Merge branch 'akpm' (Andrew's patch-bomb)
[omap4-v4l2-camera:dvachevs-omap4-v4l2-camera.git] / fs / proc / base.c
1 /*
2  *  linux/fs/proc/base.c
3  *
4  *  Copyright (C) 1991, 1992 Linus Torvalds
5  *
6  *  proc base directory handling functions
7  *
8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9  *  Instead of using magical inumbers to determine the kind of object
10  *  we allocate and fill in-core inodes upon lookup. They don't even
11  *  go into icache. We cache the reference to task_struct upon lookup too.
12  *  Eventually it should become a filesystem in its own. We don't use the
13  *  rest of procfs anymore.
14  *
15  *
16  *  Changelog:
17  *  17-Jan-2005
18  *  Allan Bezerra
19  *  Bruna Moreira <bruna.moreira@indt.org.br>
20  *  Edjard Mota <edjard.mota@indt.org.br>
21  *  Ilias Biris <ilias.biris@indt.org.br>
22  *  Mauricio Lin <mauricio.lin@indt.org.br>
23  *
24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25  *
26  *  A new process specific entry (smaps) included in /proc. It shows the
27  *  size of rss for each memory area. The maps entry lacks information
28  *  about physical memory size (rss) for each mapped file, i.e.,
29  *  rss information for executables and library files.
30  *  This additional information is useful for any tools that need to know
31  *  about physical memory consumption for a process specific library.
32  *
33  *  Changelog:
34  *  21-Feb-2005
35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36  *  Pud inclusion in the page table walking.
37  *
38  *  ChangeLog:
39  *  10-Mar-2005
40  *  10LE Instituto Nokia de Tecnologia - INdT:
41  *  A better way to walks through the page table as suggested by Hugh Dickins.
42  *
43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44  *  Smaps information related to shared, private, clean and dirty pages.
45  *
46  *  Paul Mundt <paul.mundt@nokia.com>:
47  *  Overall revision about smaps.
48  */
49
50 #include <asm/uaccess.h>
51
52 #include <linux/errno.h>
53 #include <linux/time.h>
54 #include <linux/proc_fs.h>
55 #include <linux/stat.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/init.h>
58 #include <linux/capability.h>
59 #include <linux/file.h>
60 #include <linux/fdtable.h>
61 #include <linux/string.h>
62 #include <linux/seq_file.h>
63 #include <linux/namei.h>
64 #include <linux/mnt_namespace.h>
65 #include <linux/mm.h>
66 #include <linux/swap.h>
67 #include <linux/rcupdate.h>
68 #include <linux/kallsyms.h>
69 #include <linux/stacktrace.h>
70 #include <linux/resource.h>
71 #include <linux/module.h>
72 #include <linux/mount.h>
73 #include <linux/security.h>
74 #include <linux/ptrace.h>
75 #include <linux/tracehook.h>
76 #include <linux/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/fs_struct.h>
85 #include <linux/slab.h>
86 #include <linux/flex_array.h>
87 #ifdef CONFIG_HARDWALL
88 #include <asm/hardwall.h>
89 #endif
90 #include <trace/events/oom.h>
91 #include "internal.h"
92
93 /* NOTE:
94  *      Implementing inode permission operations in /proc is almost
95  *      certainly an error.  Permission checks need to happen during
96  *      each system call not at open time.  The reason is that most of
97  *      what we wish to check for permissions in /proc varies at runtime.
98  *
99  *      The classic example of a problem is opening file descriptors
100  *      in /proc for a task before it execs a suid executable.
101  */
102
103 struct pid_entry {
104         char *name;
105         int len;
106         umode_t mode;
107         const struct inode_operations *iop;
108         const struct file_operations *fop;
109         union proc_op op;
110 };
111
112 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
113         .name = (NAME),                                 \
114         .len  = sizeof(NAME) - 1,                       \
115         .mode = MODE,                                   \
116         .iop  = IOP,                                    \
117         .fop  = FOP,                                    \
118         .op   = OP,                                     \
119 }
120
121 #define DIR(NAME, MODE, iops, fops)     \
122         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
123 #define LNK(NAME, get_link)                                     \
124         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
125                 &proc_pid_link_inode_operations, NULL,          \
126                 { .proc_get_link = get_link } )
127 #define REG(NAME, MODE, fops)                           \
128         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
129 #define INF(NAME, MODE, read)                           \
130         NOD(NAME, (S_IFREG|(MODE)),                     \
131                 NULL, &proc_info_file_operations,       \
132                 { .proc_read = read } )
133 #define ONE(NAME, MODE, show)                           \
134         NOD(NAME, (S_IFREG|(MODE)),                     \
135                 NULL, &proc_single_file_operations,     \
136                 { .proc_show = show } )
137
138 static int proc_fd_permission(struct inode *inode, int mask);
139
140 /*
141  * Count the number of hardlinks for the pid_entry table, excluding the .
142  * and .. links.
143  */
144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
145         unsigned int n)
146 {
147         unsigned int i;
148         unsigned int count;
149
150         count = 0;
151         for (i = 0; i < n; ++i) {
152                 if (S_ISDIR(entries[i].mode))
153                         ++count;
154         }
155
156         return count;
157 }
158
159 static int get_task_root(struct task_struct *task, struct path *root)
160 {
161         int result = -ENOENT;
162
163         task_lock(task);
164         if (task->fs) {
165                 get_fs_root(task->fs, root);
166                 result = 0;
167         }
168         task_unlock(task);
169         return result;
170 }
171
172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
173 {
174         struct task_struct *task = get_proc_task(dentry->d_inode);
175         int result = -ENOENT;
176
177         if (task) {
178                 task_lock(task);
179                 if (task->fs) {
180                         get_fs_pwd(task->fs, path);
181                         result = 0;
182                 }
183                 task_unlock(task);
184                 put_task_struct(task);
185         }
186         return result;
187 }
188
189 static int proc_root_link(struct dentry *dentry, struct path *path)
190 {
191         struct task_struct *task = get_proc_task(dentry->d_inode);
192         int result = -ENOENT;
193
194         if (task) {
195                 result = get_task_root(task, path);
196                 put_task_struct(task);
197         }
198         return result;
199 }
200
201 struct mm_struct *mm_for_maps(struct task_struct *task)
202 {
203         return mm_access(task, PTRACE_MODE_READ);
204 }
205
206 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
207 {
208         int res = 0;
209         unsigned int len;
210         struct mm_struct *mm = get_task_mm(task);
211         if (!mm)
212                 goto out;
213         if (!mm->arg_end)
214                 goto out_mm;    /* Shh! No looking before we're done */
215
216         len = mm->arg_end - mm->arg_start;
217  
218         if (len > PAGE_SIZE)
219                 len = PAGE_SIZE;
220  
221         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
222
223         // If the nul at the end of args has been overwritten, then
224         // assume application is using setproctitle(3).
225         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
226                 len = strnlen(buffer, res);
227                 if (len < res) {
228                     res = len;
229                 } else {
230                         len = mm->env_end - mm->env_start;
231                         if (len > PAGE_SIZE - res)
232                                 len = PAGE_SIZE - res;
233                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
234                         res = strnlen(buffer, res);
235                 }
236         }
237 out_mm:
238         mmput(mm);
239 out:
240         return res;
241 }
242
243 static int proc_pid_auxv(struct task_struct *task, char *buffer)
244 {
245         struct mm_struct *mm = mm_for_maps(task);
246         int res = PTR_ERR(mm);
247         if (mm && !IS_ERR(mm)) {
248                 unsigned int nwords = 0;
249                 do {
250                         nwords += 2;
251                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
252                 res = nwords * sizeof(mm->saved_auxv[0]);
253                 if (res > PAGE_SIZE)
254                         res = PAGE_SIZE;
255                 memcpy(buffer, mm->saved_auxv, res);
256                 mmput(mm);
257         }
258         return res;
259 }
260
261
262 #ifdef CONFIG_KALLSYMS
263 /*
264  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
265  * Returns the resolved symbol.  If that fails, simply return the address.
266  */
267 static int proc_pid_wchan(struct task_struct *task, char *buffer)
268 {
269         unsigned long wchan;
270         char symname[KSYM_NAME_LEN];
271
272         wchan = get_wchan(task);
273
274         if (lookup_symbol_name(wchan, symname) < 0)
275                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
276                         return 0;
277                 else
278                         return sprintf(buffer, "%lu", wchan);
279         else
280                 return sprintf(buffer, "%s", symname);
281 }
282 #endif /* CONFIG_KALLSYMS */
283
284 static int lock_trace(struct task_struct *task)
285 {
286         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
287         if (err)
288                 return err;
289         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
290                 mutex_unlock(&task->signal->cred_guard_mutex);
291                 return -EPERM;
292         }
293         return 0;
294 }
295
296 static void unlock_trace(struct task_struct *task)
297 {
298         mutex_unlock(&task->signal->cred_guard_mutex);
299 }
300
301 #ifdef CONFIG_STACKTRACE
302
303 #define MAX_STACK_TRACE_DEPTH   64
304
305 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
306                           struct pid *pid, struct task_struct *task)
307 {
308         struct stack_trace trace;
309         unsigned long *entries;
310         int err;
311         int i;
312
313         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
314         if (!entries)
315                 return -ENOMEM;
316
317         trace.nr_entries        = 0;
318         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
319         trace.entries           = entries;
320         trace.skip              = 0;
321
322         err = lock_trace(task);
323         if (!err) {
324                 save_stack_trace_tsk(task, &trace);
325
326                 for (i = 0; i < trace.nr_entries; i++) {
327                         seq_printf(m, "[<%pK>] %pS\n",
328                                    (void *)entries[i], (void *)entries[i]);
329                 }
330                 unlock_trace(task);
331         }
332         kfree(entries);
333
334         return err;
335 }
336 #endif
337
338 #ifdef CONFIG_SCHEDSTATS
339 /*
340  * Provides /proc/PID/schedstat
341  */
342 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
343 {
344         return sprintf(buffer, "%llu %llu %lu\n",
345                         (unsigned long long)task->se.sum_exec_runtime,
346                         (unsigned long long)task->sched_info.run_delay,
347                         task->sched_info.pcount);
348 }
349 #endif
350
351 #ifdef CONFIG_LATENCYTOP
352 static int lstats_show_proc(struct seq_file *m, void *v)
353 {
354         int i;
355         struct inode *inode = m->private;
356         struct task_struct *task = get_proc_task(inode);
357
358         if (!task)
359                 return -ESRCH;
360         seq_puts(m, "Latency Top version : v0.1\n");
361         for (i = 0; i < 32; i++) {
362                 struct latency_record *lr = &task->latency_record[i];
363                 if (lr->backtrace[0]) {
364                         int q;
365                         seq_printf(m, "%i %li %li",
366                                    lr->count, lr->time, lr->max);
367                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
368                                 unsigned long bt = lr->backtrace[q];
369                                 if (!bt)
370                                         break;
371                                 if (bt == ULONG_MAX)
372                                         break;
373                                 seq_printf(m, " %ps", (void *)bt);
374                         }
375                         seq_putc(m, '\n');
376                 }
377
378         }
379         put_task_struct(task);
380         return 0;
381 }
382
383 static int lstats_open(struct inode *inode, struct file *file)
384 {
385         return single_open(file, lstats_show_proc, inode);
386 }
387
388 static ssize_t lstats_write(struct file *file, const char __user *buf,
389                             size_t count, loff_t *offs)
390 {
391         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
392
393         if (!task)
394                 return -ESRCH;
395         clear_all_latency_tracing(task);
396         put_task_struct(task);
397
398         return count;
399 }
400
401 static const struct file_operations proc_lstats_operations = {
402         .open           = lstats_open,
403         .read           = seq_read,
404         .write          = lstats_write,
405         .llseek         = seq_lseek,
406         .release        = single_release,
407 };
408
409 #endif
410
411 static int proc_oom_score(struct task_struct *task, char *buffer)
412 {
413         unsigned long points = 0;
414
415         read_lock(&tasklist_lock);
416         if (pid_alive(task))
417                 points = oom_badness(task, NULL, NULL,
418                                         totalram_pages + total_swap_pages);
419         read_unlock(&tasklist_lock);
420         return sprintf(buffer, "%lu\n", points);
421 }
422
423 struct limit_names {
424         char *name;
425         char *unit;
426 };
427
428 static const struct limit_names lnames[RLIM_NLIMITS] = {
429         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
430         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
431         [RLIMIT_DATA] = {"Max data size", "bytes"},
432         [RLIMIT_STACK] = {"Max stack size", "bytes"},
433         [RLIMIT_CORE] = {"Max core file size", "bytes"},
434         [RLIMIT_RSS] = {"Max resident set", "bytes"},
435         [RLIMIT_NPROC] = {"Max processes", "processes"},
436         [RLIMIT_NOFILE] = {"Max open files", "files"},
437         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
438         [RLIMIT_AS] = {"Max address space", "bytes"},
439         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
440         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
441         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
442         [RLIMIT_NICE] = {"Max nice priority", NULL},
443         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
444         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
445 };
446
447 /* Display limits for a process */
448 static int proc_pid_limits(struct task_struct *task, char *buffer)
449 {
450         unsigned int i;
451         int count = 0;
452         unsigned long flags;
453         char *bufptr = buffer;
454
455         struct rlimit rlim[RLIM_NLIMITS];
456
457         if (!lock_task_sighand(task, &flags))
458                 return 0;
459         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
460         unlock_task_sighand(task, &flags);
461
462         /*
463          * print the file header
464          */
465         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
466                         "Limit", "Soft Limit", "Hard Limit", "Units");
467
468         for (i = 0; i < RLIM_NLIMITS; i++) {
469                 if (rlim[i].rlim_cur == RLIM_INFINITY)
470                         count += sprintf(&bufptr[count], "%-25s %-20s ",
471                                          lnames[i].name, "unlimited");
472                 else
473                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
474                                          lnames[i].name, rlim[i].rlim_cur);
475
476                 if (rlim[i].rlim_max == RLIM_INFINITY)
477                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
478                 else
479                         count += sprintf(&bufptr[count], "%-20lu ",
480                                          rlim[i].rlim_max);
481
482                 if (lnames[i].unit)
483                         count += sprintf(&bufptr[count], "%-10s\n",
484                                          lnames[i].unit);
485                 else
486                         count += sprintf(&bufptr[count], "\n");
487         }
488
489         return count;
490 }
491
492 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
493 static int proc_pid_syscall(struct task_struct *task, char *buffer)
494 {
495         long nr;
496         unsigned long args[6], sp, pc;
497         int res = lock_trace(task);
498         if (res)
499                 return res;
500
501         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
502                 res = sprintf(buffer, "running\n");
503         else if (nr < 0)
504                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
505         else
506                 res = sprintf(buffer,
507                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
508                        nr,
509                        args[0], args[1], args[2], args[3], args[4], args[5],
510                        sp, pc);
511         unlock_trace(task);
512         return res;
513 }
514 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
515
516 /************************************************************************/
517 /*                       Here the fs part begins                        */
518 /************************************************************************/
519
520 /* permission checks */
521 static int proc_fd_access_allowed(struct inode *inode)
522 {
523         struct task_struct *task;
524         int allowed = 0;
525         /* Allow access to a task's file descriptors if it is us or we
526          * may use ptrace attach to the process and find out that
527          * information.
528          */
529         task = get_proc_task(inode);
530         if (task) {
531                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
532                 put_task_struct(task);
533         }
534         return allowed;
535 }
536
537 int proc_setattr(struct dentry *dentry, struct iattr *attr)
538 {
539         int error;
540         struct inode *inode = dentry->d_inode;
541
542         if (attr->ia_valid & ATTR_MODE)
543                 return -EPERM;
544
545         error = inode_change_ok(inode, attr);
546         if (error)
547                 return error;
548
549         if ((attr->ia_valid & ATTR_SIZE) &&
550             attr->ia_size != i_size_read(inode)) {
551                 error = vmtruncate(inode, attr->ia_size);
552                 if (error)
553                         return error;
554         }
555
556         setattr_copy(inode, attr);
557         mark_inode_dirty(inode);
558         return 0;
559 }
560
561 /*
562  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
563  * or euid/egid (for hide_pid_min=2)?
564  */
565 static bool has_pid_permissions(struct pid_namespace *pid,
566                                  struct task_struct *task,
567                                  int hide_pid_min)
568 {
569         if (pid->hide_pid < hide_pid_min)
570                 return true;
571         if (in_group_p(pid->pid_gid))
572                 return true;
573         return ptrace_may_access(task, PTRACE_MODE_READ);
574 }
575
576
577 static int proc_pid_permission(struct inode *inode, int mask)
578 {
579         struct pid_namespace *pid = inode->i_sb->s_fs_info;
580         struct task_struct *task;
581         bool has_perms;
582
583         task = get_proc_task(inode);
584         if (!task)
585                 return -ESRCH;
586         has_perms = has_pid_permissions(pid, task, 1);
587         put_task_struct(task);
588
589         if (!has_perms) {
590                 if (pid->hide_pid == 2) {
591                         /*
592                          * Let's make getdents(), stat(), and open()
593                          * consistent with each other.  If a process
594                          * may not stat() a file, it shouldn't be seen
595                          * in procfs at all.
596                          */
597                         return -ENOENT;
598                 }
599
600                 return -EPERM;
601         }
602         return generic_permission(inode, mask);
603 }
604
605
606
607 static const struct inode_operations proc_def_inode_operations = {
608         .setattr        = proc_setattr,
609 };
610
611 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
612
613 static ssize_t proc_info_read(struct file * file, char __user * buf,
614                           size_t count, loff_t *ppos)
615 {
616         struct inode * inode = file->f_path.dentry->d_inode;
617         unsigned long page;
618         ssize_t length;
619         struct task_struct *task = get_proc_task(inode);
620
621         length = -ESRCH;
622         if (!task)
623                 goto out_no_task;
624
625         if (count > PROC_BLOCK_SIZE)
626                 count = PROC_BLOCK_SIZE;
627
628         length = -ENOMEM;
629         if (!(page = __get_free_page(GFP_TEMPORARY)))
630                 goto out;
631
632         length = PROC_I(inode)->op.proc_read(task, (char*)page);
633
634         if (length >= 0)
635                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
636         free_page(page);
637 out:
638         put_task_struct(task);
639 out_no_task:
640         return length;
641 }
642
643 static const struct file_operations proc_info_file_operations = {
644         .read           = proc_info_read,
645         .llseek         = generic_file_llseek,
646 };
647
648 static int proc_single_show(struct seq_file *m, void *v)
649 {
650         struct inode *inode = m->private;
651         struct pid_namespace *ns;
652         struct pid *pid;
653         struct task_struct *task;
654         int ret;
655
656         ns = inode->i_sb->s_fs_info;
657         pid = proc_pid(inode);
658         task = get_pid_task(pid, PIDTYPE_PID);
659         if (!task)
660                 return -ESRCH;
661
662         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
663
664         put_task_struct(task);
665         return ret;
666 }
667
668 static int proc_single_open(struct inode *inode, struct file *filp)
669 {
670         return single_open(filp, proc_single_show, inode);
671 }
672
673 static const struct file_operations proc_single_file_operations = {
674         .open           = proc_single_open,
675         .read           = seq_read,
676         .llseek         = seq_lseek,
677         .release        = single_release,
678 };
679
680 static int mem_open(struct inode* inode, struct file* file)
681 {
682         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
683         struct mm_struct *mm;
684
685         if (!task)
686                 return -ESRCH;
687
688         mm = mm_access(task, PTRACE_MODE_ATTACH);
689         put_task_struct(task);
690
691         if (IS_ERR(mm))
692                 return PTR_ERR(mm);
693
694         if (mm) {
695                 /* ensure this mm_struct can't be freed */
696                 atomic_inc(&mm->mm_count);
697                 /* but do not pin its memory */
698                 mmput(mm);
699         }
700
701         /* OK to pass negative loff_t, we can catch out-of-range */
702         file->f_mode |= FMODE_UNSIGNED_OFFSET;
703         file->private_data = mm;
704
705         return 0;
706 }
707
708 static ssize_t mem_rw(struct file *file, char __user *buf,
709                         size_t count, loff_t *ppos, int write)
710 {
711         struct mm_struct *mm = file->private_data;
712         unsigned long addr = *ppos;
713         ssize_t copied;
714         char *page;
715
716         if (!mm)
717                 return 0;
718
719         page = (char *)__get_free_page(GFP_TEMPORARY);
720         if (!page)
721                 return -ENOMEM;
722
723         copied = 0;
724         if (!atomic_inc_not_zero(&mm->mm_users))
725                 goto free;
726
727         while (count > 0) {
728                 int this_len = min_t(int, count, PAGE_SIZE);
729
730                 if (write && copy_from_user(page, buf, this_len)) {
731                         copied = -EFAULT;
732                         break;
733                 }
734
735                 this_len = access_remote_vm(mm, addr, page, this_len, write);
736                 if (!this_len) {
737                         if (!copied)
738                                 copied = -EIO;
739                         break;
740                 }
741
742                 if (!write && copy_to_user(buf, page, this_len)) {
743                         copied = -EFAULT;
744                         break;
745                 }
746
747                 buf += this_len;
748                 addr += this_len;
749                 copied += this_len;
750                 count -= this_len;
751         }
752         *ppos = addr;
753
754         mmput(mm);
755 free:
756         free_page((unsigned long) page);
757         return copied;
758 }
759
760 static ssize_t mem_read(struct file *file, char __user *buf,
761                         size_t count, loff_t *ppos)
762 {
763         return mem_rw(file, buf, count, ppos, 0);
764 }
765
766 static ssize_t mem_write(struct file *file, const char __user *buf,
767                          size_t count, loff_t *ppos)
768 {
769         return mem_rw(file, (char __user*)buf, count, ppos, 1);
770 }
771
772 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
773 {
774         switch (orig) {
775         case 0:
776                 file->f_pos = offset;
777                 break;
778         case 1:
779                 file->f_pos += offset;
780                 break;
781         default:
782                 return -EINVAL;
783         }
784         force_successful_syscall_return();
785         return file->f_pos;
786 }
787
788 static int mem_release(struct inode *inode, struct file *file)
789 {
790         struct mm_struct *mm = file->private_data;
791         if (mm)
792                 mmdrop(mm);
793         return 0;
794 }
795
796 static const struct file_operations proc_mem_operations = {
797         .llseek         = mem_lseek,
798         .read           = mem_read,
799         .write          = mem_write,
800         .open           = mem_open,
801         .release        = mem_release,
802 };
803
804 static ssize_t environ_read(struct file *file, char __user *buf,
805                         size_t count, loff_t *ppos)
806 {
807         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
808         char *page;
809         unsigned long src = *ppos;
810         int ret = -ESRCH;
811         struct mm_struct *mm;
812
813         if (!task)
814                 goto out_no_task;
815
816         ret = -ENOMEM;
817         page = (char *)__get_free_page(GFP_TEMPORARY);
818         if (!page)
819                 goto out;
820
821
822         mm = mm_for_maps(task);
823         ret = PTR_ERR(mm);
824         if (!mm || IS_ERR(mm))
825                 goto out_free;
826
827         ret = 0;
828         while (count > 0) {
829                 int this_len, retval, max_len;
830
831                 this_len = mm->env_end - (mm->env_start + src);
832
833                 if (this_len <= 0)
834                         break;
835
836                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
837                 this_len = (this_len > max_len) ? max_len : this_len;
838
839                 retval = access_process_vm(task, (mm->env_start + src),
840                         page, this_len, 0);
841
842                 if (retval <= 0) {
843                         ret = retval;
844                         break;
845                 }
846
847                 if (copy_to_user(buf, page, retval)) {
848                         ret = -EFAULT;
849                         break;
850                 }
851
852                 ret += retval;
853                 src += retval;
854                 buf += retval;
855                 count -= retval;
856         }
857         *ppos = src;
858
859         mmput(mm);
860 out_free:
861         free_page((unsigned long) page);
862 out:
863         put_task_struct(task);
864 out_no_task:
865         return ret;
866 }
867
868 static const struct file_operations proc_environ_operations = {
869         .read           = environ_read,
870         .llseek         = generic_file_llseek,
871 };
872
873 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
874                                 size_t count, loff_t *ppos)
875 {
876         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877         char buffer[PROC_NUMBUF];
878         size_t len;
879         int oom_adjust = OOM_DISABLE;
880         unsigned long flags;
881
882         if (!task)
883                 return -ESRCH;
884
885         if (lock_task_sighand(task, &flags)) {
886                 oom_adjust = task->signal->oom_adj;
887                 unlock_task_sighand(task, &flags);
888         }
889
890         put_task_struct(task);
891
892         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
893
894         return simple_read_from_buffer(buf, count, ppos, buffer, len);
895 }
896
897 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
898                                 size_t count, loff_t *ppos)
899 {
900         struct task_struct *task;
901         char buffer[PROC_NUMBUF];
902         int oom_adjust;
903         unsigned long flags;
904         int err;
905
906         memset(buffer, 0, sizeof(buffer));
907         if (count > sizeof(buffer) - 1)
908                 count = sizeof(buffer) - 1;
909         if (copy_from_user(buffer, buf, count)) {
910                 err = -EFAULT;
911                 goto out;
912         }
913
914         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
915         if (err)
916                 goto out;
917         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
918              oom_adjust != OOM_DISABLE) {
919                 err = -EINVAL;
920                 goto out;
921         }
922
923         task = get_proc_task(file->f_path.dentry->d_inode);
924         if (!task) {
925                 err = -ESRCH;
926                 goto out;
927         }
928
929         task_lock(task);
930         if (!task->mm) {
931                 err = -EINVAL;
932                 goto err_task_lock;
933         }
934
935         if (!lock_task_sighand(task, &flags)) {
936                 err = -ESRCH;
937                 goto err_task_lock;
938         }
939
940         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
941                 err = -EACCES;
942                 goto err_sighand;
943         }
944
945         /*
946          * Warn that /proc/pid/oom_adj is deprecated, see
947          * Documentation/feature-removal-schedule.txt.
948          */
949         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
950                   current->comm, task_pid_nr(current), task_pid_nr(task),
951                   task_pid_nr(task));
952         task->signal->oom_adj = oom_adjust;
953         /*
954          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
955          * value is always attainable.
956          */
957         if (task->signal->oom_adj == OOM_ADJUST_MAX)
958                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
959         else
960                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
961                                                                 -OOM_DISABLE;
962         trace_oom_score_adj_update(task);
963 err_sighand:
964         unlock_task_sighand(task, &flags);
965 err_task_lock:
966         task_unlock(task);
967         put_task_struct(task);
968 out:
969         return err < 0 ? err : count;
970 }
971
972 static const struct file_operations proc_oom_adjust_operations = {
973         .read           = oom_adjust_read,
974         .write          = oom_adjust_write,
975         .llseek         = generic_file_llseek,
976 };
977
978 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
979                                         size_t count, loff_t *ppos)
980 {
981         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
982         char buffer[PROC_NUMBUF];
983         int oom_score_adj = OOM_SCORE_ADJ_MIN;
984         unsigned long flags;
985         size_t len;
986
987         if (!task)
988                 return -ESRCH;
989         if (lock_task_sighand(task, &flags)) {
990                 oom_score_adj = task->signal->oom_score_adj;
991                 unlock_task_sighand(task, &flags);
992         }
993         put_task_struct(task);
994         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
995         return simple_read_from_buffer(buf, count, ppos, buffer, len);
996 }
997
998 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
999                                         size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task;
1002         char buffer[PROC_NUMBUF];
1003         unsigned long flags;
1004         int oom_score_adj;
1005         int err;
1006
1007         memset(buffer, 0, sizeof(buffer));
1008         if (count > sizeof(buffer) - 1)
1009                 count = sizeof(buffer) - 1;
1010         if (copy_from_user(buffer, buf, count)) {
1011                 err = -EFAULT;
1012                 goto out;
1013         }
1014
1015         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016         if (err)
1017                 goto out;
1018         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         task = get_proc_task(file->f_path.dentry->d_inode);
1025         if (!task) {
1026                 err = -ESRCH;
1027                 goto out;
1028         }
1029
1030         task_lock(task);
1031         if (!task->mm) {
1032                 err = -EINVAL;
1033                 goto err_task_lock;
1034         }
1035
1036         if (!lock_task_sighand(task, &flags)) {
1037                 err = -ESRCH;
1038                 goto err_task_lock;
1039         }
1040
1041         if (oom_score_adj < task->signal->oom_score_adj_min &&
1042                         !capable(CAP_SYS_RESOURCE)) {
1043                 err = -EACCES;
1044                 goto err_sighand;
1045         }
1046
1047         task->signal->oom_score_adj = oom_score_adj;
1048         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049                 task->signal->oom_score_adj_min = oom_score_adj;
1050         trace_oom_score_adj_update(task);
1051         /*
1052          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053          * always attainable.
1054          */
1055         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056                 task->signal->oom_adj = OOM_DISABLE;
1057         else
1058                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059                                                         OOM_SCORE_ADJ_MAX;
1060 err_sighand:
1061         unlock_task_sighand(task, &flags);
1062 err_task_lock:
1063         task_unlock(task);
1064         put_task_struct(task);
1065 out:
1066         return err < 0 ? err : count;
1067 }
1068
1069 static const struct file_operations proc_oom_score_adj_operations = {
1070         .read           = oom_score_adj_read,
1071         .write          = oom_score_adj_write,
1072         .llseek         = default_llseek,
1073 };
1074
1075 #ifdef CONFIG_AUDITSYSCALL
1076 #define TMPBUFLEN 21
1077 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078                                   size_t count, loff_t *ppos)
1079 {
1080         struct inode * inode = file->f_path.dentry->d_inode;
1081         struct task_struct *task = get_proc_task(inode);
1082         ssize_t length;
1083         char tmpbuf[TMPBUFLEN];
1084
1085         if (!task)
1086                 return -ESRCH;
1087         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088                                 audit_get_loginuid(task));
1089         put_task_struct(task);
1090         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091 }
1092
1093 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094                                    size_t count, loff_t *ppos)
1095 {
1096         struct inode * inode = file->f_path.dentry->d_inode;
1097         char *page, *tmp;
1098         ssize_t length;
1099         uid_t loginuid;
1100
1101         rcu_read_lock();
1102         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103                 rcu_read_unlock();
1104                 return -EPERM;
1105         }
1106         rcu_read_unlock();
1107
1108         if (count >= PAGE_SIZE)
1109                 count = PAGE_SIZE - 1;
1110
1111         if (*ppos != 0) {
1112                 /* No partial writes. */
1113                 return -EINVAL;
1114         }
1115         page = (char*)__get_free_page(GFP_TEMPORARY);
1116         if (!page)
1117                 return -ENOMEM;
1118         length = -EFAULT;
1119         if (copy_from_user(page, buf, count))
1120                 goto out_free_page;
1121
1122         page[count] = '\0';
1123         loginuid = simple_strtoul(page, &tmp, 10);
1124         if (tmp == page) {
1125                 length = -EINVAL;
1126                 goto out_free_page;
1127
1128         }
1129         length = audit_set_loginuid(loginuid);
1130         if (likely(length == 0))
1131                 length = count;
1132
1133 out_free_page:
1134         free_page((unsigned long) page);
1135         return length;
1136 }
1137
1138 static const struct file_operations proc_loginuid_operations = {
1139         .read           = proc_loginuid_read,
1140         .write          = proc_loginuid_write,
1141         .llseek         = generic_file_llseek,
1142 };
1143
1144 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145                                   size_t count, loff_t *ppos)
1146 {
1147         struct inode * inode = file->f_path.dentry->d_inode;
1148         struct task_struct *task = get_proc_task(inode);
1149         ssize_t length;
1150         char tmpbuf[TMPBUFLEN];
1151
1152         if (!task)
1153                 return -ESRCH;
1154         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155                                 audit_get_sessionid(task));
1156         put_task_struct(task);
1157         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158 }
1159
1160 static const struct file_operations proc_sessionid_operations = {
1161         .read           = proc_sessionid_read,
1162         .llseek         = generic_file_llseek,
1163 };
1164 #endif
1165
1166 #ifdef CONFIG_FAULT_INJECTION
1167 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168                                       size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171         char buffer[PROC_NUMBUF];
1172         size_t len;
1173         int make_it_fail;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         make_it_fail = task->make_it_fail;
1178         put_task_struct(task);
1179
1180         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183 }
1184
1185 static ssize_t proc_fault_inject_write(struct file * file,
1186                         const char __user * buf, size_t count, loff_t *ppos)
1187 {
1188         struct task_struct *task;
1189         char buffer[PROC_NUMBUF], *end;
1190         int make_it_fail;
1191
1192         if (!capable(CAP_SYS_RESOURCE))
1193                 return -EPERM;
1194         memset(buffer, 0, sizeof(buffer));
1195         if (count > sizeof(buffer) - 1)
1196                 count = sizeof(buffer) - 1;
1197         if (copy_from_user(buffer, buf, count))
1198                 return -EFAULT;
1199         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200         if (*end)
1201                 return -EINVAL;
1202         task = get_proc_task(file->f_dentry->d_inode);
1203         if (!task)
1204                 return -ESRCH;
1205         task->make_it_fail = make_it_fail;
1206         put_task_struct(task);
1207
1208         return count;
1209 }
1210
1211 static const struct file_operations proc_fault_inject_operations = {
1212         .read           = proc_fault_inject_read,
1213         .write          = proc_fault_inject_write,
1214         .llseek         = generic_file_llseek,
1215 };
1216 #endif
1217
1218
1219 #ifdef CONFIG_SCHED_DEBUG
1220 /*
1221  * Print out various scheduling related per-task fields:
1222  */
1223 static int sched_show(struct seq_file *m, void *v)
1224 {
1225         struct inode *inode = m->private;
1226         struct task_struct *p;
1227
1228         p = get_proc_task(inode);
1229         if (!p)
1230                 return -ESRCH;
1231         proc_sched_show_task(p, m);
1232
1233         put_task_struct(p);
1234
1235         return 0;
1236 }
1237
1238 static ssize_t
1239 sched_write(struct file *file, const char __user *buf,
1240             size_t count, loff_t *offset)
1241 {
1242         struct inode *inode = file->f_path.dentry->d_inode;
1243         struct task_struct *p;
1244
1245         p = get_proc_task(inode);
1246         if (!p)
1247                 return -ESRCH;
1248         proc_sched_set_task(p);
1249
1250         put_task_struct(p);
1251
1252         return count;
1253 }
1254
1255 static int sched_open(struct inode *inode, struct file *filp)
1256 {
1257         return single_open(filp, sched_show, inode);
1258 }
1259
1260 static const struct file_operations proc_pid_sched_operations = {
1261         .open           = sched_open,
1262         .read           = seq_read,
1263         .write          = sched_write,
1264         .llseek         = seq_lseek,
1265         .release        = single_release,
1266 };
1267
1268 #endif
1269
1270 #ifdef CONFIG_SCHED_AUTOGROUP
1271 /*
1272  * Print out autogroup related information:
1273  */
1274 static int sched_autogroup_show(struct seq_file *m, void *v)
1275 {
1276         struct inode *inode = m->private;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_autogroup_show_task(p, m);
1283
1284         put_task_struct(p);
1285
1286         return 0;
1287 }
1288
1289 static ssize_t
1290 sched_autogroup_write(struct file *file, const char __user *buf,
1291             size_t count, loff_t *offset)
1292 {
1293         struct inode *inode = file->f_path.dentry->d_inode;
1294         struct task_struct *p;
1295         char buffer[PROC_NUMBUF];
1296         int nice;
1297         int err;
1298
1299         memset(buffer, 0, sizeof(buffer));
1300         if (count > sizeof(buffer) - 1)
1301                 count = sizeof(buffer) - 1;
1302         if (copy_from_user(buffer, buf, count))
1303                 return -EFAULT;
1304
1305         err = kstrtoint(strstrip(buffer), 0, &nice);
1306         if (err < 0)
1307                 return err;
1308
1309         p = get_proc_task(inode);
1310         if (!p)
1311                 return -ESRCH;
1312
1313         err = proc_sched_autogroup_set_nice(p, nice);
1314         if (err)
1315                 count = err;
1316
1317         put_task_struct(p);
1318
1319         return count;
1320 }
1321
1322 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323 {
1324         int ret;
1325
1326         ret = single_open(filp, sched_autogroup_show, NULL);
1327         if (!ret) {
1328                 struct seq_file *m = filp->private_data;
1329
1330                 m->private = inode;
1331         }
1332         return ret;
1333 }
1334
1335 static const struct file_operations proc_pid_sched_autogroup_operations = {
1336         .open           = sched_autogroup_open,
1337         .read           = seq_read,
1338         .write          = sched_autogroup_write,
1339         .llseek         = seq_lseek,
1340         .release        = single_release,
1341 };
1342
1343 #endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345 static ssize_t comm_write(struct file *file, const char __user *buf,
1346                                 size_t count, loff_t *offset)
1347 {
1348         struct inode *inode = file->f_path.dentry->d_inode;
1349         struct task_struct *p;
1350         char buffer[TASK_COMM_LEN];
1351
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361
1362         if (same_thread_group(current, p))
1363                 set_task_comm(p, buffer);
1364         else
1365                 count = -EINVAL;
1366
1367         put_task_struct(p);
1368
1369         return count;
1370 }
1371
1372 static int comm_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380
1381         task_lock(p);
1382         seq_printf(m, "%s\n", p->comm);
1383         task_unlock(p);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static int comm_open(struct inode *inode, struct file *filp)
1391 {
1392         return single_open(filp, comm_show, inode);
1393 }
1394
1395 static const struct file_operations proc_pid_set_comm_operations = {
1396         .open           = comm_open,
1397         .read           = seq_read,
1398         .write          = comm_write,
1399         .llseek         = seq_lseek,
1400         .release        = single_release,
1401 };
1402
1403 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404 {
1405         struct task_struct *task;
1406         struct mm_struct *mm;
1407         struct file *exe_file;
1408
1409         task = get_proc_task(dentry->d_inode);
1410         if (!task)
1411                 return -ENOENT;
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414         if (!mm)
1415                 return -ENOENT;
1416         exe_file = get_mm_exe_file(mm);
1417         mmput(mm);
1418         if (exe_file) {
1419                 *exe_path = exe_file->f_path;
1420                 path_get(&exe_file->f_path);
1421                 fput(exe_file);
1422                 return 0;
1423         } else
1424                 return -ENOENT;
1425 }
1426
1427 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428 {
1429         struct inode *inode = dentry->d_inode;
1430         int error = -EACCES;
1431
1432         /* We don't need a base pointer in the /proc filesystem */
1433         path_put(&nd->path);
1434
1435         /* Are we allowed to snoop on the tasks file descriptors? */
1436         if (!proc_fd_access_allowed(inode))
1437                 goto out;
1438
1439         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1440 out:
1441         return ERR_PTR(error);
1442 }
1443
1444 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1445 {
1446         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1447         char *pathname;
1448         int len;
1449
1450         if (!tmp)
1451                 return -ENOMEM;
1452
1453         pathname = d_path(path, tmp, PAGE_SIZE);
1454         len = PTR_ERR(pathname);
1455         if (IS_ERR(pathname))
1456                 goto out;
1457         len = tmp + PAGE_SIZE - 1 - pathname;
1458
1459         if (len > buflen)
1460                 len = buflen;
1461         if (copy_to_user(buffer, pathname, len))
1462                 len = -EFAULT;
1463  out:
1464         free_page((unsigned long)tmp);
1465         return len;
1466 }
1467
1468 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1469 {
1470         int error = -EACCES;
1471         struct inode *inode = dentry->d_inode;
1472         struct path path;
1473
1474         /* Are we allowed to snoop on the tasks file descriptors? */
1475         if (!proc_fd_access_allowed(inode))
1476                 goto out;
1477
1478         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1479         if (error)
1480                 goto out;
1481
1482         error = do_proc_readlink(&path, buffer, buflen);
1483         path_put(&path);
1484 out:
1485         return error;
1486 }
1487
1488 static const struct inode_operations proc_pid_link_inode_operations = {
1489         .readlink       = proc_pid_readlink,
1490         .follow_link    = proc_pid_follow_link,
1491         .setattr        = proc_setattr,
1492 };
1493
1494
1495 /* building an inode */
1496
1497 static int task_dumpable(struct task_struct *task)
1498 {
1499         int dumpable = 0;
1500         struct mm_struct *mm;
1501
1502         task_lock(task);
1503         mm = task->mm;
1504         if (mm)
1505                 dumpable = get_dumpable(mm);
1506         task_unlock(task);
1507         if(dumpable == 1)
1508                 return 1;
1509         return 0;
1510 }
1511
1512 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1513 {
1514         struct inode * inode;
1515         struct proc_inode *ei;
1516         const struct cred *cred;
1517
1518         /* We need a new inode */
1519
1520         inode = new_inode(sb);
1521         if (!inode)
1522                 goto out;
1523
1524         /* Common stuff */
1525         ei = PROC_I(inode);
1526         inode->i_ino = get_next_ino();
1527         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528         inode->i_op = &proc_def_inode_operations;
1529
1530         /*
1531          * grab the reference to task.
1532          */
1533         ei->pid = get_task_pid(task, PIDTYPE_PID);
1534         if (!ei->pid)
1535                 goto out_unlock;
1536
1537         if (task_dumpable(task)) {
1538                 rcu_read_lock();
1539                 cred = __task_cred(task);
1540                 inode->i_uid = cred->euid;
1541                 inode->i_gid = cred->egid;
1542                 rcu_read_unlock();
1543         }
1544         security_task_to_inode(task, inode);
1545
1546 out:
1547         return inode;
1548
1549 out_unlock:
1550         iput(inode);
1551         return NULL;
1552 }
1553
1554 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1555 {
1556         struct inode *inode = dentry->d_inode;
1557         struct task_struct *task;
1558         const struct cred *cred;
1559         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1560
1561         generic_fillattr(inode, stat);
1562
1563         rcu_read_lock();
1564         stat->uid = 0;
1565         stat->gid = 0;
1566         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1567         if (task) {
1568                 if (!has_pid_permissions(pid, task, 2)) {
1569                         rcu_read_unlock();
1570                         /*
1571                          * This doesn't prevent learning whether PID exists,
1572                          * it only makes getattr() consistent with readdir().
1573                          */
1574                         return -ENOENT;
1575                 }
1576                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1577                     task_dumpable(task)) {
1578                         cred = __task_cred(task);
1579                         stat->uid = cred->euid;
1580                         stat->gid = cred->egid;
1581                 }
1582         }
1583         rcu_read_unlock();
1584         return 0;
1585 }
1586
1587 /* dentry stuff */
1588
1589 /*
1590  *      Exceptional case: normally we are not allowed to unhash a busy
1591  * directory. In this case, however, we can do it - no aliasing problems
1592  * due to the way we treat inodes.
1593  *
1594  * Rewrite the inode's ownerships here because the owning task may have
1595  * performed a setuid(), etc.
1596  *
1597  * Before the /proc/pid/status file was created the only way to read
1598  * the effective uid of a /process was to stat /proc/pid.  Reading
1599  * /proc/pid/status is slow enough that procps and other packages
1600  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1601  * made this apply to all per process world readable and executable
1602  * directories.
1603  */
1604 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1605 {
1606         struct inode *inode;
1607         struct task_struct *task;
1608         const struct cred *cred;
1609
1610         if (nd && nd->flags & LOOKUP_RCU)
1611                 return -ECHILD;
1612
1613         inode = dentry->d_inode;
1614         task = get_proc_task(inode);
1615
1616         if (task) {
1617                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1618                     task_dumpable(task)) {
1619                         rcu_read_lock();
1620                         cred = __task_cred(task);
1621                         inode->i_uid = cred->euid;
1622                         inode->i_gid = cred->egid;
1623                         rcu_read_unlock();
1624                 } else {
1625                         inode->i_uid = 0;
1626                         inode->i_gid = 0;
1627                 }
1628                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1629                 security_task_to_inode(task, inode);
1630                 put_task_struct(task);
1631                 return 1;
1632         }
1633         d_drop(dentry);
1634         return 0;
1635 }
1636
1637 static int pid_delete_dentry(const struct dentry * dentry)
1638 {
1639         /* Is the task we represent dead?
1640          * If so, then don't put the dentry on the lru list,
1641          * kill it immediately.
1642          */
1643         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1644 }
1645
1646 const struct dentry_operations pid_dentry_operations =
1647 {
1648         .d_revalidate   = pid_revalidate,
1649         .d_delete       = pid_delete_dentry,
1650 };
1651
1652 /* Lookups */
1653
1654 /*
1655  * Fill a directory entry.
1656  *
1657  * If possible create the dcache entry and derive our inode number and
1658  * file type from dcache entry.
1659  *
1660  * Since all of the proc inode numbers are dynamically generated, the inode
1661  * numbers do not exist until the inode is cache.  This means creating the
1662  * the dcache entry in readdir is necessary to keep the inode numbers
1663  * reported by readdir in sync with the inode numbers reported
1664  * by stat.
1665  */
1666 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1667         const char *name, int len,
1668         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669 {
1670         struct dentry *child, *dir = filp->f_path.dentry;
1671         struct inode *inode;
1672         struct qstr qname;
1673         ino_t ino = 0;
1674         unsigned type = DT_UNKNOWN;
1675
1676         qname.name = name;
1677         qname.len  = len;
1678         qname.hash = full_name_hash(name, len);
1679
1680         child = d_lookup(dir, &qname);
1681         if (!child) {
1682                 struct dentry *new;
1683                 new = d_alloc(dir, &qname);
1684                 if (new) {
1685                         child = instantiate(dir->d_inode, new, task, ptr);
1686                         if (child)
1687                                 dput(new);
1688                         else
1689                                 child = new;
1690                 }
1691         }
1692         if (!child || IS_ERR(child) || !child->d_inode)
1693                 goto end_instantiate;
1694         inode = child->d_inode;
1695         if (inode) {
1696                 ino = inode->i_ino;
1697                 type = inode->i_mode >> 12;
1698         }
1699         dput(child);
1700 end_instantiate:
1701         if (!ino)
1702                 ino = find_inode_number(dir, &qname);
1703         if (!ino)
1704                 ino = 1;
1705         return filldir(dirent, name, len, filp->f_pos, ino, type);
1706 }
1707
1708 static unsigned name_to_int(struct dentry *dentry)
1709 {
1710         const char *name = dentry->d_name.name;
1711         int len = dentry->d_name.len;
1712         unsigned n = 0;
1713
1714         if (len > 1 && *name == '0')
1715                 goto out;
1716         while (len-- > 0) {
1717                 unsigned c = *name++ - '0';
1718                 if (c > 9)
1719                         goto out;
1720                 if (n >= (~0U-9)/10)
1721                         goto out;
1722                 n *= 10;
1723                 n += c;
1724         }
1725         return n;
1726 out:
1727         return ~0U;
1728 }
1729
1730 #define PROC_FDINFO_MAX 64
1731
1732 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1733 {
1734         struct task_struct *task = get_proc_task(inode);
1735         struct files_struct *files = NULL;
1736         struct file *file;
1737         int fd = proc_fd(inode);
1738
1739         if (task) {
1740                 files = get_files_struct(task);
1741                 put_task_struct(task);
1742         }
1743         if (files) {
1744                 /*
1745                  * We are not taking a ref to the file structure, so we must
1746                  * hold ->file_lock.
1747                  */
1748                 spin_lock(&files->file_lock);
1749                 file = fcheck_files(files, fd);
1750                 if (file) {
1751                         unsigned int f_flags;
1752                         struct fdtable *fdt;
1753
1754                         fdt = files_fdtable(files);
1755                         f_flags = file->f_flags & ~O_CLOEXEC;
1756                         if (close_on_exec(fd, fdt))
1757                                 f_flags |= O_CLOEXEC;
1758
1759                         if (path) {
1760                                 *path = file->f_path;
1761                                 path_get(&file->f_path);
1762                         }
1763                         if (info)
1764                                 snprintf(info, PROC_FDINFO_MAX,
1765                                          "pos:\t%lli\n"
1766                                          "flags:\t0%o\n",
1767                                          (long long) file->f_pos,
1768                                          f_flags);
1769                         spin_unlock(&files->file_lock);
1770                         put_files_struct(files);
1771                         return 0;
1772                 }
1773                 spin_unlock(&files->file_lock);
1774                 put_files_struct(files);
1775         }
1776         return -ENOENT;
1777 }
1778
1779 static int proc_fd_link(struct dentry *dentry, struct path *path)
1780 {
1781         return proc_fd_info(dentry->d_inode, path, NULL);
1782 }
1783
1784 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1785 {
1786         struct inode *inode;
1787         struct task_struct *task;
1788         int fd;
1789         struct files_struct *files;
1790         const struct cred *cred;
1791
1792         if (nd && nd->flags & LOOKUP_RCU)
1793                 return -ECHILD;
1794
1795         inode = dentry->d_inode;
1796         task = get_proc_task(inode);
1797         fd = proc_fd(inode);
1798
1799         if (task) {
1800                 files = get_files_struct(task);
1801                 if (files) {
1802                         struct file *file;
1803                         rcu_read_lock();
1804                         file = fcheck_files(files, fd);
1805                         if (file) {
1806                                 unsigned i_mode, f_mode = file->f_mode;
1807
1808                                 rcu_read_unlock();
1809                                 put_files_struct(files);
1810
1811                                 if (task_dumpable(task)) {
1812                                         rcu_read_lock();
1813                                         cred = __task_cred(task);
1814                                         inode->i_uid = cred->euid;
1815                                         inode->i_gid = cred->egid;
1816                                         rcu_read_unlock();
1817                                 } else {
1818                                         inode->i_uid = 0;
1819                                         inode->i_gid = 0;
1820                                 }
1821
1822                                 i_mode = S_IFLNK;
1823                                 if (f_mode & FMODE_READ)
1824                                         i_mode |= S_IRUSR | S_IXUSR;
1825                                 if (f_mode & FMODE_WRITE)
1826                                         i_mode |= S_IWUSR | S_IXUSR;
1827                                 inode->i_mode = i_mode;
1828
1829                                 security_task_to_inode(task, inode);
1830                                 put_task_struct(task);
1831                                 return 1;
1832                         }
1833                         rcu_read_unlock();
1834                         put_files_struct(files);
1835                 }
1836                 put_task_struct(task);
1837         }
1838         d_drop(dentry);
1839         return 0;
1840 }
1841
1842 static const struct dentry_operations tid_fd_dentry_operations =
1843 {
1844         .d_revalidate   = tid_fd_revalidate,
1845         .d_delete       = pid_delete_dentry,
1846 };
1847
1848 static struct dentry *proc_fd_instantiate(struct inode *dir,
1849         struct dentry *dentry, struct task_struct *task, const void *ptr)
1850 {
1851         unsigned fd = *(const unsigned *)ptr;
1852         struct inode *inode;
1853         struct proc_inode *ei;
1854         struct dentry *error = ERR_PTR(-ENOENT);
1855
1856         inode = proc_pid_make_inode(dir->i_sb, task);
1857         if (!inode)
1858                 goto out;
1859         ei = PROC_I(inode);
1860         ei->fd = fd;
1861
1862         inode->i_op = &proc_pid_link_inode_operations;
1863         inode->i_size = 64;
1864         ei->op.proc_get_link = proc_fd_link;
1865         d_set_d_op(dentry, &tid_fd_dentry_operations);
1866         d_add(dentry, inode);
1867         /* Close the race of the process dying before we return the dentry */
1868         if (tid_fd_revalidate(dentry, NULL))
1869                 error = NULL;
1870
1871  out:
1872         return error;
1873 }
1874
1875 static struct dentry *proc_lookupfd_common(struct inode *dir,
1876                                            struct dentry *dentry,
1877                                            instantiate_t instantiate)
1878 {
1879         struct task_struct *task = get_proc_task(dir);
1880         unsigned fd = name_to_int(dentry);
1881         struct dentry *result = ERR_PTR(-ENOENT);
1882
1883         if (!task)
1884                 goto out_no_task;
1885         if (fd == ~0U)
1886                 goto out;
1887
1888         result = instantiate(dir, dentry, task, &fd);
1889 out:
1890         put_task_struct(task);
1891 out_no_task:
1892         return result;
1893 }
1894
1895 static int proc_readfd_common(struct file * filp, void * dirent,
1896                               filldir_t filldir, instantiate_t instantiate)
1897 {
1898         struct dentry *dentry = filp->f_path.dentry;
1899         struct inode *inode = dentry->d_inode;
1900         struct task_struct *p = get_proc_task(inode);
1901         unsigned int fd, ino;
1902         int retval;
1903         struct files_struct * files;
1904
1905         retval = -ENOENT;
1906         if (!p)
1907                 goto out_no_task;
1908         retval = 0;
1909
1910         fd = filp->f_pos;
1911         switch (fd) {
1912                 case 0:
1913                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1914                                 goto out;
1915                         filp->f_pos++;
1916                 case 1:
1917                         ino = parent_ino(dentry);
1918                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1919                                 goto out;
1920                         filp->f_pos++;
1921                 default:
1922                         files = get_files_struct(p);
1923                         if (!files)
1924                                 goto out;
1925                         rcu_read_lock();
1926                         for (fd = filp->f_pos-2;
1927                              fd < files_fdtable(files)->max_fds;
1928                              fd++, filp->f_pos++) {
1929                                 char name[PROC_NUMBUF];
1930                                 int len;
1931
1932                                 if (!fcheck_files(files, fd))
1933                                         continue;
1934                                 rcu_read_unlock();
1935
1936                                 len = snprintf(name, sizeof(name), "%d", fd);
1937                                 if (proc_fill_cache(filp, dirent, filldir,
1938                                                     name, len, instantiate,
1939                                                     p, &fd) < 0) {
1940                                         rcu_read_lock();
1941                                         break;
1942                                 }
1943                                 rcu_read_lock();
1944                         }
1945                         rcu_read_unlock();
1946                         put_files_struct(files);
1947         }
1948 out:
1949         put_task_struct(p);
1950 out_no_task:
1951         return retval;
1952 }
1953
1954 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1955                                     struct nameidata *nd)
1956 {
1957         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1958 }
1959
1960 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1961 {
1962         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1963 }
1964
1965 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1966                                       size_t len, loff_t *ppos)
1967 {
1968         char tmp[PROC_FDINFO_MAX];
1969         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1970         if (!err)
1971                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1972         return err;
1973 }
1974
1975 static const struct file_operations proc_fdinfo_file_operations = {
1976         .open           = nonseekable_open,
1977         .read           = proc_fdinfo_read,
1978         .llseek         = no_llseek,
1979 };
1980
1981 static const struct file_operations proc_fd_operations = {
1982         .read           = generic_read_dir,
1983         .readdir        = proc_readfd,
1984         .llseek         = default_llseek,
1985 };
1986
1987 #ifdef CONFIG_CHECKPOINT_RESTORE
1988
1989 /*
1990  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1991  * which represent vma start and end addresses.
1992  */
1993 static int dname_to_vma_addr(struct dentry *dentry,
1994                              unsigned long *start, unsigned long *end)
1995 {
1996         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1997                 return -EINVAL;
1998
1999         return 0;
2000 }
2001
2002 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2003 {
2004         unsigned long vm_start, vm_end;
2005         bool exact_vma_exists = false;
2006         struct mm_struct *mm = NULL;
2007         struct task_struct *task;
2008         const struct cred *cred;
2009         struct inode *inode;
2010         int status = 0;
2011
2012         if (nd && nd->flags & LOOKUP_RCU)
2013                 return -ECHILD;
2014
2015         if (!capable(CAP_SYS_ADMIN)) {
2016                 status = -EACCES;
2017                 goto out_notask;
2018         }
2019
2020         inode = dentry->d_inode;
2021         task = get_proc_task(inode);
2022         if (!task)
2023                 goto out_notask;
2024
2025         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2026                 goto out;
2027
2028         mm = get_task_mm(task);
2029         if (!mm)
2030                 goto out;
2031
2032         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2033                 down_read(&mm->mmap_sem);
2034                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2035                 up_read(&mm->mmap_sem);
2036         }
2037
2038         mmput(mm);
2039
2040         if (exact_vma_exists) {
2041                 if (task_dumpable(task)) {
2042                         rcu_read_lock();
2043                         cred = __task_cred(task);
2044                         inode->i_uid = cred->euid;
2045                         inode->i_gid = cred->egid;
2046                         rcu_read_unlock();
2047                 } else {
2048                         inode->i_uid = 0;
2049                         inode->i_gid = 0;
2050                 }
2051                 security_task_to_inode(task, inode);
2052                 status = 1;
2053         }
2054
2055 out:
2056         put_task_struct(task);
2057
2058 out_notask:
2059         if (status <= 0)
2060                 d_drop(dentry);
2061
2062         return status;
2063 }
2064
2065 static const struct dentry_operations tid_map_files_dentry_operations = {
2066         .d_revalidate   = map_files_d_revalidate,
2067         .d_delete       = pid_delete_dentry,
2068 };
2069
2070 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2071 {
2072         unsigned long vm_start, vm_end;
2073         struct vm_area_struct *vma;
2074         struct task_struct *task;
2075         struct mm_struct *mm;
2076         int rc;
2077
2078         rc = -ENOENT;
2079         task = get_proc_task(dentry->d_inode);
2080         if (!task)
2081                 goto out;
2082
2083         mm = get_task_mm(task);
2084         put_task_struct(task);
2085         if (!mm)
2086                 goto out;
2087
2088         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2089         if (rc)
2090                 goto out_mmput;
2091
2092         down_read(&mm->mmap_sem);
2093         vma = find_exact_vma(mm, vm_start, vm_end);
2094         if (vma && vma->vm_file) {
2095                 *path = vma->vm_file->f_path;
2096                 path_get(path);
2097                 rc = 0;
2098         }
2099         up_read(&mm->mmap_sem);
2100
2101 out_mmput:
2102         mmput(mm);
2103 out:
2104         return rc;
2105 }
2106
2107 struct map_files_info {
2108         struct file     *file;
2109         unsigned long   len;
2110         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2111 };
2112
2113 static struct dentry *
2114 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2115                            struct task_struct *task, const void *ptr)
2116 {
2117         const struct file *file = ptr;
2118         struct proc_inode *ei;
2119         struct inode *inode;
2120
2121         if (!file)
2122                 return ERR_PTR(-ENOENT);
2123
2124         inode = proc_pid_make_inode(dir->i_sb, task);
2125         if (!inode)
2126                 return ERR_PTR(-ENOENT);
2127
2128         ei = PROC_I(inode);
2129         ei->op.proc_get_link = proc_map_files_get_link;
2130
2131         inode->i_op = &proc_pid_link_inode_operations;
2132         inode->i_size = 64;
2133         inode->i_mode = S_IFLNK;
2134
2135         if (file->f_mode & FMODE_READ)
2136                 inode->i_mode |= S_IRUSR;
2137         if (file->f_mode & FMODE_WRITE)
2138                 inode->i_mode |= S_IWUSR;
2139
2140         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2141         d_add(dentry, inode);
2142
2143         return NULL;
2144 }
2145
2146 static struct dentry *proc_map_files_lookup(struct inode *dir,
2147                 struct dentry *dentry, struct nameidata *nd)
2148 {
2149         unsigned long vm_start, vm_end;
2150         struct vm_area_struct *vma;
2151         struct task_struct *task;
2152         struct dentry *result;
2153         struct mm_struct *mm;
2154
2155         result = ERR_PTR(-EACCES);
2156         if (!capable(CAP_SYS_ADMIN))
2157                 goto out;
2158
2159         result = ERR_PTR(-ENOENT);
2160         task = get_proc_task(dir);
2161         if (!task)
2162                 goto out;
2163
2164         result = ERR_PTR(-EACCES);
2165         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2166                 goto out_put_task;
2167
2168         result = ERR_PTR(-ENOENT);
2169         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2170                 goto out_put_task;
2171
2172         mm = get_task_mm(task);
2173         if (!mm)
2174                 goto out_put_task;
2175
2176         down_read(&mm->mmap_sem);
2177         vma = find_exact_vma(mm, vm_start, vm_end);
2178         if (!vma)
2179                 goto out_no_vma;
2180
2181         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2182
2183 out_no_vma:
2184         up_read(&mm->mmap_sem);
2185         mmput(mm);
2186 out_put_task:
2187         put_task_struct(task);
2188 out:
2189         return result;
2190 }
2191
2192 static const struct inode_operations proc_map_files_inode_operations = {
2193         .lookup         = proc_map_files_lookup,
2194         .permission     = proc_fd_permission,
2195         .setattr        = proc_setattr,
2196 };
2197
2198 static int
2199 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2200 {
2201         struct dentry *dentry = filp->f_path.dentry;
2202         struct inode *inode = dentry->d_inode;
2203         struct vm_area_struct *vma;
2204         struct task_struct *task;
2205         struct mm_struct *mm;
2206         ino_t ino;
2207         int ret;
2208
2209         ret = -EACCES;
2210         if (!capable(CAP_SYS_ADMIN))
2211                 goto out;
2212
2213         ret = -ENOENT;
2214         task = get_proc_task(inode);
2215         if (!task)
2216                 goto out;
2217
2218         ret = -EACCES;
2219         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2220                 goto out_put_task;
2221
2222         ret = 0;
2223         switch (filp->f_pos) {
2224         case 0:
2225                 ino = inode->i_ino;
2226                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2227                         goto out_put_task;
2228                 filp->f_pos++;
2229         case 1:
2230                 ino = parent_ino(dentry);
2231                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2232                         goto out_put_task;
2233                 filp->f_pos++;
2234         default:
2235         {
2236                 unsigned long nr_files, pos, i;
2237                 struct flex_array *fa = NULL;
2238                 struct map_files_info info;
2239                 struct map_files_info *p;
2240
2241                 mm = get_task_mm(task);
2242                 if (!mm)
2243                         goto out_put_task;
2244                 down_read(&mm->mmap_sem);
2245
2246                 nr_files = 0;
2247
2248                 /*
2249                  * We need two passes here:
2250                  *
2251                  *  1) Collect vmas of mapped files with mmap_sem taken
2252                  *  2) Release mmap_sem and instantiate entries
2253                  *
2254                  * otherwise we get lockdep complained, since filldir()
2255                  * routine might require mmap_sem taken in might_fault().
2256                  */
2257
2258                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2259                         if (vma->vm_file && ++pos > filp->f_pos)
2260                                 nr_files++;
2261                 }
2262
2263                 if (nr_files) {
2264                         fa = flex_array_alloc(sizeof(info), nr_files,
2265                                                 GFP_KERNEL);
2266                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2267                                                         GFP_KERNEL)) {
2268                                 ret = -ENOMEM;
2269                                 if (fa)
2270                                         flex_array_free(fa);
2271                                 up_read(&mm->mmap_sem);
2272                                 mmput(mm);
2273                                 goto out_put_task;
2274                         }
2275                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2276                                         vma = vma->vm_next) {
2277                                 if (!vma->vm_file)
2278                                         continue;
2279                                 if (++pos <= filp->f_pos)
2280                                         continue;
2281
2282                                 get_file(vma->vm_file);
2283                                 info.file = vma->vm_file;
2284                                 info.len = snprintf(info.name,
2285                                                 sizeof(info.name), "%lx-%lx",
2286                                                 vma->vm_start, vma->vm_end);
2287                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2288                                         BUG();
2289                         }
2290                 }
2291                 up_read(&mm->mmap_sem);
2292
2293                 for (i = 0; i < nr_files; i++) {
2294                         p = flex_array_get(fa, i);
2295                         ret = proc_fill_cache(filp, dirent, filldir,
2296                                               p->name, p->len,
2297                                               proc_map_files_instantiate,
2298                                               task, p->file);
2299                         if (ret)
2300                                 break;
2301                         filp->f_pos++;
2302                         fput(p->file);
2303                 }
2304                 for (; i < nr_files; i++) {
2305                         /*
2306                          * In case of error don't forget
2307                          * to put rest of file refs.
2308                          */
2309                         p = flex_array_get(fa, i);
2310                         fput(p->file);
2311                 }
2312                 if (fa)
2313                         flex_array_free(fa);
2314                 mmput(mm);
2315         }
2316         }
2317
2318 out_put_task:
2319         put_task_struct(task);
2320 out:
2321         return ret;
2322 }
2323
2324 static const struct file_operations proc_map_files_operations = {
2325         .read           = generic_read_dir,
2326         .readdir        = proc_map_files_readdir,
2327         .llseek         = default_llseek,
2328 };
2329
2330 #endif /* CONFIG_CHECKPOINT_RESTORE */
2331
2332 /*
2333  * /proc/pid/fd needs a special permission handler so that a process can still
2334  * access /proc/self/fd after it has executed a setuid().
2335  */
2336 static int proc_fd_permission(struct inode *inode, int mask)
2337 {
2338         int rv = generic_permission(inode, mask);
2339         if (rv == 0)
2340                 return 0;
2341         if (task_pid(current) == proc_pid(inode))
2342                 rv = 0;
2343         return rv;
2344 }
2345
2346 /*
2347  * proc directories can do almost nothing..
2348  */
2349 static const struct inode_operations proc_fd_inode_operations = {
2350         .lookup         = proc_lookupfd,
2351         .permission     = proc_fd_permission,
2352         .setattr        = proc_setattr,
2353 };
2354
2355 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2356         struct dentry *dentry, struct task_struct *task, const void *ptr)
2357 {
2358         unsigned fd = *(unsigned *)ptr;
2359         struct inode *inode;
2360         struct proc_inode *ei;
2361         struct dentry *error = ERR_PTR(-ENOENT);
2362
2363         inode = proc_pid_make_inode(dir->i_sb, task);
2364         if (!inode)
2365                 goto out;
2366         ei = PROC_I(inode);
2367         ei->fd = fd;
2368         inode->i_mode = S_IFREG | S_IRUSR;
2369         inode->i_fop = &proc_fdinfo_file_operations;
2370         d_set_d_op(dentry, &tid_fd_dentry_operations);
2371         d_add(dentry, inode);
2372         /* Close the race of the process dying before we return the dentry */
2373         if (tid_fd_revalidate(dentry, NULL))
2374                 error = NULL;
2375
2376  out:
2377         return error;
2378 }
2379
2380 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2381                                         struct dentry *dentry,
2382                                         struct nameidata *nd)
2383 {
2384         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2385 }
2386
2387 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2388 {
2389         return proc_readfd_common(filp, dirent, filldir,
2390                                   proc_fdinfo_instantiate);
2391 }
2392
2393 static const struct file_operations proc_fdinfo_operations = {
2394         .read           = generic_read_dir,
2395         .readdir        = proc_readfdinfo,
2396         .llseek         = default_llseek,
2397 };
2398
2399 /*
2400  * proc directories can do almost nothing..
2401  */
2402 static const struct inode_operations proc_fdinfo_inode_operations = {
2403         .lookup         = proc_lookupfdinfo,
2404         .setattr        = proc_setattr,
2405 };
2406
2407
2408 static struct dentry *proc_pident_instantiate(struct inode *dir,
2409         struct dentry *dentry, struct task_struct *task, const void *ptr)
2410 {
2411         const struct pid_entry *p = ptr;
2412         struct inode *inode;
2413         struct proc_inode *ei;
2414         struct dentry *error = ERR_PTR(-ENOENT);
2415
2416         inode = proc_pid_make_inode(dir->i_sb, task);
2417         if (!inode)
2418                 goto out;
2419
2420         ei = PROC_I(inode);
2421         inode->i_mode = p->mode;
2422         if (S_ISDIR(inode->i_mode))
2423                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2424         if (p->iop)
2425                 inode->i_op = p->iop;
2426         if (p->fop)
2427                 inode->i_fop = p->fop;
2428         ei->op = p->op;
2429         d_set_d_op(dentry, &pid_dentry_operations);
2430         d_add(dentry, inode);
2431         /* Close the race of the process dying before we return the dentry */
2432         if (pid_revalidate(dentry, NULL))
2433                 error = NULL;
2434 out:
2435         return error;
2436 }
2437
2438 static struct dentry *proc_pident_lookup(struct inode *dir, 
2439                                          struct dentry *dentry,
2440                                          const struct pid_entry *ents,
2441                                          unsigned int nents)
2442 {
2443         struct dentry *error;
2444         struct task_struct *task = get_proc_task(dir);
2445         const struct pid_entry *p, *last;
2446
2447         error = ERR_PTR(-ENOENT);
2448
2449         if (!task)
2450                 goto out_no_task;
2451
2452         /*
2453          * Yes, it does not scale. And it should not. Don't add
2454          * new entries into /proc/<tgid>/ without very good reasons.
2455          */
2456         last = &ents[nents - 1];
2457         for (p = ents; p <= last; p++) {
2458                 if (p->len != dentry->d_name.len)
2459                         continue;
2460                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2461                         break;
2462         }
2463         if (p > last)
2464                 goto out;
2465
2466         error = proc_pident_instantiate(dir, dentry, task, p);
2467 out:
2468         put_task_struct(task);
2469 out_no_task:
2470         return error;
2471 }
2472
2473 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2474         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2475 {
2476         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2477                                 proc_pident_instantiate, task, p);
2478 }
2479
2480 static int proc_pident_readdir(struct file *filp,
2481                 void *dirent, filldir_t filldir,
2482                 const struct pid_entry *ents, unsigned int nents)
2483 {
2484         int i;
2485         struct dentry *dentry = filp->f_path.dentry;
2486         struct inode *inode = dentry->d_inode;
2487         struct task_struct *task = get_proc_task(inode);
2488         const struct pid_entry *p, *last;
2489         ino_t ino;
2490         int ret;
2491
2492         ret = -ENOENT;
2493         if (!task)
2494                 goto out_no_task;
2495
2496         ret = 0;