procfs: make errno values consistent when open pident vs exit(2) race occurs
[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/rcupdate.h>
67 #include <linux/kallsyms.h>
68 #include <linux/stacktrace.h>
69 #include <linux/resource.h>
70 #include <linux/module.h>
71 #include <linux/mount.h>
72 #include <linux/security.h>
73 #include <linux/ptrace.h>
74 #include <linux/tracehook.h>
75 #include <linux/cgroup.h>
76 #include <linux/cpuset.h>
77 #include <linux/audit.h>
78 #include <linux/poll.h>
79 #include <linux/nsproxy.h>
80 #include <linux/oom.h>
81 #include <linux/elf.h>
82 #include <linux/pid_namespace.h>
83 #include <linux/fs_struct.h>
84 #include "internal.h"
85
86 /* NOTE:
87  *      Implementing inode permission operations in /proc is almost
88  *      certainly an error.  Permission checks need to happen during
89  *      each system call not at open time.  The reason is that most of
90  *      what we wish to check for permissions in /proc varies at runtime.
91  *
92  *      The classic example of a problem is opening file descriptors
93  *      in /proc for a task before it execs a suid executable.
94  */
95
96 struct pid_entry {
97         char *name;
98         int len;
99         mode_t mode;
100         const struct inode_operations *iop;
101         const struct file_operations *fop;
102         union proc_op op;
103 };
104
105 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
106         .name = (NAME),                                 \
107         .len  = sizeof(NAME) - 1,                       \
108         .mode = MODE,                                   \
109         .iop  = IOP,                                    \
110         .fop  = FOP,                                    \
111         .op   = OP,                                     \
112 }
113
114 #define DIR(NAME, MODE, iops, fops)     \
115         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
116 #define LNK(NAME, get_link)                                     \
117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
118                 &proc_pid_link_inode_operations, NULL,          \
119                 { .proc_get_link = get_link } )
120 #define REG(NAME, MODE, fops)                           \
121         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
122 #define INF(NAME, MODE, read)                           \
123         NOD(NAME, (S_IFREG|(MODE)),                     \
124                 NULL, &proc_info_file_operations,       \
125                 { .proc_read = read } )
126 #define ONE(NAME, MODE, show)                           \
127         NOD(NAME, (S_IFREG|(MODE)),                     \
128                 NULL, &proc_single_file_operations,     \
129                 { .proc_show = show } )
130
131 /*
132  * Count the number of hardlinks for the pid_entry table, excluding the .
133  * and .. links.
134  */
135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
136         unsigned int n)
137 {
138         unsigned int i;
139         unsigned int count;
140
141         count = 0;
142         for (i = 0; i < n; ++i) {
143                 if (S_ISDIR(entries[i].mode))
144                         ++count;
145         }
146
147         return count;
148 }
149
150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
151 {
152         struct fs_struct *fs;
153         int result = -ENOENT;
154
155         task_lock(task);
156         fs = task->fs;
157         if (fs) {
158                 read_lock(&fs->lock);
159                 *path = root ? fs->root : fs->pwd;
160                 path_get(path);
161                 read_unlock(&fs->lock);
162                 result = 0;
163         }
164         task_unlock(task);
165         return result;
166 }
167
168 static int get_nr_threads(struct task_struct *tsk)
169 {
170         unsigned long flags;
171         int count = 0;
172
173         if (lock_task_sighand(tsk, &flags)) {
174                 count = atomic_read(&tsk->signal->count);
175                 unlock_task_sighand(tsk, &flags);
176         }
177         return count;
178 }
179
180 static int proc_cwd_link(struct inode *inode, struct path *path)
181 {
182         struct task_struct *task = get_proc_task(inode);
183         int result = -ENOENT;
184
185         if (task) {
186                 result = get_fs_path(task, path, 0);
187                 put_task_struct(task);
188         }
189         return result;
190 }
191
192 static int proc_root_link(struct inode *inode, struct path *path)
193 {
194         struct task_struct *task = get_proc_task(inode);
195         int result = -ENOENT;
196
197         if (task) {
198                 result = get_fs_path(task, path, 1);
199                 put_task_struct(task);
200         }
201         return result;
202 }
203
204 /*
205  * Return zero if current may access user memory in @task, -error if not.
206  */
207 static int check_mem_permission(struct task_struct *task)
208 {
209         /*
210          * A task can always look at itself, in case it chooses
211          * to use system calls instead of load instructions.
212          */
213         if (task == current)
214                 return 0;
215
216         /*
217          * If current is actively ptrace'ing, and would also be
218          * permitted to freshly attach with ptrace now, permit it.
219          */
220         if (task_is_stopped_or_traced(task)) {
221                 int match;
222                 rcu_read_lock();
223                 match = (tracehook_tracer_task(task) == current);
224                 rcu_read_unlock();
225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
226                         return 0;
227         }
228
229         /*
230          * Noone else is allowed.
231          */
232         return -EPERM;
233 }
234
235 struct mm_struct *mm_for_maps(struct task_struct *task)
236 {
237         struct mm_struct *mm = get_task_mm(task);
238         if (!mm)
239                 return NULL;
240         down_read(&mm->mmap_sem);
241         task_lock(task);
242         if (task->mm != mm)
243                 goto out;
244         if (task->mm != current->mm &&
245             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
246                 goto out;
247         task_unlock(task);
248         return mm;
249 out:
250         task_unlock(task);
251         up_read(&mm->mmap_sem);
252         mmput(mm);
253         return NULL;
254 }
255
256 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
257 {
258         int res = 0;
259         unsigned int len;
260         struct mm_struct *mm = get_task_mm(task);
261         if (!mm)
262                 goto out;
263         if (!mm->arg_end)
264                 goto out_mm;    /* Shh! No looking before we're done */
265
266         len = mm->arg_end - mm->arg_start;
267  
268         if (len > PAGE_SIZE)
269                 len = PAGE_SIZE;
270  
271         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
272
273         // If the nul at the end of args has been overwritten, then
274         // assume application is using setproctitle(3).
275         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
276                 len = strnlen(buffer, res);
277                 if (len < res) {
278                     res = len;
279                 } else {
280                         len = mm->env_end - mm->env_start;
281                         if (len > PAGE_SIZE - res)
282                                 len = PAGE_SIZE - res;
283                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
284                         res = strnlen(buffer, res);
285                 }
286         }
287 out_mm:
288         mmput(mm);
289 out:
290         return res;
291 }
292
293 static int proc_pid_auxv(struct task_struct *task, char *buffer)
294 {
295         int res = 0;
296         struct mm_struct *mm = get_task_mm(task);
297         if (mm) {
298                 unsigned int nwords = 0;
299                 do {
300                         nwords += 2;
301                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
302                 res = nwords * sizeof(mm->saved_auxv[0]);
303                 if (res > PAGE_SIZE)
304                         res = PAGE_SIZE;
305                 memcpy(buffer, mm->saved_auxv, res);
306                 mmput(mm);
307         }
308         return res;
309 }
310
311
312 #ifdef CONFIG_KALLSYMS
313 /*
314  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
315  * Returns the resolved symbol.  If that fails, simply return the address.
316  */
317 static int proc_pid_wchan(struct task_struct *task, char *buffer)
318 {
319         unsigned long wchan;
320         char symname[KSYM_NAME_LEN];
321
322         wchan = get_wchan(task);
323
324         if (lookup_symbol_name(wchan, symname) < 0)
325                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
326                         return 0;
327                 else
328                         return sprintf(buffer, "%lu", wchan);
329         else
330                 return sprintf(buffer, "%s", symname);
331 }
332 #endif /* CONFIG_KALLSYMS */
333
334 #ifdef CONFIG_STACKTRACE
335
336 #define MAX_STACK_TRACE_DEPTH   64
337
338 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
339                           struct pid *pid, struct task_struct *task)
340 {
341         struct stack_trace trace;
342         unsigned long *entries;
343         int i;
344
345         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
346         if (!entries)
347                 return -ENOMEM;
348
349         trace.nr_entries        = 0;
350         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
351         trace.entries           = entries;
352         trace.skip              = 0;
353         save_stack_trace_tsk(task, &trace);
354
355         for (i = 0; i < trace.nr_entries; i++) {
356                 seq_printf(m, "[<%p>] %pS\n",
357                            (void *)entries[i], (void *)entries[i]);
358         }
359         kfree(entries);
360
361         return 0;
362 }
363 #endif
364
365 #ifdef CONFIG_SCHEDSTATS
366 /*
367  * Provides /proc/PID/schedstat
368  */
369 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
370 {
371         return sprintf(buffer, "%llu %llu %lu\n",
372                         (unsigned long long)task->se.sum_exec_runtime,
373                         (unsigned long long)task->sched_info.run_delay,
374                         task->sched_info.pcount);
375 }
376 #endif
377
378 #ifdef CONFIG_LATENCYTOP
379 static int lstats_show_proc(struct seq_file *m, void *v)
380 {
381         int i;
382         struct inode *inode = m->private;
383         struct task_struct *task = get_proc_task(inode);
384
385         if (!task)
386                 return -ESRCH;
387         seq_puts(m, "Latency Top version : v0.1\n");
388         for (i = 0; i < 32; i++) {
389                 if (task->latency_record[i].backtrace[0]) {
390                         int q;
391                         seq_printf(m, "%i %li %li ",
392                                 task->latency_record[i].count,
393                                 task->latency_record[i].time,
394                                 task->latency_record[i].max);
395                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
396                                 char sym[KSYM_SYMBOL_LEN];
397                                 char *c;
398                                 if (!task->latency_record[i].backtrace[q])
399                                         break;
400                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
401                                         break;
402                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
403                                 c = strchr(sym, '+');
404                                 if (c)
405                                         *c = 0;
406                                 seq_printf(m, "%s ", sym);
407                         }
408                         seq_printf(m, "\n");
409                 }
410
411         }
412         put_task_struct(task);
413         return 0;
414 }
415
416 static int lstats_open(struct inode *inode, struct file *file)
417 {
418         return single_open(file, lstats_show_proc, inode);
419 }
420
421 static ssize_t lstats_write(struct file *file, const char __user *buf,
422                             size_t count, loff_t *offs)
423 {
424         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
425
426         if (!task)
427                 return -ESRCH;
428         clear_all_latency_tracing(task);
429         put_task_struct(task);
430
431         return count;
432 }
433
434 static const struct file_operations proc_lstats_operations = {
435         .open           = lstats_open,
436         .read           = seq_read,
437         .write          = lstats_write,
438         .llseek         = seq_lseek,
439         .release        = single_release,
440 };
441
442 #endif
443
444 /* The badness from the OOM killer */
445 unsigned long badness(struct task_struct *p, unsigned long uptime);
446 static int proc_oom_score(struct task_struct *task, char *buffer)
447 {
448         unsigned long points;
449         struct timespec uptime;
450
451         do_posix_clock_monotonic_gettime(&uptime);
452         read_lock(&tasklist_lock);
453         points = badness(task, uptime.tv_sec);
454         read_unlock(&tasklist_lock);
455         return sprintf(buffer, "%lu\n", points);
456 }
457
458 struct limit_names {
459         char *name;
460         char *unit;
461 };
462
463 static const struct limit_names lnames[RLIM_NLIMITS] = {
464         [RLIMIT_CPU] = {"Max cpu time", "ms"},
465         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
466         [RLIMIT_DATA] = {"Max data size", "bytes"},
467         [RLIMIT_STACK] = {"Max stack size", "bytes"},
468         [RLIMIT_CORE] = {"Max core file size", "bytes"},
469         [RLIMIT_RSS] = {"Max resident set", "bytes"},
470         [RLIMIT_NPROC] = {"Max processes", "processes"},
471         [RLIMIT_NOFILE] = {"Max open files", "files"},
472         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
473         [RLIMIT_AS] = {"Max address space", "bytes"},
474         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
475         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
476         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
477         [RLIMIT_NICE] = {"Max nice priority", NULL},
478         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
479         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
480 };
481
482 /* Display limits for a process */
483 static int proc_pid_limits(struct task_struct *task, char *buffer)
484 {
485         unsigned int i;
486         int count = 0;
487         unsigned long flags;
488         char *bufptr = buffer;
489
490         struct rlimit rlim[RLIM_NLIMITS];
491
492         if (!lock_task_sighand(task, &flags))
493                 return 0;
494         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
495         unlock_task_sighand(task, &flags);
496
497         /*
498          * print the file header
499          */
500         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
501                         "Limit", "Soft Limit", "Hard Limit", "Units");
502
503         for (i = 0; i < RLIM_NLIMITS; i++) {
504                 if (rlim[i].rlim_cur == RLIM_INFINITY)
505                         count += sprintf(&bufptr[count], "%-25s %-20s ",
506                                          lnames[i].name, "unlimited");
507                 else
508                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
509                                          lnames[i].name, rlim[i].rlim_cur);
510
511                 if (rlim[i].rlim_max == RLIM_INFINITY)
512                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
513                 else
514                         count += sprintf(&bufptr[count], "%-20lu ",
515                                          rlim[i].rlim_max);
516
517                 if (lnames[i].unit)
518                         count += sprintf(&bufptr[count], "%-10s\n",
519                                          lnames[i].unit);
520                 else
521                         count += sprintf(&bufptr[count], "\n");
522         }
523
524         return count;
525 }
526
527 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
528 static int proc_pid_syscall(struct task_struct *task, char *buffer)
529 {
530         long nr;
531         unsigned long args[6], sp, pc;
532
533         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
534                 return sprintf(buffer, "running\n");
535
536         if (nr < 0)
537                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
538
539         return sprintf(buffer,
540                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
541                        nr,
542                        args[0], args[1], args[2], args[3], args[4], args[5],
543                        sp, pc);
544 }
545 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
546
547 /************************************************************************/
548 /*                       Here the fs part begins                        */
549 /************************************************************************/
550
551 /* permission checks */
552 static int proc_fd_access_allowed(struct inode *inode)
553 {
554         struct task_struct *task;
555         int allowed = 0;
556         /* Allow access to a task's file descriptors if it is us or we
557          * may use ptrace attach to the process and find out that
558          * information.
559          */
560         task = get_proc_task(inode);
561         if (task) {
562                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
563                 put_task_struct(task);
564         }
565         return allowed;
566 }
567
568 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
569 {
570         int error;
571         struct inode *inode = dentry->d_inode;
572
573         if (attr->ia_valid & ATTR_MODE)
574                 return -EPERM;
575
576         error = inode_change_ok(inode, attr);
577         if (!error)
578                 error = inode_setattr(inode, attr);
579         return error;
580 }
581
582 static const struct inode_operations proc_def_inode_operations = {
583         .setattr        = proc_setattr,
584 };
585
586 static int mounts_open_common(struct inode *inode, struct file *file,
587                               const struct seq_operations *op)
588 {
589         struct task_struct *task = get_proc_task(inode);
590         struct nsproxy *nsp;
591         struct mnt_namespace *ns = NULL;
592         struct path root;
593         struct proc_mounts *p;
594         int ret = -EINVAL;
595
596         if (task) {
597                 rcu_read_lock();
598                 nsp = task_nsproxy(task);
599                 if (nsp) {
600                         ns = nsp->mnt_ns;
601                         if (ns)
602                                 get_mnt_ns(ns);
603                 }
604                 rcu_read_unlock();
605                 if (ns && get_fs_path(task, &root, 1) == 0)
606                         ret = 0;
607                 put_task_struct(task);
608         }
609
610         if (!ns)
611                 goto err;
612         if (ret)
613                 goto err_put_ns;
614
615         ret = -ENOMEM;
616         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
617         if (!p)
618                 goto err_put_path;
619
620         file->private_data = &p->m;
621         ret = seq_open(file, op);
622         if (ret)
623                 goto err_free;
624
625         p->m.private = p;
626         p->ns = ns;
627         p->root = root;
628         p->event = ns->event;
629
630         return 0;
631
632  err_free:
633         kfree(p);
634  err_put_path:
635         path_put(&root);
636  err_put_ns:
637         put_mnt_ns(ns);
638  err:
639         return ret;
640 }
641
642 static int mounts_release(struct inode *inode, struct file *file)
643 {
644         struct proc_mounts *p = file->private_data;
645         path_put(&p->root);
646         put_mnt_ns(p->ns);
647         return seq_release(inode, file);
648 }
649
650 static unsigned mounts_poll(struct file *file, poll_table *wait)
651 {
652         struct proc_mounts *p = file->private_data;
653         struct mnt_namespace *ns = p->ns;
654         unsigned res = POLLIN | POLLRDNORM;
655
656         poll_wait(file, &ns->poll, wait);
657
658         spin_lock(&vfsmount_lock);
659         if (p->event != ns->event) {
660                 p->event = ns->event;
661                 res |= POLLERR | POLLPRI;
662         }
663         spin_unlock(&vfsmount_lock);
664
665         return res;
666 }
667
668 static int mounts_open(struct inode *inode, struct file *file)
669 {
670         return mounts_open_common(inode, file, &mounts_op);
671 }
672
673 static const struct file_operations proc_mounts_operations = {
674         .open           = mounts_open,
675         .read           = seq_read,
676         .llseek         = seq_lseek,
677         .release        = mounts_release,
678         .poll           = mounts_poll,
679 };
680
681 static int mountinfo_open(struct inode *inode, struct file *file)
682 {
683         return mounts_open_common(inode, file, &mountinfo_op);
684 }
685
686 static const struct file_operations proc_mountinfo_operations = {
687         .open           = mountinfo_open,
688         .read           = seq_read,
689         .llseek         = seq_lseek,
690         .release        = mounts_release,
691         .poll           = mounts_poll,
692 };
693
694 static int mountstats_open(struct inode *inode, struct file *file)
695 {
696         return mounts_open_common(inode, file, &mountstats_op);
697 }
698
699 static const struct file_operations proc_mountstats_operations = {
700         .open           = mountstats_open,
701         .read           = seq_read,
702         .llseek         = seq_lseek,
703         .release        = mounts_release,
704 };
705
706 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
707
708 static ssize_t proc_info_read(struct file * file, char __user * buf,
709                           size_t count, loff_t *ppos)
710 {
711         struct inode * inode = file->f_path.dentry->d_inode;
712         unsigned long page;
713         ssize_t length;
714         struct task_struct *task = get_proc_task(inode);
715
716         length = -ESRCH;
717         if (!task)
718                 goto out_no_task;
719
720         if (count > PROC_BLOCK_SIZE)
721                 count = PROC_BLOCK_SIZE;
722
723         length = -ENOMEM;
724         if (!(page = __get_free_page(GFP_TEMPORARY)))
725                 goto out;
726
727         length = PROC_I(inode)->op.proc_read(task, (char*)page);
728
729         if (length >= 0)
730                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
731         free_page(page);
732 out:
733         put_task_struct(task);
734 out_no_task:
735         return length;
736 }
737
738 static const struct file_operations proc_info_file_operations = {
739         .read           = proc_info_read,
740 };
741
742 static int proc_single_show(struct seq_file *m, void *v)
743 {
744         struct inode *inode = m->private;
745         struct pid_namespace *ns;
746         struct pid *pid;
747         struct task_struct *task;
748         int ret;
749
750         ns = inode->i_sb->s_fs_info;
751         pid = proc_pid(inode);
752         task = get_pid_task(pid, PIDTYPE_PID);
753         if (!task)
754                 return -ESRCH;
755
756         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
757
758         put_task_struct(task);
759         return ret;
760 }
761
762 static int proc_single_open(struct inode *inode, struct file *filp)
763 {
764         int ret;
765         ret = single_open(filp, proc_single_show, NULL);
766         if (!ret) {
767                 struct seq_file *m = filp->private_data;
768
769                 m->private = inode;
770         }
771         return ret;
772 }
773
774 static const struct file_operations proc_single_file_operations = {
775         .open           = proc_single_open,
776         .read           = seq_read,
777         .llseek         = seq_lseek,
778         .release        = single_release,
779 };
780
781 static int mem_open(struct inode* inode, struct file* file)
782 {
783         file->private_data = (void*)((long)current->self_exec_id);
784         return 0;
785 }
786
787 static ssize_t mem_read(struct file * file, char __user * buf,
788                         size_t count, loff_t *ppos)
789 {
790         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
791         char *page;
792         unsigned long src = *ppos;
793         int ret = -ESRCH;
794         struct mm_struct *mm;
795
796         if (!task)
797                 goto out_no_task;
798
799         if (check_mem_permission(task))
800                 goto out;
801
802         ret = -ENOMEM;
803         page = (char *)__get_free_page(GFP_TEMPORARY);
804         if (!page)
805                 goto out;
806
807         ret = 0;
808  
809         mm = get_task_mm(task);
810         if (!mm)
811                 goto out_free;
812
813         ret = -EIO;
814  
815         if (file->private_data != (void*)((long)current->self_exec_id))
816                 goto out_put;
817
818         ret = 0;
819  
820         while (count > 0) {
821                 int this_len, retval;
822
823                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
824                 retval = access_process_vm(task, src, page, this_len, 0);
825                 if (!retval || check_mem_permission(task)) {
826                         if (!ret)
827                                 ret = -EIO;
828                         break;
829                 }
830
831                 if (copy_to_user(buf, page, retval)) {
832                         ret = -EFAULT;
833                         break;
834                 }
835  
836                 ret += retval;
837                 src += retval;
838                 buf += retval;
839                 count -= retval;
840         }
841         *ppos = src;
842
843 out_put:
844         mmput(mm);
845 out_free:
846         free_page((unsigned long) page);
847 out:
848         put_task_struct(task);
849 out_no_task:
850         return ret;
851 }
852
853 #define mem_write NULL
854
855 #ifndef mem_write
856 /* This is a security hazard */
857 static ssize_t mem_write(struct file * file, const char __user *buf,
858                          size_t count, loff_t *ppos)
859 {
860         int copied;
861         char *page;
862         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
863         unsigned long dst = *ppos;
864
865         copied = -ESRCH;
866         if (!task)
867                 goto out_no_task;
868
869         if (check_mem_permission(task))
870                 goto out;
871
872         copied = -ENOMEM;
873         page = (char *)__get_free_page(GFP_TEMPORARY);
874         if (!page)
875                 goto out;
876
877         copied = 0;
878         while (count > 0) {
879                 int this_len, retval;
880
881                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
882                 if (copy_from_user(page, buf, this_len)) {
883                         copied = -EFAULT;
884                         break;
885                 }
886                 retval = access_process_vm(task, dst, page, this_len, 1);
887                 if (!retval) {
888                         if (!copied)
889                                 copied = -EIO;
890                         break;
891                 }
892                 copied += retval;
893                 buf += retval;
894                 dst += retval;
895                 count -= retval;                        
896         }
897         *ppos = dst;
898         free_page((unsigned long) page);
899 out:
900         put_task_struct(task);
901 out_no_task:
902         return copied;
903 }
904 #endif
905
906 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
907 {
908         switch (orig) {
909         case 0:
910                 file->f_pos = offset;
911                 break;
912         case 1:
913                 file->f_pos += offset;
914                 break;
915         default:
916                 return -EINVAL;
917         }
918         force_successful_syscall_return();
919         return file->f_pos;
920 }
921
922 static const struct file_operations proc_mem_operations = {
923         .llseek         = mem_lseek,
924         .read           = mem_read,
925         .write          = mem_write,
926         .open           = mem_open,
927 };
928
929 static ssize_t environ_read(struct file *file, char __user *buf,
930                         size_t count, loff_t *ppos)
931 {
932         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
933         char *page;
934         unsigned long src = *ppos;
935         int ret = -ESRCH;
936         struct mm_struct *mm;
937
938         if (!task)
939                 goto out_no_task;
940
941         if (!ptrace_may_access(task, PTRACE_MODE_READ))
942                 goto out;
943
944         ret = -ENOMEM;
945         page = (char *)__get_free_page(GFP_TEMPORARY);
946         if (!page)
947                 goto out;
948
949         ret = 0;
950
951         mm = get_task_mm(task);
952         if (!mm)
953                 goto out_free;
954
955         while (count > 0) {
956                 int this_len, retval, max_len;
957
958                 this_len = mm->env_end - (mm->env_start + src);
959
960                 if (this_len <= 0)
961                         break;
962
963                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
964                 this_len = (this_len > max_len) ? max_len : this_len;
965
966                 retval = access_process_vm(task, (mm->env_start + src),
967                         page, this_len, 0);
968
969                 if (retval <= 0) {
970                         ret = retval;
971                         break;
972                 }
973
974                 if (copy_to_user(buf, page, retval)) {
975                         ret = -EFAULT;
976                         break;
977                 }
978
979                 ret += retval;
980                 src += retval;
981                 buf += retval;
982                 count -= retval;
983         }
984         *ppos = src;
985
986         mmput(mm);
987 out_free:
988         free_page((unsigned long) page);
989 out:
990         put_task_struct(task);
991 out_no_task:
992         return ret;
993 }
994
995 static const struct file_operations proc_environ_operations = {
996         .read           = environ_read,
997 };
998
999 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1000                                 size_t count, loff_t *ppos)
1001 {
1002         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1003         char buffer[PROC_NUMBUF];
1004         size_t len;
1005         int oom_adjust;
1006
1007         if (!task)
1008                 return -ESRCH;
1009         oom_adjust = task->oomkilladj;
1010         put_task_struct(task);
1011
1012         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1013
1014         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1015 }
1016
1017 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1018                                 size_t count, loff_t *ppos)
1019 {
1020         struct task_struct *task;
1021         char buffer[PROC_NUMBUF], *end;
1022         int oom_adjust;
1023
1024         memset(buffer, 0, sizeof(buffer));
1025         if (count > sizeof(buffer) - 1)
1026                 count = sizeof(buffer) - 1;
1027         if (copy_from_user(buffer, buf, count))
1028                 return -EFAULT;
1029         oom_adjust = simple_strtol(buffer, &end, 0);
1030         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1031              oom_adjust != OOM_DISABLE)
1032                 return -EINVAL;
1033         if (*end == '\n')
1034                 end++;
1035         task = get_proc_task(file->f_path.dentry->d_inode);
1036         if (!task)
1037                 return -ESRCH;
1038         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1039                 put_task_struct(task);
1040                 return -EACCES;
1041         }
1042         task->oomkilladj = oom_adjust;
1043         put_task_struct(task);
1044         if (end - buffer == 0)
1045                 return -EIO;
1046         return end - buffer;
1047 }
1048
1049 static const struct file_operations proc_oom_adjust_operations = {
1050         .read           = oom_adjust_read,
1051         .write          = oom_adjust_write,
1052 };
1053
1054 #ifdef CONFIG_AUDITSYSCALL
1055 #define TMPBUFLEN 21
1056 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1057                                   size_t count, loff_t *ppos)
1058 {
1059         struct inode * inode = file->f_path.dentry->d_inode;
1060         struct task_struct *task = get_proc_task(inode);
1061         ssize_t length;
1062         char tmpbuf[TMPBUFLEN];
1063
1064         if (!task)
1065                 return -ESRCH;
1066         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1067                                 audit_get_loginuid(task));
1068         put_task_struct(task);
1069         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1070 }
1071
1072 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1073                                    size_t count, loff_t *ppos)
1074 {
1075         struct inode * inode = file->f_path.dentry->d_inode;
1076         char *page, *tmp;
1077         ssize_t length;
1078         uid_t loginuid;
1079
1080         if (!capable(CAP_AUDIT_CONTROL))
1081                 return -EPERM;
1082
1083         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1084                 return -EPERM;
1085
1086         if (count >= PAGE_SIZE)
1087                 count = PAGE_SIZE - 1;
1088
1089         if (*ppos != 0) {
1090                 /* No partial writes. */
1091                 return -EINVAL;
1092         }
1093         page = (char*)__get_free_page(GFP_TEMPORARY);
1094         if (!page)
1095                 return -ENOMEM;
1096         length = -EFAULT;
1097         if (copy_from_user(page, buf, count))
1098                 goto out_free_page;
1099
1100         page[count] = '\0';
1101         loginuid = simple_strtoul(page, &tmp, 10);
1102         if (tmp == page) {
1103                 length = -EINVAL;
1104                 goto out_free_page;
1105
1106         }
1107         length = audit_set_loginuid(current, loginuid);
1108         if (likely(length == 0))
1109                 length = count;
1110
1111 out_free_page:
1112         free_page((unsigned long) page);
1113         return length;
1114 }
1115
1116 static const struct file_operations proc_loginuid_operations = {
1117         .read           = proc_loginuid_read,
1118         .write          = proc_loginuid_write,
1119 };
1120
1121 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1122                                   size_t count, loff_t *ppos)
1123 {
1124         struct inode * inode = file->f_path.dentry->d_inode;
1125         struct task_struct *task = get_proc_task(inode);
1126         ssize_t length;
1127         char tmpbuf[TMPBUFLEN];
1128
1129         if (!task)
1130                 return -ESRCH;
1131         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1132                                 audit_get_sessionid(task));
1133         put_task_struct(task);
1134         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1135 }
1136
1137 static const struct file_operations proc_sessionid_operations = {
1138         .read           = proc_sessionid_read,
1139 };
1140 #endif
1141
1142 #ifdef CONFIG_FAULT_INJECTION
1143 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1144                                       size_t count, loff_t *ppos)
1145 {
1146         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1147         char buffer[PROC_NUMBUF];
1148         size_t len;
1149         int make_it_fail;
1150
1151         if (!task)
1152                 return -ESRCH;
1153         make_it_fail = task->make_it_fail;
1154         put_task_struct(task);
1155
1156         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1157
1158         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1159 }
1160
1161 static ssize_t proc_fault_inject_write(struct file * file,
1162                         const char __user * buf, size_t count, loff_t *ppos)
1163 {
1164         struct task_struct *task;
1165         char buffer[PROC_NUMBUF], *end;
1166         int make_it_fail;
1167
1168         if (!capable(CAP_SYS_RESOURCE))
1169                 return -EPERM;
1170         memset(buffer, 0, sizeof(buffer));
1171         if (count > sizeof(buffer) - 1)
1172                 count = sizeof(buffer) - 1;
1173         if (copy_from_user(buffer, buf, count))
1174                 return -EFAULT;
1175         make_it_fail = simple_strtol(buffer, &end, 0);
1176         if (*end == '\n')
1177                 end++;
1178         task = get_proc_task(file->f_dentry->d_inode);
1179         if (!task)
1180                 return -ESRCH;
1181         task->make_it_fail = make_it_fail;
1182         put_task_struct(task);
1183         if (end - buffer == 0)
1184                 return -EIO;
1185         return end - buffer;
1186 }
1187
1188 static const struct file_operations proc_fault_inject_operations = {
1189         .read           = proc_fault_inject_read,
1190         .write          = proc_fault_inject_write,
1191 };
1192 #endif
1193
1194
1195 #ifdef CONFIG_SCHED_DEBUG
1196 /*
1197  * Print out various scheduling related per-task fields:
1198  */
1199 static int sched_show(struct seq_file *m, void *v)
1200 {
1201         struct inode *inode = m->private;
1202         struct task_struct *p;
1203
1204         p = get_proc_task(inode);
1205         if (!p)
1206                 return -ESRCH;
1207         proc_sched_show_task(p, m);
1208
1209         put_task_struct(p);
1210
1211         return 0;
1212 }
1213
1214 static ssize_t
1215 sched_write(struct file *file, const char __user *buf,
1216             size_t count, loff_t *offset)
1217 {
1218         struct inode *inode = file->f_path.dentry->d_inode;
1219         struct task_struct *p;
1220
1221         p = get_proc_task(inode);
1222         if (!p)
1223                 return -ESRCH;
1224         proc_sched_set_task(p);
1225
1226         put_task_struct(p);
1227
1228         return count;
1229 }
1230
1231 static int sched_open(struct inode *inode, struct file *filp)
1232 {
1233         int ret;
1234
1235         ret = single_open(filp, sched_show, NULL);
1236         if (!ret) {
1237                 struct seq_file *m = filp->private_data;
1238
1239                 m->private = inode;
1240         }
1241         return ret;
1242 }
1243
1244 static const struct file_operations proc_pid_sched_operations = {
1245         .open           = sched_open,
1246         .read           = seq_read,
1247         .write          = sched_write,
1248         .llseek         = seq_lseek,
1249         .release        = single_release,
1250 };
1251
1252 #endif
1253
1254 /*
1255  * We added or removed a vma mapping the executable. The vmas are only mapped
1256  * during exec and are not mapped with the mmap system call.
1257  * Callers must hold down_write() on the mm's mmap_sem for these
1258  */
1259 void added_exe_file_vma(struct mm_struct *mm)
1260 {
1261         mm->num_exe_file_vmas++;
1262 }
1263
1264 void removed_exe_file_vma(struct mm_struct *mm)
1265 {
1266         mm->num_exe_file_vmas--;
1267         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1268                 fput(mm->exe_file);
1269                 mm->exe_file = NULL;
1270         }
1271
1272 }
1273
1274 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1275 {
1276         if (new_exe_file)
1277                 get_file(new_exe_file);
1278         if (mm->exe_file)
1279                 fput(mm->exe_file);
1280         mm->exe_file = new_exe_file;
1281         mm->num_exe_file_vmas = 0;
1282 }
1283
1284 struct file *get_mm_exe_file(struct mm_struct *mm)
1285 {
1286         struct file *exe_file;
1287
1288         /* We need mmap_sem to protect against races with removal of
1289          * VM_EXECUTABLE vmas */
1290         down_read(&mm->mmap_sem);
1291         exe_file = mm->exe_file;
1292         if (exe_file)
1293                 get_file(exe_file);
1294         up_read(&mm->mmap_sem);
1295         return exe_file;
1296 }
1297
1298 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1299 {
1300         /* It's safe to write the exe_file pointer without exe_file_lock because
1301          * this is called during fork when the task is not yet in /proc */
1302         newmm->exe_file = get_mm_exe_file(oldmm);
1303 }
1304
1305 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1306 {
1307         struct task_struct *task;
1308         struct mm_struct *mm;
1309         struct file *exe_file;
1310
1311         task = get_proc_task(inode);
1312         if (!task)
1313                 return -ENOENT;
1314         mm = get_task_mm(task);
1315         put_task_struct(task);
1316         if (!mm)
1317                 return -ENOENT;
1318         exe_file = get_mm_exe_file(mm);
1319         mmput(mm);
1320         if (exe_file) {
1321                 *exe_path = exe_file->f_path;
1322                 path_get(&exe_file->f_path);
1323                 fput(exe_file);
1324                 return 0;
1325         } else
1326                 return -ENOENT;
1327 }
1328
1329 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1330 {
1331         struct inode *inode = dentry->d_inode;
1332         int error = -EACCES;
1333
1334         /* We don't need a base pointer in the /proc filesystem */
1335         path_put(&nd->path);
1336
1337         /* Are we allowed to snoop on the tasks file descriptors? */
1338         if (!proc_fd_access_allowed(inode))
1339                 goto out;
1340
1341         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1342         nd->last_type = LAST_BIND;
1343 out:
1344         return ERR_PTR(error);
1345 }
1346
1347 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1348 {
1349         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1350         char *pathname;
1351         int len;
1352
1353         if (!tmp)
1354                 return -ENOMEM;
1355
1356         pathname = d_path(path, tmp, PAGE_SIZE);
1357         len = PTR_ERR(pathname);
1358         if (IS_ERR(pathname))
1359                 goto out;
1360         len = tmp + PAGE_SIZE - 1 - pathname;
1361
1362         if (len > buflen)
1363                 len = buflen;
1364         if (copy_to_user(buffer, pathname, len))
1365                 len = -EFAULT;
1366  out:
1367         free_page((unsigned long)tmp);
1368         return len;
1369 }
1370
1371 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1372 {
1373         int error = -EACCES;
1374         struct inode *inode = dentry->d_inode;
1375         struct path path;
1376
1377         /* Are we allowed to snoop on the tasks file descriptors? */
1378         if (!proc_fd_access_allowed(inode))
1379                 goto out;
1380
1381         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1382         if (error)
1383                 goto out;
1384
1385         error = do_proc_readlink(&path, buffer, buflen);
1386         path_put(&path);
1387 out:
1388         return error;
1389 }
1390
1391 static const struct inode_operations proc_pid_link_inode_operations = {
1392         .readlink       = proc_pid_readlink,
1393         .follow_link    = proc_pid_follow_link,
1394         .setattr        = proc_setattr,
1395 };
1396
1397
1398 /* building an inode */
1399
1400 static int task_dumpable(struct task_struct *task)
1401 {
1402         int dumpable = 0;
1403         struct mm_struct *mm;
1404
1405         task_lock(task);
1406         mm = task->mm;
1407         if (mm)
1408                 dumpable = get_dumpable(mm);
1409         task_unlock(task);
1410         if(dumpable == 1)
1411                 return 1;
1412         return 0;
1413 }
1414
1415
1416 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1417 {
1418         struct inode * inode;
1419         struct proc_inode *ei;
1420         const struct cred *cred;
1421
1422         /* We need a new inode */
1423
1424         inode = new_inode(sb);
1425         if (!inode)
1426                 goto out;
1427
1428         /* Common stuff */
1429         ei = PROC_I(inode);
1430         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1431         inode->i_op = &proc_def_inode_operations;
1432
1433         /*
1434          * grab the reference to task.
1435          */
1436         ei->pid = get_task_pid(task, PIDTYPE_PID);
1437         if (!ei->pid)
1438                 goto out_unlock;
1439
1440         if (task_dumpable(task)) {
1441                 rcu_read_lock();
1442                 cred = __task_cred(task);
1443                 inode->i_uid = cred->euid;
1444                 inode->i_gid = cred->egid;
1445                 rcu_read_unlock();
1446         }
1447         security_task_to_inode(task, inode);
1448
1449 out:
1450         return inode;
1451
1452 out_unlock:
1453         iput(inode);
1454         return NULL;
1455 }
1456
1457 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1458 {
1459         struct inode *inode = dentry->d_inode;
1460         struct task_struct *task;
1461         const struct cred *cred;
1462
1463         generic_fillattr(inode, stat);
1464
1465         rcu_read_lock();
1466         stat->uid = 0;
1467         stat->gid = 0;
1468         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1469         if (task) {
1470                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1471                     task_dumpable(task)) {
1472                         cred = __task_cred(task);
1473                         stat->uid = cred->euid;
1474                         stat->gid = cred->egid;
1475                 }
1476         }
1477         rcu_read_unlock();
1478         return 0;
1479 }
1480
1481 /* dentry stuff */
1482
1483 /*
1484  *      Exceptional case: normally we are not allowed to unhash a busy
1485  * directory. In this case, however, we can do it - no aliasing problems
1486  * due to the way we treat inodes.
1487  *
1488  * Rewrite the inode's ownerships here because the owning task may have
1489  * performed a setuid(), etc.
1490  *
1491  * Before the /proc/pid/status file was created the only way to read
1492  * the effective uid of a /process was to stat /proc/pid.  Reading
1493  * /proc/pid/status is slow enough that procps and other packages
1494  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1495  * made this apply to all per process world readable and executable
1496  * directories.
1497  */
1498 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1499 {
1500         struct inode *inode = dentry->d_inode;
1501         struct task_struct *task = get_proc_task(inode);
1502         const struct cred *cred;
1503
1504         if (task) {
1505                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1506                     task_dumpable(task)) {
1507                         rcu_read_lock();
1508                         cred = __task_cred(task);
1509                         inode->i_uid = cred->euid;
1510                         inode->i_gid = cred->egid;
1511                         rcu_read_unlock();
1512                 } else {
1513                         inode->i_uid = 0;
1514                         inode->i_gid = 0;
1515                 }
1516                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1517                 security_task_to_inode(task, inode);
1518                 put_task_struct(task);
1519                 return 1;
1520         }
1521         d_drop(dentry);
1522         return 0;
1523 }
1524
1525 static int pid_delete_dentry(struct dentry * dentry)
1526 {
1527         /* Is the task we represent dead?
1528          * If so, then don't put the dentry on the lru list,
1529          * kill it immediately.
1530          */
1531         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1532 }
1533
1534 static const struct dentry_operations pid_dentry_operations =
1535 {
1536         .d_revalidate   = pid_revalidate,
1537         .d_delete       = pid_delete_dentry,
1538 };
1539
1540 /* Lookups */
1541
1542 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1543                                 struct task_struct *, const void *);
1544
1545 /*
1546  * Fill a directory entry.
1547  *
1548  * If possible create the dcache entry and derive our inode number and
1549  * file type from dcache entry.
1550  *
1551  * Since all of the proc inode numbers are dynamically generated, the inode
1552  * numbers do not exist until the inode is cache.  This means creating the
1553  * the dcache entry in readdir is necessary to keep the inode numbers
1554  * reported by readdir in sync with the inode numbers reported
1555  * by stat.
1556  */
1557 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1558         char *name, int len,
1559         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1560 {
1561         struct dentry *child, *dir = filp->f_path.dentry;
1562         struct inode *inode;
1563         struct qstr qname;
1564         ino_t ino = 0;
1565         unsigned type = DT_UNKNOWN;
1566
1567         qname.name = name;
1568         qname.len  = len;
1569         qname.hash = full_name_hash(name, len);
1570
1571         child = d_lookup(dir, &qname);
1572         if (!child) {
1573                 struct dentry *new;
1574                 new = d_alloc(dir, &qname);
1575                 if (new) {
1576                         child = instantiate(dir->d_inode, new, task, ptr);
1577                         if (child)
1578                                 dput(new);
1579                         else
1580                                 child = new;
1581                 }
1582         }
1583         if (!child || IS_ERR(child) || !child->d_inode)
1584                 goto end_instantiate;
1585         inode = child->d_inode;
1586         if (inode) {
1587                 ino = inode->i_ino;
1588                 type = inode->i_mode >> 12;
1589         }
1590         dput(child);
1591 end_instantiate:
1592         if (!ino)
1593                 ino = find_inode_number(dir, &qname);
1594         if (!ino)
1595                 ino = 1;
1596         return filldir(dirent, name, len, filp->f_pos, ino, type);
1597 }
1598
1599 static unsigned name_to_int(struct dentry *dentry)
1600 {
1601         const char *name = dentry->d_name.name;
1602         int len = dentry->d_name.len;
1603         unsigned n = 0;
1604
1605         if (len > 1 && *name == '0')
1606                 goto out;
1607         while (len-- > 0) {
1608                 unsigned c = *name++ - '0';
1609                 if (c > 9)
1610                         goto out;
1611                 if (n >= (~0U-9)/10)
1612                         goto out;
1613                 n *= 10;
1614                 n += c;
1615         }
1616         return n;
1617 out:
1618         return ~0U;
1619 }
1620
1621 #define PROC_FDINFO_MAX 64
1622
1623 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1624 {
1625         struct task_struct *task = get_proc_task(inode);
1626         struct files_struct *files = NULL;
1627         struct file *file;
1628         int fd = proc_fd(inode);
1629
1630         if (task) {
1631                 files = get_files_struct(task);
1632                 put_task_struct(task);
1633         }
1634         if (files) {
1635                 /*
1636                  * We are not taking a ref to the file structure, so we must
1637                  * hold ->file_lock.
1638                  */
1639                 spin_lock(&files->file_lock);
1640                 file = fcheck_files(files, fd);
1641                 if (file) {
1642                         if (path) {
1643                                 *path = file->f_path;
1644                                 path_get(&file->f_path);
1645                         }
1646                         if (info)
1647                                 snprintf(info, PROC_FDINFO_MAX,
1648                                          "pos:\t%lli\n"
1649                                          "flags:\t0%o\n",
1650                                          (long long) file->f_pos,
1651                                          file->f_flags);
1652                         spin_unlock(&files->file_lock);
1653                         put_files_struct(files);
1654                         return 0;
1655                 }
1656                 spin_unlock(&files->file_lock);
1657                 put_files_struct(files);
1658         }
1659         return -ENOENT;
1660 }
1661
1662 static int proc_fd_link(struct inode *inode, struct path *path)
1663 {
1664         return proc_fd_info(inode, path, NULL);
1665 }
1666
1667 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1668 {
1669         struct inode *inode = dentry->d_inode;
1670         struct task_struct *task = get_proc_task(inode);
1671         int fd = proc_fd(inode);
1672         struct files_struct *files;
1673         const struct cred *cred;
1674
1675         if (task) {
1676                 files = get_files_struct(task);
1677                 if (files) {
1678                         rcu_read_lock();
1679                         if (fcheck_files(files, fd)) {
1680                                 rcu_read_unlock();
1681                                 put_files_struct(files);
1682                                 if (task_dumpable(task)) {
1683                                         rcu_read_lock();
1684                                         cred = __task_cred(task);
1685                                         inode->i_uid = cred->euid;
1686                                         inode->i_gid = cred->egid;
1687                                         rcu_read_unlock();
1688                                 } else {
1689                                         inode->i_uid = 0;
1690                                         inode->i_gid = 0;
1691                                 }
1692                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1693                                 security_task_to_inode(task, inode);
1694                                 put_task_struct(task);
1695                                 return 1;
1696                         }
1697                         rcu_read_unlock();
1698                         put_files_struct(files);
1699                 }
1700                 put_task_struct(task);
1701         }
1702         d_drop(dentry);
1703         return 0;
1704 }
1705
1706 static const struct dentry_operations tid_fd_dentry_operations =
1707 {
1708         .d_revalidate   = tid_fd_revalidate,
1709         .d_delete       = pid_delete_dentry,
1710 };
1711
1712 static struct dentry *proc_fd_instantiate(struct inode *dir,
1713         struct dentry *dentry, struct task_struct *task, const void *ptr)
1714 {
1715         unsigned fd = *(const unsigned *)ptr;
1716         struct file *file;
1717         struct files_struct *files;
1718         struct inode *inode;
1719         struct proc_inode *ei;
1720         struct dentry *error = ERR_PTR(-ENOENT);
1721
1722         inode = proc_pid_make_inode(dir->i_sb, task);
1723         if (!inode)
1724                 goto out;
1725         ei = PROC_I(inode);
1726         ei->fd = fd;
1727         files = get_files_struct(task);
1728         if (!files)
1729                 goto out_iput;
1730         inode->i_mode = S_IFLNK;
1731
1732         /*
1733          * We are not taking a ref to the file structure, so we must
1734          * hold ->file_lock.
1735          */
1736         spin_lock(&files->file_lock);
1737         file = fcheck_files(files, fd);
1738         if (!file)
1739                 goto out_unlock;
1740         if (file->f_mode & FMODE_READ)
1741                 inode->i_mode |= S_IRUSR | S_IXUSR;
1742         if (file->f_mode & FMODE_WRITE)
1743                 inode->i_mode |= S_IWUSR | S_IXUSR;
1744         spin_unlock(&files->file_lock);
1745         put_files_struct(files);
1746
1747         inode->i_op = &proc_pid_link_inode_operations;
1748         inode->i_size = 64;
1749         ei->op.proc_get_link = proc_fd_link;
1750         dentry->d_op = &tid_fd_dentry_operations;
1751         d_add(dentry, inode);
1752         /* Close the race of the process dying before we return the dentry */
1753         if (tid_fd_revalidate(dentry, NULL))
1754                 error = NULL;
1755
1756  out:
1757         return error;
1758 out_unlock:
1759         spin_unlock(&files->file_lock);
1760         put_files_struct(files);
1761 out_iput:
1762         iput(inode);
1763         goto out;
1764 }
1765
1766 static struct dentry *proc_lookupfd_common(struct inode *dir,
1767                                            struct dentry *dentry,
1768                                            instantiate_t instantiate)
1769 {
1770         struct task_struct *task = get_proc_task(dir);
1771         unsigned fd = name_to_int(dentry);
1772         struct dentry *result = ERR_PTR(-ENOENT);
1773
1774         if (!task)
1775                 goto out_no_task;
1776         if (fd == ~0U)
1777                 goto out;
1778
1779         result = instantiate(dir, dentry, task, &fd);
1780 out:
1781         put_task_struct(task);
1782 out_no_task:
1783         return result;
1784 }
1785
1786 static int proc_readfd_common(struct file * filp, void * dirent,
1787                               filldir_t filldir, instantiate_t instantiate)
1788 {
1789         struct dentry *dentry = filp->f_path.dentry;
1790         struct inode *inode = dentry->d_inode;
1791         struct task_struct *p = get_proc_task(inode);
1792         unsigned int fd, ino;
1793         int retval;
1794         struct files_struct * files;
1795
1796         retval = -ENOENT;
1797         if (!p)
1798                 goto out_no_task;
1799         retval = 0;
1800
1801         fd = filp->f_pos;
1802         switch (fd) {
1803                 case 0:
1804                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1805                                 goto out;
1806                         filp->f_pos++;
1807                 case 1:
1808                         ino = parent_ino(dentry);
1809                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1810                                 goto out;
1811                         filp->f_pos++;
1812                 default:
1813                         files = get_files_struct(p);
1814                         if (!files)
1815                                 goto out;
1816                         rcu_read_lock();
1817                         for (fd = filp->f_pos-2;
1818                              fd < files_fdtable(files)->max_fds;
1819                              fd++, filp->f_pos++) {
1820                                 char name[PROC_NUMBUF];
1821                                 int len;
1822
1823                                 if (!fcheck_files(files, fd))
1824                                         continue;
1825                                 rcu_read_unlock();
1826
1827                                 len = snprintf(name, sizeof(name), "%d", fd);
1828                                 if (proc_fill_cache(filp, dirent, filldir,
1829                                                     name, len, instantiate,
1830                                                     p, &fd) < 0) {
1831                                         rcu_read_lock();
1832                                         break;
1833                                 }
1834                                 rcu_read_lock();
1835                         }
1836                         rcu_read_unlock();
1837                         put_files_struct(files);
1838         }
1839 out:
1840         put_task_struct(p);
1841 out_no_task:
1842         return retval;
1843 }
1844
1845 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1846                                     struct nameidata *nd)
1847 {
1848         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1849 }
1850
1851 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1852 {
1853         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1854 }
1855
1856 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1857                                       size_t len, loff_t *ppos)
1858 {
1859         char tmp[PROC_FDINFO_MAX];
1860         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1861         if (!err)
1862                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1863         return err;
1864 }
1865
1866 static const struct file_operations proc_fdinfo_file_operations = {
1867         .open           = nonseekable_open,
1868         .read           = proc_fdinfo_read,
1869 };
1870
1871 static const struct file_operations proc_fd_operations = {
1872         .read           = generic_read_dir,
1873         .readdir        = proc_readfd,
1874 };
1875
1876 /*
1877  * /proc/pid/fd needs a special permission handler so that a process can still
1878  * access /proc/self/fd after it has executed a setuid().
1879  */
1880 static int proc_fd_permission(struct inode *inode, int mask)
1881 {
1882         int rv;
1883
1884         rv = generic_permission(inode, mask, NULL);
1885         if (rv == 0)
1886                 return 0;
1887         if (task_pid(current) == proc_pid(inode))
1888                 rv = 0;
1889         return rv;
1890 }
1891
1892 /*
1893  * proc directories can do almost nothing..
1894  */
1895 static const struct inode_operations proc_fd_inode_operations = {
1896         .lookup         = proc_lookupfd,
1897         .permission     = proc_fd_permission,
1898         .setattr        = proc_setattr,
1899 };
1900
1901 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1902         struct dentry *dentry, struct task_struct *task, const void *ptr)
1903 {
1904         unsigned fd = *(unsigned *)ptr;
1905         struct inode *inode;
1906         struct proc_inode *ei;
1907         struct dentry *error = ERR_PTR(-ENOENT);
1908
1909         inode = proc_pid_make_inode(dir->i_sb, task);
1910         if (!inode)
1911                 goto out;
1912         ei = PROC_I(inode);
1913         ei->fd = fd;
1914         inode->i_mode = S_IFREG | S_IRUSR;
1915         inode->i_fop = &proc_fdinfo_file_operations;
1916         dentry->d_op = &tid_fd_dentry_operations;
1917         d_add(dentry, inode);
1918         /* Close the race of the process dying before we return the dentry */
1919         if (tid_fd_revalidate(dentry, NULL))
1920                 error = NULL;
1921
1922  out:
1923         return error;
1924 }
1925
1926 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1927                                         struct dentry *dentry,
1928                                         struct nameidata *nd)
1929 {
1930         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1931 }
1932
1933 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1934 {
1935         return proc_readfd_common(filp, dirent, filldir,
1936                                   proc_fdinfo_instantiate);
1937 }
1938
1939 static const struct file_operations proc_fdinfo_operations = {
1940         .read           = generic_read_dir,
1941         .readdir        = proc_readfdinfo,
1942 };
1943
1944 /*
1945  * proc directories can do almost nothing..
1946  */
1947 static const struct inode_operations proc_fdinfo_inode_operations = {
1948         .lookup         = proc_lookupfdinfo,
1949         .setattr        = proc_setattr,
1950 };
1951
1952
1953 static struct dentry *proc_pident_instantiate(struct inode *dir,
1954         struct dentry *dentry, struct task_struct *task, const void *ptr)
1955 {
1956         const struct pid_entry *p = ptr;
1957         struct inode *inode;
1958         struct proc_inode *ei;
1959         struct dentry *error = ERR_PTR(-ENOENT);
1960
1961         inode = proc_pid_make_inode(dir->i_sb, task);
1962         if (!inode)
1963                 goto out;
1964
1965         ei = PROC_I(inode);
1966         inode->i_mode = p->mode;
1967         if (S_ISDIR(inode->i_mode))
1968                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1969         if (p->iop)
1970                 inode->i_op = p->iop;
1971         if (p->fop)
1972                 inode->i_fop = p->fop;
1973         ei->op = p->op;
1974         dentry->d_op = &pid_dentry_operations;
1975         d_add(dentry, inode);
1976         /* Close the race of the process dying before we return the dentry */
1977         if (pid_revalidate(dentry, NULL))
1978                 error = NULL;
1979 out:
1980         return error;
1981 }
1982
1983 static struct dentry *proc_pident_lookup(struct inode *dir, 
1984                                          struct dentry *dentry,
1985                                          const struct pid_entry *ents,
1986                                          unsigned int nents)
1987 {
1988         struct dentry *error;
1989         struct task_struct *task = get_proc_task(dir);
1990         const struct pid_entry *p, *last;
1991
1992         error = ERR_PTR(-ENOENT);
1993
1994         if (!task)
1995                 goto out_no_task;
1996
1997         /*
1998          * Yes, it does not scale. And it should not. Don't add
1999          * new entries into /proc/<tgid>/ without very good reasons.
2000          */
2001         last = &ents[nents - 1];
2002         for (p = ents; p <= last; p++) {
2003                 if (p->len != dentry->d_name.len)
2004                         continue;
2005                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2006                         break;
2007         }
2008         if (p > last)
2009                 goto out;
2010
2011         error = proc_pident_instantiate(dir, dentry, task, p);
2012 out:
2013         put_task_struct(task);
2014 out_no_task:
2015         return error;
2016 }
2017
2018 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2019         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2020 {
2021         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2022                                 proc_pident_instantiate, task, p);
2023 }
2024
2025 static int proc_pident_readdir(struct file *filp,
2026                 void *dirent, filldir_t filldir,
2027                 const struct pid_entry *ents, unsigned int nents)
2028 {
2029         int i;
2030         struct dentry *dentry = filp->f_path.dentry;
2031         struct inode *inode = dentry->d_inode;
2032         struct task_struct *task = get_proc_task(inode);
2033         const struct pid_entry *p, *last;
2034         ino_t ino;
2035         int ret;
2036
2037         ret = -ENOENT;
2038         if (!task)
2039                 goto out_no_task;
2040
2041         ret = 0;
2042         i = filp->f_pos;
2043         switch (i) {
2044         case 0:
2045                 ino = inode->i_ino;
2046                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2047                         goto out;
2048                 i++;
2049                 filp->f_pos++;
2050                 /* fall through */
2051         case 1:
2052                 ino = parent_ino(dentry);
2053                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2054                         goto out;
2055                 i++;
2056                 filp->f_pos++;
2057                 /* fall through */
2058         default:
2059                 i -= 2;
2060                 if (i >= nents) {
2061                         ret = 1;
2062                         goto out;
2063                 }
2064                 p = ents + i;
2065                 last = &ents[nents - 1];
2066                 while (p <= last) {
2067                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2068                                 goto out;
2069                         filp->f_pos++;
2070                         p++;
2071                 }
2072         }
2073
2074         ret = 1;
2075 out:
2076         put_task_struct(task);
2077 out_no_task:
2078         return ret;
2079 }
2080
2081 #ifdef CONFIG_SECURITY
2082 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2083                                   size_t count, loff_t *ppos)
2084 {
2085         struct inode * inode = file->f_path.dentry->d_inode;
2086         char *p = NULL;
2087         ssize_t length;
2088         struct task_struct *task = get_proc_task(inode);
2089
2090         if (!task)
2091                 return -ESRCH;
2092
2093         length = security_getprocattr(task,
2094                                       (char*)file->f_path.dentry->d_name.name,
2095                                       &p);
2096         put_task_struct(task);
2097         if (length > 0)
2098                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2099         kfree(p);
2100         return length;
2101 }
2102
2103 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2104                                    size_t count, loff_t *ppos)
2105 {
2106         struct inode * inode = file->f_path.dentry->d_inode;
2107         char *page;
2108         ssize_t length;
2109         struct task_struct *task = get_proc_task(inode);
2110
2111         length = -ESRCH;
2112         if (!task)
2113                 goto out_no_task;
2114         if (count > PAGE_SIZE)
2115                 count = PAGE_SIZE;
2116
2117         /* No partial writes. */
2118         length = -EINVAL;
2119         if (*ppos != 0)
2120                 goto out;
2121
2122         length = -ENOMEM;
2123         page = (char*)__get_free_page(GFP_TEMPORARY);
2124         if (!page)
2125                 goto out;
2126
2127         length = -EFAULT;
2128         if (copy_from_user(page, buf, count))
2129                 goto out_free;
2130
2131         length = security_setprocattr(task,
2132                                       (char*)file->f_path.dentry->d_name.name,
2133                                       (void*)page, count);
2134 out_free:
2135         free_page((unsigned long) page);
2136 out:
2137         put_task_struct(task);
2138 out_no_task:
2139         return length;
2140 }
2141
2142 static const struct file_operations proc_pid_attr_operations = {
2143         .read           = proc_pid_attr_read,
2144         .write          = proc_pid_attr_write,
2145 };
2146
2147 static const struct pid_entry attr_dir_stuff[] = {
2148         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2149         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2150         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2151         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2152         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2153         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2154 };
2155
2156 static int proc_attr_dir_readdir(struct file * filp,
2157                              void * dirent, filldir_t filldir)
2158 {
2159         return proc_pident_readdir(filp,dirent,filldir,
2160                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2161 }
2162
2163 static const struct file_operations proc_attr_dir_operations = {
2164         .read           = generic_read_dir,
2165         .readdir        = proc_attr_dir_readdir,
2166 };
2167
2168 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2169                                 struct dentry *dentry, struct nameidata *nd)
2170 {
2171         return proc_pident_lookup(dir, dentry,
2172                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2173 }
2174
2175 static const struct inode_operations proc_attr_dir_inode_operations = {
2176         .lookup         = proc_attr_dir_lookup,
2177         .getattr        = pid_getattr,
2178         .setattr        = proc_setattr,
2179 };
2180
2181 #endif
2182
2183 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2184 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2185                                          size_t count, loff_t *ppos)
2186 {
2187         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2188         struct mm_struct *mm;
2189         char buffer[PROC_NUMBUF];
2190         size_t len;
2191         int ret;
2192
2193         if (!task)
2194                 return -ESRCH;
2195
2196         ret = 0;
2197         mm = get_task_mm(task);
2198         if (mm) {
2199                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2200                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2201                                 MMF_DUMP_FILTER_SHIFT));
2202                 mmput(mm);
2203                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2204         }
2205
2206         put_task_struct(task);
2207
2208         return ret;
2209 }
2210
2211 static ssize_t proc_coredump_filter_write(struct file *file,
2212                                           const char __user *buf,
2213                                           size_t count,
2214                                           loff_t *ppos)
2215 {
2216         struct task_struct *task;
2217         struct mm_struct *mm;
2218         char buffer[PROC_NUMBUF], *end;
2219         unsigned int val;
2220         int ret;
2221         int i;
2222         unsigned long mask;
2223
2224         ret = -EFAULT;
2225         memset(buffer, 0, sizeof(buffer));
2226         if (count > sizeof(buffer) - 1)
2227                 count = sizeof(buffer) - 1;
2228         if (copy_from_user(buffer, buf, count))
2229                 goto out_no_task;
2230
2231         ret = -EINVAL;
2232         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2233         if (*end == '\n')
2234                 end++;
2235         if (end - buffer == 0)
2236                 goto out_no_task;
2237
2238         ret = -ESRCH;
2239         task = get_proc_task(file->f_dentry->d_inode);
2240         if (!task)
2241                 goto out_no_task;
2242
2243         ret = end - buffer;
2244         mm = get_task_mm(task);
2245         if (!mm)
2246                 goto out_no_mm;
2247
2248         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2249                 if (val & mask)
2250                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2251                 else
2252                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2253         }
2254
2255         mmput(mm);
2256  out_no_mm:
2257         put_task_struct(task);
2258  out_no_task:
2259         return ret;
2260 }
2261
2262 static const struct file_operations proc_coredump_filter_operations = {
2263         .read           = proc_coredump_filter_read,
2264         .write          = proc_coredump_filter_write,
2265 };
2266 #endif
2267
2268 /*
2269  * /proc/self:
2270  */
2271 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2272                               int buflen)
2273 {
2274         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2275         pid_t tgid = task_tgid_nr_ns(current, ns);
2276         char tmp[PROC_NUMBUF];
2277         if (!tgid)
2278                 return -ENOENT;
2279         sprintf(tmp, "%d", tgid);
2280         return vfs_readlink(dentry,buffer,buflen,tmp);
2281 }
2282
2283 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2284 {
2285         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2286         pid_t tgid = task_tgid_nr_ns(current, ns);
2287         char tmp[PROC_NUMBUF];
2288         if (!tgid)
2289                 return ERR_PTR(-ENOENT);
2290         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2291         return ERR_PTR(vfs_follow_link(nd,tmp));
2292 }
2293
2294 static const struct inode_operations proc_self_inode_operations = {
2295         .readlink       = proc_self_readlink,
2296         .follow_link    = proc_self_follow_link,
2297 };
2298
2299 /*
2300  * proc base
2301  *
2302  * These are the directory entries in the root directory of /proc
2303  * that properly belong to the /proc filesystem, as they describe
2304  * describe something that is process related.
2305  */
2306 static const struct pid_entry proc_base_stuff[] = {
2307         NOD("self", S_IFLNK|S_IRWXUGO,
2308                 &proc_self_inode_operations, NULL, {}),
2309 };
2310
2311 /*
2312  *      Exceptional case: normally we are not allowed to unhash a busy
2313  * directory. In this case, however, we can do it - no aliasing problems
2314  * due to the way we treat inodes.
2315  */
2316 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2317 {
2318         struct inode *inode = dentry->d_inode;
2319         struct task_struct *task = get_proc_task(inode);
2320         if (task) {
2321                 put_task_struct(task);
2322                 return 1;
2323         }
2324         d_drop(dentry);
2325         return 0;
2326 }
2327
2328 static const struct dentry_operations proc_base_dentry_operations =
2329 {
2330         .d_revalidate   = proc_base_revalidate,
2331         .d_delete       = pid_delete_dentry,
2332 };
2333
2334 static struct dentry *proc_base_instantiate(struct inode *dir,
2335         struct dentry *dentry, struct task_struct *task, const void *ptr)
2336 {
2337         const struct pid_entry *p = ptr;
2338         struct inode *inode;
2339         struct proc_inode *ei;
2340         struct dentry *error = ERR_PTR(-EINVAL);
2341
2342         /* Allocate the inode */
2343         error = ERR_PTR(-ENOMEM);
2344         inode = new_inode(dir->i_sb);
2345         if (!inode)
2346                 goto out;
2347
2348         /* Initialize the inode */
2349         ei = PROC_I(inode);
2350         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2351
2352         /*
2353          * grab the reference to the task.
2354          */
2355         ei->pid = get_task_pid(task, PIDTYPE_PID);
2356         if (!ei->pid)
2357                 goto out_iput;
2358
2359         inode->i_mode = p->mode;
2360         if (S_ISDIR(inode->i_mode))
2361                 inode->i_nlink = 2;
2362         if (S_ISLNK(inode->i_mode))
2363                 inode->i_size = 64;
2364         if (p->iop)
2365                 inode->i_op = p->iop;
2366         if (p->fop)
2367                 inode->i_fop = p->fop;
2368         ei->op = p->op;
2369         dentry->d_op = &proc_base_dentry_operations;
2370         d_add(dentry, inode);
2371         error = NULL;
2372 out:
2373         return error;
2374 out_iput:
2375         iput(inode);
2376         goto out;
2377 }
2378
2379 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2380 {
2381         struct dentry *error;
2382         struct task_struct *task = get_proc_task(dir);
2383         const struct pid_entry *p, *last;
2384
2385         error = ERR_PTR(-ENOENT);
2386
2387         if (!task)
2388                 goto out_no_task;
2389
2390         /* Lookup the directory entry */
2391         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2392         for (p = proc_base_stuff; p <= last; p++) {
2393                 if (p->len != dentry->d_name.len)
2394                         continue;
2395                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2396                         break;
2397         }
2398         if (p > last)
2399                 goto out;
2400
2401         error = proc_base_instantiate(dir, dentry, task, p);
2402
2403 out:
2404         put_task_struct(task);
2405 out_no_task:
2406         return error;
2407 }
2408
2409 static int proc_base_fill_cache(struct file *filp, void *dirent,
2410         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2411 {
2412         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2413                                 proc_base_instantiate, task, p);
2414 }
2415
2416 #ifdef CONFIG_TASK_IO_ACCOUNTING
2417 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2418 {
2419         struct task_io_accounting acct = task->ioac;
2420         unsigned long flags;
2421
2422         if (whole && lock_task_sighand(task, &flags)) {
2423                 struct task_struct *t = task;
2424
2425                 task_io_accounting_add(&acct, &task->signal->ioac);
2426                 while_each_thread(task, t)
2427                         task_io_accounting_add(&acct, &t->ioac);
2428
2429                 unlock_task_sighand(task, &flags);
2430         }
2431         return sprintf(buffer,
2432                         "rchar: %llu\n"
2433                         "wchar: %llu\n"
2434                         "syscr: %llu\n"
2435                         "syscw: %llu\n"
2436                         "read_bytes: %llu\n"
2437                         "write_bytes: %llu\n"
2438                         "cancelled_write_bytes: %llu\n",
2439                         (unsigned long long)acct.rchar,
2440                         (unsigned long long)acct.wchar,
2441                         (unsigned long long)acct.syscr,
2442                         (unsigned long long)acct.syscw,
2443                         (unsigned long long)acct.read_bytes,
2444                         (unsigned long long)acct.write_bytes,
2445                         (unsigned long long)acct.cancelled_write_bytes);
2446 }
2447
2448 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2449 {
2450         return do_io_accounting(task, buffer, 0);
2451 }
2452
2453 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2454 {
2455         return do_io_accounting(task, buffer, 1);
2456 }
2457 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2458
2459 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2460                                 struct pid *pid, struct task_struct *task)
2461 {
2462         seq_printf(m, "%08x\n", task->personality);
2463         return 0;
2464 }
2465
2466 /*
2467  * Thread groups
2468  */
2469 static const struct file_operations proc_task_operations;
2470 static const struct inode_operations proc_task_inode_operations;
2471
2472 static const struct pid_entry tgid_base_stuff[] = {
2473         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2474         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2475         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2476 #ifdef CONFIG_NET
2477         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2478 #endif
2479         REG("environ",    S_IRUSR, proc_environ_operations),
2480         INF("auxv",       S_IRUSR, proc_pid_auxv),
2481         ONE("status",     S_IRUGO, proc_pid_status),
2482         ONE("personality", S_IRUSR, proc_pid_personality),
2483         INF("limits",     S_IRUSR, proc_pid_limits),
2484 #ifdef CONFIG_SCHED_DEBUG
2485         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2486 #endif
2487 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2488         INF("syscall",    S_IRUSR, proc_pid_syscall),
2489 #endif
2490         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2491         ONE("stat",       S_IRUGO, proc_tgid_stat),
2492         ONE("statm",      S_IRUGO, proc_pid_statm),
2493         REG("maps",       S_IRUGO, proc_maps_operations),
2494 #ifdef CONFIG_NUMA
2495         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2496 #endif
2497         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2498         LNK("cwd",        proc_cwd_link),
2499         LNK("root",       proc_root_link),
2500         LNK("exe",        proc_exe_link),
2501         REG("mounts",     S_IRUGO, proc_mounts_operations),
2502         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2503         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2504 #ifdef CONFIG_PROC_PAGE_MONITOR
2505         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2506         REG("smaps",      S_IRUGO, proc_smaps_operations),
2507         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2508 #endif
2509 #ifdef CONFIG_SECURITY
2510         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2511 #endif
2512 #ifdef CONFIG_KALLSYMS
2513         INF("wchan",      S_IRUGO, proc_pid_wchan),
2514 #endif
2515 #ifdef CONFIG_STACKTRACE
2516         ONE("stack",      S_IRUSR, proc_pid_stack),
2517 #endif
2518 #ifdef CONFIG_SCHEDSTATS
2519         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2520 #endif
2521 #ifdef CONFIG_LATENCYTOP
2522         REG("latency",  S_IRUGO, proc_lstats_operations),
2523 #endif
2524 #ifdef CONFIG_PROC_PID_CPUSET
2525         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2526 #endif
2527 #ifdef CONFIG_CGROUPS
2528         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2529 #endif
2530         INF("oom_score",  S_IRUGO, proc_oom_score),
2531         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2532 #ifdef CONFIG_AUDITSYSCALL
2533         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2534         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2535 #endif
2536 #ifdef CONFIG_FAULT_INJECTION
2537         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2538 #endif
2539 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2540         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2541 #endif
2542 #ifdef CONFIG_TASK_IO_ACCOUNTING
2543         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2544 #endif
2545 };
2546
2547 static int proc_tgid_base_readdir(struct file * filp,
2548                              void * dirent, filldir_t filldir)
2549 {
2550         return proc_pident_readdir(filp,dirent,filldir,
2551                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2552 }
2553
2554 static const struct file_operations proc_tgid_base_operations = {
2555         .read           = generic_read_dir,
2556         .readdir        = proc_tgid_base_readdir,
2557 };
2558
2559 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2560         return proc_pident_lookup(dir, dentry,
2561                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2562 }
2563
2564 static const struct inode_operations proc_tgid_base_inode_operations = {
2565         .lookup         = proc_tgid_base_lookup,
2566         .getattr        = pid_getattr,
2567         .setattr        = proc_setattr,
2568 };
2569
2570 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2571 {
2572         struct dentry *dentry, *leader, *dir;
2573         char buf[PROC_NUMBUF];
2574         struct qstr name;
2575
2576         name.name = buf;
2577         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2578         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2579         if (dentry) {
2580                 if (!(current->flags & PF_EXITING))
2581                         shrink_dcache_parent(dentry);
2582                 d_drop(dentry);
2583                 dput(dentry);
2584         }
2585
2586         if (tgid == 0)
2587                 goto out;
2588
2589         name.name = buf;
2590         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2591         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2592         if (!leader)
2593                 goto out;
2594
2595         name.name = "task";
2596         name.len = strlen(name.name);
2597         dir = d_hash_and_lookup(leader, &name);
2598         if (!dir)
2599                 goto out_put_leader;
2600
2601         name.name = buf;
2602         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2603         dentry = d_hash_and_lookup(dir, &name);
2604         if (dentry) {
2605                 shrink_dcache_parent(dentry);
2606                 d_drop(dentry);
2607                 dput(dentry);
2608         }
2609
2610         dput(dir);
2611 out_put_leader:
2612         dput(leader);
2613 out:
2614         return;
2615 }
2616
2617 /**
2618  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2619  * @task: task that should be flushed.
2620  *
2621  * When flushing dentries from proc, one needs to flush them from global
2622  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2623  * in. This call is supposed to do all of this job.
2624  *
2625  * Looks in the dcache for
2626  * /proc/@pid
2627  * /proc/@tgid/task/@pid
2628  * if either directory is present flushes it and all of it'ts children
2629  * from the dcache.
2630  *
2631  * It is safe and reasonable to cache /proc entries for a task until
2632  * that task exits.  After that they just clog up the dcache with
2633  * useless entries, possibly causing useful dcache entries to be
2634  * flushed instead.  This routine is proved to flush those useless
2635  * dcache entries at process exit time.
2636  *
2637  * NOTE: This routine is just an optimization so it does not guarantee
2638  *       that no dcache entries will exist at process exit time it
2639  *       just makes it very unlikely that any will persist.
2640  */
2641
2642 void proc_flush_task(struct task_struct *task)
2643 {
2644         int i;
2645         struct pid *pid, *tgid = NULL;
2646         struct upid *upid;
2647
2648         pid = task_pid(task);
2649         if (thread_group_leader(task))
2650                 tgid = task_tgid(task);
2651
2652         for (i = 0; i <= pid->level; i++) {
2653                 upid = &pid->numbers[i];
2654                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2655                         tgid ? tgid->numbers[i].nr : 0);
2656         }
2657
2658         upid = &pid->numbers[pid->level];
2659         if (upid->nr == 1)
2660                 pid_ns_release_proc(upid->ns);
2661 }
2662
2663 static struct dentry *proc_pid_instantiate(struct inode *dir,
2664                                            struct dentry * dentry,
2665                                            struct task_struct *task, const void *ptr)
2666 {
2667         struct dentry *error = ERR_PTR(-ENOENT);
2668         struct inode *inode;
2669
2670         inode = proc_pid_make_inode(dir->i_sb, task);
2671         if (!inode)
2672                 goto out;
2673
2674         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2675         inode->i_op = &proc_tgid_base_inode_operations;
2676         inode->i_fop = &proc_tgid_base_operations;
2677         inode->i_flags|=S_IMMUTABLE;
2678
2679         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2680                 ARRAY_SIZE(tgid_base_stuff));
2681
2682         dentry->d_op = &pid_dentry_operations;
2683
2684         d_add(dentry, inode);
2685         /* Close the race of the process dying before we return the dentry */
2686         if (pid_revalidate(dentry, NULL))
2687                 error = NULL;
2688 out:
2689         return error;
2690 }
2691
2692 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2693 {
2694         struct dentry *result = ERR_PTR(-ENOENT);
2695         struct task_struct *task;
2696         unsigned tgid;
2697         struct pid_namespace *ns;
2698
2699         result = proc_base_lookup(dir, dentry);
2700         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2701                 goto out;
2702
2703         tgid = name_to_int(dentry);
2704         if (tgid == ~0U)
2705                 goto out;
2706
2707         ns = dentry->d_sb->s_fs_info;
2708         rcu_read_lock();
2709         task = find_task_by_pid_ns(tgid, ns);
2710         if (task)
2711                 get_task_struct(task);
2712         rcu_read_unlock();
2713         if (!task)
2714                 goto out;
2715
2716         result = proc_pid_instantiate(dir, dentry, task, NULL);
2717         put_task_struct(task);
2718 out:
2719         return result;
2720 }
2721
2722 /*
2723  * Find the first task with tgid >= tgid
2724  *
2725  */
2726 struct tgid_iter {
2727         unsigned int tgid;
2728         struct task_struct *task;
2729 };
2730 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2731 {
2732         struct pid *pid;
2733
2734         if (iter.task)
2735                 put_task_struct(iter.task);
2736         rcu_read_lock();
2737 retry:
2738         iter.task = NULL;
2739         pid = find_ge_pid(iter.tgid, ns);
2740         if (pid) {
2741                 iter.tgid = pid_nr_ns(pid, ns);
2742                 iter.task = pid_task(pid, PIDTYPE_PID);
2743                 /* What we to know is if the pid we have find is the
2744                  * pid of a thread_group_leader.  Testing for task
2745                  * being a thread_group_leader is the obvious thing
2746                  * todo but there is a window when it fails, due to
2747                  * the pid transfer logic in de_thread.
2748                  *
2749                  * So we perform the straight forward test of seeing
2750                  * if the pid we have found is the pid of a thread
2751                  * group leader, and don't worry if the task we have
2752                  * found doesn't happen to be a thread group leader.
2753                  * As we don't care in the case of readdir.
2754                  */
2755                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2756                         iter.tgid += 1;
2757                         goto retry;
2758                 }
2759                 get_task_struct(iter.task);
2760         }
2761         rcu_read_unlock();
2762         return iter;
2763 }
2764
2765 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2766
2767 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2768         struct tgid_iter iter)
2769 {
2770         char name[PROC_NUMBUF];
2771         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2772         return proc_fill_cache(filp, dirent, filldir, name, len,
2773                                 proc_pid_instantiate, iter.task, NULL);
2774 }
2775
2776 /* for the /proc/ directory itself, after non-process stuff has been done */
2777 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2778 {
2779         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2780         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2781         struct tgid_iter iter;
2782         struct pid_namespace *ns;
2783
2784         if (!reaper)
2785                 goto out_no_task;
2786
2787         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2788                 const struct pid_entry *p = &proc_base_stuff[nr];
2789                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2790                         goto out;
2791         }
2792
2793         ns = filp->f_dentry->d_sb->s_fs_info;
2794         iter.task = NULL;
2795         iter.tgid = filp->f_pos - TGID_OFFSET;
2796         for (iter = next_tgid(ns, iter);
2797              iter.task;
2798              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2799                 filp->f_pos = iter.tgid + TGID_OFFSET;
2800                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2801                         put_task_struct(iter.task);
2802                         goto out;
2803                 }
2804         }
2805         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2806 out:
2807         put_task_struct(reaper);
2808 out_no_task:
2809         return 0;
2810 }
2811
2812 /*
2813  * Tasks
2814  */
2815 static const struct pid_entry tid_base_stuff[] = {
2816         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2817         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2818         REG("environ",   S_IRUSR, proc_environ_operations),
2819         INF("auxv",      S_IRUSR, proc_pid_auxv),
2820         ONE("status",    S_IRUGO, proc_pid_status),
2821         ONE("personality", S_IRUSR, proc_pid_personality),
2822         INF("limits",    S_IRUSR, proc_pid_limits),
2823 #ifdef CONFIG_SCHED_DEBUG
2824         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2825 #endif
2826 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2827         INF("syscall",   S_IRUSR, proc_pid_syscall),
2828 #endif
2829         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2830         ONE("stat",      S_IRUGO, proc_tid_stat),
2831         ONE("statm",     S_IRUGO, proc_pid_statm),
2832         REG("maps",      S_IRUGO, proc_maps_operations),
2833 #ifdef CONFIG_NUMA
2834         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2835 #endif
2836         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2837         LNK("cwd",       proc_cwd_link),
2838         LNK("root",      proc_root_link),
2839         LNK("exe",       proc_exe_link),
2840         REG("mounts",    S_IRUGO, proc_mounts_operations),
2841         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2842 #ifdef CONFIG_PROC_PAGE_MONITOR
2843         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2844         REG("smaps",     S_IRUGO, proc_smaps_operations),
2845         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2846 #endif
2847 #ifdef CONFIG_SECURITY
2848         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2849 #endif
2850 #ifdef CONFIG_KALLSYMS
2851         INF("wchan",     S_IRUGO, proc_pid_wchan),
2852 #endif
2853 #ifdef CONFIG_STACKTRACE
2854         ONE("stack",      S_IRUSR, proc_pid_stack),
2855 #endif
2856 #ifdef CONFIG_SCHEDSTATS
2857         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2858 #endif
2859 #ifdef CONFIG_LATENCYTOP
2860         REG("latency",  S_IRUGO, proc_lstats_operations),
2861 #endif
2862 #ifdef CONFIG_PROC_PID_CPUSET
2863         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2864 #endif
2865 #ifdef CONFIG_CGROUPS
2866         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2867 #endif
2868         INF("oom_score", S_IRUGO, proc_oom_score),
2869         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2870 #ifdef CONFIG_AUDITSYSCALL
2871         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2872         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2873 #endif
2874 #ifdef CONFIG_FAULT_INJECTION
2875         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2876 #endif
2877 #ifdef CONFIG_TASK_IO_ACCOUNTING
2878         INF("io",       S_IRUGO, proc_tid_io_accounting),
2879 #endif
2880 };
2881
2882 static int proc_tid_base_readdir(struct file * filp,
2883                              void * dirent, filldir_t filldir)
2884 {
2885         return proc_pident_readdir(filp,dirent,filldir,
2886                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2887 }
2888
2889 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2890         return proc_pident_lookup(dir, dentry,
2891                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2892 }
2893
2894 static const struct file_operations proc_tid_base_operations = {
2895         .read           = generic_read_dir,
2896         .readdir        = proc_tid_base_readdir,
2897 };
2898
2899 static const struct inode_operations proc_tid_base_inode_operations = {
2900         .lookup         = proc_tid_base_lookup,
2901         .getattr        = pid_getattr,
2902         .setattr        = proc_setattr,
2903 };
2904
2905 static struct dentry *proc_task_instantiate(struct inode *dir,
2906         struct dentry *dentry, struct task_struct *task, const void *ptr)
2907 {
2908         struct dentry *error = ERR_PTR(-ENOENT);
2909         struct inode *inode;
2910         inode = proc_pid_make_inode(dir->i_sb, task);
2911
2912         if (!inode)
2913                 goto out;
2914         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2915         inode->i_op = &proc_tid_base_inode_operations;
2916         inode->i_fop = &proc_tid_base_operations;
2917         inode->i_flags|=S_IMMUTABLE;
2918
2919         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2920                 ARRAY_SIZE(tid_base_stuff));
2921
2922         dentry->d_op = &pid_dentry_operations;
2923
2924         d_add(dentry, inode);
2925         /* Close the race of the process dying before we return the dentry */
2926         if (pid_revalidate(dentry, NULL))
2927                 error = NULL;
2928 out:
2929         return error;
2930 }
2931
2932 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2933 {
2934         struct dentry *result = ERR_PTR(-ENOENT);
2935         struct task_struct *task;
2936         struct task_struct *leader = get_proc_task(dir);
2937         unsigned tid;
2938         struct pid_namespace *ns;
2939
2940         if (!leader)
2941                 goto out_no_task;
2942
2943         tid = name_to_int(dentry);
2944         if (tid == ~0U)
2945                 goto out;
2946
2947         ns = dentry->d_sb->s_fs_info;
2948         rcu_read_lock();
2949         task = find_task_by_pid_ns(tid, ns);
2950         if (task)
2951                 get_task_struct(task);
2952         rcu_read_unlock();
2953         if (!task)
2954                 goto out;
2955         if (!same_thread_group(leader, task))
2956                 goto out_drop_task;
2957
2958         result = proc_task_instantiate(dir, dentry, task, NULL);
2959 out_drop_task:
2960         put_task_struct(task);
2961 out:
2962         put_task_struct(leader);
2963 out_no_task:
2964         return result;
2965 }
2966
2967 /*
2968  * Find the first tid of a thread group to return to user space.
2969  *
2970  * Usually this is just the thread group leader, but if the users
2971  * buffer was too small or there was a seek into the middle of the
2972  * directory we have more work todo.
2973  *
2974  * In the case of a short read we start with find_task_by_pid.
2975  *
2976  * In the case of a seek we start with the leader and walk nr
2977  * threads past it.
2978  */
2979 static struct task_struct *first_tid(struct task_struct *leader,
2980                 int tid, int nr, struct pid_namespace *ns)
2981 {
2982         struct task_struct *pos;
2983
2984         rcu_read_lock();
2985         /* Attempt to start with the pid of a thread */
2986         if (tid && (nr > 0)) {
2987                 pos = find_task_by_pid_ns(tid, ns);
2988                 if (pos && (pos->group_leader == leader))
2989                         goto found;
2990         }
2991
2992         /* If nr exceeds the number of threads there is nothing todo */
2993         pos = NULL;
2994         if (nr && nr >= get_nr_threads(leader))
2995                 goto out;
2996
2997         /* If we haven't found our starting place yet start
2998          * with the leader and walk nr threads forward.
2999          */
3000         for (pos = leader; nr > 0; --nr) {
3001                 pos = next_thread(pos);
3002                 if (pos == leader) {
3003                         pos = NULL;
3004                         goto out;
3005                 }
3006         }
3007 found:
3008         get_task_struct(pos);
3009 out:
3010         rcu_read_unlock();
3011         return pos;
3012 }
3013
3014 /*
3015  * Find the next thread in the thread list.
3016  * Return NULL if there is an error or no next thread.
3017  *
3018  * The reference to the input task_struct is released.
3019  */
3020 static struct task_struct *next_tid(struct task_struct *start)
3021 {
3022         struct task_struct *pos = NULL;
3023         rcu_read_lock();
3024         if (pid_alive(start)) {
3025                 pos = next_thread(start);
3026                 if (thread_group_leader(pos))
3027                         pos = NULL;
3028                 else
3029                         get_task_struct(pos);
3030         }
3031         rcu_read_unlock();
3032         put_task_struct(start);
3033         return pos;
3034 }
3035
3036 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3037         struct task_struct *task, int tid)
3038 {
3039         char name[PROC_NUMBUF];
3040         int len = snprintf(name, sizeof(name), "%d", tid);
3041         return proc_fill_cache(filp, dirent, filldir, name, len,
3042                                 proc_task_instantiate, task, NULL);
3043 }
3044
3045 /* for the /proc/TGID/task/ directories */
3046 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3047 {
3048         struct dentry *dentry = filp->f_path.dentry;
3049         struct inode *inode = dentry->d_inode;
3050         struct task_struct *leader = NULL;
3051         struct task_struct *task;
3052         int retval = -ENOENT;
3053         ino_t ino;
3054         int tid;
3055         struct pid_namespace *ns;
3056
3057         task = get_proc_task(inode);
3058         if (!task)
3059                 goto out_no_task;
3060         rcu_read_lock();
3061         if (pid_alive(task)) {
3062                 leader = task->group_leader;
3063                 get_task_struct(leader);
3064         }
3065         rcu_read_unlock();
3066         put_task_struct(task);
3067         if (!leader)
3068                 goto out_no_task;
3069         retval = 0;
3070
3071         switch ((unsigned long)filp->f_pos) {
3072         case 0:
3073                 ino = inode->i_ino;
3074                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3075                         goto out;
3076                 filp->f_pos++;
3077                 /* fall through */
3078         case 1:
3079                 ino = parent_ino(dentry);
3080                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3081                         goto out;
3082                 filp->f_pos++;
3083                 /* fall through */
3084         }
3085
3086         /* f_version caches the tgid value that the last readdir call couldn't
3087          * return. lseek aka telldir automagically resets f_version to 0.
3088          */
3089         ns = filp->f_dentry->d_sb->s_fs_info;
3090         tid = (int)filp->f_version;
3091         filp->f_version = 0;
3092         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3093              task;
3094              task = next_tid(task), filp->f_pos++) {
3095                 tid = task_pid_nr_ns(task, ns);
3096                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3097                         /* returning this tgid failed, save it as the first
3098                          * pid for the next readir call */
3099                         filp->f_version = (u64)tid;
3100                         put_task_struct(task);
3101                         break;
3102                 }
3103         }
3104 out:
3105         put_task_struct(leader);
3106 out_no_task:
3107         return retval;
3108 }
3109
3110 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3111 {
3112         struct inode *inode = dentry->d_inode;
3113         struct task_struct *p = get_proc_task(inode);
3114         generic_fillattr(inode, stat);
3115
3116         if (p) {
3117                 stat->nlink += get_nr_threads(p);
3118                 put_task_struct(p);
3119         }
3120
3121         return 0;
3122 }
3123
3124 static const struct inode_operations proc_task_inode_operations = {
3125         .lookup         = proc_task_lookup,
3126         .getattr        = proc_task_getattr,
3127         .setattr        = proc_setattr,
3128 };
3129
3130 static const struct file_operations proc_task_operations = {
3131         .read           = generic_read_dir,
3132         .readdir        = proc_task_readdir,
3133 };