Wrap accesses to the fd_sets in struct fdtable
[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 = nice;
1314         err = proc_sched_autogroup_set_nice(p, &err);
1315         if (err)
1316                 count = err;
1317
1318         put_task_struct(p);
1319
1320         return count;
1321 }
1322
1323 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1324 {
1325         int ret;
1326
1327         ret = single_open(filp, sched_autogroup_show, NULL);
1328         if (!ret) {
1329                 struct seq_file *m = filp->private_data;
1330
1331                 m->private = inode;
1332         }
1333         return ret;
1334 }
1335
1336 static const struct file_operations proc_pid_sched_autogroup_operations = {
1337         .open           = sched_autogroup_open,
1338         .read           = seq_read,
1339         .write          = sched_autogroup_write,
1340         .llseek         = seq_lseek,
1341         .release        = single_release,
1342 };
1343
1344 #endif /* CONFIG_SCHED_AUTOGROUP */
1345
1346 static ssize_t comm_write(struct file *file, const char __user *buf,
1347                                 size_t count, loff_t *offset)
1348 {
1349         struct inode *inode = file->f_path.dentry->d_inode;
1350         struct task_struct *p;
1351         char buffer[TASK_COMM_LEN];
1352
1353         memset(buffer, 0, sizeof(buffer));
1354         if (count > sizeof(buffer) - 1)
1355                 count = sizeof(buffer) - 1;
1356         if (copy_from_user(buffer, buf, count))
1357                 return -EFAULT;
1358
1359         p = get_proc_task(inode);
1360         if (!p)
1361                 return -ESRCH;
1362
1363         if (same_thread_group(current, p))
1364                 set_task_comm(p, buffer);
1365         else
1366                 count = -EINVAL;
1367
1368         put_task_struct(p);
1369
1370         return count;
1371 }
1372
1373 static int comm_show(struct seq_file *m, void *v)
1374 {
1375         struct inode *inode = m->private;
1376         struct task_struct *p;
1377
1378         p = get_proc_task(inode);
1379         if (!p)
1380                 return -ESRCH;
1381
1382         task_lock(p);
1383         seq_printf(m, "%s\n", p->comm);
1384         task_unlock(p);
1385
1386         put_task_struct(p);
1387
1388         return 0;
1389 }
1390
1391 static int comm_open(struct inode *inode, struct file *filp)
1392 {
1393         return single_open(filp, comm_show, inode);
1394 }
1395
1396 static const struct file_operations proc_pid_set_comm_operations = {
1397         .open           = comm_open,
1398         .read           = seq_read,
1399         .write          = comm_write,
1400         .llseek         = seq_lseek,
1401         .release        = single_release,
1402 };
1403
1404 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1405 {
1406         struct task_struct *task;
1407         struct mm_struct *mm;
1408         struct file *exe_file;
1409
1410         task = get_proc_task(dentry->d_inode);
1411         if (!task)
1412                 return -ENOENT;
1413         mm = get_task_mm(task);
1414         put_task_struct(task);
1415         if (!mm)
1416                 return -ENOENT;
1417         exe_file = get_mm_exe_file(mm);
1418         mmput(mm);
1419         if (exe_file) {
1420                 *exe_path = exe_file->f_path;
1421                 path_get(&exe_file->f_path);
1422                 fput(exe_file);
1423                 return 0;
1424         } else
1425                 return -ENOENT;
1426 }
1427
1428 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1429 {
1430         struct inode *inode = dentry->d_inode;
1431         int error = -EACCES;
1432
1433         /* We don't need a base pointer in the /proc filesystem */
1434         path_put(&nd->path);
1435
1436         /* Are we allowed to snoop on the tasks file descriptors? */
1437         if (!proc_fd_access_allowed(inode))
1438                 goto out;
1439
1440         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1441 out:
1442         return ERR_PTR(error);
1443 }
1444
1445 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1446 {
1447         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1448         char *pathname;
1449         int len;
1450
1451         if (!tmp)
1452                 return -ENOMEM;
1453
1454         pathname = d_path(path, tmp, PAGE_SIZE);
1455         len = PTR_ERR(pathname);
1456         if (IS_ERR(pathname))
1457                 goto out;
1458         len = tmp + PAGE_SIZE - 1 - pathname;
1459
1460         if (len > buflen)
1461                 len = buflen;
1462         if (copy_to_user(buffer, pathname, len))
1463                 len = -EFAULT;
1464  out:
1465         free_page((unsigned long)tmp);
1466         return len;
1467 }
1468
1469 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1470 {
1471         int error = -EACCES;
1472         struct inode *inode = dentry->d_inode;
1473         struct path path;
1474
1475         /* Are we allowed to snoop on the tasks file descriptors? */
1476         if (!proc_fd_access_allowed(inode))
1477                 goto out;
1478
1479         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1480         if (error)
1481                 goto out;
1482
1483         error = do_proc_readlink(&path, buffer, buflen);
1484         path_put(&path);
1485 out:
1486         return error;
1487 }
1488
1489 static const struct inode_operations proc_pid_link_inode_operations = {
1490         .readlink       = proc_pid_readlink,
1491         .follow_link    = proc_pid_follow_link,
1492         .setattr        = proc_setattr,
1493 };
1494
1495
1496 /* building an inode */
1497
1498 static int task_dumpable(struct task_struct *task)
1499 {
1500         int dumpable = 0;
1501         struct mm_struct *mm;
1502
1503         task_lock(task);
1504         mm = task->mm;
1505         if (mm)
1506                 dumpable = get_dumpable(mm);
1507         task_unlock(task);
1508         if(dumpable == 1)
1509                 return 1;
1510         return 0;
1511 }
1512
1513 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1514 {
1515         struct inode * inode;
1516         struct proc_inode *ei;
1517         const struct cred *cred;
1518
1519         /* We need a new inode */
1520
1521         inode = new_inode(sb);
1522         if (!inode)
1523                 goto out;
1524
1525         /* Common stuff */
1526         ei = PROC_I(inode);
1527         inode->i_ino = get_next_ino();
1528         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1529         inode->i_op = &proc_def_inode_operations;
1530
1531         /*
1532          * grab the reference to task.
1533          */
1534         ei->pid = get_task_pid(task, PIDTYPE_PID);
1535         if (!ei->pid)
1536                 goto out_unlock;
1537
1538         if (task_dumpable(task)) {
1539                 rcu_read_lock();
1540                 cred = __task_cred(task);
1541                 inode->i_uid = cred->euid;
1542                 inode->i_gid = cred->egid;
1543                 rcu_read_unlock();
1544         }
1545         security_task_to_inode(task, inode);
1546
1547 out:
1548         return inode;
1549
1550 out_unlock:
1551         iput(inode);
1552         return NULL;
1553 }
1554
1555 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1556 {
1557         struct inode *inode = dentry->d_inode;
1558         struct task_struct *task;
1559         const struct cred *cred;
1560         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1561
1562         generic_fillattr(inode, stat);
1563
1564         rcu_read_lock();
1565         stat->uid = 0;
1566         stat->gid = 0;
1567         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1568         if (task) {
1569                 if (!has_pid_permissions(pid, task, 2)) {
1570                         rcu_read_unlock();
1571                         /*
1572                          * This doesn't prevent learning whether PID exists,
1573                          * it only makes getattr() consistent with readdir().
1574                          */
1575                         return -ENOENT;
1576                 }
1577                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1578                     task_dumpable(task)) {
1579                         cred = __task_cred(task);
1580                         stat->uid = cred->euid;
1581                         stat->gid = cred->egid;
1582                 }
1583         }
1584         rcu_read_unlock();
1585         return 0;
1586 }
1587
1588 /* dentry stuff */
1589
1590 /*
1591  *      Exceptional case: normally we are not allowed to unhash a busy
1592  * directory. In this case, however, we can do it - no aliasing problems
1593  * due to the way we treat inodes.
1594  *
1595  * Rewrite the inode's ownerships here because the owning task may have
1596  * performed a setuid(), etc.
1597  *
1598  * Before the /proc/pid/status file was created the only way to read
1599  * the effective uid of a /process was to stat /proc/pid.  Reading
1600  * /proc/pid/status is slow enough that procps and other packages
1601  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1602  * made this apply to all per process world readable and executable
1603  * directories.
1604  */
1605 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1606 {
1607         struct inode *inode;
1608         struct task_struct *task;
1609         const struct cred *cred;
1610
1611         if (nd && nd->flags & LOOKUP_RCU)
1612                 return -ECHILD;
1613
1614         inode = dentry->d_inode;
1615         task = get_proc_task(inode);
1616
1617         if (task) {
1618                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1619                     task_dumpable(task)) {
1620                         rcu_read_lock();
1621                         cred = __task_cred(task);
1622                         inode->i_uid = cred->euid;
1623                         inode->i_gid = cred->egid;
1624                         rcu_read_unlock();
1625                 } else {
1626                         inode->i_uid = 0;
1627                         inode->i_gid = 0;
1628                 }
1629                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1630                 security_task_to_inode(task, inode);
1631                 put_task_struct(task);
1632                 return 1;
1633         }
1634         d_drop(dentry);
1635         return 0;
1636 }
1637
1638 static int pid_delete_dentry(const struct dentry * dentry)
1639 {
1640         /* Is the task we represent dead?
1641          * If so, then don't put the dentry on the lru list,
1642          * kill it immediately.
1643          */
1644         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1645 }
1646
1647 const struct dentry_operations pid_dentry_operations =
1648 {
1649         .d_revalidate   = pid_revalidate,
1650         .d_delete       = pid_delete_dentry,
1651 };
1652
1653 /* Lookups */
1654
1655 /*
1656  * Fill a directory entry.
1657  *
1658  * If possible create the dcache entry and derive our inode number and
1659  * file type from dcache entry.
1660  *
1661  * Since all of the proc inode numbers are dynamically generated, the inode
1662  * numbers do not exist until the inode is cache.  This means creating the
1663  * the dcache entry in readdir is necessary to keep the inode numbers
1664  * reported by readdir in sync with the inode numbers reported
1665  * by stat.
1666  */
1667 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1668         const char *name, int len,
1669         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1670 {
1671         struct dentry *child, *dir = filp->f_path.dentry;
1672         struct inode *inode;
1673         struct qstr qname;
1674         ino_t ino = 0;
1675         unsigned type = DT_UNKNOWN;
1676
1677         qname.name = name;
1678         qname.len  = len;
1679         qname.hash = full_name_hash(name, len);
1680
1681         child = d_lookup(dir, &qname);
1682         if (!child) {
1683                 struct dentry *new;
1684                 new = d_alloc(dir, &qname);
1685                 if (new) {
1686                         child = instantiate(dir->d_inode, new, task, ptr);
1687                         if (child)
1688                                 dput(new);
1689                         else
1690                                 child = new;
1691                 }
1692         }
1693         if (!child || IS_ERR(child) || !child->d_inode)
1694                 goto end_instantiate;
1695         inode = child->d_inode;
1696         if (inode) {
1697                 ino = inode->i_ino;
1698                 type = inode->i_mode >> 12;
1699         }
1700         dput(child);
1701 end_instantiate:
1702         if (!ino)
1703                 ino = find_inode_number(dir, &qname);
1704         if (!ino)
1705                 ino = 1;
1706         return filldir(dirent, name, len, filp->f_pos, ino, type);
1707 }
1708
1709 static unsigned name_to_int(struct dentry *dentry)
1710 {
1711         const char *name = dentry->d_name.name;
1712         int len = dentry->d_name.len;
1713         unsigned n = 0;
1714
1715         if (len > 1 && *name == '0')
1716                 goto out;
1717         while (len-- > 0) {
1718                 unsigned c = *name++ - '0';
1719                 if (c > 9)
1720                         goto out;
1721                 if (n >= (~0U-9)/10)
1722                         goto out;
1723                 n *= 10;
1724                 n += c;
1725         }
1726         return n;
1727 out:
1728         return ~0U;
1729 }
1730
1731 #define PROC_FDINFO_MAX 64
1732
1733 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1734 {
1735         struct task_struct *task = get_proc_task(inode);
1736         struct files_struct *files = NULL;
1737         struct file *file;
1738         int fd = proc_fd(inode);
1739
1740         if (task) {
1741                 files = get_files_struct(task);
1742                 put_task_struct(task);
1743         }
1744         if (files) {
1745                 /*
1746                  * We are not taking a ref to the file structure, so we must
1747                  * hold ->file_lock.
1748                  */
1749                 spin_lock(&files->file_lock);
1750                 file = fcheck_files(files, fd);
1751                 if (file) {
1752                         unsigned int f_flags;
1753                         struct fdtable *fdt;
1754
1755                         fdt = files_fdtable(files);
1756                         f_flags = file->f_flags & ~O_CLOEXEC;
1757                         if (close_on_exec(fd, fdt))
1758                                 f_flags |= O_CLOEXEC;
1759
1760                         if (path) {
1761                                 *path = file->f_path;
1762                                 path_get(&file->f_path);
1763                         }
1764                         if (info)
1765                                 snprintf(info, PROC_FDINFO_MAX,
1766                                          "pos:\t%lli\n"
1767                                          "flags:\t0%o\n",
1768                                          (long long) file->f_pos,
1769                                          f_flags);
1770                         spin_unlock(&files->file_lock);
1771                         put_files_struct(files);
1772                         return 0;
1773                 }
1774                 spin_unlock(&files->file_lock);
1775                 put_files_struct(files);
1776         }
1777         return -ENOENT;
1778 }
1779
1780 static int proc_fd_link(struct dentry *dentry, struct path *path)
1781 {
1782         return proc_fd_info(dentry->d_inode, path, NULL);
1783 }
1784
1785 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1786 {
1787         struct inode *inode;
1788         struct task_struct *task;
1789         int fd;
1790         struct files_struct *files;
1791         const struct cred *cred;
1792
1793         if (nd && nd->flags & LOOKUP_RCU)
1794                 return -ECHILD;
1795
1796         inode = dentry->d_inode;
1797         task = get_proc_task(inode);
1798         fd = proc_fd(inode);
1799
1800         if (task) {
1801                 files = get_files_struct(task);
1802                 if (files) {
1803                         rcu_read_lock();
1804                         if (fcheck_files(files, fd)) {
1805                                 rcu_read_unlock();
1806                                 put_files_struct(files);
1807                                 if (task_dumpable(task)) {
1808                                         rcu_read_lock();
1809                                         cred = __task_cred(task);
1810                                         inode->i_uid = cred->euid;
1811                                         inode->i_gid = cred->egid;
1812                                         rcu_read_unlock();
1813                                 } else {
1814                                         inode->i_uid = 0;
1815                                         inode->i_gid = 0;
1816                                 }
1817                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1818                                 security_task_to_inode(task, inode);
1819                                 put_task_struct(task);
1820                                 return 1;
1821                         }
1822                         rcu_read_unlock();
1823                         put_files_struct(files);
1824                 }
1825                 put_task_struct(task);
1826         }
1827         d_drop(dentry);
1828         return 0;
1829 }
1830
1831 static const struct dentry_operations tid_fd_dentry_operations =
1832 {
1833         .d_revalidate   = tid_fd_revalidate,
1834         .d_delete       = pid_delete_dentry,
1835 };
1836
1837 static struct dentry *proc_fd_instantiate(struct inode *dir,
1838         struct dentry *dentry, struct task_struct *task, const void *ptr)
1839 {
1840         unsigned fd = *(const unsigned *)ptr;
1841         struct file *file;
1842         struct files_struct *files;
1843         struct inode *inode;
1844         struct proc_inode *ei;
1845         struct dentry *error = ERR_PTR(-ENOENT);
1846
1847         inode = proc_pid_make_inode(dir->i_sb, task);
1848         if (!inode)
1849                 goto out;
1850         ei = PROC_I(inode);
1851         ei->fd = fd;
1852         files = get_files_struct(task);
1853         if (!files)
1854                 goto out_iput;
1855         inode->i_mode = S_IFLNK;
1856
1857         /*
1858          * We are not taking a ref to the file structure, so we must
1859          * hold ->file_lock.
1860          */
1861         spin_lock(&files->file_lock);
1862         file = fcheck_files(files, fd);
1863         if (!file)
1864                 goto out_unlock;
1865         if (file->f_mode & FMODE_READ)
1866                 inode->i_mode |= S_IRUSR | S_IXUSR;
1867         if (file->f_mode & FMODE_WRITE)
1868                 inode->i_mode |= S_IWUSR | S_IXUSR;
1869         spin_unlock(&files->file_lock);
1870         put_files_struct(files);
1871
1872         inode->i_op = &proc_pid_link_inode_operations;
1873         inode->i_size = 64;
1874         ei->op.proc_get_link = proc_fd_link;
1875         d_set_d_op(dentry, &tid_fd_dentry_operations);
1876         d_add(dentry, inode);
1877         /* Close the race of the process dying before we return the dentry */
1878         if (tid_fd_revalidate(dentry, NULL))
1879                 error = NULL;
1880
1881  out:
1882         return error;
1883 out_unlock:
1884         spin_unlock(&files->file_lock);
1885         put_files_struct(files);
1886 out_iput:
1887         iput(inode);
1888         goto out;
1889 }
1890
1891 static struct dentry *proc_lookupfd_common(struct inode *dir,
1892                                            struct dentry *dentry,
1893                                            instantiate_t instantiate)
1894 {
1895         struct task_struct *task = get_proc_task(dir);
1896         unsigned fd = name_to_int(dentry);
1897         struct dentry *result = ERR_PTR(-ENOENT);
1898
1899         if (!task)
1900                 goto out_no_task;
1901         if (fd == ~0U)
1902                 goto out;
1903
1904         result = instantiate(dir, dentry, task, &fd);
1905 out:
1906         put_task_struct(task);
1907 out_no_task:
1908         return result;
1909 }
1910
1911 static int proc_readfd_common(struct file * filp, void * dirent,
1912                               filldir_t filldir, instantiate_t instantiate)
1913 {
1914         struct dentry *dentry = filp->f_path.dentry;
1915         struct inode *inode = dentry->d_inode;
1916         struct task_struct *p = get_proc_task(inode);
1917         unsigned int fd, ino;
1918         int retval;
1919         struct files_struct * files;
1920
1921         retval = -ENOENT;
1922         if (!p)
1923                 goto out_no_task;
1924         retval = 0;
1925
1926         fd = filp->f_pos;
1927         switch (fd) {
1928                 case 0:
1929                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1930                                 goto out;
1931                         filp->f_pos++;
1932                 case 1:
1933                         ino = parent_ino(dentry);
1934                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1935                                 goto out;
1936                         filp->f_pos++;
1937                 default:
1938                         files = get_files_struct(p);
1939                         if (!files)
1940                                 goto out;
1941                         rcu_read_lock();
1942                         for (fd = filp->f_pos-2;
1943                              fd < files_fdtable(files)->max_fds;
1944                              fd++, filp->f_pos++) {
1945                                 char name[PROC_NUMBUF];
1946                                 int len;
1947
1948                                 if (!fcheck_files(files, fd))
1949                                         continue;
1950                                 rcu_read_unlock();
1951
1952                                 len = snprintf(name, sizeof(name), "%d", fd);
1953                                 if (proc_fill_cache(filp, dirent, filldir,
1954                                                     name, len, instantiate,
1955                                                     p, &fd) < 0) {
1956                                         rcu_read_lock();
1957                                         break;
1958                                 }
1959                                 rcu_read_lock();
1960                         }
1961                         rcu_read_unlock();
1962                         put_files_struct(files);
1963         }
1964 out:
1965         put_task_struct(p);
1966 out_no_task:
1967         return retval;
1968 }
1969
1970 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1971                                     struct nameidata *nd)
1972 {
1973         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1974 }
1975
1976 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1977 {
1978         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1979 }
1980
1981 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1982                                       size_t len, loff_t *ppos)
1983 {
1984         char tmp[PROC_FDINFO_MAX];
1985         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1986         if (!err)
1987                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1988         return err;
1989 }
1990
1991 static const struct file_operations proc_fdinfo_file_operations = {
1992         .open           = nonseekable_open,
1993         .read           = proc_fdinfo_read,
1994         .llseek         = no_llseek,
1995 };
1996
1997 static const struct file_operations proc_fd_operations = {
1998         .read           = generic_read_dir,
1999         .readdir        = proc_readfd,
2000         .llseek         = default_llseek,
2001 };
2002
2003 #ifdef CONFIG_CHECKPOINT_RESTORE
2004
2005 /*
2006  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2007  * which represent vma start and end addresses.
2008  */
2009 static int dname_to_vma_addr(struct dentry *dentry,
2010                              unsigned long *start, unsigned long *end)
2011 {
2012         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2013                 return -EINVAL;
2014
2015         return 0;
2016 }
2017
2018 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2019 {
2020         unsigned long vm_start, vm_end;
2021         bool exact_vma_exists = false;
2022         struct mm_struct *mm = NULL;
2023         struct task_struct *task;
2024         const struct cred *cred;
2025         struct inode *inode;
2026         int status = 0;
2027
2028         if (nd && nd->flags & LOOKUP_RCU)
2029                 return -ECHILD;
2030
2031         if (!capable(CAP_SYS_ADMIN)) {
2032                 status = -EACCES;
2033                 goto out_notask;
2034         }
2035
2036         inode = dentry->d_inode;
2037         task = get_proc_task(inode);
2038         if (!task)
2039                 goto out_notask;
2040
2041         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2042                 goto out;
2043
2044         mm = get_task_mm(task);
2045         if (!mm)
2046                 goto out;
2047
2048         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2049                 down_read(&mm->mmap_sem);
2050                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2051                 up_read(&mm->mmap_sem);
2052         }
2053
2054         mmput(mm);
2055
2056         if (exact_vma_exists) {
2057                 if (task_dumpable(task)) {
2058                         rcu_read_lock();
2059                         cred = __task_cred(task);
2060                         inode->i_uid = cred->euid;
2061                         inode->i_gid = cred->egid;
2062                         rcu_read_unlock();
2063                 } else {
2064                         inode->i_uid = 0;
2065                         inode->i_gid = 0;
2066                 }
2067                 security_task_to_inode(task, inode);
2068                 status = 1;
2069         }
2070
2071 out:
2072         put_task_struct(task);
2073
2074 out_notask:
2075         if (status <= 0)
2076                 d_drop(dentry);
2077
2078         return status;
2079 }
2080
2081 static const struct dentry_operations tid_map_files_dentry_operations = {
2082         .d_revalidate   = map_files_d_revalidate,
2083         .d_delete       = pid_delete_dentry,
2084 };
2085
2086 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2087 {
2088         unsigned long vm_start, vm_end;
2089         struct vm_area_struct *vma;
2090         struct task_struct *task;
2091         struct mm_struct *mm;
2092         int rc;
2093
2094         rc = -ENOENT;
2095         task = get_proc_task(dentry->d_inode);
2096         if (!task)
2097                 goto out;
2098
2099         mm = get_task_mm(task);
2100         put_task_struct(task);
2101         if (!mm)
2102                 goto out;
2103
2104         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2105         if (rc)
2106                 goto out_mmput;
2107
2108         down_read(&mm->mmap_sem);
2109         vma = find_exact_vma(mm, vm_start, vm_end);
2110         if (vma && vma->vm_file) {
2111                 *path = vma->vm_file->f_path;
2112                 path_get(path);
2113                 rc = 0;
2114         }
2115         up_read(&mm->mmap_sem);
2116
2117 out_mmput:
2118         mmput(mm);
2119 out:
2120         return rc;
2121 }
2122
2123 struct map_files_info {
2124         struct file     *file;
2125         unsigned long   len;
2126         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2127 };
2128
2129 static struct dentry *
2130 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2131                            struct task_struct *task, const void *ptr)
2132 {
2133         const struct file *file = ptr;
2134         struct proc_inode *ei;
2135         struct inode *inode;
2136
2137         if (!file)
2138                 return ERR_PTR(-ENOENT);
2139
2140         inode = proc_pid_make_inode(dir->i_sb, task);
2141         if (!inode)
2142                 return ERR_PTR(-ENOENT);
2143
2144         ei = PROC_I(inode);
2145         ei->op.proc_get_link = proc_map_files_get_link;
2146
2147         inode->i_op = &proc_pid_link_inode_operations;
2148         inode->i_size = 64;
2149         inode->i_mode = S_IFLNK;
2150
2151         if (file->f_mode & FMODE_READ)
2152                 inode->i_mode |= S_IRUSR;
2153         if (file->f_mode & FMODE_WRITE)
2154                 inode->i_mode |= S_IWUSR;
2155
2156         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2157         d_add(dentry, inode);
2158
2159         return NULL;
2160 }
2161
2162 static struct dentry *proc_map_files_lookup(struct inode *dir,
2163                 struct dentry *dentry, struct nameidata *nd)
2164 {
2165         unsigned long vm_start, vm_end;
2166         struct vm_area_struct *vma;
2167         struct task_struct *task;
2168         struct dentry *result;
2169         struct mm_struct *mm;
2170
2171         result = ERR_PTR(-EACCES);
2172         if (!capable(CAP_SYS_ADMIN))
2173                 goto out;
2174
2175         result = ERR_PTR(-ENOENT);
2176         task = get_proc_task(dir);
2177         if (!task)
2178                 goto out;
2179
2180         result = ERR_PTR(-EACCES);
2181         if (lock_trace(task))
2182                 goto out_put_task;
2183
2184         result = ERR_PTR(-ENOENT);
2185         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2186                 goto out_unlock;
2187
2188         mm = get_task_mm(task);
2189         if (!mm)
2190                 goto out_unlock;
2191
2192         down_read(&mm->mmap_sem);
2193         vma = find_exact_vma(mm, vm_start, vm_end);
2194         if (!vma)
2195                 goto out_no_vma;
2196
2197         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2198
2199 out_no_vma:
2200         up_read(&mm->mmap_sem);
2201         mmput(mm);
2202 out_unlock:
2203         unlock_trace(task);
2204 out_put_task:
2205         put_task_struct(task);
2206 out:
2207         return result;
2208 }
2209
2210 static const struct inode_operations proc_map_files_inode_operations = {
2211         .lookup         = proc_map_files_lookup,
2212         .permission     = proc_fd_permission,
2213         .setattr        = proc_setattr,
2214 };
2215
2216 static int
2217 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2218 {
2219         struct dentry *dentry = filp->f_path.dentry;
2220         struct inode *inode = dentry->d_inode;
2221         struct vm_area_struct *vma;
2222         struct task_struct *task;
2223         struct mm_struct *mm;
2224         ino_t ino;
2225         int ret;
2226
2227         ret = -EACCES;
2228         if (!capable(CAP_SYS_ADMIN))
2229                 goto out;
2230
2231         ret = -ENOENT;
2232         task = get_proc_task(inode);
2233         if (!task)
2234                 goto out;
2235
2236         ret = -EACCES;
2237         if (lock_trace(task))
2238                 goto out_put_task;
2239
2240         ret = 0;
2241         switch (filp->f_pos) {
2242         case 0:
2243                 ino = inode->i_ino;
2244                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2245                         goto out_unlock;
2246                 filp->f_pos++;
2247         case 1:
2248                 ino = parent_ino(dentry);
2249                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2250                         goto out_unlock;
2251                 filp->f_pos++;
2252         default:
2253         {
2254                 unsigned long nr_files, pos, i;
2255                 struct flex_array *fa = NULL;
2256                 struct map_files_info info;
2257                 struct map_files_info *p;
2258
2259                 mm = get_task_mm(task);
2260                 if (!mm)
2261                         goto out_unlock;
2262                 down_read(&mm->mmap_sem);
2263
2264                 nr_files = 0;
2265
2266                 /*
2267                  * We need two passes here:
2268                  *
2269                  *  1) Collect vmas of mapped files with mmap_sem taken
2270                  *  2) Release mmap_sem and instantiate entries
2271                  *
2272                  * otherwise we get lockdep complained, since filldir()
2273                  * routine might require mmap_sem taken in might_fault().
2274                  */
2275
2276                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2277                         if (vma->vm_file && ++pos > filp->f_pos)
2278                                 nr_files++;
2279                 }
2280
2281                 if (nr_files) {
2282                         fa = flex_array_alloc(sizeof(info), nr_files,
2283                                                 GFP_KERNEL);
2284                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2285                                                         GFP_KERNEL)) {
2286                                 ret = -ENOMEM;
2287                                 if (fa)
2288                                         flex_array_free(fa);
2289                                 up_read(&mm->mmap_sem);
2290                                 mmput(mm);
2291                                 goto out_unlock;
2292                         }
2293                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2294                                         vma = vma->vm_next) {
2295                                 if (!vma->vm_file)
2296                                         continue;
2297                                 if (++pos <= filp->f_pos)
2298                                         continue;
2299
2300                                 get_file(vma->vm_file);
2301                                 info.file = vma->vm_file;
2302                                 info.len = snprintf(info.name,
2303                                                 sizeof(info.name), "%lx-%lx",
2304                                                 vma->vm_start, vma->vm_end);
2305                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2306                                         BUG();
2307                         }
2308                 }
2309                 up_read(&mm->mmap_sem);
2310
2311                 for (i = 0; i < nr_files; i++) {
2312                         p = flex_array_get(fa, i);
2313                         ret = proc_fill_cache(filp, dirent, filldir,
2314                                               p->name, p->len,
2315                                               proc_map_files_instantiate,
2316                                               task, p->file);
2317                         if (ret)
2318                                 break;
2319                         filp->f_pos++;
2320                         fput(p->file);
2321                 }
2322                 for (; i < nr_files; i++) {
2323                         /*
2324                          * In case of error don't forget
2325                          * to put rest of file refs.
2326                          */
2327                         p = flex_array_get(fa, i);
2328                         fput(p->file);
2329                 }
2330                 if (fa)
2331                         flex_array_free(fa);
2332                 mmput(mm);
2333         }
2334         }
2335
2336 out_unlock:
2337         unlock_trace(task);
2338 out_put_task:
2339         put_task_struct(task);
2340 out:
2341         return ret;
2342 }
2343
2344 static const struct file_operations proc_map_files_operations = {
2345         .read           = generic_read_dir,
2346         .readdir        = proc_map_files_readdir,
2347         .llseek         = default_llseek,
2348 };
2349
2350 #endif /* CONFIG_CHECKPOINT_RESTORE */
2351
2352 /*
2353  * /proc/pid/fd needs a special permission handler so that a process can still
2354  * access /proc/self/fd after it has executed a setuid().
2355  */
2356 static int proc_fd_permission(struct inode *inode, int mask)
2357 {
2358         int rv = generic_permission(inode, mask);
2359         if (rv == 0)
2360                 return 0;
2361         if (task_pid(current) == proc_pid(inode))
2362                 rv = 0;
2363         return rv;
2364 }
2365
2366 /*
2367  * proc directories can do almost nothing..
2368  */
2369 static const struct inode_operations proc_fd_inode_operations = {
2370         .lookup         = proc_lookupfd,
2371         .permission     = proc_fd_permission,
2372         .setattr        = proc_setattr,
2373 };
2374
2375 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2376         struct dentry *dentry, struct task_struct *task, const void *ptr)
2377 {
2378         unsigned fd = *(unsigned *)ptr;
2379         struct inode *inode;
2380         struct proc_inode *ei;
2381         struct dentry *error = ERR_PTR(-ENOENT);
2382
2383         inode = proc_pid_make_inode(dir->i_sb, task);
2384         if (!inode)
2385                 goto out;
2386         ei = PROC_I(inode);
2387         ei->fd = fd;
2388         inode->i_mode = S_IFREG | S_IRUSR;
2389         inode->i_fop = &proc_fdinfo_file_operations;
2390         d_set_d_op(dentry, &tid_fd_dentry_operations);
2391         d_add(dentry, inode);
2392         /* Close the race of the process dying before we return the dentry */
2393         if (tid_fd_revalidate(dentry, NULL))
2394                 error = NULL;
2395
2396  out:
2397         return error;
2398 }
2399
2400 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2401                                         struct dentry *dentry,
2402                                         struct nameidata *nd)
2403 {
2404         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2405 }
2406
2407 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2408 {
2409         return proc_readfd_common(filp, dirent, filldir,
2410                                   proc_fdinfo_instantiate);
2411 }
2412
2413 static const struct file_operations proc_fdinfo_operations = {
2414         .read           = generic_read_dir,
2415         .readdir        = proc_readfdinfo,
2416         .llseek         = default_llseek,
2417 };
2418
2419 /*
2420  * proc directories can do almost nothing..
2421  */
2422 static const struct inode_operations proc_fdinfo_inode_operations = {
2423         .lookup         = proc_lookupfdinfo,
2424         .setattr        = proc_setattr,
2425 };
2426
2427
2428 static struct dentry *proc_pident_instantiate(struct inode *dir,
2429         struct dentry *dentry, struct task_struct *task, const void *ptr)
2430 {
2431         const struct pid_entry *p = ptr;
2432         struct inode *inode;
2433         struct proc_inode *ei;
2434         struct dentry *error = ERR_PTR(-ENOENT);
2435
2436         inode = proc_pid_make_inode(dir->i_sb, task);
2437         if (!inode)
2438                 goto out;
2439
2440         ei = PROC_I(inode);
2441         inode->i_mode = p->mode;
2442         if (S_ISDIR(inode->i_mode))
2443                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2444         if (p->iop)
2445                 inode->i_op = p->iop;
2446         if (p->fop)
2447                 inode->i_fop = p->fop;
2448         ei->op = p->op;
2449         d_set_d_op(dentry, &pid_dentry_operations);
2450         d_add(dentry, inode);
2451         /* Close the race of the process dying before we return the dentry */
2452         if (pid_revalidate(dentry, NULL))
2453                 error = NULL;
2454 out:
2455         return error;
2456 }
2457
2458 static struct dentry *proc_pident_lookup(struct inode *dir, 
2459                                          struct dentry *dentry,
2460                                          const struct pid_entry *ents,
2461                                          unsigned int nents)
2462 {
2463         struct dentry *error;
2464         struct task_struct *task = get_proc_task(dir);
2465         const struct pid_entry *p, *last;
2466
2467         error = ERR_PTR(-ENOENT);
2468
2469         if (!task)
2470                 goto out_no_task;
2471
2472         /*
2473          * Yes, it does not scale. And it should not. Don't add
2474          * new entries into /proc/<tgid>/ without very good reasons.
2475          */
2476         last = &ents[nents - 1];
2477         for (p = ents; p <= last; p++) {
2478                 if (p->len != dentry->d_name.len)
2479                         continue;
2480                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2481                         break;
2482         }
2483         if (p > last)
2484                 goto out;
2485
2486         error = proc_pident_instantiate(dir, dentry, task, p);
2487 out:
2488         put_task_struct(task);
2489 out_no_task:
2490         return error;
2491 }
2492
2493 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2494         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2495 {
2496         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2497                                 proc_pident_instantiate, task, p);
2498 }
2499
2500 static int proc_pident_readdir(struct file *filp,
2501                 void *dirent, filldir_t filldir,
2502                 const struct pid_entry *ents, unsigned int nents)
2503 {
2504         int i;
2505         struct dentry *dentry = filp->f_path.dentry;
2506         struct inode *inode = dentry->d_inode;
2507         struct task_struct *task = get_proc_task(inode);
2508         const struct pid_entry *p, *last;
2509         ino_t ino;
2510         int ret;
2511
2512         ret = -ENOENT;
2513         if (!task)
2514                 goto out_no_task;
2515
2516         ret = 0;
2517         i = filp->f_pos;
2518         switch (i) {
2519         case 0:
2520                 ino = inode->i_ino;
2521                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2522                         goto out;
2523                 i++;
2524                 filp->f_pos++;
2525                 /* fall through */
2526         case 1:
2527                 ino = parent_ino(dentry);
2528                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2529                         goto out;
2530                 i++;
2531                 filp->f_pos++;
2532                 /* fall through */
2533         default:
2534                 i -= 2;
2535                 if (i >= nents) {
2536                         ret = 1;
2537                         goto out;
2538                 }
2539                 p = ents + i;
2540                 last = &ents[nents - 1];
2541                 while (p <= last) {
2542                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2543                                 goto out;
2544                         filp->f_pos++;
2545                         p++;
2546                 }
2547         }
2548
2549         ret = 1;
2550 out:
2551         put_task_struct(task);
2552 out_no_task:
2553         return ret;
2554 }
2555
2556 #ifdef CONFIG_SECURITY
2557 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2558                                   size_t count, loff_t *ppos)
2559 {
2560         struct inode * inode = file->f_path.dentry->d_inode;
2561         char *p = NULL;
2562         ssize_t length;
2563         struct task_struct *task = get_proc_task(inode);
2564
2565         if (!task)
2566                 return -ESRCH;
2567
2568         length = security_getprocattr(task,
2569                                       (char*)file->f_path.dentry->d_name.name,
2570                                       &p);
2571         put_task_struct(task);
2572         if (length > 0)
2573                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2574         kfree(p);
2575         return length;
2576 }
2577
2578 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2579                                    size_t count, loff_t *ppos)
2580 {
2581         struct inode * inode = file->f_path.dentry->d_inode;
2582         char *page;
2583         ssize_t length;
2584         struct task_struct *task = get_proc_task(inode);
2585
2586         length = -ESRCH;
2587         if (!task)
2588                 goto out_no_task;
2589         if (count > PAGE_SIZE)
2590                 count = PAGE_SIZE;
2591
2592         /* No partial writes. */
2593         length = -EINVAL;
2594         if (*ppos != 0)
2595                 goto out;
2596
2597         length = -ENOMEM;
2598         page = (char*)__get_free_page(GFP_TEMPORARY);
2599         if (!page)
2600                 goto out;
2601
2602         length = -EFAULT;
2603         if (copy_from_user(page, buf, count))
2604                 goto out_free;
2605
2606         /* Guard against adverse ptrace interaction */
2607         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2608         if (length < 0)
2609                 goto out_free;
2610
2611         length = security_setprocattr(task,
2612                                       (char*)file->f_path.dentry->d_name.name,
2613                                       (void*)page, count);
2614         mutex_unlock(&task->signal->cred_guard_mutex);
2615 out_free:
2616         free_page((unsigned long) page);
2617 out:
2618         put_task_struct(task);
2619 out_no_task:
2620         return length;
2621 }
2622
2623 static const struct file_operations proc_pid_attr_operations = {
2624         .read           = proc_pid_attr_read,
2625         .write          = proc_pid_attr_write,
2626         .llseek         = generic_file_llseek,
2627 };
2628
2629 static const struct pid_entry attr_dir_stuff[] = {
2630         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2631         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2632         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2633         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2634         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2635         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2636 };
2637
2638 static int proc_attr_dir_readdir(struct file * filp,
2639                              void * dirent, filldir_t filldir)
2640 {
2641         return proc_pident_readdir(filp,dirent,filldir,
2642                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2643 }
2644
2645 static const struct file_operations proc_attr_dir_operations = {
2646         .read           = generic_read_dir,
2647         .readdir        = proc_attr_dir_readdir,
2648         .llseek         = default_llseek,
2649 };
2650
2651 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2652                                 struct dentry *dentry, struct nameidata *nd)
2653 {
2654         return proc_pident_lookup(dir, dentry,
2655                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2656 }
2657
2658 static const struct inode_operations proc_attr_dir_inode_operations = {
2659         .lookup         = proc_attr_dir_lookup,
2660         .getattr        = pid_getattr,
2661         .setattr        = proc_setattr,
2662 };
2663
2664 #endif
2665
2666 #ifdef CONFIG_ELF_CORE
2667 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2668                                          size_t count, loff_t *ppos)
2669 {
2670         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2671         struct mm_struct *mm;
2672         char buffer[PROC_NUMBUF];
2673         size_t len;
2674         int ret;
2675
2676         if (!task)
2677                 return -ESRCH;
2678
2679         ret = 0;
2680         mm = get_task_mm(task);
2681         if (mm) {
2682                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2683                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2684                                 MMF_DUMP_FILTER_SHIFT));
2685                 mmput(mm);
2686                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2687         }
2688
2689         put_task_struct(task);
2690
2691         return ret;
2692 }
2693
2694 static ssize_t proc_coredump_filter_write(struct file *file,
2695                                           const char __user *buf,
2696                                           size_t count,
2697                                           loff_t *ppos)
2698 {
2699         struct task_struct *task;
2700         struct mm_struct *mm;
2701         char buffer[PROC_NUMBUF], *end;
2702         unsigned int val;
2703         int ret;
2704         int i;
2705         unsigned long mask;
2706
2707         ret = -EFAULT;
2708         memset(buffer, 0, sizeof(buffer));
2709         if (count > sizeof(buffer) - 1)
2710                 count = sizeof(buffer) - 1;
2711         if (copy_from_user(buffer, buf, count))
2712                 goto out_no_task;
2713
2714         ret = -EINVAL;
2715         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2716         if (*end == '\n')
2717                 end++;
2718         if (end - buffer == 0)
2719                 goto out_no_task;
2720
2721         ret = -ESRCH;
2722         task = get_proc_task(file->f_dentry->d_inode);
2723         if (!task)
2724                 goto out_no_task;
2725
2726         ret = end - buffer;
2727         mm = get_task_mm(task);
2728         if (!mm)
2729                 goto out_no_mm;
2730
2731         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2732                 if (val & mask)
2733                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2734                 else
2735                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2736         }
2737
2738         mmput(mm);
2739  out_no_mm:
2740         put_task_struct(task);
2741  out_no_task:
2742         return ret;
2743 }
2744
2745 static const struct file_operations proc_coredump_filter_operations = {
2746         .read           = proc_coredump_filter_read,
2747         .write          = proc_coredump_filter_write,
2748         .llseek         = generic_file_llseek,
2749 };
2750 #endif
2751
2752 /*
2753  * /proc/self:
2754  */
2755 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2756                               int buflen)
2757 {
2758         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2759         pid_t tgid = task_tgid_nr_ns(current, ns);
2760         char tmp[PROC_NUMBUF];
2761         if (!tgid)
2762                 return -ENOENT;
2763         sprintf(tmp, "%d", tgid);
2764         return vfs_readlink(dentry,buffer,buflen,tmp);
2765 }
2766
2767 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2768 {
2769         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2770         pid_t tgid = task_tgid_nr_ns(current, ns);
2771         char *name = ERR_PTR(-ENOENT);
2772         if (tgid) {
2773                 name = __getname();
2774                 if (!name)
2775                         name = ERR_PTR(-ENOMEM);
2776                 else
2777                         sprintf(name, "%d", tgid);
2778         }
2779         nd_set_link(nd, name);
2780         return NULL;
2781 }
2782
2783 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2784                                 void *cookie)
2785 {
2786         char *s = nd_get_link(nd);
2787         if (!IS_ERR(s))
2788                 __putname(s);
2789 }
2790
2791 static const struct inode_operations proc_self_inode_operations = {
2792         .readlink       = proc_self_readlink,
2793         .follow_link    = proc_self_follow_link,
2794         .put_link       = proc_self_put_link,
2795 };
2796
2797 /*
2798  * proc base
2799  *
2800  * These are the directory entries in the root directory of /proc
2801  * that properly belong to the /proc filesystem, as they describe
2802  * describe something that is process related.
2803  */
2804 static const struct pid_entry proc_base_stuff[] = {
2805         NOD("self", S_IFLNK|S_IRWXUGO,
2806                 &proc_self_inode_operations, NULL, {}),
2807 };
2808
2809 static struct dentry *proc_base_instantiate(struct inode *dir,
2810         struct dentry *dentry, struct task_struct *task, const void *ptr)
2811 {
2812         const struct pid_entry *p = ptr;
2813         struct inode *inode;
2814         struct proc_inode *ei;
2815         struct dentry *error;
2816
2817         /* Allocate the inode */
2818         error = ERR_PTR(-ENOMEM);
2819         inode = new_inode(dir->i_sb);
2820         if (!inode)
2821                 goto out;
2822
2823         /* Initialize the inode */
2824         ei = PROC_I(inode);
2825         inode->i_ino = get_next_ino();
2826         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2827
2828         /*
2829          * grab the reference to the task.
2830          */
2831         ei->pid = get_task_pid(task, PIDTYPE_PID);
2832         if (!ei->pid)
2833                 goto out_iput;
2834
2835         inode->i_mode = p->mode;
2836         if (S_ISDIR(inode->i_mode))
2837                 set_nlink(inode, 2);
2838         if (S_ISLNK(inode->i_mode))
2839                 inode->i_size = 64;
2840         if (p->iop)
2841                 inode->i_op = p->iop;
2842         if (p->fop)
2843                 inode->i_fop = p->fop;
2844         ei->op = p->op;
2845         d_add(dentry, inode);
2846         error = NULL;
2847 out:
2848         return error;
2849 out_iput:
2850         iput(inode);
2851         goto out;
2852 }
2853
2854 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2855 {
2856         struct dentry *error;
2857         struct task_struct *task = get_proc_task(dir);
2858         const struct pid_entry *p, *last;
2859
2860         error = ERR_PTR(-ENOENT);
2861
2862         if (!task)
2863                 goto out_no_task;
2864
2865         /* Lookup the directory entry */
2866         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2867         for (p = proc_base_stuff; p <= last; p++) {
2868                 if (p->len != dentry->d_name.len)
2869                         continue;
2870                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2871                         break;
2872         }
2873         if (p > last)
2874                 goto out;
2875
2876         error = proc_base_instantiate(dir, dentry, task, p);
2877
2878 out:
2879         put_task_struct(task);
2880 out_no_task:
2881         return error;
2882 }
2883
2884 static int proc_base_fill_cache(struct file *filp, void *dirent,
2885         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2886 {
2887         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2888                                 proc_base_instantiate, task, p);
2889 }
2890
2891 #ifdef CONFIG_TASK_IO_ACCOUNTING
2892 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2893 {
2894         struct task_io_accounting acct = task->ioac;
2895         unsigned long flags;
2896         int result;
2897
2898         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2899         if (result)
2900                 return result;
2901
2902         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2903                 result = -EACCES;
2904                 goto out_unlock;
2905         }
2906
2907         if (whole && lock_task_sighand(task, &flags)) {
2908                 struct task_struct *t = task;
2909
2910                 task_io_accounting_add(&acct, &task->signal->ioac);
2911                 while_each_thread(task, t)
2912                         task_io_accounting_add(&acct, &t->ioac);
2913
2914                 unlock_task_sighand(task, &flags);
2915         }
2916         result = sprintf(buffer,
2917                         "rchar: %llu\n"
2918                         "wchar: %llu\n"
2919                         "syscr: %llu\n"
2920                         "syscw: %llu\n"
2921                         "read_bytes: %llu\n"
2922                         "write_bytes: %llu\n"
2923                         "cancelled_write_bytes: %llu\n",
2924                         (unsigned long long)acct.rchar,
2925                         (unsigned long long)acct.wchar,
2926                         (unsigned long long)acct.syscr,
2927                         (unsigned long long)acct.syscw,
2928                         (unsigned long long)acct.read_bytes,
2929                         (unsigned long long)acct.write_bytes,
2930                         (unsigned long long)acct.cancelled_write_bytes);
2931 out_unlock:
2932         mutex_unlock(&task->signal->cred_guard_mutex);
2933         return result;
2934 }
2935
2936 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2937 {
2938         return do_io_accounting(task, buffer, 0);
2939 }
2940
2941 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2942 {
2943         return do_io_accounting(task, buffer, 1);
2944 }
2945 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2946
2947 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2948                                 struct pid *pid, struct task_struct *task)
2949 {
2950         int err = lock_trace(task);
2951         if (!err) {
2952                 seq_printf(m, "%08x\n", task->personality);
2953                 unlock_trace(task);
2954         }
2955         return err;
2956 }
2957
2958 /*
2959  * Thread groups
2960  */
2961 static const struct file_operations proc_task_operations;
2962 static const struct inode_operations proc_task_inode_operations;
2963
2964 static const struct pid_entry tgid_base_stuff[] = {
2965         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2966         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2967 #ifdef CONFIG_CHECKPOINT_RESTORE
2968         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2969 #endif
2970         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2971         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2972 #ifdef CONFIG_NET
2973         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2974 #endif
2975         REG("environ",    S_IRUSR, proc_environ_operations),
2976         INF("auxv",       S_IRUSR, proc_pid_auxv),
2977         ONE("status",     S_IRUGO, proc_pid_status),
2978         ONE("personality", S_IRUGO, proc_pid_personality),
2979         INF("limits",     S_IRUGO, proc_pid_limits),
2980 #ifdef CONFIG_SCHED_DEBUG
2981         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2982 #endif
2983 #ifdef CONFIG_SCHED_AUTOGROUP
2984         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2985 #endif
2986         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2987 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2988         INF("syscall",    S_IRUGO, proc_pid_syscall),
2989 #endif
2990         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2991         ONE("stat",       S_IRUGO, proc_tgid_stat),
2992         ONE("statm",      S_IRUGO, proc_pid_statm),
2993         REG("maps",       S_IRUGO, proc_maps_operations),
2994 #ifdef CONFIG_NUMA
2995         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2996 #endif
2997         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2998         LNK("cwd",        proc_cwd_link),
2999         LNK("root",       proc_root_link),
3000         LNK("exe",        proc_exe_link),
3001         REG("mounts",     S_IRUGO, proc_mounts_operations),
3002         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3003         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3004 #ifdef CONFIG_PROC_PAGE_MONITOR
3005         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3006         REG("smaps",      S_IRUGO, proc_smaps_operations),
3007         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3008 #endif
3009 #ifdef CONFIG_SECURITY
3010         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3011 #endif
3012 #ifdef CONFIG_KALLSYMS
3013         INF("wchan",      S_IRUGO, proc_pid_wchan),
3014 #endif
3015 #ifdef CONFIG_STACKTRACE
3016         ONE("stack",      S_IRUGO, proc_pid_stack),
3017 #endif
3018 #ifdef CONFIG_SCHEDSTATS
3019         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3020 #endif
3021 #ifdef CONFIG_LATENCYTOP
3022         REG("latency",  S_IRUGO, proc_lstats_operations),
3023 #endif
3024 #ifdef CONFIG_PROC_PID_CPUSET
3025         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3026 #endif
3027 #ifdef CONFIG_CGROUPS
3028         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3029 #endif
3030         INF("oom_score",  S_IRUGO, proc_oom_score),
3031         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3032         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3033 #ifdef CONFIG_AUDITSYSCALL
3034         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3035         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3036 #endif
3037 #ifdef CONFIG_FAULT_INJECTION
3038         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3039 #endif
3040 #ifdef CONFIG_ELF_CORE
3041         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3042 #endif
3043 #ifdef CONFIG_TASK_IO_ACCOUNTING
3044         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3045 #endif
3046 #ifdef CONFIG_HARDWALL
3047         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3048 #endif
3049 };
3050
3051 static int proc_tgid_base_readdir(struct file * filp,
3052                              void * dirent, filldir_t filldir)
3053 {
3054         return proc_pident_readdir(filp,dirent,filldir,
3055                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3056 }
3057
3058 static const struct file_operations proc_tgid_base_operations = {
3059         .read           = generic_read_dir,
3060         .readdir        = proc_tgid_base_readdir,
3061         .llseek         = default_llseek,
3062 };
3063
3064 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3065         return proc_pident_lookup(dir, dentry,
3066                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3067 }
3068
3069 static const struct inode_operations proc_tgid_base_inode_operations = {
3070         .lookup         = proc_tgid_base_lookup,
3071         .getattr        = pid_getattr,
3072         .setattr        = proc_setattr,
3073         .permission     = proc_pid_permission,
3074 };
3075
3076 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3077 {
3078         struct dentry *dentry, *leader, *dir;
3079         char buf[PROC_NUMBUF];
3080         struct qstr name;
3081
3082         name.name = buf;
3083         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3084         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3085         if (dentry) {
3086                 shrink_dcache_parent(dentry);
3087                 d_drop(dentry);
3088                 dput(dentry);
3089         }
3090
3091         name.name = buf;
3092         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3093         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3094         if (!leader)
3095                 goto out;
3096
3097         name.name = "task";
3098         name.len = strlen(name.name);
3099         dir = d_hash_and_lookup(leader, &name);
3100         if (!dir)
3101                 goto out_put_leader;
3102
3103         name.name = buf;
3104         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3105         dentry = d_hash_and_lookup(dir, &name);
3106         if (dentry) {
3107                 shrink_dcache_parent(dentry);
3108                 d_drop(dentry);
3109                 dput(dentry);
3110         }
3111
3112         dput(dir);
3113 out_put_leader:
3114         dput(leader);
3115 out:
3116         return;
3117 }
3118
3119 /**
3120  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3121  * @task: task that should be flushed.
3122  *
3123  * When flushing dentries from proc, one needs to flush them from global
3124  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3125  * in. This call is supposed to do all of this job.
3126  *
3127  * Looks in the dcache for
3128  * /proc/@pid
3129  * /proc/@tgid/task/@pid
3130  * if either directory is present flushes it and all of it'ts children
3131  * from the dcache.
3132  *
3133  * It is safe and reasonable to cache /proc entries for a task until
3134  * that task exits.  After that they just clog up the dcache with
3135  * useless entries, possibly causing useful dcache entries to be
3136  * flushed instead.  This routine is proved to flush those useless
3137  * dcache entries at process exit time.
3138  *
3139  * NOTE: This routine is just an optimization so it does not guarantee
3140  *       that no dcache entries will exist at process exit time it
3141  *       just makes it very unlikely that any will persist.
3142  */
3143
3144 void proc_flush_task(struct task_struct *task)
3145 {
3146         int i;
3147         struct pid *pid, *tgid;
3148         struct upid *upid;
3149
3150         pid = task_pid(task);
3151         tgid = task_tgid(task);
3152
3153         for (i = 0; i <= pid->level; i++) {
3154                 upid = &pid->numbers[i];
3155                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3156                                         tgid->numbers[i].nr);
3157         }
3158
3159         upid = &pid->numbers[pid->level];
3160         if (upid->nr == 1)
3161                 pid_ns_release_proc(upid->ns);
3162 }
3163
3164 static struct dentry *proc_pid_instantiate(struct inode *dir,
3165                                            struct dentry * dentry,
3166                                            struct task_struct *task, const void *ptr)
3167 {
3168         struct dentry *error = ERR_PTR(-ENOENT);
3169         struct inode *inode;
3170
3171         inode = proc_pid_make_inode(dir->i_sb, task);
3172         if (!inode)
3173                 goto out;
3174
3175         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3176         inode->i_op = &proc_tgid_base_inode_operations;
3177         inode->i_fop = &proc_tgid_base_operations;
3178         inode->i_flags|=S_IMMUTABLE;
3179
3180         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3181                                                   ARRAY_SIZE(tgid_base_stuff)));
3182
3183         d_set_d_op(dentry, &pid_dentry_operations);
3184
3185         d_add(dentry, inode);
3186         /* Close the race of the process dying before we return the dentry */
3187         if (pid_revalidate(dentry, NULL))
3188                 error = NULL;
3189 out:
3190         return error;
3191 }
3192
3193 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3194 {
3195         struct dentry *result;
3196         struct task_struct *task;
3197         unsigned tgid;
3198         struct pid_namespace *ns;
3199
3200         result = proc_base_lookup(dir, dentry);
3201         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3202                 goto out;
3203
3204         tgid = name_to_int(dentry);
3205         if (tgid == ~0U)
3206                 goto out;
3207
3208         ns = dentry->d_sb->s_fs_info;
3209         rcu_read_lock();
3210         task = find_task_by_pid_ns(tgid, ns);
3211         if (task)
3212                 get_task_struct(task);
3213         rcu_read_unlock();
3214         if (!task)
3215                 goto out;
3216
3217         result = proc_pid_instantiate(dir, dentry, task, NULL);
3218         put_task_struct(task);
3219 out:
3220         return result;
3221 }
3222
3223 /*
3224  * Find the first task with tgid >= tgid
3225  *
3226  */
3227 struct tgid_iter {
3228         unsigned int tgid;
3229         struct task_struct *task;
3230 };
3231 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3232 {
3233         struct pid *pid;
3234
3235         if (iter.task)
3236                 put_task_struct(iter.task);
3237         rcu_read_lock();
3238 retry:
3239         iter.task = NULL;
3240         pid = find_ge_pid(iter.tgid, ns);
3241         if (pid) {
3242                 iter.tgid = pid_nr_ns(pid, ns);
3243                 iter.task = pid_task(pid, PIDTYPE_PID);
3244                 /* What we to know is if the pid we have find is the
3245                  * pid of a thread_group_leader.  Testing for task
3246                  * being a thread_group_leader is the obvious thing
3247                  * todo but there is a window when it fails, due to
3248                  * the pid transfer logic in de_thread.
3249                  *
3250                  * So we perform the straight forward test of seeing
3251                  * if the pid we have found is the pid of a thread
3252                  * group leader, and don't worry if the task we have
3253                  * found doesn't happen to be a thread group leader.
3254                  * As we don't care in the case of readdir.
3255                  */
3256                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3257                         iter.tgid += 1;
3258                         goto retry;
3259                 }
3260                 get_task_struct(iter.task);
3261         }
3262         rcu_read_unlock();
3263         return iter;
3264 }
3265
3266 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3267
3268 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3269         struct tgid_iter iter)
3270 {
3271         char name[PROC_NUMBUF];
3272         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3273         return proc_fill_cache(filp, dirent, filldir, name, len,
3274                                 proc_pid_instantiate, iter.task, NULL);
3275 }
3276
3277 static int fake_filldir(void *buf, const char *name, int namelen,
3278                         loff_t offset, u64 ino, unsigned d_type)
3279 {
3280         return 0;
3281 }
3282
3283 /* for the /proc/ directory itself, after non-process stuff has been done */
3284 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3285 {
3286         unsigned int nr;
3287         struct task_struct *reaper;
3288         struct tgid_iter iter;
3289         struct pid_namespace *ns;
3290         filldir_t __filldir;
3291
3292         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3293                 goto out_no_task;
3294         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3295
3296         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3297         if (!reaper)
3298                 goto out_no_task;
3299
3300         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3301                 const struct pid_entry *p = &proc_base_stuff[nr];
3302                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3303                         goto out;
3304         }
3305
3306         ns = filp->f_dentry->d_sb->s_fs_info;
3307         iter.task = NULL;
3308         iter.tgid = filp->f_pos - TGID_OFFSET;
3309         for (iter = next_tgid(ns, iter);
3310              iter.task;
3311              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3312                 if (has_pid_permissions(ns, iter.task, 2))
3313                         __filldir = filldir;
3314                 else
3315                         __filldir = fake_filldir;
3316
3317                 filp->f_pos = iter.tgid + TGID_OFFSET;
3318                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3319                         put_task_struct(iter.task);
3320                         goto out;
3321                 }
3322         }
3323         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3324 out:
3325         put_task_struct(reaper);
3326 out_no_task:
3327         return 0;
3328 }
3329
3330 /*
3331  * Tasks
3332  */
3333 static const struct pid_entry tid_base_stuff[] = {
3334         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3335         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3336         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3337         REG("environ",   S_IRUSR, proc_environ_operations),
3338         INF("auxv",      S_IRUSR, proc_pid_auxv),
3339         ONE("status",    S_IRUGO, proc_pid_status),
3340         ONE("personality", S_IRUGO, proc_pid_personality),
3341         INF("limits",    S_IRUGO, proc_pid_limits),
3342 #ifdef CONFIG_SCHED_DEBUG
3343         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3344 #endif
3345         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3346 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3347         INF("syscall",   S_IRUGO, proc_pid_syscall),
3348 #endif
3349         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3350         ONE("stat",      S_IRUGO, proc_tid_stat),
3351         ONE("statm",     S_IRUGO, proc_pid_statm),
3352         REG("maps",      S_IRUGO, proc_maps_operations),
3353 #ifdef CONFIG_NUMA
3354         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3355 #endif
3356         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3357         LNK("cwd",       proc_cwd_link),
3358         LNK("root",      proc_root_link),
3359         LNK("exe",       proc_exe_link),
3360         REG("mounts",    S_IRUGO, proc_mounts_operations),
3361         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3362 #ifdef CONFIG_PROC_PAGE_MONITOR
3363         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3364         REG("smaps",     S_IRUGO, proc_smaps_operations),
3365         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3366 #endif
3367 #ifdef CONFIG_SECURITY
3368         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3369 #endif
3370 #ifdef CONFIG_KALLSYMS
3371         INF("wchan",     S_IRUGO, proc_pid_wchan),
3372 #endif
3373 #ifdef CONFIG_STACKTRACE
3374         ONE("stack",      S_IRUGO, proc_pid_stack),
3375 #endif
3376 #ifdef CONFIG_SCHEDSTATS
3377         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3378 #endif
3379 #ifdef CONFIG_LATENCYTOP
3380         REG("latency",  S_IRUGO, proc_lstats_operations),
3381 #endif
3382 #ifdef CONFIG_PROC_PID_CPUSET
3383         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3384 #endif
3385 #ifdef CONFIG_CGROUPS
3386         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3387 #endif
3388         INF("oom_score", S_IRUGO, proc_oom_score),
3389         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3390         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3391 #ifdef CONFIG_AUDITSYSCALL
3392         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3393         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3394 #endif
3395 #ifdef CONFIG_FAULT_INJECTION
3396         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3397 #endif
3398 #ifdef CONFIG_TASK_IO_ACCOUNTING
3399         INF("io",       S_IRUSR, proc_tid_io_accounting),
3400 #endif
3401 #ifdef CONFIG_HARDWALL
3402         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3403 #endif
3404 };
3405
3406 static int proc_tid_base_readdir(struct file * filp,
3407                              void * dirent, filldir_t filldir)
3408 {
3409         return proc_pident_readdir(filp,dirent,filldir,
3410                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3411 }
3412
3413 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3414         return proc_pident_lookup(dir, dentry,
3415                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3416 }
3417
3418 static const struct file_operations proc_tid_base_operations = {
3419         .read           = generic_read_dir,
3420         .readdir        = proc_tid_base_readdir,
3421         .llseek         = default_llseek,
3422 };
3423
3424 static const struct inode_operations proc_tid_base_inode_operations = {
3425         .lookup         = proc_tid_base_lookup,
3426         .getattr        = pid_getattr,
3427         .setattr        = proc_setattr,
3428 };
3429
3430 static struct dentry *proc_task_instantiate(struct inode *dir,
3431         struct dentry *dentry, struct task_struct *task, const void *ptr)
3432 {
3433         struct dentry *error = ERR_PTR(-ENOENT);
3434         struct inode *inode;
3435         inode = proc_pid_make_inode(dir->i_sb, task);
3436
3437         if (!inode)
3438                 goto out;
3439         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3440         inode->i_op = &proc_tid_base_inode_operations;
3441         inode->i_fop = &proc_tid_base_operations;
3442         inode->i_flags|=S_IMMUTABLE;
3443
3444         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3445                                                   ARRAY_SIZE(tid_base_stuff)));
3446
3447         d_set_d_op(dentry, &pid_dentry_operations);
3448
3449         d_add(dentry, inode);
3450         /* Close the race of the process dying before we return the dentry */
3451         if (pid_revalidate(dentry, NULL))
3452                 error = NULL;
3453 out:
3454         return error;
3455 }
3456
3457 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3458 {
3459         struct dentry *result = ERR_PTR(-ENOENT);
3460         struct task_struct *task;
3461         struct task_struct *leader = get_proc_task(dir);
3462         unsigned tid;
3463         struct pid_namespace *ns;
3464
3465         if (!leader)
3466                 goto out_no_task;
3467
3468         tid = name_to_int(dentry);
3469         if (tid == ~0U)
3470                 goto out;
3471
3472         ns = dentry->d_sb->s_fs_info;
3473         rcu_read_lock();
3474         task = find_task_by_pid_ns(tid, ns);
3475         if (task)
3476                 get_task_struct(task);
3477         rcu_read_unlock();
3478         if (!task)
3479                 goto out;
3480         if (!same_thread_group(leader, task))
3481                 goto out_drop_task;
3482
3483         result = proc_task_instantiate(dir, dentry, task, NULL);
3484 out_drop_task:
3485         put_task_struct(task);
3486 out:
3487         put_task_struct(leader);
3488 out_no_task:
3489         return result;
3490 }
3491
3492 /*
3493  * Find the first tid of a thread group to return to user space.
3494  *
3495  * Usually this is just the thread group leader, but if the users
3496  * buffer was too small or there was a seek into the middle of the
3497  * directory we have more work todo.
3498  *
3499  * In the case of a short read we start with find_task_by_pid.
3500  *
3501  * In the case of a seek we start with the leader and walk nr
3502  * threads past it.
3503  */
3504 static struct task_struct *first_tid(struct task_struct *leader,
3505                 int tid, int nr, struct pid_namespace *ns)
3506 {
3507         struct task_struct *pos;
3508
3509         rcu_read_lock();
3510         /* Attempt to start with the pid of a thread */
3511         if (tid && (nr > 0)) {
3512                 pos = find_task_by_pid_ns(tid, ns);
3513                 if (pos && (pos->group_leader == leader))
3514                         goto found;
3515         }
3516
3517         /* If nr exceeds the number of threads there is nothing todo */
3518         pos = NULL;
3519         if (nr && nr >= get_nr_threads(leader))
3520                 goto out;
3521
3522         /* If we haven't found our starting place yet start
3523          * with the leader and walk nr threads forward.
3524          */
3525         for (pos = leader; nr > 0; --nr) {
3526                 pos = next_thread(pos);
3527                 if (pos == leader) {
3528                         pos = NULL;
3529                         goto out;
3530                 }
3531         }
3532 found:
3533         get_task_struct(pos);
3534 out:
3535         rcu_read_unlock();
3536         return pos;
3537 }
3538
3539 /*
3540  * Find the next thread in the thread list.
3541  * Return NULL if there is an error or no next thread.
3542  *
3543  * The reference to the input task_struct is released.
3544  */
3545 static struct task_struct *next_tid(struct task_struct *start)
3546 {
3547         struct task_struct *pos = NULL;
3548         rcu_read_lock();
3549         if (pid_alive(start)) {
3550                 pos = next_thread(start);
3551                 if (thread_group_leader(pos))
3552                         pos = NULL;
3553                 else
3554                         get_task_struct(pos);
3555         }
3556         rcu_read_unlock();
3557         put_task_struct(start);
3558         return pos;
3559 }
3560
3561 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3562         struct task_struct *task, int tid)
3563 {
3564         char name[PROC_NUMBUF];
3565         int len = snprintf(name, sizeof(name), "%d", tid);
3566         return proc_fill_cache(filp, dirent, filldir, name, len,
3567                                 proc_task_instantiate, task, NULL);
3568 }
3569
3570 /* for the /proc/TGID/task/ directories */
3571 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3572 {
3573         struct dentry *dentry = filp->f_path.dentry;
3574         struct inode *inode = dentry->d_inode;
3575         struct task_struct *leader = NULL;
3576         struct task_struct *task;
3577         int retval = -ENOENT;
3578         ino_t ino;
3579         int tid;
3580         struct pid_namespace *ns;
3581
3582         task = get_proc_task(inode);
3583         if (!task)
3584                 goto out_no_task;
3585         rcu_read_lock();
3586         if (pid_alive(task)) {
3587                 leader = task->group_leader;
3588                 get_task_struct(leader);
3589         }
3590         rcu_read_unlock();
3591         put_task_struct(task);
3592         if (!leader)
3593                 goto out_no_task;
3594         retval = 0;
3595
3596         switch ((unsigned long)filp->f_pos) {
3597         case 0:
3598                 ino = inode->i_ino;
3599                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3600                         goto out;
3601                 filp->f_pos++;
3602                 /* fall through */
3603         case 1:
3604                 ino = parent_ino(dentry);
3605                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3606                         goto out;
3607                 filp->f_pos++;
3608                 /* fall through */
3609         }
3610
3611         /* f_version caches the tgid value that the last readdir call couldn't
3612          * return. lseek aka telldir automagically resets f_version to 0.
3613          */
3614         ns = filp->f_dentry->d_sb->s_fs_info;
3615         tid = (int)filp->f_version;
3616         filp->f_version = 0;
3617         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3618              task;
3619              task = next_tid(task), filp->f_pos++) {
3620                 tid = task_pid_nr_ns(task, ns);
3621                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3622                         /* returning this tgid failed, save it as the first
3623                          * pid for the next readir call */
3624                         filp->f_version = (u64)tid;
3625                         put_task_struct(task);
3626                         break;
3627                 }
3628         }
3629 out:
3630         put_task_struct(leader);
3631 out_no_task:
3632         return retval;
3633 }
3634
3635 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3636 {
3637         struct inode *inode = dentry->d_inode;
3638         struct task_struct *p = get_proc_task(inode);
3639         generic_fillattr(inode, stat);
3640
3641         if (p) {
3642                 stat->nlink += get_nr_threads(p);
3643                 put_task_struct(p);
3644         }
3645
3646         return 0;
3647 }
3648
3649 static const struct inode_operations proc_task_inode_operations = {
3650         .lookup         = proc_task_lookup,
3651         .getattr        = proc_task_getattr,
3652         .setattr        = proc_setattr,
3653         .permission     = proc_pid_permission,
3654 };
3655
3656 static const struct file_operations proc_task_operations = {
3657         .read           = generic_read_dir,
3658         .readdir        = proc_task_readdir,
3659         .llseek         = default_llseek,
3660 };