
563                                 get_mnt_ns(ns);
564                 }
565                 rcu_read_unlock();
566                 if (ns)
567                         fs = get_fs_struct(task);
568                 put_task_struct(task);
569         }
570
571         if (!ns)
572                 goto err;
573         if (!fs)
574                 goto err_put_ns;
575
576         read_lock(&fs->lock);
577         root = fs->root;
578         path_get(&root);
579         read_unlock(&fs->lock);
580         put_fs_struct(fs);
581
582         ret = -ENOMEM;
583         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
584         if (!p)
585                 goto err_put_path;
586
587         file->private_data = &p->m;
588         ret = seq_open(file, op);
589         if (ret)
590                 goto err_free;
591
592         p->m.private = p;
593         p->ns = ns;
594         p->root = root;
595         p->event = ns->event;
596
597         return 0;
598
599  err_free:
600         kfree(p);
601  err_put_path:
602         path_put(&root);
603  err_put_ns:
604         put_mnt_ns(ns);
605  err:
606         return ret;
607 }
608
609 static int mounts_release(struct inode *inode, struct file *file)
610 {
611         struct proc_mounts *p = file->private_data;
612         path_put(&p->root);
613         put_mnt_ns(p->ns);
614         return seq_release(inode, file);
615 }
616
617 static unsigned mounts_poll(struct file *file, poll_table *wait)
618 {
619         struct proc_mounts *p = file->private_data;
620         struct mnt_namespace *ns = p->ns;
621         unsigned res = 0;
622
623         poll_wait(file, &ns->poll, wait);
624
625         spin_lock(&vfsmount_lock);
626         if (p->event != ns->event) {
627                 p->event = ns->event;
628                 res = POLLERR;
629         }
630         spin_unlock(&vfsmount_lock);
631
632         return res;
633 }
634
635 static int mounts_open(struct inode *inode, struct file *file)
636 {
637         return mounts_open_common(inode, file, &mounts_op);
638 }
639
640 static const struct file_operations proc_mounts_operations = {
641         .open           = mounts_open,
642         .read           = seq_read,
643         .llseek         = seq_lseek,
644         .release        = mounts_release,
645         .poll           = mounts_poll,
646 };
647
648 static int mountinfo_open(struct inode *inode, struct file *file)
649 {
650         return mounts_open_common(inode, file, &mountinfo_op);
651 }
652
653 static const struct file_operations proc_mountinfo_operations = {
654         .open           = mountinfo_open,
655         .read           = seq_read,
656         .llseek         = seq_lseek,
657         .release        = mounts_release,
658         .poll           = mounts_poll,
659 };
660
661 static int mountstats_open(struct inode *inode, struct file *file)
662 {
663         return mounts_open_common(inode, file, &mountstats_op);
664 }
665
666 static const struct file_operations proc_mountstats_operations = {
667         .open           = mountstats_open,
668         .read           = seq_read,
669         .llseek         = seq_lseek,
670         .release        = mounts_release,
671 };
672
673 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
674
675 static ssize_t proc_info_read(struct file * file, char __user * buf,
676                           size_t count, loff_t *ppos)
677 {
678         struct inode * inode = file->f_path.dentry->d_inode;
679         unsigned long page;
680         ssize_t length;
681         struct task_struct *task = get_proc_task(inode);
682
683         length = -ESRCH;
684         if (!task)
685                 goto out_no_task;
686
687         if (count > PROC_BLOCK_SIZE)
688                 count = PROC_BLOCK_SIZE;
689
690         length = -ENOMEM;
691         if (!(page = __get_free_page(GFP_TEMPORARY)))
692                 goto out;
693
694         length = PROC_I(inode)->op.proc_read(task, (char*)page);
695
696         if (length >= 0)
697                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
698         free_page(page);
699 out:
700         put_task_struct(task);
701 out_no_task:
702         return length;
703 }
704
705 static const struct file_operations proc_info_file_operations = {
706         .read           = proc_info_read,
707 };
708
709 static int proc_single_show(struct seq_file *m, void *v)
710 {
711         struct inode *inode = m->private;
712         struct pid_namespace *ns;
713         struct pid *pid;
714         struct task_struct *task;
715         int ret;
716
717         ns = inode->i_sb->s_fs_info;
718         pid = proc_pid(inode);
719         task = get_pid_task(pid, PIDTYPE_PID);
720         if (!task)
721                 return -ESRCH;
722
723         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
724
725         put_task_struct(task);
726         return ret;
727 }
728
729 static int proc_single_open(struct inode *inode, struct file *filp)
730 {
731         int ret;
732         ret = single_open(filp, proc_single_show, NULL);
733         if (!ret) {
734                 struct seq_file *m = filp->private_data;
735
736                 m->private = inode;
737         }
738         return ret;
739 }
740
741 static const struct file_operations proc_single_file_operations = {
742         .open           = proc_single_open,
743         .read           = seq_read,
744         .llseek         = seq_lseek,
745         .release        = single_release,
746 };
747
748 static int mem_open(struct inode* inode, struct file* file)
749 {
750         file->private_data = (void*)((long)current->self_exec_id);
751         return 0;
752 }
753
754 static ssize_t mem_read(struct file * file, char __user * buf,
755                         size_t count, loff_t *ppos)
756 {
757         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
758         char *page;
759         unsigned long src = *ppos;
760         int ret = -ESRCH;
761         struct mm_struct *mm;
762
763         if (!task)
764                 goto out_no_task;
765
766         if (check_mem_permission(task))
767                 goto out;
768
769         ret = -ENOMEM;
770         page = (char *)__get_free_page(GFP_TEMPORARY);
771         if (!page)
772                 goto out;
773
774         ret = 0;
775  
776         mm = get_task_mm(task);
777         if (!mm)
778                 goto out_free;
779
780         ret = -EIO;
781  
782         if (file->private_data != (void*)((long)current->self_exec_id))
783                 goto out_put;
784
785         ret = 0;
786  
787         while (count > 0) {
788                 int this_len, retval;
789
790                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
791                 retval = access_process_vm(task, src, page, this_len, 0);
792                 if (!retval || check_mem_permission(task)) {
793                         if (!ret)
794                                 ret = -EIO;
795                         break;
796                 }
797
798                 if (copy_to_user(buf, page, retval)) {
799                         ret = -EFAULT;
800                         break;
801                 }
802  
803                 ret += retval;
804                 src += retval;
805                 buf += retval;
806                 count -= retval;
807         }
808         *ppos = src;
809
810 out_put:
811         mmput(mm);
812 out_free:
813         free_page((unsigned long) page);
814 out:
815         put_task_struct(task);
816 out_no_task:
817         return ret;
818 }
819
820 #define mem_write NULL
821
822 #ifndef mem_write
823 /* This is a security hazard */
824 static ssize_t mem_write(struct file * file, const char __user *buf,
825                          size_t count, loff_t *ppos)
826 {
827         int copied;
828         char *page;
829         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
830         unsigned long dst = *ppos;
831
832         copied = -ESRCH;
833         if (!task)
834                 goto out_no_task;
835
836         if (check_mem_permission(task))
837                 goto out;
838
839         copied = -ENOMEM;
840         page = (char *)__get_free_page(GFP_TEMPORARY);
841         if (!page)
842                 goto out;
843
844         copied = 0;
845         while (count > 0) {
846                 int this_len, retval;
847
848                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
849                 if (copy_from_user(page, buf, this_len)) {
850                         copied = -EFAULT;
851                         break;
852                 }
853                 retval = access_process_vm(task, dst, page, this_len, 1);
854                 if (!retval) {
855                         if (!copied)
856                                 copied = -EIO;
857                         break;
858                 }
859                 copied += retval;
860                 buf += retval;
861                 dst += retval;
862                 count -= retval;                        
863         }
864         *ppos = dst;
865         free_page((unsigned long) page);
866 out:
867         put_task_struct(task);
868 out_no_task:
869         return copied;
870 }
871 #endif
872
873 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
874 {
875         switch (orig) {
876         case 0:
877                 file->f_pos = offset;
878                 break;
879         case 1:
880                 file->f_pos += offset;
881                 break;
882         default:
883                 return -EINVAL;
884         }
885         force_successful_syscall_return();
886         return file->f_pos;
887 }
888
889 static const struct file_operations proc_mem_operations = {
890         .llseek         = mem_lseek,
891         .read           = mem_read,
892         .write          = mem_write,
893         .open           = mem_open,
894 };
895
896 static ssize_t environ_read(struct file *file, char __user *buf,
897                         size_t count, loff_t *ppos)
898 {
899         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
900         char *page;
901         unsigned long src = *ppos;
902         int ret = -ESRCH;
903         struct mm_struct *mm;
904
905         if (!task)
906                 goto out_no_task;
907
908         if (!ptrace_may_access(task, PTRACE_MODE_READ))
909                 goto out;
910
911         ret = -ENOMEM;
912         page = (char *)__get_free_page(GFP_TEMPORARY);
913         if (!page)
914                 goto out;
915
916         ret = 0;
917
918         mm = get_task_mm(task);
919         if (!mm)
920                 goto out_free;
921
922         while (count > 0) {
923                 int this_len, retval, max_len;
924
925                 this_len = mm->env_end - (mm->env_start + src);
926
927                 if (this_len <= 0)
928                         break;
929
930                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
931                 this_len = (this_len > max_len) ? max_len : this_len;
932
933                 retval = access_process_vm(task, (mm->env_start + src),
934                         page, this_len, 0);
935
936                 if (retval <= 0) {
937                         ret = retval;
938                         break;
939                 }
940
941                 if (copy_to_user(buf, page, retval)) {
942                         ret = -EFAULT;
943                         break;
944                 }
945
946                 ret += retval;
947                 src += retval;
948                 buf += retval;
949                 count -= retval;
950         }
951         *ppos = src;
952
953         mmput(mm);
954 out_free:
955         free_page((unsigned long) page);
956 out:
957         put_task_struct(task);
958 out_no_task:
959         return ret;
960 }
961
962 static const struct file_operations proc_environ_operations = {
963         .read           = environ_read,
964 };
965
966 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
967                                 size_t count, loff_t *ppos)
968 {
969         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
970         char buffer[PROC_NUMBUF];
971         size_t len;
972         int oom_adjust;
973
974         if (!task)
975                 return -ESRCH;
976         oom_adjust = task->oomkilladj;
977         put_task_struct(task);
978
979         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
980
981         return simple_read_from_buffer(buf, count, ppos, buffer, len);
982 }
983
984 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
985                                 size_t count, loff_t *ppos)
986 {
987         struct task_struct *task;
988         char buffer[PROC_NUMBUF], *end;
989         int oom_adjust;
990
991         memset(buffer, 0, sizeof(buffer));
992         if (count > sizeof(buffer) - 1)
993                 count = sizeof(buffer) - 1;
994         if (copy_from_user(buffer, buf, count))
995                 return -EFAULT;
996         oom_adjust = simple_strtol(buffer, &end, 0);
997         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
998              oom_adjust != OOM_DISABLE)
999                 return -EINVAL;
1000         if (*end == '\n')
1001                 end++;
1002         task = get_proc_task(file->f_path.dentry->d_inode);
1003         if (!task)
1004                 return -ESRCH;
1005         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1006                 put_task_struct(task);
1007                 return -EACCES;
1008         }
1009         task->oomkilladj = oom_adjust;
1010         put_task_struct(task);
1011         if (end - buffer == 0)
1012                 return -EIO;
1013         return end - buffer;
1014 }
1015
1016 static const struct file_operations proc_oom_adjust_operations = {
1017         .read           = oom_adjust_read,
1018         .write          = oom_adjust_write,
1019 };
1020
1021 #ifdef CONFIG_AUDITSYSCALL
1022 #define TMPBUFLEN 21
1023 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1024                                   size_t count, loff_t *ppos)
1025 {
1026         struct inode * inode = file->f_path.dentry->d_inode;
1027         struct task_struct *task = get_proc_task(inode);
1028         ssize_t length;
1029         char tmpbuf[TMPBUFLEN];
1030
1031         if (!task)
1032                 return -ESRCH;
1033         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1034                                 audit_get_loginuid(task));
1035         put_task_struct(task);
1036         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1037 }
1038
1039 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1040                                    size_t count, loff_t *ppos)
1041 {
1042         struct inode * inode = file->f_path.dentry->d_inode;
1043         char *page, *tmp;
1044         ssize_t length;
1045         uid_t loginuid;
1046
1047         if (!capable(CAP_AUDIT_CONTROL))
1048                 return -EPERM;
1049
1050         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1051                 return -EPERM;
1052
1053         if (count >= PAGE_SIZE)
1054                 count = PAGE_SIZE - 1;
1055
1056         if (*ppos != 0) {
1057                 /* No partial writes. */
1058                 return -EINVAL;
1059         }
1060         page = (char*)__get_free_page(GFP_TEMPORARY);
1061         if (!page)
1062                 return -ENOMEM;
1063         length = -EFAULT;
1064         if (copy_from_user(page, buf, count))
1065                 goto out_free_page;
1066
1067         page[count] = '\0';
1068         loginuid = simple_strtoul(page, &tmp, 10);
1069         if (tmp == page) {
1070                 length = -EINVAL;
1071                 goto out_free_page;
1072
1073         }
1074         length = audit_set_loginuid(current, loginuid);
1075         if (likely(length == 0))
1076                 length = count;
1077
1078 out_free_page:
1079         free_page((unsigned long) page);
1080         return length;
1081 }
1082
1083 static const struct file_operations proc_loginuid_operations = {
1084         .read           = proc_loginuid_read,
1085         .write          = proc_loginuid_write,
1086 };
1087
1088 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1089                                   size_t count, loff_t *ppos)
1090 {
1091         struct inode * inode = file->f_path.dentry->d_inode;
1092         struct task_struct *task = get_proc_task(inode);
1093         ssize_t length;
1094         char tmpbuf[TMPBUFLEN];
1095
1096         if (!task)
1097                 return -ESRCH;
1098         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1099                                 audit_get_sessionid(task));
1100         put_task_struct(task);
1101         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1102 }
1103
1104 static const struct file_operations proc_sessionid_operations = {
1105         .read           = proc_sessionid_read,
1106 };
1107 #endif
1108
1109 #ifdef CONFIG_FAULT_INJECTION
1110 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1111                                       size_t count, loff_t *ppos)
1112 {
1113         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1114         char buffer[PROC_NUMBUF];
1115         size_t len;
1116         int make_it_fail;
1117
1118         if (!task)
1119                 return -ESRCH;
1120         make_it_fail = task->make_it_fail;
1121         put_task_struct(task);
1122
1123         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1124
1125         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1126 }
1127
1128 static ssize_t proc_fault_inject_write(struct file * file,
1129                         const char __user * buf, size_t count, loff_t *ppos)
1130 {
1131         struct task_struct *task;
1132         char buffer[PROC_NUMBUF], *end;
1133         int make_it_fail;
1134
1135         if (!capable(CAP_SYS_RESOURCE))
1136                 return -EPERM;
1137         memset(buffer, 0, sizeof(buffer));
1138         if (count > sizeof(buffer) - 1)
1139                 count = sizeof(buffer) - 1;
1140         if (copy_from_user(buffer, buf, count))
1141                 return -EFAULT;
1142         make_it_fail = simple_strtol(buffer, &end, 0);
1143         if (*end == '\n')
1144                 end++;
1145         task = get_proc_task(file->f_dentry->d_inode);
1146         if (!task)
1147                 return -ESRCH;
1148         task->make_it_fail = make_it_fail;
1149         put_task_struct(task);
1150         if (end - buffer == 0)
1151                 return -EIO;
1152         return end - buffer;
1153 }
1154
1155 static const struct file_operations proc_fault_inject_operations = {
1156         .read           = proc_fault_inject_read,
1157         .write          = proc_fault_inject_write,
1158 };
1159 #endif
1160
1161
1162 #ifdef CONFIG_SCHED_DEBUG
1163 /*
1164  * Print out various scheduling related per-task fields:
1165  */
1166 static int sched_show(struct seq_file *m, void *v)
1167 {
1168         struct inode *inode = m->private;
1169         struct task_struct *p;
1170
1171         WARN_ON(!inode);
1172
1173         p = get_proc_task(inode);
1174         if (!p)
1175                 return -ESRCH;
1176         proc_sched_show_task(p, m);
1177
1178         put_task_struct(p);
1179
1180         return 0;
1181 }
1182
1183 static ssize_t
1184 sched_write(struct file *file, const char __user *buf,
1185             size_t count, loff_t *offset)
1186 {
1187         struct inode *inode = file->f_path.dentry->d_inode;
1188         struct task_struct *p;
1189
1190         WARN_ON(!inode);
1191
1192         p = get_proc_task(inode);
1193         if (!p)
1194                 return -ESRCH;
1195         proc_sched_set_task(p);
1196
1197         put_task_struct(p);
1198
1199         return count;
1200 }
1201
1202 static int sched_open(struct inode *inode, struct file *filp)
1203 {
1204         int ret;
1205
1206         ret = single_open(filp, sched_show, NULL);
1207         if (!ret) {
1208                 struct seq_file *m = filp->private_data;
1209
1210                 m->private = inode;
1211         }
1212         return ret;
1213 }
1214
1215 static const struct file_operations proc_pid_sched_operations = {
1216         .open           = sched_open,
1217         .read           = seq_read,
1218         .write          = sched_write,
1219         .llseek         = seq_lseek,
1220         .release        = single_release,
1221 };
1222
1223 #endif
1224
1225 /*
1226  * We added or removed a vma mapping the executable. The vmas are only mapped
1227  * during exec and are not mapped with the mmap system call.
1228  * Callers must hold down_write() on the mm's mmap_sem for these
1229  */
1230 void added_exe_file_vma(struct mm_struct *mm)
1231 {
1232         mm->num_exe_file_vmas++;
1233 }
1234
1235 void removed_exe_file_vma(struct mm_struct *mm)
1236 {
1237         mm->num_exe_file_vmas--;
1238         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1239                 fput(mm->exe_file);
1240                 mm->exe_file = NULL;
1241         }
1242
1243 }
1244
1245 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1246 {
1247         if (new_exe_file)
1248                 get_file(new_exe_file);
1249         if (mm->exe_file)
1250                 fput(mm->exe_file);
1251         mm->exe_file = new_exe_file;
1252         mm->num_exe_file_vmas = 0;
1253 }
1254
1255 struct file *get_mm_exe_file(struct mm_struct *mm)
1256 {
1257         struct file *exe_file;
1258
1259         /* We need mmap_sem to protect against races with removal of
1260          * VM_EXECUTABLE vmas */
1261         down_read(&mm->mmap_sem);
1262         exe_file = mm->exe_file;
1263         if (exe_file)
1264                 get_file(exe_file);
1265         up_read(&mm->mmap_sem);
1266         return exe_file;
1267 }
1268
1269 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1270 {
1271         /* It's safe to write the exe_file pointer without exe_file_lock because
1272          * this is called during fork when the task is not yet in /proc */
1273         newmm->exe_file = get_mm_exe_file(oldmm);
1274 }
1275
1276 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1277 {
1278         struct task_struct *task;
1279         struct mm_struct *mm;
1280         struct file *exe_file;
1281
1282         task = get_proc_task(inode);
1283         if (!task)
1284                 return -ENOENT;
1285         mm = get_task_mm(task);
1286         put_task_struct(task);
1287         if (!mm)
1288                 return -ENOENT;
1289         exe_file = get_mm_exe_file(mm);
1290         mmput(mm);
1291         if (exe_file) {
1292                 *exe_path = exe_file->f_path;
1293                 path_get(&exe_file->f_path);
1294                 fput(exe_file);
1295                 return 0;
1296         } else
1297                 return -ENOENT;
1298 }
1299
1300 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1301 {
1302         struct inode *inode = dentry->d_inode;
1303         int error = -EACCES;
1304
1305         /* We don't need a base pointer in the /proc filesystem */
1306         path_put(&nd->path);
1307
1308         /* Are we allowed to snoop on the tasks file descriptors? */
1309         if (!proc_fd_access_allowed(inode))
1310                 goto out;
1311
1312         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1313         nd->last_type = LAST_BIND;
1314 out:
1315         return ERR_PTR(error);
1316 }
1317
1318 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1319 {
1320         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1321         char *pathname;
1322         int len;
1323
1324         if (!tmp)
1325                 return -ENOMEM;
1326
1327         pathname = d_path(path, tmp, PAGE_SIZE);
1328         len = PTR_ERR(pathname);
1329         if (IS_ERR(pathname))
1330                 goto out;
1331         len = tmp + PAGE_SIZE - 1 - pathname;
1332
1333         if (len > buflen)
1334                 len = buflen;
1335         if (copy_to_user(buffer, pathname, len))
1336                 len = -EFAULT;
1337  out:
1338         free_page((unsigned long)tmp);
1339         return len;
1340 }
1341
1342 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1343 {
1344         int error = -EACCES;
1345         struct inode *inode = dentry->d_inode;
1346         struct path path;
1347
1348         /* Are we allowed to snoop on the tasks file descriptors? */
1349         if (!proc_fd_access_allowed(inode))
1350                 goto out;
1351
1352         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1353         if (error)
1354                 goto out;
1355
1356         error = do_proc_readlink(&path, buffer, buflen);
1357         path_put(&path);
1358 out:
1359         return error;
1360 }
1361
1362 static const struct inode_operations proc_pid_link_inode_operations = {
1363         .readlink       = proc_pid_readlink,
1364         .follow_link    = proc_pid_follow_link,
1365         .setattr        = proc_setattr,
1366 };
1367
1368
1369 /* building an inode */
1370
1371 static int task_dumpable(struct task_struct *task)
1372 {
1373         int dumpable = 0;
1374         struct mm_struct *mm;
1375
1376         task_lock(task);
1377         mm = task->mm;
1378         if (mm)
1379                 dumpable = get_dumpable(mm);
1380         task_unlock(task);
1381         if(dumpable == 1)
1382                 return 1;
1383         return 0;
1384 }
1385
1386
1387 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1388 {
1389         struct inode * inode;
1390         struct proc_inode *ei;
1391
1392         /* We need a new inode */
1393
1394         inode = new_inode(sb);
1395         if (!inode)
1396                 goto out;
1397
1398         /* Common stuff */
1399         ei = PROC_I(inode);
1400         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1401         inode->i_op = &proc_def_inode_operations;
1402
1403         /*
1404          * grab the reference to task.
1405          */
1406         ei->pid = get_task_pid(task, PIDTYPE_PID);
1407         if (!ei->pid)
1408                 goto out_unlock;
1409
1410         inode->i_uid = 0;
1411         inode->i_gid = 0;
1412         if (task_dumpable(task)) {
1413                 inode->i_uid = task->euid;
1414                 inode->i_gid = task->egid;
1415         }
1416         security_task_to_inode(task, inode);
1417
1418 out:
1419         return inode;
1420
1421 out_unlock:
1422         iput(inode);
1423         return NULL;
1424 }
1425
1426 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1427 {
1428         struct inode *inode = dentry->d_inode;
1429         struct task_struct *task;
1430         generic_fillattr(inode, stat);
1431
1432         rcu_read_lock();
1433         stat->uid = 0;
1434         stat->gid = 0;
1435         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1436         if (task) {
1437                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1438                     task_dumpable(task)) {
1439                         stat->uid = task->euid;
1440                         stat->gid = task->egid;
1441                 }
1442         }
1443         rcu_read_unlock();
1444         return 0;
1445 }
1446
1447 /* dentry stuff */
1448
1449 /*
1450  *      Exceptional case: normally we are not allowed to unhash a busy
1451  * directory. In this case, however, we can do it - no aliasing problems
1452  * due to the way we treat inodes.
1453  *
1454  * Rewrite the inode's ownerships here because the owning task may have
1455  * performed a setuid(), etc.
1456  *
1457  * Before the /proc/pid/status file was created the only way to read
1458  * the effective uid of a /process was to stat /proc/pid.  Reading
1459  * /proc/pid/status is slow enough that procps and other packages
1460  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1461  * made this apply to all per process world readable and executable
1462  * directories.
1463  */
1464 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1465 {
1466         struct inode *inode = dentry->d_inode;
1467         struct task_struct *task = get_proc_task(inode);
1468         if (task) {
1469                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1470                     task_dumpable(task)) {
1471                         inode->i_uid = task->euid;
1472                         inode->i_gid = task->egid;
1473                 } else {
1474                         inode->i_uid = 0;
1475                         inode->i_gid = 0;
1476                 }
1477                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1478                 security_task_to_inode(task, inode);
1479                 put_task_struct(task);
1480                 return 1;
1481         }
1482         d_drop(dentry);
1483         return 0;
1484 }
1485
1486 static int pid_delete_dentry(struct dentry * dentry)
1487 {
1488         /* Is the task we represent dead?
1489          * If so, then don't put the dentry on the lru list,
1490          * kill it immediately.
1491          */
1492         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1493 }
1494
1495 static struct dentry_operations pid_dentry_operations =
1496 {
1497         .d_revalidate   = pid_revalidate,
1498         .d_delete       = pid_delete_dentry,
1499 };
1500
1501 /* Lookups */
1502
1503 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1504                                 struct task_struct *, const void *);
1505
1506 /*
1507  * Fill a directory entry.
1508  *
1509  * If possible create the dcache entry and derive our inode number and
1510  * file type from dcache entry.
1511  *
1512  * Since all of the proc inode numbers are dynamically generated, the inode
1513  * numbers do not exist until the inode is cache.  This means creating the
1514  * the dcache entry in readdir is necessary to keep the inode numbers
1515  * reported by readdir in sync with the inode numbers reported
1516  * by stat.
1517  */
1518 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1519         char *name, int len,
1520         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1521 {
1522         struct dentry *child, *dir = filp->f_path.dentry;
1523         struct inode *inode;
1524         struct qstr qname;
1525         ino_t ino = 0;
1526         unsigned type = DT_UNKNOWN;
1527
1528         qname.name = name;
1529         qname.len  = len;
1530         qname.hash = full_name_hash(name, len);
1531
1532         child = d_lookup(dir, &qname);
1533         if (!child) {
1534                 struct dentry *new;
1535                 new = d_alloc(dir, &qname);
1536                 if (new) {
1537                         child = instantiate(dir->d_inode, new, task, ptr);
1538                         if (child)
1539                                 dput(new);
1540                         else
1541                                 child = new;
1542                 }
1543         }
1544         if (!child || IS_ERR(child) || !child->d_inode)
1545                 goto end_instantiate;
1546         inode = child->d_inode;
1547         if (inode) {
1548                 ino = inode->i_ino;
1549                 type = inode->i_mode >> 12;
1550         }
1551         dput(child);
1552 end_instantiate:
1553         if (!ino)
1554                 ino = find_inode_number(dir, &qname);
1555         if (!ino)
1556                 ino = 1;
1557         return filldir(dirent, name, len, filp->f_pos, ino, type);
1558 }
1559
1560 static unsigned name_to_int(struct dentry *dentry)
1561 {
1562         const char *name = dentry->d_name.name;
1563         int len = dentry->d_name.len;
1564         unsigned n = 0;
1565
1566         if (len > 1 && *name == '0')
1567                 goto out;
1568         while (len-- > 0) {
1569                 unsigned c = *name++ - '0';
1570                 if (c > 9)
1571                         goto out;
1572                 if (n >= (~0U-9)/10)
1573                         goto out;
1574                 n *= 10;
1575                 n += c;
1576         }
1577         return n;
1578 out:
1579         return ~0U;
1580 }
1581
1582 #define PROC_FDINFO_MAX 64
1583
1584 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1585 {
1586         struct task_struct *task = get_proc_task(inode);
1587         struct files_struct *files = NULL;
1588         struct file *file;
1589         int fd = proc_fd(inode);
1590
1591         if (task) {
1592                 files = get_files_struct(task);
1593                 put_task_struct(task);
1594         }
1595         if (files) {
1596                 /*
1597                  * We are not taking a ref to the file structure, so we must
1598                  * hold ->file_lock.
1599                  */
1600                 spin_lock(&files->file_lock);
1601                 file = fcheck_files(files, fd);
1602                 if (file) {
1603                         if (path) {
1604                                 *path = file->f_path;
1605                                 path_get(&file->f_path);
1606                         }
1607                         if (info)
1608                                 snprintf(info, PROC_FDINFO_MAX,
1609                                          "pos:\t%lli\n"
1610                                          "flags:\t0%o\n",
1611                                          (long long) file->f_pos,
1612                                          file->f_flags);
1613                         spin_unlock(&files->file_lock);
1614                         put_files_struct(files);
1615                         return 0;
1616                 }
1617                 spin_unlock(&files->file_lock);
1618                 put_files_struct(files);
1619         }
1620         return -ENOENT;
1621 }
1622
1623 static int proc_fd_link(struct inode *inode, struct path *path)
1624 {
1625         return proc_fd_info(inode, path, NULL);
1626 }
1627
1628 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1629 {
1630         struct inode *inode = dentry->d_inode;
1631         struct task_struct *task = get_proc_task(inode);
1632         int fd = proc_fd(inode);
1633         struct files_struct *files;
1634
1635         if (task) {
1636                 files = get_files_struct(task);
1637                 if (files) {
1638                         rcu_read_lock();
1639                         if (fcheck_files(files, fd)) {
1640                                 rcu_read_unlock();
1641                                 put_files_struct(files);
1642                                 if (task_dumpable(task)) {
1643                                         inode->i_uid = task->euid;
1644                                         inode->i_gid = task->egid;
1645                                 } else {
1646                                         inode->i_uid = 0;
1647                                         inode->i_gid = 0;
1648                                 }
1649                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1650                                 security_task_to_inode(task, inode);
1651                                 put_task_struct(task);
1652                                 return 1;
1653                         }
1654                         rcu_read_unlock();
1655                         put_files_struct(files);
1656                 }
1657                 put_task_struct(task);
1658         }
1659         d_drop(dentry);
1660         return 0;
1661 }
1662
1663 static struct dentry_operations tid_fd_dentry_operations =
1664 {
1665         .d_revalidate   = tid_fd_revalidate,
1666         .d_delete       = pid_delete_dentry,
1667 };
1668
1669 static struct dentry *proc_fd_instantiate(struct inode *dir,
1670         struct dentry *dentry, struct task_struct *task, const void *ptr)
1671 {
1672         unsigned fd = *(const unsigned *)ptr;
1673         struct file *file;
1674         struct files_struct *files;
1675         struct inode *inode;
1676         struct proc_inode *ei;
1677         struct dentry *error = ERR_PTR(-ENOENT);
1678
1679         inode = proc_pid_make_inode(dir->i_sb, task);
1680         if (!inode)
1681                 goto out;
1682         ei = PROC_I(inode);
1683         ei->fd = fd;
1684         files = get_files_struct(task);
1685         if (!files)
1686                 goto out_iput;
1687         inode->i_mode = S_IFLNK;
1688
1689         /*
1690          * We are not taking a ref to the file structure, so we must
1691          * hold ->file_lock.
1692          */
1693         spin_lock(&files->file_lock);
1694         file = fcheck_files(files, fd);
1695         if (!file)
1696                 goto out_unlock;
1697         if (file->f_mode & 1)
1698                 inode->i_mode |= S_IRUSR | S_IXUSR;
1699         if (file->f_mode & 2)
1700                 inode->i_mode |= S_IWUSR | S_IXUSR;
1701         spin_unlock(&files->file_lock);
1702         put_files_struct(files);
1703
1704         inode->i_op = &proc_pid_link_inode_operations;
1705         inode->i_size = 64;
1706         ei->op.proc_get_link = proc_fd_link;
1707         dentry->d_op = &tid_fd_dentry_operations;
1708         d_add(dentry, inode);
1709         /* Close the race of the process dying before we return the dentry */
1710         if (tid_fd_revalidate(dentry, NULL))
1711                 error = NULL;
1712
1713  out:
1714         return error;
1715 out_unlock:
1716         spin_unlock(&files->file_lock);
1717         put_files_struct(files);
1718 out_iput:
1719         iput(inode);
1720         goto out;
1721 }
1722
1723 static struct dentry *proc_lookupfd_common(struct inode *dir,
1724                                            struct dentry *dentry,
1725                                            instantiate_t instantiate)
1726 {
1727         struct task_struct *task = get_proc_task(dir);
1728         unsigned fd = name_to_int(dentry);
1729         struct dentry *result = ERR_PTR(-ENOENT);
1730
1731         if (!task)
1732                 goto out_no_task;
1733         if (fd == ~0U)
1734                 goto out;
1735
1736         result = instantiate(dir, dentry, task, &fd);
1737 out:
1738         put_task_struct(task);
1739 out_no_task:
1740         return result;
1741 }
1742
1743 static int proc_readfd_common(struct file * filp, void * dirent,
1744                               filldir_t filldir, instantiate_t instantiate)
1745 {
1746         struct dentry *dentry = filp->f_path.dentry;
1747         struct inode *inode = dentry->d_inode;
1748         struct task_struct *p = get_proc_task(inode);
1749         unsigned int fd, ino;
1750         int retval;
1751         struct files_struct * files;
1752
1753         retval = -ENOENT;
1754         if (!p)
1755                 goto out_no_task;
1756         retval = 0;
1757
1758         fd = filp->f_pos;
1759         switch (fd) {
1760                 case 0:
1761                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1762                                 goto out;
1763                         filp->f_pos++;
1764                 case 1:
1765                         ino = parent_ino(dentry);
1766                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1767                                 goto out;
1768                         filp->f_pos++;
1769                 default:
1770                         files = get_files_struct(p);
1771                         if (!files)
1772                                 goto out;
1773                         rcu_read_lock();
1774                         for (fd = filp->f_pos-2;
1775                              fd < files_fdtable(files)->max_fds;
1776                              fd++, filp->f_pos++) {
1777                                 char name[PROC_NUMBUF];
1778                                 int len;
1779
1780                                 if (!fcheck_files(files, fd))
1781                                         continue;
1782                                 rcu_read_unlock();
1783
1784                                 len = snprintf(name, sizeof(name), "%d", fd);
1785                                 if (proc_fill_cache(filp, dirent, filldir,
1786                                                     name, len, instantiate,
1787                                                     p, &fd) < 0) {
1788                                         rcu_read_lock();
1789                                         break;
1790                                 }
1791                                 rcu_read_lock();
1792                         }
1793                         rcu_read_unlock();
1794                         put_files_struct(files);
1795         }
1796 out:
1797         put_task_struct(p);
1798 out_no_task:
1799         return retval;
1800 }
1801
1802 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1803                                     struct nameidata *nd)
1804 {
1805         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1806 }
1807
1808 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1809 {
1810         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1811 }
1812
1813 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1814                                       size_t len, loff_t *ppos)
1815 {
1816         char tmp[PROC_FDINFO_MAX];
1817         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1818         if (!err)
1819                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1820         return err;
1821 }
1822
1823 static const struct file_operations proc_fdinfo_file_operations = {
1824         .open           = nonseekable_open,
1825         .read           = proc_fdinfo_read,
1826 };
1827
1828 static const struct file_operations proc_fd_operations = {
1829         .read           = generic_read_dir,
1830         .readdir        = proc_readfd,
1831 };
1832
1833 /*
1834  * /proc/pid/fd needs a special permission handler so that a process can still
1835  * access /proc/self/fd after it has executed a setuid().
1836  */
1837 static int proc_fd_permission(struct inode *inode, int mask,
1838                                 struct nameidata *nd)
1839 {
1840         int rv;
1841
1842         rv = generic_permission(inode, mask, NULL);
1843         if (rv == 0)
1844                 return 0;
1845         if (task_pid(current) == proc_pid(inode))
1846                 rv = 0;
1847         return rv;
1848 }
1849
1850 /*
1851  * proc directories can do almost nothing..
1852  */
1853 static const struct inode_operations proc_fd_inode_operations = {
1854         .lookup         = proc_lookupfd,
1855         .permission     = proc_fd_permission,
1856         .setattr        = proc_setattr,
1857 };
1858
1859 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1860         struct dentry *dentry, struct task_struct *task, const void *ptr)
1861 {
1862         unsigned fd = *(unsigned *)ptr;
1863         struct inode *inode;
1864         struct proc_inode *ei;
1865         struct dentry *error = ERR_PTR(-ENOENT);
1866
1867         inode = proc_pid_make_inode(dir->i_sb, task);
1868         if (!inode)
1869                 goto out;
1870         ei = PROC_I(inode);
1871         ei->fd = fd;
1872         inode->i_mode = S_IFREG | S_IRUSR;
1873         inode->i_fop = &proc_fdinfo_file_operations;
1874         dentry->d_op = &tid_fd_dentry_operations;
1875         d_add(dentry, inode);
1876         /* Close the race of the process dying before we return the dentry */
1877         if (tid_fd_revalidate(dentry, NULL))
1878                 error = NULL;
1879
1880  out:
1881         return error;
1882 }
1883
1884 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1885                                         struct dentry *dentry,
1886                                         struct nameidata *nd)
1887 {
1888         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1889 }
1890
1891 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1892 {
1893         return proc_readfd_common(filp, dirent, filldir,
1894                                   proc_fdinfo_instantiate);
1895 }
1896
1897 static const struct file_operations proc_fdinfo_operations = {
1898         .read           = generic_read_dir,
1899         .readdir        = proc_readfdinfo,
1900 };
1901
1902 /*
1903  * proc directories can do almost nothing..
1904  */
1905 static const struct inode_operations proc_fdinfo_inode_operations = {
1906         .lookup         = proc_lookupfdinfo,
1907         .setattr        = proc_setattr,
1908 };
1909
1910
1911 static struct dentry *proc_pident_instantiate(struct inode *dir,
1912         struct dentry *dentry, struct task_struct *task, const void *ptr)
1913 {
1914         const struct pid_entry *p = ptr;
1915         struct inode *inode;
1916         struct proc_inode *ei;
1917         struct dentry *error = ERR_PTR(-EINVAL);
1918
1919         inode = proc_pid_make_inode(dir->i_sb, task);
1920         if (!inode)
1921                 goto out;
1922
1923         ei = PROC_I(inode);
1924         inode->i_mode = p->mode;
1925         if (S_ISDIR(inode->i_mode))
1926                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1927         if (p->iop)
1928                 inode->i_op = p->iop;
1929         if (p->fop)
1930                 inode->i_fop = p->fop;
1931         ei->op = p->op;
1932         dentry->d_op = &pid_dentry_operations;
1933         d_add(dentry, inode);
1934         /* Close the race of the process dying before we return the dentry */
1935         if (pid_revalidate(dentry, NULL))
1936                 error = NULL;
1937 out:
1938         return error;
1939 }
1940
1941 static struct dentry *proc_pident_lookup(struct inode *dir, 
1942                                          struct dentry *dentry,
1943                                          const struct pid_entry *ents,
1944                                          unsigned int nents)
1945 {
1946         struct inode *inode;
1947         struct dentry *error;
1948         struct task_struct *task = get_proc_task(dir);
1949         const struct pid_entry *p, *last;
1950
1951         error = ERR_PTR(-ENOENT);
1952         inode = NULL;
1953
1954         if (!task)
1955                 goto out_no_task;
1956
1957         /*
1958          * Yes, it does not scale. And it should not. Don't add
1959          * new entries into /proc/<tgid>/ without very good reasons.
1960          */
1961         last = &ents[nents - 1];
1962         for (p = ents; p <= last; p++) {
1963                 if (p->len != dentry->d_name.len)
1964                         continue;
1965                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1966                         break;
1967         }
1968         if (p > last)
1969                 goto out;
1970
1971         error = proc_pident_instantiate(dir, dentry, task, p);
1972 out:
1973         put_task_struct(task);
1974 out_no_task:
1975         return error;
1976 }
1977
1978 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1979         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1980 {
1981         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1982                                 proc_pident_instantiate, task, p);
1983 }
1984
1985 static int proc_pident_readdir(struct file *filp,
1986                 void *dirent, filldir_t filldir,
1987                 const struct pid_entry *ents, unsigned int nents)
1988 {
1989         int i;
1990         struct dentry *dentry = filp->f_path.dentry;
1991         struct inode *inode = dentry->d_inode;
1992         struct task_struct *task = get_proc_task(inode);
1993         const struct pid_entry *p, *last;
1994         ino_t ino;
1995         int ret;
1996
1997         ret = -ENOENT;
1998         if (!task)
1999                 goto out_no_task;
2000
2001         ret = 0;
2002         i = filp->f_pos;
2003         switch (i) {
2004         case 0:
2005                 ino = inode->i_ino;
2006                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2007                         goto out;
2008                 i++;
2009                 filp->f_pos++;
2010                 /* fall through */
2011         case 1:
2012                 ino = parent_ino(dentry);
2013                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2014                         goto out;
2015                 i++;
2016                 filp->f_pos++;
2017                 /* fall through */
2018         default:
2019                 i -= 2;
2020                 if (i >= nents) {
2021                         ret = 1;
2022                         goto out;
2023                 }
2024                 p = ents + i;
2025                 last = &ents[nents - 1];
2026                 while (p <= last) {
2027                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2028                                 goto out;
2029                         filp->f_pos++;
2030                         p++;
2031                 }
2032         }
2033
2034         ret = 1;
2035 out:
2036         put_task_struct(task);
2037 out_no_task:
2038         return ret;
2039 }
2040
2041 #ifdef CONFIG_SECURITY
2042 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2043                                   size_t count, loff_t *ppos)
2044 {
2045         struct inode * inode = file->f_path.dentry->d_inode;
2046         char *p = NULL;
2047         ssize_t length;
2048         struct task_struct *task = get_proc_task(inode);
2049
2050         if (!task)
2051                 return -ESRCH;
2052
2053         length = security_getprocattr(task,
2054                                       (char*)file->f_path.dentry->d_name.name,
2055                                       &p);
2056         put_task_struct(task);
2057         if (length > 0)
2058                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2059         kfree(p);
2060         return length;
2061 }
2062
2063 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2064                                    size_t count, loff_t *ppos)
2065 {
2066         struct inode * inode = file->f_path.dentry->d_inode;
2067         char *page;
2068         ssize_t length;
2069         struct task_struct *task = get_proc_task(inode);
2070
2071         length = -ESRCH;
2072         if (!task)
2073                 goto out_no_task;
2074         if (count > PAGE_SIZE)
2075                 count = PAGE_SIZE;
2076
2077         /* No partial writes. */
2078         length = -EINVAL;
2079         if (*ppos != 0)
2080                 goto out;
2081
2082         length = -ENOMEM;
2083         page = (char*)__get_free_page(GFP_TEMPORARY);
2084         if (!page)
2085                 goto out;
2086
2087         length = -EFAULT;
2088         if (copy_from_user(page, buf, count))
2089                 goto out_free;
2090
2091         length = security_setprocattr(task,
2092                                       (char*)file->f_path.dentry->d_name.name,
2093                                       (void*)page, count);
2094 out_free:
2095         free_page((unsigned long) page);
2096 out:
2097         put_task_struct(task);
2098 out_no_task:
2099         return length;
2100 }
2101
2102 static const struct file_operations proc_pid_attr_operations = {
2103         .read           = proc_pid_attr_read,
2104         .write          = proc_pid_attr_write,
2105 };
2106
2107 static const struct pid_entry attr_dir_stuff[] = {
2108         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
2109         REG("prev",       S_IRUGO,         pid_attr),
2110         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
2111         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
2112         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
2113         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
2114 };
2115
2116 static int proc_attr_dir_readdir(struct file * filp,
2117                              void * dirent, filldir_t filldir)
2118 {
2119         return proc_pident_readdir(filp,dirent,filldir,
2120                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2121 }
2122
2123 static const struct file_operations proc_attr_dir_operations = {
2124         .read           = generic_read_dir,
2125         .readdir        = proc_attr_dir_readdir,
2126 };
2127
2128 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2129                                 struct dentry *dentry, struct nameidata *nd)
2130 {
2131         return proc_pident_lookup(dir, dentry,
2132                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2133 }
2134
2135 static const struct inode_operations proc_attr_dir_inode_operations = {
2136         .lookup         = proc_attr_dir_lookup,
2137         .getattr        = pid_getattr,
2138         .setattr        = proc_setattr,
2139 };
2140
2141 #endif
2142
2143 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2144 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2145                                          size_t count, loff_t *ppos)
2146 {
2147         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2148         struct mm_struct *mm;
2149         char buffer[PROC_NUMBUF];
2150         size_t len;
2151         int ret;
2152
2153         if (!task)
2154                 return -ESRCH;
2155
2156         ret = 0;
2157         mm = get_task_mm(task);
2158         if (mm) {
2159                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2160                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2161                                 MMF_DUMP_FILTER_SHIFT));
2162                 mmput(mm);
2163                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2164         }
2165
2166         put_task_struct(task);
2167
2168         return ret;
2169 }
2170
2171 static ssize_t proc_coredump_filter_write(struct file *file,
2172                                           const char __user *buf,
2173                                           size_t count,
2174                                           loff_t *ppos)
2175 {
2176         struct task_struct *task;
2177         struct mm_struct *mm;
2178         char buffer[PROC_NUMBUF], *end;
2179         unsigned int val;
2180         int ret;
2181         int i;
2182         unsigned long mask;
2183
2184         ret = -EFAULT;
2185         memset(buffer, 0, sizeof(buffer));
2186         if (count > sizeof(buffer) - 1)
2187                 count = sizeof(buffer) - 1;
2188         if (copy_from_user(buffer, buf, count))
2189                 goto out_no_task;
2190
2191         ret = -EINVAL;
2192         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2193         if (*end == '\n')
2194                 end++;
2195         if (end - buffer == 0)
2196                 goto out_no_task;
2197
2198         ret = -ESRCH;
2199         task = get_proc_task(file->f_dentry->d_inode);
2200         if (!task)
2201                 goto out_no_task;
2202
2203         ret = end - buffer;
2204         mm = get_task_mm(task);
2205         if (!mm)
2206                 goto out_no_mm;
2207
2208         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2209                 if (val & mask)
2210                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2211                 else
2212                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2213         }
2214
2215         mmput(mm);
2216  out_no_mm:
2217         put_task_struct(task);
2218  out_no_task:
2219         return ret;
2220 }
2221
2222 static const struct file_operations proc_coredump_filter_operations = {
2223         .read           = proc_coredump_filter_read,
2224         .write          = proc_coredump_filter_write,
2225 };
2226 #endif
2227
2228 /*
2229  * /proc/self:
2230  */
2231 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2232                               int buflen)
2233 {
2234         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2235         pid_t tgid = task_tgid_nr_ns(current, ns);
2236         char tmp[PROC_NUMBUF];
2237         if (!tgid)
2238                 return -ENOENT;
2239         sprintf(tmp, "%d", tgid);
2240         return vfs_readlink(dentry,buffer,buflen,tmp);
2241 }
2242
2243 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2244 {
2245         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2246         pid_t tgid = task_tgid_nr_ns(current, ns);
2247         char tmp[PROC_NUMBUF];
2248         if (!tgid)
2249                 return ERR_PTR(-ENOENT);
2250         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2251         return ERR_PTR(vfs_follow_link(nd,tmp));
2252 }
2253
2254 static const struct inode_operations proc_self_inode_operations = {
2255         .readlink       = proc_self_readlink,
2256         .follow_link    = proc_self_follow_link,
2257 };
2258
2259 /*
2260  * proc base
2261  *
2262  * These are the directory entries in the root directory of /proc
2263  * that properly belong to the /proc filesystem, as they describe
2264  * describe something that is process related.
2265  */
2266 static const struct pid_entry proc_base_stuff[] = {
2267         NOD("self", S_IFLNK|S_IRWXUGO,
2268                 &proc_self_inode_operations, NULL, {}),
2269 };
2270
2271 /*
2272  *      Exceptional case: normally we are not allowed to unhash a busy
2273  * directory. In this case, however, we can do it - no aliasing problems
2274  * due to the way we treat inodes.
2275  */
2276 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2277 {
2278         struct inode *inode = dentry->d_inode;
2279         struct task_struct *task = get_proc_task(inode);
2280         if (task) {
2281                 put_task_struct(task);
2282                 return 1;
2283         }
2284         d_drop(dentry);
2285         return 0;
2286 }
2287
2288 static struct dentry_operations proc_base_dentry_operations =
2289 {
2290         .d_revalidate   = proc_base_revalidate,
2291         .d_delete       = pid_delete_dentry,
2292 };
2293
2294 static struct dentry *proc_base_instantiate(struct inode *dir,
2295         struct dentry *dentry, struct task_struct *task, const void *ptr)
2296 {
2297         const struct pid_entry *p = ptr;
2298         struct inode *inode;
2299         struct proc_inode *ei;
2300         struct dentry *error = ERR_PTR(-EINVAL);
2301
2302         /* Allocate the inode */
2303         error = ERR_PTR(-ENOMEM);
2304         inode = new_inode(dir->i_sb);
2305         if (!inode)
2306                 goto out;
2307
2308         /* Initialize the inode */
2309         ei = PROC_I(inode);
2310         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2311
2312         /*
2313          * grab the reference to the task.
2314          */
2315         ei->pid = get_task_pid(task, PIDTYPE_PID);
2316         if (!ei->pid)
2317                 goto out_iput;
2318
2319         inode->i_uid = 0;
2320         inode->i_gid = 0;
2321         inode->i_mode = p->mode;
2322         if (S_ISDIR(inode->i_mode))
2323                 inode->i_nlink = 2;
2324         if (S_ISLNK(inode->i_mode))
2325                 inode->i_size = 64;
2326         if (p->iop)
2327                 inode->i_op = p->iop;
2328         if (p->fop)
2329                 inode->i_fop = p->fop;
2330         ei->op = p->op;
2331         dentry->d_op = &proc_base_dentry_operations;
2332         d_add(dentry, inode);
2333         error = NULL;
2334 out:
2335         return error;
2336 out_iput:
2337         iput(inode);
2338         goto out;
2339 }
2340
2341 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2342 {
2343         struct dentry *error;
2344         struct task_struct *task = get_proc_task(dir);
2345         const struct pid_entry *p, *last;
2346
2347         error = ERR_PTR(-ENOENT);
2348
2349         if (!task)
2350                 goto out_no_task;
2351
2352         /* Lookup the directory entry */
2353         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2354         for (p = proc_base_stuff; p <= last; p++) {
2355                 if (p->len != dentry->d_name.len)
2356                         continue;
2357                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2358                         break;
2359         }
2360         if (p > last)
2361                 goto out;
2362
2363         error = proc_base_instantiate(dir, dentry, task, p);
2364
2365 out:
2366         put_task_struct(task);
2367 out_no_task:
2368         return error;
2369 }
2370
2371 static int proc_base_fill_cache(struct file *filp, void *dirent,
2372         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2373 {
2374         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2375                                 proc_base_instantiate, task, p);
2376 }
2377
2378 #ifdef CONFIG_TASK_IO_ACCOUNTING
2379 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2380 {
2381         u64 rchar, wchar, syscr, syscw;
2382         struct task_io_accounting ioac;
2383
2384         if (!whole) {
2385                 rchar = task->rchar;
2386                 wchar = task->wchar;
2387                 syscr = task->syscr;
2388                 syscw = task->syscw;
2389                 memcpy(&ioac, &task->ioac, sizeof(ioac));
2390         } else {
2391                 unsigned long flags;
2392                 struct task_struct *t = task;
2393                 rchar = wchar = syscr = syscw = 0;
2394                 memset(&ioac, 0, sizeof(ioac));
2395
2396                 rcu_read_lock();
2397                 do {
2398                         rchar += t->rchar;
2399                         wchar += t->wchar;
2400                         syscr += t->syscr;
2401                         syscw += t->syscw;
2402
2403                         ioac.read_bytes += t->ioac.read_bytes;
2404                         ioac.write_bytes += t->ioac.write_bytes;
2405                         ioac.cancelled_write_bytes +=
2406                                         t->ioac.cancelled_write_bytes;
2407                         t = next_thread(t);
2408                 } while (t != task);
2409                 rcu_read_unlock();
2410
2411                 if (lock_task_sighand(task, &flags)) {
2412                         struct signal_struct *sig = task->signal;
2413
2414                         rchar += sig->rchar;
2415                         wchar += sig->wchar;
2416                         syscr += sig->syscr;
2417                         syscw += sig->syscw;
2418
2419                         ioac.read_bytes += sig->ioac.read_bytes;
2420                         ioac.write_bytes += sig->ioac.write_bytes;
2421                         ioac.cancelled_write_bytes +=
2422                                         sig->ioac.cancelled_write_bytes;
2423
2424                         unlock_task_sighand(task, &flags);
2425                 }
2426         }
2427
2428         return sprintf(buffer,
2429                         "rchar: %llu\n"
2430                         "wchar: %llu\n"
2431                         "syscr: %llu\n"
2432                         "syscw: %llu\n"
2433                         "read_bytes: %llu\n"
2434                         "write_bytes: %llu\n"
2435                         "cancelled_write_bytes: %llu\n",
2436                         (unsigned long long)rchar,
2437                         (unsigned long long)wchar,
2438                         (unsigned long long)syscr,
2439                         (unsigned long long)syscw,
2440                         (unsigned long long)ioac.read_bytes,
2441                         (unsigned long long)ioac.write_bytes,
2442                         (unsigned long long)ioac.cancelled_write_bytes);
2443 }
2444
2445 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2446 {
2447         return do_io_accounting(task, buffer, 0);
2448 }
2449
2450 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2451 {
2452         return do_io_accounting(task, buffer, 1);
2453 }
2454 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2455
2456 /*
2457  * Thread groups
2458  */
2459 static const struct file_operations proc_task_operations;
2460 static const struct inode_operations proc_task_inode_operations;
2461
2462 static const struct pid_entry tgid_base_stuff[] = {
2463         DIR("task",       S_IRUGO|S_IXUGO, task),
2464         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2465         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2466 #ifdef CONFIG_NET
2467         DIR("net",        S_IRUGO|S_IXUGO, net),
2468 #endif
2469         REG("environ",    S_IRUSR, environ),
2470         INF("auxv",       S_IRUSR, pid_auxv),
2471         ONE("status",     S_IRUGO, pid_status),
2472         INF("limits",     S_IRUSR, pid_limits),
2473 #ifdef CONFIG_SCHED_DEBUG
2474         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2475 #endif
2476         INF("cmdline",    S_IRUGO, pid_cmdline),
2477         ONE("stat",       S_IRUGO, tgid_stat),
2478         ONE("statm",      S_IRUGO, pid_statm),
2479         REG("maps",       S_IRUGO, maps),
2480 #ifdef CONFIG_NUMA
2481         REG("numa_maps",  S_IRUGO, numa_maps),
2482 #endif
2483         REG("mem",        S_IRUSR|S_IWUSR, mem),
2484         LNK("cwd",        cwd),
2485         LNK("root",       root),
2486         LNK("exe",        exe),
2487         REG("mounts",     S_IRUGO, mounts),
2488         REG("mountinfo",  S_IRUGO, mountinfo),
2489         REG("mountstats", S_IRUSR, mountstats),
2490 #ifdef CONFIG_PROC_PAGE_MONITOR
2491         REG("clear_refs", S_IWUSR, clear_refs),
2492         REG("smaps",      S_IRUGO, smaps),
2493         REG("pagemap",    S_IRUSR, pagemap),
2494 #endif
2495 #ifdef CONFIG_SECURITY
2496         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2497 #endif
2498 #ifdef CONFIG_KALLSYMS
2499         INF("wchan",      S_IRUGO, pid_wchan),
2500 #endif
2501 #ifdef CONFIG_SCHEDSTATS
2502         INF("schedstat",  S_IRUGO, pid_schedstat),
2503 #endif
2504 #ifdef CONFIG_LATENCYTOP
2505         REG("latency",  S_IRUGO, lstats),
2506 #endif
2507 #ifdef CONFIG_PROC_PID_CPUSET
2508         REG("cpuset",     S_IRUGO, cpuset),
2509 #endif
2510 #ifdef CONFIG_CGROUPS
2511         REG("cgroup",  S_IRUGO, cgroup),
2512 #endif
2513         INF("oom_score",  S_IRUGO, oom_score),
2514         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2515 #ifdef CONFIG_AUDITSYSCALL
2516         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2517         REG("sessionid",  S_IRUGO, sessionid),
2518 #endif
2519 #ifdef CONFIG_FAULT_INJECTION
2520         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2521 #endif
2522 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2523         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2524 #endif
2525 #ifdef CONFIG_TASK_IO_ACCOUNTING
2526         INF("io",       S_IRUGO, tgid_io_accounting),
2527 #endif
2528 };
2529
2530 static int proc_tgid_base_readdir(struct file * filp,
2531                              void * dirent, filldir_t filldir)
2532 {
2533         return proc_pident_readdir(filp,dirent,filldir,
2534                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2535 }
2536
2537 static const struct file_operations proc_tgid_base_operations = {
2538         .read           = generic_read_dir,
2539         .readdir        = proc_tgid_base_readdir,
2540 };
2541
2542 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2543         return proc_pident_lookup(dir, dentry,
2544                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2545 }
2546
2547 static const struct inode_operations proc_tgid_base_inode_operations = {
2548         .lookup         = proc_tgid_base_lookup,
2549         .getattr        = pid_getattr,
2550         .setattr        = proc_setattr,
2551 };
2552
2553 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2554 {
2555         struct dentry *dentry, *leader, *dir;
2556         char buf[PROC_NUMBUF];
2557         struct qstr name;
2558
2559         name.name = buf;
2560         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2561         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2562         if (dentry) {
2563                 if (!(current->flags & PF_EXITING))
2564                         shrink_dcache_parent(dentry);
2565                 d_drop(dentry);
2566                 dput(dentry);
2567         }
2568
2569         if (tgid == 0)
2570                 goto out;
2571
2572         name.name = buf;
2573         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2574         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2575         if (!leader)
2576                 goto out;
2577
2578         name.name = "task";
2579         name.len = strlen(name.name);
2580         dir = d_hash_and_lookup(leader, &name);
2581         if (!dir)
2582                 goto out_put_leader;
2583
2584         name.name = buf;
2585         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2586         dentry = d_hash_and_lookup(dir, &name);
2587         if (dentry) {
2588                 shrink_dcache_parent(dentry);
2589                 d_drop(dentry);
2590                 dput(dentry);
2591         }
2592
2593         dput(dir);
2594 out_put_leader:
2595         dput(leader);
2596 out:
2597         return;
2598 }
2599
2600 /**
2601  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2602  * @task: task that should be flushed.
2603  *
2604  * When flushing dentries from proc, one needs to flush them from global
2605  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2606  * in. This call is supposed to do all of this job.
2607  *
2608  * Looks in the dcache for
2609  * /proc/@pid
2610  * /proc/@tgid/task/@pid
2611  * if either directory is present flushes it and all of it'ts children
2612  * from the dcache.
2613  *
2614  * It is safe and reasonable to cache /proc entries for a task until
2615  * that task exits.  After that they just clog up the dcache with
2616  * useless entries, possibly causing useful dcache entries to be
2617  * flushed instead.  This routine is proved to flush those useless
2618  * dcache entries at process exit time.
2619  *
2620  * NOTE: This routine is just an optimization so it does not guarantee
2621  *       that no dcache entries will exist at process exit time it
2622  *       just makes it very unlikely that any will persist.
2623  */
2624
2625 void proc_flush_task(struct task_struct *task)
2626 {
2627         int i;
2628         struct pid *pid, *tgid = NULL;
2629         struct upid *upid;
2630
2631         pid = task_pid(task);
2632         if (thread_group_leader(task))
2633                 tgid = task_tgid(task);
2634
2635         for (i = 0; i <= pid->level; i++) {
2636                 upid = &pid->numbers[i];
2637                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2638                         tgid ? tgid->numbers[i].nr : 0);
2639         }
2640
2641         upid = &pid->numbers[pid->level];
2642         if (upid->nr == 1)
2643                 pid_ns_release_proc(upid->ns);
2644 }
2645
2646 static struct dentry *proc_pid_instantiate(struct inode *dir,
2647                                            struct dentry * dentry,
2648                                            struct task_struct *task, const void *ptr)
2649 {
2650         struct dentry *error = ERR_PTR(-ENOENT);
2651         struct inode *inode;
2652
2653         inode = proc_pid_make_inode(dir->i_sb, task);
2654         if (!inode)
2655                 goto out;
2656
2657         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2658         inode->i_op = &proc_tgid_base_inode_operations;
2659         inode->i_fop = &proc_tgid_base_operations;
2660         inode->i_flags|=S_IMMUTABLE;
2661
2662         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2663                 ARRAY_SIZE(tgid_base_stuff));
2664
2665         dentry->d_op = &pid_dentry_operations;
2666
2667         d_add(dentry, inode);
2668         /* Close the race of the process dying before we return the dentry */
2669         if (pid_revalidate(dentry, NULL))
2670                 error = NULL;
2671 out:
2672         return error;
2673 }
2674
2675 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2676 {
2677         struct dentry *result = ERR_PTR(-ENOENT);
2678         struct task_struct *task;
2679         unsigned tgid;
2680         struct pid_namespace *ns;
2681
2682         result = proc_base_lookup(dir, dentry);
2683         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2684                 goto out;
2685
2686         tgid = name_to_int(dentry);
2687         if (tgid == ~0U)
2688                 goto out;
2689
2690         ns = dentry->d_sb->s_fs_info;
2691         rcu_read_lock();
2692         task = find_task_by_pid_ns(tgid, ns);
2693         if (task)
2694                 get_task_struct(task);
2695         rcu_read_unlock();
2696         if (!task)
2697                 goto out;
2698
2699         result = proc_pid_instantiate(dir, dentry, task, NULL);
2700         put_task_struct(task);
2701 out:
2702         return result;
2703 }
2704
2705 /*
2706  * Find the first task with tgid >= tgid
2707  *
2708  */
2709 struct tgid_iter {
2710         unsigned int tgid;
2711         struct task_struct *task;
2712 };
2713 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2714 {
2715         struct pid *pid;
2716
2717         if (iter.task)
2718                 put_task_struct(iter.task);
2719         rcu_read_lock();
2720 retry:
2721         iter.task = NULL;
2722         pid = find_ge_pid(iter.tgid, ns);
2723         if (pid) {
2724                 iter.tgid = pid_nr_ns(pid, ns);
2725                 iter.task = pid_task(pid, PIDTYPE_PID);
2726                 /* What we to know is if the pid we have find is the
2727                  * pid of a thread_group_leader.  Testing for task
2728                  * being a thread_group_leader is the obvious thing
2729                  * todo but there is a window when it fails, due to
2730                  * the pid transfer logic in de_thread.
2731                  *
2732                  * So we perform the straight forward test of seeing
2733                  * if the pid we have found is the pid of a thread
2734                  * group leader, and don't worry if the task we have
2735                  * found doesn't happen to be a thread group leader.
2736                  * As we don't care in the case of readdir.
2737                  */
2738                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2739                         iter.tgid += 1;
2740                         goto retry;
2741                 }
2742                 get_task_struct(iter.task);
2743         }
2744         rcu_read_unlock();
2745         return iter;
2746 }
2747
2748 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2749
2750 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2751         struct tgid_iter iter)
2752 {
2753         char name[PROC_NUMBUF];
2754         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2755         return proc_fill_cache(filp, dirent, filldir, name, len,
2756                                 proc_pid_instantiate, iter.task, NULL);
2757 }
2758
2759 /* for the /proc/ directory itself, after non-process stuff has been done */
2760 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2761 {
2762         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2763         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2764         struct tgid_iter iter;
2765         struct pid_namespace *ns;
2766
2767         if (!reaper)
2768                 goto out_no_task;
2769
2770         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2771                 const struct pid_entry *p = &proc_base_stuff[nr];
2772                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2773                         goto out;
2774         }
2775
2776         ns = filp->f_dentry->d_sb->s_fs_info;
2777         iter.task = NULL;
2778         iter.tgid = filp->f_pos - TGID_OFFSET;
2779         for (iter = next_tgid(ns, iter);
2780              iter.task;
2781              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2782                 filp->f_pos = iter.tgid + TGID_OFFSET;
2783                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2784                         put_task_struct(iter.task);
2785                         goto out;
2786                 }
2787         }
2788         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2789 out:
2790         put_task_struct(reaper);
2791 out_no_task:
2792         return 0;
2793 }
2794
2795 /*
2796  * Tasks
2797  */
2798 static const struct pid_entry tid_base_stuff[] = {
2799         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2800         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2801         REG("environ",   S_IRUSR, environ),
2802         INF("auxv",      S_IRUSR, pid_auxv),
2803         ONE("status",    S_IRUGO, pid_status),
2804         INF("limits",    S_IRUSR, pid_limits),
2805 #ifdef CONFIG_SCHED_DEBUG
2806         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2807 #endif
2808         INF("cmdline",   S_IRUGO, pid_cmdline),
2809         ONE("stat",      S_IRUGO, tid_stat),
2810         ONE("statm",     S_IRUGO, pid_statm),
2811         REG("maps",      S_IRUGO, maps),
2812 #ifdef CONFIG_NUMA
2813         REG("numa_maps", S_IRUGO, numa_maps),
2814 #endif
2815         REG("mem",       S_IRUSR|S_IWUSR, mem),
2816         LNK("cwd",       cwd),
2817         LNK("root",      root),
2818         LNK("exe",       exe),
2819         REG("mounts",    S_IRUGO, mounts),
2820         REG("mountinfo",  S_IRUGO, mountinfo),
2821 #ifdef CONFIG_PROC_PAGE_MONITOR
2822         REG("clear_refs", S_IWUSR, clear_refs),
2823         REG("smaps",     S_IRUGO, smaps),
2824         REG("pagemap",    S_IRUSR, pagemap),
2825 #endif
2826 #ifdef CONFIG_SECURITY
2827         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2828 #endif
2829 #ifdef CONFIG_KALLSYMS
2830         INF("wchan",     S_IRUGO, pid_wchan),
2831 #endif
2832 #ifdef CONFIG_SCHEDSTATS
2833         INF("schedstat", S_IRUGO, pid_schedstat),
2834 #endif
2835 #ifdef CONFIG_LATENCYTOP
2836         REG("latency",  S_IRUGO, lstats),
2837 #endif
2838 #ifdef CONFIG_PROC_PID_CPUSET
2839         REG("cpuset",    S_IRUGO, cpuset),
2840 #endif
2841 #ifdef CONFIG_CGROUPS
2842         REG("cgroup",  S_IRUGO, cgroup),
2843 #endif
2844         INF("oom_score", S_IRUGO, oom_score),
2845         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2846 #ifdef CONFIG_AUDITSYSCALL
2847         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2848         REG("sessionid",  S_IRUSR, sessionid),
2849 #endif
2850 #ifdef CONFIG_FAULT_INJECTION
2851         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2852 #endif
2853 #ifdef CONFIG_TASK_IO_ACCOUNTING
2854         INF("io",       S_IRUGO, tid_io_accounting),
2855 #endif
2856 };
2857
2858 static int proc_tid_base_readdir(struct file * filp,
2859                              void * dirent, filldir_t filldir)
2860 {
2861         return proc_pident_readdir(filp,dirent,filldir,
2862                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2863 }
2864
2865 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2866         return proc_pident_lookup(dir, dentry,
2867                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2868 }
2869
2870 static const struct file_operations proc_tid_base_operations = {
2871         .read           = generic_read_dir,
2872         .readdir        = proc_tid_base_readdir,
2873 };
2874
2875 static const struct inode_operations proc_tid_base_inode_operations = {
2876         .lookup         = proc_tid_base_lookup,
2877         .getattr        = pid_getattr,
2878         .setattr        = proc_setattr,
2879 };
2880
2881 static struct dentry *proc_task_instantiate(struct inode *dir,
2882         struct dentry *dentry, struct task_struct *task, const void *ptr)
2883 {
2884         struct dentry *error = ERR_PTR(-ENOENT);
2885         struct inode *inode;
2886         inode = proc_pid_make_inode(dir->i_sb, task);
2887
2888         if (!inode)
2889                 goto out;
2890         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2891         inode->i_op = &proc_tid_base_inode_operations;
2892         inode->i_fop = &proc_tid_base_operations;
2893         inode->i_flags|=S_IMMUTABLE;
2894
2895         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2896                 ARRAY_SIZE(tid_base_stuff));
2897
2898         dentry->d_op = &pid_dentry_operations;
2899
2900         d_add(dentry, inode);
2901         /* Close the race of the process dying before we return the dentry */
2902         if (pid_revalidate(dentry, NULL))
2903                 error = NULL;
2904 out:
2905         return error;
2906 }
2907
2908 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2909 {
2910         struct dentry *result = ERR_PTR(-ENOENT);
2911         struct task_struct *task;
2912         struct task_struct *leader = get_proc_task(dir);
2913         unsigned tid;
2914         struct pid_namespace *ns;
2915
2916         if (!leader)
2917                 goto out_no_task;
2918
2919         tid = name_to_int(dentry);
2920         if (tid == ~0U)
2921                 goto out;
2922
2923         ns = dentry->d_sb->s_fs_info;
2924         rcu_read_lock();
2925         task = find_task_by_pid_ns(tid, ns);
2926         if (task)
2927                 get_task_struct(task);
2928         rcu_read_unlock();
2929         if (!task)
2930                 goto out;
2931         if (!same_thread_group(leader, task))
2932                 goto out_drop_task;
2933
2934         result = proc_task_instantiate(dir, dentry, task, NULL);
2935 out_drop_task:
2936         put_task_struct(task);
2937 out:
2938         put_task_struct(leader);
2939 out_no_task:
2940         return result;
2941 }
2942
2943 /*
2944  * Find the first tid of a thread group to return to user space.
2945  *
2946  * Usually this is just the thread group leader, but if the users
2947  * buffer was too small or there was a seek into the middle of the
2948  * directory we have more work todo.
2949  *
2950  * In the case of a short read we start with find_task_by_pid.
2951  *
2952  * In the case of a seek we start with the leader and walk nr
2953  * threads past it.
2954  */
2955 static struct task_struct *first_tid(struct task_struct *leader,
2956                 int tid, int nr, struct pid_namespace *ns)
2957 {
2958         struct task_struct *pos;
2959
2960         rcu_read_lock();
2961         /* Attempt to start with the pid of a thread */
2962         if (tid && (nr > 0)) {
2963                 pos = find_task_by_pid_ns(tid, ns);
2964                 if (pos && (pos->group_leader == leader))
2965                         goto found;
2966         }
2967
2968         /* If nr exceeds the number of threads there is nothing todo */
2969         pos = NULL;
2970         if (nr && nr >= get_nr_threads(leader))
2971                 goto out;
2972
2973         /* If we haven't found our starting place yet start
2974          * with the leader and walk nr threads forward.
2975          */
2976         for (pos = leader; nr > 0; --nr) {
2977                 pos = next_thread(pos);
2978                 if (pos == leader) {
2979                         pos = NULL;
2980                         goto out;
2981                 }
2982         }
2983 found:
2984         get_task_struct(pos);
2985 out:
2986         rcu_read_unlock();
2987         return pos;
2988 }
2989
2990 /*
2991  * Find the next thread in the thread list.
2992  * Return NULL if there is an error or no next thread.
2993  *
2994  * The reference to the input task_struct is released.
2995  */
2996 static struct task_struct *next_tid(struct task_struct *start)
2997 {
2998         struct task_struct *pos = NULL;
2999         rcu_read_lock();
3000         if (pid_alive(start)) {
3001                 pos = next_thread(start);
3002                 if (thread_group_leader(pos))
3003                         pos = NULL;
3004                 else
3005                         get_task_struct(pos);
3006         }
3007         rcu_read_unlock();
3008         put_task_struct(start);
3009         return pos;
3010 }
3011
3012 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3013         struct task_struct *task, int tid)
3014 {
3015         char name[PROC_NUMBUF];
3016         int len = snprintf(name, sizeof(name), "%d", tid);
3017         return proc_fill_cache(filp, dirent, filldir, name, len,
3018                                 proc_task_instantiate, task, NULL);
3019 }
3020
3021 /* for the /proc/TGID/task/ directories */
3022 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3023 {
3024         struct dentry *dentry = filp->f_path.dentry;
3025         struct inode *inode = dentry->d_inode;
3026         struct task_struct *leader = NULL;
3027         struct task_struct *task;
3028         int retval = -ENOENT;
3029         ino_t ino;
3030         int tid;
3031         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
3032         struct pid_namespace *ns;
3033
3034         task = get_proc_task(inode);
3035         if (!task)
3036                 goto out_no_task;
3037         rcu_read_lock();
3038         if (pid_alive(task)) {
3039                 leader = task->group_leader;
3040                 get_task_struct(leader);
3041         }
3042         rcu_read_unlock();
3043         put_task_struct(task);
3044         if (!leader)
3045                 goto out_no_task;
3046         retval = 0;
3047
3048         switch (pos) {
3049         case 0:
3050                 ino = inode->i_ino;
3051                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
3052                         goto out;
3053                 pos++;
3054                 /* fall through */
3055         case 1:
3056                 ino = parent_ino(dentry);
3057                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
3058                         goto out;
3059                 pos++;
3060                 /* fall through */
3061         }
3062
3063         /* f_version caches the tgid value that the last readdir call couldn't
3064          * return. lseek aka telldir automagically resets f_version to 0.
3065          */
3066         ns = filp->f_dentry->d_sb->s_fs_info;
3067         tid = (int)filp->f_version;
3068         filp->f_version = 0;
3069         for (task = first_tid(leader, tid, pos - 2, ns);
3070              task;
3071              task = next_tid(task), pos++) {
3072                 tid = task_pid_nr_ns(task, ns);
3073                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3074                         /* returning this tgid failed, save it as the first
3075                          * pid for the next readir call */
3076                         filp->f_version = (u64)tid;
3077                         put_task_struct(task);
3078                         break;
3079                 }
3080         }
3081 out:
3082         filp->f_pos = pos;
3083         put_task_struct(leader);
3084 out_no_task:
3085         return retval;
3086 }
3087
3088 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3089 {
3090         struct inode *inode = dentry->d_inode;
3091         struct task_struct *p = get_proc_task(inode);
3092         generic_fillattr(inode, stat);
3093
3094         if (p) {
3095                 rcu_read_lock();
3096                 stat->nlink += get_nr_threads(p);
3097                 rcu_read_unlock();
3098                 put_task_struct(p);
3099         }
3100
3101         return 0;
3102 }
3103
3104 static const struct inode_operations proc_task_inode_operations = {
3105         .lookup         = proc_task_lookup,
3106         .getattr        = proc_task_getattr,
3107         .setattr        = proc_setattr,
3108 };
3109
3110 static const struct file_operations proc_task_operations = {
3111         .read           = generic_read_dir,
3112         .readdir        = proc_task_readdir,
3113 };