fix setuid sometimes wouldn't
[linux-2.6.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/rcupdate.h>
67 #include <linux/kallsyms.h>
68 #include <linux/stacktrace.h>
69 #include <linux/resource.h>
70 #include <linux/module.h>
71 #include <linux/mount.h>
72 #include <linux/security.h>
73 #include <linux/ptrace.h>
74 #include <linux/tracehook.h>
75 #include <linux/cgroup.h>
76 #include <linux/cpuset.h>
77 #include <linux/audit.h>
78 #include <linux/poll.h>
79 #include <linux/nsproxy.h>
80 #include <linux/oom.h>
81 #include <linux/elf.h>
82 #include <linux/pid_namespace.h>
83 #include "internal.h"
84
85 /* NOTE:
86  *      Implementing inode permission operations in /proc is almost
87  *      certainly an error.  Permission checks need to happen during
88  *      each system call not at open time.  The reason is that most of
89  *      what we wish to check for permissions in /proc varies at runtime.
90  *
91  *      The classic example of a problem is opening file descriptors
92  *      in /proc for a task before it execs a suid executable.
93  */
94
95 struct pid_entry {
96         char *name;
97         int len;
98         mode_t mode;
99         const struct inode_operations *iop;
100         const struct file_operations *fop;
101         union proc_op op;
102 };
103
104 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
105         .name = (NAME),                                 \
106         .len  = sizeof(NAME) - 1,                       \
107         .mode = MODE,                                   \
108         .iop  = IOP,                                    \
109         .fop  = FOP,                                    \
110         .op   = OP,                                     \
111 }
112
113 #define DIR(NAME, MODE, iops, fops)     \
114         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
115 #define LNK(NAME, get_link)                                     \
116         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
117                 &proc_pid_link_inode_operations, NULL,          \
118                 { .proc_get_link = get_link } )
119 #define REG(NAME, MODE, fops)                           \
120         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
121 #define INF(NAME, MODE, read)                           \
122         NOD(NAME, (S_IFREG|(MODE)),                     \
123                 NULL, &proc_info_file_operations,       \
124                 { .proc_read = read } )
125 #define ONE(NAME, MODE, show)                           \
126         NOD(NAME, (S_IFREG|(MODE)),                     \
127                 NULL, &proc_single_file_operations,     \
128                 { .proc_show = show } )
129
130 /*
131  * Count the number of hardlinks for the pid_entry table, excluding the .
132  * and .. links.
133  */
134 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
135         unsigned int n)
136 {
137         unsigned int i;
138         unsigned int count;
139
140         count = 0;
141         for (i = 0; i < n; ++i) {
142                 if (S_ISDIR(entries[i].mode))
143                         ++count;
144         }
145
146         return count;
147 }
148
149 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
150 {
151         struct fs_struct *fs;
152         int result = -ENOENT;
153
154         task_lock(task);
155         fs = task->fs;
156         if (fs) {
157                 read_lock(&fs->lock);
158                 *path = root ? fs->root : fs->pwd;
159                 path_get(path);
160                 read_unlock(&fs->lock);
161                 result = 0;
162         }
163         task_unlock(task);
164         return result;
165 }
166
167 static int get_nr_threads(struct task_struct *tsk)
168 {
169         unsigned long flags;
170         int count = 0;
171
172         if (lock_task_sighand(tsk, &flags)) {
173                 count = atomic_read(&tsk->signal->count);
174                 unlock_task_sighand(tsk, &flags);
175         }
176         return count;
177 }
178
179 static int proc_cwd_link(struct inode *inode, struct path *path)
180 {
181         struct task_struct *task = get_proc_task(inode);
182         int result = -ENOENT;
183
184         if (task) {
185                 result = get_fs_path(task, path, 0);
186                 put_task_struct(task);
187         }
188         return result;
189 }
190
191 static int proc_root_link(struct inode *inode, struct path *path)
192 {
193         struct task_struct *task = get_proc_task(inode);
194         int result = -ENOENT;
195
196         if (task) {
197                 result = get_fs_path(task, path, 1);
198                 put_task_struct(task);
199         }
200         return result;
201 }
202
203 /*
204  * Return zero if current may access user memory in @task, -error if not.
205  */
206 static int check_mem_permission(struct task_struct *task)
207 {
208         /*
209          * A task can always look at itself, in case it chooses
210          * to use system calls instead of load instructions.
211          */
212         if (task == current)
213                 return 0;
214
215         /*
216          * If current is actively ptrace'ing, and would also be
217          * permitted to freshly attach with ptrace now, permit it.
218          */
219         if (task_is_stopped_or_traced(task)) {
220                 int match;
221                 rcu_read_lock();
222                 match = (tracehook_tracer_task(task) == current);
223                 rcu_read_unlock();
224                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
225                         return 0;
226         }
227
228         /*
229          * Noone else is allowed.
230          */
231         return -EPERM;
232 }
233
234 struct mm_struct *mm_for_maps(struct task_struct *task)
235 {
236         struct mm_struct *mm = get_task_mm(task);
237         if (!mm)
238                 return NULL;
239         down_read(&mm->mmap_sem);
240         task_lock(task);
241         if (task->mm != mm)
242                 goto out;
243         if (task->mm != current->mm &&
244             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
245                 goto out;
246         task_unlock(task);
247         return mm;
248 out:
249         task_unlock(task);
250         up_read(&mm->mmap_sem);
251         mmput(mm);
252         return NULL;
253 }
254
255 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
256 {
257         int res = 0;
258         unsigned int len;
259         struct mm_struct *mm = get_task_mm(task);
260         if (!mm)
261                 goto out;
262         if (!mm->arg_end)
263                 goto out_mm;    /* Shh! No looking before we're done */
264
265         len = mm->arg_end - mm->arg_start;
266  
267         if (len > PAGE_SIZE)
268                 len = PAGE_SIZE;
269  
270         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
271
272         // If the nul at the end of args has been overwritten, then
273         // assume application is using setproctitle(3).
274         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
275                 len = strnlen(buffer, res);
276                 if (len < res) {
277                     res = len;
278                 } else {
279                         len = mm->env_end - mm->env_start;
280                         if (len > PAGE_SIZE - res)
281                                 len = PAGE_SIZE - res;
282                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
283                         res = strnlen(buffer, res);
284                 }
285         }
286 out_mm:
287         mmput(mm);
288 out:
289         return res;
290 }
291
292 static int proc_pid_auxv(struct task_struct *task, char *buffer)
293 {
294         int res = 0;
295         struct mm_struct *mm = get_task_mm(task);
296         if (mm) {
297                 unsigned int nwords = 0;
298                 do {
299                         nwords += 2;
300                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
301                 res = nwords * sizeof(mm->saved_auxv[0]);
302                 if (res > PAGE_SIZE)
303                         res = PAGE_SIZE;
304                 memcpy(buffer, mm->saved_auxv, res);
305                 mmput(mm);
306         }
307         return res;
308 }
309
310
311 #ifdef CONFIG_KALLSYMS
312 /*
313  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
314  * Returns the resolved symbol.  If that fails, simply return the address.
315  */
316 static int proc_pid_wchan(struct task_struct *task, char *buffer)
317 {
318         unsigned long wchan;
319         char symname[KSYM_NAME_LEN];
320
321         wchan = get_wchan(task);
322
323         if (lookup_symbol_name(wchan, symname) < 0)
324                 return sprintf(buffer, "%lu", wchan);
325         else
326                 return sprintf(buffer, "%s", symname);
327 }
328 #endif /* CONFIG_KALLSYMS */
329
330 #ifdef CONFIG_STACKTRACE
331
332 #define MAX_STACK_TRACE_DEPTH   64
333
334 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
335                           struct pid *pid, struct task_struct *task)
336 {
337         struct stack_trace trace;
338         unsigned long *entries;
339         int i;
340
341         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
342         if (!entries)
343                 return -ENOMEM;
344
345         trace.nr_entries        = 0;
346         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
347         trace.entries           = entries;
348         trace.skip              = 0;
349         save_stack_trace_tsk(task, &trace);
350
351         for (i = 0; i < trace.nr_entries; i++) {
352                 seq_printf(m, "[<%p>] %pS\n",
353                            (void *)entries[i], (void *)entries[i]);
354         }
355         kfree(entries);
356
357         return 0;
358 }
359 #endif
360
361 #ifdef CONFIG_SCHEDSTATS
362 /*
363  * Provides /proc/PID/schedstat
364  */
365 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
366 {
367         return sprintf(buffer, "%llu %llu %lu\n",
368                         (unsigned long long)task->se.sum_exec_runtime,
369                         (unsigned long long)task->sched_info.run_delay,
370                         task->sched_info.pcount);
371 }
372 #endif
373
374 #ifdef CONFIG_LATENCYTOP
375 static int lstats_show_proc(struct seq_file *m, void *v)
376 {
377         int i;
378         struct inode *inode = m->private;
379         struct task_struct *task = get_proc_task(inode);
380
381         if (!task)
382                 return -ESRCH;
383         seq_puts(m, "Latency Top version : v0.1\n");
384         for (i = 0; i < 32; i++) {
385                 if (task->latency_record[i].backtrace[0]) {
386                         int q;
387                         seq_printf(m, "%i %li %li ",
388                                 task->latency_record[i].count,
389                                 task->latency_record[i].time,
390                                 task->latency_record[i].max);
391                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
392                                 char sym[KSYM_SYMBOL_LEN];
393                                 char *c;
394                                 if (!task->latency_record[i].backtrace[q])
395                                         break;
396                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
397                                         break;
398                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
399                                 c = strchr(sym, '+');
400                                 if (c)
401                                         *c = 0;
402                                 seq_printf(m, "%s ", sym);
403                         }
404                         seq_printf(m, "\n");
405                 }
406
407         }
408         put_task_struct(task);
409         return 0;
410 }
411
412 static int lstats_open(struct inode *inode, struct file *file)
413 {
414         return single_open(file, lstats_show_proc, inode);
415 }
416
417 static ssize_t lstats_write(struct file *file, const char __user *buf,
418                             size_t count, loff_t *offs)
419 {
420         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
421
422         if (!task)
423                 return -ESRCH;
424         clear_all_latency_tracing(task);
425         put_task_struct(task);
426
427         return count;
428 }
429
430 static const struct file_operations proc_lstats_operations = {
431         .open           = lstats_open,
432         .read           = seq_read,
433         .write          = lstats_write,
434         .llseek         = seq_lseek,
435         .release        = single_release,
436 };
437
438 #endif
439
440 /* The badness from the OOM killer */
441 unsigned long badness(struct task_struct *p, unsigned long uptime);
442 static int proc_oom_score(struct task_struct *task, char *buffer)
443 {
444         unsigned long points;
445         struct timespec uptime;
446
447         do_posix_clock_monotonic_gettime(&uptime);
448         read_lock(&tasklist_lock);
449         points = badness(task, uptime.tv_sec);
450         read_unlock(&tasklist_lock);
451         return sprintf(buffer, "%lu\n", points);
452 }
453
454 struct limit_names {
455         char *name;
456         char *unit;
457 };
458
459 static const struct limit_names lnames[RLIM_NLIMITS] = {
460         [RLIMIT_CPU] = {"Max cpu time", "ms"},
461         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
462         [RLIMIT_DATA] = {"Max data size", "bytes"},
463         [RLIMIT_STACK] = {"Max stack size", "bytes"},
464         [RLIMIT_CORE] = {"Max core file size", "bytes"},
465         [RLIMIT_RSS] = {"Max resident set", "bytes"},
466         [RLIMIT_NPROC] = {"Max processes", "processes"},
467         [RLIMIT_NOFILE] = {"Max open files", "files"},
468         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
469         [RLIMIT_AS] = {"Max address space", "bytes"},
470         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
471         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
472         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
473         [RLIMIT_NICE] = {"Max nice priority", NULL},
474         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
475         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
476 };
477
478 /* Display limits for a process */
479 static int proc_pid_limits(struct task_struct *task, char *buffer)
480 {
481         unsigned int i;
482         int count = 0;
483         unsigned long flags;
484         char *bufptr = buffer;
485
486         struct rlimit rlim[RLIM_NLIMITS];
487
488         if (!lock_task_sighand(task, &flags))
489                 return 0;
490         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
491         unlock_task_sighand(task, &flags);
492
493         /*
494          * print the file header
495          */
496         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
497                         "Limit", "Soft Limit", "Hard Limit", "Units");
498
499         for (i = 0; i < RLIM_NLIMITS; i++) {
500                 if (rlim[i].rlim_cur == RLIM_INFINITY)
501                         count += sprintf(&bufptr[count], "%-25s %-20s ",
502                                          lnames[i].name, "unlimited");
503                 else
504                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
505                                          lnames[i].name, rlim[i].rlim_cur);
506
507                 if (rlim[i].rlim_max == RLIM_INFINITY)
508                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
509                 else
510                         count += sprintf(&bufptr[count], "%-20lu ",
511                                          rlim[i].rlim_max);
512
513                 if (lnames[i].unit)
514                         count += sprintf(&bufptr[count], "%-10s\n",
515                                          lnames[i].unit);
516                 else
517                         count += sprintf(&bufptr[count], "\n");
518         }
519
520         return count;
521 }
522
523 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
524 static int proc_pid_syscall(struct task_struct *task, char *buffer)
525 {
526         long nr;
527         unsigned long args[6], sp, pc;
528
529         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
530                 return sprintf(buffer, "running\n");
531
532         if (nr < 0)
533                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
534
535         return sprintf(buffer,
536                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
537                        nr,
538                        args[0], args[1], args[2], args[3], args[4], args[5],
539                        sp, pc);
540 }
541 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
542
543 /************************************************************************/
544 /*                       Here the fs part begins                        */
545 /************************************************************************/
546
547 /* permission checks */
548 static int proc_fd_access_allowed(struct inode *inode)
549 {
550         struct task_struct *task;
551         int allowed = 0;
552         /* Allow access to a task's file descriptors if it is us or we
553          * may use ptrace attach to the process and find out that
554          * information.
555          */
556         task = get_proc_task(inode);
557         if (task) {
558                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
559                 put_task_struct(task);
560         }
561         return allowed;
562 }
563
564 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
565 {
566         int error;
567         struct inode *inode = dentry->d_inode;
568
569         if (attr->ia_valid & ATTR_MODE)
570                 return -EPERM;
571
572         error = inode_change_ok(inode, attr);
573         if (!error)
574                 error = inode_setattr(inode, attr);
575         return error;
576 }
577
578 static const struct inode_operations proc_def_inode_operations = {
579         .setattr        = proc_setattr,
580 };
581
582 static int mounts_open_common(struct inode *inode, struct file *file,
583                               const struct seq_operations *op)
584 {
585         struct task_struct *task = get_proc_task(inode);
586         struct nsproxy *nsp;
587         struct mnt_namespace *ns = NULL;
588         struct path root;
589         struct proc_mounts *p;
590         int ret = -EINVAL;
591
592         if (task) {
593                 rcu_read_lock();
594                 nsp = task_nsproxy(task);
595                 if (nsp) {
596                         ns = nsp->mnt_ns;
597                         if (ns)
598                                 get_mnt_ns(ns);
599                 }
600                 rcu_read_unlock();
601                 if (ns && get_fs_path(task, &root, 1) == 0)
602                         ret = 0;
603                 put_task_struct(task);
604         }
605
606         if (!ns)
607                 goto err;
608         if (ret)
609                 goto err_put_ns;
610
611         ret = -ENOMEM;
612         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
613         if (!p)
614                 goto err_put_path;
615
616         file->private_data = &p->m;
617         ret = seq_open(file, op);
618         if (ret)
619                 goto err_free;
620
621         p->m.private = p;
622         p->ns = ns;
623         p->root = root;
624         p->event = ns->event;
625
626         return 0;
627
628  err_free:
629         kfree(p);
630  err_put_path:
631         path_put(&root);
632  err_put_ns:
633         put_mnt_ns(ns);
634  err:
635         return ret;
636 }
637
638 static int mounts_release(struct inode *inode, struct file *file)
639 {
640         struct proc_mounts *p = file->private_data;
641         path_put(&p->root);
642         put_mnt_ns(p->ns);
643         return seq_release(inode, file);
644 }
645
646 static unsigned mounts_poll(struct file *file, poll_table *wait)
647 {
648         struct proc_mounts *p = file->private_data;
649         struct mnt_namespace *ns = p->ns;
650         unsigned res = 0;
651
652         poll_wait(file, &ns->poll, wait);
653
654         spin_lock(&vfsmount_lock);
655         if (p->event != ns->event) {
656                 p->event = ns->event;
657                 res = POLLERR;
658         }
659         spin_unlock(&vfsmount_lock);
660
661         return res;
662 }
663
664 static int mounts_open(struct inode *inode, struct file *file)
665 {
666         return mounts_open_common(inode, file, &mounts_op);
667 }
668
669 static const struct file_operations proc_mounts_operations = {
670         .open           = mounts_open,
671         .read           = seq_read,
672         .llseek         = seq_lseek,
673         .release        = mounts_release,
674         .poll           = mounts_poll,
675 };
676
677 static int mountinfo_open(struct inode *inode, struct file *file)
678 {
679         return mounts_open_common(inode, file, &mountinfo_op);
680 }
681
682 static const struct file_operations proc_mountinfo_operations = {
683         .open           = mountinfo_open,
684         .read           = seq_read,
685         .llseek         = seq_lseek,
686         .release        = mounts_release,
687         .poll           = mounts_poll,
688 };
689
690 static int mountstats_open(struct inode *inode, struct file *file)
691 {
692         return mounts_open_common(inode, file, &mountstats_op);
693 }
694
695 static const struct file_operations proc_mountstats_operations = {
696         .open           = mountstats_open,
697         .read           = seq_read,
698         .llseek         = seq_lseek,
699         .release        = mounts_release,
700 };
701
702 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
703
704 static ssize_t proc_info_read(struct file * file, char __user * buf,
705                           size_t count, loff_t *ppos)
706 {
707         struct inode * inode = file->f_path.dentry->d_inode;
708         unsigned long page;
709         ssize_t length;
710         struct task_struct *task = get_proc_task(inode);
711
712         length = -ESRCH;
713         if (!task)
714                 goto out_no_task;
715
716         if (count > PROC_BLOCK_SIZE)
717                 count = PROC_BLOCK_SIZE;
718
719         length = -ENOMEM;
720         if (!(page = __get_free_page(GFP_TEMPORARY)))
721                 goto out;
722
723         length = PROC_I(inode)->op.proc_read(task, (char*)page);
724
725         if (length >= 0)
726                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
727         free_page(page);
728 out:
729         put_task_struct(task);
730 out_no_task:
731         return length;
732 }
733
734 static const struct file_operations proc_info_file_operations = {
735         .read           = proc_info_read,
736 };
737
738 static int proc_single_show(struct seq_file *m, void *v)
739 {
740         struct inode *inode = m->private;
741         struct pid_namespace *ns;
742         struct pid *pid;
743         struct task_struct *task;
744         int ret;
745
746         ns = inode->i_sb->s_fs_info;
747         pid = proc_pid(inode);
748         task = get_pid_task(pid, PIDTYPE_PID);
749         if (!task)
750                 return -ESRCH;
751
752         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
753
754         put_task_struct(task);
755         return ret;
756 }
757
758 static int proc_single_open(struct inode *inode, struct file *filp)
759 {
760         int ret;
761         ret = single_open(filp, proc_single_show, NULL);
762         if (!ret) {
763                 struct seq_file *m = filp->private_data;
764
765                 m->private = inode;
766         }
767         return ret;
768 }
769
770 static const struct file_operations proc_single_file_operations = {
771         .open           = proc_single_open,
772         .read           = seq_read,
773         .llseek         = seq_lseek,
774         .release        = single_release,
775 };
776
777 static int mem_open(struct inode* inode, struct file* file)
778 {
779         file->private_data = (void*)((long)current->self_exec_id);
780         return 0;
781 }
782
783 static ssize_t mem_read(struct file * file, char __user * buf,
784                         size_t count, loff_t *ppos)
785 {
786         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
787         char *page;
788         unsigned long src = *ppos;
789         int ret = -ESRCH;
790         struct mm_struct *mm;
791
792         if (!task)
793                 goto out_no_task;
794
795         if (check_mem_permission(task))
796                 goto out;
797
798         ret = -ENOMEM;
799         page = (char *)__get_free_page(GFP_TEMPORARY);
800         if (!page)
801                 goto out;
802
803         ret = 0;
804  
805         mm = get_task_mm(task);
806         if (!mm)
807                 goto out_free;
808
809         ret = -EIO;
810  
811         if (file->private_data != (void*)((long)current->self_exec_id))
812                 goto out_put;
813
814         ret = 0;
815  
816         while (count > 0) {
817                 int this_len, retval;
818
819                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
820                 retval = access_process_vm(task, src, page, this_len, 0);
821                 if (!retval || check_mem_permission(task)) {
822                         if (!ret)
823                                 ret = -EIO;
824                         break;
825                 }
826
827                 if (copy_to_user(buf, page, retval)) {
828                         ret = -EFAULT;
829                         break;
830                 }
831  
832                 ret += retval;
833                 src += retval;
834                 buf += retval;
835                 count -= retval;
836         }
837         *ppos = src;
838
839 out_put:
840         mmput(mm);
841 out_free:
842         free_page((unsigned long) page);
843 out:
844         put_task_struct(task);
845 out_no_task:
846         return ret;
847 }
848
849 #define mem_write NULL
850
851 #ifndef mem_write
852 /* This is a security hazard */
853 static ssize_t mem_write(struct file * file, const char __user *buf,
854                          size_t count, loff_t *ppos)
855 {
856         int copied;
857         char *page;
858         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
859         unsigned long dst = *ppos;
860
861         copied = -ESRCH;
862         if (!task)
863                 goto out_no_task;
864
865         if (check_mem_permission(task))
866                 goto out;
867
868         copied = -ENOMEM;
869         page = (char *)__get_free_page(GFP_TEMPORARY);
870         if (!page)
871                 goto out;
872
873         copied = 0;
874         while (count > 0) {
875                 int this_len, retval;
876
877                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
878                 if (copy_from_user(page, buf, this_len)) {
879                         copied = -EFAULT;
880                         break;
881                 }
882                 retval = access_process_vm(task, dst, page, this_len, 1);
883                 if (!retval) {
884                         if (!copied)
885                                 copied = -EIO;
886                         break;
887                 }
888                 copied += retval;
889                 buf += retval;
890                 dst += retval;
891                 count -= retval;                        
892         }
893         *ppos = dst;
894         free_page((unsigned long) page);
895 out:
896         put_task_struct(task);
897 out_no_task:
898         return copied;
899 }
900 #endif
901
902 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
903 {
904         switch (orig) {
905         case 0:
906                 file->f_pos = offset;
907                 break;
908         case 1:
909                 file->f_pos += offset;
910                 break;
911         default:
912                 return -EINVAL;
913         }
914         force_successful_syscall_return();
915         return file->f_pos;
916 }
917
918 static const struct file_operations proc_mem_operations = {
919         .llseek         = mem_lseek,
920         .read           = mem_read,
921         .write          = mem_write,
922         .open           = mem_open,
923 };
924
925 static ssize_t environ_read(struct file *file, char __user *buf,
926                         size_t count, loff_t *ppos)
927 {
928         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
929         char *page;
930         unsigned long src = *ppos;
931         int ret = -ESRCH;
932         struct mm_struct *mm;
933
934         if (!task)
935                 goto out_no_task;
936
937         if (!ptrace_may_access(task, PTRACE_MODE_READ))
938                 goto out;
939
940         ret = -ENOMEM;
941         page = (char *)__get_free_page(GFP_TEMPORARY);
942         if (!page)
943                 goto out;
944
945         ret = 0;
946
947         mm = get_task_mm(task);
948         if (!mm)
949                 goto out_free;
950
951         while (count > 0) {
952                 int this_len, retval, max_len;
953
954                 this_len = mm->env_end - (mm->env_start + src);
955
956                 if (this_len <= 0)
957                         break;
958
959                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
960                 this_len = (this_len > max_len) ? max_len : this_len;
961
962                 retval = access_process_vm(task, (mm->env_start + src),
963                         page, this_len, 0);
964
965                 if (retval <= 0) {
966                         ret = retval;
967                         break;
968                 }
969
970                 if (copy_to_user(buf, page, retval)) {
971                         ret = -EFAULT;
972                         break;
973                 }
974
975                 ret += retval;
976                 src += retval;
977                 buf += retval;
978                 count -= retval;
979         }
980         *ppos = src;
981
982         mmput(mm);
983 out_free:
984         free_page((unsigned long) page);
985 out:
986         put_task_struct(task);
987 out_no_task:
988         return ret;
989 }
990
991 static const struct file_operations proc_environ_operations = {
992         .read           = environ_read,
993 };
994
995 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
996                                 size_t count, loff_t *ppos)
997 {
998         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
999         char buffer[PROC_NUMBUF];
1000         size_t len;
1001         int oom_adjust;
1002
1003         if (!task)
1004                 return -ESRCH;
1005         oom_adjust = task->oomkilladj;
1006         put_task_struct(task);
1007
1008         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1009
1010         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1011 }
1012
1013 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1014                                 size_t count, loff_t *ppos)
1015 {
1016         struct task_struct *task;
1017         char buffer[PROC_NUMBUF], *end;
1018         int oom_adjust;
1019
1020         memset(buffer, 0, sizeof(buffer));
1021         if (count > sizeof(buffer) - 1)
1022                 count = sizeof(buffer) - 1;
1023         if (copy_from_user(buffer, buf, count))
1024                 return -EFAULT;
1025         oom_adjust = simple_strtol(buffer, &end, 0);
1026         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1027              oom_adjust != OOM_DISABLE)
1028                 return -EINVAL;
1029         if (*end == '\n')
1030                 end++;
1031         task = get_proc_task(file->f_path.dentry->d_inode);
1032         if (!task)
1033                 return -ESRCH;
1034         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1035                 put_task_struct(task);
1036                 return -EACCES;
1037         }
1038         task->oomkilladj = oom_adjust;
1039         put_task_struct(task);
1040         if (end - buffer == 0)
1041                 return -EIO;
1042         return end - buffer;
1043 }
1044
1045 static const struct file_operations proc_oom_adjust_operations = {
1046         .read           = oom_adjust_read,
1047         .write          = oom_adjust_write,
1048 };
1049
1050 #ifdef CONFIG_AUDITSYSCALL
1051 #define TMPBUFLEN 21
1052 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1053                                   size_t count, loff_t *ppos)
1054 {
1055         struct inode * inode = file->f_path.dentry->d_inode;
1056         struct task_struct *task = get_proc_task(inode);
1057         ssize_t length;
1058         char tmpbuf[TMPBUFLEN];
1059
1060         if (!task)
1061                 return -ESRCH;
1062         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1063                                 audit_get_loginuid(task));
1064         put_task_struct(task);
1065         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1066 }
1067
1068 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1069                                    size_t count, loff_t *ppos)
1070 {
1071         struct inode * inode = file->f_path.dentry->d_inode;
1072         char *page, *tmp;
1073         ssize_t length;
1074         uid_t loginuid;
1075
1076         if (!capable(CAP_AUDIT_CONTROL))
1077                 return -EPERM;
1078
1079         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1080                 return -EPERM;
1081
1082         if (count >= PAGE_SIZE)
1083                 count = PAGE_SIZE - 1;
1084
1085         if (*ppos != 0) {
1086                 /* No partial writes. */
1087                 return -EINVAL;
1088         }
1089         page = (char*)__get_free_page(GFP_TEMPORARY);
1090         if (!page)
1091                 return -ENOMEM;
1092         length = -EFAULT;
1093         if (copy_from_user(page, buf, count))
1094                 goto out_free_page;
1095
1096         page[count] = '\0';
1097         loginuid = simple_strtoul(page, &tmp, 10);
1098         if (tmp == page) {
1099                 length = -EINVAL;
1100                 goto out_free_page;
1101
1102         }
1103         length = audit_set_loginuid(current, loginuid);
1104         if (likely(length == 0))
1105                 length = count;
1106
1107 out_free_page:
1108         free_page((unsigned long) page);
1109         return length;
1110 }
1111
1112 static const struct file_operations proc_loginuid_operations = {
1113         .read           = proc_loginuid_read,
1114         .write          = proc_loginuid_write,
1115 };
1116
1117 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1118                                   size_t count, loff_t *ppos)
1119 {
1120         struct inode * inode = file->f_path.dentry->d_inode;
1121         struct task_struct *task = get_proc_task(inode);
1122         ssize_t length;
1123         char tmpbuf[TMPBUFLEN];
1124
1125         if (!task)
1126                 return -ESRCH;
1127         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1128                                 audit_get_sessionid(task));
1129         put_task_struct(task);
1130         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1131 }
1132
1133 static const struct file_operations proc_sessionid_operations = {
1134         .read           = proc_sessionid_read,
1135 };
1136 #endif
1137
1138 #ifdef CONFIG_FAULT_INJECTION
1139 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1140                                       size_t count, loff_t *ppos)
1141 {
1142         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1143         char buffer[PROC_NUMBUF];
1144         size_t len;
1145         int make_it_fail;
1146
1147         if (!task)
1148                 return -ESRCH;
1149         make_it_fail = task->make_it_fail;
1150         put_task_struct(task);
1151
1152         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1153
1154         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1155 }
1156
1157 static ssize_t proc_fault_inject_write(struct file * file,
1158                         const char __user * buf, size_t count, loff_t *ppos)
1159 {
1160         struct task_struct *task;
1161         char buffer[PROC_NUMBUF], *end;
1162         int make_it_fail;
1163
1164         if (!capable(CAP_SYS_RESOURCE))
1165                 return -EPERM;
1166         memset(buffer, 0, sizeof(buffer));
1167         if (count > sizeof(buffer) - 1)
1168                 count = sizeof(buffer) - 1;
1169         if (copy_from_user(buffer, buf, count))
1170                 return -EFAULT;
1171         make_it_fail = simple_strtol(buffer, &end, 0);
1172         if (*end == '\n')
1173                 end++;
1174         task = get_proc_task(file->f_dentry->d_inode);
1175         if (!task)
1176                 return -ESRCH;
1177         task->make_it_fail = make_it_fail;
1178         put_task_struct(task);
1179         if (end - buffer == 0)
1180                 return -EIO;
1181         return end - buffer;
1182 }
1183
1184 static const struct file_operations proc_fault_inject_operations = {
1185         .read           = proc_fault_inject_read,
1186         .write          = proc_fault_inject_write,
1187 };
1188 #endif
1189
1190
1191 #ifdef CONFIG_SCHED_DEBUG
1192 /*
1193  * Print out various scheduling related per-task fields:
1194  */
1195 static int sched_show(struct seq_file *m, void *v)
1196 {
1197         struct inode *inode = m->private;
1198         struct task_struct *p;
1199
1200         p = get_proc_task(inode);
1201         if (!p)
1202                 return -ESRCH;
1203         proc_sched_show_task(p, m);
1204
1205         put_task_struct(p);
1206
1207         return 0;
1208 }
1209
1210 static ssize_t
1211 sched_write(struct file *file, const char __user *buf,
1212             size_t count, loff_t *offset)
1213 {
1214         struct inode *inode = file->f_path.dentry->d_inode;
1215         struct task_struct *p;
1216
1217         p = get_proc_task(inode);
1218         if (!p)
1219                 return -ESRCH;
1220         proc_sched_set_task(p);
1221
1222         put_task_struct(p);
1223
1224         return count;
1225 }
1226
1227 static int sched_open(struct inode *inode, struct file *filp)
1228 {
1229         int ret;
1230
1231         ret = single_open(filp, sched_show, NULL);
1232         if (!ret) {
1233                 struct seq_file *m = filp->private_data;
1234
1235                 m->private = inode;
1236         }
1237         return ret;
1238 }
1239
1240 static const struct file_operations proc_pid_sched_operations = {
1241         .open           = sched_open,
1242         .read           = seq_read,
1243         .write          = sched_write,
1244         .llseek         = seq_lseek,
1245         .release        = single_release,
1246 };
1247
1248 #endif
1249
1250 /*
1251  * We added or removed a vma mapping the executable. The vmas are only mapped
1252  * during exec and are not mapped with the mmap system call.
1253  * Callers must hold down_write() on the mm's mmap_sem for these
1254  */
1255 void added_exe_file_vma(struct mm_struct *mm)
1256 {
1257         mm->num_exe_file_vmas++;
1258 }
1259
1260 void removed_exe_file_vma(struct mm_struct *mm)
1261 {
1262         mm->num_exe_file_vmas--;
1263         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1264                 fput(mm->exe_file);
1265                 mm->exe_file = NULL;
1266         }
1267
1268 }
1269
1270 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1271 {
1272         if (new_exe_file)
1273                 get_file(new_exe_file);
1274         if (mm->exe_file)
1275                 fput(mm->exe_file);
1276         mm->exe_file = new_exe_file;
1277         mm->num_exe_file_vmas = 0;
1278 }
1279
1280 struct file *get_mm_exe_file(struct mm_struct *mm)
1281 {
1282         struct file *exe_file;
1283
1284         /* We need mmap_sem to protect against races with removal of
1285          * VM_EXECUTABLE vmas */
1286         down_read(&mm->mmap_sem);
1287         exe_file = mm->exe_file;
1288         if (exe_file)
1289                 get_file(exe_file);
1290         up_read(&mm->mmap_sem);
1291         return exe_file;
1292 }
1293
1294 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1295 {
1296         /* It's safe to write the exe_file pointer without exe_file_lock because
1297          * this is called during fork when the task is not yet in /proc */
1298         newmm->exe_file = get_mm_exe_file(oldmm);
1299 }
1300
1301 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1302 {
1303         struct task_struct *task;
1304         struct mm_struct *mm;
1305         struct file *exe_file;
1306
1307         task = get_proc_task(inode);
1308         if (!task)
1309                 return -ENOENT;
1310         mm = get_task_mm(task);
1311         put_task_struct(task);
1312         if (!mm)
1313                 return -ENOENT;
1314         exe_file = get_mm_exe_file(mm);
1315         mmput(mm);
1316         if (exe_file) {
1317                 *exe_path = exe_file->f_path;
1318                 path_get(&exe_file->f_path);
1319                 fput(exe_file);
1320                 return 0;
1321         } else
1322                 return -ENOENT;
1323 }
1324
1325 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1326 {
1327         struct inode *inode = dentry->d_inode;
1328         int error = -EACCES;
1329
1330         /* We don't need a base pointer in the /proc filesystem */
1331         path_put(&nd->path);
1332
1333         /* Are we allowed to snoop on the tasks file descriptors? */
1334         if (!proc_fd_access_allowed(inode))
1335                 goto out;
1336
1337         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1338         nd->last_type = LAST_BIND;
1339 out:
1340         return ERR_PTR(error);
1341 }
1342
1343 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1344 {
1345         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1346         char *pathname;
1347         int len;
1348
1349         if (!tmp)
1350                 return -ENOMEM;
1351
1352         pathname = d_path(path, tmp, PAGE_SIZE);
1353         len = PTR_ERR(pathname);
1354         if (IS_ERR(pathname))
1355                 goto out;
1356         len = tmp + PAGE_SIZE - 1 - pathname;
1357
1358         if (len > buflen)
1359                 len = buflen;
1360         if (copy_to_user(buffer, pathname, len))
1361                 len = -EFAULT;
1362  out:
1363         free_page((unsigned long)tmp);
1364         return len;
1365 }
1366
1367 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1368 {
1369         int error = -EACCES;
1370         struct inode *inode = dentry->d_inode;
1371         struct path path;
1372
1373         /* Are we allowed to snoop on the tasks file descriptors? */
1374         if (!proc_fd_access_allowed(inode))
1375                 goto out;
1376
1377         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1378         if (error)
1379                 goto out;
1380
1381         error = do_proc_readlink(&path, buffer, buflen);
1382         path_put(&path);
1383 out:
1384         return error;
1385 }
1386
1387 static const struct inode_operations proc_pid_link_inode_operations = {
1388         .readlink       = proc_pid_readlink,
1389         .follow_link    = proc_pid_follow_link,
1390         .setattr        = proc_setattr,
1391 };
1392
1393
1394 /* building an inode */
1395
1396 static int task_dumpable(struct task_struct *task)
1397 {
1398         int dumpable = 0;
1399         struct mm_struct *mm;
1400
1401         task_lock(task);
1402         mm = task->mm;
1403         if (mm)
1404                 dumpable = get_dumpable(mm);
1405         task_unlock(task);
1406         if(dumpable == 1)
1407                 return 1;
1408         return 0;
1409 }
1410
1411
1412 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1413 {
1414         struct inode * inode;
1415         struct proc_inode *ei;
1416         const struct cred *cred;
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         if (task_dumpable(task)) {
1437                 rcu_read_lock();
1438                 cred = __task_cred(task);
1439                 inode->i_uid = cred->euid;
1440                 inode->i_gid = cred->egid;
1441                 rcu_read_unlock();
1442         }
1443         security_task_to_inode(task, inode);
1444
1445 out:
1446         return inode;
1447
1448 out_unlock:
1449         iput(inode);
1450         return NULL;
1451 }
1452
1453 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1454 {
1455         struct inode *inode = dentry->d_inode;
1456         struct task_struct *task;
1457         const struct cred *cred;
1458
1459         generic_fillattr(inode, stat);
1460
1461         rcu_read_lock();
1462         stat->uid = 0;
1463         stat->gid = 0;
1464         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1465         if (task) {
1466                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1467                     task_dumpable(task)) {
1468                         cred = __task_cred(task);
1469                         stat->uid = cred->euid;
1470                         stat->gid = cred->egid;
1471                 }
1472         }
1473         rcu_read_unlock();
1474         return 0;
1475 }
1476
1477 /* dentry stuff */
1478
1479 /*
1480  *      Exceptional case: normally we are not allowed to unhash a busy
1481  * directory. In this case, however, we can do it - no aliasing problems
1482  * due to the way we treat inodes.
1483  *
1484  * Rewrite the inode's ownerships here because the owning task may have
1485  * performed a setuid(), etc.
1486  *
1487  * Before the /proc/pid/status file was created the only way to read
1488  * the effective uid of a /process was to stat /proc/pid.  Reading
1489  * /proc/pid/status is slow enough that procps and other packages
1490  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1491  * made this apply to all per process world readable and executable
1492  * directories.
1493  */
1494 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1495 {
1496         struct inode *inode = dentry->d_inode;
1497         struct task_struct *task = get_proc_task(inode);
1498         const struct cred *cred;
1499
1500         if (task) {
1501                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1502                     task_dumpable(task)) {
1503                         rcu_read_lock();
1504                         cred = __task_cred(task);
1505                         inode->i_uid = cred->euid;
1506                         inode->i_gid = cred->egid;
1507                         rcu_read_unlock();
1508                 } else {
1509                         inode->i_uid = 0;
1510                         inode->i_gid = 0;
1511                 }
1512                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1513                 security_task_to_inode(task, inode);
1514                 put_task_struct(task);
1515                 return 1;
1516         }
1517         d_drop(dentry);
1518         return 0;
1519 }
1520
1521 static int pid_delete_dentry(struct dentry * dentry)
1522 {
1523         /* Is the task we represent dead?
1524          * If so, then don't put the dentry on the lru list,
1525          * kill it immediately.
1526          */
1527         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1528 }
1529
1530 static const struct dentry_operations pid_dentry_operations =
1531 {
1532         .d_revalidate   = pid_revalidate,
1533         .d_delete       = pid_delete_dentry,
1534 };
1535
1536 /* Lookups */
1537
1538 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1539                                 struct task_struct *, const void *);
1540
1541 /*
1542  * Fill a directory entry.
1543  *
1544  * If possible create the dcache entry and derive our inode number and
1545  * file type from dcache entry.
1546  *
1547  * Since all of the proc inode numbers are dynamically generated, the inode
1548  * numbers do not exist until the inode is cache.  This means creating the
1549  * the dcache entry in readdir is necessary to keep the inode numbers
1550  * reported by readdir in sync with the inode numbers reported
1551  * by stat.
1552  */
1553 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1554         char *name, int len,
1555         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1556 {
1557         struct dentry *child, *dir = filp->f_path.dentry;
1558         struct inode *inode;
1559         struct qstr qname;
1560         ino_t ino = 0;
1561         unsigned type = DT_UNKNOWN;
1562
1563         qname.name = name;
1564         qname.len  = len;
1565         qname.hash = full_name_hash(name, len);
1566
1567         child = d_lookup(dir, &qname);
1568         if (!child) {
1569                 struct dentry *new;
1570                 new = d_alloc(dir, &qname);
1571                 if (new) {
1572                         child = instantiate(dir->d_inode, new, task, ptr);
1573                         if (child)
1574                                 dput(new);
1575                         else
1576                                 child = new;
1577                 }
1578         }
1579         if (!child || IS_ERR(child) || !child->d_inode)
1580                 goto end_instantiate;
1581         inode = child->d_inode;
1582         if (inode) {
1583                 ino = inode->i_ino;
1584                 type = inode->i_mode >> 12;
1585         }
1586         dput(child);
1587 end_instantiate:
1588         if (!ino)
1589                 ino = find_inode_number(dir, &qname);
1590         if (!ino)
1591                 ino = 1;
1592         return filldir(dirent, name, len, filp->f_pos, ino, type);
1593 }
1594
1595 static unsigned name_to_int(struct dentry *dentry)
1596 {
1597         const char *name = dentry->d_name.name;
1598         int len = dentry->d_name.len;
1599         unsigned n = 0;
1600
1601         if (len > 1 && *name == '0')
1602                 goto out;
1603         while (len-- > 0) {
1604                 unsigned c = *name++ - '0';
1605                 if (c > 9)
1606                         goto out;
1607                 if (n >= (~0U-9)/10)
1608                         goto out;
1609                 n *= 10;
1610                 n += c;
1611         }
1612         return n;
1613 out:
1614         return ~0U;
1615 }
1616
1617 #define PROC_FDINFO_MAX 64
1618
1619 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1620 {
1621         struct task_struct *task = get_proc_task(inode);
1622         struct files_struct *files = NULL;
1623         struct file *file;
1624         int fd = proc_fd(inode);
1625
1626         if (task) {
1627                 files = get_files_struct(task);
1628                 put_task_struct(task);
1629         }
1630         if (files) {
1631                 /*
1632                  * We are not taking a ref to the file structure, so we must
1633                  * hold ->file_lock.
1634                  */
1635                 spin_lock(&files->file_lock);
1636                 file = fcheck_files(files, fd);
1637                 if (file) {
1638                         if (path) {
1639                                 *path = file->f_path;
1640                                 path_get(&file->f_path);
1641                         }
1642                         if (info)
1643                                 snprintf(info, PROC_FDINFO_MAX,
1644                                          "pos:\t%lli\n"
1645                                          "flags:\t0%o\n",
1646                                          (long long) file->f_pos,
1647                                          file->f_flags);
1648                         spin_unlock(&files->file_lock);
1649                         put_files_struct(files);
1650                         return 0;
1651                 }
1652                 spin_unlock(&files->file_lock);
1653                 put_files_struct(files);
1654         }
1655         return -ENOENT;
1656 }
1657
1658 static int proc_fd_link(struct inode *inode, struct path *path)
1659 {
1660         return proc_fd_info(inode, path, NULL);
1661 }
1662
1663 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1664 {
1665         struct inode *inode = dentry->d_inode;
1666         struct task_struct *task = get_proc_task(inode);
1667         int fd = proc_fd(inode);
1668         struct files_struct *files;
1669         const struct cred *cred;
1670
1671         if (task) {
1672                 files = get_files_struct(task);
1673                 if (files) {
1674                         rcu_read_lock();
1675                         if (fcheck_files(files, fd)) {
1676                                 rcu_read_unlock();
1677                                 put_files_struct(files);
1678                                 if (task_dumpable(task)) {
1679                                         rcu_read_lock();
1680                                         cred = __task_cred(task);
1681                                         inode->i_uid = cred->euid;
1682                                         inode->i_gid = cred->egid;
1683                                         rcu_read_unlock();
1684                                 } else {
1685                                         inode->i_uid = 0;
1686                                         inode->i_gid = 0;
1687                                 }
1688                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1689                                 security_task_to_inode(task, inode);
1690                                 put_task_struct(task);
1691                                 return 1;
1692                         }
1693                         rcu_read_unlock();
1694                         put_files_struct(files);
1695                 }
1696                 put_task_struct(task);
1697         }
1698         d_drop(dentry);
1699         return 0;
1700 }
1701
1702 static const struct dentry_operations tid_fd_dentry_operations =
1703 {
1704         .d_revalidate   = tid_fd_revalidate,
1705         .d_delete       = pid_delete_dentry,
1706 };
1707
1708 static struct dentry *proc_fd_instantiate(struct inode *dir,
1709         struct dentry *dentry, struct task_struct *task, const void *ptr)
1710 {
1711         unsigned fd = *(const unsigned *)ptr;
1712         struct file *file;
1713         struct files_struct *files;
1714         struct inode *inode;
1715         struct proc_inode *ei;
1716         struct dentry *error = ERR_PTR(-ENOENT);
1717
1718         inode = proc_pid_make_inode(dir->i_sb, task);
1719         if (!inode)
1720                 goto out;
1721         ei = PROC_I(inode);
1722         ei->fd = fd;
1723         files = get_files_struct(task);
1724         if (!files)
1725                 goto out_iput;
1726         inode->i_mode = S_IFLNK;
1727
1728         /*
1729          * We are not taking a ref to the file structure, so we must
1730          * hold ->file_lock.
1731          */
1732         spin_lock(&files->file_lock);
1733         file = fcheck_files(files, fd);
1734         if (!file)
1735                 goto out_unlock;
1736         if (file->f_mode & FMODE_READ)
1737                 inode->i_mode |= S_IRUSR | S_IXUSR;
1738         if (file->f_mode & FMODE_WRITE)
1739                 inode->i_mode |= S_IWUSR | S_IXUSR;
1740         spin_unlock(&files->file_lock);
1741         put_files_struct(files);
1742
1743         inode->i_op = &proc_pid_link_inode_operations;
1744         inode->i_size = 64;
1745         ei->op.proc_get_link = proc_fd_link;
1746         dentry->d_op = &tid_fd_dentry_operations;
1747         d_add(dentry, inode);
1748         /* Close the race of the process dying before we return the dentry */
1749         if (tid_fd_revalidate(dentry, NULL))
1750                 error = NULL;
1751
1752  out:
1753         return error;
1754 out_unlock:
1755         spin_unlock(&files->file_lock);
1756         put_files_struct(files);
1757 out_iput:
1758         iput(inode);
1759         goto out;
1760 }
1761
1762 static struct dentry *proc_lookupfd_common(struct inode *dir,
1763                                            struct dentry *dentry,
1764                                            instantiate_t instantiate)
1765 {
1766         struct task_struct *task = get_proc_task(dir);
1767         unsigned fd = name_to_int(dentry);
1768         struct dentry *result = ERR_PTR(-ENOENT);
1769
1770         if (!task)
1771                 goto out_no_task;
1772         if (fd == ~0U)
1773                 goto out;
1774
1775         result = instantiate(dir, dentry, task, &fd);
1776 out:
1777         put_task_struct(task);
1778 out_no_task:
1779         return result;
1780 }
1781
1782 static int proc_readfd_common(struct file * filp, void * dirent,
1783                               filldir_t filldir, instantiate_t instantiate)
1784 {
1785         struct dentry *dentry = filp->f_path.dentry;
1786         struct inode *inode = dentry->d_inode;
1787         struct task_struct *p = get_proc_task(inode);
1788         unsigned int fd, ino;
1789         int retval;
1790         struct files_struct * files;
1791
1792         retval = -ENOENT;
1793         if (!p)
1794                 goto out_no_task;
1795         retval = 0;
1796
1797         fd = filp->f_pos;
1798         switch (fd) {
1799                 case 0:
1800                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1801                                 goto out;
1802                         filp->f_pos++;
1803                 case 1:
1804                         ino = parent_ino(dentry);
1805                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1806                                 goto out;
1807                         filp->f_pos++;
1808                 default:
1809                         files = get_files_struct(p);
1810                         if (!files)
1811                                 goto out;
1812                         rcu_read_lock();
1813                         for (fd = filp->f_pos-2;
1814                              fd < files_fdtable(files)->max_fds;
1815                              fd++, filp->f_pos++) {
1816                                 char name[PROC_NUMBUF];
1817                                 int len;
1818
1819                                 if (!fcheck_files(files, fd))
1820                                         continue;
1821                                 rcu_read_unlock();
1822
1823                                 len = snprintf(name, sizeof(name), "%d", fd);
1824                                 if (proc_fill_cache(filp, dirent, filldir,
1825                                                     name, len, instantiate,
1826                                                     p, &fd) < 0) {
1827                                         rcu_read_lock();
1828                                         break;
1829                                 }
1830                                 rcu_read_lock();
1831                         }
1832                         rcu_read_unlock();
1833                         put_files_struct(files);
1834         }
1835 out:
1836         put_task_struct(p);
1837 out_no_task:
1838         return retval;
1839 }
1840
1841 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1842                                     struct nameidata *nd)
1843 {
1844         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1845 }
1846
1847 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1848 {
1849         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1850 }
1851
1852 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1853                                       size_t len, loff_t *ppos)
1854 {
1855         char tmp[PROC_FDINFO_MAX];
1856         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1857         if (!err)
1858                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1859         return err;
1860 }
1861
1862 static const struct file_operations proc_fdinfo_file_operations = {
1863         .open           = nonseekable_open,
1864         .read           = proc_fdinfo_read,
1865 };
1866
1867 static const struct file_operations proc_fd_operations = {
1868         .read           = generic_read_dir,
1869         .readdir        = proc_readfd,
1870 };
1871
1872 /*
1873  * /proc/pid/fd needs a special permission handler so that a process can still
1874  * access /proc/self/fd after it has executed a setuid().
1875  */
1876 static int proc_fd_permission(struct inode *inode, int mask)
1877 {
1878         int rv;
1879
1880         rv = generic_permission(inode, mask, NULL);
1881         if (rv == 0)
1882                 return 0;
1883         if (task_pid(current) == proc_pid(inode))
1884                 rv = 0;
1885         return rv;
1886 }
1887
1888 /*
1889  * proc directories can do almost nothing..
1890  */
1891 static const struct inode_operations proc_fd_inode_operations = {
1892         .lookup         = proc_lookupfd,
1893         .permission     = proc_fd_permission,
1894         .setattr        = proc_setattr,
1895 };
1896
1897 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1898         struct dentry *dentry, struct task_struct *task, const void *ptr)
1899 {
1900         unsigned fd = *(unsigned *)ptr;
1901         struct inode *inode;
1902         struct proc_inode *ei;
1903         struct dentry *error = ERR_PTR(-ENOENT);
1904
1905         inode = proc_pid_make_inode(dir->i_sb, task);
1906         if (!inode)
1907                 goto out;
1908         ei = PROC_I(inode);
1909         ei->fd = fd;
1910         inode->i_mode = S_IFREG | S_IRUSR;
1911         inode->i_fop = &proc_fdinfo_file_operations;
1912         dentry->d_op = &tid_fd_dentry_operations;
1913         d_add(dentry, inode);
1914         /* Close the race of the process dying before we return the dentry */
1915         if (tid_fd_revalidate(dentry, NULL))
1916                 error = NULL;
1917
1918  out:
1919         return error;
1920 }
1921
1922 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1923                                         struct dentry *dentry,
1924                                         struct nameidata *nd)
1925 {
1926         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1927 }
1928
1929 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1930 {
1931         return proc_readfd_common(filp, dirent, filldir,
1932                                   proc_fdinfo_instantiate);
1933 }
1934
1935 static const struct file_operations proc_fdinfo_operations = {
1936         .read           = generic_read_dir,
1937         .readdir        = proc_readfdinfo,
1938 };
1939
1940 /*
1941  * proc directories can do almost nothing..
1942  */
1943 static const struct inode_operations proc_fdinfo_inode_operations = {
1944         .lookup         = proc_lookupfdinfo,
1945         .setattr        = proc_setattr,
1946 };
1947
1948
1949 static struct dentry *proc_pident_instantiate(struct inode *dir,
1950         struct dentry *dentry, struct task_struct *task, const void *ptr)
1951 {
1952         const struct pid_entry *p = ptr;
1953         struct inode *inode;
1954         struct proc_inode *ei;
1955         struct dentry *error = ERR_PTR(-EINVAL);
1956
1957         inode = proc_pid_make_inode(dir->i_sb, task);
1958         if (!inode)
1959                 goto out;
1960
1961         ei = PROC_I(inode);
1962         inode->i_mode = p->mode;
1963         if (S_ISDIR(inode->i_mode))
1964                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1965         if (p->iop)
1966                 inode->i_op = p->iop;
1967         if (p->fop)
1968                 inode->i_fop = p->fop;
1969         ei->op = p->op;
1970         dentry->d_op = &pid_dentry_operations;
1971         d_add(dentry, inode);
1972         /* Close the race of the process dying before we return the dentry */
1973         if (pid_revalidate(dentry, NULL))
1974                 error = NULL;
1975 out:
1976         return error;
1977 }
1978
1979 static struct dentry *proc_pident_lookup(struct inode *dir, 
1980                                          struct dentry *dentry,
1981                                          const struct pid_entry *ents,
1982                                          unsigned int nents)
1983 {
1984         struct dentry *error;
1985         struct task_struct *task = get_proc_task(dir);
1986         const struct pid_entry *p, *last;
1987
1988         error = ERR_PTR(-ENOENT);
1989
1990         if (!task)
1991                 goto out_no_task;
1992
1993         /*
1994          * Yes, it does not scale. And it should not. Don't add
1995          * new entries into /proc/<tgid>/ without very good reasons.
1996          */
1997         last = &ents[nents - 1];
1998         for (p = ents; p <= last; p++) {
1999                 if (p->len != dentry->d_name.len)
2000                         continue;
2001                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2002                         break;
2003         }
2004         if (p > last)
2005                 goto out;
2006
2007         error = proc_pident_instantiate(dir, dentry, task, p);
2008 out:
2009         put_task_struct(task);
2010 out_no_task:
2011         return error;
2012 }
2013
2014 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2015         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2016 {
2017         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2018                                 proc_pident_instantiate, task, p);
2019 }
2020
2021 static int proc_pident_readdir(struct file *filp,
2022                 void *dirent, filldir_t filldir,
2023                 const struct pid_entry *ents, unsigned int nents)
2024 {
2025         int i;
2026         struct dentry *dentry = filp->f_path.dentry;
2027         struct inode *inode = dentry->d_inode;
2028         struct task_struct *task = get_proc_task(inode);
2029         const struct pid_entry *p, *last;
2030         ino_t ino;
2031         int ret;
2032
2033         ret = -ENOENT;
2034         if (!task)
2035                 goto out_no_task;
2036
2037         ret = 0;
2038         i = filp->f_pos;
2039         switch (i) {
2040         case 0:
2041                 ino = inode->i_ino;
2042                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2043                         goto out;
2044                 i++;
2045                 filp->f_pos++;
2046                 /* fall through */
2047         case 1:
2048                 ino = parent_ino(dentry);
2049                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2050                         goto out;
2051                 i++;
2052                 filp->f_pos++;
2053                 /* fall through */
2054         default:
2055                 i -= 2;
2056                 if (i >= nents) {
2057                         ret = 1;
2058                         goto out;
2059                 }
2060                 p = ents + i;
2061                 last = &ents[nents - 1];
2062                 while (p <= last) {
2063                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2064                                 goto out;
2065                         filp->f_pos++;
2066                         p++;
2067                 }
2068         }
2069
2070         ret = 1;
2071 out:
2072         put_task_struct(task);
2073 out_no_task:
2074         return ret;
2075 }
2076
2077 #ifdef CONFIG_SECURITY
2078 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2079                                   size_t count, loff_t *ppos)
2080 {
2081         struct inode * inode = file->f_path.dentry->d_inode;
2082         char *p = NULL;
2083         ssize_t length;
2084         struct task_struct *task = get_proc_task(inode);
2085
2086         if (!task)
2087                 return -ESRCH;
2088
2089         length = security_getprocattr(task,
2090                                       (char*)file->f_path.dentry->d_name.name,
2091                                       &p);
2092         put_task_struct(task);
2093         if (length > 0)
2094                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2095         kfree(p);
2096         return length;
2097 }
2098
2099 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2100                                    size_t count, loff_t *ppos)
2101 {
2102         struct inode * inode = file->f_path.dentry->d_inode;
2103         char *page;
2104         ssize_t length;
2105         struct task_struct *task = get_proc_task(inode);
2106
2107         length = -ESRCH;
2108         if (!task)
2109                 goto out_no_task;
2110         if (count > PAGE_SIZE)
2111                 count = PAGE_SIZE;
2112
2113         /* No partial writes. */
2114         length = -EINVAL;
2115         if (*ppos != 0)
2116                 goto out;
2117
2118         length = -ENOMEM;
2119         page = (char*)__get_free_page(GFP_TEMPORARY);
2120         if (!page)
2121                 goto out;
2122
2123         length = -EFAULT;
2124         if (copy_from_user(page, buf, count))
2125                 goto out_free;
2126
2127         length = security_setprocattr(task,
2128                                       (char*)file->f_path.dentry->d_name.name,
2129                                       (void*)page, count);
2130 out_free:
2131         free_page((unsigned long) page);
2132 out:
2133         put_task_struct(task);
2134 out_no_task:
2135         return length;
2136 }
2137
2138 static const struct file_operations proc_pid_attr_operations = {
2139         .read           = proc_pid_attr_read,
2140         .write          = proc_pid_attr_write,
2141 };
2142
2143 static const struct pid_entry attr_dir_stuff[] = {
2144         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2145         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2146         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2147         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2148         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2149         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2150 };
2151
2152 static int proc_attr_dir_readdir(struct file * filp,
2153                              void * dirent, filldir_t filldir)
2154 {
2155         return proc_pident_readdir(filp,dirent,filldir,
2156                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2157 }
2158
2159 static const struct file_operations proc_attr_dir_operations = {
2160         .read           = generic_read_dir,
2161         .readdir        = proc_attr_dir_readdir,
2162 };
2163
2164 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2165                                 struct dentry *dentry, struct nameidata *nd)
2166 {
2167         return proc_pident_lookup(dir, dentry,
2168                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2169 }
2170
2171 static const struct inode_operations proc_attr_dir_inode_operations = {
2172         .lookup         = proc_attr_dir_lookup,
2173         .getattr        = pid_getattr,
2174         .setattr        = proc_setattr,
2175 };
2176
2177 #endif
2178
2179 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2180 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2181                                          size_t count, loff_t *ppos)
2182 {
2183         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2184         struct mm_struct *mm;
2185         char buffer[PROC_NUMBUF];
2186         size_t len;
2187         int ret;
2188
2189         if (!task)
2190                 return -ESRCH;
2191
2192         ret = 0;
2193         mm = get_task_mm(task);
2194         if (mm) {
2195                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2196                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2197                                 MMF_DUMP_FILTER_SHIFT));
2198                 mmput(mm);
2199                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2200         }
2201
2202         put_task_struct(task);
2203
2204         return ret;
2205 }
2206
2207 static ssize_t proc_coredump_filter_write(struct file *file,
2208                                           const char __user *buf,
2209                                           size_t count,
2210                                           loff_t *ppos)
2211 {
2212         struct task_struct *task;
2213         struct mm_struct *mm;
2214         char buffer[PROC_NUMBUF], *end;
2215         unsigned int val;
2216         int ret;
2217         int i;
2218         unsigned long mask;
2219
2220         ret = -EFAULT;
2221         memset(buffer, 0, sizeof(buffer));
2222         if (count > sizeof(buffer) - 1)
2223                 count = sizeof(buffer) - 1;
2224         if (copy_from_user(buffer, buf, count))
2225                 goto out_no_task;
2226
2227         ret = -EINVAL;
2228         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2229         if (*end == '\n')
2230                 end++;
2231         if (end - buffer == 0)
2232                 goto out_no_task;
2233
2234         ret = -ESRCH;
2235         task = get_proc_task(file->f_dentry->d_inode);
2236         if (!task)
2237                 goto out_no_task;
2238
2239         ret = end - buffer;
2240         mm = get_task_mm(task);
2241         if (!mm)
2242                 goto out_no_mm;
2243
2244         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2245                 if (val & mask)
2246                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2247                 else
2248                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2249         }
2250
2251         mmput(mm);
2252  out_no_mm:
2253         put_task_struct(task);
2254  out_no_task:
2255         return ret;
2256 }
2257
2258 static const struct file_operations proc_coredump_filter_operations = {
2259         .read           = proc_coredump_filter_read,
2260         .write          = proc_coredump_filter_write,
2261 };
2262 #endif
2263
2264 /*
2265  * /proc/self:
2266  */
2267 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2268                               int buflen)
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 -ENOENT;
2275         sprintf(tmp, "%d", tgid);
2276         return vfs_readlink(dentry,buffer,buflen,tmp);
2277 }
2278
2279 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2280 {
2281         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2282         pid_t tgid = task_tgid_nr_ns(current, ns);
2283         char tmp[PROC_NUMBUF];
2284         if (!tgid)
2285                 return ERR_PTR(-ENOENT);
2286         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2287         return ERR_PTR(vfs_follow_link(nd,tmp));
2288 }
2289
2290 static const struct inode_operations proc_self_inode_operations = {
2291         .readlink       = proc_self_readlink,
2292         .follow_link    = proc_self_follow_link,
2293 };
2294
2295 /*
2296  * proc base
2297  *
2298  * These are the directory entries in the root directory of /proc
2299  * that properly belong to the /proc filesystem, as they describe
2300  * describe something that is process related.
2301  */
2302 static const struct pid_entry proc_base_stuff[] = {
2303         NOD("self", S_IFLNK|S_IRWXUGO,
2304                 &proc_self_inode_operations, NULL, {}),
2305 };
2306
2307 /*
2308  *      Exceptional case: normally we are not allowed to unhash a busy
2309  * directory. In this case, however, we can do it - no aliasing problems
2310  * due to the way we treat inodes.
2311  */
2312 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2313 {
2314         struct inode *inode = dentry->d_inode;
2315         struct task_struct *task = get_proc_task(inode);
2316         if (task) {
2317                 put_task_struct(task);
2318                 return 1;
2319         }
2320         d_drop(dentry);
2321         return 0;
2322 }
2323
2324 static const struct dentry_operations proc_base_dentry_operations =
2325 {
2326         .d_revalidate   = proc_base_revalidate,
2327         .d_delete       = pid_delete_dentry,
2328 };
2329
2330 static struct dentry *proc_base_instantiate(struct inode *dir,
2331         struct dentry *dentry, struct task_struct *task, const void *ptr)
2332 {
2333         const struct pid_entry *p = ptr;
2334         struct inode *inode;
2335         struct proc_inode *ei;
2336         struct dentry *error = ERR_PTR(-EINVAL);
2337
2338         /* Allocate the inode */
2339         error = ERR_PTR(-ENOMEM);
2340         inode = new_inode(dir->i_sb);
2341         if (!inode)
2342                 goto out;
2343
2344         /* Initialize the inode */
2345         ei = PROC_I(inode);
2346         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2347
2348         /*
2349          * grab the reference to the task.
2350          */
2351         ei->pid = get_task_pid(task, PIDTYPE_PID);
2352         if (!ei->pid)
2353                 goto out_iput;
2354
2355         inode->i_mode = p->mode;
2356         if (S_ISDIR(inode->i_mode))
2357                 inode->i_nlink = 2;
2358         if (S_ISLNK(inode->i_mode))
2359                 inode->i_size = 64;
2360         if (p->iop)
2361                 inode->i_op = p->iop;
2362         if (p->fop)
2363                 inode->i_fop = p->fop;
2364         ei->op = p->op;
2365         dentry->d_op = &proc_base_dentry_operations;
2366         d_add(dentry, inode);
2367         error = NULL;
2368 out:
2369         return error;
2370 out_iput:
2371         iput(inode);
2372         goto out;
2373 }
2374
2375 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2376 {
2377         struct dentry *error;
2378         struct task_struct *task = get_proc_task(dir);
2379         const struct pid_entry *p, *last;
2380
2381         error = ERR_PTR(-ENOENT);
2382
2383         if (!task)
2384                 goto out_no_task;
2385
2386         /* Lookup the directory entry */
2387         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2388         for (p = proc_base_stuff; p <= last; p++) {
2389                 if (p->len != dentry->d_name.len)
2390                         continue;
2391                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2392                         break;
2393         }
2394         if (p > last)
2395                 goto out;
2396
2397         error = proc_base_instantiate(dir, dentry, task, p);
2398
2399 out:
2400         put_task_struct(task);
2401 out_no_task:
2402         return error;
2403 }
2404
2405 static int proc_base_fill_cache(struct file *filp, void *dirent,
2406         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2407 {
2408         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2409                                 proc_base_instantiate, task, p);
2410 }
2411
2412 #ifdef CONFIG_TASK_IO_ACCOUNTING
2413 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2414 {
2415         struct task_io_accounting acct = task->ioac;
2416         unsigned long flags;
2417
2418         if (whole && lock_task_sighand(task, &flags)) {
2419                 struct task_struct *t = task;
2420
2421                 task_io_accounting_add(&acct, &task->signal->ioac);
2422                 while_each_thread(task, t)
2423                         task_io_accounting_add(&acct, &t->ioac);
2424
2425                 unlock_task_sighand(task, &flags);
2426         }
2427         return sprintf(buffer,
2428                         "rchar: %llu\n"
2429                         "wchar: %llu\n"
2430                         "syscr: %llu\n"
2431                         "syscw: %llu\n"
2432                         "read_bytes: %llu\n"
2433                         "write_bytes: %llu\n"
2434                         "cancelled_write_bytes: %llu\n",
2435                         (unsigned long long)acct.rchar,
2436                         (unsigned long long)acct.wchar,
2437                         (unsigned long long)acct.syscr,
2438                         (unsigned long long)acct.syscw,
2439                         (unsigned long long)acct.read_bytes,
2440                         (unsigned long long)acct.write_bytes,
2441                         (unsigned long long)acct.cancelled_write_bytes);
2442 }
2443
2444 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2445 {
2446         return do_io_accounting(task, buffer, 0);
2447 }
2448
2449 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2450 {
2451         return do_io_accounting(task, buffer, 1);
2452 }
2453 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2454
2455 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2456                                 struct pid *pid, struct task_struct *task)
2457 {
2458         seq_printf(m, "%08x\n", task->personality);
2459         return 0;
2460 }
2461
2462 /*
2463  * Thread groups
2464  */
2465 static const struct file_operations proc_task_operations;
2466 static const struct inode_operations proc_task_inode_operations;
2467
2468 static const struct pid_entry tgid_base_stuff[] = {
2469         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2470         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2471         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2472 #ifdef CONFIG_NET
2473         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2474 #endif
2475         REG("environ",    S_IRUSR, proc_environ_operations),
2476         INF("auxv",       S_IRUSR, proc_pid_auxv),
2477         ONE("status",     S_IRUGO, proc_pid_status),
2478         ONE("personality", S_IRUSR, proc_pid_personality),
2479         INF("limits",     S_IRUSR, proc_pid_limits),
2480 #ifdef CONFIG_SCHED_DEBUG
2481         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2482 #endif
2483 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2484         INF("syscall",    S_IRUSR, proc_pid_syscall),
2485 #endif
2486         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2487         ONE("stat",       S_IRUGO, proc_tgid_stat),
2488         ONE("statm",      S_IRUGO, proc_pid_statm),
2489         REG("maps",       S_IRUGO, proc_maps_operations),
2490 #ifdef CONFIG_NUMA
2491         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2492 #endif
2493         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2494         LNK("cwd",        proc_cwd_link),
2495         LNK("root",       proc_root_link),
2496         LNK("exe",        proc_exe_link),
2497         REG("mounts",     S_IRUGO, proc_mounts_operations),
2498         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2499         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2500 #ifdef CONFIG_PROC_PAGE_MONITOR
2501         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2502         REG("smaps",      S_IRUGO, proc_smaps_operations),
2503         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2504 #endif
2505 #ifdef CONFIG_SECURITY
2506         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2507 #endif
2508 #ifdef CONFIG_KALLSYMS
2509         INF("wchan",      S_IRUGO, proc_pid_wchan),
2510 #endif
2511 #ifdef CONFIG_STACKTRACE
2512         ONE("stack",      S_IRUSR, proc_pid_stack),
2513 #endif
2514 #ifdef CONFIG_SCHEDSTATS
2515         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2516 #endif
2517 #ifdef CONFIG_LATENCYTOP
2518         REG("latency",  S_IRUGO, proc_lstats_operations),
2519 #endif
2520 #ifdef CONFIG_PROC_PID_CPUSET
2521         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2522 #endif
2523 #ifdef CONFIG_CGROUPS
2524         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2525 #endif
2526         INF("oom_score",  S_IRUGO, proc_oom_score),
2527         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2528 #ifdef CONFIG_AUDITSYSCALL
2529         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2530         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2531 #endif
2532 #ifdef CONFIG_FAULT_INJECTION
2533         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2534 #endif
2535 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2536         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2537 #endif
2538 #ifdef CONFIG_TASK_IO_ACCOUNTING
2539         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2540 #endif
2541 };
2542
2543 static int proc_tgid_base_readdir(struct file * filp,
2544                              void * dirent, filldir_t filldir)
2545 {
2546         return proc_pident_readdir(filp,dirent,filldir,
2547                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2548 }
2549
2550 static const struct file_operations proc_tgid_base_operations = {
2551         .read           = generic_read_dir,
2552         .readdir        = proc_tgid_base_readdir,
2553 };
2554
2555 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2556         return proc_pident_lookup(dir, dentry,
2557                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2558 }
2559
2560 static const struct inode_operations proc_tgid_base_inode_operations = {
2561         .lookup         = proc_tgid_base_lookup,
2562         .getattr        = pid_getattr,
2563         .setattr        = proc_setattr,
2564 };
2565
2566 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2567 {
2568         struct dentry *dentry, *leader, *dir;
2569         char buf[PROC_NUMBUF];
2570         struct qstr name;
2571
2572         name.name = buf;
2573         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2574         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2575         if (dentry) {
2576                 if (!(current->flags & PF_EXITING))
2577                         shrink_dcache_parent(dentry);
2578                 d_drop(dentry);
2579                 dput(dentry);
2580         }
2581
2582         if (tgid == 0)
2583                 goto out;
2584
2585         name.name = buf;
2586         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2587         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2588         if (!leader)
2589                 goto out;
2590
2591         name.name = "task";
2592         name.len = strlen(name.name);
2593         dir = d_hash_and_lookup(leader, &name);
2594         if (!dir)
2595                 goto out_put_leader;
2596
2597         name.name = buf;
2598         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2599         dentry = d_hash_and_lookup(dir, &name);
2600         if (dentry) {
2601                 shrink_dcache_parent(dentry);
2602                 d_drop(dentry);
2603                 dput(dentry);
2604         }
2605
2606         dput(dir);
2607 out_put_leader:
2608         dput(leader);
2609 out:
2610         return;
2611 }
2612
2613 /**
2614  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2615  * @task: task that should be flushed.
2616  *
2617  * When flushing dentries from proc, one needs to flush them from global
2618  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2619  * in. This call is supposed to do all of this job.
2620  *
2621  * Looks in the dcache for
2622  * /proc/@pid
2623  * /proc/@tgid/task/@pid
2624  * if either directory is present flushes it and all of it'ts children
2625  * from the dcache.
2626  *
2627  * It is safe and reasonable to cache /proc entries for a task until
2628  * that task exits.  After that they just clog up the dcache with
2629  * useless entries, possibly causing useful dcache entries to be
2630  * flushed instead.  This routine is proved to flush those useless
2631  * dcache entries at process exit time.
2632  *
2633  * NOTE: This routine is just an optimization so it does not guarantee
2634  *       that no dcache entries will exist at process exit time it
2635  *       just makes it very unlikely that any will persist.
2636  */
2637
2638 void proc_flush_task(struct task_struct *task)
2639 {
2640         int i;
2641         struct pid *pid, *tgid = NULL;
2642         struct upid *upid;
2643
2644         pid = task_pid(task);
2645         if (thread_group_leader(task))
2646                 tgid = task_tgid(task);
2647
2648         for (i = 0; i <= pid->level; i++) {
2649                 upid = &pid->numbers[i];
2650                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2651                         tgid ? tgid->numbers[i].nr : 0);
2652         }
2653
2654         upid = &pid->numbers[pid->level];
2655         if (upid->nr == 1)
2656                 pid_ns_release_proc(upid->ns);
2657 }
2658
2659 static struct dentry *proc_pid_instantiate(struct inode *dir,
2660                                            struct dentry * dentry,
2661                                            struct task_struct *task, const void *ptr)
2662 {
2663         struct dentry *error = ERR_PTR(-ENOENT);
2664         struct inode *inode;
2665
2666         inode = proc_pid_make_inode(dir->i_sb, task);
2667         if (!inode)
2668                 goto out;
2669
2670         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2671         inode->i_op = &proc_tgid_base_inode_operations;
2672         inode->i_fop = &proc_tgid_base_operations;
2673         inode->i_flags|=S_IMMUTABLE;
2674
2675         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2676                 ARRAY_SIZE(tgid_base_stuff));
2677
2678         dentry->d_op = &pid_dentry_operations;
2679
2680         d_add(dentry, inode);
2681         /* Close the race of the process dying before we return the dentry */
2682         if (pid_revalidate(dentry, NULL))
2683                 error = NULL;
2684 out:
2685         return error;
2686 }
2687
2688 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2689 {
2690         struct dentry *result = ERR_PTR(-ENOENT);
2691         struct task_struct *task;
2692         unsigned tgid;
2693         struct pid_namespace *ns;
2694
2695         result = proc_base_lookup(dir, dentry);
2696         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2697                 goto out;
2698
2699         tgid = name_to_int(dentry);
2700         if (tgid == ~0U)
2701                 goto out;
2702
2703         ns = dentry->d_sb->s_fs_info;
2704         rcu_read_lock();
2705         task = find_task_by_pid_ns(tgid, ns);
2706         if (task)
2707                 get_task_struct(task);
2708         rcu_read_unlock();
2709         if (!task)
2710                 goto out;
2711
2712         result = proc_pid_instantiate(dir, dentry, task, NULL);
2713         put_task_struct(task);
2714 out:
2715         return result;
2716 }
2717
2718 /*
2719  * Find the first task with tgid >= tgid
2720  *
2721  */
2722 struct tgid_iter {
2723         unsigned int tgid;
2724         struct task_struct *task;
2725 };
2726 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2727 {
2728         struct pid *pid;
2729
2730         if (iter.task)
2731                 put_task_struct(iter.task);
2732         rcu_read_lock();
2733 retry:
2734         iter.task = NULL;
2735         pid = find_ge_pid(iter.tgid, ns);
2736         if (pid) {
2737                 iter.tgid = pid_nr_ns(pid, ns);
2738                 iter.task = pid_task(pid, PIDTYPE_PID);
2739                 /* What we to know is if the pid we have find is the
2740                  * pid of a thread_group_leader.  Testing for task
2741                  * being a thread_group_leader is the obvious thing
2742                  * todo but there is a window when it fails, due to
2743                  * the pid transfer logic in de_thread.
2744                  *
2745                  * So we perform the straight forward test of seeing
2746                  * if the pid we have found is the pid of a thread
2747                  * group leader, and don't worry if the task we have
2748                  * found doesn't happen to be a thread group leader.
2749                  * As we don't care in the case of readdir.
2750                  */
2751                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2752                         iter.tgid += 1;
2753                         goto retry;
2754                 }
2755                 get_task_struct(iter.task);
2756         }
2757         rcu_read_unlock();
2758         return iter;
2759 }
2760
2761 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2762
2763 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2764         struct tgid_iter iter)
2765 {
2766         char name[PROC_NUMBUF];
2767         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2768         return proc_fill_cache(filp, dirent, filldir, name, len,
2769                                 proc_pid_instantiate, iter.task, NULL);
2770 }
2771
2772 /* for the /proc/ directory itself, after non-process stuff has been done */
2773 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2774 {
2775         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2776         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2777         struct tgid_iter iter;
2778         struct pid_namespace *ns;
2779
2780         if (!reaper)
2781                 goto out_no_task;
2782
2783         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2784                 const struct pid_entry *p = &proc_base_stuff[nr];
2785                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2786                         goto out;
2787         }
2788
2789         ns = filp->f_dentry->d_sb->s_fs_info;
2790         iter.task = NULL;
2791         iter.tgid = filp->f_pos - TGID_OFFSET;
2792         for (iter = next_tgid(ns, iter);
2793              iter.task;
2794              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2795                 filp->f_pos = iter.tgid + TGID_OFFSET;
2796                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2797                         put_task_struct(iter.task);
2798                         goto out;
2799                 }
2800         }
2801         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2802 out:
2803         put_task_struct(reaper);
2804 out_no_task:
2805         return 0;
2806 }
2807
2808 /*
2809  * Tasks
2810  */
2811 static const struct pid_entry tid_base_stuff[] = {
2812         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2813         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2814         REG("environ",   S_IRUSR, proc_environ_operations),
2815         INF("auxv",      S_IRUSR, proc_pid_auxv),
2816         ONE("status",    S_IRUGO, proc_pid_status),
2817         ONE("personality", S_IRUSR, proc_pid_personality),
2818         INF("limits",    S_IRUSR, proc_pid_limits),
2819 #ifdef CONFIG_SCHED_DEBUG
2820         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2821 #endif
2822 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2823         INF("syscall",   S_IRUSR, proc_pid_syscall),
2824 #endif
2825         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2826         ONE("stat",      S_IRUGO, proc_tid_stat),
2827         ONE("statm",     S_IRUGO, proc_pid_statm),
2828         REG("maps",      S_IRUGO, proc_maps_operations),
2829 #ifdef CONFIG_NUMA
2830         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2831 #endif
2832         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2833         LNK("cwd",       proc_cwd_link),
2834         LNK("root",      proc_root_link),
2835         LNK("exe",       proc_exe_link),
2836         REG("mounts",    S_IRUGO, proc_mounts_operations),
2837         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2838 #ifdef CONFIG_PROC_PAGE_MONITOR
2839         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2840         REG("smaps",     S_IRUGO, proc_smaps_operations),
2841         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2842 #endif
2843 #ifdef CONFIG_SECURITY
2844         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2845 #endif
2846 #ifdef CONFIG_KALLSYMS
2847         INF("wchan",     S_IRUGO, proc_pid_wchan),
2848 #endif
2849 #ifdef CONFIG_STACKTRACE
2850         ONE("stack",      S_IRUSR, proc_pid_stack),
2851 #endif
2852 #ifdef CONFIG_SCHEDSTATS
2853         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2854 #endif
2855 #ifdef CONFIG_LATENCYTOP
2856         REG("latency",  S_IRUGO, proc_lstats_operations),
2857 #endif
2858 #ifdef CONFIG_PROC_PID_CPUSET
2859         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2860 #endif
2861 #ifdef CONFIG_CGROUPS
2862         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2863 #endif
2864         INF("oom_score", S_IRUGO, proc_oom_score),
2865         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2866 #ifdef CONFIG_AUDITSYSCALL
2867         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2868         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2869 #endif
2870 #ifdef CONFIG_FAULT_INJECTION
2871         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2872 #endif
2873 #ifdef CONFIG_TASK_IO_ACCOUNTING
2874         INF("io",       S_IRUGO, proc_tid_io_accounting),
2875 #endif
2876 };
2877
2878 static int proc_tid_base_readdir(struct file * filp,
2879                              void * dirent, filldir_t filldir)
2880 {
2881         return proc_pident_readdir(filp,dirent,filldir,
2882                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2883 }
2884
2885 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2886         return proc_pident_lookup(dir, dentry,
2887                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2888 }
2889
2890 static const struct file_operations proc_tid_base_operations = {
2891         .read           = generic_read_dir,
2892         .readdir        = proc_tid_base_readdir,
2893 };
2894
2895 static const struct inode_operations proc_tid_base_inode_operations = {
2896         .lookup         = proc_tid_base_lookup,
2897         .getattr        = pid_getattr,
2898         .setattr        = proc_setattr,
2899 };
2900
2901 static struct dentry *proc_task_instantiate(struct inode *dir,
2902         struct dentry *dentry, struct task_struct *task, const void *ptr)
2903 {
2904         struct dentry *error = ERR_PTR(-ENOENT);
2905         struct inode *inode;
2906         inode = proc_pid_make_inode(dir->i_sb, task);
2907
2908         if (!inode)
2909                 goto out;
2910         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2911         inode->i_op = &proc_tid_base_inode_operations;
2912         inode->i_fop = &proc_tid_base_operations;
2913         inode->i_flags|=S_IMMUTABLE;
2914
2915         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2916                 ARRAY_SIZE(tid_base_stuff));
2917
2918         dentry->d_op = &pid_dentry_operations;
2919
2920         d_add(dentry, inode);
2921         /* Close the race of the process dying before we return the dentry */
2922         if (pid_revalidate(dentry, NULL))
2923                 error = NULL;
2924 out:
2925         return error;
2926 }
2927
2928 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2929 {
2930         struct dentry *result = ERR_PTR(-ENOENT);
2931         struct task_struct *task;
2932         struct task_struct *leader = get_proc_task(dir);
2933         unsigned tid;
2934         struct pid_namespace *ns;
2935
2936         if (!leader)
2937                 goto out_no_task;
2938
2939         tid = name_to_int(dentry);
2940         if (tid == ~0U)
2941                 goto out;
2942
2943         ns = dentry->d_sb->s_fs_info;
2944         rcu_read_lock();
2945         task = find_task_by_pid_ns(tid, ns);
2946         if (task)
2947                 get_task_struct(task);
2948         rcu_read_unlock();
2949         if (!task)
2950                 goto out;
2951         if (!same_thread_group(leader, task))
2952                 goto out_drop_task;
2953
2954         result = proc_task_instantiate(dir, dentry, task, NULL);
2955 out_drop_task:
2956         put_task_struct(task);
2957 out:
2958         put_task_struct(leader);
2959 out_no_task:
2960         return result;
2961 }
2962
2963 /*
2964  * Find the first tid of a thread group to return to user space.
2965  *
2966  * Usually this is just the thread group leader, but if the users
2967  * buffer was too small or there was a seek into the middle of the
2968  * directory we have more work todo.
2969  *
2970  * In the case of a short read we start with find_task_by_pid.
2971  *
2972  * In the case of a seek we start with the leader and walk nr
2973  * threads past it.
2974  */
2975 static struct task_struct *first_tid(struct task_struct *leader,
2976                 int tid, int nr, struct pid_namespace *ns)
2977 {
2978         struct task_struct *pos;
2979
2980         rcu_read_lock();
2981         /* Attempt to start with the pid of a thread */
2982         if (tid && (nr > 0)) {
2983                 pos = find_task_by_pid_ns(tid, ns);
2984                 if (pos && (pos->group_leader == leader))
2985                         goto found;
2986         }
2987
2988         /* If nr exceeds the number of threads there is nothing todo */
2989         pos = NULL;
2990         if (nr && nr >= get_nr_threads(leader))
2991                 goto out;
2992
2993         /* If we haven't found our starting place yet start
2994          * with the leader and walk nr threads forward.
2995          */
2996         for (pos = leader; nr > 0; --nr) {
2997                 pos = next_thread(pos);
2998                 if (pos == leader) {
2999                         pos = NULL;
3000                         goto out;
3001                 }
3002         }
3003 found:
3004         get_task_struct(pos);
3005 out:
3006         rcu_read_unlock();
3007         return pos;
3008 }
3009
3010 /*
3011  * Find the next thread in the thread list.
3012  * Return NULL if there is an error or no next thread.
3013  *
3014  * The reference to the input task_struct is released.
3015  */
3016 static struct task_struct *next_tid(struct task_struct *start)
3017 {
3018         struct task_struct *pos = NULL;
3019         rcu_read_lock();
3020         if (pid_alive(start)) {
3021                 pos = next_thread(start);
3022                 if (thread_group_leader(pos))
3023                         pos = NULL;
3024                 else
3025                         get_task_struct(pos);
3026         }
3027         rcu_read_unlock();
3028         put_task_struct(start);
3029         return pos;
3030 }
3031
3032 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3033         struct task_struct *task, int tid)
3034 {
3035         char name[PROC_NUMBUF];
3036         int len = snprintf(name, sizeof(name), "%d", tid);
3037         return proc_fill_cache(filp, dirent, filldir, name, len,
3038                                 proc_task_instantiate, task, NULL);
3039 }
3040
3041 /* for the /proc/TGID/task/ directories */
3042 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3043 {
3044         struct dentry *dentry = filp->f_path.dentry;
3045         struct inode *inode = dentry->d_inode;
3046         struct task_struct *leader = NULL;
3047         struct task_struct *task;
3048         int retval = -ENOENT;
3049         ino_t ino;
3050         int tid;
3051         struct pid_namespace *ns;
3052
3053         task = get_proc_task(inode);
3054         if (!task)
3055                 goto out_no_task;
3056         rcu_read_lock();
3057         if (pid_alive(task)) {
3058                 leader = task->group_leader;
3059                 get_task_struct(leader);
3060         }
3061         rcu_read_unlock();
3062         put_task_struct(task);
3063         if (!leader)
3064                 goto out_no_task;
3065         retval = 0;
3066
3067         switch ((unsigned long)filp->f_pos) {
3068         case 0:
3069                 ino = inode->i_ino;
3070                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3071                         goto out;
3072                 filp->f_pos++;
3073                 /* fall through */
3074         case 1:
3075                 ino = parent_ino(dentry);
3076                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3077                         goto out;
3078                 filp->f_pos++;
3079                 /* fall through */
3080         }
3081
3082         /* f_version caches the tgid value that the last readdir call couldn't
3083          * return. lseek aka telldir automagically resets f_version to 0.
3084          */
3085         ns = filp->f_dentry->d_sb->s_fs_info;
3086         tid = (int)filp->f_version;
3087         filp->f_version = 0;
3088         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3089              task;
3090              task = next_tid(task), filp->f_pos++) {
3091                 tid = task_pid_nr_ns(task, ns);
3092                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3093                         /* returning this tgid failed, save it as the first
3094                          * pid for the next readir call */
3095                         filp->f_version = (u64)tid;
3096                         put_task_struct(task);
3097                         break;
3098                 }
3099         }
3100 out:
3101         put_task_struct(leader);
3102 out_no_task:
3103         return retval;
3104 }
3105
3106 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3107 {
3108         struct inode *inode = dentry->d_inode;
3109         struct task_struct *p = get_proc_task(inode);
3110         generic_fillattr(inode, stat);
3111
3112         if (p) {
3113                 stat->nlink += get_nr_threads(p);
3114                 put_task_struct(p);
3115         }
3116
3117         return 0;
3118 }
3119
3120 static const struct inode_operations proc_task_inode_operations = {
3121         .lookup         = proc_task_lookup,
3122         .getattr        = proc_task_getattr,
3123         .setattr        = proc_setattr,
3124 };
3125
3126 static const struct file_operations proc_task_operations = {
3127         .read           = generic_read_dir,
3128         .readdir        = proc_task_readdir,
3129 };