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