fs/proc: clean up printks
[linux-3.10.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/printk.h>
77 #include <linux/cgroup.h>
78 #include <linux/cpuset.h>
79 #include <linux/audit.h>
80 #include <linux/poll.h>
81 #include <linux/nsproxy.h>
82 #include <linux/oom.h>
83 #include <linux/elf.h>
84 #include <linux/pid_namespace.h>
85 #include <linux/user_namespace.h>
86 #include <linux/fs_struct.h>
87 #include <linux/slab.h>
88 #include <linux/flex_array.h>
89 #ifdef CONFIG_HARDWALL
90 #include <asm/hardwall.h>
91 #endif
92 #include <trace/events/oom.h>
93 #include "internal.h"
94 #include "fd.h"
95
96 /* NOTE:
97  *      Implementing inode permission operations in /proc is almost
98  *      certainly an error.  Permission checks need to happen during
99  *      each system call not at open time.  The reason is that most of
100  *      what we wish to check for permissions in /proc varies at runtime.
101  *
102  *      The classic example of a problem is opening file descriptors
103  *      in /proc for a task before it execs a suid executable.
104  */
105
106 struct pid_entry {
107         char *name;
108         int len;
109         umode_t mode;
110         const struct inode_operations *iop;
111         const struct file_operations *fop;
112         union proc_op op;
113 };
114
115 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
116         .name = (NAME),                                 \
117         .len  = sizeof(NAME) - 1,                       \
118         .mode = MODE,                                   \
119         .iop  = IOP,                                    \
120         .fop  = FOP,                                    \
121         .op   = OP,                                     \
122 }
123
124 #define DIR(NAME, MODE, iops, fops)     \
125         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
126 #define LNK(NAME, get_link)                                     \
127         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
128                 &proc_pid_link_inode_operations, NULL,          \
129                 { .proc_get_link = get_link } )
130 #define REG(NAME, MODE, fops)                           \
131         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
132 #define INF(NAME, MODE, read)                           \
133         NOD(NAME, (S_IFREG|(MODE)),                     \
134                 NULL, &proc_info_file_operations,       \
135                 { .proc_read = read } )
136 #define ONE(NAME, MODE, show)                           \
137         NOD(NAME, (S_IFREG|(MODE)),                     \
138                 NULL, &proc_single_file_operations,     \
139                 { .proc_show = show } )
140
141 /*
142  * Count the number of hardlinks for the pid_entry table, excluding the .
143  * and .. links.
144  */
145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
146         unsigned int n)
147 {
148         unsigned int i;
149         unsigned int count;
150
151         count = 0;
152         for (i = 0; i < n; ++i) {
153                 if (S_ISDIR(entries[i].mode))
154                         ++count;
155         }
156
157         return count;
158 }
159
160 static int get_task_root(struct task_struct *task, struct path *root)
161 {
162         int result = -ENOENT;
163
164         task_lock(task);
165         if (task->fs) {
166                 get_fs_root(task->fs, root);
167                 result = 0;
168         }
169         task_unlock(task);
170         return result;
171 }
172
173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
174 {
175         struct task_struct *task = get_proc_task(dentry->d_inode);
176         int result = -ENOENT;
177
178         if (task) {
179                 task_lock(task);
180                 if (task->fs) {
181                         get_fs_pwd(task->fs, path);
182                         result = 0;
183                 }
184                 task_unlock(task);
185                 put_task_struct(task);
186         }
187         return result;
188 }
189
190 static int proc_root_link(struct dentry *dentry, struct path *path)
191 {
192         struct task_struct *task = get_proc_task(dentry->d_inode);
193         int result = -ENOENT;
194
195         if (task) {
196                 result = get_task_root(task, path);
197                 put_task_struct(task);
198         }
199         return result;
200 }
201
202 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
203 {
204         int res = 0;
205         unsigned int len;
206         struct mm_struct *mm = get_task_mm(task);
207         if (!mm)
208                 goto out;
209         if (!mm->arg_end)
210                 goto out_mm;    /* Shh! No looking before we're done */
211
212         len = mm->arg_end - mm->arg_start;
213  
214         if (len > PAGE_SIZE)
215                 len = PAGE_SIZE;
216  
217         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
218
219         // If the nul at the end of args has been overwritten, then
220         // assume application is using setproctitle(3).
221         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
222                 len = strnlen(buffer, res);
223                 if (len < res) {
224                     res = len;
225                 } else {
226                         len = mm->env_end - mm->env_start;
227                         if (len > PAGE_SIZE - res)
228                                 len = PAGE_SIZE - res;
229                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
230                         res = strnlen(buffer, res);
231                 }
232         }
233 out_mm:
234         mmput(mm);
235 out:
236         return res;
237 }
238
239 static int proc_pid_auxv(struct task_struct *task, char *buffer)
240 {
241         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
242         int res = PTR_ERR(mm);
243         if (mm && !IS_ERR(mm)) {
244                 unsigned int nwords = 0;
245                 do {
246                         nwords += 2;
247                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
248                 res = nwords * sizeof(mm->saved_auxv[0]);
249                 if (res > PAGE_SIZE)
250                         res = PAGE_SIZE;
251                 memcpy(buffer, mm->saved_auxv, res);
252                 mmput(mm);
253         }
254         return res;
255 }
256
257
258 #ifdef CONFIG_KALLSYMS
259 /*
260  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
261  * Returns the resolved symbol.  If that fails, simply return the address.
262  */
263 static int proc_pid_wchan(struct task_struct *task, char *buffer)
264 {
265         unsigned long wchan;
266         char symname[KSYM_NAME_LEN];
267
268         wchan = get_wchan(task);
269
270         if (lookup_symbol_name(wchan, symname) < 0)
271                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
272                         return 0;
273                 else
274                         return sprintf(buffer, "%lu", wchan);
275         else
276                 return sprintf(buffer, "%s", symname);
277 }
278 #endif /* CONFIG_KALLSYMS */
279
280 static int lock_trace(struct task_struct *task)
281 {
282         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
283         if (err)
284                 return err;
285         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
286                 mutex_unlock(&task->signal->cred_guard_mutex);
287                 return -EPERM;
288         }
289         return 0;
290 }
291
292 static void unlock_trace(struct task_struct *task)
293 {
294         mutex_unlock(&task->signal->cred_guard_mutex);
295 }
296
297 #ifdef CONFIG_STACKTRACE
298
299 #define MAX_STACK_TRACE_DEPTH   64
300
301 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
302                           struct pid *pid, struct task_struct *task)
303 {
304         struct stack_trace trace;
305         unsigned long *entries;
306         int err;
307         int i;
308
309         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
310         if (!entries)
311                 return -ENOMEM;
312
313         trace.nr_entries        = 0;
314         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
315         trace.entries           = entries;
316         trace.skip              = 0;
317
318         err = lock_trace(task);
319         if (!err) {
320                 save_stack_trace_tsk(task, &trace);
321
322                 for (i = 0; i < trace.nr_entries; i++) {
323                         seq_printf(m, "[<%pK>] %pS\n",
324                                    (void *)entries[i], (void *)entries[i]);
325                 }
326                 unlock_trace(task);
327         }
328         kfree(entries);
329
330         return err;
331 }
332 #endif
333
334 #ifdef CONFIG_SCHEDSTATS
335 /*
336  * Provides /proc/PID/schedstat
337  */
338 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
339 {
340         return sprintf(buffer, "%llu %llu %lu\n",
341                         (unsigned long long)task->se.sum_exec_runtime,
342                         (unsigned long long)task->sched_info.run_delay,
343                         task->sched_info.pcount);
344 }
345 #endif
346
347 #ifdef CONFIG_LATENCYTOP
348 static int lstats_show_proc(struct seq_file *m, void *v)
349 {
350         int i;
351         struct inode *inode = m->private;
352         struct task_struct *task = get_proc_task(inode);
353
354         if (!task)
355                 return -ESRCH;
356         seq_puts(m, "Latency Top version : v0.1\n");
357         for (i = 0; i < 32; i++) {
358                 struct latency_record *lr = &task->latency_record[i];
359                 if (lr->backtrace[0]) {
360                         int q;
361                         seq_printf(m, "%i %li %li",
362                                    lr->count, lr->time, lr->max);
363                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
364                                 unsigned long bt = lr->backtrace[q];
365                                 if (!bt)
366                                         break;
367                                 if (bt == ULONG_MAX)
368                                         break;
369                                 seq_printf(m, " %ps", (void *)bt);
370                         }
371                         seq_putc(m, '\n');
372                 }
373
374         }
375         put_task_struct(task);
376         return 0;
377 }
378
379 static int lstats_open(struct inode *inode, struct file *file)
380 {
381         return single_open(file, lstats_show_proc, inode);
382 }
383
384 static ssize_t lstats_write(struct file *file, const char __user *buf,
385                             size_t count, loff_t *offs)
386 {
387         struct task_struct *task = get_proc_task(file_inode(file));
388
389         if (!task)
390                 return -ESRCH;
391         clear_all_latency_tracing(task);
392         put_task_struct(task);
393
394         return count;
395 }
396
397 static const struct file_operations proc_lstats_operations = {
398         .open           = lstats_open,
399         .read           = seq_read,
400         .write          = lstats_write,
401         .llseek         = seq_lseek,
402         .release        = single_release,
403 };
404
405 #endif
406
407 static int proc_oom_score(struct task_struct *task, char *buffer)
408 {
409         unsigned long totalpages = totalram_pages + total_swap_pages;
410         unsigned long points = 0;
411
412         read_lock(&tasklist_lock);
413         if (pid_alive(task))
414                 points = oom_badness(task, NULL, NULL, totalpages) *
415                                                 1000 / totalpages;
416         read_unlock(&tasklist_lock);
417         return sprintf(buffer, "%lu\n", points);
418 }
419
420 struct limit_names {
421         char *name;
422         char *unit;
423 };
424
425 static const struct limit_names lnames[RLIM_NLIMITS] = {
426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
428         [RLIMIT_DATA] = {"Max data size", "bytes"},
429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
432         [RLIMIT_NPROC] = {"Max processes", "processes"},
433         [RLIMIT_NOFILE] = {"Max open files", "files"},
434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
435         [RLIMIT_AS] = {"Max address space", "bytes"},
436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
439         [RLIMIT_NICE] = {"Max nice priority", NULL},
440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
442 };
443
444 /* Display limits for a process */
445 static int proc_pid_limits(struct task_struct *task, char *buffer)
446 {
447         unsigned int i;
448         int count = 0;
449         unsigned long flags;
450         char *bufptr = buffer;
451
452         struct rlimit rlim[RLIM_NLIMITS];
453
454         if (!lock_task_sighand(task, &flags))
455                 return 0;
456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
457         unlock_task_sighand(task, &flags);
458
459         /*
460          * print the file header
461          */
462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
463                         "Limit", "Soft Limit", "Hard Limit", "Units");
464
465         for (i = 0; i < RLIM_NLIMITS; i++) {
466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
468                                          lnames[i].name, "unlimited");
469                 else
470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
471                                          lnames[i].name, rlim[i].rlim_cur);
472
473                 if (rlim[i].rlim_max == RLIM_INFINITY)
474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
475                 else
476                         count += sprintf(&bufptr[count], "%-20lu ",
477                                          rlim[i].rlim_max);
478
479                 if (lnames[i].unit)
480                         count += sprintf(&bufptr[count], "%-10s\n",
481                                          lnames[i].unit);
482                 else
483                         count += sprintf(&bufptr[count], "\n");
484         }
485
486         return count;
487 }
488
489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
491 {
492         long nr;
493         unsigned long args[6], sp, pc;
494         int res = lock_trace(task);
495         if (res)
496                 return res;
497
498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
499                 res = sprintf(buffer, "running\n");
500         else if (nr < 0)
501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
502         else
503                 res = sprintf(buffer,
504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
505                        nr,
506                        args[0], args[1], args[2], args[3], args[4], args[5],
507                        sp, pc);
508         unlock_trace(task);
509         return res;
510 }
511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
512
513 /************************************************************************/
514 /*                       Here the fs part begins                        */
515 /************************************************************************/
516
517 /* permission checks */
518 static int proc_fd_access_allowed(struct inode *inode)
519 {
520         struct task_struct *task;
521         int allowed = 0;
522         /* Allow access to a task's file descriptors if it is us or we
523          * may use ptrace attach to the process and find out that
524          * information.
525          */
526         task = get_proc_task(inode);
527         if (task) {
528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
529                 put_task_struct(task);
530         }
531         return allowed;
532 }
533
534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
535 {
536         int error;
537         struct inode *inode = dentry->d_inode;
538
539         if (attr->ia_valid & ATTR_MODE)
540                 return -EPERM;
541
542         error = inode_change_ok(inode, attr);
543         if (error)
544                 return error;
545
546         setattr_copy(inode, attr);
547         mark_inode_dirty(inode);
548         return 0;
549 }
550
551 /*
552  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
553  * or euid/egid (for hide_pid_min=2)?
554  */
555 static bool has_pid_permissions(struct pid_namespace *pid,
556                                  struct task_struct *task,
557                                  int hide_pid_min)
558 {
559         if (pid->hide_pid < hide_pid_min)
560                 return true;
561         if (in_group_p(pid->pid_gid))
562                 return true;
563         return ptrace_may_access(task, PTRACE_MODE_READ);
564 }
565
566
567 static int proc_pid_permission(struct inode *inode, int mask)
568 {
569         struct pid_namespace *pid = inode->i_sb->s_fs_info;
570         struct task_struct *task;
571         bool has_perms;
572
573         task = get_proc_task(inode);
574         if (!task)
575                 return -ESRCH;
576         has_perms = has_pid_permissions(pid, task, 1);
577         put_task_struct(task);
578
579         if (!has_perms) {
580                 if (pid->hide_pid == 2) {
581                         /*
582                          * Let's make getdents(), stat(), and open()
583                          * consistent with each other.  If a process
584                          * may not stat() a file, it shouldn't be seen
585                          * in procfs at all.
586                          */
587                         return -ENOENT;
588                 }
589
590                 return -EPERM;
591         }
592         return generic_permission(inode, mask);
593 }
594
595
596
597 static const struct inode_operations proc_def_inode_operations = {
598         .setattr        = proc_setattr,
599 };
600
601 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
602
603 static ssize_t proc_info_read(struct file * file, char __user * buf,
604                           size_t count, loff_t *ppos)
605 {
606         struct inode * inode = file_inode(file);
607         unsigned long page;
608         ssize_t length;
609         struct task_struct *task = get_proc_task(inode);
610
611         length = -ESRCH;
612         if (!task)
613                 goto out_no_task;
614
615         if (count > PROC_BLOCK_SIZE)
616                 count = PROC_BLOCK_SIZE;
617
618         length = -ENOMEM;
619         if (!(page = __get_free_page(GFP_TEMPORARY)))
620                 goto out;
621
622         length = PROC_I(inode)->op.proc_read(task, (char*)page);
623
624         if (length >= 0)
625                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
626         free_page(page);
627 out:
628         put_task_struct(task);
629 out_no_task:
630         return length;
631 }
632
633 static const struct file_operations proc_info_file_operations = {
634         .read           = proc_info_read,
635         .llseek         = generic_file_llseek,
636 };
637
638 static int proc_single_show(struct seq_file *m, void *v)
639 {
640         struct inode *inode = m->private;
641         struct pid_namespace *ns;
642         struct pid *pid;
643         struct task_struct *task;
644         int ret;
645
646         ns = inode->i_sb->s_fs_info;
647         pid = proc_pid(inode);
648         task = get_pid_task(pid, PIDTYPE_PID);
649         if (!task)
650                 return -ESRCH;
651
652         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
653
654         put_task_struct(task);
655         return ret;
656 }
657
658 static int proc_single_open(struct inode *inode, struct file *filp)
659 {
660         return single_open(filp, proc_single_show, inode);
661 }
662
663 static const struct file_operations proc_single_file_operations = {
664         .open           = proc_single_open,
665         .read           = seq_read,
666         .llseek         = seq_lseek,
667         .release        = single_release,
668 };
669
670 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
671 {
672         struct task_struct *task = get_proc_task(file_inode(file));
673         struct mm_struct *mm;
674
675         if (!task)
676                 return -ESRCH;
677
678         mm = mm_access(task, mode);
679         put_task_struct(task);
680
681         if (IS_ERR(mm))
682                 return PTR_ERR(mm);
683
684         if (mm) {
685                 /* ensure this mm_struct can't be freed */
686                 atomic_inc(&mm->mm_count);
687                 /* but do not pin its memory */
688                 mmput(mm);
689         }
690
691         file->private_data = mm;
692
693         return 0;
694 }
695
696 static int mem_open(struct inode *inode, struct file *file)
697 {
698         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
699
700         /* OK to pass negative loff_t, we can catch out-of-range */
701         file->f_mode |= FMODE_UNSIGNED_OFFSET;
702
703         return ret;
704 }
705
706 static ssize_t mem_rw(struct file *file, char __user *buf,
707                         size_t count, loff_t *ppos, int write)
708 {
709         struct mm_struct *mm = file->private_data;
710         unsigned long addr = *ppos;
711         ssize_t copied;
712         char *page;
713
714         if (!mm)
715                 return 0;
716
717         page = (char *)__get_free_page(GFP_TEMPORARY);
718         if (!page)
719                 return -ENOMEM;
720
721         copied = 0;
722         if (!atomic_inc_not_zero(&mm->mm_users))
723                 goto free;
724
725         while (count > 0) {
726                 int this_len = min_t(int, count, PAGE_SIZE);
727
728                 if (write && copy_from_user(page, buf, this_len)) {
729                         copied = -EFAULT;
730                         break;
731                 }
732
733                 this_len = access_remote_vm(mm, addr, page, this_len, write);
734                 if (!this_len) {
735                         if (!copied)
736                                 copied = -EIO;
737                         break;
738                 }
739
740                 if (!write && copy_to_user(buf, page, this_len)) {
741                         copied = -EFAULT;
742                         break;
743                 }
744
745                 buf += this_len;
746                 addr += this_len;
747                 copied += this_len;
748                 count -= this_len;
749         }
750         *ppos = addr;
751
752         mmput(mm);
753 free:
754         free_page((unsigned long) page);
755         return copied;
756 }
757
758 static ssize_t mem_read(struct file *file, char __user *buf,
759                         size_t count, loff_t *ppos)
760 {
761         return mem_rw(file, buf, count, ppos, 0);
762 }
763
764 static ssize_t mem_write(struct file *file, const char __user *buf,
765                          size_t count, loff_t *ppos)
766 {
767         return mem_rw(file, (char __user*)buf, count, ppos, 1);
768 }
769
770 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
771 {
772         switch (orig) {
773         case 0:
774                 file->f_pos = offset;
775                 break;
776         case 1:
777                 file->f_pos += offset;
778                 break;
779         default:
780                 return -EINVAL;
781         }
782         force_successful_syscall_return();
783         return file->f_pos;
784 }
785
786 static int mem_release(struct inode *inode, struct file *file)
787 {
788         struct mm_struct *mm = file->private_data;
789         if (mm)
790                 mmdrop(mm);
791         return 0;
792 }
793
794 static const struct file_operations proc_mem_operations = {
795         .llseek         = mem_lseek,
796         .read           = mem_read,
797         .write          = mem_write,
798         .open           = mem_open,
799         .release        = mem_release,
800 };
801
802 static int environ_open(struct inode *inode, struct file *file)
803 {
804         return __mem_open(inode, file, PTRACE_MODE_READ);
805 }
806
807 static ssize_t environ_read(struct file *file, char __user *buf,
808                         size_t count, loff_t *ppos)
809 {
810         char *page;
811         unsigned long src = *ppos;
812         int ret = 0;
813         struct mm_struct *mm = file->private_data;
814
815         if (!mm)
816                 return 0;
817
818         page = (char *)__get_free_page(GFP_TEMPORARY);
819         if (!page)
820                 return -ENOMEM;
821
822         ret = 0;
823         if (!atomic_inc_not_zero(&mm->mm_users))
824                 goto free;
825         while (count > 0) {
826                 size_t this_len, max_len;
827                 int retval;
828
829                 if (src >= (mm->env_end - mm->env_start))
830                         break;
831
832                 this_len = mm->env_end - (mm->env_start + src);
833
834                 max_len = min_t(size_t, PAGE_SIZE, count);
835                 this_len = min(max_len, this_len);
836
837                 retval = access_remote_vm(mm, (mm->env_start + src),
838                         page, this_len, 0);
839
840                 if (retval <= 0) {
841                         ret = retval;
842                         break;
843                 }
844
845                 if (copy_to_user(buf, page, retval)) {
846                         ret = -EFAULT;
847                         break;
848                 }
849
850                 ret += retval;
851                 src += retval;
852                 buf += retval;
853                 count -= retval;
854         }
855         *ppos = src;
856         mmput(mm);
857
858 free:
859         free_page((unsigned long) page);
860         return ret;
861 }
862
863 static const struct file_operations proc_environ_operations = {
864         .open           = environ_open,
865         .read           = environ_read,
866         .llseek         = generic_file_llseek,
867         .release        = mem_release,
868 };
869
870 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
871                             loff_t *ppos)
872 {
873         struct task_struct *task = get_proc_task(file_inode(file));
874         char buffer[PROC_NUMBUF];
875         int oom_adj = OOM_ADJUST_MIN;
876         size_t len;
877         unsigned long flags;
878
879         if (!task)
880                 return -ESRCH;
881         if (lock_task_sighand(task, &flags)) {
882                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
883                         oom_adj = OOM_ADJUST_MAX;
884                 else
885                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
886                                   OOM_SCORE_ADJ_MAX;
887                 unlock_task_sighand(task, &flags);
888         }
889         put_task_struct(task);
890         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
891         return simple_read_from_buffer(buf, count, ppos, buffer, len);
892 }
893
894 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
895                              size_t count, loff_t *ppos)
896 {
897         struct task_struct *task;
898         char buffer[PROC_NUMBUF];
899         int oom_adj;
900         unsigned long flags;
901         int err;
902
903         memset(buffer, 0, sizeof(buffer));
904         if (count > sizeof(buffer) - 1)
905                 count = sizeof(buffer) - 1;
906         if (copy_from_user(buffer, buf, count)) {
907                 err = -EFAULT;
908                 goto out;
909         }
910
911         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
912         if (err)
913                 goto out;
914         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
915              oom_adj != OOM_DISABLE) {
916                 err = -EINVAL;
917                 goto out;
918         }
919
920         task = get_proc_task(file_inode(file));
921         if (!task) {
922                 err = -ESRCH;
923                 goto out;
924         }
925
926         task_lock(task);
927         if (!task->mm) {
928                 err = -EINVAL;
929                 goto err_task_lock;
930         }
931
932         if (!lock_task_sighand(task, &flags)) {
933                 err = -ESRCH;
934                 goto err_task_lock;
935         }
936
937         /*
938          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
939          * value is always attainable.
940          */
941         if (oom_adj == OOM_ADJUST_MAX)
942                 oom_adj = OOM_SCORE_ADJ_MAX;
943         else
944                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
945
946         if (oom_adj < task->signal->oom_score_adj &&
947             !capable(CAP_SYS_RESOURCE)) {
948                 err = -EACCES;
949                 goto err_sighand;
950         }
951
952         /*
953          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
954          * /proc/pid/oom_score_adj instead.
955          */
956         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
957                   current->comm, task_pid_nr(current), task_pid_nr(task),
958                   task_pid_nr(task));
959
960         task->signal->oom_score_adj = oom_adj;
961         trace_oom_score_adj_update(task);
962 err_sighand:
963         unlock_task_sighand(task, &flags);
964 err_task_lock:
965         task_unlock(task);
966         put_task_struct(task);
967 out:
968         return err < 0 ? err : count;
969 }
970
971 static const struct file_operations proc_oom_adj_operations = {
972         .read           = oom_adj_read,
973         .write          = oom_adj_write,
974         .llseek         = generic_file_llseek,
975 };
976
977 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
978                                         size_t count, loff_t *ppos)
979 {
980         struct task_struct *task = get_proc_task(file_inode(file));
981         char buffer[PROC_NUMBUF];
982         short oom_score_adj = OOM_SCORE_ADJ_MIN;
983         unsigned long flags;
984         size_t len;
985
986         if (!task)
987                 return -ESRCH;
988         if (lock_task_sighand(task, &flags)) {
989                 oom_score_adj = task->signal->oom_score_adj;
990                 unlock_task_sighand(task, &flags);
991         }
992         put_task_struct(task);
993         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
994         return simple_read_from_buffer(buf, count, ppos, buffer, len);
995 }
996
997 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
998                                         size_t count, loff_t *ppos)
999 {
1000         struct task_struct *task;
1001         char buffer[PROC_NUMBUF];
1002         unsigned long flags;
1003         int oom_score_adj;
1004         int err;
1005
1006         memset(buffer, 0, sizeof(buffer));
1007         if (count > sizeof(buffer) - 1)
1008                 count = sizeof(buffer) - 1;
1009         if (copy_from_user(buffer, buf, count)) {
1010                 err = -EFAULT;
1011                 goto out;
1012         }
1013
1014         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1015         if (err)
1016                 goto out;
1017         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1018                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1019                 err = -EINVAL;
1020                 goto out;
1021         }
1022
1023         task = get_proc_task(file_inode(file));
1024         if (!task) {
1025                 err = -ESRCH;
1026                 goto out;
1027         }
1028
1029         task_lock(task);
1030         if (!task->mm) {
1031                 err = -EINVAL;
1032                 goto err_task_lock;
1033         }
1034
1035         if (!lock_task_sighand(task, &flags)) {
1036                 err = -ESRCH;
1037                 goto err_task_lock;
1038         }
1039
1040         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1041                         !capable(CAP_SYS_RESOURCE)) {
1042                 err = -EACCES;
1043                 goto err_sighand;
1044         }
1045
1046         task->signal->oom_score_adj = (short)oom_score_adj;
1047         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1048                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1049         trace_oom_score_adj_update(task);
1050
1051 err_sighand:
1052         unlock_task_sighand(task, &flags);
1053 err_task_lock:
1054         task_unlock(task);
1055         put_task_struct(task);
1056 out:
1057         return err < 0 ? err : count;
1058 }
1059
1060 static const struct file_operations proc_oom_score_adj_operations = {
1061         .read           = oom_score_adj_read,
1062         .write          = oom_score_adj_write,
1063         .llseek         = default_llseek,
1064 };
1065
1066 #ifdef CONFIG_AUDITSYSCALL
1067 #define TMPBUFLEN 21
1068 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1069                                   size_t count, loff_t *ppos)
1070 {
1071         struct inode * inode = file_inode(file);
1072         struct task_struct *task = get_proc_task(inode);
1073         ssize_t length;
1074         char tmpbuf[TMPBUFLEN];
1075
1076         if (!task)
1077                 return -ESRCH;
1078         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1079                            from_kuid(file->f_cred->user_ns,
1080                                      audit_get_loginuid(task)));
1081         put_task_struct(task);
1082         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1083 }
1084
1085 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1086                                    size_t count, loff_t *ppos)
1087 {
1088         struct inode * inode = file_inode(file);
1089         char *page, *tmp;
1090         ssize_t length;
1091         uid_t loginuid;
1092         kuid_t kloginuid;
1093
1094         rcu_read_lock();
1095         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1096                 rcu_read_unlock();
1097                 return -EPERM;
1098         }
1099         rcu_read_unlock();
1100
1101         if (count >= PAGE_SIZE)
1102                 count = PAGE_SIZE - 1;
1103
1104         if (*ppos != 0) {
1105                 /* No partial writes. */
1106                 return -EINVAL;
1107         }
1108         page = (char*)__get_free_page(GFP_TEMPORARY);
1109         if (!page)
1110                 return -ENOMEM;
1111         length = -EFAULT;
1112         if (copy_from_user(page, buf, count))
1113                 goto out_free_page;
1114
1115         page[count] = '\0';
1116         loginuid = simple_strtoul(page, &tmp, 10);
1117         if (tmp == page) {
1118                 length = -EINVAL;
1119                 goto out_free_page;
1120
1121         }
1122         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1123         if (!uid_valid(kloginuid)) {
1124                 length = -EINVAL;
1125                 goto out_free_page;
1126         }
1127
1128         length = audit_set_loginuid(kloginuid);
1129         if (likely(length == 0))
1130                 length = count;
1131
1132 out_free_page:
1133         free_page((unsigned long) page);
1134         return length;
1135 }
1136
1137 static const struct file_operations proc_loginuid_operations = {
1138         .read           = proc_loginuid_read,
1139         .write          = proc_loginuid_write,
1140         .llseek         = generic_file_llseek,
1141 };
1142
1143 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1144                                   size_t count, loff_t *ppos)
1145 {
1146         struct inode * inode = file_inode(file);
1147         struct task_struct *task = get_proc_task(inode);
1148         ssize_t length;
1149         char tmpbuf[TMPBUFLEN];
1150
1151         if (!task)
1152                 return -ESRCH;
1153         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1154                                 audit_get_sessionid(task));
1155         put_task_struct(task);
1156         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1157 }
1158
1159 static const struct file_operations proc_sessionid_operations = {
1160         .read           = proc_sessionid_read,
1161         .llseek         = generic_file_llseek,
1162 };
1163 #endif
1164
1165 #ifdef CONFIG_FAULT_INJECTION
1166 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1167                                       size_t count, loff_t *ppos)
1168 {
1169         struct task_struct *task = get_proc_task(file_inode(file));
1170         char buffer[PROC_NUMBUF];
1171         size_t len;
1172         int make_it_fail;
1173
1174         if (!task)
1175                 return -ESRCH;
1176         make_it_fail = task->make_it_fail;
1177         put_task_struct(task);
1178
1179         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1180
1181         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1182 }
1183
1184 static ssize_t proc_fault_inject_write(struct file * file,
1185                         const char __user * buf, size_t count, loff_t *ppos)
1186 {
1187         struct task_struct *task;
1188         char buffer[PROC_NUMBUF], *end;
1189         int make_it_fail;
1190
1191         if (!capable(CAP_SYS_RESOURCE))
1192                 return -EPERM;
1193         memset(buffer, 0, sizeof(buffer));
1194         if (count > sizeof(buffer) - 1)
1195                 count = sizeof(buffer) - 1;
1196         if (copy_from_user(buffer, buf, count))
1197                 return -EFAULT;
1198         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1199         if (*end)
1200                 return -EINVAL;
1201         task = get_proc_task(file_inode(file));
1202         if (!task)
1203                 return -ESRCH;
1204         task->make_it_fail = make_it_fail;
1205         put_task_struct(task);
1206
1207         return count;
1208 }
1209
1210 static const struct file_operations proc_fault_inject_operations = {
1211         .read           = proc_fault_inject_read,
1212         .write          = proc_fault_inject_write,
1213         .llseek         = generic_file_llseek,
1214 };
1215 #endif
1216
1217
1218 #ifdef CONFIG_SCHED_DEBUG
1219 /*
1220  * Print out various scheduling related per-task fields:
1221  */
1222 static int sched_show(struct seq_file *m, void *v)
1223 {
1224         struct inode *inode = m->private;
1225         struct task_struct *p;
1226
1227         p = get_proc_task(inode);
1228         if (!p)
1229                 return -ESRCH;
1230         proc_sched_show_task(p, m);
1231
1232         put_task_struct(p);
1233
1234         return 0;
1235 }
1236
1237 static ssize_t
1238 sched_write(struct file *file, const char __user *buf,
1239             size_t count, loff_t *offset)
1240 {
1241         struct inode *inode = file_inode(file);
1242         struct task_struct *p;
1243
1244         p = get_proc_task(inode);
1245         if (!p)
1246                 return -ESRCH;
1247         proc_sched_set_task(p);
1248
1249         put_task_struct(p);
1250
1251         return count;
1252 }
1253
1254 static int sched_open(struct inode *inode, struct file *filp)
1255 {
1256         return single_open(filp, sched_show, inode);
1257 }
1258
1259 static const struct file_operations proc_pid_sched_operations = {
1260         .open           = sched_open,
1261         .read           = seq_read,
1262         .write          = sched_write,
1263         .llseek         = seq_lseek,
1264         .release        = single_release,
1265 };
1266
1267 #endif
1268
1269 #ifdef CONFIG_SCHED_AUTOGROUP
1270 /*
1271  * Print out autogroup related information:
1272  */
1273 static int sched_autogroup_show(struct seq_file *m, void *v)
1274 {
1275         struct inode *inode = m->private;
1276         struct task_struct *p;
1277
1278         p = get_proc_task(inode);
1279         if (!p)
1280                 return -ESRCH;
1281         proc_sched_autogroup_show_task(p, m);
1282
1283         put_task_struct(p);
1284
1285         return 0;
1286 }
1287
1288 static ssize_t
1289 sched_autogroup_write(struct file *file, const char __user *buf,
1290             size_t count, loff_t *offset)
1291 {
1292         struct inode *inode = file_inode(file);
1293         struct task_struct *p;
1294         char buffer[PROC_NUMBUF];
1295         int nice;
1296         int err;
1297
1298         memset(buffer, 0, sizeof(buffer));
1299         if (count > sizeof(buffer) - 1)
1300                 count = sizeof(buffer) - 1;
1301         if (copy_from_user(buffer, buf, count))
1302                 return -EFAULT;
1303
1304         err = kstrtoint(strstrip(buffer), 0, &nice);
1305         if (err < 0)
1306                 return err;
1307
1308         p = get_proc_task(inode);
1309         if (!p)
1310                 return -ESRCH;
1311
1312         err = proc_sched_autogroup_set_nice(p, nice);
1313         if (err)
1314                 count = err;
1315
1316         put_task_struct(p);
1317
1318         return count;
1319 }
1320
1321 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1322 {
1323         int ret;
1324
1325         ret = single_open(filp, sched_autogroup_show, NULL);
1326         if (!ret) {
1327                 struct seq_file *m = filp->private_data;
1328
1329                 m->private = inode;
1330         }
1331         return ret;
1332 }
1333
1334 static const struct file_operations proc_pid_sched_autogroup_operations = {
1335         .open           = sched_autogroup_open,
1336         .read           = seq_read,
1337         .write          = sched_autogroup_write,
1338         .llseek         = seq_lseek,
1339         .release        = single_release,
1340 };
1341
1342 #endif /* CONFIG_SCHED_AUTOGROUP */
1343
1344 static ssize_t comm_write(struct file *file, const char __user *buf,
1345                                 size_t count, loff_t *offset)
1346 {
1347         struct inode *inode = file_inode(file);
1348         struct task_struct *p;
1349         char buffer[TASK_COMM_LEN];
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         p = get_proc_task(inode);
1358         if (!p)
1359                 return -ESRCH;
1360
1361         if (same_thread_group(current, p))
1362                 set_task_comm(p, buffer);
1363         else
1364                 count = -EINVAL;
1365
1366         put_task_struct(p);
1367
1368         return count;
1369 }
1370
1371 static int comm_show(struct seq_file *m, void *v)
1372 {
1373         struct inode *inode = m->private;
1374         struct task_struct *p;
1375
1376         p = get_proc_task(inode);
1377         if (!p)
1378                 return -ESRCH;
1379
1380         task_lock(p);
1381         seq_printf(m, "%s\n", p->comm);
1382         task_unlock(p);
1383
1384         put_task_struct(p);
1385
1386         return 0;
1387 }
1388
1389 static int comm_open(struct inode *inode, struct file *filp)
1390 {
1391         return single_open(filp, comm_show, inode);
1392 }
1393
1394 static const struct file_operations proc_pid_set_comm_operations = {
1395         .open           = comm_open,
1396         .read           = seq_read,
1397         .write          = comm_write,
1398         .llseek         = seq_lseek,
1399         .release        = single_release,
1400 };
1401
1402 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1403 {
1404         struct task_struct *task;
1405         struct mm_struct *mm;
1406         struct file *exe_file;
1407
1408         task = get_proc_task(dentry->d_inode);
1409         if (!task)
1410                 return -ENOENT;
1411         mm = get_task_mm(task);
1412         put_task_struct(task);
1413         if (!mm)
1414                 return -ENOENT;
1415         exe_file = get_mm_exe_file(mm);
1416         mmput(mm);
1417         if (exe_file) {
1418                 *exe_path = exe_file->f_path;
1419                 path_get(&exe_file->f_path);
1420                 fput(exe_file);
1421                 return 0;
1422         } else
1423                 return -ENOENT;
1424 }
1425
1426 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1427 {
1428         struct inode *inode = dentry->d_inode;
1429         struct path path;
1430         int error = -EACCES;
1431
1432         /* Are we allowed to snoop on the tasks file descriptors? */
1433         if (!proc_fd_access_allowed(inode))
1434                 goto out;
1435
1436         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1437         if (error)
1438                 goto out;
1439
1440         nd_jump_link(nd, &path);
1441         return NULL;
1442 out:
1443         return ERR_PTR(error);
1444 }
1445
1446 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1447 {
1448         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1449         char *pathname;
1450         int len;
1451
1452         if (!tmp)
1453                 return -ENOMEM;
1454
1455         pathname = d_path(path, tmp, PAGE_SIZE);
1456         len = PTR_ERR(pathname);
1457         if (IS_ERR(pathname))
1458                 goto out;
1459         len = tmp + PAGE_SIZE - 1 - pathname;
1460
1461         if (len > buflen)
1462                 len = buflen;
1463         if (copy_to_user(buffer, pathname, len))
1464                 len = -EFAULT;
1465  out:
1466         free_page((unsigned long)tmp);
1467         return len;
1468 }
1469
1470 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1471 {
1472         int error = -EACCES;
1473         struct inode *inode = dentry->d_inode;
1474         struct path path;
1475
1476         /* Are we allowed to snoop on the tasks file descriptors? */
1477         if (!proc_fd_access_allowed(inode))
1478                 goto out;
1479
1480         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1481         if (error)
1482                 goto out;
1483
1484         error = do_proc_readlink(&path, buffer, buflen);
1485         path_put(&path);
1486 out:
1487         return error;
1488 }
1489
1490 const struct inode_operations proc_pid_link_inode_operations = {
1491         .readlink       = proc_pid_readlink,
1492         .follow_link    = proc_pid_follow_link,
1493         .setattr        = proc_setattr,
1494 };
1495
1496
1497 /* building an inode */
1498
1499 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1500 {
1501         struct inode * inode;
1502         struct proc_inode *ei;
1503         const struct cred *cred;
1504
1505         /* We need a new inode */
1506
1507         inode = new_inode(sb);
1508         if (!inode)
1509                 goto out;
1510
1511         /* Common stuff */
1512         ei = PROC_I(inode);
1513         inode->i_ino = get_next_ino();
1514         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1515         inode->i_op = &proc_def_inode_operations;
1516
1517         /*
1518          * grab the reference to task.
1519          */
1520         ei->pid = get_task_pid(task, PIDTYPE_PID);
1521         if (!ei->pid)
1522                 goto out_unlock;
1523
1524         if (task_dumpable(task)) {
1525                 rcu_read_lock();
1526                 cred = __task_cred(task);
1527                 inode->i_uid = cred->euid;
1528                 inode->i_gid = cred->egid;
1529                 rcu_read_unlock();
1530         }
1531         security_task_to_inode(task, inode);
1532
1533 out:
1534         return inode;
1535
1536 out_unlock:
1537         iput(inode);
1538         return NULL;
1539 }
1540
1541 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1542 {
1543         struct inode *inode = dentry->d_inode;
1544         struct task_struct *task;
1545         const struct cred *cred;
1546         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1547
1548         generic_fillattr(inode, stat);
1549
1550         rcu_read_lock();
1551         stat->uid = GLOBAL_ROOT_UID;
1552         stat->gid = GLOBAL_ROOT_GID;
1553         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1554         if (task) {
1555                 if (!has_pid_permissions(pid, task, 2)) {
1556                         rcu_read_unlock();
1557                         /*
1558                          * This doesn't prevent learning whether PID exists,
1559                          * it only makes getattr() consistent with readdir().
1560                          */
1561                         return -ENOENT;
1562                 }
1563                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1564                     task_dumpable(task)) {
1565                         cred = __task_cred(task);
1566                         stat->uid = cred->euid;
1567                         stat->gid = cred->egid;
1568                 }
1569         }
1570         rcu_read_unlock();
1571         return 0;
1572 }
1573
1574 /* dentry stuff */
1575
1576 /*
1577  *      Exceptional case: normally we are not allowed to unhash a busy
1578  * directory. In this case, however, we can do it - no aliasing problems
1579  * due to the way we treat inodes.
1580  *
1581  * Rewrite the inode's ownerships here because the owning task may have
1582  * performed a setuid(), etc.
1583  *
1584  * Before the /proc/pid/status file was created the only way to read
1585  * the effective uid of a /process was to stat /proc/pid.  Reading
1586  * /proc/pid/status is slow enough that procps and other packages
1587  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1588  * made this apply to all per process world readable and executable
1589  * directories.
1590  */
1591 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1592 {
1593         struct inode *inode;
1594         struct task_struct *task;
1595         const struct cred *cred;
1596
1597         if (flags & LOOKUP_RCU)
1598                 return -ECHILD;
1599
1600         inode = dentry->d_inode;
1601         task = get_proc_task(inode);
1602
1603         if (task) {
1604                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1605                     task_dumpable(task)) {
1606                         rcu_read_lock();
1607                         cred = __task_cred(task);
1608                         inode->i_uid = cred->euid;
1609                         inode->i_gid = cred->egid;
1610                         rcu_read_unlock();
1611                 } else {
1612                         inode->i_uid = GLOBAL_ROOT_UID;
1613                         inode->i_gid = GLOBAL_ROOT_GID;
1614                 }
1615                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1616                 security_task_to_inode(task, inode);
1617                 put_task_struct(task);
1618                 return 1;
1619         }
1620         d_drop(dentry);
1621         return 0;
1622 }
1623
1624 const struct dentry_operations pid_dentry_operations =
1625 {
1626         .d_revalidate   = pid_revalidate,
1627         .d_delete       = pid_delete_dentry,
1628 };
1629
1630 /* Lookups */
1631
1632 /*
1633  * Fill a directory entry.
1634  *
1635  * If possible create the dcache entry and derive our inode number and
1636  * file type from dcache entry.
1637  *
1638  * Since all of the proc inode numbers are dynamically generated, the inode
1639  * numbers do not exist until the inode is cache.  This means creating the
1640  * the dcache entry in readdir is necessary to keep the inode numbers
1641  * reported by readdir in sync with the inode numbers reported
1642  * by stat.
1643  */
1644 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1645         const char *name, int len,
1646         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1647 {
1648         struct dentry *child, *dir = filp->f_path.dentry;
1649         struct inode *inode;
1650         struct qstr qname;
1651         ino_t ino = 0;
1652         unsigned type = DT_UNKNOWN;
1653
1654         qname.name = name;
1655         qname.len  = len;
1656         qname.hash = full_name_hash(name, len);
1657
1658         child = d_lookup(dir, &qname);
1659         if (!child) {
1660                 struct dentry *new;
1661                 new = d_alloc(dir, &qname);
1662                 if (new) {
1663                         child = instantiate(dir->d_inode, new, task, ptr);
1664                         if (child)
1665                                 dput(new);
1666                         else
1667                                 child = new;
1668                 }
1669         }
1670         if (!child || IS_ERR(child) || !child->d_inode)
1671                 goto end_instantiate;
1672         inode = child->d_inode;
1673         if (inode) {
1674                 ino = inode->i_ino;
1675                 type = inode->i_mode >> 12;
1676         }
1677         dput(child);
1678 end_instantiate:
1679         if (!ino)
1680                 ino = find_inode_number(dir, &qname);
1681         if (!ino)
1682                 ino = 1;
1683         return filldir(dirent, name, len, filp->f_pos, ino, type);
1684 }
1685
1686 #ifdef CONFIG_CHECKPOINT_RESTORE
1687
1688 /*
1689  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1690  * which represent vma start and end addresses.
1691  */
1692 static int dname_to_vma_addr(struct dentry *dentry,
1693                              unsigned long *start, unsigned long *end)
1694 {
1695         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1696                 return -EINVAL;
1697
1698         return 0;
1699 }
1700
1701 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1702 {
1703         unsigned long vm_start, vm_end;
1704         bool exact_vma_exists = false;
1705         struct mm_struct *mm = NULL;
1706         struct task_struct *task;
1707         const struct cred *cred;
1708         struct inode *inode;
1709         int status = 0;
1710
1711         if (flags & LOOKUP_RCU)
1712                 return -ECHILD;
1713
1714         if (!capable(CAP_SYS_ADMIN)) {
1715                 status = -EPERM;
1716                 goto out_notask;
1717         }
1718
1719         inode = dentry->d_inode;
1720         task = get_proc_task(inode);
1721         if (!task)
1722                 goto out_notask;
1723
1724         mm = mm_access(task, PTRACE_MODE_READ);
1725         if (IS_ERR_OR_NULL(mm))
1726                 goto out;
1727
1728         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1729                 down_read(&mm->mmap_sem);
1730                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1731                 up_read(&mm->mmap_sem);
1732         }
1733
1734         mmput(mm);
1735
1736         if (exact_vma_exists) {
1737                 if (task_dumpable(task)) {
1738                         rcu_read_lock();
1739                         cred = __task_cred(task);
1740                         inode->i_uid = cred->euid;
1741                         inode->i_gid = cred->egid;
1742                         rcu_read_unlock();
1743                 } else {
1744                         inode->i_uid = GLOBAL_ROOT_UID;
1745                         inode->i_gid = GLOBAL_ROOT_GID;
1746                 }
1747                 security_task_to_inode(task, inode);
1748                 status = 1;
1749         }
1750
1751 out:
1752         put_task_struct(task);
1753
1754 out_notask:
1755         if (status <= 0)
1756                 d_drop(dentry);
1757
1758         return status;
1759 }
1760
1761 static const struct dentry_operations tid_map_files_dentry_operations = {
1762         .d_revalidate   = map_files_d_revalidate,
1763         .d_delete       = pid_delete_dentry,
1764 };
1765
1766 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1767 {
1768         unsigned long vm_start, vm_end;
1769         struct vm_area_struct *vma;
1770         struct task_struct *task;
1771         struct mm_struct *mm;
1772         int rc;
1773
1774         rc = -ENOENT;
1775         task = get_proc_task(dentry->d_inode);
1776         if (!task)
1777                 goto out;
1778
1779         mm = get_task_mm(task);
1780         put_task_struct(task);
1781         if (!mm)
1782                 goto out;
1783
1784         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1785         if (rc)
1786                 goto out_mmput;
1787
1788         down_read(&mm->mmap_sem);
1789         vma = find_exact_vma(mm, vm_start, vm_end);
1790         if (vma && vma->vm_file) {
1791                 *path = vma->vm_file->f_path;
1792                 path_get(path);
1793                 rc = 0;
1794         }
1795         up_read(&mm->mmap_sem);
1796
1797 out_mmput:
1798         mmput(mm);
1799 out:
1800         return rc;
1801 }
1802
1803 struct map_files_info {
1804         fmode_t         mode;
1805         unsigned long   len;
1806         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1807 };
1808
1809 static struct dentry *
1810 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1811                            struct task_struct *task, const void *ptr)
1812 {
1813         fmode_t mode = (fmode_t)(unsigned long)ptr;
1814         struct proc_inode *ei;
1815         struct inode *inode;
1816
1817         inode = proc_pid_make_inode(dir->i_sb, task);
1818         if (!inode)
1819                 return ERR_PTR(-ENOENT);
1820
1821         ei = PROC_I(inode);
1822         ei->op.proc_get_link = proc_map_files_get_link;
1823
1824         inode->i_op = &proc_pid_link_inode_operations;
1825         inode->i_size = 64;
1826         inode->i_mode = S_IFLNK;
1827
1828         if (mode & FMODE_READ)
1829                 inode->i_mode |= S_IRUSR;
1830         if (mode & FMODE_WRITE)
1831                 inode->i_mode |= S_IWUSR;
1832
1833         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1834         d_add(dentry, inode);
1835
1836         return NULL;
1837 }
1838
1839 static struct dentry *proc_map_files_lookup(struct inode *dir,
1840                 struct dentry *dentry, unsigned int flags)
1841 {
1842         unsigned long vm_start, vm_end;
1843         struct vm_area_struct *vma;
1844         struct task_struct *task;
1845         struct dentry *result;
1846         struct mm_struct *mm;
1847
1848         result = ERR_PTR(-EPERM);
1849         if (!capable(CAP_SYS_ADMIN))
1850                 goto out;
1851
1852         result = ERR_PTR(-ENOENT);
1853         task = get_proc_task(dir);
1854         if (!task)
1855                 goto out;
1856
1857         result = ERR_PTR(-EACCES);
1858         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1859                 goto out_put_task;
1860
1861         result = ERR_PTR(-ENOENT);
1862         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1863                 goto out_put_task;
1864
1865         mm = get_task_mm(task);
1866         if (!mm)
1867                 goto out_put_task;
1868
1869         down_read(&mm->mmap_sem);
1870         vma = find_exact_vma(mm, vm_start, vm_end);
1871         if (!vma)
1872                 goto out_no_vma;
1873
1874         if (vma->vm_file)
1875                 result = proc_map_files_instantiate(dir, dentry, task,
1876                                 (void *)(unsigned long)vma->vm_file->f_mode);
1877
1878 out_no_vma:
1879         up_read(&mm->mmap_sem);
1880         mmput(mm);
1881 out_put_task:
1882         put_task_struct(task);
1883 out:
1884         return result;
1885 }
1886
1887 static const struct inode_operations proc_map_files_inode_operations = {
1888         .lookup         = proc_map_files_lookup,
1889         .permission     = proc_fd_permission,
1890         .setattr        = proc_setattr,
1891 };
1892
1893 static int
1894 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1895 {
1896         struct dentry *dentry = filp->f_path.dentry;
1897         struct inode *inode = dentry->d_inode;
1898         struct vm_area_struct *vma;
1899         struct task_struct *task;
1900         struct mm_struct *mm;
1901         ino_t ino;
1902         int ret;
1903
1904         ret = -EPERM;
1905         if (!capable(CAP_SYS_ADMIN))
1906                 goto out;
1907
1908         ret = -ENOENT;
1909         task = get_proc_task(inode);
1910         if (!task)
1911                 goto out;
1912
1913         ret = -EACCES;
1914         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1915                 goto out_put_task;
1916
1917         ret = 0;
1918         switch (filp->f_pos) {
1919         case 0:
1920                 ino = inode->i_ino;
1921                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1922                         goto out_put_task;
1923                 filp->f_pos++;
1924         case 1:
1925                 ino = parent_ino(dentry);
1926                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1927                         goto out_put_task;
1928                 filp->f_pos++;
1929         default:
1930         {
1931                 unsigned long nr_files, pos, i;
1932                 struct flex_array *fa = NULL;
1933                 struct map_files_info info;
1934                 struct map_files_info *p;
1935
1936                 mm = get_task_mm(task);
1937                 if (!mm)
1938                         goto out_put_task;
1939                 down_read(&mm->mmap_sem);
1940
1941                 nr_files = 0;
1942
1943                 /*
1944                  * We need two passes here:
1945                  *
1946                  *  1) Collect vmas of mapped files with mmap_sem taken
1947                  *  2) Release mmap_sem and instantiate entries
1948                  *
1949                  * otherwise we get lockdep complained, since filldir()
1950                  * routine might require mmap_sem taken in might_fault().
1951                  */
1952
1953                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1954                         if (vma->vm_file && ++pos > filp->f_pos)
1955                                 nr_files++;
1956                 }
1957
1958                 if (nr_files) {
1959                         fa = flex_array_alloc(sizeof(info), nr_files,
1960                                                 GFP_KERNEL);
1961                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1962                                                         GFP_KERNEL)) {
1963                                 ret = -ENOMEM;
1964                                 if (fa)
1965                                         flex_array_free(fa);
1966                                 up_read(&mm->mmap_sem);
1967                                 mmput(mm);
1968                                 goto out_put_task;
1969                         }
1970                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1971                                         vma = vma->vm_next) {
1972                                 if (!vma->vm_file)
1973                                         continue;
1974                                 if (++pos <= filp->f_pos)
1975                                         continue;
1976
1977                                 info.mode = vma->vm_file->f_mode;
1978                                 info.len = snprintf(info.name,
1979                                                 sizeof(info.name), "%lx-%lx",
1980                                                 vma->vm_start, vma->vm_end);
1981                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1982                                         BUG();
1983                         }
1984                 }
1985                 up_read(&mm->mmap_sem);
1986
1987                 for (i = 0; i < nr_files; i++) {
1988                         p = flex_array_get(fa, i);
1989                         ret = proc_fill_cache(filp, dirent, filldir,
1990                                               p->name, p->len,
1991                                               proc_map_files_instantiate,
1992                                               task,
1993                                               (void *)(unsigned long)p->mode);
1994                         if (ret)
1995                                 break;
1996                         filp->f_pos++;
1997                 }
1998                 if (fa)
1999                         flex_array_free(fa);
2000                 mmput(mm);
2001         }
2002         }
2003
2004 out_put_task:
2005         put_task_struct(task);
2006 out:
2007         return ret;
2008 }
2009
2010 static const struct file_operations proc_map_files_operations = {
2011         .read           = generic_read_dir,
2012         .readdir        = proc_map_files_readdir,
2013         .llseek         = default_llseek,
2014 };
2015
2016 #endif /* CONFIG_CHECKPOINT_RESTORE */
2017
2018 static struct dentry *proc_pident_instantiate(struct inode *dir,
2019         struct dentry *dentry, struct task_struct *task, const void *ptr)
2020 {
2021         const struct pid_entry *p = ptr;
2022         struct inode *inode;
2023         struct proc_inode *ei;
2024         struct dentry *error = ERR_PTR(-ENOENT);
2025
2026         inode = proc_pid_make_inode(dir->i_sb, task);
2027         if (!inode)
2028                 goto out;
2029
2030         ei = PROC_I(inode);
2031         inode->i_mode = p->mode;
2032         if (S_ISDIR(inode->i_mode))
2033                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2034         if (p->iop)
2035                 inode->i_op = p->iop;
2036         if (p->fop)
2037                 inode->i_fop = p->fop;
2038         ei->op = p->op;
2039         d_set_d_op(dentry, &pid_dentry_operations);
2040         d_add(dentry, inode);
2041         /* Close the race of the process dying before we return the dentry */
2042         if (pid_revalidate(dentry, 0))
2043                 error = NULL;
2044 out:
2045         return error;
2046 }
2047
2048 static struct dentry *proc_pident_lookup(struct inode *dir, 
2049                                          struct dentry *dentry,
2050                                          const struct pid_entry *ents,
2051                                          unsigned int nents)
2052 {
2053         struct dentry *error;
2054         struct task_struct *task = get_proc_task(dir);
2055         const struct pid_entry *p, *last;
2056
2057         error = ERR_PTR(-ENOENT);
2058
2059         if (!task)
2060                 goto out_no_task;
2061
2062         /*
2063          * Yes, it does not scale. And it should not. Don't add
2064          * new entries into /proc/<tgid>/ without very good reasons.
2065          */
2066         last = &ents[nents - 1];
2067         for (p = ents; p <= last; p++) {
2068                 if (p->len != dentry->d_name.len)
2069                         continue;
2070                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2071                         break;
2072         }
2073         if (p > last)
2074                 goto out;
2075
2076         error = proc_pident_instantiate(dir, dentry, task, p);
2077 out:
2078         put_task_struct(task);
2079 out_no_task:
2080         return error;
2081 }
2082
2083 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2084         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2085 {
2086         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2087                                 proc_pident_instantiate, task, p);
2088 }
2089
2090 static int proc_pident_readdir(struct file *filp,
2091                 void *dirent, filldir_t filldir,
2092                 const struct pid_entry *ents, unsigned int nents)
2093 {
2094         int i;
2095         struct dentry *dentry = filp->f_path.dentry;
2096         struct inode *inode = dentry->d_inode;
2097         struct task_struct *task = get_proc_task(inode);
2098         const struct pid_entry *p, *last;
2099         ino_t ino;
2100         int ret;
2101
2102         ret = -ENOENT;
2103         if (!task)
2104                 goto out_no_task;
2105
2106         ret = 0;
2107         i = filp->f_pos;
2108         switch (i) {
2109         case 0:
2110                 ino = inode->i_ino;
2111                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2112                         goto out;
2113                 i++;
2114                 filp->f_pos++;
2115                 /* fall through */
2116         case 1:
2117                 ino = parent_ino(dentry);
2118                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2119                         goto out;
2120                 i++;
2121                 filp->f_pos++;
2122                 /* fall through */
2123         default:
2124                 i -= 2;
2125                 if (i >= nents) {
2126                         ret = 1;
2127                         goto out;
2128                 }
2129                 p = ents + i;
2130                 last = &ents[nents - 1];
2131                 while (p <= last) {
2132                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2133                                 goto out;
2134                         filp->f_pos++;
2135                         p++;
2136                 }
2137         }
2138
2139         ret = 1;
2140 out:
2141         put_task_struct(task);
2142 out_no_task:
2143         return ret;
2144 }
2145
2146 #ifdef CONFIG_SECURITY
2147 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2148                                   size_t count, loff_t *ppos)
2149 {
2150         struct inode * inode = file_inode(file);
2151         char *p = NULL;
2152         ssize_t length;
2153         struct task_struct *task = get_proc_task(inode);
2154
2155         if (!task)
2156                 return -ESRCH;
2157
2158         length = security_getprocattr(task,
2159                                       (char*)file->f_path.dentry->d_name.name,
2160                                       &p);
2161         put_task_struct(task);
2162         if (length > 0)
2163                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2164         kfree(p);
2165         return length;
2166 }
2167
2168 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2169                                    size_t count, loff_t *ppos)
2170 {
2171         struct inode * inode = file_inode(file);
2172         char *page;
2173         ssize_t length;
2174         struct task_struct *task = get_proc_task(inode);
2175
2176         length = -ESRCH;
2177         if (!task)
2178                 goto out_no_task;
2179         if (count > PAGE_SIZE)
2180                 count = PAGE_SIZE;
2181
2182         /* No partial writes. */
2183         length = -EINVAL;
2184         if (*ppos != 0)
2185                 goto out;
2186
2187         length = -ENOMEM;
2188         page = (char*)__get_free_page(GFP_TEMPORARY);
2189         if (!page)
2190                 goto out;
2191
2192         length = -EFAULT;
2193         if (copy_from_user(page, buf, count))
2194                 goto out_free;
2195
2196         /* Guard against adverse ptrace interaction */
2197         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2198         if (length < 0)
2199                 goto out_free;
2200
2201         length = security_setprocattr(task,
2202                                       (char*)file->f_path.dentry->d_name.name,
2203                                       (void*)page, count);
2204         mutex_unlock(&task->signal->cred_guard_mutex);
2205 out_free:
2206         free_page((unsigned long) page);
2207 out:
2208         put_task_struct(task);
2209 out_no_task:
2210         return length;
2211 }
2212
2213 static const struct file_operations proc_pid_attr_operations = {
2214         .read           = proc_pid_attr_read,
2215         .write          = proc_pid_attr_write,
2216         .llseek         = generic_file_llseek,
2217 };
2218
2219 static const struct pid_entry attr_dir_stuff[] = {
2220         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2221         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2222         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2223         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2224         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2225         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2226 };
2227
2228 static int proc_attr_dir_readdir(struct file * filp,
2229                              void * dirent, filldir_t filldir)
2230 {
2231         return proc_pident_readdir(filp,dirent,filldir,
2232                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2233 }
2234
2235 static const struct file_operations proc_attr_dir_operations = {
2236         .read           = generic_read_dir,
2237         .readdir        = proc_attr_dir_readdir,
2238         .llseek         = default_llseek,
2239 };
2240
2241 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2242                                 struct dentry *dentry, unsigned int flags)
2243 {
2244         return proc_pident_lookup(dir, dentry,
2245                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2246 }
2247
2248 static const struct inode_operations proc_attr_dir_inode_operations = {
2249         .lookup         = proc_attr_dir_lookup,
2250         .getattr        = pid_getattr,
2251         .setattr        = proc_setattr,
2252 };
2253
2254 #endif
2255
2256 #ifdef CONFIG_ELF_CORE
2257 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2258                                          size_t count, loff_t *ppos)
2259 {
2260         struct task_struct *task = get_proc_task(file_inode(file));
2261         struct mm_struct *mm;
2262         char buffer[PROC_NUMBUF];
2263         size_t len;
2264         int ret;
2265
2266         if (!task)
2267                 return -ESRCH;
2268
2269         ret = 0;
2270         mm = get_task_mm(task);
2271         if (mm) {
2272                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2273                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2274                                 MMF_DUMP_FILTER_SHIFT));
2275                 mmput(mm);
2276                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2277         }
2278
2279         put_task_struct(task);
2280
2281         return ret;
2282 }
2283
2284 static ssize_t proc_coredump_filter_write(struct file *file,
2285                                           const char __user *buf,
2286                                           size_t count,
2287                                           loff_t *ppos)
2288 {
2289         struct task_struct *task;
2290         struct mm_struct *mm;
2291         char buffer[PROC_NUMBUF], *end;
2292         unsigned int val;
2293         int ret;
2294         int i;
2295         unsigned long mask;
2296
2297         ret = -EFAULT;
2298         memset(buffer, 0, sizeof(buffer));
2299         if (count > sizeof(buffer) - 1)
2300                 count = sizeof(buffer) - 1;
2301         if (copy_from_user(buffer, buf, count))
2302                 goto out_no_task;
2303
2304         ret = -EINVAL;
2305         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2306         if (*end == '\n')
2307                 end++;
2308         if (end - buffer == 0)
2309                 goto out_no_task;
2310
2311         ret = -ESRCH;
2312         task = get_proc_task(file_inode(file));
2313         if (!task)
2314                 goto out_no_task;
2315
2316         ret = end - buffer;
2317         mm = get_task_mm(task);
2318         if (!mm)
2319                 goto out_no_mm;
2320
2321         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2322                 if (val & mask)
2323                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2324                 else
2325                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2326         }
2327
2328         mmput(mm);
2329  out_no_mm:
2330         put_task_struct(task);
2331  out_no_task:
2332         return ret;
2333 }
2334
2335 static const struct file_operations proc_coredump_filter_operations = {
2336         .read           = proc_coredump_filter_read,
2337         .write          = proc_coredump_filter_write,
2338         .llseek         = generic_file_llseek,
2339 };
2340 #endif
2341
2342 #ifdef CONFIG_TASK_IO_ACCOUNTING
2343 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2344 {
2345         struct task_io_accounting acct = task->ioac;
2346         unsigned long flags;
2347         int result;
2348
2349         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2350         if (result)
2351                 return result;
2352
2353         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2354                 result = -EACCES;
2355                 goto out_unlock;
2356         }
2357
2358         if (whole && lock_task_sighand(task, &flags)) {
2359                 struct task_struct *t = task;
2360
2361                 task_io_accounting_add(&acct, &task->signal->ioac);
2362                 while_each_thread(task, t)
2363                         task_io_accounting_add(&acct, &t->ioac);
2364
2365                 unlock_task_sighand(task, &flags);
2366         }
2367         result = sprintf(buffer,
2368                         "rchar: %llu\n"
2369                         "wchar: %llu\n"
2370                         "syscr: %llu\n"
2371                         "syscw: %llu\n"
2372                         "read_bytes: %llu\n"
2373                         "write_bytes: %llu\n"
2374                         "cancelled_write_bytes: %llu\n",
2375                         (unsigned long long)acct.rchar,
2376                         (unsigned long long)acct.wchar,
2377                         (unsigned long long)acct.syscr,
2378                         (unsigned long long)acct.syscw,
2379                         (unsigned long long)acct.read_bytes,
2380                         (unsigned long long)acct.write_bytes,
2381                         (unsigned long long)acct.cancelled_write_bytes);
2382 out_unlock:
2383         mutex_unlock(&task->signal->cred_guard_mutex);
2384         return result;
2385 }
2386
2387 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2388 {
2389         return do_io_accounting(task, buffer, 0);
2390 }
2391
2392 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2393 {
2394         return do_io_accounting(task, buffer, 1);
2395 }
2396 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2397
2398 #ifdef CONFIG_USER_NS
2399 static int proc_id_map_open(struct inode *inode, struct file *file,
2400         struct seq_operations *seq_ops)
2401 {
2402         struct user_namespace *ns = NULL;
2403         struct task_struct *task;
2404         struct seq_file *seq;
2405         int ret = -EINVAL;
2406
2407         task = get_proc_task(inode);
2408         if (task) {
2409                 rcu_read_lock();
2410                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2411                 rcu_read_unlock();
2412                 put_task_struct(task);
2413         }
2414         if (!ns)
2415                 goto err;
2416
2417         ret = seq_open(file, seq_ops);
2418         if (ret)
2419                 goto err_put_ns;
2420
2421         seq = file->private_data;
2422         seq->private = ns;
2423
2424         return 0;
2425 err_put_ns:
2426         put_user_ns(ns);
2427 err:
2428         return ret;
2429 }
2430
2431 static int proc_id_map_release(struct inode *inode, struct file *file)
2432 {
2433         struct seq_file *seq = file->private_data;
2434         struct user_namespace *ns = seq->private;
2435         put_user_ns(ns);
2436         return seq_release(inode, file);
2437 }
2438
2439 static int proc_uid_map_open(struct inode *inode, struct file *file)
2440 {
2441         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2442 }
2443
2444 static int proc_gid_map_open(struct inode *inode, struct file *file)
2445 {
2446         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2447 }
2448
2449 static int proc_projid_map_open(struct inode *inode, struct file *file)
2450 {
2451         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2452 }
2453
2454 static const struct file_operations proc_uid_map_operations = {
2455         .open           = proc_uid_map_open,
2456         .write          = proc_uid_map_write,
2457         .read           = seq_read,
2458         .llseek         = seq_lseek,
2459         .release        = proc_id_map_release,
2460 };
2461
2462 static const struct file_operations proc_gid_map_operations = {
2463         .open           = proc_gid_map_open,
2464         .write          = proc_gid_map_write,
2465         .read           = seq_read,
2466         .llseek         = seq_lseek,
2467         .release        = proc_id_map_release,
2468 };
2469
2470 static const struct file_operations proc_projid_map_operations = {
2471         .open           = proc_projid_map_open,
2472         .write          = proc_projid_map_write,
2473         .read           = seq_read,
2474         .llseek         = seq_lseek,
2475         .release        = proc_id_map_release,
2476 };
2477 #endif /* CONFIG_USER_NS */
2478
2479 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2480                                 struct pid *pid, struct task_struct *task)
2481 {
2482         int err = lock_trace(task);
2483         if (!err) {
2484                 seq_printf(m, "%08x\n", task->personality);
2485                 unlock_trace(task);
2486         }
2487         return err;
2488 }
2489
2490 /*
2491  * Thread groups
2492  */
2493 static const struct file_operations proc_task_operations;
2494 static const struct inode_operations proc_task_inode_operations;
2495
2496 static const struct pid_entry tgid_base_stuff[] = {
2497         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2498         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2499 #ifdef CONFIG_CHECKPOINT_RESTORE
2500         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2501 #endif
2502         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2503         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2504 #ifdef CONFIG_NET
2505         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2506 #endif
2507         REG("environ",    S_IRUSR, proc_environ_operations),
2508         INF("auxv",       S_IRUSR, proc_pid_auxv),
2509         ONE("status",     S_IRUGO, proc_pid_status),
2510         ONE("personality", S_IRUGO, proc_pid_personality),
2511         INF("limits",     S_IRUGO, proc_pid_limits),
2512 #ifdef CONFIG_SCHED_DEBUG
2513         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2514 #endif
2515 #ifdef CONFIG_SCHED_AUTOGROUP
2516         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2517 #endif
2518         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2519 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2520         INF("syscall",    S_IRUGO, proc_pid_syscall),
2521 #endif
2522         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2523         ONE("stat",       S_IRUGO, proc_tgid_stat),
2524         ONE("statm",      S_IRUGO, proc_pid_statm),
2525         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2526 #ifdef CONFIG_NUMA
2527         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2528 #endif
2529         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2530         LNK("cwd",        proc_cwd_link),
2531         LNK("root",       proc_root_link),
2532         LNK("exe",        proc_exe_link),
2533         REG("mounts",     S_IRUGO, proc_mounts_operations),
2534         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2535         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2536 #ifdef CONFIG_PROC_PAGE_MONITOR
2537         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2538         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2539         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2540 #endif
2541 #ifdef CONFIG_SECURITY
2542         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2543 #endif
2544 #ifdef CONFIG_KALLSYMS
2545         INF("wchan",      S_IRUGO, proc_pid_wchan),
2546 #endif
2547 #ifdef CONFIG_STACKTRACE
2548         ONE("stack",      S_IRUGO, proc_pid_stack),
2549 #endif
2550 #ifdef CONFIG_SCHEDSTATS
2551         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2552 #endif
2553 #ifdef CONFIG_LATENCYTOP
2554         REG("latency",  S_IRUGO, proc_lstats_operations),
2555 #endif
2556 #ifdef CONFIG_PROC_PID_CPUSET
2557         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2558 #endif
2559 #ifdef CONFIG_CGROUPS
2560         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2561 #endif
2562         INF("oom_score",  S_IRUGO, proc_oom_score),
2563         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2564         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2565 #ifdef CONFIG_AUDITSYSCALL
2566         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2567         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2568 #endif
2569 #ifdef CONFIG_FAULT_INJECTION
2570         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2571 #endif
2572 #ifdef CONFIG_ELF_CORE
2573         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2574 #endif
2575 #ifdef CONFIG_TASK_IO_ACCOUNTING
2576         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2577 #endif
2578 #ifdef CONFIG_HARDWALL
2579         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2580 #endif
2581 #ifdef CONFIG_USER_NS
2582         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2583         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2584         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2585 #endif
2586 };
2587
2588 static int proc_tgid_base_readdir(struct file * filp,
2589                              void * dirent, filldir_t filldir)
2590 {
2591         return proc_pident_readdir(filp,dirent,filldir,
2592                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2593 }
2594
2595 static const struct file_operations proc_tgid_base_operations = {
2596         .read           = generic_read_dir,
2597         .readdir        = proc_tgid_base_readdir,
2598         .llseek         = default_llseek,
2599 };
2600
2601 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2602 {
2603         return proc_pident_lookup(dir, dentry,
2604                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2605 }
2606
2607 static const struct inode_operations proc_tgid_base_inode_operations = {
2608         .lookup         = proc_tgid_base_lookup,
2609         .getattr        = pid_getattr,
2610         .setattr        = proc_setattr,
2611         .permission     = proc_pid_permission,
2612 };
2613
2614 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2615 {
2616         struct dentry *dentry, *leader, *dir;
2617         char buf[PROC_NUMBUF];
2618         struct qstr name;
2619
2620         name.name = buf;
2621         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2622         /* no ->d_hash() rejects on procfs */
2623         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2624         if (dentry) {
2625                 shrink_dcache_parent(dentry);
2626                 d_drop(dentry);
2627                 dput(dentry);
2628         }
2629
2630         name.name = buf;
2631         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2632         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2633         if (!leader)
2634                 goto out;
2635
2636         name.name = "task";
2637         name.len = strlen(name.name);
2638         dir = d_hash_and_lookup(leader, &name);
2639         if (!dir)
2640                 goto out_put_leader;
2641
2642         name.name = buf;
2643         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2644         dentry = d_hash_and_lookup(dir, &name);
2645         if (dentry) {
2646                 shrink_dcache_parent(dentry);
2647                 d_drop(dentry);
2648                 dput(dentry);
2649         }
2650
2651         dput(dir);
2652 out_put_leader:
2653         dput(leader);
2654 out:
2655         return;
2656 }
2657
2658 /**
2659  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2660  * @task: task that should be flushed.
2661  *
2662  * When flushing dentries from proc, one needs to flush them from global
2663  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2664  * in. This call is supposed to do all of this job.
2665  *
2666  * Looks in the dcache for
2667  * /proc/@pid
2668  * /proc/@tgid/task/@pid
2669  * if either directory is present flushes it and all of it'ts children
2670  * from the dcache.
2671  *
2672  * It is safe and reasonable to cache /proc entries for a task until
2673  * that task exits.  After that they just clog up the dcache with
2674  * useless entries, possibly causing useful dcache entries to be
2675  * flushed instead.  This routine is proved to flush those useless
2676  * dcache entries at process exit time.
2677  *
2678  * NOTE: This routine is just an optimization so it does not guarantee
2679  *       that no dcache entries will exist at process exit time it
2680  *       just makes it very unlikely that any will persist.
2681  */
2682
2683 void proc_flush_task(struct task_struct *task)
2684 {
2685         int i;
2686         struct pid *pid, *tgid;
2687         struct upid *upid;
2688
2689         pid = task_pid(task);
2690         tgid = task_tgid(task);
2691
2692         for (i = 0; i <= pid->level; i++) {
2693                 upid = &pid->numbers[i];
2694                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2695                                         tgid->numbers[i].nr);
2696         }
2697 }
2698
2699 static struct dentry *proc_pid_instantiate(struct inode *dir,
2700                                            struct dentry * dentry,
2701                                            struct task_struct *task, const void *ptr)
2702 {
2703         struct dentry *error = ERR_PTR(-ENOENT);
2704         struct inode *inode;
2705
2706         inode = proc_pid_make_inode(dir->i_sb, task);
2707         if (!inode)
2708                 goto out;
2709
2710         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2711         inode->i_op = &proc_tgid_base_inode_operations;
2712         inode->i_fop = &proc_tgid_base_operations;
2713         inode->i_flags|=S_IMMUTABLE;
2714
2715         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2716                                                   ARRAY_SIZE(tgid_base_stuff)));
2717
2718         d_set_d_op(dentry, &pid_dentry_operations);
2719
2720         d_add(dentry, inode);
2721         /* Close the race of the process dying before we return the dentry */
2722         if (pid_revalidate(dentry, 0))
2723                 error = NULL;
2724 out:
2725         return error;
2726 }
2727
2728 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2729 {
2730         struct dentry *result = NULL;
2731         struct task_struct *task;
2732         unsigned tgid;
2733         struct pid_namespace *ns;
2734
2735         tgid = name_to_int(dentry);
2736         if (tgid == ~0U)
2737                 goto out;
2738
2739         ns = dentry->d_sb->s_fs_info;
2740         rcu_read_lock();
2741         task = find_task_by_pid_ns(tgid, ns);
2742         if (task)
2743                 get_task_struct(task);
2744         rcu_read_unlock();
2745         if (!task)
2746                 goto out;
2747
2748         result = proc_pid_instantiate(dir, dentry, task, NULL);
2749         put_task_struct(task);
2750 out:
2751         return result;
2752 }
2753
2754 /*
2755  * Find the first task with tgid >= tgid
2756  *
2757  */
2758 struct tgid_iter {
2759         unsigned int tgid;
2760         struct task_struct *task;
2761 };
2762 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2763 {
2764         struct pid *pid;
2765
2766         if (iter.task)
2767                 put_task_struct(iter.task);
2768         rcu_read_lock();
2769 retry:
2770         iter.task = NULL;
2771         pid = find_ge_pid(iter.tgid, ns);
2772         if (pid) {
2773                 iter.tgid = pid_nr_ns(pid, ns);
2774                 iter.task = pid_task(pid, PIDTYPE_PID);
2775                 /* What we to know is if the pid we have find is the
2776                  * pid of a thread_group_leader.  Testing for task
2777                  * being a thread_group_leader is the obvious thing
2778                  * todo but there is a window when it fails, due to
2779                  * the pid transfer logic in de_thread.
2780                  *
2781                  * So we perform the straight forward test of seeing
2782                  * if the pid we have found is the pid of a thread
2783                  * group leader, and don't worry if the task we have
2784                  * found doesn't happen to be a thread group leader.
2785                  * As we don't care in the case of readdir.
2786                  */
2787                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2788                         iter.tgid += 1;
2789                         goto retry;
2790                 }
2791                 get_task_struct(iter.task);
2792         }
2793         rcu_read_unlock();
2794         return iter;
2795 }
2796
2797 #define TGID_OFFSET (FIRST_PROCESS_ENTRY)
2798
2799 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2800         struct tgid_iter iter)
2801 {
2802         char name[PROC_NUMBUF];
2803         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2804         return proc_fill_cache(filp, dirent, filldir, name, len,
2805                                 proc_pid_instantiate, iter.task, NULL);
2806 }
2807
2808 static int fake_filldir(void *buf, const char *name, int namelen,
2809                         loff_t offset, u64 ino, unsigned d_type)
2810 {
2811         return 0;
2812 }
2813
2814 /* for the /proc/ directory itself, after non-process stuff has been done */
2815 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2816 {
2817         struct tgid_iter iter;
2818         struct pid_namespace *ns;
2819         filldir_t __filldir;
2820
2821         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2822                 goto out;
2823
2824         ns = filp->f_dentry->d_sb->s_fs_info;
2825         iter.task = NULL;
2826         iter.tgid = filp->f_pos - TGID_OFFSET;
2827         for (iter = next_tgid(ns, iter);
2828              iter.task;
2829              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2830                 if (has_pid_permissions(ns, iter.task, 2))
2831                         __filldir = filldir;
2832                 else
2833                         __filldir = fake_filldir;
2834
2835                 filp->f_pos = iter.tgid + TGID_OFFSET;
2836                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
2837                         put_task_struct(iter.task);
2838                         goto out;
2839                 }
2840         }
2841         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2842 out:
2843         return 0;
2844 }
2845
2846 /*
2847  * Tasks
2848  */
2849 static const struct pid_entry tid_base_stuff[] = {
2850         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2851         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2852         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2853         REG("environ",   S_IRUSR, proc_environ_operations),
2854         INF("auxv",      S_IRUSR, proc_pid_auxv),
2855         ONE("status",    S_IRUGO, proc_pid_status),
2856         ONE("personality", S_IRUGO, proc_pid_personality),
2857         INF("limits",    S_IRUGO, proc_pid_limits),
2858 #ifdef CONFIG_SCHED_DEBUG
2859         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2860 #endif
2861         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2862 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2863         INF("syscall",   S_IRUGO, proc_pid_syscall),
2864 #endif
2865         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2866         ONE("stat",      S_IRUGO, proc_tid_stat),
2867         ONE("statm",     S_IRUGO, proc_pid_statm),
2868         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2869 #ifdef CONFIG_CHECKPOINT_RESTORE
2870         REG("children",  S_IRUGO, proc_tid_children_operations),
2871 #endif
2872 #ifdef CONFIG_NUMA
2873         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2874 #endif
2875         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2876         LNK("cwd",       proc_cwd_link),
2877         LNK("root",      proc_root_link),
2878         LNK("exe",       proc_exe_link),
2879         REG("mounts",    S_IRUGO, proc_mounts_operations),
2880         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2881 #ifdef CONFIG_PROC_PAGE_MONITOR
2882         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2883         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2884         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2885 #endif
2886 #ifdef CONFIG_SECURITY
2887         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2888 #endif
2889 #ifdef CONFIG_KALLSYMS
2890         INF("wchan",     S_IRUGO, proc_pid_wchan),
2891 #endif
2892 #ifdef CONFIG_STACKTRACE
2893         ONE("stack",      S_IRUGO, proc_pid_stack),
2894 #endif
2895 #ifdef CONFIG_SCHEDSTATS
2896         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2897 #endif
2898 #ifdef CONFIG_LATENCYTOP
2899         REG("latency",  S_IRUGO, proc_lstats_operations),
2900 #endif
2901 #ifdef CONFIG_PROC_PID_CPUSET
2902         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2903 #endif
2904 #ifdef CONFIG_CGROUPS
2905         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2906 #endif
2907         INF("oom_score", S_IRUGO, proc_oom_score),
2908         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2909         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2910 #ifdef CONFIG_AUDITSYSCALL
2911         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2912         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2913 #endif
2914 #ifdef CONFIG_FAULT_INJECTION
2915         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2916 #endif
2917 #ifdef CONFIG_TASK_IO_ACCOUNTING
2918         INF("io",       S_IRUSR, proc_tid_io_accounting),
2919 #endif
2920 #ifdef CONFIG_HARDWALL
2921         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2922 #endif
2923 #ifdef CONFIG_USER_NS
2924         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2925         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2926         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2927 #endif
2928 };
2929
2930 static int proc_tid_base_readdir(struct file * filp,
2931                              void * dirent, filldir_t filldir)
2932 {
2933         return proc_pident_readdir(filp,dirent,filldir,
2934                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2935 }
2936
2937 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2938 {
2939         return proc_pident_lookup(dir, dentry,
2940                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2941 }
2942
2943 static const struct file_operations proc_tid_base_operations = {
2944         .read           = generic_read_dir,
2945         .readdir        = proc_tid_base_readdir,
2946         .llseek         = default_llseek,
2947 };
2948
2949 static const struct inode_operations proc_tid_base_inode_operations = {
2950         .lookup         = proc_tid_base_lookup,
2951         .getattr        = pid_getattr,
2952         .setattr        = proc_setattr,
2953 };
2954
2955 static struct dentry *proc_task_instantiate(struct inode *dir,
2956         struct dentry *dentry, struct task_struct *task, const void *ptr)
2957 {
2958         struct dentry *error = ERR_PTR(-ENOENT);
2959         struct inode *inode;
2960         inode = proc_pid_make_inode(dir->i_sb, task);
2961
2962         if (!inode)
2963                 goto out;
2964         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2965         inode->i_op = &proc_tid_base_inode_operations;
2966         inode->i_fop = &proc_tid_base_operations;
2967         inode->i_flags|=S_IMMUTABLE;
2968
2969         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
2970                                                   ARRAY_SIZE(tid_base_stuff)));
2971
2972         d_set_d_op(dentry, &pid_dentry_operations);
2973
2974         d_add(dentry, inode);
2975         /* Close the race of the process dying before we return the dentry */
2976         if (pid_revalidate(dentry, 0))
2977                 error = NULL;
2978 out:
2979         return error;
2980 }
2981
2982 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2983 {
2984         struct dentry *result = ERR_PTR(-ENOENT);
2985         struct task_struct *task;
2986         struct task_struct *leader = get_proc_task(dir);
2987         unsigned tid;
2988         struct pid_namespace *ns;
2989
2990         if (!leader)
2991                 goto out_no_task;
2992
2993         tid = name_to_int(dentry);
2994         if (tid == ~0U)
2995                 goto out;
2996
2997         ns = dentry->d_sb->s_fs_info;
2998         rcu_read_lock();
2999         task = find_task_by_pid_ns(tid, ns);
3000         if (task)
3001                 get_task_struct(task);
3002         rcu_read_unlock();
3003         if (!task)
3004                 goto out;
3005         if (!same_thread_group(leader, task))
3006                 goto out_drop_task;
3007
3008         result = proc_task_instantiate(dir, dentry, task, NULL);
3009 out_drop_task:
3010         put_task_struct(task);
3011 out:
3012         put_task_struct(leader);
3013 out_no_task:
3014         return result;
3015 }
3016
3017 /*
3018  * Find the first tid of a thread group to return to user space.
3019  *
3020  * Usually this is just the thread group leader, but if the users
3021  * buffer was too small or there was a seek into the middle of the
3022  * directory we have more work todo.
3023  *
3024  * In the case of a short read we start with find_task_by_pid.
3025  *
3026  * In the case of a seek we start with the leader and walk nr
3027  * threads past it.
3028  */
3029 static struct task_struct *first_tid(struct task_struct *leader,
3030                 int tid, int nr, struct pid_namespace *ns)
3031 {
3032         struct task_struct *pos;
3033
3034         rcu_read_lock();
3035         /* Attempt to start with the pid of a thread */
3036         if (tid && (nr > 0)) {
3037                 pos = find_task_by_pid_ns(tid, ns);
3038                 if (pos && (pos->group_leader == leader))
3039                         goto found;
3040         }
3041
3042         /* If nr exceeds the number of threads there is nothing todo */
3043         pos = NULL;
3044         if (nr && nr >= get_nr_threads(leader))
3045                 goto out;
3046
3047         /* If we haven't found our starting place yet start
3048          * with the leader and walk nr threads forward.
3049          */
3050         for (pos = leader; nr > 0; --nr) {
3051                 pos = next_thread(pos);
3052                 if (pos == leader) {
3053                         pos = NULL;
3054                         goto out;
3055                 }
3056         }
3057 found:
3058         get_task_struct(pos);
3059 out:
3060         rcu_read_unlock();
3061         return pos;
3062 }
3063
3064 /*
3065  * Find the next thread in the thread list.
3066  * Return NULL if there is an error or no next thread.
3067  *
3068  * The reference to the input task_struct is released.
3069  */
3070 static struct task_struct *next_tid(struct task_struct *start)
3071 {
3072         struct task_struct *pos = NULL;
3073         rcu_read_lock();
3074         if (pid_alive(start)) {
3075                 pos = next_thread(start);
3076                 if (thread_group_leader(pos))
3077                         pos = NULL;
3078                 else
3079                         get_task_struct(pos);
3080         }
3081         rcu_read_unlock();
3082         put_task_struct(start);
3083         return pos;
3084 }
3085
3086 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3087         struct task_struct *task, int tid)
3088 {
3089         char name[PROC_NUMBUF];
3090         int len = snprintf(name, sizeof(name), "%d", tid);
3091         return proc_fill_cache(filp, dirent, filldir, name, len,
3092                                 proc_task_instantiate, task, NULL);
3093 }
3094
3095 /* for the /proc/TGID/task/ directories */
3096 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3097 {
3098         struct dentry *dentry = filp->f_path.dentry;
3099         struct inode *inode = dentry->d_inode;
3100         struct task_struct *leader = NULL;
3101         struct task_struct *task;
3102         int retval = -ENOENT;
3103         ino_t ino;
3104         int tid;
3105         struct pid_namespace *ns;
3106
3107         task = get_proc_task(inode);
3108         if (!task)
3109                 goto out_no_task;
3110         rcu_read_lock();
3111         if (pid_alive(task)) {
3112                 leader = task->group_leader;
3113                 get_task_struct(leader);
3114         }
3115         rcu_read_unlock();
3116         put_task_struct(task);
3117         if (!leader)
3118                 goto out_no_task;
3119         retval = 0;
3120
3121         switch ((unsigned long)filp->f_pos) {
3122         case 0:
3123                 ino = inode->i_ino;
3124                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3125                         goto out;
3126                 filp->f_pos++;
3127                 /* fall through */
3128         case 1:
3129                 ino = parent_ino(dentry);
3130                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3131                         goto out;
3132                 filp->f_pos++;
3133                 /* fall through */
3134         }
3135
3136         /* f_version caches the tgid value that the last readdir call couldn't
3137          * return. lseek aka telldir automagically resets f_version to 0.
3138          */
3139         ns = filp->f_dentry->d_sb->s_fs_info;
3140         tid = (int)filp->f_version;
3141         filp->f_version = 0;
3142         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3143              task;
3144              task = next_tid(task), filp->f_pos++) {
3145                 tid = task_pid_nr_ns(task, ns);
3146                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3147                         /* returning this tgid failed, save it as the first
3148                          * pid for the next readir call */
3149                         filp->f_version = (u64)tid;
3150                         put_task_struct(task);
3151                         break;
3152                 }
3153         }
3154 out:
3155         put_task_struct(leader);
3156 out_no_task:
3157         return retval;
3158 }
3159
3160 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3161 {
3162         struct inode *inode = dentry->d_inode;
3163         struct task_struct *p = get_proc_task(inode);
3164         generic_fillattr(inode, stat);
3165
3166         if (p) {
3167                 stat->nlink += get_nr_threads(p);
3168                 put_task_struct(p);
3169         }
3170
3171         return 0;
3172 }
3173
3174 static const struct inode_operations proc_task_inode_operations = {
3175         .lookup         = proc_task_lookup,
3176         .getattr        = proc_task_getattr,
3177         .setattr        = proc_setattr,
3178         .permission     = proc_pid_permission,
3179 };
3180
3181 static const struct file_operations proc_task_operations = {
3182         .read           = generic_read_dir,
3183         .readdir        = proc_task_readdir,
3184         .llseek         = default_llseek,
3185 };