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