stop passing nameidata * to ->d_revalidate()
[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/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/user_namespace.h>
85 #include <linux/fs_struct.h>
86 #include <linux/slab.h>
87 #include <linux/flex_array.h>
88 #ifdef CONFIG_HARDWALL
89 #include <asm/hardwall.h>
90 #endif
91 #include <trace/events/oom.h>
92 #include "internal.h"
93
94 /* NOTE:
95  *      Implementing inode permission operations in /proc is almost
96  *      certainly an error.  Permission checks need to happen during
97  *      each system call not at open time.  The reason is that most of
98  *      what we wish to check for permissions in /proc varies at runtime.
99  *
100  *      The classic example of a problem is opening file descriptors
101  *      in /proc for a task before it execs a suid executable.
102  */
103
104 struct pid_entry {
105         char *name;
106         int len;
107         umode_t mode;
108         const struct inode_operations *iop;
109         const struct file_operations *fop;
110         union proc_op op;
111 };
112
113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
114         .name = (NAME),                                 \
115         .len  = sizeof(NAME) - 1,                       \
116         .mode = MODE,                                   \
117         .iop  = IOP,                                    \
118         .fop  = FOP,                                    \
119         .op   = OP,                                     \
120 }
121
122 #define DIR(NAME, MODE, iops, fops)     \
123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
124 #define LNK(NAME, get_link)                                     \
125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
126                 &proc_pid_link_inode_operations, NULL,          \
127                 { .proc_get_link = get_link } )
128 #define REG(NAME, MODE, fops)                           \
129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
130 #define INF(NAME, MODE, read)                           \
131         NOD(NAME, (S_IFREG|(MODE)),                     \
132                 NULL, &proc_info_file_operations,       \
133                 { .proc_read = read } )
134 #define ONE(NAME, MODE, show)                           \
135         NOD(NAME, (S_IFREG|(MODE)),                     \
136                 NULL, &proc_single_file_operations,     \
137                 { .proc_show = show } )
138
139 static int proc_fd_permission(struct inode *inode, int mask);
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->f_dentry->d_inode);
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         if ((attr->ia_valid & ATTR_SIZE) &&
547             attr->ia_size != i_size_read(inode)) {
548                 error = vmtruncate(inode, attr->ia_size);
549                 if (error)
550                         return error;
551         }
552
553         setattr_copy(inode, attr);
554         mark_inode_dirty(inode);
555         return 0;
556 }
557
558 /*
559  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
560  * or euid/egid (for hide_pid_min=2)?
561  */
562 static bool has_pid_permissions(struct pid_namespace *pid,
563                                  struct task_struct *task,
564                                  int hide_pid_min)
565 {
566         if (pid->hide_pid < hide_pid_min)
567                 return true;
568         if (in_group_p(pid->pid_gid))
569                 return true;
570         return ptrace_may_access(task, PTRACE_MODE_READ);
571 }
572
573
574 static int proc_pid_permission(struct inode *inode, int mask)
575 {
576         struct pid_namespace *pid = inode->i_sb->s_fs_info;
577         struct task_struct *task;
578         bool has_perms;
579
580         task = get_proc_task(inode);
581         if (!task)
582                 return -ESRCH;
583         has_perms = has_pid_permissions(pid, task, 1);
584         put_task_struct(task);
585
586         if (!has_perms) {
587                 if (pid->hide_pid == 2) {
588                         /*
589                          * Let's make getdents(), stat(), and open()
590                          * consistent with each other.  If a process
591                          * may not stat() a file, it shouldn't be seen
592                          * in procfs at all.
593                          */
594                         return -ENOENT;
595                 }
596
597                 return -EPERM;
598         }
599         return generic_permission(inode, mask);
600 }
601
602
603
604 static const struct inode_operations proc_def_inode_operations = {
605         .setattr        = proc_setattr,
606 };
607
608 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
609
610 static ssize_t proc_info_read(struct file * file, char __user * buf,
611                           size_t count, loff_t *ppos)
612 {
613         struct inode * inode = file->f_path.dentry->d_inode;
614         unsigned long page;
615         ssize_t length;
616         struct task_struct *task = get_proc_task(inode);
617
618         length = -ESRCH;
619         if (!task)
620                 goto out_no_task;
621
622         if (count > PROC_BLOCK_SIZE)
623                 count = PROC_BLOCK_SIZE;
624
625         length = -ENOMEM;
626         if (!(page = __get_free_page(GFP_TEMPORARY)))
627                 goto out;
628
629         length = PROC_I(inode)->op.proc_read(task, (char*)page);
630
631         if (length >= 0)
632                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
633         free_page(page);
634 out:
635         put_task_struct(task);
636 out_no_task:
637         return length;
638 }
639
640 static const struct file_operations proc_info_file_operations = {
641         .read           = proc_info_read,
642         .llseek         = generic_file_llseek,
643 };
644
645 static int proc_single_show(struct seq_file *m, void *v)
646 {
647         struct inode *inode = m->private;
648         struct pid_namespace *ns;
649         struct pid *pid;
650         struct task_struct *task;
651         int ret;
652
653         ns = inode->i_sb->s_fs_info;
654         pid = proc_pid(inode);
655         task = get_pid_task(pid, PIDTYPE_PID);
656         if (!task)
657                 return -ESRCH;
658
659         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
660
661         put_task_struct(task);
662         return ret;
663 }
664
665 static int proc_single_open(struct inode *inode, struct file *filp)
666 {
667         return single_open(filp, proc_single_show, inode);
668 }
669
670 static const struct file_operations proc_single_file_operations = {
671         .open           = proc_single_open,
672         .read           = seq_read,
673         .llseek         = seq_lseek,
674         .release        = single_release,
675 };
676
677 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
678 {
679         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
680         struct mm_struct *mm;
681
682         if (!task)
683                 return -ESRCH;
684
685         mm = mm_access(task, mode);
686         put_task_struct(task);
687
688         if (IS_ERR(mm))
689                 return PTR_ERR(mm);
690
691         if (mm) {
692                 /* ensure this mm_struct can't be freed */
693                 atomic_inc(&mm->mm_count);
694                 /* but do not pin its memory */
695                 mmput(mm);
696         }
697
698         /* OK to pass negative loff_t, we can catch out-of-range */
699         file->f_mode |= FMODE_UNSIGNED_OFFSET;
700         file->private_data = mm;
701
702         return 0;
703 }
704
705 static int mem_open(struct inode *inode, struct file *file)
706 {
707         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
708 }
709
710 static ssize_t mem_rw(struct file *file, char __user *buf,
711                         size_t count, loff_t *ppos, int write)
712 {
713         struct mm_struct *mm = file->private_data;
714         unsigned long addr = *ppos;
715         ssize_t copied;
716         char *page;
717
718         if (!mm)
719                 return 0;
720
721         page = (char *)__get_free_page(GFP_TEMPORARY);
722         if (!page)
723                 return -ENOMEM;
724
725         copied = 0;
726         if (!atomic_inc_not_zero(&mm->mm_users))
727                 goto free;
728
729         while (count > 0) {
730                 int this_len = min_t(int, count, PAGE_SIZE);
731
732                 if (write && copy_from_user(page, buf, this_len)) {
733                         copied = -EFAULT;
734                         break;
735                 }
736
737                 this_len = access_remote_vm(mm, addr, page, this_len, write);
738                 if (!this_len) {
739                         if (!copied)
740                                 copied = -EIO;
741                         break;
742                 }
743
744                 if (!write && copy_to_user(buf, page, this_len)) {
745                         copied = -EFAULT;
746                         break;
747                 }
748
749                 buf += this_len;
750                 addr += this_len;
751                 copied += this_len;
752                 count -= this_len;
753         }
754         *ppos = addr;
755
756         mmput(mm);
757 free:
758         free_page((unsigned long) page);
759         return copied;
760 }
761
762 static ssize_t mem_read(struct file *file, char __user *buf,
763                         size_t count, loff_t *ppos)
764 {
765         return mem_rw(file, buf, count, ppos, 0);
766 }
767
768 static ssize_t mem_write(struct file *file, const char __user *buf,
769                          size_t count, loff_t *ppos)
770 {
771         return mem_rw(file, (char __user*)buf, count, ppos, 1);
772 }
773
774 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
775 {
776         switch (orig) {
777         case 0:
778                 file->f_pos = offset;
779                 break;
780         case 1:
781                 file->f_pos += offset;
782                 break;
783         default:
784                 return -EINVAL;
785         }
786         force_successful_syscall_return();
787         return file->f_pos;
788 }
789
790 static int mem_release(struct inode *inode, struct file *file)
791 {
792         struct mm_struct *mm = file->private_data;
793         if (mm)
794                 mmdrop(mm);
795         return 0;
796 }
797
798 static const struct file_operations proc_mem_operations = {
799         .llseek         = mem_lseek,
800         .read           = mem_read,
801         .write          = mem_write,
802         .open           = mem_open,
803         .release        = mem_release,
804 };
805
806 static int environ_open(struct inode *inode, struct file *file)
807 {
808         return __mem_open(inode, file, PTRACE_MODE_READ);
809 }
810
811 static ssize_t environ_read(struct file *file, char __user *buf,
812                         size_t count, loff_t *ppos)
813 {
814         char *page;
815         unsigned long src = *ppos;
816         int ret = 0;
817         struct mm_struct *mm = file->private_data;
818
819         if (!mm)
820                 return 0;
821
822         page = (char *)__get_free_page(GFP_TEMPORARY);
823         if (!page)
824                 return -ENOMEM;
825
826         ret = 0;
827         if (!atomic_inc_not_zero(&mm->mm_users))
828                 goto free;
829         while (count > 0) {
830                 int this_len, retval, max_len;
831
832                 this_len = mm->env_end - (mm->env_start + src);
833
834                 if (this_len <= 0)
835                         break;
836
837                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
838                 this_len = (this_len > max_len) ? max_len : this_len;
839
840                 retval = access_remote_vm(mm, (mm->env_start + src),
841                         page, this_len, 0);
842
843                 if (retval <= 0) {
844                         ret = retval;
845                         break;
846                 }
847
848                 if (copy_to_user(buf, page, retval)) {
849                         ret = -EFAULT;
850                         break;
851                 }
852
853                 ret += retval;
854                 src += retval;
855                 buf += retval;
856                 count -= retval;
857         }
858         *ppos = src;
859         mmput(mm);
860
861 free:
862         free_page((unsigned long) page);
863         return ret;
864 }
865
866 static const struct file_operations proc_environ_operations = {
867         .open           = environ_open,
868         .read           = environ_read,
869         .llseek         = generic_file_llseek,
870         .release        = mem_release,
871 };
872
873 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
874                                 size_t count, loff_t *ppos)
875 {
876         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877         char buffer[PROC_NUMBUF];
878         size_t len;
879         int oom_adjust = OOM_DISABLE;
880         unsigned long flags;
881
882         if (!task)
883                 return -ESRCH;
884
885         if (lock_task_sighand(task, &flags)) {
886                 oom_adjust = task->signal->oom_adj;
887                 unlock_task_sighand(task, &flags);
888         }
889
890         put_task_struct(task);
891
892         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
893
894         return simple_read_from_buffer(buf, count, ppos, buffer, len);
895 }
896
897 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
898                                 size_t count, loff_t *ppos)
899 {
900         struct task_struct *task;
901         char buffer[PROC_NUMBUF];
902         int oom_adjust;
903         unsigned long flags;
904         int err;
905
906         memset(buffer, 0, sizeof(buffer));
907         if (count > sizeof(buffer) - 1)
908                 count = sizeof(buffer) - 1;
909         if (copy_from_user(buffer, buf, count)) {
910                 err = -EFAULT;
911                 goto out;
912         }
913
914         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
915         if (err)
916                 goto out;
917         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
918              oom_adjust != OOM_DISABLE) {
919                 err = -EINVAL;
920                 goto out;
921         }
922
923         task = get_proc_task(file->f_path.dentry->d_inode);
924         if (!task) {
925                 err = -ESRCH;
926                 goto out;
927         }
928
929         task_lock(task);
930         if (!task->mm) {
931                 err = -EINVAL;
932                 goto err_task_lock;
933         }
934
935         if (!lock_task_sighand(task, &flags)) {
936                 err = -ESRCH;
937                 goto err_task_lock;
938         }
939
940         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
941                 err = -EACCES;
942                 goto err_sighand;
943         }
944
945         /*
946          * Warn that /proc/pid/oom_adj is deprecated, see
947          * Documentation/feature-removal-schedule.txt.
948          */
949         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
950                   current->comm, task_pid_nr(current), task_pid_nr(task),
951                   task_pid_nr(task));
952         task->signal->oom_adj = oom_adjust;
953         /*
954          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
955          * value is always attainable.
956          */
957         if (task->signal->oom_adj == OOM_ADJUST_MAX)
958                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
959         else
960                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
961                                                                 -OOM_DISABLE;
962         trace_oom_score_adj_update(task);
963 err_sighand:
964         unlock_task_sighand(task, &flags);
965 err_task_lock:
966         task_unlock(task);
967         put_task_struct(task);
968 out:
969         return err < 0 ? err : count;
970 }
971
972 static const struct file_operations proc_oom_adjust_operations = {
973         .read           = oom_adjust_read,
974         .write          = oom_adjust_write,
975         .llseek         = generic_file_llseek,
976 };
977
978 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
979                                         size_t count, loff_t *ppos)
980 {
981         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
982         char buffer[PROC_NUMBUF];
983         int oom_score_adj = OOM_SCORE_ADJ_MIN;
984         unsigned long flags;
985         size_t len;
986
987         if (!task)
988                 return -ESRCH;
989         if (lock_task_sighand(task, &flags)) {
990                 oom_score_adj = task->signal->oom_score_adj;
991                 unlock_task_sighand(task, &flags);
992         }
993         put_task_struct(task);
994         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
995         return simple_read_from_buffer(buf, count, ppos, buffer, len);
996 }
997
998 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
999                                         size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task;
1002         char buffer[PROC_NUMBUF];
1003         unsigned long flags;
1004         int oom_score_adj;
1005         int err;
1006
1007         memset(buffer, 0, sizeof(buffer));
1008         if (count > sizeof(buffer) - 1)
1009                 count = sizeof(buffer) - 1;
1010         if (copy_from_user(buffer, buf, count)) {
1011                 err = -EFAULT;
1012                 goto out;
1013         }
1014
1015         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016         if (err)
1017                 goto out;
1018         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         task = get_proc_task(file->f_path.dentry->d_inode);
1025         if (!task) {
1026                 err = -ESRCH;
1027                 goto out;
1028         }
1029
1030         task_lock(task);
1031         if (!task->mm) {
1032                 err = -EINVAL;
1033                 goto err_task_lock;
1034         }
1035
1036         if (!lock_task_sighand(task, &flags)) {
1037                 err = -ESRCH;
1038                 goto err_task_lock;
1039         }
1040
1041         if (oom_score_adj < task->signal->oom_score_adj_min &&
1042                         !capable(CAP_SYS_RESOURCE)) {
1043                 err = -EACCES;
1044                 goto err_sighand;
1045         }
1046
1047         task->signal->oom_score_adj = oom_score_adj;
1048         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049                 task->signal->oom_score_adj_min = oom_score_adj;
1050         trace_oom_score_adj_update(task);
1051         /*
1052          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053          * always attainable.
1054          */
1055         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056                 task->signal->oom_adj = OOM_DISABLE;
1057         else
1058                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059                                                         OOM_SCORE_ADJ_MAX;
1060 err_sighand:
1061         unlock_task_sighand(task, &flags);
1062 err_task_lock:
1063         task_unlock(task);
1064         put_task_struct(task);
1065 out:
1066         return err < 0 ? err : count;
1067 }
1068
1069 static const struct file_operations proc_oom_score_adj_operations = {
1070         .read           = oom_score_adj_read,
1071         .write          = oom_score_adj_write,
1072         .llseek         = default_llseek,
1073 };
1074
1075 #ifdef CONFIG_AUDITSYSCALL
1076 #define TMPBUFLEN 21
1077 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078                                   size_t count, loff_t *ppos)
1079 {
1080         struct inode * inode = file->f_path.dentry->d_inode;
1081         struct task_struct *task = get_proc_task(inode);
1082         ssize_t length;
1083         char tmpbuf[TMPBUFLEN];
1084
1085         if (!task)
1086                 return -ESRCH;
1087         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088                                 audit_get_loginuid(task));
1089         put_task_struct(task);
1090         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091 }
1092
1093 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094                                    size_t count, loff_t *ppos)
1095 {
1096         struct inode * inode = file->f_path.dentry->d_inode;
1097         char *page, *tmp;
1098         ssize_t length;
1099         uid_t loginuid;
1100
1101         rcu_read_lock();
1102         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103                 rcu_read_unlock();
1104                 return -EPERM;
1105         }
1106         rcu_read_unlock();
1107
1108         if (count >= PAGE_SIZE)
1109                 count = PAGE_SIZE - 1;
1110
1111         if (*ppos != 0) {
1112                 /* No partial writes. */
1113                 return -EINVAL;
1114         }
1115         page = (char*)__get_free_page(GFP_TEMPORARY);
1116         if (!page)
1117                 return -ENOMEM;
1118         length = -EFAULT;
1119         if (copy_from_user(page, buf, count))
1120                 goto out_free_page;
1121
1122         page[count] = '\0';
1123         loginuid = simple_strtoul(page, &tmp, 10);
1124         if (tmp == page) {
1125                 length = -EINVAL;
1126                 goto out_free_page;
1127
1128         }
1129         length = audit_set_loginuid(loginuid);
1130         if (likely(length == 0))
1131                 length = count;
1132
1133 out_free_page:
1134         free_page((unsigned long) page);
1135         return length;
1136 }
1137
1138 static const struct file_operations proc_loginuid_operations = {
1139         .read           = proc_loginuid_read,
1140         .write          = proc_loginuid_write,
1141         .llseek         = generic_file_llseek,
1142 };
1143
1144 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145                                   size_t count, loff_t *ppos)
1146 {
1147         struct inode * inode = file->f_path.dentry->d_inode;
1148         struct task_struct *task = get_proc_task(inode);
1149         ssize_t length;
1150         char tmpbuf[TMPBUFLEN];
1151
1152         if (!task)
1153                 return -ESRCH;
1154         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155                                 audit_get_sessionid(task));
1156         put_task_struct(task);
1157         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158 }
1159
1160 static const struct file_operations proc_sessionid_operations = {
1161         .read           = proc_sessionid_read,
1162         .llseek         = generic_file_llseek,
1163 };
1164 #endif
1165
1166 #ifdef CONFIG_FAULT_INJECTION
1167 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168                                       size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171         char buffer[PROC_NUMBUF];
1172         size_t len;
1173         int make_it_fail;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         make_it_fail = task->make_it_fail;
1178         put_task_struct(task);
1179
1180         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183 }
1184
1185 static ssize_t proc_fault_inject_write(struct file * file,
1186                         const char __user * buf, size_t count, loff_t *ppos)
1187 {
1188         struct task_struct *task;
1189         char buffer[PROC_NUMBUF], *end;
1190         int make_it_fail;
1191
1192         if (!capable(CAP_SYS_RESOURCE))
1193                 return -EPERM;
1194         memset(buffer, 0, sizeof(buffer));
1195         if (count > sizeof(buffer) - 1)
1196                 count = sizeof(buffer) - 1;
1197         if (copy_from_user(buffer, buf, count))
1198                 return -EFAULT;
1199         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200         if (*end)
1201                 return -EINVAL;
1202         task = get_proc_task(file->f_dentry->d_inode);
1203         if (!task)
1204                 return -ESRCH;
1205         task->make_it_fail = make_it_fail;
1206         put_task_struct(task);
1207
1208         return count;
1209 }
1210
1211 static const struct file_operations proc_fault_inject_operations = {
1212         .read           = proc_fault_inject_read,
1213         .write          = proc_fault_inject_write,
1214         .llseek         = generic_file_llseek,
1215 };
1216 #endif
1217
1218
1219 #ifdef CONFIG_SCHED_DEBUG
1220 /*
1221  * Print out various scheduling related per-task fields:
1222  */
1223 static int sched_show(struct seq_file *m, void *v)
1224 {
1225         struct inode *inode = m->private;
1226         struct task_struct *p;
1227
1228         p = get_proc_task(inode);
1229         if (!p)
1230                 return -ESRCH;
1231         proc_sched_show_task(p, m);
1232
1233         put_task_struct(p);
1234
1235         return 0;
1236 }
1237
1238 static ssize_t
1239 sched_write(struct file *file, const char __user *buf,
1240             size_t count, loff_t *offset)
1241 {
1242         struct inode *inode = file->f_path.dentry->d_inode;
1243         struct task_struct *p;
1244
1245         p = get_proc_task(inode);
1246         if (!p)
1247                 return -ESRCH;
1248         proc_sched_set_task(p);
1249
1250         put_task_struct(p);
1251
1252         return count;
1253 }
1254
1255 static int sched_open(struct inode *inode, struct file *filp)
1256 {
1257         return single_open(filp, sched_show, inode);
1258 }
1259
1260 static const struct file_operations proc_pid_sched_operations = {
1261         .open           = sched_open,
1262         .read           = seq_read,
1263         .write          = sched_write,
1264         .llseek         = seq_lseek,
1265         .release        = single_release,
1266 };
1267
1268 #endif
1269
1270 #ifdef CONFIG_SCHED_AUTOGROUP
1271 /*
1272  * Print out autogroup related information:
1273  */
1274 static int sched_autogroup_show(struct seq_file *m, void *v)
1275 {
1276         struct inode *inode = m->private;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_autogroup_show_task(p, m);
1283
1284         put_task_struct(p);
1285
1286         return 0;
1287 }
1288
1289 static ssize_t
1290 sched_autogroup_write(struct file *file, const char __user *buf,
1291             size_t count, loff_t *offset)
1292 {
1293         struct inode *inode = file->f_path.dentry->d_inode;
1294         struct task_struct *p;
1295         char buffer[PROC_NUMBUF];
1296         int nice;
1297         int err;
1298
1299         memset(buffer, 0, sizeof(buffer));
1300         if (count > sizeof(buffer) - 1)
1301                 count = sizeof(buffer) - 1;
1302         if (copy_from_user(buffer, buf, count))
1303                 return -EFAULT;
1304
1305         err = kstrtoint(strstrip(buffer), 0, &nice);
1306         if (err < 0)
1307                 return err;
1308
1309         p = get_proc_task(inode);
1310         if (!p)
1311                 return -ESRCH;
1312
1313         err = proc_sched_autogroup_set_nice(p, nice);
1314         if (err)
1315                 count = err;
1316
1317         put_task_struct(p);
1318
1319         return count;
1320 }
1321
1322 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323 {
1324         int ret;
1325
1326         ret = single_open(filp, sched_autogroup_show, NULL);
1327         if (!ret) {
1328                 struct seq_file *m = filp->private_data;
1329
1330                 m->private = inode;
1331         }
1332         return ret;
1333 }
1334
1335 static const struct file_operations proc_pid_sched_autogroup_operations = {
1336         .open           = sched_autogroup_open,
1337         .read           = seq_read,
1338         .write          = sched_autogroup_write,
1339         .llseek         = seq_lseek,
1340         .release        = single_release,
1341 };
1342
1343 #endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345 static ssize_t comm_write(struct file *file, const char __user *buf,
1346                                 size_t count, loff_t *offset)
1347 {
1348         struct inode *inode = file->f_path.dentry->d_inode;
1349         struct task_struct *p;
1350         char buffer[TASK_COMM_LEN];
1351
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361
1362         if (same_thread_group(current, p))
1363                 set_task_comm(p, buffer);
1364         else
1365                 count = -EINVAL;
1366
1367         put_task_struct(p);
1368
1369         return count;
1370 }
1371
1372 static int comm_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380
1381         task_lock(p);
1382         seq_printf(m, "%s\n", p->comm);
1383         task_unlock(p);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static int comm_open(struct inode *inode, struct file *filp)
1391 {
1392         return single_open(filp, comm_show, inode);
1393 }
1394
1395 static const struct file_operations proc_pid_set_comm_operations = {
1396         .open           = comm_open,
1397         .read           = seq_read,
1398         .write          = comm_write,
1399         .llseek         = seq_lseek,
1400         .release        = single_release,
1401 };
1402
1403 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404 {
1405         struct task_struct *task;
1406         struct mm_struct *mm;
1407         struct file *exe_file;
1408
1409         task = get_proc_task(dentry->d_inode);
1410         if (!task)
1411                 return -ENOENT;
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414         if (!mm)
1415                 return -ENOENT;
1416         exe_file = get_mm_exe_file(mm);
1417         mmput(mm);
1418         if (exe_file) {
1419                 *exe_path = exe_file->f_path;
1420                 path_get(&exe_file->f_path);
1421                 fput(exe_file);
1422                 return 0;
1423         } else
1424                 return -ENOENT;
1425 }
1426
1427 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428 {
1429         struct inode *inode = dentry->d_inode;
1430         int error = -EACCES;
1431
1432         /* We don't need a base pointer in the /proc filesystem */
1433         path_put(&nd->path);
1434
1435         /* Are we allowed to snoop on the tasks file descriptors? */
1436         if (!proc_fd_access_allowed(inode))
1437                 goto out;
1438
1439         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1440 out:
1441         return ERR_PTR(error);
1442 }
1443
1444 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1445 {
1446         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1447         char *pathname;
1448         int len;
1449
1450         if (!tmp)
1451                 return -ENOMEM;
1452
1453         pathname = d_path(path, tmp, PAGE_SIZE);
1454         len = PTR_ERR(pathname);
1455         if (IS_ERR(pathname))
1456                 goto out;
1457         len = tmp + PAGE_SIZE - 1 - pathname;
1458
1459         if (len > buflen)
1460                 len = buflen;
1461         if (copy_to_user(buffer, pathname, len))
1462                 len = -EFAULT;
1463  out:
1464         free_page((unsigned long)tmp);
1465         return len;
1466 }
1467
1468 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1469 {
1470         int error = -EACCES;
1471         struct inode *inode = dentry->d_inode;
1472         struct path path;
1473
1474         /* Are we allowed to snoop on the tasks file descriptors? */
1475         if (!proc_fd_access_allowed(inode))
1476                 goto out;
1477
1478         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1479         if (error)
1480                 goto out;
1481
1482         error = do_proc_readlink(&path, buffer, buflen);
1483         path_put(&path);
1484 out:
1485         return error;
1486 }
1487
1488 static const struct inode_operations proc_pid_link_inode_operations = {
1489         .readlink       = proc_pid_readlink,
1490         .follow_link    = proc_pid_follow_link,
1491         .setattr        = proc_setattr,
1492 };
1493
1494
1495 /* building an inode */
1496
1497 static int task_dumpable(struct task_struct *task)
1498 {
1499         int dumpable = 0;
1500         struct mm_struct *mm;
1501
1502         task_lock(task);
1503         mm = task->mm;
1504         if (mm)
1505                 dumpable = get_dumpable(mm);
1506         task_unlock(task);
1507         if(dumpable == 1)
1508                 return 1;
1509         return 0;
1510 }
1511
1512 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1513 {
1514         struct inode * inode;
1515         struct proc_inode *ei;
1516         const struct cred *cred;
1517
1518         /* We need a new inode */
1519
1520         inode = new_inode(sb);
1521         if (!inode)
1522                 goto out;
1523
1524         /* Common stuff */
1525         ei = PROC_I(inode);
1526         inode->i_ino = get_next_ino();
1527         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528         inode->i_op = &proc_def_inode_operations;
1529
1530         /*
1531          * grab the reference to task.
1532          */
1533         ei->pid = get_task_pid(task, PIDTYPE_PID);
1534         if (!ei->pid)
1535                 goto out_unlock;
1536
1537         if (task_dumpable(task)) {
1538                 rcu_read_lock();
1539                 cred = __task_cred(task);
1540                 inode->i_uid = cred->euid;
1541                 inode->i_gid = cred->egid;
1542                 rcu_read_unlock();
1543         }
1544         security_task_to_inode(task, inode);
1545
1546 out:
1547         return inode;
1548
1549 out_unlock:
1550         iput(inode);
1551         return NULL;
1552 }
1553
1554 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1555 {
1556         struct inode *inode = dentry->d_inode;
1557         struct task_struct *task;
1558         const struct cred *cred;
1559         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1560
1561         generic_fillattr(inode, stat);
1562
1563         rcu_read_lock();
1564         stat->uid = GLOBAL_ROOT_UID;
1565         stat->gid = GLOBAL_ROOT_GID;
1566         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1567         if (task) {
1568                 if (!has_pid_permissions(pid, task, 2)) {
1569                         rcu_read_unlock();
1570                         /*
1571                          * This doesn't prevent learning whether PID exists,
1572                          * it only makes getattr() consistent with readdir().
1573                          */
1574                         return -ENOENT;
1575                 }
1576                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1577                     task_dumpable(task)) {
1578                         cred = __task_cred(task);
1579                         stat->uid = cred->euid;
1580                         stat->gid = cred->egid;
1581                 }
1582         }
1583         rcu_read_unlock();
1584         return 0;
1585 }
1586
1587 /* dentry stuff */
1588
1589 /*
1590  *      Exceptional case: normally we are not allowed to unhash a busy
1591  * directory. In this case, however, we can do it - no aliasing problems
1592  * due to the way we treat inodes.
1593  *
1594  * Rewrite the inode's ownerships here because the owning task may have
1595  * performed a setuid(), etc.
1596  *
1597  * Before the /proc/pid/status file was created the only way to read
1598  * the effective uid of a /process was to stat /proc/pid.  Reading
1599  * /proc/pid/status is slow enough that procps and other packages
1600  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1601  * made this apply to all per process world readable and executable
1602  * directories.
1603  */
1604 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1605 {
1606         struct inode *inode;
1607         struct task_struct *task;
1608         const struct cred *cred;
1609
1610         if (flags & LOOKUP_RCU)
1611                 return -ECHILD;
1612
1613         inode = dentry->d_inode;
1614         task = get_proc_task(inode);
1615
1616         if (task) {
1617                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1618                     task_dumpable(task)) {
1619                         rcu_read_lock();
1620                         cred = __task_cred(task);
1621                         inode->i_uid = cred->euid;
1622                         inode->i_gid = cred->egid;
1623                         rcu_read_unlock();
1624                 } else {
1625                         inode->i_uid = GLOBAL_ROOT_UID;
1626                         inode->i_gid = GLOBAL_ROOT_GID;
1627                 }
1628                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1629                 security_task_to_inode(task, inode);
1630                 put_task_struct(task);
1631                 return 1;
1632         }
1633         d_drop(dentry);
1634         return 0;
1635 }
1636
1637 static int pid_delete_dentry(const struct dentry * dentry)
1638 {
1639         /* Is the task we represent dead?
1640          * If so, then don't put the dentry on the lru list,
1641          * kill it immediately.
1642          */
1643         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1644 }
1645
1646 const struct dentry_operations pid_dentry_operations =
1647 {
1648         .d_revalidate   = pid_revalidate,
1649         .d_delete       = pid_delete_dentry,
1650 };
1651
1652 /* Lookups */
1653
1654 /*
1655  * Fill a directory entry.
1656  *
1657  * If possible create the dcache entry and derive our inode number and
1658  * file type from dcache entry.
1659  *
1660  * Since all of the proc inode numbers are dynamically generated, the inode
1661  * numbers do not exist until the inode is cache.  This means creating the
1662  * the dcache entry in readdir is necessary to keep the inode numbers
1663  * reported by readdir in sync with the inode numbers reported
1664  * by stat.
1665  */
1666 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1667         const char *name, int len,
1668         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669 {
1670         struct dentry *child, *dir = filp->f_path.dentry;
1671         struct inode *inode;
1672         struct qstr qname;
1673         ino_t ino = 0;
1674         unsigned type = DT_UNKNOWN;
1675
1676         qname.name = name;
1677         qname.len  = len;
1678         qname.hash = full_name_hash(name, len);
1679
1680         child = d_lookup(dir, &qname);
1681         if (!child) {
1682                 struct dentry *new;
1683                 new = d_alloc(dir, &qname);
1684                 if (new) {
1685                         child = instantiate(dir->d_inode, new, task, ptr);
1686                         if (child)
1687                                 dput(new);
1688                         else
1689                                 child = new;
1690                 }
1691         }
1692         if (!child || IS_ERR(child) || !child->d_inode)
1693                 goto end_instantiate;
1694         inode = child->d_inode;
1695         if (inode) {
1696                 ino = inode->i_ino;
1697                 type = inode->i_mode >> 12;
1698         }
1699         dput(child);
1700 end_instantiate:
1701         if (!ino)
1702                 ino = find_inode_number(dir, &qname);
1703         if (!ino)
1704                 ino = 1;
1705         return filldir(dirent, name, len, filp->f_pos, ino, type);
1706 }
1707
1708 static unsigned name_to_int(struct dentry *dentry)
1709 {
1710         const char *name = dentry->d_name.name;
1711         int len = dentry->d_name.len;
1712         unsigned n = 0;
1713
1714         if (len > 1 && *name == '0')
1715                 goto out;
1716         while (len-- > 0) {
1717                 unsigned c = *name++ - '0';
1718                 if (c > 9)
1719                         goto out;
1720                 if (n >= (~0U-9)/10)
1721                         goto out;
1722                 n *= 10;
1723                 n += c;
1724         }
1725         return n;
1726 out:
1727         return ~0U;
1728 }
1729
1730 #define PROC_FDINFO_MAX 64
1731
1732 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1733 {
1734         struct task_struct *task = get_proc_task(inode);
1735         struct files_struct *files = NULL;
1736         struct file *file;
1737         int fd = proc_fd(inode);
1738
1739         if (task) {
1740                 files = get_files_struct(task);
1741                 put_task_struct(task);
1742         }
1743         if (files) {
1744                 /*
1745                  * We are not taking a ref to the file structure, so we must
1746                  * hold ->file_lock.
1747                  */
1748                 spin_lock(&files->file_lock);
1749                 file = fcheck_files(files, fd);
1750                 if (file) {
1751                         unsigned int f_flags;
1752                         struct fdtable *fdt;
1753
1754                         fdt = files_fdtable(files);
1755                         f_flags = file->f_flags & ~O_CLOEXEC;
1756                         if (close_on_exec(fd, fdt))
1757                                 f_flags |= O_CLOEXEC;
1758
1759                         if (path) {
1760                                 *path = file->f_path;
1761                                 path_get(&file->f_path);
1762                         }
1763                         if (info)
1764                                 snprintf(info, PROC_FDINFO_MAX,
1765                                          "pos:\t%lli\n"
1766                                          "flags:\t0%o\n",
1767                                          (long long) file->f_pos,
1768                                          f_flags);
1769                         spin_unlock(&files->file_lock);
1770                         put_files_struct(files);
1771                         return 0;
1772                 }
1773                 spin_unlock(&files->file_lock);
1774                 put_files_struct(files);
1775         }
1776         return -ENOENT;
1777 }
1778
1779 static int proc_fd_link(struct dentry *dentry, struct path *path)
1780 {
1781         return proc_fd_info(dentry->d_inode, path, NULL);
1782 }
1783
1784 static int tid_fd_revalidate(struct dentry *dentry, unsigned int flags)
1785 {
1786         struct inode *inode;
1787         struct task_struct *task;
1788         int fd;
1789         struct files_struct *files;
1790         const struct cred *cred;
1791
1792         if (flags & LOOKUP_RCU)
1793                 return -ECHILD;
1794
1795         inode = dentry->d_inode;
1796         task = get_proc_task(inode);
1797         fd = proc_fd(inode);
1798
1799         if (task) {
1800                 files = get_files_struct(task);
1801                 if (files) {
1802                         struct file *file;
1803                         rcu_read_lock();
1804                         file = fcheck_files(files, fd);
1805                         if (file) {
1806                                 unsigned f_mode = file->f_mode;
1807
1808                                 rcu_read_unlock();
1809                                 put_files_struct(files);
1810
1811                                 if (task_dumpable(task)) {
1812                                         rcu_read_lock();
1813                                         cred = __task_cred(task);
1814                                         inode->i_uid = cred->euid;
1815                                         inode->i_gid = cred->egid;
1816                                         rcu_read_unlock();
1817                                 } else {
1818                                         inode->i_uid = GLOBAL_ROOT_UID;
1819                                         inode->i_gid = GLOBAL_ROOT_GID;
1820                                 }
1821
1822                                 if (S_ISLNK(inode->i_mode)) {
1823                                         unsigned i_mode = S_IFLNK;
1824                                         if (f_mode & FMODE_READ)
1825                                                 i_mode |= S_IRUSR | S_IXUSR;
1826                                         if (f_mode & FMODE_WRITE)
1827                                                 i_mode |= S_IWUSR | S_IXUSR;
1828                                         inode->i_mode = i_mode;
1829                                 }
1830
1831                                 security_task_to_inode(task, inode);
1832                                 put_task_struct(task);
1833                                 return 1;
1834                         }
1835                         rcu_read_unlock();
1836                         put_files_struct(files);
1837                 }
1838                 put_task_struct(task);
1839         }
1840         d_drop(dentry);
1841         return 0;
1842 }
1843
1844 static const struct dentry_operations tid_fd_dentry_operations =
1845 {
1846         .d_revalidate   = tid_fd_revalidate,
1847         .d_delete       = pid_delete_dentry,
1848 };
1849
1850 static struct dentry *proc_fd_instantiate(struct inode *dir,
1851         struct dentry *dentry, struct task_struct *task, const void *ptr)
1852 {
1853         unsigned fd = (unsigned long)ptr;
1854         struct inode *inode;
1855         struct proc_inode *ei;
1856         struct dentry *error = ERR_PTR(-ENOENT);
1857
1858         inode = proc_pid_make_inode(dir->i_sb, task);
1859         if (!inode)
1860                 goto out;
1861         ei = PROC_I(inode);
1862         ei->fd = fd;
1863
1864         inode->i_mode = S_IFLNK;
1865         inode->i_op = &proc_pid_link_inode_operations;
1866         inode->i_size = 64;
1867         ei->op.proc_get_link = proc_fd_link;
1868         d_set_d_op(dentry, &tid_fd_dentry_operations);
1869         d_add(dentry, inode);
1870         /* Close the race of the process dying before we return the dentry */
1871         if (tid_fd_revalidate(dentry, 0))
1872                 error = NULL;
1873
1874  out:
1875         return error;
1876 }
1877
1878 static struct dentry *proc_lookupfd_common(struct inode *dir,
1879                                            struct dentry *dentry,
1880                                            instantiate_t instantiate)
1881 {
1882         struct task_struct *task = get_proc_task(dir);
1883         unsigned fd = name_to_int(dentry);
1884         struct dentry *result = ERR_PTR(-ENOENT);
1885
1886         if (!task)
1887                 goto out_no_task;
1888         if (fd == ~0U)
1889                 goto out;
1890
1891         result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1892 out:
1893         put_task_struct(task);
1894 out_no_task:
1895         return result;
1896 }
1897
1898 static int proc_readfd_common(struct file * filp, void * dirent,
1899                               filldir_t filldir, instantiate_t instantiate)
1900 {
1901         struct dentry *dentry = filp->f_path.dentry;
1902         struct inode *inode = dentry->d_inode;
1903         struct task_struct *p = get_proc_task(inode);
1904         unsigned int fd, ino;
1905         int retval;
1906         struct files_struct * files;
1907
1908         retval = -ENOENT;
1909         if (!p)
1910                 goto out_no_task;
1911         retval = 0;
1912
1913         fd = filp->f_pos;
1914         switch (fd) {
1915                 case 0:
1916                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1917                                 goto out;
1918                         filp->f_pos++;
1919                 case 1:
1920                         ino = parent_ino(dentry);
1921                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1922                                 goto out;
1923                         filp->f_pos++;
1924                 default:
1925                         files = get_files_struct(p);
1926                         if (!files)
1927                                 goto out;
1928                         rcu_read_lock();
1929                         for (fd = filp->f_pos-2;
1930                              fd < files_fdtable(files)->max_fds;
1931                              fd++, filp->f_pos++) {
1932                                 char name[PROC_NUMBUF];
1933                                 int len;
1934                                 int rv;
1935
1936                                 if (!fcheck_files(files, fd))
1937                                         continue;
1938                                 rcu_read_unlock();
1939
1940                                 len = snprintf(name, sizeof(name), "%d", fd);
1941                                 rv = proc_fill_cache(filp, dirent, filldir,
1942                                                      name, len, instantiate, p,
1943                                                      (void *)(unsigned long)fd);
1944                                 if (rv < 0)
1945                                         goto out_fd_loop;
1946                                 rcu_read_lock();
1947                         }
1948                         rcu_read_unlock();
1949 out_fd_loop:
1950                         put_files_struct(files);
1951         }
1952 out:
1953         put_task_struct(p);
1954 out_no_task:
1955         return retval;
1956 }
1957
1958 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1959                                     struct nameidata *nd)
1960 {
1961         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1962 }
1963
1964 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1965 {
1966         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1967 }
1968
1969 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1970                                       size_t len, loff_t *ppos)
1971 {
1972         char tmp[PROC_FDINFO_MAX];
1973         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1974         if (!err)
1975                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1976         return err;
1977 }
1978
1979 static const struct file_operations proc_fdinfo_file_operations = {
1980         .open           = nonseekable_open,
1981         .read           = proc_fdinfo_read,
1982         .llseek         = no_llseek,
1983 };
1984
1985 static const struct file_operations proc_fd_operations = {
1986         .read           = generic_read_dir,
1987         .readdir        = proc_readfd,
1988         .llseek         = default_llseek,
1989 };
1990
1991 #ifdef CONFIG_CHECKPOINT_RESTORE
1992
1993 /*
1994  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1995  * which represent vma start and end addresses.
1996  */
1997 static int dname_to_vma_addr(struct dentry *dentry,
1998                              unsigned long *start, unsigned long *end)
1999 {
2000         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2001                 return -EINVAL;
2002
2003         return 0;
2004 }
2005
2006 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2007 {
2008         unsigned long vm_start, vm_end;
2009         bool exact_vma_exists = false;
2010         struct mm_struct *mm = NULL;
2011         struct task_struct *task;
2012         const struct cred *cred;
2013         struct inode *inode;
2014         int status = 0;
2015
2016         if (flags & LOOKUP_RCU)
2017                 return -ECHILD;
2018
2019         if (!capable(CAP_SYS_ADMIN)) {
2020                 status = -EACCES;
2021                 goto out_notask;
2022         }
2023
2024         inode = dentry->d_inode;
2025         task = get_proc_task(inode);
2026         if (!task)
2027                 goto out_notask;
2028
2029         mm = mm_access(task, PTRACE_MODE_READ);
2030         if (IS_ERR_OR_NULL(mm))
2031                 goto out;
2032
2033         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2034                 down_read(&mm->mmap_sem);
2035                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2036                 up_read(&mm->mmap_sem);
2037         }
2038
2039         mmput(mm);
2040
2041         if (exact_vma_exists) {
2042                 if (task_dumpable(task)) {
2043                         rcu_read_lock();
2044                         cred = __task_cred(task);
2045                         inode->i_uid = cred->euid;
2046                         inode->i_gid = cred->egid;
2047                         rcu_read_unlock();
2048                 } else {
2049                         inode->i_uid = GLOBAL_ROOT_UID;
2050                         inode->i_gid = GLOBAL_ROOT_GID;
2051                 }
2052                 security_task_to_inode(task, inode);
2053                 status = 1;
2054         }
2055
2056 out:
2057         put_task_struct(task);
2058
2059 out_notask:
2060         if (status <= 0)
2061                 d_drop(dentry);
2062
2063         return status;
2064 }
2065
2066 static const struct dentry_operations tid_map_files_dentry_operations = {
2067         .d_revalidate   = map_files_d_revalidate,
2068         .d_delete       = pid_delete_dentry,
2069 };
2070
2071 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2072 {
2073         unsigned long vm_start, vm_end;
2074         struct vm_area_struct *vma;
2075         struct task_struct *task;
2076         struct mm_struct *mm;
2077         int rc;
2078
2079         rc = -ENOENT;
2080         task = get_proc_task(dentry->d_inode);
2081         if (!task)
2082                 goto out;
2083
2084         mm = get_task_mm(task);
2085         put_task_struct(task);
2086         if (!mm)
2087                 goto out;
2088
2089         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2090         if (rc)
2091                 goto out_mmput;
2092
2093         down_read(&mm->mmap_sem);
2094         vma = find_exact_vma(mm, vm_start, vm_end);
2095         if (vma && vma->vm_file) {
2096                 *path = vma->vm_file->f_path;
2097                 path_get(path);
2098                 rc = 0;
2099         }
2100         up_read(&mm->mmap_sem);
2101
2102 out_mmput:
2103         mmput(mm);
2104 out:
2105         return rc;
2106 }
2107
2108 struct map_files_info {
2109         struct file     *file;
2110         unsigned long   len;
2111         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2112 };
2113
2114 static struct dentry *
2115 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2116                            struct task_struct *task, const void *ptr)
2117 {
2118         const struct file *file = ptr;
2119         struct proc_inode *ei;
2120         struct inode *inode;
2121
2122         if (!file)
2123                 return ERR_PTR(-ENOENT);
2124
2125         inode = proc_pid_make_inode(dir->i_sb, task);
2126         if (!inode)
2127                 return ERR_PTR(-ENOENT);
2128
2129         ei = PROC_I(inode);
2130         ei->op.proc_get_link = proc_map_files_get_link;
2131
2132         inode->i_op = &proc_pid_link_inode_operations;
2133         inode->i_size = 64;
2134         inode->i_mode = S_IFLNK;
2135
2136         if (file->f_mode & FMODE_READ)
2137                 inode->i_mode |= S_IRUSR;
2138         if (file->f_mode & FMODE_WRITE)
2139                 inode->i_mode |= S_IWUSR;
2140
2141         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2142         d_add(dentry, inode);
2143
2144         return NULL;
2145 }
2146
2147 static struct dentry *proc_map_files_lookup(struct inode *dir,
2148                 struct dentry *dentry, struct nameidata *nd)
2149 {
2150         unsigned long vm_start, vm_end;
2151         struct vm_area_struct *vma;
2152         struct task_struct *task;
2153         struct dentry *result;
2154         struct mm_struct *mm;
2155
2156         result = ERR_PTR(-EACCES);
2157         if (!capable(CAP_SYS_ADMIN))
2158                 goto out;
2159
2160         result = ERR_PTR(-ENOENT);
2161         task = get_proc_task(dir);
2162         if (!task)
2163                 goto out;
2164
2165         result = ERR_PTR(-EACCES);
2166         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2167                 goto out_put_task;
2168
2169         result = ERR_PTR(-ENOENT);
2170         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2171                 goto out_put_task;
2172
2173         mm = get_task_mm(task);
2174         if (!mm)
2175                 goto out_put_task;
2176
2177         down_read(&mm->mmap_sem);
2178         vma = find_exact_vma(mm, vm_start, vm_end);
2179         if (!vma)
2180                 goto out_no_vma;
2181
2182         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2183
2184 out_no_vma:
2185         up_read(&mm->mmap_sem);
2186         mmput(mm);
2187 out_put_task:
2188         put_task_struct(task);
2189 out:
2190         return result;
2191 }
2192
2193 static const struct inode_operations proc_map_files_inode_operations = {
2194         .lookup         = proc_map_files_lookup,
2195         .permission     = proc_fd_permission,
2196         .setattr        = proc_setattr,
2197 };
2198
2199 static int
2200 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2201 {
2202         struct dentry *dentry = filp->f_path.dentry;
2203         struct inode *inode = dentry->d_inode;
2204         struct vm_area_struct *vma;
2205         struct task_struct *task;
2206         struct mm_struct *mm;
2207         ino_t ino;
2208         int ret;
2209
2210         ret = -EACCES;
2211         if (!capable(CAP_SYS_ADMIN))
2212                 goto out;
2213
2214         ret = -ENOENT;
2215         task = get_proc_task(inode);
2216         if (!task)
2217                 goto out;
2218
2219         ret = -EACCES;
2220         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2221                 goto out_put_task;
2222
2223         ret = 0;
2224         switch (filp->f_pos) {
2225         case 0:
2226                 ino = inode->i_ino;
2227                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2228                         goto out_put_task;
2229                 filp->f_pos++;
2230         case 1:
2231                 ino = parent_ino(dentry);
2232                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2233                         goto out_put_task;
2234                 filp->f_pos++;
2235         default:
2236         {
2237                 unsigned long nr_files, pos, i;
2238                 struct flex_array *fa = NULL;
2239                 struct map_files_info info;
2240                 struct map_files_info *p;
2241
2242                 mm = get_task_mm(task);
2243                 if (!mm)
2244                         goto out_put_task;
2245                 down_read(&mm->mmap_sem);
2246
2247                 nr_files = 0;
2248
2249                 /*
2250                  * We need two passes here:
2251                  *
2252                  *  1) Collect vmas of mapped files with mmap_sem taken
2253                  *  2) Release mmap_sem and instantiate entries
2254                  *
2255                  * otherwise we get lockdep complained, since filldir()
2256                  * routine might require mmap_sem taken in might_fault().
2257                  */
2258
2259                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2260                         if (vma->vm_file && ++pos > filp->f_pos)
2261                                 nr_files++;
2262                 }
2263
2264                 if (nr_files) {
2265                         fa = flex_array_alloc(sizeof(info), nr_files,
2266                                                 GFP_KERNEL);
2267                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2268                                                         GFP_KERNEL)) {
2269                                 ret = -ENOMEM;
2270                                 if (fa)
2271                                         flex_array_free(fa);
2272                                 up_read(&mm->mmap_sem);
2273                                 mmput(mm);
2274                                 goto out_put_task;
2275                         }
2276                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2277                                         vma = vma->vm_next) {
2278                                 if (!vma->vm_file)
2279                                         continue;
2280                                 if (++pos <= filp->f_pos)
2281                                         continue;
2282
2283                                 get_file(vma->vm_file);
2284                                 info.file = vma->vm_file;
2285                                 info.len = snprintf(info.name,
2286                                                 sizeof(info.name), "%lx-%lx",
2287                                                 vma->vm_start, vma->vm_end);
2288                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2289                                         BUG();
2290                         }
2291                 }
2292                 up_read(&mm->mmap_sem);
2293
2294                 for (i = 0; i < nr_files; i++) {
2295                         p = flex_array_get(fa, i);
2296                         ret = proc_fill_cache(filp, dirent, filldir,
2297                                               p->name, p->len,
2298                                               proc_map_files_instantiate,
2299                                               task, p->file);
2300                         if (ret)
2301                                 break;
2302                         filp->f_pos++;
2303                         fput(p->file);
2304                 }
2305                 for (; i < nr_files; i++) {
2306                         /*
2307                          * In case of error don't forget
2308                          * to put rest of file refs.
2309                          */
2310                         p = flex_array_get(fa, i);
2311                         fput(p->file);
2312                 }
2313                 if (fa)
2314                         flex_array_free(fa);
2315                 mmput(mm);
2316         }
2317         }
2318
2319 out_put_task:
2320         put_task_struct(task);
2321 out:
2322         return ret;
2323 }
2324
2325 static const struct file_operations proc_map_files_operations = {
2326         .read           = generic_read_dir,
2327         .readdir        = proc_map_files_readdir,
2328         .llseek         = default_llseek,
2329 };
2330
2331 #endif /* CONFIG_CHECKPOINT_RESTORE */
2332
2333 /*
2334  * /proc/pid/fd needs a special permission handler so that a process can still
2335  * access /proc/self/fd after it has executed a setuid().
2336  */
2337 static int proc_fd_permission(struct inode *inode, int mask)
2338 {
2339         int rv = generic_permission(inode, mask);
2340         if (rv == 0)
2341                 return 0;
2342         if (task_pid(current) == proc_pid(inode))
2343                 rv = 0;
2344         return rv;
2345 }
2346
2347 /*
2348  * proc directories can do almost nothing..
2349  */
2350 static const struct inode_operations proc_fd_inode_operations = {
2351         .lookup         = proc_lookupfd,
2352         .permission     = proc_fd_permission,
2353         .setattr        = proc_setattr,
2354 };
2355
2356 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2357         struct dentry *dentry, struct task_struct *task, const void *ptr)
2358 {
2359         unsigned fd = (unsigned long)ptr;
2360         struct inode *inode;
2361         struct proc_inode *ei;
2362         struct dentry *error = ERR_PTR(-ENOENT);
2363
2364         inode = proc_pid_make_inode(dir->i_sb, task);
2365         if (!inode)
2366                 goto out;
2367         ei = PROC_I(inode);
2368         ei->fd = fd;
2369         inode->i_mode = S_IFREG | S_IRUSR;
2370         inode->i_fop = &proc_fdinfo_file_operations;
2371         d_set_d_op(dentry, &tid_fd_dentry_operations);
2372         d_add(dentry, inode);
2373         /* Close the race of the process dying before we return the dentry */
2374         if (tid_fd_revalidate(dentry, 0))
2375                 error = NULL;
2376
2377  out:
2378         return error;
2379 }
2380
2381 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2382                                         struct dentry *dentry,
2383                                         struct nameidata *nd)
2384 {
2385         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2386 }
2387
2388 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2389 {
2390         return proc_readfd_common(filp, dirent, filldir,
2391                                   proc_fdinfo_instantiate);
2392 }
2393
2394 static const struct file_operations proc_fdinfo_operations = {
2395         .read           = generic_read_dir,
2396         .readdir        = proc_readfdinfo,
2397         .llseek         = default_llseek,
2398 };
2399
2400 /*
2401  * proc directories can do almost nothing..
2402  */
2403 static const struct inode_operations proc_fdinfo_inode_operations = {
2404         .lookup         = proc_lookupfdinfo,
2405         .setattr        = proc_setattr,
2406 };
2407
2408
2409 static struct dentry *proc_pident_instantiate(struct inode *dir,
2410         struct dentry *dentry, struct task_struct *task, const void *ptr)
2411 {
2412         const struct pid_entry *p = ptr;
2413         struct inode *inode;
2414         struct proc_inode *ei;
2415         struct dentry *error = ERR_PTR(-ENOENT);
2416
2417         inode = proc_pid_make_inode(dir->i_sb, task);
2418         if (!inode)
2419                 goto out;
2420
2421         ei = PROC_I(inode);
2422         inode->i_mode = p->mode;
2423         if (S_ISDIR(inode->i_mode))
2424                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2425         if (p->iop)
2426                 inode->i_op = p->iop;
2427         if (p->fop)
2428                 inode->i_fop = p->fop;
2429         ei->op = p->op;
2430         d_set_d_op(dentry, &pid_dentry_operations);
2431         d_add(dentry, inode);
2432         /* Close the race of the process dying before we return the dentry */
2433         if (pid_revalidate(dentry, 0))
2434                 error = NULL;
2435 out:
2436         return error;
2437 }
2438
2439 static struct dentry *proc_pident_lookup(struct inode *dir, 
2440                                          struct dentry *dentry,
2441                                          const struct pid_entry *ents,
2442                                          unsigned int nents)
2443 {
2444         struct dentry *error;
2445         struct task_struct *task = get_proc_task(dir);
2446         const struct pid_entry *p, *last;
2447
2448         error = ERR_PTR(-ENOENT);
2449
2450         if (!task)
2451                 goto out_no_task;
2452
2453         /*
2454          * Yes, it does not scale. And it should not. Don't add
2455          * new entries into /proc/<tgid>/ without very good reasons.
2456          */
2457         last = &ents[nents - 1];
2458         for (p = ents; p <= last; p++) {
2459                 if (p->len != dentry->d_name.len)
2460                         continue;
2461                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2462                         break;
2463         }
2464         if (p > last)
2465                 goto out;
2466
2467         error = proc_pident_instantiate(dir, dentry, task, p);
2468 out:
2469         put_task_struct(task);
2470 out_no_task:
2471         return error;
2472 }
2473
2474 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2475         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2476 {
2477         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2478                                 proc_pident_instantiate, task, p);
2479 }
2480
2481 static int proc_pident_readdir(struct file *filp,
2482                 void *dirent, filldir_t filldir,
2483                 const struct pid_entry *ents, unsigned int nents)
2484 {
2485         int i;
2486         struct dentry *dentry = filp->f_path.dentry;
2487         struct inode *inode = dentry->d_inode;
2488         struct task_struct *task = get_proc_task(inode);
2489         const struct pid_entry *p, *last;
2490         ino_t ino;
2491         int ret;
2492
2493         ret = -ENOENT;
2494         if (!task)
2495                 goto out_no_task;
2496
2497         ret = 0;
2498         i = filp->f_pos;
2499         switch (i) {
2500         case 0:
2501                 ino = inode->i_ino;
2502                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2503                         goto out;
2504                 i++;
2505                 filp->f_pos++;
2506                 /* fall through */
2507         case 1:
2508                 ino = parent_ino(dentry);
2509                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2510                         goto out;
2511                 i++;
2512                 filp->f_pos++;
2513                 /* fall through */
2514         default:
2515                 i -= 2;
2516                 if (i >= nents) {
2517                         ret = 1;
2518                         goto out;
2519                 }
2520                 p = ents + i;
2521                 last = &ents[nents - 1];
2522                 while (p <= last) {
2523                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2524                                 goto out;
2525                         filp->f_pos++;
2526                         p++;
2527                 }
2528         }
2529
2530         ret = 1;
2531 out:
2532         put_task_struct(task);
2533 out_no_task:
2534         return ret;
2535 }
2536
2537 #ifdef CONFIG_SECURITY
2538 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2539                                   size_t count, loff_t *ppos)
2540 {
2541         struct inode * inode = file->f_path.dentry->d_inode;
2542         char *p = NULL;
2543         ssize_t length;
2544         struct task_struct *task = get_proc_task(inode);
2545
2546         if (!task)
2547                 return -ESRCH;
2548
2549         length = security_getprocattr(task,
2550                                       (char*)file->f_path.dentry->d_name.name,
2551                                       &p);
2552         put_task_struct(task);
2553         if (length > 0)
2554                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2555         kfree(p);
2556         return length;
2557 }
2558
2559 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2560                                    size_t count, loff_t *ppos)
2561 {
2562         struct inode * inode = file->f_path.dentry->d_inode;
2563         char *page;
2564         ssize_t length;
2565         struct task_struct *task = get_proc_task(inode);
2566
2567         length = -ESRCH;
2568         if (!task)
2569                 goto out_no_task;
2570         if (count > PAGE_SIZE)
2571                 count = PAGE_SIZE;
2572
2573         /* No partial writes. */
2574         length = -EINVAL;
2575         if (*ppos != 0)
2576                 goto out;
2577
2578         length = -ENOMEM;
2579         page = (char*)__get_free_page(GFP_TEMPORARY);
2580         if (!page)
2581                 goto out;
2582
2583         length = -EFAULT;
2584         if (copy_from_user(page, buf, count))
2585                 goto out_free;
2586
2587         /* Guard against adverse ptrace interaction */
2588         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2589         if (length < 0)
2590                 goto out_free;
2591
2592         length = security_setprocattr(task,
2593                                       (char*)file->f_path.dentry->d_name.name,
2594                                       (void*)page, count);
2595         mutex_unlock(&task->signal->cred_guard_mutex);
2596 out_free:
2597         free_page((unsigned long) page);
2598 out:
2599         put_task_struct(task);
2600 out_no_task:
2601         return length;
2602 }
2603
2604 static const struct file_operations proc_pid_attr_operations = {
2605         .read           = proc_pid_attr_read,
2606         .write          = proc_pid_attr_write,
2607         .llseek         = generic_file_llseek,
2608 };
2609
2610 static const struct pid_entry attr_dir_stuff[] = {
2611         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2612         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2613         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2614         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2617 };
2618
2619 static int proc_attr_dir_readdir(struct file * filp,
2620                              void * dirent, filldir_t filldir)
2621 {
2622         return proc_pident_readdir(filp,dirent,filldir,
2623                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2624 }
2625
2626 static const struct file_operations proc_attr_dir_operations = {
2627         .read           = generic_read_dir,
2628         .readdir        = proc_attr_dir_readdir,
2629         .llseek         = default_llseek,
2630 };
2631
2632 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2633                                 struct dentry *dentry, struct nameidata *nd)
2634 {
2635         return proc_pident_lookup(dir, dentry,
2636                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2637 }
2638
2639 static const struct inode_operations proc_attr_dir_inode_operations = {
2640         .lookup         = proc_attr_dir_lookup,
2641         .getattr        = pid_getattr,
2642         .setattr        = proc_setattr,
2643 };
2644
2645 #endif
2646
2647 #ifdef CONFIG_ELF_CORE
2648 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2649                                          size_t count, loff_t *ppos)
2650 {
2651         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2652         struct mm_struct *mm;
2653         char buffer[PROC_NUMBUF];
2654         size_t len;
2655         int ret;
2656
2657         if (!task)
2658                 return -ESRCH;
2659
2660         ret = 0;
2661         mm = get_task_mm(task);
2662         if (mm) {
2663                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2664                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2665                                 MMF_DUMP_FILTER_SHIFT));
2666                 mmput(mm);
2667                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2668         }
2669
2670         put_task_struct(task);
2671
2672         return ret;
2673 }
2674
2675 static ssize_t proc_coredump_filter_write(struct file *file,
2676                                           const char __user *buf,
2677                                           size_t count,
2678                                           loff_t *ppos)
2679 {
2680         struct task_struct *task;
2681         struct mm_struct *mm;
2682         char buffer[PROC_NUMBUF], *end;
2683         unsigned int val;
2684         int ret;
2685         int i;
2686         unsigned long mask;
2687
2688         ret = -EFAULT;
2689         memset(buffer, 0, sizeof(buffer));
2690         if (count > sizeof(buffer) - 1)
2691                 count = sizeof(buffer) - 1;
2692         if (copy_from_user(buffer, buf, count))
2693                 goto out_no_task;
2694
2695         ret = -EINVAL;
2696         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2697         if (*end == '\n')
2698                 end++;
2699         if (end - buffer == 0)
2700                 goto out_no_task;
2701
2702         ret = -ESRCH;
2703         task = get_proc_task(file->f_dentry->d_inode);
2704         if (!task)
2705                 goto out_no_task;
2706
2707         ret = end - buffer;
2708         mm = get_task_mm(task);
2709         if (!mm)
2710                 goto out_no_mm;
2711
2712         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2713                 if (val & mask)
2714                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2715                 else
2716                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2717         }
2718
2719         mmput(mm);
2720  out_no_mm:
2721         put_task_struct(task);
2722  out_no_task:
2723         return ret;
2724 }
2725
2726 static const struct file_operations proc_coredump_filter_operations = {
2727         .read           = proc_coredump_filter_read,
2728         .write          = proc_coredump_filter_write,
2729         .llseek         = generic_file_llseek,
2730 };
2731 #endif
2732
2733 /*
2734  * /proc/self:
2735  */
2736 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2737                               int buflen)
2738 {
2739         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2740         pid_t tgid = task_tgid_nr_ns(current, ns);
2741         char tmp[PROC_NUMBUF];
2742         if (!tgid)
2743                 return -ENOENT;
2744         sprintf(tmp, "%d", tgid);
2745         return vfs_readlink(dentry,buffer,buflen,tmp);
2746 }
2747
2748 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2749 {
2750         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2751         pid_t tgid = task_tgid_nr_ns(current, ns);
2752         char *name = ERR_PTR(-ENOENT);
2753         if (tgid) {
2754                 name = __getname();
2755                 if (!name)
2756                         name = ERR_PTR(-ENOMEM);
2757                 else
2758                         sprintf(name, "%d", tgid);
2759         }
2760         nd_set_link(nd, name);
2761         return NULL;
2762 }
2763
2764 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2765                                 void *cookie)
2766 {
2767         char *s = nd_get_link(nd);
2768         if (!IS_ERR(s))
2769                 __putname(s);
2770 }
2771
2772 static const struct inode_operations proc_self_inode_operations = {
2773         .readlink       = proc_self_readlink,
2774         .follow_link    = proc_self_follow_link,
2775         .put_link       = proc_self_put_link,
2776 };
2777
2778 /*
2779  * proc base
2780  *
2781  * These are the directory entries in the root directory of /proc
2782  * that properly belong to the /proc filesystem, as they describe
2783  * describe something that is process related.
2784  */
2785 static const struct pid_entry proc_base_stuff[] = {
2786         NOD("self", S_IFLNK|S_IRWXUGO,
2787                 &proc_self_inode_operations, NULL, {}),
2788 };
2789
2790 static struct dentry *proc_base_instantiate(struct inode *dir,
2791         struct dentry *dentry, struct task_struct *task, const void *ptr)
2792 {
2793         const struct pid_entry *p = ptr;
2794         struct inode *inode;
2795         struct proc_inode *ei;
2796         struct dentry *error;
2797
2798         /* Allocate the inode */
2799         error = ERR_PTR(-ENOMEM);
2800         inode = new_inode(dir->i_sb);
2801         if (!inode)
2802                 goto out;
2803
2804         /* Initialize the inode */
2805         ei = PROC_I(inode);
2806         inode->i_ino = get_next_ino();
2807         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2808
2809         /*
2810          * grab the reference to the task.
2811          */
2812         ei->pid = get_task_pid(task, PIDTYPE_PID);
2813         if (!ei->pid)
2814                 goto out_iput;
2815
2816         inode->i_mode = p->mode;
2817         if (S_ISDIR(inode->i_mode))
2818                 set_nlink(inode, 2);
2819         if (S_ISLNK(inode->i_mode))
2820                 inode->i_size = 64;
2821         if (p->iop)
2822                 inode->i_op = p->iop;
2823         if (p->fop)
2824                 inode->i_fop = p->fop;
2825         ei->op = p->op;
2826         d_add(dentry, inode);
2827         error = NULL;
2828 out:
2829         return error;
2830 out_iput:
2831         iput(inode);
2832         goto out;
2833 }
2834
2835 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2836 {
2837         struct dentry *error;
2838         struct task_struct *task = get_proc_task(dir);
2839         const struct pid_entry *p, *last;
2840
2841         error = ERR_PTR(-ENOENT);
2842
2843         if (!task)
2844                 goto out_no_task;
2845
2846         /* Lookup the directory entry */
2847         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2848         for (p = proc_base_stuff; p <= last; p++) {
2849                 if (p->len != dentry->d_name.len)
2850                         continue;
2851                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2852                         break;
2853         }
2854         if (p > last)
2855                 goto out;
2856
2857         error = proc_base_instantiate(dir, dentry, task, p);
2858
2859 out:
2860         put_task_struct(task);
2861 out_no_task:
2862         return error;
2863 }
2864
2865 static int proc_base_fill_cache(struct file *filp, void *dirent,
2866         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2867 {
2868         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2869                                 proc_base_instantiate, task, p);
2870 }
2871
2872 #ifdef CONFIG_TASK_IO_ACCOUNTING
2873 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2874 {
2875         struct task_io_accounting acct = task->ioac;
2876         unsigned long flags;
2877         int result;
2878
2879         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2880         if (result)
2881                 return result;
2882
2883         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2884                 result = -EACCES;
2885                 goto out_unlock;
2886         }
2887
2888         if (whole && lock_task_sighand(task, &flags)) {
2889                 struct task_struct *t = task;
2890
2891                 task_io_accounting_add(&acct, &task->signal->ioac);
2892                 while_each_thread(task, t)
2893                         task_io_accounting_add(&acct, &t->ioac);
2894
2895                 unlock_task_sighand(task, &flags);
2896         }
2897         result = sprintf(buffer,
2898                         "rchar: %llu\n"
2899                         "wchar: %llu\n"
2900                         "syscr: %llu\n"
2901                         "syscw: %llu\n"
2902                         "read_bytes: %llu\n"
2903                         "write_bytes: %llu\n"
2904                         "cancelled_write_bytes: %llu\n",
2905                         (unsigned long long)acct.rchar,
2906                         (unsigned long long)acct.wchar,
2907                         (unsigned long long)acct.syscr,
2908                         (unsigned long long)acct.syscw,
2909                         (unsigned long long)acct.read_bytes,
2910                         (unsigned long long)acct.write_bytes,
2911                         (unsigned long long)acct.cancelled_write_bytes);
2912 out_unlock:
2913         mutex_unlock(&task->signal->cred_guard_mutex);
2914         return result;
2915 }
2916
2917 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2918 {
2919         return do_io_accounting(task, buffer, 0);
2920 }
2921
2922 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2923 {
2924         return do_io_accounting(task, buffer, 1);
2925 }
2926 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2927
2928 #ifdef CONFIG_USER_NS
2929 static int proc_id_map_open(struct inode *inode, struct file *file,
2930         struct seq_operations *seq_ops)
2931 {
2932         struct user_namespace *ns = NULL;
2933         struct task_struct *task;
2934         struct seq_file *seq;
2935         int ret = -EINVAL;
2936
2937         task = get_proc_task(inode);
2938         if (task) {
2939                 rcu_read_lock();
2940                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2941                 rcu_read_unlock();
2942                 put_task_struct(task);
2943         }
2944         if (!ns)
2945                 goto err;
2946
2947         ret = seq_open(file, seq_ops);
2948         if (ret)
2949                 goto err_put_ns;
2950
2951         seq = file->private_data;
2952         seq->private = ns;
2953
2954         return 0;
2955 err_put_ns:
2956         put_user_ns(ns);
2957 err:
2958         return ret;
2959 }
2960
2961 static int proc_id_map_release(struct inode *inode, struct file *file)
2962 {
2963         struct seq_file *seq = file->private_data;
2964         struct user_namespace *ns = seq->private;
2965         put_user_ns(ns);
2966         return seq_release(inode, file);
2967 }
2968
2969 static int proc_uid_map_open(struct inode *inode, struct file *file)
2970 {
2971         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2972 }
2973
2974 static int proc_gid_map_open(struct inode *inode, struct file *file)
2975 {
2976         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2977 }
2978
2979 static const struct file_operations proc_uid_map_operations = {
2980         .open           = proc_uid_map_open,
2981         .write          = proc_uid_map_write,
2982         .read           = seq_read,
2983         .llseek         = seq_lseek,
2984         .release        = proc_id_map_release,
2985 };
2986
2987 static const struct file_operations proc_gid_map_operations = {
2988         .open           = proc_gid_map_open,
2989         .write          = proc_gid_map_write,
2990         .read           = seq_read,
2991         .llseek         = seq_lseek,
2992         .release        = proc_id_map_release,
2993 };
2994 #endif /* CONFIG_USER_NS */
2995
2996 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2997                                 struct pid *pid, struct task_struct *task)
2998 {
2999         int err = lock_trace(task);
3000         if (!err) {
3001                 seq_printf(m, "%08x\n", task->personality);
3002                 unlock_trace(task);
3003         }
3004         return err;
3005 }
3006
3007 /*
3008  * Thread groups
3009  */
3010 static const struct file_operations proc_task_operations;
3011 static const struct inode_operations proc_task_inode_operations;
3012
3013 static const struct pid_entry tgid_base_stuff[] = {
3014         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3015         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3016 #ifdef CONFIG_CHECKPOINT_RESTORE
3017         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3018 #endif
3019         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3020         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3021 #ifdef CONFIG_NET
3022         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3023 #endif
3024         REG("environ",    S_IRUSR, proc_environ_operations),
3025         INF("auxv",       S_IRUSR, proc_pid_auxv),
3026         ONE("status",     S_IRUGO, proc_pid_status),
3027         ONE("personality", S_IRUGO, proc_pid_personality),
3028         INF("limits",     S_IRUGO, proc_pid_limits),
3029 #ifdef CONFIG_SCHED_DEBUG
3030         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3031 #endif
3032 #ifdef CONFIG_SCHED_AUTOGROUP
3033         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3034 #endif
3035         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3036 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3037         INF("syscall",    S_IRUGO, proc_pid_syscall),
3038 #endif
3039         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3040         ONE("stat",       S_IRUGO, proc_tgid_stat),
3041         ONE("statm",      S_IRUGO, proc_pid_statm),
3042         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3043 #ifdef CONFIG_NUMA
3044         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3045 #endif
3046         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3047         LNK("cwd",        proc_cwd_link),
3048         LNK("root",       proc_root_link),
3049         LNK("exe",        proc_exe_link),
3050         REG("mounts",     S_IRUGO, proc_mounts_operations),
3051         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3052         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3053 #ifdef CONFIG_PROC_PAGE_MONITOR
3054         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3055         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3056         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3057 #endif
3058 #ifdef CONFIG_SECURITY
3059         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3060 #endif
3061 #ifdef CONFIG_KALLSYMS
3062         INF("wchan",      S_IRUGO, proc_pid_wchan),
3063 #endif
3064 #ifdef CONFIG_STACKTRACE
3065         ONE("stack",      S_IRUGO, proc_pid_stack),
3066 #endif
3067 #ifdef CONFIG_SCHEDSTATS
3068         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3069 #endif
3070 #ifdef CONFIG_LATENCYTOP
3071         REG("latency",  S_IRUGO, proc_lstats_operations),
3072 #endif
3073 #ifdef CONFIG_PROC_PID_CPUSET
3074         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3075 #endif
3076 #ifdef CONFIG_CGROUPS
3077         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3078 #endif
3079         INF("oom_score",  S_IRUGO, proc_oom_score),
3080         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3081         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3082 #ifdef CONFIG_AUDITSYSCALL
3083         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3084         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3085 #endif
3086 #ifdef CONFIG_FAULT_INJECTION
3087         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3088 #endif
3089 #ifdef CONFIG_ELF_CORE
3090         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3091 #endif
3092 #ifdef CONFIG_TASK_IO_ACCOUNTING
3093         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3094 #endif
3095 #ifdef CONFIG_HARDWALL
3096         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3097 #endif
3098 #ifdef CONFIG_USER_NS
3099         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3100         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3101 #endif
3102 };
3103
3104 static int proc_tgid_base_readdir(struct file * filp,
3105                              void * dirent, filldir_t filldir)
3106 {
3107         return proc_pident_readdir(filp,dirent,filldir,
3108                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3109 }
3110
3111 static const struct file_operations proc_tgid_base_operations = {
3112         .read           = generic_read_dir,
3113         .readdir        = proc_tgid_base_readdir,
3114         .llseek         = default_llseek,
3115 };
3116
3117 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3118         return proc_pident_lookup(dir, dentry,
3119                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3120 }
3121
3122 static const struct inode_operations proc_tgid_base_inode_operations = {
3123         .lookup         = proc_tgid_base_lookup,
3124         .getattr        = pid_getattr,
3125         .setattr        = proc_setattr,
3126         .permission     = proc_pid_permission,
3127 };
3128
3129 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3130 {
3131         struct dentry *dentry, *leader, *dir;
3132         char buf[PROC_NUMBUF];
3133         struct qstr name;
3134
3135         name.name = buf;
3136         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3137         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3138         if (dentry) {
3139                 shrink_dcache_parent(dentry);
3140                 d_drop(dentry);
3141                 dput(dentry);
3142         }
3143
3144         name.name = buf;
3145         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3146         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3147         if (!leader)
3148                 goto out;
3149
3150         name.name = "task";
3151         name.len = strlen(name.name);
3152         dir = d_hash_and_lookup(leader, &name);
3153         if (!dir)
3154                 goto out_put_leader;
3155
3156         name.name = buf;
3157         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3158         dentry = d_hash_and_lookup(dir, &name);
3159         if (dentry) {
3160                 shrink_dcache_parent(dentry);
3161                 d_drop(dentry);
3162                 dput(dentry);
3163         }
3164
3165         dput(dir);
3166 out_put_leader:
3167         dput(leader);
3168 out:
3169         return;
3170 }
3171
3172 /**
3173  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3174  * @task: task that should be flushed.
3175  *
3176  * When flushing dentries from proc, one needs to flush them from global
3177  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3178  * in. This call is supposed to do all of this job.
3179  *
3180  * Looks in the dcache for
3181  * /proc/@pid
3182  * /proc/@tgid/task/@pid
3183  * if either directory is present flushes it and all of it'ts children
3184  * from the dcache.
3185  *
3186  * It is safe and reasonable to cache /proc entries for a task until
3187  * that task exits.  After that they just clog up the dcache with
3188  * useless entries, possibly causing useful dcache entries to be
3189  * flushed instead.  This routine is proved to flush those useless
3190  * dcache entries at process exit time.
3191  *
3192  * NOTE: This routine is just an optimization so it does not guarantee
3193  *       that no dcache entries will exist at process exit time it
3194  *       just makes it very unlikely that any will persist.
3195  */
3196
3197 void proc_flush_task(struct task_struct *task)
3198 {
3199         int i;
3200         struct pid *pid, *tgid;
3201         struct upid *upid;
3202
3203         pid = task_pid(task);
3204         tgid = task_tgid(task);
3205
3206         for (i = 0; i <= pid->level; i++) {
3207                 upid = &pid->numbers[i];
3208                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3209                                         tgid->numbers[i].nr);
3210         }
3211
3212         upid = &pid->numbers[pid->level];
3213         if (upid->nr == 1)
3214                 pid_ns_release_proc(upid->ns);
3215 }
3216
3217 static struct dentry *proc_pid_instantiate(struct inode *dir,
3218                                            struct dentry * dentry,
3219                                            struct task_struct *task, const void *ptr)
3220 {
3221         struct dentry *error = ERR_PTR(-ENOENT);
3222         struct inode *inode;
3223
3224         inode = proc_pid_make_inode(dir->i_sb, task);
3225         if (!inode)
3226                 goto out;
3227
3228         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3229         inode->i_op = &proc_tgid_base_inode_operations;
3230         inode->i_fop = &proc_tgid_base_operations;
3231         inode->i_flags|=S_IMMUTABLE;
3232
3233         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3234                                                   ARRAY_SIZE(tgid_base_stuff)));
3235
3236         d_set_d_op(dentry, &pid_dentry_operations);
3237
3238         d_add(dentry, inode);
3239         /* Close the race of the process dying before we return the dentry */
3240         if (pid_revalidate(dentry, 0))
3241                 error = NULL;
3242 out:
3243         return error;
3244 }
3245
3246 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3247 {
3248         struct dentry *result;
3249         struct task_struct *task;
3250         unsigned tgid;
3251         struct pid_namespace *ns;
3252
3253         result = proc_base_lookup(dir, dentry);
3254         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3255                 goto out;
3256
3257         tgid = name_to_int(dentry);
3258         if (tgid == ~0U)
3259                 goto out;
3260
3261         ns = dentry->d_sb->s_fs_info;
3262         rcu_read_lock();
3263         task = find_task_by_pid_ns(tgid, ns);
3264         if (task)
3265                 get_task_struct(task);
3266         rcu_read_unlock();
3267         if (!task)
3268                 goto out;
3269
3270         result = proc_pid_instantiate(dir, dentry, task, NULL);
3271         put_task_struct(task);
3272 out:
3273         return result;
3274 }
3275
3276 /*
3277  * Find the first task with tgid >= tgid
3278  *
3279  */
3280 struct tgid_iter {
3281         unsigned int tgid;
3282         struct task_struct *task;
3283 };
3284 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3285 {
3286         struct pid *pid;
3287
3288         if (iter.task)
3289                 put_task_struct(iter.task);
3290         rcu_read_lock();
3291 retry:
3292         iter.task = NULL;
3293         pid = find_ge_pid(iter.tgid, ns);
3294         if (pid) {
3295                 iter.tgid = pid_nr_ns(pid, ns);
3296                 iter.task = pid_task(pid, PIDTYPE_PID);
3297                 /* What we to know is if the pid we have find is the
3298                  * pid of a thread_group_leader.  Testing for task
3299                  * being a thread_group_leader is the obvious thing
3300                  * todo but there is a window when it fails, due to
3301                  * the pid transfer logic in de_thread.
3302                  *
3303                  * So we perform the straight forward test of seeing
3304                  * if the pid we have found is the pid of a thread
3305                  * group leader, and don't worry if the task we have
3306                  * found doesn't happen to be a thread group leader.
3307                  * As we don't care in the case of readdir.
3308                  */
3309                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3310                         iter.tgid += 1;
3311                         goto retry;
3312                 }
3313                 get_task_struct(iter.task);
3314         }
3315         rcu_read_unlock();
3316         return iter;
3317 }
3318
3319 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3320
3321 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3322         struct tgid_iter iter)
3323 {
3324         char name[PROC_NUMBUF];
3325         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3326         return proc_fill_cache(filp, dirent, filldir, name, len,
3327                                 proc_pid_instantiate, iter.task, NULL);
3328 }
3329
3330 static int fake_filldir(void *buf, const char *name, int namelen,
3331                         loff_t offset, u64 ino, unsigned d_type)
3332 {
3333         return 0;
3334 }
3335
3336 /* for the /proc/ directory itself, after non-process stuff has been done */
3337 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3338 {
3339         unsigned int nr;
3340         struct task_struct *reaper;
3341         struct tgid_iter iter;
3342         struct pid_namespace *ns;
3343         filldir_t __filldir;
3344
3345         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3346                 goto out_no_task;
3347         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3348
3349         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3350         if (!reaper)
3351                 goto out_no_task;
3352
3353         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3354                 const struct pid_entry *p = &proc_base_stuff[nr];
3355                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3356                         goto out;
3357         }
3358
3359         ns = filp->f_dentry->d_sb->s_fs_info;
3360         iter.task = NULL;
3361         iter.tgid = filp->f_pos - TGID_OFFSET;
3362         for (iter = next_tgid(ns, iter);
3363              iter.task;
3364              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3365                 if (has_pid_permissions(ns, iter.task, 2))
3366                         __filldir = filldir;
3367                 else
3368                         __filldir = fake_filldir;
3369
3370                 filp->f_pos = iter.tgid + TGID_OFFSET;
3371                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3372                         put_task_struct(iter.task);
3373                         goto out;
3374                 }
3375         }
3376         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3377 out:
3378         put_task_struct(reaper);
3379 out_no_task:
3380         return 0;
3381 }
3382
3383 /*
3384  * Tasks
3385  */
3386 static const struct pid_entry tid_base_stuff[] = {
3387         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3388         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3389         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3390         REG("environ",   S_IRUSR, proc_environ_operations),
3391         INF("auxv",      S_IRUSR, proc_pid_auxv),
3392         ONE("status",    S_IRUGO, proc_pid_status),
3393         ONE("personality", S_IRUGO, proc_pid_personality),
3394         INF("limits",    S_IRUGO, proc_pid_limits),
3395 #ifdef CONFIG_SCHED_DEBUG
3396         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3397 #endif
3398         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3399 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3400         INF("syscall",   S_IRUGO, proc_pid_syscall),
3401 #endif
3402         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3403         ONE("stat",      S_IRUGO, proc_tid_stat),
3404         ONE("statm",     S_IRUGO, proc_pid_statm),
3405         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3406 #ifdef CONFIG_CHECKPOINT_RESTORE
3407         REG("children",  S_IRUGO, proc_tid_children_operations),
3408 #endif
3409 #ifdef CONFIG_NUMA
3410         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3411 #endif
3412         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3413         LNK("cwd",       proc_cwd_link),
3414         LNK("root",      proc_root_link),
3415         LNK("exe",       proc_exe_link),
3416         REG("mounts",    S_IRUGO, proc_mounts_operations),
3417         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3418 #ifdef CONFIG_PROC_PAGE_MONITOR
3419         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3420         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3421         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3422 #endif
3423 #ifdef CONFIG_SECURITY
3424         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3425 #endif
3426 #ifdef CONFIG_KALLSYMS
3427         INF("wchan",     S_IRUGO, proc_pid_wchan),
3428 #endif
3429 #ifdef CONFIG_STACKTRACE
3430         ONE("stack",      S_IRUGO, proc_pid_stack),
3431 #endif
3432 #ifdef CONFIG_SCHEDSTATS
3433         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3434 #endif
3435 #ifdef CONFIG_LATENCYTOP
3436         REG("latency",  S_IRUGO, proc_lstats_operations),
3437 #endif
3438 #ifdef CONFIG_PROC_PID_CPUSET
3439         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3440 #endif
3441 #ifdef CONFIG_CGROUPS
3442         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3443 #endif
3444         INF("oom_score", S_IRUGO, proc_oom_score),
3445         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3446         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3447 #ifdef CONFIG_AUDITSYSCALL
3448         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3449         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3450 #endif
3451 #ifdef CONFIG_FAULT_INJECTION
3452         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3453 #endif
3454 #ifdef CONFIG_TASK_IO_ACCOUNTING
3455         INF("io",       S_IRUSR, proc_tid_io_accounting),
3456 #endif
3457 #ifdef CONFIG_HARDWALL
3458         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3459 #endif
3460 #ifdef CONFIG_USER_NS
3461         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3462         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3463 #endif
3464 };
3465
3466 static int proc_tid_base_readdir(struct file * filp,
3467                              void * dirent, filldir_t filldir)
3468 {
3469         return proc_pident_readdir(filp,dirent,filldir,
3470                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3471 }
3472
3473 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3474         return proc_pident_lookup(dir, dentry,
3475                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3476 }
3477
3478 static const struct file_operations proc_tid_base_operations = {
3479         .read           = generic_read_dir,
3480         .readdir        = proc_tid_base_readdir,
3481         .llseek         = default_llseek,
3482 };
3483
3484 static const struct inode_operations proc_tid_base_inode_operations = {
3485         .lookup         = proc_tid_base_lookup,
3486         .getattr        = pid_getattr,
3487         .setattr        = proc_setattr,
3488 };
3489
3490 static struct dentry *proc_task_instantiate(struct inode *dir,
3491         struct dentry *dentry, struct task_struct *task, const void *ptr)
3492 {
3493         struct dentry *error = ERR_PTR(-ENOENT);
3494         struct inode *inode;
3495         inode = proc_pid_make_inode(dir->i_sb, task);
3496
3497         if (!inode)
3498                 goto out;
3499         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3500         inode->i_op = &proc_tid_base_inode_operations;
3501         inode->i_fop = &proc_tid_base_operations;
3502         inode->i_flags|=S_IMMUTABLE;
3503
3504         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3505                                                   ARRAY_SIZE(tid_base_stuff)));
3506
3507         d_set_d_op(dentry, &pid_dentry_operations);
3508
3509         d_add(dentry, inode);
3510         /* Close the race of the process dying before we return the dentry */
3511         if (pid_revalidate(dentry, 0))
3512                 error = NULL;
3513 out:
3514         return error;
3515 }
3516
3517 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3518 {
3519         struct dentry *result = ERR_PTR(-ENOENT);
3520         struct task_struct *task;
3521         struct task_struct *leader = get_proc_task(dir);
3522         unsigned tid;
3523         struct pid_namespace *ns;
3524
3525         if (!leader)
3526                 goto out_no_task;
3527
3528         tid = name_to_int(dentry);
3529         if (tid == ~0U)
3530                 goto out;
3531
3532         ns = dentry->d_sb->s_fs_info;
3533         rcu_read_lock();
3534         task = find_task_by_pid_ns(tid, ns);
3535         if (task)
3536                 get_task_struct(task);
3537         rcu_read_unlock();
3538         if (!task)
3539                 goto out;
3540         if (!same_thread_group(leader, task))
3541                 goto out_drop_task;
3542
3543         result = proc_task_instantiate(dir, dentry, task, NULL);
3544 out_drop_task:
3545         put_task_struct(task);
3546 out:
3547         put_task_struct(leader);
3548 out_no_task:
3549         return result;
3550 }
3551
3552 /*
3553  * Find the first tid of a thread group to return to user space.
3554  *
3555  * Usually this is just the thread group leader, but if the users
3556  * buffer was too small or there was a seek into the middle of the
3557  * directory we have more work todo.
3558  *
3559  * In the case of a short read we start with find_task_by_pid.
3560  *
3561  * In the case of a seek we start with the leader and walk nr
3562  * threads past it.
3563  */
3564 static struct task_struct *first_tid(struct task_struct *leader,
3565                 int tid, int nr, struct pid_namespace *ns)
3566 {
3567         struct task_struct *pos;
3568
3569         rcu_read_lock();
3570         /* Attempt to start with the pid of a thread */
3571         if (tid && (nr > 0)) {
3572                 pos = find_task_by_pid_ns(tid, ns);
3573                 if (pos && (pos->group_leader == leader))
3574                         goto found;
3575         }
3576
3577         /* If nr exceeds the number of threads there is nothing todo */
3578         pos = NULL;
3579         if (nr && nr >= get_nr_threads(leader))
3580                 goto out;
3581
3582         /* If we haven't found our starting place yet start
3583          * with the leader and walk nr threads forward.
3584          */
3585         for (pos = leader; nr > 0; --nr) {
3586                 pos = next_thread(pos);
3587                 if (pos == leader) {
3588                         pos = NULL;
3589                         goto out;
3590                 }
3591         }
3592 found:
3593         get_task_struct(pos);
3594 out:
3595         rcu_read_unlock();
3596         return pos;
3597 }
3598
3599 /*
3600  * Find the next thread in the thread list.
3601  * Return NULL if there is an error or no next thread.
3602  *
3603  * The reference to the input task_struct is released.
3604  */
3605 static struct task_struct *next_tid(struct task_struct *start)
3606 {
3607         struct task_struct *pos = NULL;
3608         rcu_read_lock();
3609         if (pid_alive(start)) {
3610                 pos = next_thread(start);
3611                 if (thread_group_leader(pos))
3612                         pos = NULL;
3613                 else
3614                         get_task_struct(pos);
3615         }
3616         rcu_read_unlock();
3617         put_task_struct(start);
3618         return pos;
3619 }
3620
3621 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3622         struct task_struct *task, int tid)
3623 {
3624         char name[PROC_NUMBUF];
3625         int len = snprintf(name, sizeof(name), "%d", tid);
3626         return proc_fill_cache(filp, dirent, filldir, name, len,
3627                                 proc_task_instantiate, task, NULL);
3628 }
3629
3630 /* for the /proc/TGID/task/ directories */
3631 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3632 {
3633         struct dentry *dentry = filp->f_path.dentry;
3634         struct inode *inode = dentry->d_inode;
3635         struct task_struct *leader = NULL;
3636         struct task_struct *task;
3637         int retval = -ENOENT;
3638         ino_t ino;
3639         int tid;
3640         struct pid_namespace *ns;
3641
3642         task = get_proc_task(inode);
3643         if (!task)
3644                 goto out_no_task;
3645         rcu_read_lock();
3646         if (pid_alive(task)) {
3647                 leader = task->group_leader;
3648                 get_task_struct(leader);
3649         }
3650         rcu_read_unlock();
3651         put_task_struct(task);
3652         if (!leader)
3653                 goto out_no_task;
3654         retval = 0;
3655
3656         switch ((unsigned long)filp->f_pos) {
3657         case 0:
3658                 ino = inode->i_ino;
3659                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3660                         goto out;
3661                 filp->f_pos++;
3662                 /* fall through */
3663         case 1:
3664                 ino = parent_ino(dentry);
3665                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3666                         goto out;
3667                 filp->f_pos++;
3668                 /* fall through */
3669         }
3670
3671         /* f_version caches the tgid value that the last readdir call couldn't
3672          * return. lseek aka telldir automagically resets f_version to 0.
3673          */
3674         ns = filp->f_dentry->d_sb->s_fs_info;
3675         tid = (int)filp->f_version;
3676         filp->f_version = 0;
3677         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3678              task;
3679              task = next_tid(task), filp->f_pos++) {
3680                 tid = task_pid_nr_ns(task, ns);
3681                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3682                         /* returning this tgid failed, save it as the first
3683                          * pid for the next readir call */
3684                         filp->f_version = (u64)tid;
3685                         put_task_struct(task);
3686                         break;
3687                 }
3688         }
3689 out:
3690         put_task_struct(leader);
3691 out_no_task:
3692         return retval;
3693 }
3694
3695 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3696 {
3697         struct inode *inode = dentry->d_inode;
3698         struct task_struct *p = get_proc_task(inode);
3699         generic_fillattr(inode, stat);
3700
3701         if (p) {
3702                 stat->nlink += get_nr_threads(p);
3703                 put_task_struct(p);
3704         }
3705
3706         return 0;
3707 }
3708
3709 static const struct inode_operations proc_task_inode_operations = {
3710         .lookup         = proc_task_lookup,
3711         .getattr        = proc_task_getattr,
3712         .setattr        = proc_setattr,
3713         .permission     = proc_pid_permission,
3714 };
3715
3716 static const struct file_operations proc_task_operations = {
3717         .read           = generic_read_dir,
3718         .readdir        = proc_task_readdir,
3719         .llseek         = default_llseek,
3720 };