sched: Add 'autogroup' scheduling feature: automated per session task groups
[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/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/fs_struct.h>
85 #include <linux/slab.h>
86 #include "internal.h"
87
88 /* NOTE:
89  *      Implementing inode permission operations in /proc is almost
90  *      certainly an error.  Permission checks need to happen during
91  *      each system call not at open time.  The reason is that most of
92  *      what we wish to check for permissions in /proc varies at runtime.
93  *
94  *      The classic example of a problem is opening file descriptors
95  *      in /proc for a task before it execs a suid executable.
96  */
97
98 struct pid_entry {
99         char *name;
100         int len;
101         mode_t mode;
102         const struct inode_operations *iop;
103         const struct file_operations *fop;
104         union proc_op op;
105 };
106
107 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
108         .name = (NAME),                                 \
109         .len  = sizeof(NAME) - 1,                       \
110         .mode = MODE,                                   \
111         .iop  = IOP,                                    \
112         .fop  = FOP,                                    \
113         .op   = OP,                                     \
114 }
115
116 #define DIR(NAME, MODE, iops, fops)     \
117         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
118 #define LNK(NAME, get_link)                                     \
119         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
120                 &proc_pid_link_inode_operations, NULL,          \
121                 { .proc_get_link = get_link } )
122 #define REG(NAME, MODE, fops)                           \
123         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
124 #define INF(NAME, MODE, read)                           \
125         NOD(NAME, (S_IFREG|(MODE)),                     \
126                 NULL, &proc_info_file_operations,       \
127                 { .proc_read = read } )
128 #define ONE(NAME, MODE, show)                           \
129         NOD(NAME, (S_IFREG|(MODE)),                     \
130                 NULL, &proc_single_file_operations,     \
131                 { .proc_show = show } )
132
133 /*
134  * Count the number of hardlinks for the pid_entry table, excluding the .
135  * and .. links.
136  */
137 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
138         unsigned int n)
139 {
140         unsigned int i;
141         unsigned int count;
142
143         count = 0;
144         for (i = 0; i < n; ++i) {
145                 if (S_ISDIR(entries[i].mode))
146                         ++count;
147         }
148
149         return count;
150 }
151
152 static int get_task_root(struct task_struct *task, struct path *root)
153 {
154         int result = -ENOENT;
155
156         task_lock(task);
157         if (task->fs) {
158                 get_fs_root(task->fs, root);
159                 result = 0;
160         }
161         task_unlock(task);
162         return result;
163 }
164
165 static int proc_cwd_link(struct inode *inode, struct path *path)
166 {
167         struct task_struct *task = get_proc_task(inode);
168         int result = -ENOENT;
169
170         if (task) {
171                 task_lock(task);
172                 if (task->fs) {
173                         get_fs_pwd(task->fs, path);
174                         result = 0;
175                 }
176                 task_unlock(task);
177                 put_task_struct(task);
178         }
179         return result;
180 }
181
182 static int proc_root_link(struct inode *inode, struct path *path)
183 {
184         struct task_struct *task = get_proc_task(inode);
185         int result = -ENOENT;
186
187         if (task) {
188                 result = get_task_root(task, path);
189                 put_task_struct(task);
190         }
191         return result;
192 }
193
194 /*
195  * Return zero if current may access user memory in @task, -error if not.
196  */
197 static int check_mem_permission(struct task_struct *task)
198 {
199         /*
200          * A task can always look at itself, in case it chooses
201          * to use system calls instead of load instructions.
202          */
203         if (task == current)
204                 return 0;
205
206         /*
207          * If current is actively ptrace'ing, and would also be
208          * permitted to freshly attach with ptrace now, permit it.
209          */
210         if (task_is_stopped_or_traced(task)) {
211                 int match;
212                 rcu_read_lock();
213                 match = (tracehook_tracer_task(task) == current);
214                 rcu_read_unlock();
215                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
216                         return 0;
217         }
218
219         /*
220          * Noone else is allowed.
221          */
222         return -EPERM;
223 }
224
225 struct mm_struct *mm_for_maps(struct task_struct *task)
226 {
227         struct mm_struct *mm;
228
229         if (mutex_lock_killable(&task->signal->cred_guard_mutex))
230                 return NULL;
231
232         mm = get_task_mm(task);
233         if (mm && mm != current->mm &&
234                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
235                 mmput(mm);
236                 mm = NULL;
237         }
238         mutex_unlock(&task->signal->cred_guard_mutex);
239
240         return mm;
241 }
242
243 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
244 {
245         int res = 0;
246         unsigned int len;
247         struct mm_struct *mm = get_task_mm(task);
248         if (!mm)
249                 goto out;
250         if (!mm->arg_end)
251                 goto out_mm;    /* Shh! No looking before we're done */
252
253         len = mm->arg_end - mm->arg_start;
254  
255         if (len > PAGE_SIZE)
256                 len = PAGE_SIZE;
257  
258         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
259
260         // If the nul at the end of args has been overwritten, then
261         // assume application is using setproctitle(3).
262         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
263                 len = strnlen(buffer, res);
264                 if (len < res) {
265                     res = len;
266                 } else {
267                         len = mm->env_end - mm->env_start;
268                         if (len > PAGE_SIZE - res)
269                                 len = PAGE_SIZE - res;
270                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
271                         res = strnlen(buffer, res);
272                 }
273         }
274 out_mm:
275         mmput(mm);
276 out:
277         return res;
278 }
279
280 static int proc_pid_auxv(struct task_struct *task, char *buffer)
281 {
282         int res = 0;
283         struct mm_struct *mm = get_task_mm(task);
284         if (mm) {
285                 unsigned int nwords = 0;
286                 do {
287                         nwords += 2;
288                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
289                 res = nwords * sizeof(mm->saved_auxv[0]);
290                 if (res > PAGE_SIZE)
291                         res = PAGE_SIZE;
292                 memcpy(buffer, mm->saved_auxv, res);
293                 mmput(mm);
294         }
295         return res;
296 }
297
298
299 #ifdef CONFIG_KALLSYMS
300 /*
301  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
302  * Returns the resolved symbol.  If that fails, simply return the address.
303  */
304 static int proc_pid_wchan(struct task_struct *task, char *buffer)
305 {
306         unsigned long wchan;
307         char symname[KSYM_NAME_LEN];
308
309         wchan = get_wchan(task);
310
311         if (lookup_symbol_name(wchan, symname) < 0)
312                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
313                         return 0;
314                 else
315                         return sprintf(buffer, "%lu", wchan);
316         else
317                 return sprintf(buffer, "%s", symname);
318 }
319 #endif /* CONFIG_KALLSYMS */
320
321 #ifdef CONFIG_STACKTRACE
322
323 #define MAX_STACK_TRACE_DEPTH   64
324
325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
326                           struct pid *pid, struct task_struct *task)
327 {
328         struct stack_trace trace;
329         unsigned long *entries;
330         int i;
331
332         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
333         if (!entries)
334                 return -ENOMEM;
335
336         trace.nr_entries        = 0;
337         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
338         trace.entries           = entries;
339         trace.skip              = 0;
340         save_stack_trace_tsk(task, &trace);
341
342         for (i = 0; i < trace.nr_entries; i++) {
343                 seq_printf(m, "[<%p>] %pS\n",
344                            (void *)entries[i], (void *)entries[i]);
345         }
346         kfree(entries);
347
348         return 0;
349 }
350 #endif
351
352 #ifdef CONFIG_SCHEDSTATS
353 /*
354  * Provides /proc/PID/schedstat
355  */
356 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
357 {
358         return sprintf(buffer, "%llu %llu %lu\n",
359                         (unsigned long long)task->se.sum_exec_runtime,
360                         (unsigned long long)task->sched_info.run_delay,
361                         task->sched_info.pcount);
362 }
363 #endif
364
365 #ifdef CONFIG_LATENCYTOP
366 static int lstats_show_proc(struct seq_file *m, void *v)
367 {
368         int i;
369         struct inode *inode = m->private;
370         struct task_struct *task = get_proc_task(inode);
371
372         if (!task)
373                 return -ESRCH;
374         seq_puts(m, "Latency Top version : v0.1\n");
375         for (i = 0; i < 32; i++) {
376                 if (task->latency_record[i].backtrace[0]) {
377                         int q;
378                         seq_printf(m, "%i %li %li ",
379                                 task->latency_record[i].count,
380                                 task->latency_record[i].time,
381                                 task->latency_record[i].max);
382                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
383                                 char sym[KSYM_SYMBOL_LEN];
384                                 char *c;
385                                 if (!task->latency_record[i].backtrace[q])
386                                         break;
387                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
388                                         break;
389                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
390                                 c = strchr(sym, '+');
391                                 if (c)
392                                         *c = 0;
393                                 seq_printf(m, "%s ", sym);
394                         }
395                         seq_printf(m, "\n");
396                 }
397
398         }
399         put_task_struct(task);
400         return 0;
401 }
402
403 static int lstats_open(struct inode *inode, struct file *file)
404 {
405         return single_open(file, lstats_show_proc, inode);
406 }
407
408 static ssize_t lstats_write(struct file *file, const char __user *buf,
409                             size_t count, loff_t *offs)
410 {
411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
412
413         if (!task)
414                 return -ESRCH;
415         clear_all_latency_tracing(task);
416         put_task_struct(task);
417
418         return count;
419 }
420
421 static const struct file_operations proc_lstats_operations = {
422         .open           = lstats_open,
423         .read           = seq_read,
424         .write          = lstats_write,
425         .llseek         = seq_lseek,
426         .release        = single_release,
427 };
428
429 #endif
430
431 static int proc_oom_score(struct task_struct *task, char *buffer)
432 {
433         unsigned long points = 0;
434
435         read_lock(&tasklist_lock);
436         if (pid_alive(task))
437                 points = oom_badness(task, NULL, NULL,
438                                         totalram_pages + total_swap_pages);
439         read_unlock(&tasklist_lock);
440         return sprintf(buffer, "%lu\n", points);
441 }
442
443 struct limit_names {
444         char *name;
445         char *unit;
446 };
447
448 static const struct limit_names lnames[RLIM_NLIMITS] = {
449         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
450         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
451         [RLIMIT_DATA] = {"Max data size", "bytes"},
452         [RLIMIT_STACK] = {"Max stack size", "bytes"},
453         [RLIMIT_CORE] = {"Max core file size", "bytes"},
454         [RLIMIT_RSS] = {"Max resident set", "bytes"},
455         [RLIMIT_NPROC] = {"Max processes", "processes"},
456         [RLIMIT_NOFILE] = {"Max open files", "files"},
457         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
458         [RLIMIT_AS] = {"Max address space", "bytes"},
459         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
460         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
461         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
462         [RLIMIT_NICE] = {"Max nice priority", NULL},
463         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
464         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
465 };
466
467 /* Display limits for a process */
468 static int proc_pid_limits(struct task_struct *task, char *buffer)
469 {
470         unsigned int i;
471         int count = 0;
472         unsigned long flags;
473         char *bufptr = buffer;
474
475         struct rlimit rlim[RLIM_NLIMITS];
476
477         if (!lock_task_sighand(task, &flags))
478                 return 0;
479         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
480         unlock_task_sighand(task, &flags);
481
482         /*
483          * print the file header
484          */
485         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
486                         "Limit", "Soft Limit", "Hard Limit", "Units");
487
488         for (i = 0; i < RLIM_NLIMITS; i++) {
489                 if (rlim[i].rlim_cur == RLIM_INFINITY)
490                         count += sprintf(&bufptr[count], "%-25s %-20s ",
491                                          lnames[i].name, "unlimited");
492                 else
493                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
494                                          lnames[i].name, rlim[i].rlim_cur);
495
496                 if (rlim[i].rlim_max == RLIM_INFINITY)
497                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
498                 else
499                         count += sprintf(&bufptr[count], "%-20lu ",
500                                          rlim[i].rlim_max);
501
502                 if (lnames[i].unit)
503                         count += sprintf(&bufptr[count], "%-10s\n",
504                                          lnames[i].unit);
505                 else
506                         count += sprintf(&bufptr[count], "\n");
507         }
508
509         return count;
510 }
511
512 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
513 static int proc_pid_syscall(struct task_struct *task, char *buffer)
514 {
515         long nr;
516         unsigned long args[6], sp, pc;
517
518         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
519                 return sprintf(buffer, "running\n");
520
521         if (nr < 0)
522                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
523
524         return sprintf(buffer,
525                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
526                        nr,
527                        args[0], args[1], args[2], args[3], args[4], args[5],
528                        sp, pc);
529 }
530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
531
532 /************************************************************************/
533 /*                       Here the fs part begins                        */
534 /************************************************************************/
535
536 /* permission checks */
537 static int proc_fd_access_allowed(struct inode *inode)
538 {
539         struct task_struct *task;
540         int allowed = 0;
541         /* Allow access to a task's file descriptors if it is us or we
542          * may use ptrace attach to the process and find out that
543          * information.
544          */
545         task = get_proc_task(inode);
546         if (task) {
547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
548                 put_task_struct(task);
549         }
550         return allowed;
551 }
552
553 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
554 {
555         int error;
556         struct inode *inode = dentry->d_inode;
557
558         if (attr->ia_valid & ATTR_MODE)
559                 return -EPERM;
560
561         error = inode_change_ok(inode, attr);
562         if (error)
563                 return error;
564
565         if ((attr->ia_valid & ATTR_SIZE) &&
566             attr->ia_size != i_size_read(inode)) {
567                 error = vmtruncate(inode, attr->ia_size);
568                 if (error)
569                         return error;
570         }
571
572         setattr_copy(inode, attr);
573         mark_inode_dirty(inode);
574         return 0;
575 }
576
577 static const struct inode_operations proc_def_inode_operations = {
578         .setattr        = proc_setattr,
579 };
580
581 static int mounts_open_common(struct inode *inode, struct file *file,
582                               const struct seq_operations *op)
583 {
584         struct task_struct *task = get_proc_task(inode);
585         struct nsproxy *nsp;
586         struct mnt_namespace *ns = NULL;
587         struct path root;
588         struct proc_mounts *p;
589         int ret = -EINVAL;
590
591         if (task) {
592                 rcu_read_lock();
593                 nsp = task_nsproxy(task);
594                 if (nsp) {
595                         ns = nsp->mnt_ns;
596                         if (ns)
597                                 get_mnt_ns(ns);
598                 }
599                 rcu_read_unlock();
600                 if (ns && get_task_root(task, &root) == 0)
601                         ret = 0;
602                 put_task_struct(task);
603         }
604
605         if (!ns)
606                 goto err;
607         if (ret)
608                 goto err_put_ns;
609
610         ret = -ENOMEM;
611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
612         if (!p)
613                 goto err_put_path;
614
615         file->private_data = &p->m;
616         ret = seq_open(file, op);
617         if (ret)
618                 goto err_free;
619
620         p->m.private = p;
621         p->ns = ns;
622         p->root = root;
623         p->event = ns->event;
624
625         return 0;
626
627  err_free:
628         kfree(p);
629  err_put_path:
630         path_put(&root);
631  err_put_ns:
632         put_mnt_ns(ns);
633  err:
634         return ret;
635 }
636
637 static int mounts_release(struct inode *inode, struct file *file)
638 {
639         struct proc_mounts *p = file->private_data;
640         path_put(&p->root);
641         put_mnt_ns(p->ns);
642         return seq_release(inode, file);
643 }
644
645 static unsigned mounts_poll(struct file *file, poll_table *wait)
646 {
647         struct proc_mounts *p = file->private_data;
648         unsigned res = POLLIN | POLLRDNORM;
649
650         poll_wait(file, &p->ns->poll, wait);
651         if (mnt_had_events(p))
652                 res |= POLLERR | POLLPRI;
653
654         return res;
655 }
656
657 static int mounts_open(struct inode *inode, struct file *file)
658 {
659         return mounts_open_common(inode, file, &mounts_op);
660 }
661
662 static const struct file_operations proc_mounts_operations = {
663         .open           = mounts_open,
664         .read           = seq_read,
665         .llseek         = seq_lseek,
666         .release        = mounts_release,
667         .poll           = mounts_poll,
668 };
669
670 static int mountinfo_open(struct inode *inode, struct file *file)
671 {
672         return mounts_open_common(inode, file, &mountinfo_op);
673 }
674
675 static const struct file_operations proc_mountinfo_operations = {
676         .open           = mountinfo_open,
677         .read           = seq_read,
678         .llseek         = seq_lseek,
679         .release        = mounts_release,
680         .poll           = mounts_poll,
681 };
682
683 static int mountstats_open(struct inode *inode, struct file *file)
684 {
685         return mounts_open_common(inode, file, &mountstats_op);
686 }
687
688 static const struct file_operations proc_mountstats_operations = {
689         .open           = mountstats_open,
690         .read           = seq_read,
691         .llseek         = seq_lseek,
692         .release        = mounts_release,
693 };
694
695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
696
697 static ssize_t proc_info_read(struct file * file, char __user * buf,
698                           size_t count, loff_t *ppos)
699 {
700         struct inode * inode = file->f_path.dentry->d_inode;
701         unsigned long page;
702         ssize_t length;
703         struct task_struct *task = get_proc_task(inode);
704
705         length = -ESRCH;
706         if (!task)
707                 goto out_no_task;
708
709         if (count > PROC_BLOCK_SIZE)
710                 count = PROC_BLOCK_SIZE;
711
712         length = -ENOMEM;
713         if (!(page = __get_free_page(GFP_TEMPORARY)))
714                 goto out;
715
716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
717
718         if (length >= 0)
719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
720         free_page(page);
721 out:
722         put_task_struct(task);
723 out_no_task:
724         return length;
725 }
726
727 static const struct file_operations proc_info_file_operations = {
728         .read           = proc_info_read,
729         .llseek         = generic_file_llseek,
730 };
731
732 static int proc_single_show(struct seq_file *m, void *v)
733 {
734         struct inode *inode = m->private;
735         struct pid_namespace *ns;
736         struct pid *pid;
737         struct task_struct *task;
738         int ret;
739
740         ns = inode->i_sb->s_fs_info;
741         pid = proc_pid(inode);
742         task = get_pid_task(pid, PIDTYPE_PID);
743         if (!task)
744                 return -ESRCH;
745
746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
747
748         put_task_struct(task);
749         return ret;
750 }
751
752 static int proc_single_open(struct inode *inode, struct file *filp)
753 {
754         int ret;
755         ret = single_open(filp, proc_single_show, NULL);
756         if (!ret) {
757                 struct seq_file *m = filp->private_data;
758
759                 m->private = inode;
760         }
761         return ret;
762 }
763
764 static const struct file_operations proc_single_file_operations = {
765         .open           = proc_single_open,
766         .read           = seq_read,
767         .llseek         = seq_lseek,
768         .release        = single_release,
769 };
770
771 static int mem_open(struct inode* inode, struct file* file)
772 {
773         file->private_data = (void*)((long)current->self_exec_id);
774         /* OK to pass negative loff_t, we can catch out-of-range */
775         file->f_mode |= FMODE_UNSIGNED_OFFSET;
776         return 0;
777 }
778
779 static ssize_t mem_read(struct file * file, char __user * buf,
780                         size_t count, loff_t *ppos)
781 {
782         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
783         char *page;
784         unsigned long src = *ppos;
785         int ret = -ESRCH;
786         struct mm_struct *mm;
787
788         if (!task)
789                 goto out_no_task;
790
791         if (check_mem_permission(task))
792                 goto out;
793
794         ret = -ENOMEM;
795         page = (char *)__get_free_page(GFP_TEMPORARY);
796         if (!page)
797                 goto out;
798
799         ret = 0;
800  
801         mm = get_task_mm(task);
802         if (!mm)
803                 goto out_free;
804
805         ret = -EIO;
806  
807         if (file->private_data != (void*)((long)current->self_exec_id))
808                 goto out_put;
809
810         ret = 0;
811  
812         while (count > 0) {
813                 int this_len, retval;
814
815                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
816                 retval = access_process_vm(task, src, page, this_len, 0);
817                 if (!retval || check_mem_permission(task)) {
818                         if (!ret)
819                                 ret = -EIO;
820                         break;
821                 }
822
823                 if (copy_to_user(buf, page, retval)) {
824                         ret = -EFAULT;
825                         break;
826                 }
827  
828                 ret += retval;
829                 src += retval;
830                 buf += retval;
831                 count -= retval;
832         }
833         *ppos = src;
834
835 out_put:
836         mmput(mm);
837 out_free:
838         free_page((unsigned long) page);
839 out:
840         put_task_struct(task);
841 out_no_task:
842         return ret;
843 }
844
845 #define mem_write NULL
846
847 #ifndef mem_write
848 /* This is a security hazard */
849 static ssize_t mem_write(struct file * file, const char __user *buf,
850                          size_t count, loff_t *ppos)
851 {
852         int copied;
853         char *page;
854         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
855         unsigned long dst = *ppos;
856
857         copied = -ESRCH;
858         if (!task)
859                 goto out_no_task;
860
861         if (check_mem_permission(task))
862                 goto out;
863
864         copied = -ENOMEM;
865         page = (char *)__get_free_page(GFP_TEMPORARY);
866         if (!page)
867                 goto out;
868
869         copied = 0;
870         while (count > 0) {
871                 int this_len, retval;
872
873                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
874                 if (copy_from_user(page, buf, this_len)) {
875                         copied = -EFAULT;
876                         break;
877                 }
878                 retval = access_process_vm(task, dst, page, this_len, 1);
879                 if (!retval) {
880                         if (!copied)
881                                 copied = -EIO;
882                         break;
883                 }
884                 copied += retval;
885                 buf += retval;
886                 dst += retval;
887                 count -= retval;                        
888         }
889         *ppos = dst;
890         free_page((unsigned long) page);
891 out:
892         put_task_struct(task);
893 out_no_task:
894         return copied;
895 }
896 #endif
897
898 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
899 {
900         switch (orig) {
901         case 0:
902                 file->f_pos = offset;
903                 break;
904         case 1:
905                 file->f_pos += offset;
906                 break;
907         default:
908                 return -EINVAL;
909         }
910         force_successful_syscall_return();
911         return file->f_pos;
912 }
913
914 static const struct file_operations proc_mem_operations = {
915         .llseek         = mem_lseek,
916         .read           = mem_read,
917         .write          = mem_write,
918         .open           = mem_open,
919 };
920
921 static ssize_t environ_read(struct file *file, char __user *buf,
922                         size_t count, loff_t *ppos)
923 {
924         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
925         char *page;
926         unsigned long src = *ppos;
927         int ret = -ESRCH;
928         struct mm_struct *mm;
929
930         if (!task)
931                 goto out_no_task;
932
933         if (!ptrace_may_access(task, PTRACE_MODE_READ))
934                 goto out;
935
936         ret = -ENOMEM;
937         page = (char *)__get_free_page(GFP_TEMPORARY);
938         if (!page)
939                 goto out;
940
941         ret = 0;
942
943         mm = get_task_mm(task);
944         if (!mm)
945                 goto out_free;
946
947         while (count > 0) {
948                 int this_len, retval, max_len;
949
950                 this_len = mm->env_end - (mm->env_start + src);
951
952                 if (this_len <= 0)
953                         break;
954
955                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
956                 this_len = (this_len > max_len) ? max_len : this_len;
957
958                 retval = access_process_vm(task, (mm->env_start + src),
959                         page, this_len, 0);
960
961                 if (retval <= 0) {
962                         ret = retval;
963                         break;
964                 }
965
966                 if (copy_to_user(buf, page, retval)) {
967                         ret = -EFAULT;
968                         break;
969                 }
970
971                 ret += retval;
972                 src += retval;
973                 buf += retval;
974                 count -= retval;
975         }
976         *ppos = src;
977
978         mmput(mm);
979 out_free:
980         free_page((unsigned long) page);
981 out:
982         put_task_struct(task);
983 out_no_task:
984         return ret;
985 }
986
987 static const struct file_operations proc_environ_operations = {
988         .read           = environ_read,
989         .llseek         = generic_file_llseek,
990 };
991
992 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
993                                 size_t count, loff_t *ppos)
994 {
995         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
996         char buffer[PROC_NUMBUF];
997         size_t len;
998         int oom_adjust = OOM_DISABLE;
999         unsigned long flags;
1000
1001         if (!task)
1002                 return -ESRCH;
1003
1004         if (lock_task_sighand(task, &flags)) {
1005                 oom_adjust = task->signal->oom_adj;
1006                 unlock_task_sighand(task, &flags);
1007         }
1008
1009         put_task_struct(task);
1010
1011         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1012
1013         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1014 }
1015
1016 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1017                                 size_t count, loff_t *ppos)
1018 {
1019         struct task_struct *task;
1020         char buffer[PROC_NUMBUF];
1021         long oom_adjust;
1022         unsigned long flags;
1023         int err;
1024
1025         memset(buffer, 0, sizeof(buffer));
1026         if (count > sizeof(buffer) - 1)
1027                 count = sizeof(buffer) - 1;
1028         if (copy_from_user(buffer, buf, count)) {
1029                 err = -EFAULT;
1030                 goto out;
1031         }
1032
1033         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1034         if (err)
1035                 goto out;
1036         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1037              oom_adjust != OOM_DISABLE) {
1038                 err = -EINVAL;
1039                 goto out;
1040         }
1041
1042         task = get_proc_task(file->f_path.dentry->d_inode);
1043         if (!task) {
1044                 err = -ESRCH;
1045                 goto out;
1046         }
1047
1048         task_lock(task);
1049         if (!task->mm) {
1050                 err = -EINVAL;
1051                 goto err_task_lock;
1052         }
1053
1054         if (!lock_task_sighand(task, &flags)) {
1055                 err = -ESRCH;
1056                 goto err_task_lock;
1057         }
1058
1059         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1060                 err = -EACCES;
1061                 goto err_sighand;
1062         }
1063
1064         if (oom_adjust != task->signal->oom_adj) {
1065                 if (oom_adjust == OOM_DISABLE)
1066                         atomic_inc(&task->mm->oom_disable_count);
1067                 if (task->signal->oom_adj == OOM_DISABLE)
1068                         atomic_dec(&task->mm->oom_disable_count);
1069         }
1070
1071         /*
1072          * Warn that /proc/pid/oom_adj is deprecated, see
1073          * Documentation/feature-removal-schedule.txt.
1074          */
1075         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1076                         "please use /proc/%d/oom_score_adj instead.\n",
1077                         current->comm, task_pid_nr(current),
1078                         task_pid_nr(task), task_pid_nr(task));
1079         task->signal->oom_adj = oom_adjust;
1080         /*
1081          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1082          * value is always attainable.
1083          */
1084         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1085                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1086         else
1087                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1088                                                                 -OOM_DISABLE;
1089 err_sighand:
1090         unlock_task_sighand(task, &flags);
1091 err_task_lock:
1092         task_unlock(task);
1093         put_task_struct(task);
1094 out:
1095         return err < 0 ? err : count;
1096 }
1097
1098 static const struct file_operations proc_oom_adjust_operations = {
1099         .read           = oom_adjust_read,
1100         .write          = oom_adjust_write,
1101         .llseek         = generic_file_llseek,
1102 };
1103
1104 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1105                                         size_t count, loff_t *ppos)
1106 {
1107         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1108         char buffer[PROC_NUMBUF];
1109         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1110         unsigned long flags;
1111         size_t len;
1112
1113         if (!task)
1114                 return -ESRCH;
1115         if (lock_task_sighand(task, &flags)) {
1116                 oom_score_adj = task->signal->oom_score_adj;
1117                 unlock_task_sighand(task, &flags);
1118         }
1119         put_task_struct(task);
1120         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1121         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1122 }
1123
1124 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1125                                         size_t count, loff_t *ppos)
1126 {
1127         struct task_struct *task;
1128         char buffer[PROC_NUMBUF];
1129         unsigned long flags;
1130         long oom_score_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1142         if (err)
1143                 goto out;
1144         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1145                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         task = get_proc_task(file->f_path.dentry->d_inode);
1151         if (!task) {
1152                 err = -ESRCH;
1153                 goto out;
1154         }
1155
1156         task_lock(task);
1157         if (!task->mm) {
1158                 err = -EINVAL;
1159                 goto err_task_lock;
1160         }
1161
1162         if (!lock_task_sighand(task, &flags)) {
1163                 err = -ESRCH;
1164                 goto err_task_lock;
1165         }
1166
1167         if (oom_score_adj < task->signal->oom_score_adj &&
1168                         !capable(CAP_SYS_RESOURCE)) {
1169                 err = -EACCES;
1170                 goto err_sighand;
1171         }
1172
1173         if (oom_score_adj != task->signal->oom_score_adj) {
1174                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1175                         atomic_inc(&task->mm->oom_disable_count);
1176                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1177                         atomic_dec(&task->mm->oom_disable_count);
1178         }
1179         task->signal->oom_score_adj = oom_score_adj;
1180         /*
1181          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1182          * always attainable.
1183          */
1184         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1185                 task->signal->oom_adj = OOM_DISABLE;
1186         else
1187                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1188                                                         OOM_SCORE_ADJ_MAX;
1189 err_sighand:
1190         unlock_task_sighand(task, &flags);
1191 err_task_lock:
1192         task_unlock(task);
1193         put_task_struct(task);
1194 out:
1195         return err < 0 ? err : count;
1196 }
1197
1198 static const struct file_operations proc_oom_score_adj_operations = {
1199         .read           = oom_score_adj_read,
1200         .write          = oom_score_adj_write,
1201         .llseek         = default_llseek,
1202 };
1203
1204 #ifdef CONFIG_AUDITSYSCALL
1205 #define TMPBUFLEN 21
1206 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1207                                   size_t count, loff_t *ppos)
1208 {
1209         struct inode * inode = file->f_path.dentry->d_inode;
1210         struct task_struct *task = get_proc_task(inode);
1211         ssize_t length;
1212         char tmpbuf[TMPBUFLEN];
1213
1214         if (!task)
1215                 return -ESRCH;
1216         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1217                                 audit_get_loginuid(task));
1218         put_task_struct(task);
1219         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1220 }
1221
1222 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1223                                    size_t count, loff_t *ppos)
1224 {
1225         struct inode * inode = file->f_path.dentry->d_inode;
1226         char *page, *tmp;
1227         ssize_t length;
1228         uid_t loginuid;
1229
1230         if (!capable(CAP_AUDIT_CONTROL))
1231                 return -EPERM;
1232
1233         rcu_read_lock();
1234         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1235                 rcu_read_unlock();
1236                 return -EPERM;
1237         }
1238         rcu_read_unlock();
1239
1240         if (count >= PAGE_SIZE)
1241                 count = PAGE_SIZE - 1;
1242
1243         if (*ppos != 0) {
1244                 /* No partial writes. */
1245                 return -EINVAL;
1246         }
1247         page = (char*)__get_free_page(GFP_TEMPORARY);
1248         if (!page)
1249                 return -ENOMEM;
1250         length = -EFAULT;
1251         if (copy_from_user(page, buf, count))
1252                 goto out_free_page;
1253
1254         page[count] = '\0';
1255         loginuid = simple_strtoul(page, &tmp, 10);
1256         if (tmp == page) {
1257                 length = -EINVAL;
1258                 goto out_free_page;
1259
1260         }
1261         length = audit_set_loginuid(current, loginuid);
1262         if (likely(length == 0))
1263                 length = count;
1264
1265 out_free_page:
1266         free_page((unsigned long) page);
1267         return length;
1268 }
1269
1270 static const struct file_operations proc_loginuid_operations = {
1271         .read           = proc_loginuid_read,
1272         .write          = proc_loginuid_write,
1273         .llseek         = generic_file_llseek,
1274 };
1275
1276 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1277                                   size_t count, loff_t *ppos)
1278 {
1279         struct inode * inode = file->f_path.dentry->d_inode;
1280         struct task_struct *task = get_proc_task(inode);
1281         ssize_t length;
1282         char tmpbuf[TMPBUFLEN];
1283
1284         if (!task)
1285                 return -ESRCH;
1286         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1287                                 audit_get_sessionid(task));
1288         put_task_struct(task);
1289         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1290 }
1291
1292 static const struct file_operations proc_sessionid_operations = {
1293         .read           = proc_sessionid_read,
1294         .llseek         = generic_file_llseek,
1295 };
1296 #endif
1297
1298 #ifdef CONFIG_FAULT_INJECTION
1299 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1300                                       size_t count, loff_t *ppos)
1301 {
1302         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1303         char buffer[PROC_NUMBUF];
1304         size_t len;
1305         int make_it_fail;
1306
1307         if (!task)
1308                 return -ESRCH;
1309         make_it_fail = task->make_it_fail;
1310         put_task_struct(task);
1311
1312         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1313
1314         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1315 }
1316
1317 static ssize_t proc_fault_inject_write(struct file * file,
1318                         const char __user * buf, size_t count, loff_t *ppos)
1319 {
1320         struct task_struct *task;
1321         char buffer[PROC_NUMBUF], *end;
1322         int make_it_fail;
1323
1324         if (!capable(CAP_SYS_RESOURCE))
1325                 return -EPERM;
1326         memset(buffer, 0, sizeof(buffer));
1327         if (count > sizeof(buffer) - 1)
1328                 count = sizeof(buffer) - 1;
1329         if (copy_from_user(buffer, buf, count))
1330                 return -EFAULT;
1331         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1332         if (*end)
1333                 return -EINVAL;
1334         task = get_proc_task(file->f_dentry->d_inode);
1335         if (!task)
1336                 return -ESRCH;
1337         task->make_it_fail = make_it_fail;
1338         put_task_struct(task);
1339
1340         return count;
1341 }
1342
1343 static const struct file_operations proc_fault_inject_operations = {
1344         .read           = proc_fault_inject_read,
1345         .write          = proc_fault_inject_write,
1346         .llseek         = generic_file_llseek,
1347 };
1348 #endif
1349
1350
1351 #ifdef CONFIG_SCHED_DEBUG
1352 /*
1353  * Print out various scheduling related per-task fields:
1354  */
1355 static int sched_show(struct seq_file *m, void *v)
1356 {
1357         struct inode *inode = m->private;
1358         struct task_struct *p;
1359
1360         p = get_proc_task(inode);
1361         if (!p)
1362                 return -ESRCH;
1363         proc_sched_show_task(p, m);
1364
1365         put_task_struct(p);
1366
1367         return 0;
1368 }
1369
1370 static ssize_t
1371 sched_write(struct file *file, const char __user *buf,
1372             size_t count, loff_t *offset)
1373 {
1374         struct inode *inode = file->f_path.dentry->d_inode;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380         proc_sched_set_task(p);
1381
1382         put_task_struct(p);
1383
1384         return count;
1385 }
1386
1387 static int sched_open(struct inode *inode, struct file *filp)
1388 {
1389         int ret;
1390
1391         ret = single_open(filp, sched_show, NULL);
1392         if (!ret) {
1393                 struct seq_file *m = filp->private_data;
1394
1395                 m->private = inode;
1396         }
1397         return ret;
1398 }
1399
1400 static const struct file_operations proc_pid_sched_operations = {
1401         .open           = sched_open,
1402         .read           = seq_read,
1403         .write          = sched_write,
1404         .llseek         = seq_lseek,
1405         .release        = single_release,
1406 };
1407
1408 #endif
1409
1410 #ifdef CONFIG_SCHED_AUTOGROUP
1411 /*
1412  * Print out autogroup related information:
1413  */
1414 static int sched_autogroup_show(struct seq_file *m, void *v)
1415 {
1416         struct inode *inode = m->private;
1417         struct task_struct *p;
1418
1419         p = get_proc_task(inode);
1420         if (!p)
1421                 return -ESRCH;
1422         proc_sched_autogroup_show_task(p, m);
1423
1424         put_task_struct(p);
1425
1426         return 0;
1427 }
1428
1429 static ssize_t
1430 sched_autogroup_write(struct file *file, const char __user *buf,
1431             size_t count, loff_t *offset)
1432 {
1433         struct inode *inode = file->f_path.dentry->d_inode;
1434         struct task_struct *p;
1435         char buffer[PROC_NUMBUF];
1436         long nice;
1437         int err;
1438
1439         memset(buffer, 0, sizeof(buffer));
1440         if (count > sizeof(buffer) - 1)
1441                 count = sizeof(buffer) - 1;
1442         if (copy_from_user(buffer, buf, count))
1443                 return -EFAULT;
1444
1445         err = strict_strtol(strstrip(buffer), 0, &nice);
1446         if (err)
1447                 return -EINVAL;
1448
1449         p = get_proc_task(inode);
1450         if (!p)
1451                 return -ESRCH;
1452
1453         err = nice;
1454         err = proc_sched_autogroup_set_nice(p, &err);
1455         if (err)
1456                 count = err;
1457
1458         put_task_struct(p);
1459
1460         return count;
1461 }
1462
1463 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1464 {
1465         int ret;
1466
1467         ret = single_open(filp, sched_autogroup_show, NULL);
1468         if (!ret) {
1469                 struct seq_file *m = filp->private_data;
1470
1471                 m->private = inode;
1472         }
1473         return ret;
1474 }
1475
1476 static const struct file_operations proc_pid_sched_autogroup_operations = {
1477         .open           = sched_autogroup_open,
1478         .read           = seq_read,
1479         .write          = sched_autogroup_write,
1480         .llseek         = seq_lseek,
1481         .release        = single_release,
1482 };
1483
1484 #endif /* CONFIG_SCHED_AUTOGROUP */
1485
1486 static ssize_t comm_write(struct file *file, const char __user *buf,
1487                                 size_t count, loff_t *offset)
1488 {
1489         struct inode *inode = file->f_path.dentry->d_inode;
1490         struct task_struct *p;
1491         char buffer[TASK_COMM_LEN];
1492
1493         memset(buffer, 0, sizeof(buffer));
1494         if (count > sizeof(buffer) - 1)
1495                 count = sizeof(buffer) - 1;
1496         if (copy_from_user(buffer, buf, count))
1497                 return -EFAULT;
1498
1499         p = get_proc_task(inode);
1500         if (!p)
1501                 return -ESRCH;
1502
1503         if (same_thread_group(current, p))
1504                 set_task_comm(p, buffer);
1505         else
1506                 count = -EINVAL;
1507
1508         put_task_struct(p);
1509
1510         return count;
1511 }
1512
1513 static int comm_show(struct seq_file *m, void *v)
1514 {
1515         struct inode *inode = m->private;
1516         struct task_struct *p;
1517
1518         p = get_proc_task(inode);
1519         if (!p)
1520                 return -ESRCH;
1521
1522         task_lock(p);
1523         seq_printf(m, "%s\n", p->comm);
1524         task_unlock(p);
1525
1526         put_task_struct(p);
1527
1528         return 0;
1529 }
1530
1531 static int comm_open(struct inode *inode, struct file *filp)
1532 {
1533         int ret;
1534
1535         ret = single_open(filp, comm_show, NULL);
1536         if (!ret) {
1537                 struct seq_file *m = filp->private_data;
1538
1539                 m->private = inode;
1540         }
1541         return ret;
1542 }
1543
1544 static const struct file_operations proc_pid_set_comm_operations = {
1545         .open           = comm_open,
1546         .read           = seq_read,
1547         .write          = comm_write,
1548         .llseek         = seq_lseek,
1549         .release        = single_release,
1550 };
1551
1552 /*
1553  * We added or removed a vma mapping the executable. The vmas are only mapped
1554  * during exec and are not mapped with the mmap system call.
1555  * Callers must hold down_write() on the mm's mmap_sem for these
1556  */
1557 void added_exe_file_vma(struct mm_struct *mm)
1558 {
1559         mm->num_exe_file_vmas++;
1560 }
1561
1562 void removed_exe_file_vma(struct mm_struct *mm)
1563 {
1564         mm->num_exe_file_vmas--;
1565         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1566                 fput(mm->exe_file);
1567                 mm->exe_file = NULL;
1568         }
1569
1570 }
1571
1572 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1573 {
1574         if (new_exe_file)
1575                 get_file(new_exe_file);
1576         if (mm->exe_file)
1577                 fput(mm->exe_file);
1578         mm->exe_file = new_exe_file;
1579         mm->num_exe_file_vmas = 0;
1580 }
1581
1582 struct file *get_mm_exe_file(struct mm_struct *mm)
1583 {
1584         struct file *exe_file;
1585
1586         /* We need mmap_sem to protect against races with removal of
1587          * VM_EXECUTABLE vmas */
1588         down_read(&mm->mmap_sem);
1589         exe_file = mm->exe_file;
1590         if (exe_file)
1591                 get_file(exe_file);
1592         up_read(&mm->mmap_sem);
1593         return exe_file;
1594 }
1595
1596 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1597 {
1598         /* It's safe to write the exe_file pointer without exe_file_lock because
1599          * this is called during fork when the task is not yet in /proc */
1600         newmm->exe_file = get_mm_exe_file(oldmm);
1601 }
1602
1603 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1604 {
1605         struct task_struct *task;
1606         struct mm_struct *mm;
1607         struct file *exe_file;
1608
1609         task = get_proc_task(inode);
1610         if (!task)
1611                 return -ENOENT;
1612         mm = get_task_mm(task);
1613         put_task_struct(task);
1614         if (!mm)
1615                 return -ENOENT;
1616         exe_file = get_mm_exe_file(mm);
1617         mmput(mm);
1618         if (exe_file) {
1619                 *exe_path = exe_file->f_path;
1620                 path_get(&exe_file->f_path);
1621                 fput(exe_file);
1622                 return 0;
1623         } else
1624                 return -ENOENT;
1625 }
1626
1627 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1628 {
1629         struct inode *inode = dentry->d_inode;
1630         int error = -EACCES;
1631
1632         /* We don't need a base pointer in the /proc filesystem */
1633         path_put(&nd->path);
1634
1635         /* Are we allowed to snoop on the tasks file descriptors? */
1636         if (!proc_fd_access_allowed(inode))
1637                 goto out;
1638
1639         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1640 out:
1641         return ERR_PTR(error);
1642 }
1643
1644 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1645 {
1646         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1647         char *pathname;
1648         int len;
1649
1650         if (!tmp)
1651                 return -ENOMEM;
1652
1653         pathname = d_path_with_unreachable(path, tmp, PAGE_SIZE);
1654         len = PTR_ERR(pathname);
1655         if (IS_ERR(pathname))
1656                 goto out;
1657         len = tmp + PAGE_SIZE - 1 - pathname;
1658
1659         if (len > buflen)
1660                 len = buflen;
1661         if (copy_to_user(buffer, pathname, len))
1662                 len = -EFAULT;
1663  out:
1664         free_page((unsigned long)tmp);
1665         return len;
1666 }
1667
1668 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1669 {
1670         int error = -EACCES;
1671         struct inode *inode = dentry->d_inode;
1672         struct path path;
1673
1674         /* Are we allowed to snoop on the tasks file descriptors? */
1675         if (!proc_fd_access_allowed(inode))
1676                 goto out;
1677
1678         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1679         if (error)
1680                 goto out;
1681
1682         error = do_proc_readlink(&path, buffer, buflen);
1683         path_put(&path);
1684 out:
1685         return error;
1686 }
1687
1688 static const struct inode_operations proc_pid_link_inode_operations = {
1689         .readlink       = proc_pid_readlink,
1690         .follow_link    = proc_pid_follow_link,
1691         .setattr        = proc_setattr,
1692 };
1693
1694
1695 /* building an inode */
1696
1697 static int task_dumpable(struct task_struct *task)
1698 {
1699         int dumpable = 0;
1700         struct mm_struct *mm;
1701
1702         task_lock(task);
1703         mm = task->mm;
1704         if (mm)
1705                 dumpable = get_dumpable(mm);
1706         task_unlock(task);
1707         if(dumpable == 1)
1708                 return 1;
1709         return 0;
1710 }
1711
1712
1713 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1714 {
1715         struct inode * inode;
1716         struct proc_inode *ei;
1717         const struct cred *cred;
1718
1719         /* We need a new inode */
1720
1721         inode = new_inode(sb);
1722         if (!inode)
1723                 goto out;
1724
1725         /* Common stuff */
1726         ei = PROC_I(inode);
1727         inode->i_ino = get_next_ino();
1728         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1729         inode->i_op = &proc_def_inode_operations;
1730
1731         /*
1732          * grab the reference to task.
1733          */
1734         ei->pid = get_task_pid(task, PIDTYPE_PID);
1735         if (!ei->pid)
1736                 goto out_unlock;
1737
1738         if (task_dumpable(task)) {
1739                 rcu_read_lock();
1740                 cred = __task_cred(task);
1741                 inode->i_uid = cred->euid;
1742                 inode->i_gid = cred->egid;
1743                 rcu_read_unlock();
1744         }
1745         security_task_to_inode(task, inode);
1746
1747 out:
1748         return inode;
1749
1750 out_unlock:
1751         iput(inode);
1752         return NULL;
1753 }
1754
1755 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1756 {
1757         struct inode *inode = dentry->d_inode;
1758         struct task_struct *task;
1759         const struct cred *cred;
1760
1761         generic_fillattr(inode, stat);
1762
1763         rcu_read_lock();
1764         stat->uid = 0;
1765         stat->gid = 0;
1766         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1767         if (task) {
1768                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1769                     task_dumpable(task)) {
1770                         cred = __task_cred(task);
1771                         stat->uid = cred->euid;
1772                         stat->gid = cred->egid;
1773                 }
1774         }
1775         rcu_read_unlock();
1776         return 0;
1777 }
1778
1779 /* dentry stuff */
1780
1781 /*
1782  *      Exceptional case: normally we are not allowed to unhash a busy
1783  * directory. In this case, however, we can do it - no aliasing problems
1784  * due to the way we treat inodes.
1785  *
1786  * Rewrite the inode's ownerships here because the owning task may have
1787  * performed a setuid(), etc.
1788  *
1789  * Before the /proc/pid/status file was created the only way to read
1790  * the effective uid of a /process was to stat /proc/pid.  Reading
1791  * /proc/pid/status is slow enough that procps and other packages
1792  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1793  * made this apply to all per process world readable and executable
1794  * directories.
1795  */
1796 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1797 {
1798         struct inode *inode = dentry->d_inode;
1799         struct task_struct *task = get_proc_task(inode);
1800         const struct cred *cred;
1801
1802         if (task) {
1803                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1804                     task_dumpable(task)) {
1805                         rcu_read_lock();
1806                         cred = __task_cred(task);
1807                         inode->i_uid = cred->euid;
1808                         inode->i_gid = cred->egid;
1809                         rcu_read_unlock();
1810                 } else {
1811                         inode->i_uid = 0;
1812                         inode->i_gid = 0;
1813                 }
1814                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1815                 security_task_to_inode(task, inode);
1816                 put_task_struct(task);
1817                 return 1;
1818         }
1819         d_drop(dentry);
1820         return 0;
1821 }
1822
1823 static int pid_delete_dentry(struct dentry * dentry)
1824 {
1825         /* Is the task we represent dead?
1826          * If so, then don't put the dentry on the lru list,
1827          * kill it immediately.
1828          */
1829         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1830 }
1831
1832 static const struct dentry_operations pid_dentry_operations =
1833 {
1834         .d_revalidate   = pid_revalidate,
1835         .d_delete       = pid_delete_dentry,
1836 };
1837
1838 /* Lookups */
1839
1840 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1841                                 struct task_struct *, const void *);
1842
1843 /*
1844  * Fill a directory entry.
1845  *
1846  * If possible create the dcache entry and derive our inode number and
1847  * file type from dcache entry.
1848  *
1849  * Since all of the proc inode numbers are dynamically generated, the inode
1850  * numbers do not exist until the inode is cache.  This means creating the
1851  * the dcache entry in readdir is necessary to keep the inode numbers
1852  * reported by readdir in sync with the inode numbers reported
1853  * by stat.
1854  */
1855 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1856         char *name, int len,
1857         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1858 {
1859         struct dentry *child, *dir = filp->f_path.dentry;
1860         struct inode *inode;
1861         struct qstr qname;
1862         ino_t ino = 0;
1863         unsigned type = DT_UNKNOWN;
1864
1865         qname.name = name;
1866         qname.len  = len;
1867         qname.hash = full_name_hash(name, len);
1868
1869         child = d_lookup(dir, &qname);
1870         if (!child) {
1871                 struct dentry *new;
1872                 new = d_alloc(dir, &qname);
1873                 if (new) {
1874                         child = instantiate(dir->d_inode, new, task, ptr);
1875                         if (child)
1876                                 dput(new);
1877                         else
1878                                 child = new;
1879                 }
1880         }
1881         if (!child || IS_ERR(child) || !child->d_inode)
1882                 goto end_instantiate;
1883         inode = child->d_inode;
1884         if (inode) {
1885                 ino = inode->i_ino;
1886                 type = inode->i_mode >> 12;
1887         }
1888         dput(child);
1889 end_instantiate:
1890         if (!ino)
1891                 ino = find_inode_number(dir, &qname);
1892         if (!ino)
1893                 ino = 1;
1894         return filldir(dirent, name, len, filp->f_pos, ino, type);
1895 }
1896
1897 static unsigned name_to_int(struct dentry *dentry)
1898 {
1899         const char *name = dentry->d_name.name;
1900         int len = dentry->d_name.len;
1901         unsigned n = 0;
1902
1903         if (len > 1 && *name == '0')
1904                 goto out;
1905         while (len-- > 0) {
1906                 unsigned c = *name++ - '0';
1907                 if (c > 9)
1908                         goto out;
1909                 if (n >= (~0U-9)/10)
1910                         goto out;
1911                 n *= 10;
1912                 n += c;
1913         }
1914         return n;
1915 out:
1916         return ~0U;
1917 }
1918
1919 #define PROC_FDINFO_MAX 64
1920
1921 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1922 {
1923         struct task_struct *task = get_proc_task(inode);
1924         struct files_struct *files = NULL;
1925         struct file *file;
1926         int fd = proc_fd(inode);
1927
1928         if (task) {
1929                 files = get_files_struct(task);
1930                 put_task_struct(task);
1931         }
1932         if (files) {
1933                 /*
1934                  * We are not taking a ref to the file structure, so we must
1935                  * hold ->file_lock.
1936                  */
1937                 spin_lock(&files->file_lock);
1938                 file = fcheck_files(files, fd);
1939                 if (file) {
1940                         if (path) {
1941                                 *path = file->f_path;
1942                                 path_get(&file->f_path);
1943                         }
1944                         if (info)
1945                                 snprintf(info, PROC_FDINFO_MAX,
1946                                          "pos:\t%lli\n"
1947                                          "flags:\t0%o\n",
1948                                          (long long) file->f_pos,
1949                                          file->f_flags);
1950                         spin_unlock(&files->file_lock);
1951                         put_files_struct(files);
1952                         return 0;
1953                 }
1954                 spin_unlock(&files->file_lock);
1955                 put_files_struct(files);
1956         }
1957         return -ENOENT;
1958 }
1959
1960 static int proc_fd_link(struct inode *inode, struct path *path)
1961 {
1962         return proc_fd_info(inode, path, NULL);
1963 }
1964
1965 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1966 {
1967         struct inode *inode = dentry->d_inode;
1968         struct task_struct *task = get_proc_task(inode);
1969         int fd = proc_fd(inode);
1970         struct files_struct *files;
1971         const struct cred *cred;
1972
1973         if (task) {
1974                 files = get_files_struct(task);
1975                 if (files) {
1976                         rcu_read_lock();
1977                         if (fcheck_files(files, fd)) {
1978                                 rcu_read_unlock();
1979                                 put_files_struct(files);
1980                                 if (task_dumpable(task)) {
1981                                         rcu_read_lock();
1982                                         cred = __task_cred(task);
1983                                         inode->i_uid = cred->euid;
1984                                         inode->i_gid = cred->egid;
1985                                         rcu_read_unlock();
1986                                 } else {
1987                                         inode->i_uid = 0;
1988                                         inode->i_gid = 0;
1989                                 }
1990                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1991                                 security_task_to_inode(task, inode);
1992                                 put_task_struct(task);
1993                                 return 1;
1994                         }
1995                         rcu_read_unlock();
1996                         put_files_struct(files);
1997                 }
1998                 put_task_struct(task);
1999         }
2000         d_drop(dentry);
2001         return 0;
2002 }
2003
2004 static const struct dentry_operations tid_fd_dentry_operations =
2005 {
2006         .d_revalidate   = tid_fd_revalidate,
2007         .d_delete       = pid_delete_dentry,
2008 };
2009
2010 static struct dentry *proc_fd_instantiate(struct inode *dir,
2011         struct dentry *dentry, struct task_struct *task, const void *ptr)
2012 {
2013         unsigned fd = *(const unsigned *)ptr;
2014         struct file *file;
2015         struct files_struct *files;
2016         struct inode *inode;
2017         struct proc_inode *ei;
2018         struct dentry *error = ERR_PTR(-ENOENT);
2019
2020         inode = proc_pid_make_inode(dir->i_sb, task);
2021         if (!inode)
2022                 goto out;
2023         ei = PROC_I(inode);
2024         ei->fd = fd;
2025         files = get_files_struct(task);
2026         if (!files)
2027                 goto out_iput;
2028         inode->i_mode = S_IFLNK;
2029
2030         /*
2031          * We are not taking a ref to the file structure, so we must
2032          * hold ->file_lock.
2033          */
2034         spin_lock(&files->file_lock);
2035         file = fcheck_files(files, fd);
2036         if (!file)
2037                 goto out_unlock;
2038         if (file->f_mode & FMODE_READ)
2039                 inode->i_mode |= S_IRUSR | S_IXUSR;
2040         if (file->f_mode & FMODE_WRITE)
2041                 inode->i_mode |= S_IWUSR | S_IXUSR;
2042         spin_unlock(&files->file_lock);
2043         put_files_struct(files);
2044
2045         inode->i_op = &proc_pid_link_inode_operations;
2046         inode->i_size = 64;
2047         ei->op.proc_get_link = proc_fd_link;
2048         dentry->d_op = &tid_fd_dentry_operations;
2049         d_add(dentry, inode);
2050         /* Close the race of the process dying before we return the dentry */
2051         if (tid_fd_revalidate(dentry, NULL))
2052                 error = NULL;
2053
2054  out:
2055         return error;
2056 out_unlock:
2057         spin_unlock(&files->file_lock);
2058         put_files_struct(files);
2059 out_iput:
2060         iput(inode);
2061         goto out;
2062 }
2063
2064 static struct dentry *proc_lookupfd_common(struct inode *dir,
2065                                            struct dentry *dentry,
2066                                            instantiate_t instantiate)
2067 {
2068         struct task_struct *task = get_proc_task(dir);
2069         unsigned fd = name_to_int(dentry);
2070         struct dentry *result = ERR_PTR(-ENOENT);
2071
2072         if (!task)
2073                 goto out_no_task;
2074         if (fd == ~0U)
2075                 goto out;
2076
2077         result = instantiate(dir, dentry, task, &fd);
2078 out:
2079         put_task_struct(task);
2080 out_no_task:
2081         return result;
2082 }
2083
2084 static int proc_readfd_common(struct file * filp, void * dirent,
2085                               filldir_t filldir, instantiate_t instantiate)
2086 {
2087         struct dentry *dentry = filp->f_path.dentry;
2088         struct inode *inode = dentry->d_inode;
2089         struct task_struct *p = get_proc_task(inode);
2090         unsigned int fd, ino;
2091         int retval;
2092         struct files_struct * files;
2093
2094         retval = -ENOENT;
2095         if (!p)
2096                 goto out_no_task;
2097         retval = 0;
2098
2099         fd = filp->f_pos;
2100         switch (fd) {
2101                 case 0:
2102                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2103                                 goto out;
2104                         filp->f_pos++;
2105                 case 1:
2106                         ino = parent_ino(dentry);
2107                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2108                                 goto out;
2109                         filp->f_pos++;
2110                 default:
2111                         files = get_files_struct(p);
2112                         if (!files)
2113                                 goto out;
2114                         rcu_read_lock();
2115                         for (fd = filp->f_pos-2;
2116                              fd < files_fdtable(files)->max_fds;
2117                              fd++, filp->f_pos++) {
2118                                 char name[PROC_NUMBUF];
2119                                 int len;
2120
2121                                 if (!fcheck_files(files, fd))
2122                                         continue;
2123                                 rcu_read_unlock();
2124
2125                                 len = snprintf(name, sizeof(name), "%d", fd);
2126                                 if (proc_fill_cache(filp, dirent, filldir,
2127                                                     name, len, instantiate,
2128                                                     p, &fd) < 0) {
2129                                         rcu_read_lock();
2130                                         break;
2131                                 }
2132                                 rcu_read_lock();
2133                         }
2134                         rcu_read_unlock();
2135                         put_files_struct(files);
2136         }
2137 out:
2138         put_task_struct(p);
2139 out_no_task:
2140         return retval;
2141 }
2142
2143 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2144                                     struct nameidata *nd)
2145 {
2146         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2147 }
2148
2149 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2150 {
2151         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2152 }
2153
2154 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2155                                       size_t len, loff_t *ppos)
2156 {
2157         char tmp[PROC_FDINFO_MAX];
2158         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2159         if (!err)
2160                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2161         return err;
2162 }
2163
2164 static const struct file_operations proc_fdinfo_file_operations = {
2165         .open           = nonseekable_open,
2166         .read           = proc_fdinfo_read,
2167         .llseek         = no_llseek,
2168 };
2169
2170 static const struct file_operations proc_fd_operations = {
2171         .read           = generic_read_dir,
2172         .readdir        = proc_readfd,
2173         .llseek         = default_llseek,
2174 };
2175
2176 /*
2177  * /proc/pid/fd needs a special permission handler so that a process can still
2178  * access /proc/self/fd after it has executed a setuid().
2179  */
2180 static int proc_fd_permission(struct inode *inode, int mask)
2181 {
2182         int rv;
2183
2184         rv = generic_permission(inode, mask, NULL);
2185         if (rv == 0)
2186                 return 0;
2187         if (task_pid(current) == proc_pid(inode))
2188                 rv = 0;
2189         return rv;
2190 }
2191
2192 /*
2193  * proc directories can do almost nothing..
2194  */
2195 static const struct inode_operations proc_fd_inode_operations = {
2196         .lookup         = proc_lookupfd,
2197         .permission     = proc_fd_permission,
2198         .setattr        = proc_setattr,
2199 };
2200
2201 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2202         struct dentry *dentry, struct task_struct *task, const void *ptr)
2203 {
2204         unsigned fd = *(unsigned *)ptr;
2205         struct inode *inode;
2206         struct proc_inode *ei;
2207         struct dentry *error = ERR_PTR(-ENOENT);
2208
2209         inode = proc_pid_make_inode(dir->i_sb, task);
2210         if (!inode)
2211                 goto out;
2212         ei = PROC_I(inode);
2213         ei->fd = fd;
2214         inode->i_mode = S_IFREG | S_IRUSR;
2215         inode->i_fop = &proc_fdinfo_file_operations;
2216         dentry->d_op = &tid_fd_dentry_operations;
2217         d_add(dentry, inode);
2218         /* Close the race of the process dying before we return the dentry */
2219         if (tid_fd_revalidate(dentry, NULL))
2220                 error = NULL;
2221
2222  out:
2223         return error;
2224 }
2225
2226 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2227                                         struct dentry *dentry,
2228                                         struct nameidata *nd)
2229 {
2230         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2231 }
2232
2233 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2234 {
2235         return proc_readfd_common(filp, dirent, filldir,
2236                                   proc_fdinfo_instantiate);
2237 }
2238
2239 static const struct file_operations proc_fdinfo_operations = {
2240         .read           = generic_read_dir,
2241         .readdir        = proc_readfdinfo,
2242         .llseek         = default_llseek,
2243 };
2244
2245 /*
2246  * proc directories can do almost nothing..
2247  */
2248 static const struct inode_operations proc_fdinfo_inode_operations = {
2249         .lookup         = proc_lookupfdinfo,
2250         .setattr        = proc_setattr,
2251 };
2252
2253
2254 static struct dentry *proc_pident_instantiate(struct inode *dir,
2255         struct dentry *dentry, struct task_struct *task, const void *ptr)
2256 {
2257         const struct pid_entry *p = ptr;
2258         struct inode *inode;
2259         struct proc_inode *ei;
2260         struct dentry *error = ERR_PTR(-ENOENT);
2261
2262         inode = proc_pid_make_inode(dir->i_sb, task);
2263         if (!inode)
2264                 goto out;
2265
2266         ei = PROC_I(inode);
2267         inode->i_mode = p->mode;
2268         if (S_ISDIR(inode->i_mode))
2269                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2270         if (p->iop)
2271                 inode->i_op = p->iop;
2272         if (p->fop)
2273                 inode->i_fop = p->fop;
2274         ei->op = p->op;
2275         dentry->d_op = &pid_dentry_operations;
2276         d_add(dentry, inode);
2277         /* Close the race of the process dying before we return the dentry */
2278         if (pid_revalidate(dentry, NULL))
2279                 error = NULL;
2280 out:
2281         return error;
2282 }
2283
2284 static struct dentry *proc_pident_lookup(struct inode *dir, 
2285                                          struct dentry *dentry,
2286                                          const struct pid_entry *ents,
2287                                          unsigned int nents)
2288 {
2289         struct dentry *error;
2290         struct task_struct *task = get_proc_task(dir);
2291         const struct pid_entry *p, *last;
2292
2293         error = ERR_PTR(-ENOENT);
2294
2295         if (!task)
2296                 goto out_no_task;
2297
2298         /*
2299          * Yes, it does not scale. And it should not. Don't add
2300          * new entries into /proc/<tgid>/ without very good reasons.
2301          */
2302         last = &ents[nents - 1];
2303         for (p = ents; p <= last; p++) {
2304                 if (p->len != dentry->d_name.len)
2305                         continue;
2306                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2307                         break;
2308         }
2309         if (p > last)
2310                 goto out;
2311
2312         error = proc_pident_instantiate(dir, dentry, task, p);
2313 out:
2314         put_task_struct(task);
2315 out_no_task:
2316         return error;
2317 }
2318
2319 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2320         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2321 {
2322         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2323                                 proc_pident_instantiate, task, p);
2324 }
2325
2326 static int proc_pident_readdir(struct file *filp,
2327                 void *dirent, filldir_t filldir,
2328                 const struct pid_entry *ents, unsigned int nents)
2329 {
2330         int i;
2331         struct dentry *dentry = filp->f_path.dentry;
2332         struct inode *inode = dentry->d_inode;
2333         struct task_struct *task = get_proc_task(inode);
2334         const struct pid_entry *p, *last;
2335         ino_t ino;
2336         int ret;
2337
2338         ret = -ENOENT;
2339         if (!task)
2340                 goto out_no_task;
2341
2342         ret = 0;
2343         i = filp->f_pos;
2344         switch (i) {
2345         case 0:
2346                 ino = inode->i_ino;
2347                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2348                         goto out;
2349                 i++;
2350                 filp->f_pos++;
2351                 /* fall through */
2352         case 1:
2353                 ino = parent_ino(dentry);
2354                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2355                         goto out;
2356                 i++;
2357                 filp->f_pos++;
2358                 /* fall through */
2359         default:
2360                 i -= 2;
2361                 if (i >= nents) {
2362                         ret = 1;
2363                         goto out;
2364                 }
2365                 p = ents + i;
2366                 last = &ents[nents - 1];
2367                 while (p <= last) {
2368                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2369                                 goto out;
2370                         filp->f_pos++;
2371                         p++;
2372                 }
2373         }
2374
2375         ret = 1;
2376 out:
2377         put_task_struct(task);
2378 out_no_task:
2379         return ret;
2380 }
2381
2382 #ifdef CONFIG_SECURITY
2383 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2384                                   size_t count, loff_t *ppos)
2385 {
2386         struct inode * inode = file->f_path.dentry->d_inode;
2387         char *p = NULL;
2388         ssize_t length;
2389         struct task_struct *task = get_proc_task(inode);
2390
2391         if (!task)
2392                 return -ESRCH;
2393
2394         length = security_getprocattr(task,
2395                                       (char*)file->f_path.dentry->d_name.name,
2396                                       &p);
2397         put_task_struct(task);
2398         if (length > 0)
2399                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2400         kfree(p);
2401         return length;
2402 }
2403
2404 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2405                                    size_t count, loff_t *ppos)
2406 {
2407         struct inode * inode = file->f_path.dentry->d_inode;
2408         char *page;
2409         ssize_t length;
2410         struct task_struct *task = get_proc_task(inode);
2411
2412         length = -ESRCH;
2413         if (!task)
2414                 goto out_no_task;
2415         if (count > PAGE_SIZE)
2416                 count = PAGE_SIZE;
2417
2418         /* No partial writes. */
2419         length = -EINVAL;
2420         if (*ppos != 0)
2421                 goto out;
2422
2423         length = -ENOMEM;
2424         page = (char*)__get_free_page(GFP_TEMPORARY);
2425         if (!page)
2426                 goto out;
2427
2428         length = -EFAULT;
2429         if (copy_from_user(page, buf, count))
2430                 goto out_free;
2431
2432         /* Guard against adverse ptrace interaction */
2433         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2434         if (length < 0)
2435                 goto out_free;
2436
2437         length = security_setprocattr(task,
2438                                       (char*)file->f_path.dentry->d_name.name,
2439                                       (void*)page, count);
2440         mutex_unlock(&task->signal->cred_guard_mutex);
2441 out_free:
2442         free_page((unsigned long) page);
2443 out:
2444         put_task_struct(task);
2445 out_no_task:
2446         return length;
2447 }
2448
2449 static const struct file_operations proc_pid_attr_operations = {
2450         .read           = proc_pid_attr_read,
2451         .write          = proc_pid_attr_write,
2452         .llseek         = generic_file_llseek,
2453 };
2454
2455 static const struct pid_entry attr_dir_stuff[] = {
2456         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2457         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2458         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2459         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2460         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2461         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2462 };
2463
2464 static int proc_attr_dir_readdir(struct file * filp,
2465                              void * dirent, filldir_t filldir)
2466 {
2467         return proc_pident_readdir(filp,dirent,filldir,
2468                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2469 }
2470
2471 static const struct file_operations proc_attr_dir_operations = {
2472         .read           = generic_read_dir,
2473         .readdir        = proc_attr_dir_readdir,
2474         .llseek         = default_llseek,
2475 };
2476
2477 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2478                                 struct dentry *dentry, struct nameidata *nd)
2479 {
2480         return proc_pident_lookup(dir, dentry,
2481                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2482 }
2483
2484 static const struct inode_operations proc_attr_dir_inode_operations = {
2485         .lookup         = proc_attr_dir_lookup,
2486         .getattr        = pid_getattr,
2487         .setattr        = proc_setattr,
2488 };
2489
2490 #endif
2491
2492 #ifdef CONFIG_ELF_CORE
2493 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2494                                          size_t count, loff_t *ppos)
2495 {
2496         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2497         struct mm_struct *mm;
2498         char buffer[PROC_NUMBUF];
2499         size_t len;
2500         int ret;
2501
2502         if (!task)
2503                 return -ESRCH;
2504
2505         ret = 0;
2506         mm = get_task_mm(task);
2507         if (mm) {
2508                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2509                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2510                                 MMF_DUMP_FILTER_SHIFT));
2511                 mmput(mm);
2512                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2513         }
2514
2515         put_task_struct(task);
2516
2517         return ret;
2518 }
2519
2520 static ssize_t proc_coredump_filter_write(struct file *file,
2521                                           const char __user *buf,
2522                                           size_t count,
2523                                           loff_t *ppos)
2524 {
2525         struct task_struct *task;
2526         struct mm_struct *mm;
2527         char buffer[PROC_NUMBUF], *end;
2528         unsigned int val;
2529         int ret;
2530         int i;
2531         unsigned long mask;
2532
2533         ret = -EFAULT;
2534         memset(buffer, 0, sizeof(buffer));
2535         if (count > sizeof(buffer) - 1)
2536                 count = sizeof(buffer) - 1;
2537         if (copy_from_user(buffer, buf, count))
2538                 goto out_no_task;
2539
2540         ret = -EINVAL;
2541         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2542         if (*end == '\n')
2543                 end++;
2544         if (end - buffer == 0)
2545                 goto out_no_task;
2546
2547         ret = -ESRCH;
2548         task = get_proc_task(file->f_dentry->d_inode);
2549         if (!task)
2550                 goto out_no_task;
2551
2552         ret = end - buffer;
2553         mm = get_task_mm(task);
2554         if (!mm)
2555                 goto out_no_mm;
2556
2557         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2558                 if (val & mask)
2559                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2560                 else
2561                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2562         }
2563
2564         mmput(mm);
2565  out_no_mm:
2566         put_task_struct(task);
2567  out_no_task:
2568         return ret;
2569 }
2570
2571 static const struct file_operations proc_coredump_filter_operations = {
2572         .read           = proc_coredump_filter_read,
2573         .write          = proc_coredump_filter_write,
2574         .llseek         = generic_file_llseek,
2575 };
2576 #endif
2577
2578 /*
2579  * /proc/self:
2580  */
2581 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2582                               int buflen)
2583 {
2584         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2585         pid_t tgid = task_tgid_nr_ns(current, ns);
2586         char tmp[PROC_NUMBUF];
2587         if (!tgid)
2588                 return -ENOENT;
2589         sprintf(tmp, "%d", tgid);
2590         return vfs_readlink(dentry,buffer,buflen,tmp);
2591 }
2592
2593 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2594 {
2595         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2596         pid_t tgid = task_tgid_nr_ns(current, ns);
2597         char *name = ERR_PTR(-ENOENT);
2598         if (tgid) {
2599                 name = __getname();
2600                 if (!name)
2601                         name = ERR_PTR(-ENOMEM);
2602                 else
2603                         sprintf(name, "%d", tgid);
2604         }
2605         nd_set_link(nd, name);
2606         return NULL;
2607 }
2608
2609 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2610                                 void *cookie)
2611 {
2612         char *s = nd_get_link(nd);
2613         if (!IS_ERR(s))
2614                 __putname(s);
2615 }
2616
2617 static const struct inode_operations proc_self_inode_operations = {
2618         .readlink       = proc_self_readlink,
2619         .follow_link    = proc_self_follow_link,
2620         .put_link       = proc_self_put_link,
2621 };
2622
2623 /*
2624  * proc base
2625  *
2626  * These are the directory entries in the root directory of /proc
2627  * that properly belong to the /proc filesystem, as they describe
2628  * describe something that is process related.
2629  */
2630 static const struct pid_entry proc_base_stuff[] = {
2631         NOD("self", S_IFLNK|S_IRWXUGO,
2632                 &proc_self_inode_operations, NULL, {}),
2633 };
2634
2635 /*
2636  *      Exceptional case: normally we are not allowed to unhash a busy
2637  * directory. In this case, however, we can do it - no aliasing problems
2638  * due to the way we treat inodes.
2639  */
2640 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2641 {
2642         struct inode *inode = dentry->d_inode;
2643         struct task_struct *task = get_proc_task(inode);
2644         if (task) {
2645                 put_task_struct(task);
2646                 return 1;
2647         }
2648         d_drop(dentry);
2649         return 0;
2650 }
2651
2652 static const struct dentry_operations proc_base_dentry_operations =
2653 {
2654         .d_revalidate   = proc_base_revalidate,
2655         .d_delete       = pid_delete_dentry,
2656 };
2657
2658 static struct dentry *proc_base_instantiate(struct inode *dir,
2659         struct dentry *dentry, struct task_struct *task, const void *ptr)
2660 {
2661         const struct pid_entry *p = ptr;
2662         struct inode *inode;
2663         struct proc_inode *ei;
2664         struct dentry *error;
2665
2666         /* Allocate the inode */
2667         error = ERR_PTR(-ENOMEM);
2668         inode = new_inode(dir->i_sb);
2669         if (!inode)
2670                 goto out;
2671
2672         /* Initialize the inode */
2673         ei = PROC_I(inode);
2674         inode->i_ino = get_next_ino();
2675         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2676
2677         /*
2678          * grab the reference to the task.
2679          */
2680         ei->pid = get_task_pid(task, PIDTYPE_PID);
2681         if (!ei->pid)
2682                 goto out_iput;
2683
2684         inode->i_mode = p->mode;
2685         if (S_ISDIR(inode->i_mode))
2686                 inode->i_nlink = 2;
2687         if (S_ISLNK(inode->i_mode))
2688                 inode->i_size = 64;
2689         if (p->iop)
2690                 inode->i_op = p->iop;
2691         if (p->fop)
2692                 inode->i_fop = p->fop;
2693         ei->op = p->op;
2694         dentry->d_op = &proc_base_dentry_operations;
2695         d_add(dentry, inode);
2696         error = NULL;
2697 out:
2698         return error;
2699 out_iput:
2700         iput(inode);
2701         goto out;
2702 }
2703
2704 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2705 {
2706         struct dentry *error;
2707         struct task_struct *task = get_proc_task(dir);
2708         const struct pid_entry *p, *last;
2709
2710         error = ERR_PTR(-ENOENT);
2711
2712         if (!task)
2713                 goto out_no_task;
2714
2715         /* Lookup the directory entry */
2716         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2717         for (p = proc_base_stuff; p <= last; p++) {
2718                 if (p->len != dentry->d_name.len)
2719                         continue;
2720                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2721                         break;
2722         }
2723         if (p > last)
2724                 goto out;
2725
2726         error = proc_base_instantiate(dir, dentry, task, p);
2727
2728 out:
2729         put_task_struct(task);
2730 out_no_task:
2731         return error;
2732 }
2733
2734 static int proc_base_fill_cache(struct file *filp, void *dirent,
2735         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2736 {
2737         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2738                                 proc_base_instantiate, task, p);
2739 }
2740
2741 #ifdef CONFIG_TASK_IO_ACCOUNTING
2742 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2743 {
2744         struct task_io_accounting acct = task->ioac;
2745         unsigned long flags;
2746
2747         if (whole && lock_task_sighand(task, &flags)) {
2748                 struct task_struct *t = task;
2749
2750                 task_io_accounting_add(&acct, &task->signal->ioac);
2751                 while_each_thread(task, t)
2752                         task_io_accounting_add(&acct, &t->ioac);
2753
2754                 unlock_task_sighand(task, &flags);
2755         }
2756         return sprintf(buffer,
2757                         "rchar: %llu\n"
2758                         "wchar: %llu\n"
2759                         "syscr: %llu\n"
2760                         "syscw: %llu\n"
2761                         "read_bytes: %llu\n"
2762                         "write_bytes: %llu\n"
2763                         "cancelled_write_bytes: %llu\n",
2764                         (unsigned long long)acct.rchar,
2765                         (unsigned long long)acct.wchar,
2766                         (unsigned long long)acct.syscr,
2767                         (unsigned long long)acct.syscw,
2768                         (unsigned long long)acct.read_bytes,
2769                         (unsigned long long)acct.write_bytes,
2770                         (unsigned long long)acct.cancelled_write_bytes);
2771 }
2772
2773 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2774 {
2775         return do_io_accounting(task, buffer, 0);
2776 }
2777
2778 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2779 {
2780         return do_io_accounting(task, buffer, 1);
2781 }
2782 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2783
2784 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2785                                 struct pid *pid, struct task_struct *task)
2786 {
2787         seq_printf(m, "%08x\n", task->personality);
2788         return 0;
2789 }
2790
2791 /*
2792  * Thread groups
2793  */
2794 static const struct file_operations proc_task_operations;
2795 static const struct inode_operations proc_task_inode_operations;
2796
2797 static const struct pid_entry tgid_base_stuff[] = {
2798         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2799         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2800         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2801 #ifdef CONFIG_NET
2802         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2803 #endif
2804         REG("environ",    S_IRUSR, proc_environ_operations),
2805         INF("auxv",       S_IRUSR, proc_pid_auxv),
2806         ONE("status",     S_IRUGO, proc_pid_status),
2807         ONE("personality", S_IRUSR, proc_pid_personality),
2808         INF("limits",     S_IRUGO, proc_pid_limits),
2809 #ifdef CONFIG_SCHED_DEBUG
2810         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2811 #endif
2812 #ifdef CONFIG_SCHED_AUTOGROUP
2813         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2814 #endif
2815         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2816 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2817         INF("syscall",    S_IRUSR, proc_pid_syscall),
2818 #endif
2819         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2820         ONE("stat",       S_IRUGO, proc_tgid_stat),
2821         ONE("statm",      S_IRUGO, proc_pid_statm),
2822         REG("maps",       S_IRUGO, proc_maps_operations),
2823 #ifdef CONFIG_NUMA
2824         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2825 #endif
2826         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2827         LNK("cwd",        proc_cwd_link),
2828         LNK("root",       proc_root_link),
2829         LNK("exe",        proc_exe_link),
2830         REG("mounts",     S_IRUGO, proc_mounts_operations),
2831         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2832         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2833 #ifdef CONFIG_PROC_PAGE_MONITOR
2834         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2835         REG("smaps",      S_IRUGO, proc_smaps_operations),
2836         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2837 #endif
2838 #ifdef CONFIG_SECURITY
2839         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2840 #endif
2841 #ifdef CONFIG_KALLSYMS
2842         INF("wchan",      S_IRUGO, proc_pid_wchan),
2843 #endif
2844 #ifdef CONFIG_STACKTRACE
2845         ONE("stack",      S_IRUSR, proc_pid_stack),
2846 #endif
2847 #ifdef CONFIG_SCHEDSTATS
2848         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2849 #endif
2850 #ifdef CONFIG_LATENCYTOP
2851         REG("latency",  S_IRUGO, proc_lstats_operations),
2852 #endif
2853 #ifdef CONFIG_PROC_PID_CPUSET
2854         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2855 #endif
2856 #ifdef CONFIG_CGROUPS
2857         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2858 #endif
2859         INF("oom_score",  S_IRUGO, proc_oom_score),
2860         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2861         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2862 #ifdef CONFIG_AUDITSYSCALL
2863         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2864         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2865 #endif
2866 #ifdef CONFIG_FAULT_INJECTION
2867         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2868 #endif
2869 #ifdef CONFIG_ELF_CORE
2870         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2871 #endif
2872 #ifdef CONFIG_TASK_IO_ACCOUNTING
2873         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2874 #endif
2875 };
2876
2877 static int proc_tgid_base_readdir(struct file * filp,
2878                              void * dirent, filldir_t filldir)
2879 {
2880         return proc_pident_readdir(filp,dirent,filldir,
2881                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2882 }
2883
2884 static const struct file_operations proc_tgid_base_operations = {
2885         .read           = generic_read_dir,
2886         .readdir        = proc_tgid_base_readdir,
2887         .llseek         = default_llseek,
2888 };
2889
2890 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2891         return proc_pident_lookup(dir, dentry,
2892                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2893 }
2894
2895 static const struct inode_operations proc_tgid_base_inode_operations = {
2896         .lookup         = proc_tgid_base_lookup,
2897         .getattr        = pid_getattr,
2898         .setattr        = proc_setattr,
2899 };
2900
2901 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2902 {
2903         struct dentry *dentry, *leader, *dir;
2904         char buf[PROC_NUMBUF];
2905         struct qstr name;
2906
2907         name.name = buf;
2908         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2909         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2910         if (dentry) {
2911                 shrink_dcache_parent(dentry);
2912                 d_drop(dentry);
2913                 dput(dentry);
2914         }
2915
2916         name.name = buf;
2917         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2918         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2919         if (!leader)
2920                 goto out;
2921
2922         name.name = "task";
2923         name.len = strlen(name.name);
2924         dir = d_hash_and_lookup(leader, &name);
2925         if (!dir)
2926                 goto out_put_leader;
2927
2928         name.name = buf;
2929         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2930         dentry = d_hash_and_lookup(dir, &name);
2931         if (dentry) {
2932                 shrink_dcache_parent(dentry);
2933                 d_drop(dentry);
2934                 dput(dentry);
2935         }
2936
2937         dput(dir);
2938 out_put_leader:
2939         dput(leader);
2940 out:
2941         return;
2942 }
2943
2944 /**
2945  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2946  * @task: task that should be flushed.
2947  *
2948  * When flushing dentries from proc, one needs to flush them from global
2949  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2950  * in. This call is supposed to do all of this job.
2951  *
2952  * Looks in the dcache for
2953  * /proc/@pid
2954  * /proc/@tgid/task/@pid
2955  * if either directory is present flushes it and all of it'ts children
2956  * from the dcache.
2957  *
2958  * It is safe and reasonable to cache /proc entries for a task until
2959  * that task exits.  After that they just clog up the dcache with
2960  * useless entries, possibly causing useful dcache entries to be
2961  * flushed instead.  This routine is proved to flush those useless
2962  * dcache entries at process exit time.
2963  *
2964  * NOTE: This routine is just an optimization so it does not guarantee
2965  *       that no dcache entries will exist at process exit time it
2966  *       just makes it very unlikely that any will persist.
2967  */
2968
2969 void proc_flush_task(struct task_struct *task)
2970 {
2971         int i;
2972         struct pid *pid, *tgid;
2973         struct upid *upid;
2974
2975         pid = task_pid(task);
2976         tgid = task_tgid(task);
2977
2978         for (i = 0; i <= pid->level; i++) {
2979                 upid = &pid->numbers[i];
2980                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2981                                         tgid->numbers[i].nr);
2982         }
2983
2984         upid = &pid->numbers[pid->level];
2985         if (upid->nr == 1)
2986                 pid_ns_release_proc(upid->ns);
2987 }
2988
2989 static struct dentry *proc_pid_instantiate(struct inode *dir,
2990                                            struct dentry * dentry,
2991                                            struct task_struct *task, const void *ptr)
2992 {
2993         struct dentry *error = ERR_PTR(-ENOENT);
2994         struct inode *inode;
2995
2996         inode = proc_pid_make_inode(dir->i_sb, task);
2997         if (!inode)
2998                 goto out;
2999
3000         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3001         inode->i_op = &proc_tgid_base_inode_operations;
3002         inode->i_fop = &proc_tgid_base_operations;
3003         inode->i_flags|=S_IMMUTABLE;
3004
3005         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
3006                 ARRAY_SIZE(tgid_base_stuff));
3007
3008         dentry->d_op = &pid_dentry_operations;
3009
3010         d_add(dentry, inode);
3011         /* Close the race of the process dying before we return the dentry */
3012         if (pid_revalidate(dentry, NULL))
3013                 error = NULL;
3014 out:
3015         return error;
3016 }
3017
3018 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3019 {
3020         struct dentry *result;
3021         struct task_struct *task;
3022         unsigned tgid;
3023         struct pid_namespace *ns;
3024
3025         result = proc_base_lookup(dir, dentry);
3026         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3027                 goto out;
3028
3029         tgid = name_to_int(dentry);
3030         if (tgid == ~0U)
3031                 goto out;
3032
3033         ns = dentry->d_sb->s_fs_info;
3034         rcu_read_lock();
3035         task = find_task_by_pid_ns(tgid, ns);
3036         if (task)
3037                 get_task_struct(task);
3038         rcu_read_unlock();
3039         if (!task)
3040                 goto out;
3041
3042         result = proc_pid_instantiate(dir, dentry, task, NULL);
3043         put_task_struct(task);
3044 out:
3045         return result;
3046 }
3047
3048 /*
3049  * Find the first task with tgid >= tgid
3050  *
3051  */
3052 struct tgid_iter {
3053         unsigned int tgid;
3054         struct task_struct *task;
3055 };
3056 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3057 {
3058         struct pid *pid;
3059
3060         if (iter.task)
3061                 put_task_struct(iter.task);
3062         rcu_read_lock();
3063 retry:
3064         iter.task = NULL;
3065         pid = find_ge_pid(iter.tgid, ns);
3066         if (pid) {
3067                 iter.tgid = pid_nr_ns(pid, ns);
3068                 iter.task = pid_task(pid, PIDTYPE_PID);
3069                 /* What we to know is if the pid we have find is the
3070                  * pid of a thread_group_leader.  Testing for task
3071                  * being a thread_group_leader is the obvious thing
3072                  * todo but there is a window when it fails, due to
3073                  * the pid transfer logic in de_thread.
3074                  *
3075                  * So we perform the straight forward test of seeing
3076                  * if the pid we have found is the pid of a thread
3077                  * group leader, and don't worry if the task we have
3078                  * found doesn't happen to be a thread group leader.
3079                  * As we don't care in the case of readdir.
3080                  */
3081                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3082                         iter.tgid += 1;
3083                         goto retry;
3084                 }
3085                 get_task_struct(iter.task);
3086         }
3087         rcu_read_unlock();
3088         return iter;
3089 }
3090
3091 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3092
3093 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3094         struct tgid_iter iter)
3095 {
3096         char name[PROC_NUMBUF];
3097         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3098         return proc_fill_cache(filp, dirent, filldir, name, len,
3099                                 proc_pid_instantiate, iter.task, NULL);
3100 }
3101
3102 /* for the /proc/ directory itself, after non-process stuff has been done */
3103 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3104 {
3105         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3106         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3107         struct tgid_iter iter;
3108         struct pid_namespace *ns;
3109
3110         if (!reaper)
3111                 goto out_no_task;
3112
3113         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3114                 const struct pid_entry *p = &proc_base_stuff[nr];
3115                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3116                         goto out;
3117         }
3118
3119         ns = filp->f_dentry->d_sb->s_fs_info;
3120         iter.task = NULL;
3121         iter.tgid = filp->f_pos - TGID_OFFSET;
3122         for (iter = next_tgid(ns, iter);
3123              iter.task;
3124              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3125                 filp->f_pos = iter.tgid + TGID_OFFSET;
3126                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3127                         put_task_struct(iter.task);
3128                         goto out;
3129                 }
3130         }
3131         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3132 out:
3133         put_task_struct(reaper);
3134 out_no_task:
3135         return 0;
3136 }
3137
3138 /*
3139  * Tasks
3140  */
3141 static const struct pid_entry tid_base_stuff[] = {
3142         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3143         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3144         REG("environ",   S_IRUSR, proc_environ_operations),
3145         INF("auxv",      S_IRUSR, proc_pid_auxv),
3146         ONE("status",    S_IRUGO, proc_pid_status),
3147         ONE("personality", S_IRUSR, proc_pid_personality),
3148         INF("limits",    S_IRUGO, proc_pid_limits),
3149 #ifdef CONFIG_SCHED_DEBUG
3150         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3151 #endif
3152         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3153 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3154         INF("syscall",   S_IRUSR, proc_pid_syscall),
3155 #endif
3156         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3157         ONE("stat",      S_IRUGO, proc_tid_stat),
3158         ONE("statm",     S_IRUGO, proc_pid_statm),
3159         REG("maps",      S_IRUGO, proc_maps_operations),
3160 #ifdef CONFIG_NUMA
3161         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3162 #endif
3163         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3164         LNK("cwd",       proc_cwd_link),
3165         LNK("root",      proc_root_link),
3166         LNK("exe",       proc_exe_link),
3167         REG("mounts",    S_IRUGO, proc_mounts_operations),
3168         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3169 #ifdef CONFIG_PROC_PAGE_MONITOR
3170         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3171         REG("smaps",     S_IRUGO, proc_smaps_operations),
3172         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3173 #endif
3174 #ifdef CONFIG_SECURITY
3175         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3176 #endif
3177 #ifdef CONFIG_KALLSYMS
3178         INF("wchan",     S_IRUGO, proc_pid_wchan),
3179 #endif
3180 #ifdef CONFIG_STACKTRACE
3181         ONE("stack",      S_IRUSR, proc_pid_stack),
3182 #endif
3183 #ifdef CONFIG_SCHEDSTATS
3184         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3185 #endif
3186 #ifdef CONFIG_LATENCYTOP
3187         REG("latency",  S_IRUGO, proc_lstats_operations),
3188 #endif
3189 #ifdef CONFIG_PROC_PID_CPUSET
3190         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3191 #endif
3192 #ifdef CONFIG_CGROUPS
3193         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3194 #endif
3195         INF("oom_score", S_IRUGO, proc_oom_score),
3196         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3197         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3198 #ifdef CONFIG_AUDITSYSCALL
3199         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3200         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3201 #endif
3202 #ifdef CONFIG_FAULT_INJECTION
3203         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3204 #endif
3205 #ifdef CONFIG_TASK_IO_ACCOUNTING
3206         INF("io",       S_IRUGO, proc_tid_io_accounting),
3207 #endif
3208 };
3209
3210 static int proc_tid_base_readdir(struct file * filp,
3211                              void * dirent, filldir_t filldir)
3212 {
3213         return proc_pident_readdir(filp,dirent,filldir,
3214                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3215 }
3216
3217 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3218         return proc_pident_lookup(dir, dentry,
3219                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3220 }
3221
3222 static const struct file_operations proc_tid_base_operations = {
3223         .read           = generic_read_dir,
3224         .readdir        = proc_tid_base_readdir,
3225         .llseek         = default_llseek,
3226 };
3227
3228 static const struct inode_operations proc_tid_base_inode_operations = {
3229         .lookup         = proc_tid_base_lookup,
3230         .getattr        = pid_getattr,
3231         .setattr        = proc_setattr,
3232 };
3233
3234 static struct dentry *proc_task_instantiate(struct inode *dir,
3235         struct dentry *dentry, struct task_struct *task, const void *ptr)
3236 {
3237         struct dentry *error = ERR_PTR(-ENOENT);
3238         struct inode *inode;
3239         inode = proc_pid_make_inode(dir->i_sb, task);
3240
3241         if (!inode)
3242                 goto out;
3243         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3244         inode->i_op = &proc_tid_base_inode_operations;
3245         inode->i_fop = &proc_tid_base_operations;
3246         inode->i_flags|=S_IMMUTABLE;
3247
3248         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3249                 ARRAY_SIZE(tid_base_stuff));
3250
3251         dentry->d_op = &pid_dentry_operations;
3252
3253         d_add(dentry, inode);
3254         /* Close the race of the process dying before we return the dentry */
3255         if (pid_revalidate(dentry, NULL))
3256                 error = NULL;
3257 out:
3258         return error;
3259 }
3260
3261 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3262 {
3263         struct dentry *result = ERR_PTR(-ENOENT);
3264         struct task_struct *task;
3265         struct task_struct *leader = get_proc_task(dir);
3266         unsigned tid;
3267         struct pid_namespace *ns;
3268
3269         if (!leader)
3270                 goto out_no_task;
3271
3272         tid = name_to_int(dentry);
3273         if (tid == ~0U)
3274                 goto out;
3275
3276         ns = dentry->d_sb->s_fs_info;
3277         rcu_read_lock();
3278         task = find_task_by_pid_ns(tid, ns);
3279         if (task)
3280                 get_task_struct(task);
3281         rcu_read_unlock();
3282         if (!task)
3283                 goto out;
3284         if (!same_thread_group(leader, task))
3285                 goto out_drop_task;
3286
3287         result = proc_task_instantiate(dir, dentry, task, NULL);
3288 out_drop_task:
3289         put_task_struct(task);
3290 out:
3291         put_task_struct(leader);
3292 out_no_task:
3293         return result;
3294 }
3295
3296 /*
3297  * Find the first tid of a thread group to return to user space.
3298  *
3299  * Usually this is just the thread group leader, but if the users
3300  * buffer was too small or there was a seek into the middle of the
3301  * directory we have more work todo.
3302  *
3303  * In the case of a short read we start with find_task_by_pid.
3304  *
3305  * In the case of a seek we start with the leader and walk nr
3306  * threads past it.
3307  */
3308 static struct task_struct *first_tid(struct task_struct *leader,
3309                 int tid, int nr, struct pid_namespace *ns)
3310 {
3311         struct task_struct *pos;
3312
3313         rcu_read_lock();
3314         /* Attempt to start with the pid of a thread */
3315         if (tid && (nr > 0)) {
3316                 pos = find_task_by_pid_ns(tid, ns);
3317                 if (pos && (pos->group_leader == leader))
3318                         goto found;
3319         }
3320
3321         /* If nr exceeds the number of threads there is nothing todo */
3322         pos = NULL;
3323         if (nr && nr >= get_nr_threads(leader))
3324                 goto out;
3325
3326         /* If we haven't found our starting place yet start
3327          * with the leader and walk nr threads forward.
3328          */
3329         for (pos = leader; nr > 0; --nr) {
3330                 pos = next_thread(pos);
3331                 if (pos == leader) {
3332                         pos = NULL;
3333                         goto out;
3334                 }
3335         }
3336 found:
3337         get_task_struct(pos);
3338 out:
3339         rcu_read_unlock();
3340         return pos;
3341 }
3342
3343 /*
3344  * Find the next thread in the thread list.
3345  * Return NULL if there is an error or no next thread.
3346  *
3347  * The reference to the input task_struct is released.
3348  */
3349 static struct task_struct *next_tid(struct task_struct *start)
3350 {
3351         struct task_struct *pos = NULL;
3352         rcu_read_lock();
3353         if (pid_alive(start)) {
3354                 pos = next_thread(start);
3355                 if (thread_group_leader(pos))
3356                         pos = NULL;
3357                 else
3358                         get_task_struct(pos);
3359         }
3360         rcu_read_unlock();
3361         put_task_struct(start);
3362         return pos;
3363 }
3364
3365 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3366         struct task_struct *task, int tid)
3367 {
3368         char name[PROC_NUMBUF];
3369         int len = snprintf(name, sizeof(name), "%d", tid);
3370         return proc_fill_cache(filp, dirent, filldir, name, len,
3371                                 proc_task_instantiate, task, NULL);
3372 }
3373
3374 /* for the /proc/TGID/task/ directories */
3375 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3376 {
3377         struct dentry *dentry = filp->f_path.dentry;
3378         struct inode *inode = dentry->d_inode;
3379         struct task_struct *leader = NULL;
3380         struct task_struct *task;
3381         int retval = -ENOENT;
3382         ino_t ino;
3383         int tid;
3384         struct pid_namespace *ns;
3385
3386         task = get_proc_task(inode);
3387         if (!task)
3388                 goto out_no_task;
3389         rcu_read_lock();
3390         if (pid_alive(task)) {
3391                 leader = task->group_leader;
3392                 get_task_struct(leader);
3393         }
3394         rcu_read_unlock();
3395         put_task_struct(task);
3396         if (!leader)
3397                 goto out_no_task;
3398         retval = 0;
3399
3400         switch ((unsigned long)filp->f_pos) {
3401         case 0:
3402                 ino = inode->i_ino;
3403                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3404                         goto out;
3405                 filp->f_pos++;
3406                 /* fall through */
3407         case 1:
3408                 ino = parent_ino(dentry);
3409                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3410                         goto out;
3411                 filp->f_pos++;
3412                 /* fall through */
3413         }
3414
3415         /* f_version caches the tgid value that the last readdir call couldn't
3416          * return. lseek aka telldir automagically resets f_version to 0.
3417          */
3418         ns = filp->f_dentry->d_sb->s_fs_info;
3419         tid = (int)filp->f_version;
3420         filp->f_version = 0;
3421         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3422              task;
3423              task = next_tid(task), filp->f_pos++) {
3424                 tid = task_pid_nr_ns(task, ns);
3425                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3426                         /* returning this tgid failed, save it as the first
3427                          * pid for the next readir call */
3428                         filp->f_version = (u64)tid;
3429                         put_task_struct(task);
3430                         break;
3431                 }
3432         }
3433 out:
3434         put_task_struct(leader);
3435 out_no_task:
3436         return retval;
3437 }
3438
3439 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3440 {
3441         struct inode *inode = dentry->d_inode;
3442         struct task_struct *p = get_proc_task(inode);
3443         generic_fillattr(inode, stat);
3444
3445         if (p) {
3446                 stat->nlink += get_nr_threads(p);
3447                 put_task_struct(p);
3448         }
3449
3450         return 0;
3451 }
3452
3453 static const struct inode_operations proc_task_inode_operations = {
3454         .lookup         = proc_task_lookup,
3455         .getattr        = proc_task_getattr,
3456         .setattr        = proc_setattr,
3457 };
3458
3459 static const struct file_operations proc_task_operations = {
3460         .read           = generic_read_dir,
3461         .readdir        = proc_task_readdir,
3462         .llseek         = default_llseek,
3463 };