sched: Clean up parameter passing of proc_sched_autogroup_set_nice()
[linux-2.6.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/fs_struct.h>
85 #include <linux/slab.h>
86 #include <linux/flex_array.h>
87 #ifdef CONFIG_HARDWALL
88 #include <asm/hardwall.h>
89 #endif
90 #include <trace/events/oom.h>
91 #include "internal.h"
92
93 /* NOTE:
94  *      Implementing inode permission operations in /proc is almost
95  *      certainly an error.  Permission checks need to happen during
96  *      each system call not at open time.  The reason is that most of
97  *      what we wish to check for permissions in /proc varies at runtime.
98  *
99  *      The classic example of a problem is opening file descriptors
100  *      in /proc for a task before it execs a suid executable.
101  */
102
103 struct pid_entry {
104         char *name;
105         int len;
106         umode_t mode;
107         const struct inode_operations *iop;
108         const struct file_operations *fop;
109         union proc_op op;
110 };
111
112 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
113         .name = (NAME),                                 \
114         .len  = sizeof(NAME) - 1,                       \
115         .mode = MODE,                                   \
116         .iop  = IOP,                                    \
117         .fop  = FOP,                                    \
118         .op   = OP,                                     \
119 }
120
121 #define DIR(NAME, MODE, iops, fops)     \
122         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
123 #define LNK(NAME, get_link)                                     \
124         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
125                 &proc_pid_link_inode_operations, NULL,          \
126                 { .proc_get_link = get_link } )
127 #define REG(NAME, MODE, fops)                           \
128         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
129 #define INF(NAME, MODE, read)                           \
130         NOD(NAME, (S_IFREG|(MODE)),                     \
131                 NULL, &proc_info_file_operations,       \
132                 { .proc_read = read } )
133 #define ONE(NAME, MODE, show)                           \
134         NOD(NAME, (S_IFREG|(MODE)),                     \
135                 NULL, &proc_single_file_operations,     \
136                 { .proc_show = show } )
137
138 static int proc_fd_permission(struct inode *inode, int mask);
139
140 /*
141  * Count the number of hardlinks for the pid_entry table, excluding the .
142  * and .. links.
143  */
144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
145         unsigned int n)
146 {
147         unsigned int i;
148         unsigned int count;
149
150         count = 0;
151         for (i = 0; i < n; ++i) {
152                 if (S_ISDIR(entries[i].mode))
153                         ++count;
154         }
155
156         return count;
157 }
158
159 static int get_task_root(struct task_struct *task, struct path *root)
160 {
161         int result = -ENOENT;
162
163         task_lock(task);
164         if (task->fs) {
165                 get_fs_root(task->fs, root);
166                 result = 0;
167         }
168         task_unlock(task);
169         return result;
170 }
171
172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
173 {
174         struct task_struct *task = get_proc_task(dentry->d_inode);
175         int result = -ENOENT;
176
177         if (task) {
178                 task_lock(task);
179                 if (task->fs) {
180                         get_fs_pwd(task->fs, path);
181                         result = 0;
182                 }
183                 task_unlock(task);
184                 put_task_struct(task);
185         }
186         return result;
187 }
188
189 static int proc_root_link(struct dentry *dentry, struct path *path)
190 {
191         struct task_struct *task = get_proc_task(dentry->d_inode);
192         int result = -ENOENT;
193
194         if (task) {
195                 result = get_task_root(task, path);
196                 put_task_struct(task);
197         }
198         return result;
199 }
200
201 struct mm_struct *mm_for_maps(struct task_struct *task)
202 {
203         return mm_access(task, PTRACE_MODE_READ);
204 }
205
206 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
207 {
208         int res = 0;
209         unsigned int len;
210         struct mm_struct *mm = get_task_mm(task);
211         if (!mm)
212                 goto out;
213         if (!mm->arg_end)
214                 goto out_mm;    /* Shh! No looking before we're done */
215
216         len = mm->arg_end - mm->arg_start;
217  
218         if (len > PAGE_SIZE)
219                 len = PAGE_SIZE;
220  
221         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
222
223         // If the nul at the end of args has been overwritten, then
224         // assume application is using setproctitle(3).
225         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
226                 len = strnlen(buffer, res);
227                 if (len < res) {
228                     res = len;
229                 } else {
230                         len = mm->env_end - mm->env_start;
231                         if (len > PAGE_SIZE - res)
232                                 len = PAGE_SIZE - res;
233                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
234                         res = strnlen(buffer, res);
235                 }
236         }
237 out_mm:
238         mmput(mm);
239 out:
240         return res;
241 }
242
243 static int proc_pid_auxv(struct task_struct *task, char *buffer)
244 {
245         struct mm_struct *mm = mm_for_maps(task);
246         int res = PTR_ERR(mm);
247         if (mm && !IS_ERR(mm)) {
248                 unsigned int nwords = 0;
249                 do {
250                         nwords += 2;
251                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
252                 res = nwords * sizeof(mm->saved_auxv[0]);
253                 if (res > PAGE_SIZE)
254                         res = PAGE_SIZE;
255                 memcpy(buffer, mm->saved_auxv, res);
256                 mmput(mm);
257         }
258         return res;
259 }
260
261
262 #ifdef CONFIG_KALLSYMS
263 /*
264  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
265  * Returns the resolved symbol.  If that fails, simply return the address.
266  */
267 static int proc_pid_wchan(struct task_struct *task, char *buffer)
268 {
269         unsigned long wchan;
270         char symname[KSYM_NAME_LEN];
271
272         wchan = get_wchan(task);
273
274         if (lookup_symbol_name(wchan, symname) < 0)
275                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
276                         return 0;
277                 else
278                         return sprintf(buffer, "%lu", wchan);
279         else
280                 return sprintf(buffer, "%s", symname);
281 }
282 #endif /* CONFIG_KALLSYMS */
283
284 static int lock_trace(struct task_struct *task)
285 {
286         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
287         if (err)
288                 return err;
289         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
290                 mutex_unlock(&task->signal->cred_guard_mutex);
291                 return -EPERM;
292         }
293         return 0;
294 }
295
296 static void unlock_trace(struct task_struct *task)
297 {
298         mutex_unlock(&task->signal->cred_guard_mutex);
299 }
300
301 #ifdef CONFIG_STACKTRACE
302
303 #define MAX_STACK_TRACE_DEPTH   64
304
305 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
306                           struct pid *pid, struct task_struct *task)
307 {
308         struct stack_trace trace;
309         unsigned long *entries;
310         int err;
311         int i;
312
313         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
314         if (!entries)
315                 return -ENOMEM;
316
317         trace.nr_entries        = 0;
318         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
319         trace.entries           = entries;
320         trace.skip              = 0;
321
322         err = lock_trace(task);
323         if (!err) {
324                 save_stack_trace_tsk(task, &trace);
325
326                 for (i = 0; i < trace.nr_entries; i++) {
327                         seq_printf(m, "[<%pK>] %pS\n",
328                                    (void *)entries[i], (void *)entries[i]);
329                 }
330                 unlock_trace(task);
331         }
332         kfree(entries);
333
334         return err;
335 }
336 #endif
337
338 #ifdef CONFIG_SCHEDSTATS
339 /*
340  * Provides /proc/PID/schedstat
341  */
342 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
343 {
344         return sprintf(buffer, "%llu %llu %lu\n",
345                         (unsigned long long)task->se.sum_exec_runtime,
346                         (unsigned long long)task->sched_info.run_delay,
347                         task->sched_info.pcount);
348 }
349 #endif
350
351 #ifdef CONFIG_LATENCYTOP
352 static int lstats_show_proc(struct seq_file *m, void *v)
353 {
354         int i;
355         struct inode *inode = m->private;
356         struct task_struct *task = get_proc_task(inode);
357
358         if (!task)
359                 return -ESRCH;
360         seq_puts(m, "Latency Top version : v0.1\n");
361         for (i = 0; i < 32; i++) {
362                 struct latency_record *lr = &task->latency_record[i];
363                 if (lr->backtrace[0]) {
364                         int q;
365                         seq_printf(m, "%i %li %li",
366                                    lr->count, lr->time, lr->max);
367                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
368                                 unsigned long bt = lr->backtrace[q];
369                                 if (!bt)
370                                         break;
371                                 if (bt == ULONG_MAX)
372                                         break;
373                                 seq_printf(m, " %ps", (void *)bt);
374                         }
375                         seq_putc(m, '\n');
376                 }
377
378         }
379         put_task_struct(task);
380         return 0;
381 }
382
383 static int lstats_open(struct inode *inode, struct file *file)
384 {
385         return single_open(file, lstats_show_proc, inode);
386 }
387
388 static ssize_t lstats_write(struct file *file, const char __user *buf,
389                             size_t count, loff_t *offs)
390 {
391         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
392
393         if (!task)
394                 return -ESRCH;
395         clear_all_latency_tracing(task);
396         put_task_struct(task);
397
398         return count;
399 }
400
401 static const struct file_operations proc_lstats_operations = {
402         .open           = lstats_open,
403         .read           = seq_read,
404         .write          = lstats_write,
405         .llseek         = seq_lseek,
406         .release        = single_release,
407 };
408
409 #endif
410
411 static int proc_oom_score(struct task_struct *task, char *buffer)
412 {
413         unsigned long points = 0;
414
415         read_lock(&tasklist_lock);
416         if (pid_alive(task))
417                 points = oom_badness(task, NULL, NULL,
418                                         totalram_pages + total_swap_pages);
419         read_unlock(&tasklist_lock);
420         return sprintf(buffer, "%lu\n", points);
421 }
422
423 struct limit_names {
424         char *name;
425         char *unit;
426 };
427
428 static const struct limit_names lnames[RLIM_NLIMITS] = {
429         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
430         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
431         [RLIMIT_DATA] = {"Max data size", "bytes"},
432         [RLIMIT_STACK] = {"Max stack size", "bytes"},
433         [RLIMIT_CORE] = {"Max core file size", "bytes"},
434         [RLIMIT_RSS] = {"Max resident set", "bytes"},
435         [RLIMIT_NPROC] = {"Max processes", "processes"},
436         [RLIMIT_NOFILE] = {"Max open files", "files"},
437         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
438         [RLIMIT_AS] = {"Max address space", "bytes"},
439         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
440         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
441         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
442         [RLIMIT_NICE] = {"Max nice priority", NULL},
443         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
444         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
445 };
446
447 /* Display limits for a process */
448 static int proc_pid_limits(struct task_struct *task, char *buffer)
449 {
450         unsigned int i;
451         int count = 0;
452         unsigned long flags;
453         char *bufptr = buffer;
454
455         struct rlimit rlim[RLIM_NLIMITS];
456
457         if (!lock_task_sighand(task, &flags))
458                 return 0;
459         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
460         unlock_task_sighand(task, &flags);
461
462         /*
463          * print the file header
464          */
465         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
466                         "Limit", "Soft Limit", "Hard Limit", "Units");
467
468         for (i = 0; i < RLIM_NLIMITS; i++) {
469                 if (rlim[i].rlim_cur == RLIM_INFINITY)
470                         count += sprintf(&bufptr[count], "%-25s %-20s ",
471                                          lnames[i].name, "unlimited");
472                 else
473                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
474                                          lnames[i].name, rlim[i].rlim_cur);
475
476                 if (rlim[i].rlim_max == RLIM_INFINITY)
477                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
478                 else
479                         count += sprintf(&bufptr[count], "%-20lu ",
480                                          rlim[i].rlim_max);
481
482                 if (lnames[i].unit)
483                         count += sprintf(&bufptr[count], "%-10s\n",
484                                          lnames[i].unit);
485                 else
486                         count += sprintf(&bufptr[count], "\n");
487         }
488
489         return count;
490 }
491
492 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
493 static int proc_pid_syscall(struct task_struct *task, char *buffer)
494 {
495         long nr;
496         unsigned long args[6], sp, pc;
497         int res = lock_trace(task);
498         if (res)
499                 return res;
500
501         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
502                 res = sprintf(buffer, "running\n");
503         else if (nr < 0)
504                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
505         else
506                 res = sprintf(buffer,
507                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
508                        nr,
509                        args[0], args[1], args[2], args[3], args[4], args[5],
510                        sp, pc);
511         unlock_trace(task);
512         return res;
513 }
514 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
515
516 /************************************************************************/
517 /*                       Here the fs part begins                        */
518 /************************************************************************/
519
520 /* permission checks */
521 static int proc_fd_access_allowed(struct inode *inode)
522 {
523         struct task_struct *task;
524         int allowed = 0;
525         /* Allow access to a task's file descriptors if it is us or we
526          * may use ptrace attach to the process and find out that
527          * information.
528          */
529         task = get_proc_task(inode);
530         if (task) {
531                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
532                 put_task_struct(task);
533         }
534         return allowed;
535 }
536
537 int proc_setattr(struct dentry *dentry, struct iattr *attr)
538 {
539         int error;
540         struct inode *inode = dentry->d_inode;
541
542         if (attr->ia_valid & ATTR_MODE)
543                 return -EPERM;
544
545         error = inode_change_ok(inode, attr);
546         if (error)
547                 return error;
548
549         if ((attr->ia_valid & ATTR_SIZE) &&
550             attr->ia_size != i_size_read(inode)) {
551                 error = vmtruncate(inode, attr->ia_size);
552                 if (error)
553                         return error;
554         }
555
556         setattr_copy(inode, attr);
557         mark_inode_dirty(inode);
558         return 0;
559 }
560
561 /*
562  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
563  * or euid/egid (for hide_pid_min=2)?
564  */
565 static bool has_pid_permissions(struct pid_namespace *pid,
566                                  struct task_struct *task,
567                                  int hide_pid_min)
568 {
569         if (pid->hide_pid < hide_pid_min)
570                 return true;
571         if (in_group_p(pid->pid_gid))
572                 return true;
573         return ptrace_may_access(task, PTRACE_MODE_READ);
574 }
575
576
577 static int proc_pid_permission(struct inode *inode, int mask)
578 {
579         struct pid_namespace *pid = inode->i_sb->s_fs_info;
580         struct task_struct *task;
581         bool has_perms;
582
583         task = get_proc_task(inode);
584         if (!task)
585                 return -ESRCH;
586         has_perms = has_pid_permissions(pid, task, 1);
587         put_task_struct(task);
588
589         if (!has_perms) {
590                 if (pid->hide_pid == 2) {
591                         /*
592                          * Let's make getdents(), stat(), and open()
593                          * consistent with each other.  If a process
594                          * may not stat() a file, it shouldn't be seen
595                          * in procfs at all.
596                          */
597                         return -ENOENT;
598                 }
599
600                 return -EPERM;
601         }
602         return generic_permission(inode, mask);
603 }
604
605
606
607 static const struct inode_operations proc_def_inode_operations = {
608         .setattr        = proc_setattr,
609 };
610
611 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
612
613 static ssize_t proc_info_read(struct file * file, char __user * buf,
614                           size_t count, loff_t *ppos)
615 {
616         struct inode * inode = file->f_path.dentry->d_inode;
617         unsigned long page;
618         ssize_t length;
619         struct task_struct *task = get_proc_task(inode);
620
621         length = -ESRCH;
622         if (!task)
623                 goto out_no_task;
624
625         if (count > PROC_BLOCK_SIZE)
626                 count = PROC_BLOCK_SIZE;
627
628         length = -ENOMEM;
629         if (!(page = __get_free_page(GFP_TEMPORARY)))
630                 goto out;
631
632         length = PROC_I(inode)->op.proc_read(task, (char*)page);
633
634         if (length >= 0)
635                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
636         free_page(page);
637 out:
638         put_task_struct(task);
639 out_no_task:
640         return length;
641 }
642
643 static const struct file_operations proc_info_file_operations = {
644         .read           = proc_info_read,
645         .llseek         = generic_file_llseek,
646 };
647
648 static int proc_single_show(struct seq_file *m, void *v)
649 {
650         struct inode *inode = m->private;
651         struct pid_namespace *ns;
652         struct pid *pid;
653         struct task_struct *task;
654         int ret;
655
656         ns = inode->i_sb->s_fs_info;
657         pid = proc_pid(inode);
658         task = get_pid_task(pid, PIDTYPE_PID);
659         if (!task)
660                 return -ESRCH;
661
662         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
663
664         put_task_struct(task);
665         return ret;
666 }
667
668 static int proc_single_open(struct inode *inode, struct file *filp)
669 {
670         return single_open(filp, proc_single_show, inode);
671 }
672
673 static const struct file_operations proc_single_file_operations = {
674         .open           = proc_single_open,
675         .read           = seq_read,
676         .llseek         = seq_lseek,
677         .release        = single_release,
678 };
679
680 static int mem_open(struct inode* inode, struct file* file)
681 {
682         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
683         struct mm_struct *mm;
684
685         if (!task)
686                 return -ESRCH;
687
688         mm = mm_access(task, PTRACE_MODE_ATTACH);
689         put_task_struct(task);
690
691         if (IS_ERR(mm))
692                 return PTR_ERR(mm);
693
694         if (mm) {
695                 /* ensure this mm_struct can't be freed */
696                 atomic_inc(&mm->mm_count);
697                 /* but do not pin its memory */
698                 mmput(mm);
699         }
700
701         /* OK to pass negative loff_t, we can catch out-of-range */
702         file->f_mode |= FMODE_UNSIGNED_OFFSET;
703         file->private_data = mm;
704
705         return 0;
706 }
707
708 static ssize_t mem_rw(struct file *file, char __user *buf,
709                         size_t count, loff_t *ppos, int write)
710 {
711         struct mm_struct *mm = file->private_data;
712         unsigned long addr = *ppos;
713         ssize_t copied;
714         char *page;
715
716         if (!mm)
717                 return 0;
718
719         page = (char *)__get_free_page(GFP_TEMPORARY);
720         if (!page)
721                 return -ENOMEM;
722
723         copied = 0;
724         if (!atomic_inc_not_zero(&mm->mm_users))
725                 goto free;
726
727         while (count > 0) {
728                 int this_len = min_t(int, count, PAGE_SIZE);
729
730                 if (write && copy_from_user(page, buf, this_len)) {
731                         copied = -EFAULT;
732                         break;
733                 }
734
735                 this_len = access_remote_vm(mm, addr, page, this_len, write);
736                 if (!this_len) {
737                         if (!copied)
738                                 copied = -EIO;
739                         break;
740                 }
741
742                 if (!write && copy_to_user(buf, page, this_len)) {
743                         copied = -EFAULT;
744                         break;
745                 }
746
747                 buf += this_len;
748                 addr += this_len;
749                 copied += this_len;
750                 count -= this_len;
751         }
752         *ppos = addr;
753
754         mmput(mm);
755 free:
756         free_page((unsigned long) page);
757         return copied;
758 }
759
760 static ssize_t mem_read(struct file *file, char __user *buf,
761                         size_t count, loff_t *ppos)
762 {
763         return mem_rw(file, buf, count, ppos, 0);
764 }
765
766 static ssize_t mem_write(struct file *file, const char __user *buf,
767                          size_t count, loff_t *ppos)
768 {
769         return mem_rw(file, (char __user*)buf, count, ppos, 1);
770 }
771
772 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
773 {
774         switch (orig) {
775         case 0:
776                 file->f_pos = offset;
777                 break;
778         case 1:
779                 file->f_pos += offset;
780                 break;
781         default:
782                 return -EINVAL;
783         }
784         force_successful_syscall_return();
785         return file->f_pos;
786 }
787
788 static int mem_release(struct inode *inode, struct file *file)
789 {
790         struct mm_struct *mm = file->private_data;
791         if (mm)
792                 mmdrop(mm);
793         return 0;
794 }
795
796 static const struct file_operations proc_mem_operations = {
797         .llseek         = mem_lseek,
798         .read           = mem_read,
799         .write          = mem_write,
800         .open           = mem_open,
801         .release        = mem_release,
802 };
803
804 static ssize_t environ_read(struct file *file, char __user *buf,
805                         size_t count, loff_t *ppos)
806 {
807         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
808         char *page;
809         unsigned long src = *ppos;
810         int ret = -ESRCH;
811         struct mm_struct *mm;
812
813         if (!task)
814                 goto out_no_task;
815
816         ret = -ENOMEM;
817         page = (char *)__get_free_page(GFP_TEMPORARY);
818         if (!page)
819                 goto out;
820
821
822         mm = mm_for_maps(task);
823         ret = PTR_ERR(mm);
824         if (!mm || IS_ERR(mm))
825                 goto out_free;
826
827         ret = 0;
828         while (count > 0) {
829                 int this_len, retval, max_len;
830
831                 this_len = mm->env_end - (mm->env_start + src);
832
833                 if (this_len <= 0)
834                         break;
835
836                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
837                 this_len = (this_len > max_len) ? max_len : this_len;
838
839                 retval = access_process_vm(task, (mm->env_start + src),
840                         page, this_len, 0);
841
842                 if (retval <= 0) {
843                         ret = retval;
844                         break;
845                 }
846
847                 if (copy_to_user(buf, page, retval)) {
848                         ret = -EFAULT;
849                         break;
850                 }
851
852                 ret += retval;
853                 src += retval;
854                 buf += retval;
855                 count -= retval;
856         }
857         *ppos = src;
858
859         mmput(mm);
860 out_free:
861         free_page((unsigned long) page);
862 out:
863         put_task_struct(task);
864 out_no_task:
865         return ret;
866 }
867
868 static const struct file_operations proc_environ_operations = {
869         .read           = environ_read,
870         .llseek         = generic_file_llseek,
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 = 0;
1565         stat->gid = 0;
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, struct nameidata *nd)
1605 {
1606         struct inode *inode;
1607         struct task_struct *task;
1608         const struct cred *cred;
1609
1610         if (nd && nd->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 = 0;
1626                         inode->i_gid = 0;
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 (FD_ISSET(fd, fdt->close_on_exec))
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, struct nameidata *nd)
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 (nd && nd->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                         rcu_read_lock();
1803                         if (fcheck_files(files, fd)) {
1804                                 rcu_read_unlock();
1805                                 put_files_struct(files);
1806                                 if (task_dumpable(task)) {
1807                                         rcu_read_lock();
1808                                         cred = __task_cred(task);
1809                                         inode->i_uid = cred->euid;
1810                                         inode->i_gid = cred->egid;
1811                                         rcu_read_unlock();
1812                                 } else {
1813                                         inode->i_uid = 0;
1814                                         inode->i_gid = 0;
1815                                 }
1816                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1817                                 security_task_to_inode(task, inode);
1818                                 put_task_struct(task);
1819                                 return 1;
1820                         }
1821                         rcu_read_unlock();
1822                         put_files_struct(files);
1823                 }
1824                 put_task_struct(task);
1825         }
1826         d_drop(dentry);
1827         return 0;
1828 }
1829
1830 static const struct dentry_operations tid_fd_dentry_operations =
1831 {
1832         .d_revalidate   = tid_fd_revalidate,
1833         .d_delete       = pid_delete_dentry,
1834 };
1835
1836 static struct dentry *proc_fd_instantiate(struct inode *dir,
1837         struct dentry *dentry, struct task_struct *task, const void *ptr)
1838 {
1839         unsigned fd = *(const unsigned *)ptr;
1840         struct file *file;
1841         struct files_struct *files;
1842         struct inode *inode;
1843         struct proc_inode *ei;
1844         struct dentry *error = ERR_PTR(-ENOENT);
1845
1846         inode = proc_pid_make_inode(dir->i_sb, task);
1847         if (!inode)
1848                 goto out;
1849         ei = PROC_I(inode);
1850         ei->fd = fd;
1851         files = get_files_struct(task);
1852         if (!files)
1853                 goto out_iput;
1854         inode->i_mode = S_IFLNK;
1855
1856         /*
1857          * We are not taking a ref to the file structure, so we must
1858          * hold ->file_lock.
1859          */
1860         spin_lock(&files->file_lock);
1861         file = fcheck_files(files, fd);
1862         if (!file)
1863                 goto out_unlock;
1864         if (file->f_mode & FMODE_READ)
1865                 inode->i_mode |= S_IRUSR | S_IXUSR;
1866         if (file->f_mode & FMODE_WRITE)
1867                 inode->i_mode |= S_IWUSR | S_IXUSR;
1868         spin_unlock(&files->file_lock);
1869         put_files_struct(files);
1870
1871         inode->i_op = &proc_pid_link_inode_operations;
1872         inode->i_size = 64;
1873         ei->op.proc_get_link = proc_fd_link;
1874         d_set_d_op(dentry, &tid_fd_dentry_operations);
1875         d_add(dentry, inode);
1876         /* Close the race of the process dying before we return the dentry */
1877         if (tid_fd_revalidate(dentry, NULL))
1878                 error = NULL;
1879
1880  out:
1881         return error;
1882 out_unlock:
1883         spin_unlock(&files->file_lock);
1884         put_files_struct(files);
1885 out_iput:
1886         iput(inode);
1887         goto out;
1888 }
1889
1890 static struct dentry *proc_lookupfd_common(struct inode *dir,
1891                                            struct dentry *dentry,
1892                                            instantiate_t instantiate)
1893 {
1894         struct task_struct *task = get_proc_task(dir);
1895         unsigned fd = name_to_int(dentry);
1896         struct dentry *result = ERR_PTR(-ENOENT);
1897
1898         if (!task)
1899                 goto out_no_task;
1900         if (fd == ~0U)
1901                 goto out;
1902
1903         result = instantiate(dir, dentry, task, &fd);
1904 out:
1905         put_task_struct(task);
1906 out_no_task:
1907         return result;
1908 }
1909
1910 static int proc_readfd_common(struct file * filp, void * dirent,
1911                               filldir_t filldir, instantiate_t instantiate)
1912 {
1913         struct dentry *dentry = filp->f_path.dentry;
1914         struct inode *inode = dentry->d_inode;
1915         struct task_struct *p = get_proc_task(inode);
1916         unsigned int fd, ino;
1917         int retval;
1918         struct files_struct * files;
1919
1920         retval = -ENOENT;
1921         if (!p)
1922                 goto out_no_task;
1923         retval = 0;
1924
1925         fd = filp->f_pos;
1926         switch (fd) {
1927                 case 0:
1928                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1929                                 goto out;
1930                         filp->f_pos++;
1931                 case 1:
1932                         ino = parent_ino(dentry);
1933                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1934                                 goto out;
1935                         filp->f_pos++;
1936                 default:
1937                         files = get_files_struct(p);
1938                         if (!files)
1939                                 goto out;
1940                         rcu_read_lock();
1941                         for (fd = filp->f_pos-2;
1942                              fd < files_fdtable(files)->max_fds;
1943                              fd++, filp->f_pos++) {
1944                                 char name[PROC_NUMBUF];
1945                                 int len;
1946
1947                                 if (!fcheck_files(files, fd))
1948                                         continue;
1949                                 rcu_read_unlock();
1950
1951                                 len = snprintf(name, sizeof(name), "%d", fd);
1952                                 if (proc_fill_cache(filp, dirent, filldir,
1953                                                     name, len, instantiate,
1954                                                     p, &fd) < 0) {
1955                                         rcu_read_lock();
1956                                         break;
1957                                 }
1958                                 rcu_read_lock();
1959                         }
1960                         rcu_read_unlock();
1961                         put_files_struct(files);
1962         }
1963 out:
1964         put_task_struct(p);
1965 out_no_task:
1966         return retval;
1967 }
1968
1969 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1970                                     struct nameidata *nd)
1971 {
1972         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1973 }
1974
1975 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1976 {
1977         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1978 }
1979
1980 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1981                                       size_t len, loff_t *ppos)
1982 {
1983         char tmp[PROC_FDINFO_MAX];
1984         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1985         if (!err)
1986                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1987         return err;
1988 }
1989
1990 static const struct file_operations proc_fdinfo_file_operations = {
1991         .open           = nonseekable_open,
1992         .read           = proc_fdinfo_read,
1993         .llseek         = no_llseek,
1994 };
1995
1996 static const struct file_operations proc_fd_operations = {
1997         .read           = generic_read_dir,
1998         .readdir        = proc_readfd,
1999         .llseek         = default_llseek,
2000 };
2001
2002 #ifdef CONFIG_CHECKPOINT_RESTORE
2003
2004 /*
2005  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2006  * which represent vma start and end addresses.
2007  */
2008 static int dname_to_vma_addr(struct dentry *dentry,
2009                              unsigned long *start, unsigned long *end)
2010 {
2011         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2012                 return -EINVAL;
2013
2014         return 0;
2015 }
2016
2017 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2018 {
2019         unsigned long vm_start, vm_end;
2020         bool exact_vma_exists = false;
2021         struct mm_struct *mm = NULL;
2022         struct task_struct *task;
2023         const struct cred *cred;
2024         struct inode *inode;
2025         int status = 0;
2026
2027         if (nd && nd->flags & LOOKUP_RCU)
2028                 return -ECHILD;
2029
2030         if (!capable(CAP_SYS_ADMIN)) {
2031                 status = -EACCES;
2032                 goto out_notask;
2033         }
2034
2035         inode = dentry->d_inode;
2036         task = get_proc_task(inode);
2037         if (!task)
2038                 goto out_notask;
2039
2040         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2041                 goto out;
2042
2043         mm = get_task_mm(task);
2044         if (!mm)
2045                 goto out;
2046
2047         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2048                 down_read(&mm->mmap_sem);
2049                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2050                 up_read(&mm->mmap_sem);
2051         }
2052
2053         mmput(mm);
2054
2055         if (exact_vma_exists) {
2056                 if (task_dumpable(task)) {
2057                         rcu_read_lock();
2058                         cred = __task_cred(task);
2059                         inode->i_uid = cred->euid;
2060                         inode->i_gid = cred->egid;
2061                         rcu_read_unlock();
2062                 } else {
2063                         inode->i_uid = 0;
2064                         inode->i_gid = 0;
2065                 }
2066                 security_task_to_inode(task, inode);
2067                 status = 1;
2068         }
2069
2070 out:
2071         put_task_struct(task);
2072
2073 out_notask:
2074         if (status <= 0)
2075                 d_drop(dentry);
2076
2077         return status;
2078 }
2079
2080 static const struct dentry_operations tid_map_files_dentry_operations = {
2081         .d_revalidate   = map_files_d_revalidate,
2082         .d_delete       = pid_delete_dentry,
2083 };
2084
2085 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2086 {
2087         unsigned long vm_start, vm_end;
2088         struct vm_area_struct *vma;
2089         struct task_struct *task;
2090         struct mm_struct *mm;
2091         int rc;
2092
2093         rc = -ENOENT;
2094         task = get_proc_task(dentry->d_inode);
2095         if (!task)
2096                 goto out;
2097
2098         mm = get_task_mm(task);
2099         put_task_struct(task);
2100         if (!mm)
2101                 goto out;
2102
2103         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2104         if (rc)
2105                 goto out_mmput;
2106
2107         down_read(&mm->mmap_sem);
2108         vma = find_exact_vma(mm, vm_start, vm_end);
2109         if (vma && vma->vm_file) {
2110                 *path = vma->vm_file->f_path;
2111                 path_get(path);
2112                 rc = 0;
2113         }
2114         up_read(&mm->mmap_sem);
2115
2116 out_mmput:
2117         mmput(mm);
2118 out:
2119         return rc;
2120 }
2121
2122 struct map_files_info {
2123         struct file     *file;
2124         unsigned long   len;
2125         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2126 };
2127
2128 static struct dentry *
2129 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2130                            struct task_struct *task, const void *ptr)
2131 {
2132         const struct file *file = ptr;
2133         struct proc_inode *ei;
2134         struct inode *inode;
2135
2136         if (!file)
2137                 return ERR_PTR(-ENOENT);
2138
2139         inode = proc_pid_make_inode(dir->i_sb, task);
2140         if (!inode)
2141                 return ERR_PTR(-ENOENT);
2142
2143         ei = PROC_I(inode);
2144         ei->op.proc_get_link = proc_map_files_get_link;
2145
2146         inode->i_op = &proc_pid_link_inode_operations;
2147         inode->i_size = 64;
2148         inode->i_mode = S_IFLNK;
2149
2150         if (file->f_mode & FMODE_READ)
2151                 inode->i_mode |= S_IRUSR;
2152         if (file->f_mode & FMODE_WRITE)
2153                 inode->i_mode |= S_IWUSR;
2154
2155         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2156         d_add(dentry, inode);
2157
2158         return NULL;
2159 }
2160
2161 static struct dentry *proc_map_files_lookup(struct inode *dir,
2162                 struct dentry *dentry, struct nameidata *nd)
2163 {
2164         unsigned long vm_start, vm_end;
2165         struct vm_area_struct *vma;
2166         struct task_struct *task;
2167         struct dentry *result;
2168         struct mm_struct *mm;
2169
2170         result = ERR_PTR(-EACCES);
2171         if (!capable(CAP_SYS_ADMIN))
2172                 goto out;
2173
2174         result = ERR_PTR(-ENOENT);
2175         task = get_proc_task(dir);
2176         if (!task)
2177                 goto out;
2178
2179         result = ERR_PTR(-EACCES);
2180         if (lock_trace(task))
2181                 goto out_put_task;
2182
2183         result = ERR_PTR(-ENOENT);
2184         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2185                 goto out_unlock;
2186
2187         mm = get_task_mm(task);
2188         if (!mm)
2189                 goto out_unlock;
2190
2191         down_read(&mm->mmap_sem);
2192         vma = find_exact_vma(mm, vm_start, vm_end);
2193         if (!vma)
2194                 goto out_no_vma;
2195
2196         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2197
2198 out_no_vma:
2199         up_read(&mm->mmap_sem);
2200         mmput(mm);
2201 out_unlock:
2202         unlock_trace(task);
2203 out_put_task:
2204         put_task_struct(task);
2205 out:
2206         return result;
2207 }
2208
2209 static const struct inode_operations proc_map_files_inode_operations = {
2210         .lookup         = proc_map_files_lookup,
2211         .permission     = proc_fd_permission,
2212         .setattr        = proc_setattr,
2213 };
2214
2215 static int
2216 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2217 {
2218         struct dentry *dentry = filp->f_path.dentry;
2219         struct inode *inode = dentry->d_inode;
2220         struct vm_area_struct *vma;
2221         struct task_struct *task;
2222         struct mm_struct *mm;
2223         ino_t ino;
2224         int ret;
2225
2226         ret = -EACCES;
2227         if (!capable(CAP_SYS_ADMIN))
2228                 goto out;
2229
2230         ret = -ENOENT;
2231         task = get_proc_task(inode);
2232         if (!task)
2233                 goto out;
2234
2235         ret = -EACCES;
2236         if (lock_trace(task))
2237                 goto out_put_task;
2238
2239         ret = 0;
2240         switch (filp->f_pos) {
2241         case 0:
2242                 ino = inode->i_ino;
2243                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2244                         goto out_unlock;
2245                 filp->f_pos++;
2246         case 1:
2247                 ino = parent_ino(dentry);
2248                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2249                         goto out_unlock;
2250                 filp->f_pos++;
2251         default:
2252         {
2253                 unsigned long nr_files, pos, i;
2254                 struct flex_array *fa = NULL;
2255                 struct map_files_info info;
2256                 struct map_files_info *p;
2257
2258                 mm = get_task_mm(task);
2259                 if (!mm)
2260                         goto out_unlock;
2261                 down_read(&mm->mmap_sem);
2262
2263                 nr_files = 0;
2264
2265                 /*
2266                  * We need two passes here:
2267                  *
2268                  *  1) Collect vmas of mapped files with mmap_sem taken
2269                  *  2) Release mmap_sem and instantiate entries
2270                  *
2271                  * otherwise we get lockdep complained, since filldir()
2272                  * routine might require mmap_sem taken in might_fault().
2273                  */
2274
2275                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2276                         if (vma->vm_file && ++pos > filp->f_pos)
2277                                 nr_files++;
2278                 }
2279
2280                 if (nr_files) {
2281                         fa = flex_array_alloc(sizeof(info), nr_files,
2282                                                 GFP_KERNEL);
2283                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2284                                                         GFP_KERNEL)) {
2285                                 ret = -ENOMEM;
2286                                 if (fa)
2287                                         flex_array_free(fa);
2288                                 up_read(&mm->mmap_sem);
2289                                 mmput(mm);
2290                                 goto out_unlock;
2291                         }
2292                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2293                                         vma = vma->vm_next) {
2294                                 if (!vma->vm_file)
2295                                         continue;
2296                                 if (++pos <= filp->f_pos)
2297                                         continue;
2298
2299                                 get_file(vma->vm_file);
2300                                 info.file = vma->vm_file;
2301                                 info.len = snprintf(info.name,
2302                                                 sizeof(info.name), "%lx-%lx",
2303                                                 vma->vm_start, vma->vm_end);
2304                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2305                                         BUG();
2306                         }
2307                 }
2308                 up_read(&mm->mmap_sem);
2309
2310                 for (i = 0; i < nr_files; i++) {
2311                         p = flex_array_get(fa, i);
2312                         ret = proc_fill_cache(filp, dirent, filldir,
2313                                               p->name, p->len,
2314                                               proc_map_files_instantiate,
2315                                               task, p->file);
2316                         if (ret)
2317                                 break;
2318                         filp->f_pos++;
2319                         fput(p->file);
2320                 }
2321                 for (; i < nr_files; i++) {
2322                         /*
2323                          * In case of error don't forget
2324                          * to put rest of file refs.
2325                          */
2326                         p = flex_array_get(fa, i);
2327                         fput(p->file);
2328                 }
2329                 if (fa)
2330                         flex_array_free(fa);
2331                 mmput(mm);
2332         }
2333         }
2334
2335 out_unlock:
2336         unlock_trace(task);
2337 out_put_task:
2338         put_task_struct(task);
2339 out:
2340         return ret;
2341 }
2342
2343 static const struct file_operations proc_map_files_operations = {
2344         .read           = generic_read_dir,
2345         .readdir        = proc_map_files_readdir,
2346         .llseek         = default_llseek,
2347 };
2348
2349 #endif /* CONFIG_CHECKPOINT_RESTORE */
2350
2351 /*
2352  * /proc/pid/fd needs a special permission handler so that a process can still
2353  * access /proc/self/fd after it has executed a setuid().
2354  */
2355 static int proc_fd_permission(struct inode *inode, int mask)
2356 {
2357         int rv = generic_permission(inode, mask);
2358         if (rv == 0)
2359                 return 0;
2360         if (task_pid(current) == proc_pid(inode))
2361                 rv = 0;
2362         return rv;
2363 }
2364
2365 /*
2366  * proc directories can do almost nothing..
2367  */
2368 static const struct inode_operations proc_fd_inode_operations = {
2369         .lookup         = proc_lookupfd,
2370         .permission     = proc_fd_permission,
2371         .setattr        = proc_setattr,
2372 };
2373
2374 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2375         struct dentry *dentry, struct task_struct *task, const void *ptr)
2376 {
2377         unsigned fd = *(unsigned *)ptr;
2378         struct inode *inode;
2379         struct proc_inode *ei;
2380         struct dentry *error = ERR_PTR(-ENOENT);
2381
2382         inode = proc_pid_make_inode(dir->i_sb, task);
2383         if (!inode)
2384                 goto out;
2385         ei = PROC_I(inode);
2386         ei->fd = fd;
2387         inode->i_mode = S_IFREG | S_IRUSR;
2388         inode->i_fop = &proc_fdinfo_file_operations;
2389         d_set_d_op(dentry, &tid_fd_dentry_operations);
2390         d_add(dentry, inode);
2391         /* Close the race of the process dying before we return the dentry */
2392         if (tid_fd_revalidate(dentry, NULL))
2393                 error = NULL;
2394
2395  out:
2396         return error;
2397 }
2398
2399 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2400                                         struct dentry *dentry,
2401                                         struct nameidata *nd)
2402 {
2403         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2404 }
2405
2406 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2407 {
2408         return proc_readfd_common(filp, dirent, filldir,
2409                                   proc_fdinfo_instantiate);
2410 }
2411
2412 static const struct file_operations proc_fdinfo_operations = {
2413         .read           = generic_read_dir,
2414         .readdir        = proc_readfdinfo,
2415         .llseek         = default_llseek,
2416 };
2417
2418 /*
2419  * proc directories can do almost nothing..
2420  */
2421 static const struct inode_operations proc_fdinfo_inode_operations = {
2422         .lookup         = proc_lookupfdinfo,
2423         .setattr        = proc_setattr,
2424 };
2425
2426
2427 static struct dentry *proc_pident_instantiate(struct inode *dir,
2428         struct dentry *dentry, struct task_struct *task, const void *ptr)
2429 {
2430         const struct pid_entry *p = ptr;
2431         struct inode *inode;
2432         struct proc_inode *ei;
2433         struct dentry *error = ERR_PTR(-ENOENT);
2434
2435         inode = proc_pid_make_inode(dir->i_sb, task);
2436         if (!inode)
2437                 goto out;
2438
2439         ei = PROC_I(inode);
2440         inode->i_mode = p->mode;
2441         if (S_ISDIR(inode->i_mode))
2442                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2443         if (p->iop)
2444                 inode->i_op = p->iop;
2445         if (p->fop)
2446                 inode->i_fop = p->fop;
2447         ei->op = p->op;
2448         d_set_d_op(dentry, &pid_dentry_operations);
2449         d_add(dentry, inode);
2450         /* Close the race of the process dying before we return the dentry */
2451         if (pid_revalidate(dentry, NULL))
2452                 error = NULL;
2453 out:
2454         return error;
2455 }
2456
2457 static struct dentry *proc_pident_lookup(struct inode *dir, 
2458                                          struct dentry *dentry,
2459                                          const struct pid_entry *ents,
2460                                          unsigned int nents)
2461 {
2462         struct dentry *error;
2463         struct task_struct *task = get_proc_task(dir);
2464         const struct pid_entry *p, *last;
2465
2466         error = ERR_PTR(-ENOENT);
2467
2468         if (!task)
2469                 goto out_no_task;
2470
2471         /*
2472          * Yes, it does not scale. And it should not. Don't add
2473          * new entries into /proc/<tgid>/ without very good reasons.
2474          */
2475         last = &ents[nents - 1];
2476         for (p = ents; p <= last; p++) {
2477                 if (p->len != dentry->d_name.len)
2478                         continue;
2479                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2480                         break;
2481         }
2482         if (p > last)
2483                 goto out;
2484
2485         error = proc_pident_instantiate(dir, dentry, task, p);
2486 out:
2487         put_task_struct(task);
2488 out_no_task:
2489         return error;
2490 }
2491
2492 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2493         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2494 {
2495         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2496                                 proc_pident_instantiate, task, p);
2497 }
2498
2499 static int proc_pident_readdir(struct file *filp,
2500                 void *dirent, filldir_t filldir,
2501                 const struct pid_entry *ents, unsigned int nents)
2502 {
2503         int i;
2504         struct dentry *dentry = filp->f_path.dentry;
2505         struct inode *inode = dentry->d_inode;
2506         struct task_struct *task = get_proc_task(inode);
2507         const struct pid_entry *p, *last;
2508         ino_t ino;
2509         int ret;
2510
2511         ret = -ENOENT;
2512         if (!task)
2513                 goto out_no_task;
2514
2515         ret = 0;
2516         i = filp->f_pos;
2517         switch (i) {
2518         case 0:
2519                 ino = inode->i_ino;
2520                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2521                         goto out;
2522                 i++;
2523                 filp->f_pos++;
2524                 /* fall through */
2525         case 1:
2526                 ino = parent_ino(dentry);
2527                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2528                         goto out;
2529                 i++;
2530                 filp->f_pos++;
2531                 /* fall through */
2532         default:
2533                 i -= 2;
2534                 if (i >= nents) {
2535                         ret = 1;
2536                         goto out;
2537                 }
2538                 p = ents + i;
2539                 last = &ents[nents - 1];
2540                 while (p <= last) {
2541                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2542                                 goto out;
2543                         filp->f_pos++;
2544                         p++;
2545                 }
2546         }
2547
2548         ret = 1;
2549 out:
2550         put_task_struct(task);
2551 out_no_task:
2552         return ret;
2553 }
2554
2555 #ifdef CONFIG_SECURITY
2556 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2557                                   size_t count, loff_t *ppos)
2558 {
2559         struct inode * inode = file->f_path.dentry->d_inode;
2560         char *p = NULL;
2561         ssize_t length;
2562         struct task_struct *task = get_proc_task(inode);
2563
2564         if (!task)
2565                 return -ESRCH;
2566
2567         length = security_getprocattr(task,
2568                                       (char*)file->f_path.dentry->d_name.name,
2569                                       &p);
2570         put_task_struct(task);
2571         if (length > 0)
2572                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2573         kfree(p);
2574         return length;
2575 }
2576
2577 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2578                                    size_t count, loff_t *ppos)
2579 {
2580         struct inode * inode = file->f_path.dentry->d_inode;
2581         char *page;
2582         ssize_t length;
2583         struct task_struct *task = get_proc_task(inode);
2584
2585         length = -ESRCH;
2586         if (!task)
2587                 goto out_no_task;
2588         if (count > PAGE_SIZE)
2589                 count = PAGE_SIZE;
2590
2591         /* No partial writes. */
2592         length = -EINVAL;
2593         if (*ppos != 0)
2594                 goto out;
2595
2596         length = -ENOMEM;
2597         page = (char*)__get_free_page(GFP_TEMPORARY);
2598         if (!page)
2599                 goto out;
2600
2601         length = -EFAULT;
2602         if (copy_from_user(page, buf, count))
2603                 goto out_free;
2604
2605         /* Guard against adverse ptrace interaction */
2606         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2607         if (length < 0)
2608                 goto out_free;
2609
2610         length = security_setprocattr(task,
2611                                       (char*)file->f_path.dentry->d_name.name,
2612                                       (void*)page, count);
2613         mutex_unlock(&task->signal->cred_guard_mutex);
2614 out_free:
2615         free_page((unsigned long) page);
2616 out:
2617         put_task_struct(task);
2618 out_no_task:
2619         return length;
2620 }
2621
2622 static const struct file_operations proc_pid_attr_operations = {
2623         .read           = proc_pid_attr_read,
2624         .write          = proc_pid_attr_write,
2625         .llseek         = generic_file_llseek,
2626 };
2627
2628 static const struct pid_entry attr_dir_stuff[] = {
2629         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2630         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2631         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2632         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2633         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2634         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2635 };
2636
2637 static int proc_attr_dir_readdir(struct file * filp,
2638                              void * dirent, filldir_t filldir)
2639 {
2640         return proc_pident_readdir(filp,dirent,filldir,
2641                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2642 }
2643
2644 static const struct file_operations proc_attr_dir_operations = {
2645         .read           = generic_read_dir,
2646         .readdir        = proc_attr_dir_readdir,
2647         .llseek         = default_llseek,
2648 };
2649
2650 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2651                                 struct dentry *dentry, struct nameidata *nd)
2652 {
2653         return proc_pident_lookup(dir, dentry,
2654                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2655 }
2656
2657 static const struct inode_operations proc_attr_dir_inode_operations = {
2658         .lookup         = proc_attr_dir_lookup,
2659         .getattr        = pid_getattr,
2660         .setattr        = proc_setattr,
2661 };
2662
2663 #endif
2664
2665 #ifdef CONFIG_ELF_CORE
2666 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2667                                          size_t count, loff_t *ppos)
2668 {
2669         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2670         struct mm_struct *mm;
2671         char buffer[PROC_NUMBUF];
2672         size_t len;
2673         int ret;
2674
2675         if (!task)
2676                 return -ESRCH;
2677
2678         ret = 0;
2679         mm = get_task_mm(task);
2680         if (mm) {
2681                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2682                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2683                                 MMF_DUMP_FILTER_SHIFT));
2684                 mmput(mm);
2685                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2686         }
2687
2688         put_task_struct(task);
2689
2690         return ret;
2691 }
2692
2693 static ssize_t proc_coredump_filter_write(struct file *file,
2694                                           const char __user *buf,
2695                                           size_t count,
2696                                           loff_t *ppos)
2697 {
2698         struct task_struct *task;
2699         struct mm_struct *mm;
2700         char buffer[PROC_NUMBUF], *end;
2701         unsigned int val;
2702         int ret;
2703         int i;
2704         unsigned long mask;
2705
2706         ret = -EFAULT;
2707         memset(buffer, 0, sizeof(buffer));
2708         if (count > sizeof(buffer) - 1)
2709                 count = sizeof(buffer) - 1;
2710         if (copy_from_user(buffer, buf, count))
2711                 goto out_no_task;
2712
2713         ret = -EINVAL;
2714         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2715         if (*end == '\n')
2716                 end++;
2717         if (end - buffer == 0)
2718                 goto out_no_task;
2719
2720         ret = -ESRCH;
2721         task = get_proc_task(file->f_dentry->d_inode);
2722         if (!task)
2723                 goto out_no_task;
2724
2725         ret = end - buffer;
2726         mm = get_task_mm(task);
2727         if (!mm)
2728                 goto out_no_mm;
2729
2730         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2731                 if (val & mask)
2732                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2733                 else
2734                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2735         }
2736
2737         mmput(mm);
2738  out_no_mm:
2739         put_task_struct(task);
2740  out_no_task:
2741         return ret;
2742 }
2743
2744 static const struct file_operations proc_coredump_filter_operations = {
2745         .read           = proc_coredump_filter_read,
2746         .write          = proc_coredump_filter_write,
2747         .llseek         = generic_file_llseek,
2748 };
2749 #endif
2750
2751 /*
2752  * /proc/self:
2753  */
2754 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2755                               int buflen)
2756 {
2757         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2758         pid_t tgid = task_tgid_nr_ns(current, ns);
2759         char tmp[PROC_NUMBUF];
2760         if (!tgid)
2761                 return -ENOENT;
2762         sprintf(tmp, "%d", tgid);
2763         return vfs_readlink(dentry,buffer,buflen,tmp);
2764 }
2765
2766 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2767 {
2768         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2769         pid_t tgid = task_tgid_nr_ns(current, ns);
2770         char *name = ERR_PTR(-ENOENT);
2771         if (tgid) {
2772                 name = __getname();
2773                 if (!name)
2774                         name = ERR_PTR(-ENOMEM);
2775                 else
2776                         sprintf(name, "%d", tgid);
2777         }
2778         nd_set_link(nd, name);
2779         return NULL;
2780 }
2781
2782 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2783                                 void *cookie)
2784 {
2785         char *s = nd_get_link(nd);
2786         if (!IS_ERR(s))
2787                 __putname(s);
2788 }
2789
2790 static const struct inode_operations proc_self_inode_operations = {
2791         .readlink       = proc_self_readlink,
2792         .follow_link    = proc_self_follow_link,
2793         .put_link       = proc_self_put_link,
2794 };
2795
2796 /*
2797  * proc base
2798  *
2799  * These are the directory entries in the root directory of /proc
2800  * that properly belong to the /proc filesystem, as they describe
2801  * describe something that is process related.
2802  */
2803 static const struct pid_entry proc_base_stuff[] = {
2804         NOD("self", S_IFLNK|S_IRWXUGO,
2805                 &proc_self_inode_operations, NULL, {}),
2806 };
2807
2808 static struct dentry *proc_base_instantiate(struct inode *dir,
2809         struct dentry *dentry, struct task_struct *task, const void *ptr)
2810 {
2811         const struct pid_entry *p = ptr;
2812         struct inode *inode;
2813         struct proc_inode *ei;
2814         struct dentry *error;
2815
2816         /* Allocate the inode */
2817         error = ERR_PTR(-ENOMEM);
2818         inode = new_inode(dir->i_sb);
2819         if (!inode)
2820                 goto out;
2821
2822         /* Initialize the inode */
2823         ei = PROC_I(inode);
2824         inode->i_ino = get_next_ino();
2825         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2826
2827         /*
2828          * grab the reference to the task.
2829          */
2830         ei->pid = get_task_pid(task, PIDTYPE_PID);
2831         if (!ei->pid)
2832                 goto out_iput;
2833
2834         inode->i_mode = p->mode;
2835         if (S_ISDIR(inode->i_mode))
2836                 set_nlink(inode, 2);
2837         if (S_ISLNK(inode->i_mode))
2838                 inode->i_size = 64;
2839         if (p->iop)
2840                 inode->i_op = p->iop;
2841         if (p->fop)
2842                 inode->i_fop = p->fop;
2843         ei->op = p->op;
2844         d_add(dentry, inode);
2845         error = NULL;
2846 out:
2847         return error;
2848 out_iput:
2849         iput(inode);
2850         goto out;
2851 }
2852
2853 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2854 {
2855         struct dentry *error;
2856         struct task_struct *task = get_proc_task(dir);
2857         const struct pid_entry *p, *last;
2858
2859         error = ERR_PTR(-ENOENT);
2860
2861         if (!task)
2862                 goto out_no_task;
2863
2864         /* Lookup the directory entry */
2865         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2866         for (p = proc_base_stuff; p <= last; p++) {
2867                 if (p->len != dentry->d_name.len)
2868                         continue;
2869                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2870                         break;
2871         }
2872         if (p > last)
2873                 goto out;
2874
2875         error = proc_base_instantiate(dir, dentry, task, p);
2876
2877 out:
2878         put_task_struct(task);
2879 out_no_task:
2880         return error;
2881 }
2882
2883 static int proc_base_fill_cache(struct file *filp, void *dirent,
2884         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2885 {
2886         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2887                                 proc_base_instantiate, task, p);
2888 }
2889
2890 #ifdef CONFIG_TASK_IO_ACCOUNTING
2891 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2892 {
2893         struct task_io_accounting acct = task->ioac;
2894         unsigned long flags;
2895         int result;
2896
2897         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2898         if (result)
2899                 return result;
2900
2901         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2902                 result = -EACCES;
2903                 goto out_unlock;
2904         }
2905
2906         if (whole && lock_task_sighand(task, &flags)) {
2907                 struct task_struct *t = task;
2908
2909                 task_io_accounting_add(&acct, &task->signal->ioac);
2910                 while_each_thread(task, t)
2911                         task_io_accounting_add(&acct, &t->ioac);
2912
2913                 unlock_task_sighand(task, &flags);
2914         }
2915         result = sprintf(buffer,
2916                         "rchar: %llu\n"
2917                         "wchar: %llu\n"
2918                         "syscr: %llu\n"
2919                         "syscw: %llu\n"
2920                         "read_bytes: %llu\n"
2921                         "write_bytes: %llu\n"
2922                         "cancelled_write_bytes: %llu\n",
2923                         (unsigned long long)acct.rchar,
2924                         (unsigned long long)acct.wchar,
2925                         (unsigned long long)acct.syscr,
2926                         (unsigned long long)acct.syscw,
2927                         (unsigned long long)acct.read_bytes,
2928                         (unsigned long long)acct.write_bytes,
2929                         (unsigned long long)acct.cancelled_write_bytes);
2930 out_unlock:
2931         mutex_unlock(&task->signal->cred_guard_mutex);
2932         return result;
2933 }
2934
2935 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2936 {
2937         return do_io_accounting(task, buffer, 0);
2938 }
2939
2940 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2941 {
2942         return do_io_accounting(task, buffer, 1);
2943 }
2944 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2945
2946 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2947                                 struct pid *pid, struct task_struct *task)
2948 {
2949         int err = lock_trace(task);
2950         if (!err) {
2951                 seq_printf(m, "%08x\n", task->personality);
2952                 unlock_trace(task);
2953         }
2954         return err;
2955 }
2956
2957 /*
2958  * Thread groups
2959  */
2960 static const struct file_operations proc_task_operations;
2961 static const struct inode_operations proc_task_inode_operations;
2962
2963 static const struct pid_entry tgid_base_stuff[] = {
2964         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2965         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2966 #ifdef CONFIG_CHECKPOINT_RESTORE
2967         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2968 #endif
2969         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2970         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2971 #ifdef CONFIG_NET
2972         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2973 #endif
2974         REG("environ",    S_IRUSR, proc_environ_operations),
2975         INF("auxv",       S_IRUSR, proc_pid_auxv),
2976         ONE("status",     S_IRUGO, proc_pid_status),
2977         ONE("personality", S_IRUGO, proc_pid_personality),
2978         INF("limits",     S_IRUGO, proc_pid_limits),
2979 #ifdef CONFIG_SCHED_DEBUG
2980         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2981 #endif
2982 #ifdef CONFIG_SCHED_AUTOGROUP
2983         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2984 #endif
2985         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2986 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2987         INF("syscall",    S_IRUGO, proc_pid_syscall),
2988 #endif
2989         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2990         ONE("stat",       S_IRUGO, proc_tgid_stat),
2991         ONE("statm",      S_IRUGO, proc_pid_statm),
2992         REG("maps",       S_IRUGO, proc_maps_operations),
2993 #ifdef CONFIG_NUMA
2994         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2995 #endif
2996         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2997         LNK("cwd",        proc_cwd_link),
2998         LNK("root",       proc_root_link),
2999         LNK("exe",        proc_exe_link),
3000         REG("mounts",     S_IRUGO, proc_mounts_operations),
3001         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3002         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3003 #ifdef CONFIG_PROC_PAGE_MONITOR
3004         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3005         REG("smaps",      S_IRUGO, proc_smaps_operations),
3006         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3007 #endif
3008 #ifdef CONFIG_SECURITY
3009         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3010 #endif
3011 #ifdef CONFIG_KALLSYMS
3012         INF("wchan",      S_IRUGO, proc_pid_wchan),
3013 #endif
3014 #ifdef CONFIG_STACKTRACE
3015         ONE("stack",      S_IRUGO, proc_pid_stack),
3016 #endif
3017 #ifdef CONFIG_SCHEDSTATS
3018         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3019 #endif
3020 #ifdef CONFIG_LATENCYTOP
3021         REG("latency",  S_IRUGO, proc_lstats_operations),
3022 #endif
3023 #ifdef CONFIG_PROC_PID_CPUSET
3024         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3025 #endif
3026 #ifdef CONFIG_CGROUPS
3027         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3028 #endif
3029         INF("oom_score",  S_IRUGO, proc_oom_score),
3030         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3031         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3032 #ifdef CONFIG_AUDITSYSCALL
3033         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3034         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3035 #endif
3036 #ifdef CONFIG_FAULT_INJECTION
3037         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3038 #endif
3039 #ifdef CONFIG_ELF_CORE
3040         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3041 #endif
3042 #ifdef CONFIG_TASK_IO_ACCOUNTING
3043         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3044 #endif
3045 #ifdef CONFIG_HARDWALL
3046         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3047 #endif
3048 };
3049
3050 static int proc_tgid_base_readdir(struct file * filp,
3051                              void * dirent, filldir_t filldir)
3052 {
3053         return proc_pident_readdir(filp,dirent,filldir,
3054                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3055 }
3056
3057 static const struct file_operations proc_tgid_base_operations = {
3058         .read           = generic_read_dir,
3059         .readdir        = proc_tgid_base_readdir,
3060         .llseek         = default_llseek,
3061 };
3062
3063 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3064         return proc_pident_lookup(dir, dentry,
3065                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3066 }
3067
3068 static const struct inode_operations proc_tgid_base_inode_operations = {
3069         .lookup         = proc_tgid_base_lookup,
3070         .getattr        = pid_getattr,
3071         .setattr        = proc_setattr,
3072         .permission     = proc_pid_permission,
3073 };
3074
3075 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3076 {
3077         struct dentry *dentry, *leader, *dir;
3078         char buf[PROC_NUMBUF];
3079         struct qstr name;
3080
3081         name.name = buf;
3082         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3083         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3084         if (dentry) {
3085                 shrink_dcache_parent(dentry);
3086                 d_drop(dentry);
3087                 dput(dentry);
3088         }
3089
3090         name.name = buf;
3091         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3092         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3093         if (!leader)
3094                 goto out;
3095
3096         name.name = "task";
3097         name.len = strlen(name.name);
3098         dir = d_hash_and_lookup(leader, &name);
3099         if (!dir)
3100                 goto out_put_leader;
3101
3102         name.name = buf;
3103         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3104         dentry = d_hash_and_lookup(dir, &name);
3105         if (dentry) {
3106                 shrink_dcache_parent(dentry);
3107                 d_drop(dentry);
3108                 dput(dentry);
3109         }
3110
3111         dput(dir);
3112 out_put_leader:
3113         dput(leader);
3114 out:
3115         return;
3116 }
3117
3118 /**
3119  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3120  * @task: task that should be flushed.
3121  *
3122  * When flushing dentries from proc, one needs to flush them from global
3123  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3124  * in. This call is supposed to do all of this job.
3125  *
3126  * Looks in the dcache for
3127  * /proc/@pid
3128  * /proc/@tgid/task/@pid
3129  * if either directory is present flushes it and all of it'ts children
3130  * from the dcache.
3131  *
3132  * It is safe and reasonable to cache /proc entries for a task until
3133  * that task exits.  After that they just clog up the dcache with
3134  * useless entries, possibly causing useful dcache entries to be
3135  * flushed instead.  This routine is proved to flush those useless
3136  * dcache entries at process exit time.
3137  *
3138  * NOTE: This routine is just an optimization so it does not guarantee
3139  *       that no dcache entries will exist at process exit time it
3140  *       just makes it very unlikely that any will persist.
3141  */
3142
3143 void proc_flush_task(struct task_struct *task)
3144 {
3145         int i;
3146         struct pid *pid, *tgid;
3147         struct upid *upid;
3148
3149         pid = task_pid(task);
3150         tgid = task_tgid(task);
3151
3152         for (i = 0; i <= pid->level; i++) {
3153                 upid = &pid->numbers[i];
3154                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3155                                         tgid->numbers[i].nr);
3156         }
3157
3158         upid = &pid->numbers[pid->level];
3159         if (upid->nr == 1)
3160                 pid_ns_release_proc(upid->ns);
3161 }
3162
3163 static struct dentry *proc_pid_instantiate(struct inode *dir,
3164                                            struct dentry * dentry,
3165                                            struct task_struct *task, const void *ptr)
3166 {
3167         struct dentry *error = ERR_PTR(-ENOENT);
3168         struct inode *inode;
3169
3170         inode = proc_pid_make_inode(dir->i_sb, task);
3171         if (!inode)
3172                 goto out;
3173
3174         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3175         inode->i_op = &proc_tgid_base_inode_operations;
3176         inode->i_fop = &proc_tgid_base_operations;
3177         inode->i_flags|=S_IMMUTABLE;
3178
3179         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3180                                                   ARRAY_SIZE(tgid_base_stuff)));
3181
3182         d_set_d_op(dentry, &pid_dentry_operations);
3183
3184         d_add(dentry, inode);
3185         /* Close the race of the process dying before we return the dentry */
3186         if (pid_revalidate(dentry, NULL))
3187                 error = NULL;
3188 out:
3189         return error;
3190 }
3191
3192 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3193 {
3194         struct dentry *result;
3195         struct task_struct *task;
3196         unsigned tgid;
3197         struct pid_namespace *ns;
3198
3199         result = proc_base_lookup(dir, dentry);
3200         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3201                 goto out;
3202
3203         tgid = name_to_int(dentry);
3204         if (tgid == ~0U)
3205                 goto out;
3206
3207         ns = dentry->d_sb->s_fs_info;
3208         rcu_read_lock();
3209         task = find_task_by_pid_ns(tgid, ns);
3210         if (task)
3211                 get_task_struct(task);
3212         rcu_read_unlock();
3213         if (!task)
3214                 goto out;
3215
3216         result = proc_pid_instantiate(dir, dentry, task, NULL);
3217         put_task_struct(task);
3218 out:
3219         return result;
3220 }
3221
3222 /*
3223  * Find the first task with tgid >= tgid
3224  *
3225  */
3226 struct tgid_iter {
3227         unsigned int tgid;
3228         struct task_struct *task;
3229 };
3230 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3231 {
3232         struct pid *pid;
3233
3234         if (iter.task)
3235                 put_task_struct(iter.task);
3236         rcu_read_lock();
3237 retry:
3238         iter.task = NULL;
3239         pid = find_ge_pid(iter.tgid, ns);
3240         if (pid) {
3241                 iter.tgid = pid_nr_ns(pid, ns);
3242                 iter.task = pid_task(pid, PIDTYPE_PID);
3243                 /* What we to know is if the pid we have find is the
3244                  * pid of a thread_group_leader.  Testing for task
3245                  * being a thread_group_leader is the obvious thing
3246                  * todo but there is a window when it fails, due to
3247                  * the pid transfer logic in de_thread.
3248                  *
3249                  * So we perform the straight forward test of seeing
3250                  * if the pid we have found is the pid of a thread
3251                  * group leader, and don't worry if the task we have
3252                  * found doesn't happen to be a thread group leader.
3253                  * As we don't care in the case of readdir.
3254                  */
3255                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3256                         iter.tgid += 1;
3257                         goto retry;
3258                 }
3259                 get_task_struct(iter.task);
3260         }
3261         rcu_read_unlock();
3262         return iter;
3263 }
3264
3265 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3266
3267 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3268         struct tgid_iter iter)
3269 {
3270         char name[PROC_NUMBUF];
3271         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3272         return proc_fill_cache(filp, dirent, filldir, name, len,
3273                                 proc_pid_instantiate, iter.task, NULL);
3274 }
3275
3276 static int fake_filldir(void *buf, const char *name, int namelen,
3277                         loff_t offset, u64 ino, unsigned d_type)
3278 {
3279         return 0;
3280 }
3281
3282 /* for the /proc/ directory itself, after non-process stuff has been done */
3283 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3284 {
3285         unsigned int nr;
3286         struct task_struct *reaper;
3287         struct tgid_iter iter;
3288         struct pid_namespace *ns;
3289         filldir_t __filldir;
3290
3291         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3292                 goto out_no_task;
3293         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3294
3295         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3296         if (!reaper)
3297                 goto out_no_task;
3298
3299         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3300                 const struct pid_entry *p = &proc_base_stuff[nr];
3301                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3302                         goto out;
3303         }
3304
3305         ns = filp->f_dentry->d_sb->s_fs_info;
3306         iter.task = NULL;
3307         iter.tgid = filp->f_pos - TGID_OFFSET;
3308         for (iter = next_tgid(ns, iter);
3309              iter.task;
3310              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3311                 if (has_pid_permissions(ns, iter.task, 2))
3312                         __filldir = filldir;
3313                 else
3314                         __filldir = fake_filldir;
3315
3316                 filp->f_pos = iter.tgid + TGID_OFFSET;
3317                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3318                         put_task_struct(iter.task);
3319                         goto out;
3320                 }
3321         }
3322         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3323 out:
3324         put_task_struct(reaper);
3325 out_no_task:
3326         return 0;
3327 }
3328
3329 /*
3330  * Tasks
3331  */
3332 static const struct pid_entry tid_base_stuff[] = {
3333         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3334         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3335         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3336         REG("environ",   S_IRUSR, proc_environ_operations),
3337         INF("auxv",      S_IRUSR, proc_pid_auxv),
3338         ONE("status",    S_IRUGO, proc_pid_status),
3339         ONE("personality", S_IRUGO, proc_pid_personality),
3340         INF("limits",    S_IRUGO, proc_pid_limits),
3341 #ifdef CONFIG_SCHED_DEBUG
3342         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3343 #endif
3344         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3345 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3346         INF("syscall",   S_IRUGO, proc_pid_syscall),
3347 #endif
3348         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3349         ONE("stat",      S_IRUGO, proc_tid_stat),
3350         ONE("statm",     S_IRUGO, proc_pid_statm),
3351         REG("maps",      S_IRUGO, proc_maps_operations),
3352 #ifdef CONFIG_NUMA
3353         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3354 #endif
3355         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3356         LNK("cwd",       proc_cwd_link),
3357         LNK("root",      proc_root_link),
3358         LNK("exe",       proc_exe_link),
3359         REG("mounts",    S_IRUGO, proc_mounts_operations),
3360         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3361 #ifdef CONFIG_PROC_PAGE_MONITOR
3362         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3363         REG("smaps",     S_IRUGO, proc_smaps_operations),
3364         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3365 #endif
3366 #ifdef CONFIG_SECURITY
3367         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3368 #endif
3369 #ifdef CONFIG_KALLSYMS
3370         INF("wchan",     S_IRUGO, proc_pid_wchan),
3371 #endif
3372 #ifdef CONFIG_STACKTRACE
3373         ONE("stack",      S_IRUGO, proc_pid_stack),
3374 #endif
3375 #ifdef CONFIG_SCHEDSTATS
3376         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3377 #endif
3378 #ifdef CONFIG_LATENCYTOP
3379         REG("latency",  S_IRUGO, proc_lstats_operations),
3380 #endif
3381 #ifdef CONFIG_PROC_PID_CPUSET
3382         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3383 #endif
3384 #ifdef CONFIG_CGROUPS
3385         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3386 #endif
3387         INF("oom_score", S_IRUGO, proc_oom_score),
3388         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3389         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3390 #ifdef CONFIG_AUDITSYSCALL
3391         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3392         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3393 #endif
3394 #ifdef CONFIG_FAULT_INJECTION
3395         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3396 #endif
3397 #ifdef CONFIG_TASK_IO_ACCOUNTING
3398         INF("io",       S_IRUSR, proc_tid_io_accounting),
3399 #endif
3400 #ifdef CONFIG_HARDWALL
3401         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3402 #endif
3403 };
3404
3405 static int proc_tid_base_readdir(struct file * filp,
3406                              void * dirent, filldir_t filldir)
3407 {
3408         return proc_pident_readdir(filp,dirent,filldir,
3409                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3410 }
3411
3412 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3413         return proc_pident_lookup(dir, dentry,
3414                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3415 }
3416
3417 static const struct file_operations proc_tid_base_operations = {
3418         .read           = generic_read_dir,
3419         .readdir        = proc_tid_base_readdir,
3420         .llseek         = default_llseek,
3421 };
3422
3423 static const struct inode_operations proc_tid_base_inode_operations = {
3424         .lookup         = proc_tid_base_lookup,
3425         .getattr        = pid_getattr,
3426         .setattr        = proc_setattr,
3427 };
3428
3429 static struct dentry *proc_task_instantiate(struct inode *dir,
3430         struct dentry *dentry, struct task_struct *task, const void *ptr)
3431 {
3432         struct dentry *error = ERR_PTR(-ENOENT);
3433         struct inode *inode;
3434         inode = proc_pid_make_inode(dir->i_sb, task);
3435
3436         if (!inode)
3437                 goto out;
3438         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3439         inode->i_op = &proc_tid_base_inode_operations;
3440         inode->i_fop = &proc_tid_base_operations;
3441         inode->i_flags|=S_IMMUTABLE;
3442
3443         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3444                                                   ARRAY_SIZE(tid_base_stuff)));
3445
3446         d_set_d_op(dentry, &pid_dentry_operations);
3447
3448         d_add(dentry, inode);
3449         /* Close the race of the process dying before we return the dentry */
3450         if (pid_revalidate(dentry, NULL))
3451                 error = NULL;
3452 out:
3453         return error;
3454 }
3455
3456 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3457 {
3458         struct dentry *result = ERR_PTR(-ENOENT);
3459         struct task_struct *task;
3460         struct task_struct *leader = get_proc_task(dir);
3461         unsigned tid;
3462         struct pid_namespace *ns;
3463
3464         if (!leader)
3465                 goto out_no_task;
3466
3467         tid = name_to_int(dentry);
3468         if (tid == ~0U)
3469                 goto out;
3470
3471         ns = dentry->d_sb->s_fs_info;
3472         rcu_read_lock();
3473         task = find_task_by_pid_ns(tid, ns);
3474         if (task)
3475                 get_task_struct(task);
3476         rcu_read_unlock();
3477         if (!task)
3478                 goto out;
3479         if (!same_thread_group(leader, task))
3480                 goto out_drop_task;
3481
3482         result = proc_task_instantiate(dir, dentry, task, NULL);
3483 out_drop_task:
3484         put_task_struct(task);
3485 out:
3486         put_task_struct(leader);
3487 out_no_task:
3488         return result;
3489 }
3490
3491 /*
3492  * Find the first tid of a thread group to return to user space.
3493  *
3494  * Usually this is just the thread group leader, but if the users
3495  * buffer was too small or there was a seek into the middle of the
3496  * directory we have more work todo.
3497  *
3498  * In the case of a short read we start with find_task_by_pid.
3499  *
3500  * In the case of a seek we start with the leader and walk nr
3501  * threads past it.
3502  */
3503 static struct task_struct *first_tid(struct task_struct *leader,
3504                 int tid, int nr, struct pid_namespace *ns)
3505 {
3506         struct task_struct *pos;
3507
3508         rcu_read_lock();
3509         /* Attempt to start with the pid of a thread */
3510         if (tid && (nr > 0)) {
3511                 pos = find_task_by_pid_ns(tid, ns);
3512                 if (pos && (pos->group_leader == leader))
3513                         goto found;
3514         }
3515
3516         /* If nr exceeds the number of threads there is nothing todo */
3517         pos = NULL;
3518         if (nr && nr >= get_nr_threads(leader))
3519                 goto out;
3520
3521         /* If we haven't found our starting place yet start
3522          * with the leader and walk nr threads forward.
3523          */
3524         for (pos = leader; nr > 0; --nr) {
3525                 pos = next_thread(pos);
3526                 if (pos == leader) {
3527                         pos = NULL;
3528                         goto out;
3529                 }
3530         }
3531 found:
3532         get_task_struct(pos);
3533 out:
3534         rcu_read_unlock();
3535         return pos;
3536 }
3537
3538 /*
3539  * Find the next thread in the thread list.
3540  * Return NULL if there is an error or no next thread.
3541  *
3542  * The reference to the input task_struct is released.
3543  */
3544 static struct task_struct *next_tid(struct task_struct *start)
3545 {
3546         struct task_struct *pos = NULL;
3547         rcu_read_lock();
3548         if (pid_alive(start)) {
3549                 pos = next_thread(start);
3550                 if (thread_group_leader(pos))
3551                         pos = NULL;
3552                 else
3553                         get_task_struct(pos);
3554         }
3555         rcu_read_unlock();
3556         put_task_struct(start);
3557         return pos;
3558 }
3559
3560 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3561         struct task_struct *task, int tid)
3562 {
3563         char name[PROC_NUMBUF];
3564         int len = snprintf(name, sizeof(name), "%d", tid);
3565         return proc_fill_cache(filp, dirent, filldir, name, len,
3566                                 proc_task_instantiate, task, NULL);
3567 }
3568
3569 /* for the /proc/TGID/task/ directories */
3570 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3571 {
3572         struct dentry *dentry = filp->f_path.dentry;
3573         struct inode *inode = dentry->d_inode;
3574         struct task_struct *leader = NULL;
3575         struct task_struct *task;
3576         int retval = -ENOENT;
3577         ino_t ino;
3578         int tid;
3579         struct pid_namespace *ns;
3580
3581         task = get_proc_task(inode);
3582         if (!task)
3583                 goto out_no_task;
3584         rcu_read_lock();
3585         if (pid_alive(task)) {
3586                 leader = task->group_leader;
3587                 get_task_struct(leader);
3588         }
3589         rcu_read_unlock();
3590         put_task_struct(task);
3591         if (!leader)
3592                 goto out_no_task;
3593         retval = 0;
3594
3595         switch ((unsigned long)filp->f_pos) {
3596         case 0:
3597                 ino = inode->i_ino;
3598                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3599                         goto out;
3600                 filp->f_pos++;
3601                 /* fall through */
3602         case 1:
3603                 ino = parent_ino(dentry);
3604                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3605                         goto out;
3606                 filp->f_pos++;
3607                 /* fall through */
3608         }
3609
3610         /* f_version caches the tgid value that the last readdir call couldn't
3611          * return. lseek aka telldir automagically resets f_version to 0.
3612          */
3613         ns = filp->f_dentry->d_sb->s_fs_info;
3614         tid = (int)filp->f_version;
3615         filp->f_version = 0;
3616         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3617              task;
3618              task = next_tid(task), filp->f_pos++) {
3619                 tid = task_pid_nr_ns(task, ns);
3620                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3621                         /* returning this tgid failed, save it as the first
3622                          * pid for the next readir call */
3623                         filp->f_version = (u64)tid;
3624                         put_task_struct(task);
3625                         break;
3626                 }
3627         }
3628 out:
3629         put_task_struct(leader);
3630 out_no_task:
3631         return retval;
3632 }
3633
3634 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3635 {
3636         struct inode *inode = dentry->d_inode;
3637         struct task_struct *p = get_proc_task(inode);
3638         generic_fillattr(inode, stat);
3639
3640         if (p) {
3641                 stat->nlink += get_nr_threads(p);
3642                 put_task_struct(p);
3643         }
3644
3645         return 0;
3646 }
3647
3648 static const struct inode_operations proc_task_inode_operations = {
3649         .lookup         = proc_task_lookup,
3650         .getattr        = proc_task_getattr,
3651         .setattr        = proc_setattr,
3652         .permission     = proc_pid_permission,
3653 };
3654
3655 static const struct file_operations proc_task_operations = {
3656         .read           = generic_read_dir,
3657         .readdir        = proc_task_readdir,
3658         .llseek         = default_llseek,
3659 };