close race in /proc/*/environ
[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 "internal.h"
87
88 /* NOTE:
89  *      Implementing inode permission operations in /proc is almost
90  *      certainly an error.  Permission checks need to happen during
91  *      each system call not at open time.  The reason is that most of
92  *      what we wish to check for permissions in /proc varies at runtime.
93  *
94  *      The classic example of a problem is opening file descriptors
95  *      in /proc for a task before it execs a suid executable.
96  */
97
98 struct pid_entry {
99         char *name;
100         int len;
101         mode_t mode;
102         const struct inode_operations *iop;
103         const struct file_operations *fop;
104         union proc_op op;
105 };
106
107 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
108         .name = (NAME),                                 \
109         .len  = sizeof(NAME) - 1,                       \
110         .mode = MODE,                                   \
111         .iop  = IOP,                                    \
112         .fop  = FOP,                                    \
113         .op   = OP,                                     \
114 }
115
116 #define DIR(NAME, MODE, iops, fops)     \
117         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
118 #define LNK(NAME, get_link)                                     \
119         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
120                 &proc_pid_link_inode_operations, NULL,          \
121                 { .proc_get_link = get_link } )
122 #define REG(NAME, MODE, fops)                           \
123         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
124 #define INF(NAME, MODE, read)                           \
125         NOD(NAME, (S_IFREG|(MODE)),                     \
126                 NULL, &proc_info_file_operations,       \
127                 { .proc_read = read } )
128 #define ONE(NAME, MODE, show)                           \
129         NOD(NAME, (S_IFREG|(MODE)),                     \
130                 NULL, &proc_single_file_operations,     \
131                 { .proc_show = show } )
132
133 /*
134  * Count the number of hardlinks for the pid_entry table, excluding the .
135  * and .. links.
136  */
137 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
138         unsigned int n)
139 {
140         unsigned int i;
141         unsigned int count;
142
143         count = 0;
144         for (i = 0; i < n; ++i) {
145                 if (S_ISDIR(entries[i].mode))
146                         ++count;
147         }
148
149         return count;
150 }
151
152 static int get_task_root(struct task_struct *task, struct path *root)
153 {
154         int result = -ENOENT;
155
156         task_lock(task);
157         if (task->fs) {
158                 get_fs_root(task->fs, root);
159                 result = 0;
160         }
161         task_unlock(task);
162         return result;
163 }
164
165 static int proc_cwd_link(struct inode *inode, struct path *path)
166 {
167         struct task_struct *task = get_proc_task(inode);
168         int result = -ENOENT;
169
170         if (task) {
171                 task_lock(task);
172                 if (task->fs) {
173                         get_fs_pwd(task->fs, path);
174                         result = 0;
175                 }
176                 task_unlock(task);
177                 put_task_struct(task);
178         }
179         return result;
180 }
181
182 static int proc_root_link(struct inode *inode, struct path *path)
183 {
184         struct task_struct *task = get_proc_task(inode);
185         int result = -ENOENT;
186
187         if (task) {
188                 result = get_task_root(task, path);
189                 put_task_struct(task);
190         }
191         return result;
192 }
193
194 /*
195  * Return zero if current may access user memory in @task, -error if not.
196  */
197 static int check_mem_permission(struct task_struct *task)
198 {
199         /*
200          * A task can always look at itself, in case it chooses
201          * to use system calls instead of load instructions.
202          */
203         if (task == current)
204                 return 0;
205
206         /*
207          * If current is actively ptrace'ing, and would also be
208          * permitted to freshly attach with ptrace now, permit it.
209          */
210         if (task_is_stopped_or_traced(task)) {
211                 int match;
212                 rcu_read_lock();
213                 match = (tracehook_tracer_task(task) == current);
214                 rcu_read_unlock();
215                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
216                         return 0;
217         }
218
219         /*
220          * Noone else is allowed.
221          */
222         return -EPERM;
223 }
224
225 struct mm_struct *mm_for_maps(struct task_struct *task)
226 {
227         struct mm_struct *mm;
228         int err;
229
230         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
231         if (err)
232                 return ERR_PTR(err);
233
234         mm = get_task_mm(task);
235         if (mm && mm != current->mm &&
236                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
237                 mmput(mm);
238                 mm = ERR_PTR(-EACCES);
239         }
240         mutex_unlock(&task->signal->cred_guard_mutex);
241
242         return mm;
243 }
244
245 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
246 {
247         int res = 0;
248         unsigned int len;
249         struct mm_struct *mm = get_task_mm(task);
250         if (!mm)
251                 goto out;
252         if (!mm->arg_end)
253                 goto out_mm;    /* Shh! No looking before we're done */
254
255         len = mm->arg_end - mm->arg_start;
256  
257         if (len > PAGE_SIZE)
258                 len = PAGE_SIZE;
259  
260         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
261
262         // If the nul at the end of args has been overwritten, then
263         // assume application is using setproctitle(3).
264         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
265                 len = strnlen(buffer, res);
266                 if (len < res) {
267                     res = len;
268                 } else {
269                         len = mm->env_end - mm->env_start;
270                         if (len > PAGE_SIZE - res)
271                                 len = PAGE_SIZE - res;
272                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
273                         res = strnlen(buffer, res);
274                 }
275         }
276 out_mm:
277         mmput(mm);
278 out:
279         return res;
280 }
281
282 static int proc_pid_auxv(struct task_struct *task, char *buffer)
283 {
284         int res = 0;
285         struct mm_struct *mm = get_task_mm(task);
286         if (mm) {
287                 unsigned int nwords = 0;
288                 do {
289                         nwords += 2;
290                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
291                 res = nwords * sizeof(mm->saved_auxv[0]);
292                 if (res > PAGE_SIZE)
293                         res = PAGE_SIZE;
294                 memcpy(buffer, mm->saved_auxv, res);
295                 mmput(mm);
296         }
297         return res;
298 }
299
300
301 #ifdef CONFIG_KALLSYMS
302 /*
303  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
304  * Returns the resolved symbol.  If that fails, simply return the address.
305  */
306 static int proc_pid_wchan(struct task_struct *task, char *buffer)
307 {
308         unsigned long wchan;
309         char symname[KSYM_NAME_LEN];
310
311         wchan = get_wchan(task);
312
313         if (lookup_symbol_name(wchan, symname) < 0)
314                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
315                         return 0;
316                 else
317                         return sprintf(buffer, "%lu", wchan);
318         else
319                 return sprintf(buffer, "%s", symname);
320 }
321 #endif /* CONFIG_KALLSYMS */
322
323 #ifdef CONFIG_STACKTRACE
324
325 #define MAX_STACK_TRACE_DEPTH   64
326
327 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
328                           struct pid *pid, struct task_struct *task)
329 {
330         struct stack_trace trace;
331         unsigned long *entries;
332         int i;
333
334         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
335         if (!entries)
336                 return -ENOMEM;
337
338         trace.nr_entries        = 0;
339         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
340         trace.entries           = entries;
341         trace.skip              = 0;
342         save_stack_trace_tsk(task, &trace);
343
344         for (i = 0; i < trace.nr_entries; i++) {
345                 seq_printf(m, "[<%p>] %pS\n",
346                            (void *)entries[i], (void *)entries[i]);
347         }
348         kfree(entries);
349
350         return 0;
351 }
352 #endif
353
354 #ifdef CONFIG_SCHEDSTATS
355 /*
356  * Provides /proc/PID/schedstat
357  */
358 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
359 {
360         return sprintf(buffer, "%llu %llu %lu\n",
361                         (unsigned long long)task->se.sum_exec_runtime,
362                         (unsigned long long)task->sched_info.run_delay,
363                         task->sched_info.pcount);
364 }
365 #endif
366
367 #ifdef CONFIG_LATENCYTOP
368 static int lstats_show_proc(struct seq_file *m, void *v)
369 {
370         int i;
371         struct inode *inode = m->private;
372         struct task_struct *task = get_proc_task(inode);
373
374         if (!task)
375                 return -ESRCH;
376         seq_puts(m, "Latency Top version : v0.1\n");
377         for (i = 0; i < 32; i++) {
378                 struct latency_record *lr = &task->latency_record[i];
379                 if (lr->backtrace[0]) {
380                         int q;
381                         seq_printf(m, "%i %li %li",
382                                    lr->count, lr->time, lr->max);
383                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
384                                 unsigned long bt = lr->backtrace[q];
385                                 if (!bt)
386                                         break;
387                                 if (bt == ULONG_MAX)
388                                         break;
389                                 seq_printf(m, " %ps", (void *)bt);
390                         }
391                         seq_putc(m, '\n');
392                 }
393
394         }
395         put_task_struct(task);
396         return 0;
397 }
398
399 static int lstats_open(struct inode *inode, struct file *file)
400 {
401         return single_open(file, lstats_show_proc, inode);
402 }
403
404 static ssize_t lstats_write(struct file *file, const char __user *buf,
405                             size_t count, loff_t *offs)
406 {
407         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
408
409         if (!task)
410                 return -ESRCH;
411         clear_all_latency_tracing(task);
412         put_task_struct(task);
413
414         return count;
415 }
416
417 static const struct file_operations proc_lstats_operations = {
418         .open           = lstats_open,
419         .read           = seq_read,
420         .write          = lstats_write,
421         .llseek         = seq_lseek,
422         .release        = single_release,
423 };
424
425 #endif
426
427 static int proc_oom_score(struct task_struct *task, char *buffer)
428 {
429         unsigned long points = 0;
430
431         read_lock(&tasklist_lock);
432         if (pid_alive(task))
433                 points = oom_badness(task, NULL, NULL,
434                                         totalram_pages + total_swap_pages);
435         read_unlock(&tasklist_lock);
436         return sprintf(buffer, "%lu\n", points);
437 }
438
439 struct limit_names {
440         char *name;
441         char *unit;
442 };
443
444 static const struct limit_names lnames[RLIM_NLIMITS] = {
445         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
446         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
447         [RLIMIT_DATA] = {"Max data size", "bytes"},
448         [RLIMIT_STACK] = {"Max stack size", "bytes"},
449         [RLIMIT_CORE] = {"Max core file size", "bytes"},
450         [RLIMIT_RSS] = {"Max resident set", "bytes"},
451         [RLIMIT_NPROC] = {"Max processes", "processes"},
452         [RLIMIT_NOFILE] = {"Max open files", "files"},
453         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
454         [RLIMIT_AS] = {"Max address space", "bytes"},
455         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
456         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
457         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
458         [RLIMIT_NICE] = {"Max nice priority", NULL},
459         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
460         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
461 };
462
463 /* Display limits for a process */
464 static int proc_pid_limits(struct task_struct *task, char *buffer)
465 {
466         unsigned int i;
467         int count = 0;
468         unsigned long flags;
469         char *bufptr = buffer;
470
471         struct rlimit rlim[RLIM_NLIMITS];
472
473         if (!lock_task_sighand(task, &flags))
474                 return 0;
475         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
476         unlock_task_sighand(task, &flags);
477
478         /*
479          * print the file header
480          */
481         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
482                         "Limit", "Soft Limit", "Hard Limit", "Units");
483
484         for (i = 0; i < RLIM_NLIMITS; i++) {
485                 if (rlim[i].rlim_cur == RLIM_INFINITY)
486                         count += sprintf(&bufptr[count], "%-25s %-20s ",
487                                          lnames[i].name, "unlimited");
488                 else
489                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
490                                          lnames[i].name, rlim[i].rlim_cur);
491
492                 if (rlim[i].rlim_max == RLIM_INFINITY)
493                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
494                 else
495                         count += sprintf(&bufptr[count], "%-20lu ",
496                                          rlim[i].rlim_max);
497
498                 if (lnames[i].unit)
499                         count += sprintf(&bufptr[count], "%-10s\n",
500                                          lnames[i].unit);
501                 else
502                         count += sprintf(&bufptr[count], "\n");
503         }
504
505         return count;
506 }
507
508 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
509 static int proc_pid_syscall(struct task_struct *task, char *buffer)
510 {
511         long nr;
512         unsigned long args[6], sp, pc;
513
514         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
515                 return sprintf(buffer, "running\n");
516
517         if (nr < 0)
518                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
519
520         return sprintf(buffer,
521                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
522                        nr,
523                        args[0], args[1], args[2], args[3], args[4], args[5],
524                        sp, pc);
525 }
526 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
527
528 /************************************************************************/
529 /*                       Here the fs part begins                        */
530 /************************************************************************/
531
532 /* permission checks */
533 static int proc_fd_access_allowed(struct inode *inode)
534 {
535         struct task_struct *task;
536         int allowed = 0;
537         /* Allow access to a task's file descriptors if it is us or we
538          * may use ptrace attach to the process and find out that
539          * information.
540          */
541         task = get_proc_task(inode);
542         if (task) {
543                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
544                 put_task_struct(task);
545         }
546         return allowed;
547 }
548
549 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
550 {
551         int error;
552         struct inode *inode = dentry->d_inode;
553
554         if (attr->ia_valid & ATTR_MODE)
555                 return -EPERM;
556
557         error = inode_change_ok(inode, attr);
558         if (error)
559                 return error;
560
561         if ((attr->ia_valid & ATTR_SIZE) &&
562             attr->ia_size != i_size_read(inode)) {
563                 error = vmtruncate(inode, attr->ia_size);
564                 if (error)
565                         return error;
566         }
567
568         setattr_copy(inode, attr);
569         mark_inode_dirty(inode);
570         return 0;
571 }
572
573 static const struct inode_operations proc_def_inode_operations = {
574         .setattr        = proc_setattr,
575 };
576
577 static int mounts_open_common(struct inode *inode, struct file *file,
578                               const struct seq_operations *op)
579 {
580         struct task_struct *task = get_proc_task(inode);
581         struct nsproxy *nsp;
582         struct mnt_namespace *ns = NULL;
583         struct path root;
584         struct proc_mounts *p;
585         int ret = -EINVAL;
586
587         if (task) {
588                 rcu_read_lock();
589                 nsp = task_nsproxy(task);
590                 if (nsp) {
591                         ns = nsp->mnt_ns;
592                         if (ns)
593                                 get_mnt_ns(ns);
594                 }
595                 rcu_read_unlock();
596                 if (ns && get_task_root(task, &root) == 0)
597                         ret = 0;
598                 put_task_struct(task);
599         }
600
601         if (!ns)
602                 goto err;
603         if (ret)
604                 goto err_put_ns;
605
606         ret = -ENOMEM;
607         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
608         if (!p)
609                 goto err_put_path;
610
611         file->private_data = &p->m;
612         ret = seq_open(file, op);
613         if (ret)
614                 goto err_free;
615
616         p->m.private = p;
617         p->ns = ns;
618         p->root = root;
619         p->event = ns->event;
620
621         return 0;
622
623  err_free:
624         kfree(p);
625  err_put_path:
626         path_put(&root);
627  err_put_ns:
628         put_mnt_ns(ns);
629  err:
630         return ret;
631 }
632
633 static int mounts_release(struct inode *inode, struct file *file)
634 {
635         struct proc_mounts *p = file->private_data;
636         path_put(&p->root);
637         put_mnt_ns(p->ns);
638         return seq_release(inode, file);
639 }
640
641 static unsigned mounts_poll(struct file *file, poll_table *wait)
642 {
643         struct proc_mounts *p = file->private_data;
644         unsigned res = POLLIN | POLLRDNORM;
645
646         poll_wait(file, &p->ns->poll, wait);
647         if (mnt_had_events(p))
648                 res |= POLLERR | POLLPRI;
649
650         return res;
651 }
652
653 static int mounts_open(struct inode *inode, struct file *file)
654 {
655         return mounts_open_common(inode, file, &mounts_op);
656 }
657
658 static const struct file_operations proc_mounts_operations = {
659         .open           = mounts_open,
660         .read           = seq_read,
661         .llseek         = seq_lseek,
662         .release        = mounts_release,
663         .poll           = mounts_poll,
664 };
665
666 static int mountinfo_open(struct inode *inode, struct file *file)
667 {
668         return mounts_open_common(inode, file, &mountinfo_op);
669 }
670
671 static const struct file_operations proc_mountinfo_operations = {
672         .open           = mountinfo_open,
673         .read           = seq_read,
674         .llseek         = seq_lseek,
675         .release        = mounts_release,
676         .poll           = mounts_poll,
677 };
678
679 static int mountstats_open(struct inode *inode, struct file *file)
680 {
681         return mounts_open_common(inode, file, &mountstats_op);
682 }
683
684 static const struct file_operations proc_mountstats_operations = {
685         .open           = mountstats_open,
686         .read           = seq_read,
687         .llseek         = seq_lseek,
688         .release        = mounts_release,
689 };
690
691 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
692
693 static ssize_t proc_info_read(struct file * file, char __user * buf,
694                           size_t count, loff_t *ppos)
695 {
696         struct inode * inode = file->f_path.dentry->d_inode;
697         unsigned long page;
698         ssize_t length;
699         struct task_struct *task = get_proc_task(inode);
700
701         length = -ESRCH;
702         if (!task)
703                 goto out_no_task;
704
705         if (count > PROC_BLOCK_SIZE)
706                 count = PROC_BLOCK_SIZE;
707
708         length = -ENOMEM;
709         if (!(page = __get_free_page(GFP_TEMPORARY)))
710                 goto out;
711
712         length = PROC_I(inode)->op.proc_read(task, (char*)page);
713
714         if (length >= 0)
715                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
716         free_page(page);
717 out:
718         put_task_struct(task);
719 out_no_task:
720         return length;
721 }
722
723 static const struct file_operations proc_info_file_operations = {
724         .read           = proc_info_read,
725         .llseek         = generic_file_llseek,
726 };
727
728 static int proc_single_show(struct seq_file *m, void *v)
729 {
730         struct inode *inode = m->private;
731         struct pid_namespace *ns;
732         struct pid *pid;
733         struct task_struct *task;
734         int ret;
735
736         ns = inode->i_sb->s_fs_info;
737         pid = proc_pid(inode);
738         task = get_pid_task(pid, PIDTYPE_PID);
739         if (!task)
740                 return -ESRCH;
741
742         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
743
744         put_task_struct(task);
745         return ret;
746 }
747
748 static int proc_single_open(struct inode *inode, struct file *filp)
749 {
750         return single_open(filp, proc_single_show, inode);
751 }
752
753 static const struct file_operations proc_single_file_operations = {
754         .open           = proc_single_open,
755         .read           = seq_read,
756         .llseek         = seq_lseek,
757         .release        = single_release,
758 };
759
760 static int mem_open(struct inode* inode, struct file* file)
761 {
762         file->private_data = (void*)((long)current->self_exec_id);
763         /* OK to pass negative loff_t, we can catch out-of-range */
764         file->f_mode |= FMODE_UNSIGNED_OFFSET;
765         return 0;
766 }
767
768 static ssize_t mem_read(struct file * file, char __user * buf,
769                         size_t count, loff_t *ppos)
770 {
771         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
772         char *page;
773         unsigned long src = *ppos;
774         int ret = -ESRCH;
775         struct mm_struct *mm;
776
777         if (!task)
778                 goto out_no_task;
779
780         if (check_mem_permission(task))
781                 goto out;
782
783         ret = -ENOMEM;
784         page = (char *)__get_free_page(GFP_TEMPORARY);
785         if (!page)
786                 goto out;
787
788         ret = 0;
789  
790         mm = get_task_mm(task);
791         if (!mm)
792                 goto out_free;
793
794         ret = -EIO;
795  
796         if (file->private_data != (void*)((long)current->self_exec_id))
797                 goto out_put;
798
799         ret = 0;
800  
801         while (count > 0) {
802                 int this_len, retval;
803
804                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
805                 retval = access_process_vm(task, src, page, this_len, 0);
806                 if (!retval || check_mem_permission(task)) {
807                         if (!ret)
808                                 ret = -EIO;
809                         break;
810                 }
811
812                 if (copy_to_user(buf, page, retval)) {
813                         ret = -EFAULT;
814                         break;
815                 }
816  
817                 ret += retval;
818                 src += retval;
819                 buf += retval;
820                 count -= retval;
821         }
822         *ppos = src;
823
824 out_put:
825         mmput(mm);
826 out_free:
827         free_page((unsigned long) page);
828 out:
829         put_task_struct(task);
830 out_no_task:
831         return ret;
832 }
833
834 #define mem_write NULL
835
836 #ifndef mem_write
837 /* This is a security hazard */
838 static ssize_t mem_write(struct file * file, const char __user *buf,
839                          size_t count, loff_t *ppos)
840 {
841         int copied;
842         char *page;
843         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
844         unsigned long dst = *ppos;
845
846         copied = -ESRCH;
847         if (!task)
848                 goto out_no_task;
849
850         if (check_mem_permission(task))
851                 goto out;
852
853         copied = -ENOMEM;
854         page = (char *)__get_free_page(GFP_TEMPORARY);
855         if (!page)
856                 goto out;
857
858         copied = 0;
859         while (count > 0) {
860                 int this_len, retval;
861
862                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
863                 if (copy_from_user(page, buf, this_len)) {
864                         copied = -EFAULT;
865                         break;
866                 }
867                 retval = access_process_vm(task, dst, page, this_len, 1);
868                 if (!retval) {
869                         if (!copied)
870                                 copied = -EIO;
871                         break;
872                 }
873                 copied += retval;
874                 buf += retval;
875                 dst += retval;
876                 count -= retval;                        
877         }
878         *ppos = dst;
879         free_page((unsigned long) page);
880 out:
881         put_task_struct(task);
882 out_no_task:
883         return copied;
884 }
885 #endif
886
887 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
888 {
889         switch (orig) {
890         case 0:
891                 file->f_pos = offset;
892                 break;
893         case 1:
894                 file->f_pos += offset;
895                 break;
896         default:
897                 return -EINVAL;
898         }
899         force_successful_syscall_return();
900         return file->f_pos;
901 }
902
903 static const struct file_operations proc_mem_operations = {
904         .llseek         = mem_lseek,
905         .read           = mem_read,
906         .write          = mem_write,
907         .open           = mem_open,
908 };
909
910 static ssize_t environ_read(struct file *file, char __user *buf,
911                         size_t count, loff_t *ppos)
912 {
913         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
914         char *page;
915         unsigned long src = *ppos;
916         int ret = -ESRCH;
917         struct mm_struct *mm;
918
919         if (!task)
920                 goto out_no_task;
921
922         ret = -ENOMEM;
923         page = (char *)__get_free_page(GFP_TEMPORARY);
924         if (!page)
925                 goto out;
926
927
928         mm = mm_for_maps(task);
929         ret = PTR_ERR(mm);
930         if (!mm || IS_ERR(mm))
931                 goto out_free;
932
933         ret = 0;
934         while (count > 0) {
935                 int this_len, retval, max_len;
936
937                 this_len = mm->env_end - (mm->env_start + src);
938
939                 if (this_len <= 0)
940                         break;
941
942                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
943                 this_len = (this_len > max_len) ? max_len : this_len;
944
945                 retval = access_process_vm(task, (mm->env_start + src),
946                         page, this_len, 0);
947
948                 if (retval <= 0) {
949                         ret = retval;
950                         break;
951                 }
952
953                 if (copy_to_user(buf, page, retval)) {
954                         ret = -EFAULT;
955                         break;
956                 }
957
958                 ret += retval;
959                 src += retval;
960                 buf += retval;
961                 count -= retval;
962         }
963         *ppos = src;
964
965         mmput(mm);
966 out_free:
967         free_page((unsigned long) page);
968 out:
969         put_task_struct(task);
970 out_no_task:
971         return ret;
972 }
973
974 static const struct file_operations proc_environ_operations = {
975         .read           = environ_read,
976         .llseek         = generic_file_llseek,
977 };
978
979 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
980                                 size_t count, loff_t *ppos)
981 {
982         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
983         char buffer[PROC_NUMBUF];
984         size_t len;
985         int oom_adjust = OOM_DISABLE;
986         unsigned long flags;
987
988         if (!task)
989                 return -ESRCH;
990
991         if (lock_task_sighand(task, &flags)) {
992                 oom_adjust = task->signal->oom_adj;
993                 unlock_task_sighand(task, &flags);
994         }
995
996         put_task_struct(task);
997
998         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
999
1000         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1001 }
1002
1003 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1004                                 size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task;
1007         char buffer[PROC_NUMBUF];
1008         long oom_adjust;
1009         unsigned long flags;
1010         int err;
1011
1012         memset(buffer, 0, sizeof(buffer));
1013         if (count > sizeof(buffer) - 1)
1014                 count = sizeof(buffer) - 1;
1015         if (copy_from_user(buffer, buf, count)) {
1016                 err = -EFAULT;
1017                 goto out;
1018         }
1019
1020         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1021         if (err)
1022                 goto out;
1023         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1024              oom_adjust != OOM_DISABLE) {
1025                 err = -EINVAL;
1026                 goto out;
1027         }
1028
1029         task = get_proc_task(file->f_path.dentry->d_inode);
1030         if (!task) {
1031                 err = -ESRCH;
1032                 goto out;
1033         }
1034
1035         task_lock(task);
1036         if (!task->mm) {
1037                 err = -EINVAL;
1038                 goto err_task_lock;
1039         }
1040
1041         if (!lock_task_sighand(task, &flags)) {
1042                 err = -ESRCH;
1043                 goto err_task_lock;
1044         }
1045
1046         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1047                 err = -EACCES;
1048                 goto err_sighand;
1049         }
1050
1051         if (oom_adjust != task->signal->oom_adj) {
1052                 if (oom_adjust == OOM_DISABLE)
1053                         atomic_inc(&task->mm->oom_disable_count);
1054                 if (task->signal->oom_adj == OOM_DISABLE)
1055                         atomic_dec(&task->mm->oom_disable_count);
1056         }
1057
1058         /*
1059          * Warn that /proc/pid/oom_adj is deprecated, see
1060          * Documentation/feature-removal-schedule.txt.
1061          */
1062         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1063                         "please use /proc/%d/oom_score_adj instead.\n",
1064                         current->comm, task_pid_nr(current),
1065                         task_pid_nr(task), task_pid_nr(task));
1066         task->signal->oom_adj = oom_adjust;
1067         /*
1068          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1069          * value is always attainable.
1070          */
1071         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1072                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1073         else
1074                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1075                                                                 -OOM_DISABLE;
1076 err_sighand:
1077         unlock_task_sighand(task, &flags);
1078 err_task_lock:
1079         task_unlock(task);
1080         put_task_struct(task);
1081 out:
1082         return err < 0 ? err : count;
1083 }
1084
1085 static const struct file_operations proc_oom_adjust_operations = {
1086         .read           = oom_adjust_read,
1087         .write          = oom_adjust_write,
1088         .llseek         = generic_file_llseek,
1089 };
1090
1091 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1092                                         size_t count, loff_t *ppos)
1093 {
1094         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1095         char buffer[PROC_NUMBUF];
1096         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1097         unsigned long flags;
1098         size_t len;
1099
1100         if (!task)
1101                 return -ESRCH;
1102         if (lock_task_sighand(task, &flags)) {
1103                 oom_score_adj = task->signal->oom_score_adj;
1104                 unlock_task_sighand(task, &flags);
1105         }
1106         put_task_struct(task);
1107         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1108         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1109 }
1110
1111 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1112                                         size_t count, loff_t *ppos)
1113 {
1114         struct task_struct *task;
1115         char buffer[PROC_NUMBUF];
1116         unsigned long flags;
1117         long oom_score_adj;
1118         int err;
1119
1120         memset(buffer, 0, sizeof(buffer));
1121         if (count > sizeof(buffer) - 1)
1122                 count = sizeof(buffer) - 1;
1123         if (copy_from_user(buffer, buf, count)) {
1124                 err = -EFAULT;
1125                 goto out;
1126         }
1127
1128         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1129         if (err)
1130                 goto out;
1131         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1132                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1133                 err = -EINVAL;
1134                 goto out;
1135         }
1136
1137         task = get_proc_task(file->f_path.dentry->d_inode);
1138         if (!task) {
1139                 err = -ESRCH;
1140                 goto out;
1141         }
1142
1143         task_lock(task);
1144         if (!task->mm) {
1145                 err = -EINVAL;
1146                 goto err_task_lock;
1147         }
1148
1149         if (!lock_task_sighand(task, &flags)) {
1150                 err = -ESRCH;
1151                 goto err_task_lock;
1152         }
1153
1154         if (oom_score_adj < task->signal->oom_score_adj_min &&
1155                         !capable(CAP_SYS_RESOURCE)) {
1156                 err = -EACCES;
1157                 goto err_sighand;
1158         }
1159
1160         if (oom_score_adj != task->signal->oom_score_adj) {
1161                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1162                         atomic_inc(&task->mm->oom_disable_count);
1163                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1164                         atomic_dec(&task->mm->oom_disable_count);
1165         }
1166         task->signal->oom_score_adj = oom_score_adj;
1167         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1168                 task->signal->oom_score_adj_min = oom_score_adj;
1169         /*
1170          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1171          * always attainable.
1172          */
1173         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1174                 task->signal->oom_adj = OOM_DISABLE;
1175         else
1176                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1177                                                         OOM_SCORE_ADJ_MAX;
1178 err_sighand:
1179         unlock_task_sighand(task, &flags);
1180 err_task_lock:
1181         task_unlock(task);
1182         put_task_struct(task);
1183 out:
1184         return err < 0 ? err : count;
1185 }
1186
1187 static const struct file_operations proc_oom_score_adj_operations = {
1188         .read           = oom_score_adj_read,
1189         .write          = oom_score_adj_write,
1190         .llseek         = default_llseek,
1191 };
1192
1193 #ifdef CONFIG_AUDITSYSCALL
1194 #define TMPBUFLEN 21
1195 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1196                                   size_t count, loff_t *ppos)
1197 {
1198         struct inode * inode = file->f_path.dentry->d_inode;
1199         struct task_struct *task = get_proc_task(inode);
1200         ssize_t length;
1201         char tmpbuf[TMPBUFLEN];
1202
1203         if (!task)
1204                 return -ESRCH;
1205         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1206                                 audit_get_loginuid(task));
1207         put_task_struct(task);
1208         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1209 }
1210
1211 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1212                                    size_t count, loff_t *ppos)
1213 {
1214         struct inode * inode = file->f_path.dentry->d_inode;
1215         char *page, *tmp;
1216         ssize_t length;
1217         uid_t loginuid;
1218
1219         if (!capable(CAP_AUDIT_CONTROL))
1220                 return -EPERM;
1221
1222         rcu_read_lock();
1223         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1224                 rcu_read_unlock();
1225                 return -EPERM;
1226         }
1227         rcu_read_unlock();
1228
1229         if (count >= PAGE_SIZE)
1230                 count = PAGE_SIZE - 1;
1231
1232         if (*ppos != 0) {
1233                 /* No partial writes. */
1234                 return -EINVAL;
1235         }
1236         page = (char*)__get_free_page(GFP_TEMPORARY);
1237         if (!page)
1238                 return -ENOMEM;
1239         length = -EFAULT;
1240         if (copy_from_user(page, buf, count))
1241                 goto out_free_page;
1242
1243         page[count] = '\0';
1244         loginuid = simple_strtoul(page, &tmp, 10);
1245         if (tmp == page) {
1246                 length = -EINVAL;
1247                 goto out_free_page;
1248
1249         }
1250         length = audit_set_loginuid(current, loginuid);
1251         if (likely(length == 0))
1252                 length = count;
1253
1254 out_free_page:
1255         free_page((unsigned long) page);
1256         return length;
1257 }
1258
1259 static const struct file_operations proc_loginuid_operations = {
1260         .read           = proc_loginuid_read,
1261         .write          = proc_loginuid_write,
1262         .llseek         = generic_file_llseek,
1263 };
1264
1265 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1266                                   size_t count, loff_t *ppos)
1267 {
1268         struct inode * inode = file->f_path.dentry->d_inode;
1269         struct task_struct *task = get_proc_task(inode);
1270         ssize_t length;
1271         char tmpbuf[TMPBUFLEN];
1272
1273         if (!task)
1274                 return -ESRCH;
1275         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1276                                 audit_get_sessionid(task));
1277         put_task_struct(task);
1278         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1279 }
1280
1281 static const struct file_operations proc_sessionid_operations = {
1282         .read           = proc_sessionid_read,
1283         .llseek         = generic_file_llseek,
1284 };
1285 #endif
1286
1287 #ifdef CONFIG_FAULT_INJECTION
1288 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1289                                       size_t count, loff_t *ppos)
1290 {
1291         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1292         char buffer[PROC_NUMBUF];
1293         size_t len;
1294         int make_it_fail;
1295
1296         if (!task)
1297                 return -ESRCH;
1298         make_it_fail = task->make_it_fail;
1299         put_task_struct(task);
1300
1301         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1302
1303         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1304 }
1305
1306 static ssize_t proc_fault_inject_write(struct file * file,
1307                         const char __user * buf, size_t count, loff_t *ppos)
1308 {
1309         struct task_struct *task;
1310         char buffer[PROC_NUMBUF], *end;
1311         int make_it_fail;
1312
1313         if (!capable(CAP_SYS_RESOURCE))
1314                 return -EPERM;
1315         memset(buffer, 0, sizeof(buffer));
1316         if (count > sizeof(buffer) - 1)
1317                 count = sizeof(buffer) - 1;
1318         if (copy_from_user(buffer, buf, count))
1319                 return -EFAULT;
1320         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1321         if (*end)
1322                 return -EINVAL;
1323         task = get_proc_task(file->f_dentry->d_inode);
1324         if (!task)
1325                 return -ESRCH;
1326         task->make_it_fail = make_it_fail;
1327         put_task_struct(task);
1328
1329         return count;
1330 }
1331
1332 static const struct file_operations proc_fault_inject_operations = {
1333         .read           = proc_fault_inject_read,
1334         .write          = proc_fault_inject_write,
1335         .llseek         = generic_file_llseek,
1336 };
1337 #endif
1338
1339
1340 #ifdef CONFIG_SCHED_DEBUG
1341 /*
1342  * Print out various scheduling related per-task fields:
1343  */
1344 static int sched_show(struct seq_file *m, void *v)
1345 {
1346         struct inode *inode = m->private;
1347         struct task_struct *p;
1348
1349         p = get_proc_task(inode);
1350         if (!p)
1351                 return -ESRCH;
1352         proc_sched_show_task(p, m);
1353
1354         put_task_struct(p);
1355
1356         return 0;
1357 }
1358
1359 static ssize_t
1360 sched_write(struct file *file, const char __user *buf,
1361             size_t count, loff_t *offset)
1362 {
1363         struct inode *inode = file->f_path.dentry->d_inode;
1364         struct task_struct *p;
1365
1366         p = get_proc_task(inode);
1367         if (!p)
1368                 return -ESRCH;
1369         proc_sched_set_task(p);
1370
1371         put_task_struct(p);
1372
1373         return count;
1374 }
1375
1376 static int sched_open(struct inode *inode, struct file *filp)
1377 {
1378         return single_open(filp, sched_show, inode);
1379 }
1380
1381 static const struct file_operations proc_pid_sched_operations = {
1382         .open           = sched_open,
1383         .read           = seq_read,
1384         .write          = sched_write,
1385         .llseek         = seq_lseek,
1386         .release        = single_release,
1387 };
1388
1389 #endif
1390
1391 #ifdef CONFIG_SCHED_AUTOGROUP
1392 /*
1393  * Print out autogroup related information:
1394  */
1395 static int sched_autogroup_show(struct seq_file *m, void *v)
1396 {
1397         struct inode *inode = m->private;
1398         struct task_struct *p;
1399
1400         p = get_proc_task(inode);
1401         if (!p)
1402                 return -ESRCH;
1403         proc_sched_autogroup_show_task(p, m);
1404
1405         put_task_struct(p);
1406
1407         return 0;
1408 }
1409
1410 static ssize_t
1411 sched_autogroup_write(struct file *file, const char __user *buf,
1412             size_t count, loff_t *offset)
1413 {
1414         struct inode *inode = file->f_path.dentry->d_inode;
1415         struct task_struct *p;
1416         char buffer[PROC_NUMBUF];
1417         long nice;
1418         int err;
1419
1420         memset(buffer, 0, sizeof(buffer));
1421         if (count > sizeof(buffer) - 1)
1422                 count = sizeof(buffer) - 1;
1423         if (copy_from_user(buffer, buf, count))
1424                 return -EFAULT;
1425
1426         err = strict_strtol(strstrip(buffer), 0, &nice);
1427         if (err)
1428                 return -EINVAL;
1429
1430         p = get_proc_task(inode);
1431         if (!p)
1432                 return -ESRCH;
1433
1434         err = nice;
1435         err = proc_sched_autogroup_set_nice(p, &err);
1436         if (err)
1437                 count = err;
1438
1439         put_task_struct(p);
1440
1441         return count;
1442 }
1443
1444 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1445 {
1446         int ret;
1447
1448         ret = single_open(filp, sched_autogroup_show, NULL);
1449         if (!ret) {
1450                 struct seq_file *m = filp->private_data;
1451
1452                 m->private = inode;
1453         }
1454         return ret;
1455 }
1456
1457 static const struct file_operations proc_pid_sched_autogroup_operations = {
1458         .open           = sched_autogroup_open,
1459         .read           = seq_read,
1460         .write          = sched_autogroup_write,
1461         .llseek         = seq_lseek,
1462         .release        = single_release,
1463 };
1464
1465 #endif /* CONFIG_SCHED_AUTOGROUP */
1466
1467 static ssize_t comm_write(struct file *file, const char __user *buf,
1468                                 size_t count, loff_t *offset)
1469 {
1470         struct inode *inode = file->f_path.dentry->d_inode;
1471         struct task_struct *p;
1472         char buffer[TASK_COMM_LEN];
1473
1474         memset(buffer, 0, sizeof(buffer));
1475         if (count > sizeof(buffer) - 1)
1476                 count = sizeof(buffer) - 1;
1477         if (copy_from_user(buffer, buf, count))
1478                 return -EFAULT;
1479
1480         p = get_proc_task(inode);
1481         if (!p)
1482                 return -ESRCH;
1483
1484         if (same_thread_group(current, p))
1485                 set_task_comm(p, buffer);
1486         else
1487                 count = -EINVAL;
1488
1489         put_task_struct(p);
1490
1491         return count;
1492 }
1493
1494 static int comm_show(struct seq_file *m, void *v)
1495 {
1496         struct inode *inode = m->private;
1497         struct task_struct *p;
1498
1499         p = get_proc_task(inode);
1500         if (!p)
1501                 return -ESRCH;
1502
1503         task_lock(p);
1504         seq_printf(m, "%s\n", p->comm);
1505         task_unlock(p);
1506
1507         put_task_struct(p);
1508
1509         return 0;
1510 }
1511
1512 static int comm_open(struct inode *inode, struct file *filp)
1513 {
1514         return single_open(filp, comm_show, inode);
1515 }
1516
1517 static const struct file_operations proc_pid_set_comm_operations = {
1518         .open           = comm_open,
1519         .read           = seq_read,
1520         .write          = comm_write,
1521         .llseek         = seq_lseek,
1522         .release        = single_release,
1523 };
1524
1525 /*
1526  * We added or removed a vma mapping the executable. The vmas are only mapped
1527  * during exec and are not mapped with the mmap system call.
1528  * Callers must hold down_write() on the mm's mmap_sem for these
1529  */
1530 void added_exe_file_vma(struct mm_struct *mm)
1531 {
1532         mm->num_exe_file_vmas++;
1533 }
1534
1535 void removed_exe_file_vma(struct mm_struct *mm)
1536 {
1537         mm->num_exe_file_vmas--;
1538         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1539                 fput(mm->exe_file);
1540                 mm->exe_file = NULL;
1541         }
1542
1543 }
1544
1545 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1546 {
1547         if (new_exe_file)
1548                 get_file(new_exe_file);
1549         if (mm->exe_file)
1550                 fput(mm->exe_file);
1551         mm->exe_file = new_exe_file;
1552         mm->num_exe_file_vmas = 0;
1553 }
1554
1555 struct file *get_mm_exe_file(struct mm_struct *mm)
1556 {
1557         struct file *exe_file;
1558
1559         /* We need mmap_sem to protect against races with removal of
1560          * VM_EXECUTABLE vmas */
1561         down_read(&mm->mmap_sem);
1562         exe_file = mm->exe_file;
1563         if (exe_file)
1564                 get_file(exe_file);
1565         up_read(&mm->mmap_sem);
1566         return exe_file;
1567 }
1568
1569 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1570 {
1571         /* It's safe to write the exe_file pointer without exe_file_lock because
1572          * this is called during fork when the task is not yet in /proc */
1573         newmm->exe_file = get_mm_exe_file(oldmm);
1574 }
1575
1576 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1577 {
1578         struct task_struct *task;
1579         struct mm_struct *mm;
1580         struct file *exe_file;
1581
1582         task = get_proc_task(inode);
1583         if (!task)
1584                 return -ENOENT;
1585         mm = get_task_mm(task);
1586         put_task_struct(task);
1587         if (!mm)
1588                 return -ENOENT;
1589         exe_file = get_mm_exe_file(mm);
1590         mmput(mm);
1591         if (exe_file) {
1592                 *exe_path = exe_file->f_path;
1593                 path_get(&exe_file->f_path);
1594                 fput(exe_file);
1595                 return 0;
1596         } else
1597                 return -ENOENT;
1598 }
1599
1600 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1601 {
1602         struct inode *inode = dentry->d_inode;
1603         int error = -EACCES;
1604
1605         /* We don't need a base pointer in the /proc filesystem */
1606         path_put(&nd->path);
1607
1608         /* Are we allowed to snoop on the tasks file descriptors? */
1609         if (!proc_fd_access_allowed(inode))
1610                 goto out;
1611
1612         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1613 out:
1614         return ERR_PTR(error);
1615 }
1616
1617 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1618 {
1619         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1620         char *pathname;
1621         int len;
1622
1623         if (!tmp)
1624                 return -ENOMEM;
1625
1626         pathname = d_path(path, tmp, PAGE_SIZE);
1627         len = PTR_ERR(pathname);
1628         if (IS_ERR(pathname))
1629                 goto out;
1630         len = tmp + PAGE_SIZE - 1 - pathname;
1631
1632         if (len > buflen)
1633                 len = buflen;
1634         if (copy_to_user(buffer, pathname, len))
1635                 len = -EFAULT;
1636  out:
1637         free_page((unsigned long)tmp);
1638         return len;
1639 }
1640
1641 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1642 {
1643         int error = -EACCES;
1644         struct inode *inode = dentry->d_inode;
1645         struct path path;
1646
1647         /* Are we allowed to snoop on the tasks file descriptors? */
1648         if (!proc_fd_access_allowed(inode))
1649                 goto out;
1650
1651         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1652         if (error)
1653                 goto out;
1654
1655         error = do_proc_readlink(&path, buffer, buflen);
1656         path_put(&path);
1657 out:
1658         return error;
1659 }
1660
1661 static const struct inode_operations proc_pid_link_inode_operations = {
1662         .readlink       = proc_pid_readlink,
1663         .follow_link    = proc_pid_follow_link,
1664         .setattr        = proc_setattr,
1665 };
1666
1667
1668 /* building an inode */
1669
1670 static int task_dumpable(struct task_struct *task)
1671 {
1672         int dumpable = 0;
1673         struct mm_struct *mm;
1674
1675         task_lock(task);
1676         mm = task->mm;
1677         if (mm)
1678                 dumpable = get_dumpable(mm);
1679         task_unlock(task);
1680         if(dumpable == 1)
1681                 return 1;
1682         return 0;
1683 }
1684
1685
1686 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1687 {
1688         struct inode * inode;
1689         struct proc_inode *ei;
1690         const struct cred *cred;
1691
1692         /* We need a new inode */
1693
1694         inode = new_inode(sb);
1695         if (!inode)
1696                 goto out;
1697
1698         /* Common stuff */
1699         ei = PROC_I(inode);
1700         inode->i_ino = get_next_ino();
1701         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1702         inode->i_op = &proc_def_inode_operations;
1703
1704         /*
1705          * grab the reference to task.
1706          */
1707         ei->pid = get_task_pid(task, PIDTYPE_PID);
1708         if (!ei->pid)
1709                 goto out_unlock;
1710
1711         if (task_dumpable(task)) {
1712                 rcu_read_lock();
1713                 cred = __task_cred(task);
1714                 inode->i_uid = cred->euid;
1715                 inode->i_gid = cred->egid;
1716                 rcu_read_unlock();
1717         }
1718         security_task_to_inode(task, inode);
1719
1720 out:
1721         return inode;
1722
1723 out_unlock:
1724         iput(inode);
1725         return NULL;
1726 }
1727
1728 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1729 {
1730         struct inode *inode = dentry->d_inode;
1731         struct task_struct *task;
1732         const struct cred *cred;
1733
1734         generic_fillattr(inode, stat);
1735
1736         rcu_read_lock();
1737         stat->uid = 0;
1738         stat->gid = 0;
1739         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1740         if (task) {
1741                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1742                     task_dumpable(task)) {
1743                         cred = __task_cred(task);
1744                         stat->uid = cred->euid;
1745                         stat->gid = cred->egid;
1746                 }
1747         }
1748         rcu_read_unlock();
1749         return 0;
1750 }
1751
1752 /* dentry stuff */
1753
1754 /*
1755  *      Exceptional case: normally we are not allowed to unhash a busy
1756  * directory. In this case, however, we can do it - no aliasing problems
1757  * due to the way we treat inodes.
1758  *
1759  * Rewrite the inode's ownerships here because the owning task may have
1760  * performed a setuid(), etc.
1761  *
1762  * Before the /proc/pid/status file was created the only way to read
1763  * the effective uid of a /process was to stat /proc/pid.  Reading
1764  * /proc/pid/status is slow enough that procps and other packages
1765  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1766  * made this apply to all per process world readable and executable
1767  * directories.
1768  */
1769 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1770 {
1771         struct inode *inode;
1772         struct task_struct *task;
1773         const struct cred *cred;
1774
1775         if (nd && nd->flags & LOOKUP_RCU)
1776                 return -ECHILD;
1777
1778         inode = dentry->d_inode;
1779         task = get_proc_task(inode);
1780
1781         if (task) {
1782                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1783                     task_dumpable(task)) {
1784                         rcu_read_lock();
1785                         cred = __task_cred(task);
1786                         inode->i_uid = cred->euid;
1787                         inode->i_gid = cred->egid;
1788                         rcu_read_unlock();
1789                 } else {
1790                         inode->i_uid = 0;
1791                         inode->i_gid = 0;
1792                 }
1793                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1794                 security_task_to_inode(task, inode);
1795                 put_task_struct(task);
1796                 return 1;
1797         }
1798         d_drop(dentry);
1799         return 0;
1800 }
1801
1802 static int pid_delete_dentry(const struct dentry * dentry)
1803 {
1804         /* Is the task we represent dead?
1805          * If so, then don't put the dentry on the lru list,
1806          * kill it immediately.
1807          */
1808         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1809 }
1810
1811 static const struct dentry_operations pid_dentry_operations =
1812 {
1813         .d_revalidate   = pid_revalidate,
1814         .d_delete       = pid_delete_dentry,
1815 };
1816
1817 /* Lookups */
1818
1819 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1820                                 struct task_struct *, const void *);
1821
1822 /*
1823  * Fill a directory entry.
1824  *
1825  * If possible create the dcache entry and derive our inode number and
1826  * file type from dcache entry.
1827  *
1828  * Since all of the proc inode numbers are dynamically generated, the inode
1829  * numbers do not exist until the inode is cache.  This means creating the
1830  * the dcache entry in readdir is necessary to keep the inode numbers
1831  * reported by readdir in sync with the inode numbers reported
1832  * by stat.
1833  */
1834 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1835         char *name, int len,
1836         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1837 {
1838         struct dentry *child, *dir = filp->f_path.dentry;
1839         struct inode *inode;
1840         struct qstr qname;
1841         ino_t ino = 0;
1842         unsigned type = DT_UNKNOWN;
1843
1844         qname.name = name;
1845         qname.len  = len;
1846         qname.hash = full_name_hash(name, len);
1847
1848         child = d_lookup(dir, &qname);
1849         if (!child) {
1850                 struct dentry *new;
1851                 new = d_alloc(dir, &qname);
1852                 if (new) {
1853                         child = instantiate(dir->d_inode, new, task, ptr);
1854                         if (child)
1855                                 dput(new);
1856                         else
1857                                 child = new;
1858                 }
1859         }
1860         if (!child || IS_ERR(child) || !child->d_inode)
1861                 goto end_instantiate;
1862         inode = child->d_inode;
1863         if (inode) {
1864                 ino = inode->i_ino;
1865                 type = inode->i_mode >> 12;
1866         }
1867         dput(child);
1868 end_instantiate:
1869         if (!ino)
1870                 ino = find_inode_number(dir, &qname);
1871         if (!ino)
1872                 ino = 1;
1873         return filldir(dirent, name, len, filp->f_pos, ino, type);
1874 }
1875
1876 static unsigned name_to_int(struct dentry *dentry)
1877 {
1878         const char *name = dentry->d_name.name;
1879         int len = dentry->d_name.len;
1880         unsigned n = 0;
1881
1882         if (len > 1 && *name == '0')
1883                 goto out;
1884         while (len-- > 0) {
1885                 unsigned c = *name++ - '0';
1886                 if (c > 9)
1887                         goto out;
1888                 if (n >= (~0U-9)/10)
1889                         goto out;
1890                 n *= 10;
1891                 n += c;
1892         }
1893         return n;
1894 out:
1895         return ~0U;
1896 }
1897
1898 #define PROC_FDINFO_MAX 64
1899
1900 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1901 {
1902         struct task_struct *task = get_proc_task(inode);
1903         struct files_struct *files = NULL;
1904         struct file *file;
1905         int fd = proc_fd(inode);
1906
1907         if (task) {
1908                 files = get_files_struct(task);
1909                 put_task_struct(task);
1910         }
1911         if (files) {
1912                 /*
1913                  * We are not taking a ref to the file structure, so we must
1914                  * hold ->file_lock.
1915                  */
1916                 spin_lock(&files->file_lock);
1917                 file = fcheck_files(files, fd);
1918                 if (file) {
1919                         if (path) {
1920                                 *path = file->f_path;
1921                                 path_get(&file->f_path);
1922                         }
1923                         if (info)
1924                                 snprintf(info, PROC_FDINFO_MAX,
1925                                          "pos:\t%lli\n"
1926                                          "flags:\t0%o\n",
1927                                          (long long) file->f_pos,
1928                                          file->f_flags);
1929                         spin_unlock(&files->file_lock);
1930                         put_files_struct(files);
1931                         return 0;
1932                 }
1933                 spin_unlock(&files->file_lock);
1934                 put_files_struct(files);
1935         }
1936         return -ENOENT;
1937 }
1938
1939 static int proc_fd_link(struct inode *inode, struct path *path)
1940 {
1941         return proc_fd_info(inode, path, NULL);
1942 }
1943
1944 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1945 {
1946         struct inode *inode;
1947         struct task_struct *task;
1948         int fd;
1949         struct files_struct *files;
1950         const struct cred *cred;
1951
1952         if (nd && nd->flags & LOOKUP_RCU)
1953                 return -ECHILD;
1954
1955         inode = dentry->d_inode;
1956         task = get_proc_task(inode);
1957         fd = proc_fd(inode);
1958
1959         if (task) {
1960                 files = get_files_struct(task);
1961                 if (files) {
1962                         rcu_read_lock();
1963                         if (fcheck_files(files, fd)) {
1964                                 rcu_read_unlock();
1965                                 put_files_struct(files);
1966                                 if (task_dumpable(task)) {
1967                                         rcu_read_lock();
1968                                         cred = __task_cred(task);
1969                                         inode->i_uid = cred->euid;
1970                                         inode->i_gid = cred->egid;
1971                                         rcu_read_unlock();
1972                                 } else {
1973                                         inode->i_uid = 0;
1974                                         inode->i_gid = 0;
1975                                 }
1976                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1977                                 security_task_to_inode(task, inode);
1978                                 put_task_struct(task);
1979                                 return 1;
1980                         }
1981                         rcu_read_unlock();
1982                         put_files_struct(files);
1983                 }
1984                 put_task_struct(task);
1985         }
1986         d_drop(dentry);
1987         return 0;
1988 }
1989
1990 static const struct dentry_operations tid_fd_dentry_operations =
1991 {
1992         .d_revalidate   = tid_fd_revalidate,
1993         .d_delete       = pid_delete_dentry,
1994 };
1995
1996 static struct dentry *proc_fd_instantiate(struct inode *dir,
1997         struct dentry *dentry, struct task_struct *task, const void *ptr)
1998 {
1999         unsigned fd = *(const unsigned *)ptr;
2000         struct file *file;
2001         struct files_struct *files;
2002         struct inode *inode;
2003         struct proc_inode *ei;
2004         struct dentry *error = ERR_PTR(-ENOENT);
2005
2006         inode = proc_pid_make_inode(dir->i_sb, task);
2007         if (!inode)
2008                 goto out;
2009         ei = PROC_I(inode);
2010         ei->fd = fd;
2011         files = get_files_struct(task);
2012         if (!files)
2013                 goto out_iput;
2014         inode->i_mode = S_IFLNK;
2015
2016         /*
2017          * We are not taking a ref to the file structure, so we must
2018          * hold ->file_lock.
2019          */
2020         spin_lock(&files->file_lock);
2021         file = fcheck_files(files, fd);
2022         if (!file)
2023                 goto out_unlock;
2024         if (file->f_mode & FMODE_READ)
2025                 inode->i_mode |= S_IRUSR | S_IXUSR;
2026         if (file->f_mode & FMODE_WRITE)
2027                 inode->i_mode |= S_IWUSR | S_IXUSR;
2028         spin_unlock(&files->file_lock);
2029         put_files_struct(files);
2030
2031         inode->i_op = &proc_pid_link_inode_operations;
2032         inode->i_size = 64;
2033         ei->op.proc_get_link = proc_fd_link;
2034         d_set_d_op(dentry, &tid_fd_dentry_operations);
2035         d_add(dentry, inode);
2036         /* Close the race of the process dying before we return the dentry */
2037         if (tid_fd_revalidate(dentry, NULL))
2038                 error = NULL;
2039
2040  out:
2041         return error;
2042 out_unlock:
2043         spin_unlock(&files->file_lock);
2044         put_files_struct(files);
2045 out_iput:
2046         iput(inode);
2047         goto out;
2048 }
2049
2050 static struct dentry *proc_lookupfd_common(struct inode *dir,
2051                                            struct dentry *dentry,
2052                                            instantiate_t instantiate)
2053 {
2054         struct task_struct *task = get_proc_task(dir);
2055         unsigned fd = name_to_int(dentry);
2056         struct dentry *result = ERR_PTR(-ENOENT);
2057
2058         if (!task)
2059                 goto out_no_task;
2060         if (fd == ~0U)
2061                 goto out;
2062
2063         result = instantiate(dir, dentry, task, &fd);
2064 out:
2065         put_task_struct(task);
2066 out_no_task:
2067         return result;
2068 }
2069
2070 static int proc_readfd_common(struct file * filp, void * dirent,
2071                               filldir_t filldir, instantiate_t instantiate)
2072 {
2073         struct dentry *dentry = filp->f_path.dentry;
2074         struct inode *inode = dentry->d_inode;
2075         struct task_struct *p = get_proc_task(inode);
2076         unsigned int fd, ino;
2077         int retval;
2078         struct files_struct * files;
2079
2080         retval = -ENOENT;
2081         if (!p)
2082                 goto out_no_task;
2083         retval = 0;
2084
2085         fd = filp->f_pos;
2086         switch (fd) {
2087                 case 0:
2088                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2089                                 goto out;
2090                         filp->f_pos++;
2091                 case 1:
2092                         ino = parent_ino(dentry);
2093                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2094                                 goto out;
2095                         filp->f_pos++;
2096                 default:
2097                         files = get_files_struct(p);
2098                         if (!files)
2099                                 goto out;
2100                         rcu_read_lock();
2101                         for (fd = filp->f_pos-2;
2102                              fd < files_fdtable(files)->max_fds;
2103                              fd++, filp->f_pos++) {
2104                                 char name[PROC_NUMBUF];
2105                                 int len;
2106
2107                                 if (!fcheck_files(files, fd))
2108                                         continue;
2109                                 rcu_read_unlock();
2110
2111                                 len = snprintf(name, sizeof(name), "%d", fd);
2112                                 if (proc_fill_cache(filp, dirent, filldir,
2113                                                     name, len, instantiate,
2114                                                     p, &fd) < 0) {
2115                                         rcu_read_lock();
2116                                         break;
2117                                 }
2118                                 rcu_read_lock();
2119                         }
2120                         rcu_read_unlock();
2121                         put_files_struct(files);
2122         }
2123 out:
2124         put_task_struct(p);
2125 out_no_task:
2126         return retval;
2127 }
2128
2129 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2130                                     struct nameidata *nd)
2131 {
2132         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2133 }
2134
2135 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2136 {
2137         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2138 }
2139
2140 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2141                                       size_t len, loff_t *ppos)
2142 {
2143         char tmp[PROC_FDINFO_MAX];
2144         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2145         if (!err)
2146                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2147         return err;
2148 }
2149
2150 static const struct file_operations proc_fdinfo_file_operations = {
2151         .open           = nonseekable_open,
2152         .read           = proc_fdinfo_read,
2153         .llseek         = no_llseek,
2154 };
2155
2156 static const struct file_operations proc_fd_operations = {
2157         .read           = generic_read_dir,
2158         .readdir        = proc_readfd,
2159         .llseek         = default_llseek,
2160 };
2161
2162 /*
2163  * /proc/pid/fd needs a special permission handler so that a process can still
2164  * access /proc/self/fd after it has executed a setuid().
2165  */
2166 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2167 {
2168         int rv;
2169
2170         if (flags & IPERM_FLAG_RCU)
2171                 return -ECHILD;
2172         rv = generic_permission(inode, mask, flags, NULL);
2173         if (rv == 0)
2174                 return 0;
2175         if (task_pid(current) == proc_pid(inode))
2176                 rv = 0;
2177         return rv;
2178 }
2179
2180 /*
2181  * proc directories can do almost nothing..
2182  */
2183 static const struct inode_operations proc_fd_inode_operations = {
2184         .lookup         = proc_lookupfd,
2185         .permission     = proc_fd_permission,
2186         .setattr        = proc_setattr,
2187 };
2188
2189 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2190         struct dentry *dentry, struct task_struct *task, const void *ptr)
2191 {
2192         unsigned fd = *(unsigned *)ptr;
2193         struct inode *inode;
2194         struct proc_inode *ei;
2195         struct dentry *error = ERR_PTR(-ENOENT);
2196
2197         inode = proc_pid_make_inode(dir->i_sb, task);
2198         if (!inode)
2199                 goto out;
2200         ei = PROC_I(inode);
2201         ei->fd = fd;
2202         inode->i_mode = S_IFREG | S_IRUSR;
2203         inode->i_fop = &proc_fdinfo_file_operations;
2204         d_set_d_op(dentry, &tid_fd_dentry_operations);
2205         d_add(dentry, inode);
2206         /* Close the race of the process dying before we return the dentry */
2207         if (tid_fd_revalidate(dentry, NULL))
2208                 error = NULL;
2209
2210  out:
2211         return error;
2212 }
2213
2214 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2215                                         struct dentry *dentry,
2216                                         struct nameidata *nd)
2217 {
2218         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2219 }
2220
2221 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2222 {
2223         return proc_readfd_common(filp, dirent, filldir,
2224                                   proc_fdinfo_instantiate);
2225 }
2226
2227 static const struct file_operations proc_fdinfo_operations = {
2228         .read           = generic_read_dir,
2229         .readdir        = proc_readfdinfo,
2230         .llseek         = default_llseek,
2231 };
2232
2233 /*
2234  * proc directories can do almost nothing..
2235  */
2236 static const struct inode_operations proc_fdinfo_inode_operations = {
2237         .lookup         = proc_lookupfdinfo,
2238         .setattr        = proc_setattr,
2239 };
2240
2241
2242 static struct dentry *proc_pident_instantiate(struct inode *dir,
2243         struct dentry *dentry, struct task_struct *task, const void *ptr)
2244 {
2245         const struct pid_entry *p = ptr;
2246         struct inode *inode;
2247         struct proc_inode *ei;
2248         struct dentry *error = ERR_PTR(-ENOENT);
2249
2250         inode = proc_pid_make_inode(dir->i_sb, task);
2251         if (!inode)
2252                 goto out;
2253
2254         ei = PROC_I(inode);
2255         inode->i_mode = p->mode;
2256         if (S_ISDIR(inode->i_mode))
2257                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2258         if (p->iop)
2259                 inode->i_op = p->iop;
2260         if (p->fop)
2261                 inode->i_fop = p->fop;
2262         ei->op = p->op;
2263         d_set_d_op(dentry, &pid_dentry_operations);
2264         d_add(dentry, inode);
2265         /* Close the race of the process dying before we return the dentry */
2266         if (pid_revalidate(dentry, NULL))
2267                 error = NULL;
2268 out:
2269         return error;
2270 }
2271
2272 static struct dentry *proc_pident_lookup(struct inode *dir, 
2273                                          struct dentry *dentry,
2274                                          const struct pid_entry *ents,
2275                                          unsigned int nents)
2276 {
2277         struct dentry *error;
2278         struct task_struct *task = get_proc_task(dir);
2279         const struct pid_entry *p, *last;
2280
2281         error = ERR_PTR(-ENOENT);
2282
2283         if (!task)
2284                 goto out_no_task;
2285
2286         /*
2287          * Yes, it does not scale. And it should not. Don't add
2288          * new entries into /proc/<tgid>/ without very good reasons.
2289          */
2290         last = &ents[nents - 1];
2291         for (p = ents; p <= last; p++) {
2292                 if (p->len != dentry->d_name.len)
2293                         continue;
2294                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2295                         break;
2296         }
2297         if (p > last)
2298                 goto out;
2299
2300         error = proc_pident_instantiate(dir, dentry, task, p);
2301 out:
2302         put_task_struct(task);
2303 out_no_task:
2304         return error;
2305 }
2306
2307 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2308         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2309 {
2310         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2311                                 proc_pident_instantiate, task, p);
2312 }
2313
2314 static int proc_pident_readdir(struct file *filp,
2315                 void *dirent, filldir_t filldir,
2316                 const struct pid_entry *ents, unsigned int nents)
2317 {
2318         int i;
2319         struct dentry *dentry = filp->f_path.dentry;
2320         struct inode *inode = dentry->d_inode;
2321         struct task_struct *task = get_proc_task(inode);
2322         const struct pid_entry *p, *last;
2323         ino_t ino;
2324         int ret;
2325
2326         ret = -ENOENT;
2327         if (!task)
2328                 goto out_no_task;
2329
2330         ret = 0;
2331         i = filp->f_pos;
2332         switch (i) {
2333         case 0:
2334                 ino = inode->i_ino;
2335                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2336                         goto out;
2337                 i++;
2338                 filp->f_pos++;
2339                 /* fall through */
2340         case 1:
2341                 ino = parent_ino(dentry);
2342                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2343                         goto out;
2344                 i++;
2345                 filp->f_pos++;
2346                 /* fall through */
2347         default:
2348                 i -= 2;
2349                 if (i >= nents) {
2350                         ret = 1;
2351                         goto out;
2352                 }
2353                 p = ents + i;
2354                 last = &ents[nents - 1];
2355                 while (p <= last) {
2356                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2357                                 goto out;
2358                         filp->f_pos++;
2359                         p++;
2360                 }
2361         }
2362
2363         ret = 1;
2364 out:
2365         put_task_struct(task);
2366 out_no_task:
2367         return ret;
2368 }
2369
2370 #ifdef CONFIG_SECURITY
2371 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2372                                   size_t count, loff_t *ppos)
2373 {
2374         struct inode * inode = file->f_path.dentry->d_inode;
2375         char *p = NULL;
2376         ssize_t length;
2377         struct task_struct *task = get_proc_task(inode);
2378
2379         if (!task)
2380                 return -ESRCH;
2381
2382         length = security_getprocattr(task,
2383                                       (char*)file->f_path.dentry->d_name.name,
2384                                       &p);
2385         put_task_struct(task);
2386         if (length > 0)
2387                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2388         kfree(p);
2389         return length;
2390 }
2391
2392 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2393                                    size_t count, loff_t *ppos)
2394 {
2395         struct inode * inode = file->f_path.dentry->d_inode;
2396         char *page;
2397         ssize_t length;
2398         struct task_struct *task = get_proc_task(inode);
2399
2400         length = -ESRCH;
2401         if (!task)
2402                 goto out_no_task;
2403         if (count > PAGE_SIZE)
2404                 count = PAGE_SIZE;
2405
2406         /* No partial writes. */
2407         length = -EINVAL;
2408         if (*ppos != 0)
2409                 goto out;
2410
2411         length = -ENOMEM;
2412         page = (char*)__get_free_page(GFP_TEMPORARY);
2413         if (!page)
2414                 goto out;
2415
2416         length = -EFAULT;
2417         if (copy_from_user(page, buf, count))
2418                 goto out_free;
2419
2420         /* Guard against adverse ptrace interaction */
2421         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2422         if (length < 0)
2423                 goto out_free;
2424
2425         length = security_setprocattr(task,
2426                                       (char*)file->f_path.dentry->d_name.name,
2427                                       (void*)page, count);
2428         mutex_unlock(&task->signal->cred_guard_mutex);
2429 out_free:
2430         free_page((unsigned long) page);
2431 out:
2432         put_task_struct(task);
2433 out_no_task:
2434         return length;
2435 }
2436
2437 static const struct file_operations proc_pid_attr_operations = {
2438         .read           = proc_pid_attr_read,
2439         .write          = proc_pid_attr_write,
2440         .llseek         = generic_file_llseek,
2441 };
2442
2443 static const struct pid_entry attr_dir_stuff[] = {
2444         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2445         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2446         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2447         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2448         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2449         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2450 };
2451
2452 static int proc_attr_dir_readdir(struct file * filp,
2453                              void * dirent, filldir_t filldir)
2454 {
2455         return proc_pident_readdir(filp,dirent,filldir,
2456                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2457 }
2458
2459 static const struct file_operations proc_attr_dir_operations = {
2460         .read           = generic_read_dir,
2461         .readdir        = proc_attr_dir_readdir,
2462         .llseek         = default_llseek,
2463 };
2464
2465 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2466                                 struct dentry *dentry, struct nameidata *nd)
2467 {
2468         return proc_pident_lookup(dir, dentry,
2469                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2470 }
2471
2472 static const struct inode_operations proc_attr_dir_inode_operations = {
2473         .lookup         = proc_attr_dir_lookup,
2474         .getattr        = pid_getattr,
2475         .setattr        = proc_setattr,
2476 };
2477
2478 #endif
2479
2480 #ifdef CONFIG_ELF_CORE
2481 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2482                                          size_t count, loff_t *ppos)
2483 {
2484         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2485         struct mm_struct *mm;
2486         char buffer[PROC_NUMBUF];
2487         size_t len;
2488         int ret;
2489
2490         if (!task)
2491                 return -ESRCH;
2492
2493         ret = 0;
2494         mm = get_task_mm(task);
2495         if (mm) {
2496                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2497                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2498                                 MMF_DUMP_FILTER_SHIFT));
2499                 mmput(mm);
2500                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2501         }
2502
2503         put_task_struct(task);
2504
2505         return ret;
2506 }
2507
2508 static ssize_t proc_coredump_filter_write(struct file *file,
2509                                           const char __user *buf,
2510                                           size_t count,
2511                                           loff_t *ppos)
2512 {
2513         struct task_struct *task;
2514         struct mm_struct *mm;
2515         char buffer[PROC_NUMBUF], *end;
2516         unsigned int val;
2517         int ret;
2518         int i;
2519         unsigned long mask;
2520
2521         ret = -EFAULT;
2522         memset(buffer, 0, sizeof(buffer));
2523         if (count > sizeof(buffer) - 1)
2524                 count = sizeof(buffer) - 1;
2525         if (copy_from_user(buffer, buf, count))
2526                 goto out_no_task;
2527
2528         ret = -EINVAL;
2529         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2530         if (*end == '\n')
2531                 end++;
2532         if (end - buffer == 0)
2533                 goto out_no_task;
2534
2535         ret = -ESRCH;
2536         task = get_proc_task(file->f_dentry->d_inode);
2537         if (!task)
2538                 goto out_no_task;
2539
2540         ret = end - buffer;
2541         mm = get_task_mm(task);
2542         if (!mm)
2543                 goto out_no_mm;
2544
2545         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2546                 if (val & mask)
2547                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2548                 else
2549                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2550         }
2551
2552         mmput(mm);
2553  out_no_mm:
2554         put_task_struct(task);
2555  out_no_task:
2556         return ret;
2557 }
2558
2559 static const struct file_operations proc_coredump_filter_operations = {
2560         .read           = proc_coredump_filter_read,
2561         .write          = proc_coredump_filter_write,
2562         .llseek         = generic_file_llseek,
2563 };
2564 #endif
2565
2566 /*
2567  * /proc/self:
2568  */
2569 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2570                               int buflen)
2571 {
2572         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2573         pid_t tgid = task_tgid_nr_ns(current, ns);
2574         char tmp[PROC_NUMBUF];
2575         if (!tgid)
2576                 return -ENOENT;
2577         sprintf(tmp, "%d", tgid);
2578         return vfs_readlink(dentry,buffer,buflen,tmp);
2579 }
2580
2581 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2582 {
2583         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2584         pid_t tgid = task_tgid_nr_ns(current, ns);
2585         char *name = ERR_PTR(-ENOENT);
2586         if (tgid) {
2587                 name = __getname();
2588                 if (!name)
2589                         name = ERR_PTR(-ENOMEM);
2590                 else
2591                         sprintf(name, "%d", tgid);
2592         }
2593         nd_set_link(nd, name);
2594         return NULL;
2595 }
2596
2597 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2598                                 void *cookie)
2599 {
2600         char *s = nd_get_link(nd);
2601         if (!IS_ERR(s))
2602                 __putname(s);
2603 }
2604
2605 static const struct inode_operations proc_self_inode_operations = {
2606         .readlink       = proc_self_readlink,
2607         .follow_link    = proc_self_follow_link,
2608         .put_link       = proc_self_put_link,
2609 };
2610
2611 /*
2612  * proc base
2613  *
2614  * These are the directory entries in the root directory of /proc
2615  * that properly belong to the /proc filesystem, as they describe
2616  * describe something that is process related.
2617  */
2618 static const struct pid_entry proc_base_stuff[] = {
2619         NOD("self", S_IFLNK|S_IRWXUGO,
2620                 &proc_self_inode_operations, NULL, {}),
2621 };
2622
2623 static struct dentry *proc_base_instantiate(struct inode *dir,
2624         struct dentry *dentry, struct task_struct *task, const void *ptr)
2625 {
2626         const struct pid_entry *p = ptr;
2627         struct inode *inode;
2628         struct proc_inode *ei;
2629         struct dentry *error;
2630
2631         /* Allocate the inode */
2632         error = ERR_PTR(-ENOMEM);
2633         inode = new_inode(dir->i_sb);
2634         if (!inode)
2635                 goto out;
2636
2637         /* Initialize the inode */
2638         ei = PROC_I(inode);
2639         inode->i_ino = get_next_ino();
2640         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2641
2642         /*
2643          * grab the reference to the task.
2644          */
2645         ei->pid = get_task_pid(task, PIDTYPE_PID);
2646         if (!ei->pid)
2647                 goto out_iput;
2648
2649         inode->i_mode = p->mode;
2650         if (S_ISDIR(inode->i_mode))
2651                 inode->i_nlink = 2;
2652         if (S_ISLNK(inode->i_mode))
2653                 inode->i_size = 64;
2654         if (p->iop)
2655                 inode->i_op = p->iop;
2656         if (p->fop)
2657                 inode->i_fop = p->fop;
2658         ei->op = p->op;
2659         d_add(dentry, inode);
2660         error = NULL;
2661 out:
2662         return error;
2663 out_iput:
2664         iput(inode);
2665         goto out;
2666 }
2667
2668 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2669 {
2670         struct dentry *error;
2671         struct task_struct *task = get_proc_task(dir);
2672         const struct pid_entry *p, *last;
2673
2674         error = ERR_PTR(-ENOENT);
2675
2676         if (!task)
2677                 goto out_no_task;
2678
2679         /* Lookup the directory entry */
2680         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2681         for (p = proc_base_stuff; p <= last; p++) {
2682                 if (p->len != dentry->d_name.len)
2683                         continue;
2684                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2685                         break;
2686         }
2687         if (p > last)
2688                 goto out;
2689
2690         error = proc_base_instantiate(dir, dentry, task, p);
2691
2692 out:
2693         put_task_struct(task);
2694 out_no_task:
2695         return error;
2696 }
2697
2698 static int proc_base_fill_cache(struct file *filp, void *dirent,
2699         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2700 {
2701         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2702                                 proc_base_instantiate, task, p);
2703 }
2704
2705 #ifdef CONFIG_TASK_IO_ACCOUNTING
2706 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2707 {
2708         struct task_io_accounting acct = task->ioac;
2709         unsigned long flags;
2710
2711         if (whole && lock_task_sighand(task, &flags)) {
2712                 struct task_struct *t = task;
2713
2714                 task_io_accounting_add(&acct, &task->signal->ioac);
2715                 while_each_thread(task, t)
2716                         task_io_accounting_add(&acct, &t->ioac);
2717
2718                 unlock_task_sighand(task, &flags);
2719         }
2720         return sprintf(buffer,
2721                         "rchar: %llu\n"
2722                         "wchar: %llu\n"
2723                         "syscr: %llu\n"
2724                         "syscw: %llu\n"
2725                         "read_bytes: %llu\n"
2726                         "write_bytes: %llu\n"
2727                         "cancelled_write_bytes: %llu\n",
2728                         (unsigned long long)acct.rchar,
2729                         (unsigned long long)acct.wchar,
2730                         (unsigned long long)acct.syscr,
2731                         (unsigned long long)acct.syscw,
2732                         (unsigned long long)acct.read_bytes,
2733                         (unsigned long long)acct.write_bytes,
2734                         (unsigned long long)acct.cancelled_write_bytes);
2735 }
2736
2737 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2738 {
2739         return do_io_accounting(task, buffer, 0);
2740 }
2741
2742 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2743 {
2744         return do_io_accounting(task, buffer, 1);
2745 }
2746 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2747
2748 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2749                                 struct pid *pid, struct task_struct *task)
2750 {
2751         seq_printf(m, "%08x\n", task->personality);
2752         return 0;
2753 }
2754
2755 /*
2756  * Thread groups
2757  */
2758 static const struct file_operations proc_task_operations;
2759 static const struct inode_operations proc_task_inode_operations;
2760
2761 static const struct pid_entry tgid_base_stuff[] = {
2762         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2763         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2764         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2765 #ifdef CONFIG_NET
2766         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2767 #endif
2768         REG("environ",    S_IRUSR, proc_environ_operations),
2769         INF("auxv",       S_IRUSR, proc_pid_auxv),
2770         ONE("status",     S_IRUGO, proc_pid_status),
2771         ONE("personality", S_IRUSR, proc_pid_personality),
2772         INF("limits",     S_IRUGO, proc_pid_limits),
2773 #ifdef CONFIG_SCHED_DEBUG
2774         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2775 #endif
2776 #ifdef CONFIG_SCHED_AUTOGROUP
2777         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2778 #endif
2779         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2780 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2781         INF("syscall",    S_IRUSR, proc_pid_syscall),
2782 #endif
2783         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2784         ONE("stat",       S_IRUGO, proc_tgid_stat),
2785         ONE("statm",      S_IRUGO, proc_pid_statm),
2786         REG("maps",       S_IRUGO, proc_maps_operations),
2787 #ifdef CONFIG_NUMA
2788         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2789 #endif
2790         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2791         LNK("cwd",        proc_cwd_link),
2792         LNK("root",       proc_root_link),
2793         LNK("exe",        proc_exe_link),
2794         REG("mounts",     S_IRUGO, proc_mounts_operations),
2795         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2796         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2797 #ifdef CONFIG_PROC_PAGE_MONITOR
2798         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2799         REG("smaps",      S_IRUGO, proc_smaps_operations),
2800         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2801 #endif
2802 #ifdef CONFIG_SECURITY
2803         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2804 #endif
2805 #ifdef CONFIG_KALLSYMS
2806         INF("wchan",      S_IRUGO, proc_pid_wchan),
2807 #endif
2808 #ifdef CONFIG_STACKTRACE
2809         ONE("stack",      S_IRUSR, proc_pid_stack),
2810 #endif
2811 #ifdef CONFIG_SCHEDSTATS
2812         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2813 #endif
2814 #ifdef CONFIG_LATENCYTOP
2815         REG("latency",  S_IRUGO, proc_lstats_operations),
2816 #endif
2817 #ifdef CONFIG_PROC_PID_CPUSET
2818         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2819 #endif
2820 #ifdef CONFIG_CGROUPS
2821         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2822 #endif
2823         INF("oom_score",  S_IRUGO, proc_oom_score),
2824         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2825         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2826 #ifdef CONFIG_AUDITSYSCALL
2827         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2828         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2829 #endif
2830 #ifdef CONFIG_FAULT_INJECTION
2831         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2832 #endif
2833 #ifdef CONFIG_ELF_CORE
2834         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2835 #endif
2836 #ifdef CONFIG_TASK_IO_ACCOUNTING
2837         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2838 #endif
2839 };
2840
2841 static int proc_tgid_base_readdir(struct file * filp,
2842                              void * dirent, filldir_t filldir)
2843 {
2844         return proc_pident_readdir(filp,dirent,filldir,
2845                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2846 }
2847
2848 static const struct file_operations proc_tgid_base_operations = {
2849         .read           = generic_read_dir,
2850         .readdir        = proc_tgid_base_readdir,
2851         .llseek         = default_llseek,
2852 };
2853
2854 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2855         return proc_pident_lookup(dir, dentry,
2856                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2857 }
2858
2859 static const struct inode_operations proc_tgid_base_inode_operations = {
2860         .lookup         = proc_tgid_base_lookup,
2861         .getattr        = pid_getattr,
2862         .setattr        = proc_setattr,
2863 };
2864
2865 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2866 {
2867         struct dentry *dentry, *leader, *dir;
2868         char buf[PROC_NUMBUF];
2869         struct qstr name;
2870
2871         name.name = buf;
2872         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2873         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2874         if (dentry) {
2875                 shrink_dcache_parent(dentry);
2876                 d_drop(dentry);
2877                 dput(dentry);
2878         }
2879
2880         name.name = buf;
2881         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2882         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2883         if (!leader)
2884                 goto out;
2885
2886         name.name = "task";
2887         name.len = strlen(name.name);
2888         dir = d_hash_and_lookup(leader, &name);
2889         if (!dir)
2890                 goto out_put_leader;
2891
2892         name.name = buf;
2893         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2894         dentry = d_hash_and_lookup(dir, &name);
2895         if (dentry) {
2896                 shrink_dcache_parent(dentry);
2897                 d_drop(dentry);
2898                 dput(dentry);
2899         }
2900
2901         dput(dir);
2902 out_put_leader:
2903         dput(leader);
2904 out:
2905         return;
2906 }
2907
2908 /**
2909  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2910  * @task: task that should be flushed.
2911  *
2912  * When flushing dentries from proc, one needs to flush them from global
2913  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2914  * in. This call is supposed to do all of this job.
2915  *
2916  * Looks in the dcache for
2917  * /proc/@pid
2918  * /proc/@tgid/task/@pid
2919  * if either directory is present flushes it and all of it'ts children
2920  * from the dcache.
2921  *
2922  * It is safe and reasonable to cache /proc entries for a task until
2923  * that task exits.  After that they just clog up the dcache with
2924  * useless entries, possibly causing useful dcache entries to be
2925  * flushed instead.  This routine is proved to flush those useless
2926  * dcache entries at process exit time.
2927  *
2928  * NOTE: This routine is just an optimization so it does not guarantee
2929  *       that no dcache entries will exist at process exit time it
2930  *       just makes it very unlikely that any will persist.
2931  */
2932
2933 void proc_flush_task(struct task_struct *task)
2934 {
2935         int i;
2936         struct pid *pid, *tgid;
2937         struct upid *upid;
2938
2939         pid = task_pid(task);
2940         tgid = task_tgid(task);
2941
2942         for (i = 0; i <= pid->level; i++) {
2943                 upid = &pid->numbers[i];
2944                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2945                                         tgid->numbers[i].nr);
2946         }
2947
2948         upid = &pid->numbers[pid->level];
2949         if (upid->nr == 1)
2950                 pid_ns_release_proc(upid->ns);
2951 }
2952
2953 static struct dentry *proc_pid_instantiate(struct inode *dir,
2954                                            struct dentry * dentry,
2955                                            struct task_struct *task, const void *ptr)
2956 {
2957         struct dentry *error = ERR_PTR(-ENOENT);
2958         struct inode *inode;
2959
2960         inode = proc_pid_make_inode(dir->i_sb, task);
2961         if (!inode)
2962                 goto out;
2963
2964         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2965         inode->i_op = &proc_tgid_base_inode_operations;
2966         inode->i_fop = &proc_tgid_base_operations;
2967         inode->i_flags|=S_IMMUTABLE;
2968
2969         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2970                 ARRAY_SIZE(tgid_base_stuff));
2971
2972         d_set_d_op(dentry, &pid_dentry_operations);
2973
2974         d_add(dentry, inode);
2975         /* Close the race of the process dying before we return the dentry */
2976         if (pid_revalidate(dentry, NULL))
2977                 error = NULL;
2978 out:
2979         return error;
2980 }
2981
2982 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2983 {
2984         struct dentry *result;
2985         struct task_struct *task;
2986         unsigned tgid;
2987         struct pid_namespace *ns;
2988
2989         result = proc_base_lookup(dir, dentry);
2990         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2991                 goto out;
2992
2993         tgid = name_to_int(dentry);
2994         if (tgid == ~0U)
2995                 goto out;
2996
2997         ns = dentry->d_sb->s_fs_info;
2998         rcu_read_lock();
2999         task = find_task_by_pid_ns(tgid, ns);
3000         if (task)
3001                 get_task_struct(task);
3002         rcu_read_unlock();
3003         if (!task)
3004                 goto out;
3005
3006         result = proc_pid_instantiate(dir, dentry, task, NULL);
3007         put_task_struct(task);
3008 out:
3009         return result;
3010 }
3011
3012 /*
3013  * Find the first task with tgid >= tgid
3014  *
3015  */
3016 struct tgid_iter {
3017         unsigned int tgid;
3018         struct task_struct *task;
3019 };
3020 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3021 {
3022         struct pid *pid;
3023
3024         if (iter.task)
3025                 put_task_struct(iter.task);
3026         rcu_read_lock();
3027 retry:
3028         iter.task = NULL;
3029         pid = find_ge_pid(iter.tgid, ns);
3030         if (pid) {
3031                 iter.tgid = pid_nr_ns(pid, ns);
3032                 iter.task = pid_task(pid, PIDTYPE_PID);
3033                 /* What we to know is if the pid we have find is the
3034                  * pid of a thread_group_leader.  Testing for task
3035                  * being a thread_group_leader is the obvious thing
3036                  * todo but there is a window when it fails, due to
3037                  * the pid transfer logic in de_thread.
3038                  *
3039                  * So we perform the straight forward test of seeing
3040                  * if the pid we have found is the pid of a thread
3041                  * group leader, and don't worry if the task we have
3042                  * found doesn't happen to be a thread group leader.
3043                  * As we don't care in the case of readdir.
3044                  */
3045                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3046                         iter.tgid += 1;
3047                         goto retry;
3048                 }
3049                 get_task_struct(iter.task);
3050         }
3051         rcu_read_unlock();
3052         return iter;
3053 }
3054
3055 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3056
3057 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3058         struct tgid_iter iter)
3059 {
3060         char name[PROC_NUMBUF];
3061         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3062         return proc_fill_cache(filp, dirent, filldir, name, len,
3063                                 proc_pid_instantiate, iter.task, NULL);
3064 }
3065
3066 /* for the /proc/ directory itself, after non-process stuff has been done */
3067 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3068 {
3069         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3070         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3071         struct tgid_iter iter;
3072         struct pid_namespace *ns;
3073
3074         if (!reaper)
3075                 goto out_no_task;
3076
3077         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3078                 const struct pid_entry *p = &proc_base_stuff[nr];
3079                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3080                         goto out;
3081         }
3082
3083         ns = filp->f_dentry->d_sb->s_fs_info;
3084         iter.task = NULL;
3085         iter.tgid = filp->f_pos - TGID_OFFSET;
3086         for (iter = next_tgid(ns, iter);
3087              iter.task;
3088              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3089                 filp->f_pos = iter.tgid + TGID_OFFSET;
3090                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3091                         put_task_struct(iter.task);
3092                         goto out;
3093                 }
3094         }
3095         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3096 out:
3097         put_task_struct(reaper);
3098 out_no_task:
3099         return 0;
3100 }
3101
3102 /*
3103  * Tasks
3104  */
3105 static const struct pid_entry tid_base_stuff[] = {
3106         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3107         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3108         REG("environ",   S_IRUSR, proc_environ_operations),
3109         INF("auxv",      S_IRUSR, proc_pid_auxv),
3110         ONE("status",    S_IRUGO, proc_pid_status),
3111         ONE("personality", S_IRUSR, proc_pid_personality),
3112         INF("limits",    S_IRUGO, proc_pid_limits),
3113 #ifdef CONFIG_SCHED_DEBUG
3114         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3115 #endif
3116         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3117 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3118         INF("syscall",   S_IRUSR, proc_pid_syscall),
3119 #endif
3120         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3121         ONE("stat",      S_IRUGO, proc_tid_stat),
3122         ONE("statm",     S_IRUGO, proc_pid_statm),
3123         REG("maps",      S_IRUGO, proc_maps_operations),
3124 #ifdef CONFIG_NUMA
3125         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3126 #endif
3127         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3128         LNK("cwd",       proc_cwd_link),
3129         LNK("root",      proc_root_link),
3130         LNK("exe",       proc_exe_link),
3131         REG("mounts",    S_IRUGO, proc_mounts_operations),
3132         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3133 #ifdef CONFIG_PROC_PAGE_MONITOR
3134         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3135         REG("smaps",     S_IRUGO, proc_smaps_operations),
3136         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3137 #endif
3138 #ifdef CONFIG_SECURITY
3139         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3140 #endif
3141 #ifdef CONFIG_KALLSYMS
3142         INF("wchan",     S_IRUGO, proc_pid_wchan),
3143 #endif
3144 #ifdef CONFIG_STACKTRACE
3145         ONE("stack",      S_IRUSR, proc_pid_stack),
3146 #endif
3147 #ifdef CONFIG_SCHEDSTATS
3148         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3149 #endif
3150 #ifdef CONFIG_LATENCYTOP
3151         REG("latency",  S_IRUGO, proc_lstats_operations),
3152 #endif
3153 #ifdef CONFIG_PROC_PID_CPUSET
3154         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3155 #endif
3156 #ifdef CONFIG_CGROUPS
3157         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3158 #endif
3159         INF("oom_score", S_IRUGO, proc_oom_score),
3160         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3161         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3162 #ifdef CONFIG_AUDITSYSCALL
3163         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3164         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3165 #endif
3166 #ifdef CONFIG_FAULT_INJECTION
3167         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3168 #endif
3169 #ifdef CONFIG_TASK_IO_ACCOUNTING
3170         INF("io",       S_IRUGO, proc_tid_io_accounting),
3171 #endif
3172 };
3173
3174 static int proc_tid_base_readdir(struct file * filp,
3175                              void * dirent, filldir_t filldir)
3176 {
3177         return proc_pident_readdir(filp,dirent,filldir,
3178                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3179 }
3180
3181 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3182         return proc_pident_lookup(dir, dentry,
3183                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3184 }
3185
3186 static const struct file_operations proc_tid_base_operations = {
3187         .read           = generic_read_dir,
3188         .readdir        = proc_tid_base_readdir,
3189         .llseek         = default_llseek,
3190 };
3191
3192 static const struct inode_operations proc_tid_base_inode_operations = {
3193         .lookup         = proc_tid_base_lookup,
3194         .getattr        = pid_getattr,
3195         .setattr        = proc_setattr,
3196 };
3197
3198 static struct dentry *proc_task_instantiate(struct inode *dir,
3199         struct dentry *dentry, struct task_struct *task, const void *ptr)
3200 {
3201         struct dentry *error = ERR_PTR(-ENOENT);
3202         struct inode *inode;
3203         inode = proc_pid_make_inode(dir->i_sb, task);
3204
3205         if (!inode)
3206                 goto out;
3207         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3208         inode->i_op = &proc_tid_base_inode_operations;
3209         inode->i_fop = &proc_tid_base_operations;
3210         inode->i_flags|=S_IMMUTABLE;
3211
3212         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3213                 ARRAY_SIZE(tid_base_stuff));
3214
3215         d_set_d_op(dentry, &pid_dentry_operations);
3216
3217         d_add(dentry, inode);
3218         /* Close the race of the process dying before we return the dentry */
3219         if (pid_revalidate(dentry, NULL))
3220                 error = NULL;
3221 out:
3222         return error;
3223 }
3224
3225 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3226 {
3227         struct dentry *result = ERR_PTR(-ENOENT);
3228         struct task_struct *task;
3229         struct task_struct *leader = get_proc_task(dir);
3230         unsigned tid;
3231         struct pid_namespace *ns;
3232
3233         if (!leader)
3234                 goto out_no_task;
3235
3236         tid = name_to_int(dentry);
3237         if (tid == ~0U)
3238                 goto out;
3239
3240         ns = dentry->d_sb->s_fs_info;
3241         rcu_read_lock();
3242         task = find_task_by_pid_ns(tid, ns);
3243         if (task)
3244                 get_task_struct(task);
3245         rcu_read_unlock();
3246         if (!task)
3247                 goto out;
3248         if (!same_thread_group(leader, task))
3249                 goto out_drop_task;
3250
3251         result = proc_task_instantiate(dir, dentry, task, NULL);
3252 out_drop_task:
3253         put_task_struct(task);
3254 out:
3255         put_task_struct(leader);
3256 out_no_task:
3257         return result;
3258 }
3259
3260 /*
3261  * Find the first tid of a thread group to return to user space.
3262  *
3263  * Usually this is just the thread group leader, but if the users
3264  * buffer was too small or there was a seek into the middle of the
3265  * directory we have more work todo.
3266  *
3267  * In the case of a short read we start with find_task_by_pid.
3268  *
3269  * In the case of a seek we start with the leader and walk nr
3270  * threads past it.
3271  */
3272 static struct task_struct *first_tid(struct task_struct *leader,
3273                 int tid, int nr, struct pid_namespace *ns)
3274 {
3275         struct task_struct *pos;
3276
3277         rcu_read_lock();
3278         /* Attempt to start with the pid of a thread */
3279         if (tid && (nr > 0)) {
3280                 pos = find_task_by_pid_ns(tid, ns);
3281                 if (pos && (pos->group_leader == leader))
3282                         goto found;
3283         }
3284
3285         /* If nr exceeds the number of threads there is nothing todo */
3286         pos = NULL;
3287         if (nr && nr >= get_nr_threads(leader))
3288                 goto out;
3289
3290         /* If we haven't found our starting place yet start
3291          * with the leader and walk nr threads forward.
3292          */
3293         for (pos = leader; nr > 0; --nr) {
3294                 pos = next_thread(pos);
3295                 if (pos == leader) {
3296                         pos = NULL;
3297                         goto out;
3298                 }
3299         }
3300 found:
3301         get_task_struct(pos);
3302 out:
3303         rcu_read_unlock();
3304         return pos;
3305 }
3306
3307 /*
3308  * Find the next thread in the thread list.
3309  * Return NULL if there is an error or no next thread.
3310  *
3311  * The reference to the input task_struct is released.
3312  */
3313 static struct task_struct *next_tid(struct task_struct *start)
3314 {
3315         struct task_struct *pos = NULL;
3316         rcu_read_lock();
3317         if (pid_alive(start)) {
3318                 pos = next_thread(start);
3319                 if (thread_group_leader(pos))
3320                         pos = NULL;
3321                 else
3322                         get_task_struct(pos);
3323         }
3324         rcu_read_unlock();
3325         put_task_struct(start);
3326         return pos;
3327 }
3328
3329 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3330         struct task_struct *task, int tid)
3331 {
3332         char name[PROC_NUMBUF];
3333         int len = snprintf(name, sizeof(name), "%d", tid);
3334         return proc_fill_cache(filp, dirent, filldir, name, len,
3335                                 proc_task_instantiate, task, NULL);
3336 }
3337
3338 /* for the /proc/TGID/task/ directories */
3339 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3340 {
3341         struct dentry *dentry = filp->f_path.dentry;
3342         struct inode *inode = dentry->d_inode;
3343         struct task_struct *leader = NULL;
3344         struct task_struct *task;
3345         int retval = -ENOENT;
3346         ino_t ino;
3347         int tid;
3348         struct pid_namespace *ns;
3349
3350         task = get_proc_task(inode);
3351         if (!task)
3352                 goto out_no_task;
3353         rcu_read_lock();
3354         if (pid_alive(task)) {
3355                 leader = task->group_leader;
3356                 get_task_struct(leader);
3357         }
3358         rcu_read_unlock();
3359         put_task_struct(task);
3360         if (!leader)
3361                 goto out_no_task;
3362         retval = 0;
3363
3364         switch ((unsigned long)filp->f_pos) {
3365         case 0:
3366                 ino = inode->i_ino;
3367                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3368                         goto out;
3369                 filp->f_pos++;
3370                 /* fall through */
3371         case 1:
3372                 ino = parent_ino(dentry);
3373                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3374                         goto out;
3375                 filp->f_pos++;
3376                 /* fall through */
3377         }
3378
3379         /* f_version caches the tgid value that the last readdir call couldn't
3380          * return. lseek aka telldir automagically resets f_version to 0.
3381          */
3382         ns = filp->f_dentry->d_sb->s_fs_info;
3383         tid = (int)filp->f_version;
3384         filp->f_version = 0;
3385         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3386              task;
3387              task = next_tid(task), filp->f_pos++) {
3388                 tid = task_pid_nr_ns(task, ns);
3389                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3390                         /* returning this tgid failed, save it as the first
3391                          * pid for the next readir call */
3392                         filp->f_version = (u64)tid;
3393                         put_task_struct(task);
3394                         break;
3395                 }
3396         }
3397 out:
3398         put_task_struct(leader);
3399 out_no_task:
3400         return retval;
3401 }
3402
3403 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3404 {
3405         struct inode *inode = dentry->d_inode;
3406         struct task_struct *p = get_proc_task(inode);
3407         generic_fillattr(inode, stat);
3408
3409         if (p) {
3410                 stat->nlink += get_nr_threads(p);
3411                 put_task_struct(p);
3412         }
3413
3414         return 0;
3415 }
3416
3417 static const struct inode_operations proc_task_inode_operations = {
3418         .lookup         = proc_task_lookup,
3419         .getattr        = proc_task_getattr,
3420         .setattr        = proc_setattr,
3421 };
3422
3423 static const struct file_operations proc_task_operations = {
3424         .read           = generic_read_dir,
3425         .readdir        = proc_task_readdir,
3426         .llseek         = default_llseek,
3427 };