filp->f_pos not correctly updated in proc_task_readdir
[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 "internal.h"
84
85 /* NOTE:
86  *      Implementing inode permission operations in /proc is almost
87  *      certainly an error.  Permission checks need to happen during
88  *      each system call not at open time.  The reason is that most of
89  *      what we wish to check for permissions in /proc varies at runtime.
90  *
91  *      The classic example of a problem is opening file descriptors
92  *      in /proc for a task before it execs a suid executable.
93  */
94
95 struct pid_entry {
96         char *name;
97         int len;
98         mode_t mode;
99         const struct inode_operations *iop;
100         const struct file_operations *fop;
101         union proc_op op;
102 };
103
104 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
105         .name = (NAME),                                 \
106         .len  = sizeof(NAME) - 1,                       \
107         .mode = MODE,                                   \
108         .iop  = IOP,                                    \
109         .fop  = FOP,                                    \
110         .op   = OP,                                     \
111 }
112
113 #define DIR(NAME, MODE, iops, fops)     \
114         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
115 #define LNK(NAME, get_link)                                     \
116         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
117                 &proc_pid_link_inode_operations, NULL,          \
118                 { .proc_get_link = get_link } )
119 #define REG(NAME, MODE, fops)                           \
120         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
121 #define INF(NAME, MODE, read)                           \
122         NOD(NAME, (S_IFREG|(MODE)),                     \
123                 NULL, &proc_info_file_operations,       \
124                 { .proc_read = read } )
125 #define ONE(NAME, MODE, show)                           \
126         NOD(NAME, (S_IFREG|(MODE)),                     \
127                 NULL, &proc_single_file_operations,     \
128                 { .proc_show = show } )
129
130 /*
131  * Count the number of hardlinks for the pid_entry table, excluding the .
132  * and .. links.
133  */
134 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
135         unsigned int n)
136 {
137         unsigned int i;
138         unsigned int count;
139
140         count = 0;
141         for (i = 0; i < n; ++i) {
142                 if (S_ISDIR(entries[i].mode))
143                         ++count;
144         }
145
146         return count;
147 }
148
149 static struct fs_struct *get_fs_struct(struct task_struct *task)
150 {
151         struct fs_struct *fs;
152         task_lock(task);
153         fs = task->fs;
154         if(fs)
155                 atomic_inc(&fs->count);
156         task_unlock(task);
157         return fs;
158 }
159
160 static int get_nr_threads(struct task_struct *tsk)
161 {
162         unsigned long flags;
163         int count = 0;
164
165         if (lock_task_sighand(tsk, &flags)) {
166                 count = atomic_read(&tsk->signal->count);
167                 unlock_task_sighand(tsk, &flags);
168         }
169         return count;
170 }
171
172 static int proc_cwd_link(struct inode *inode, struct path *path)
173 {
174         struct task_struct *task = get_proc_task(inode);
175         struct fs_struct *fs = NULL;
176         int result = -ENOENT;
177
178         if (task) {
179                 fs = get_fs_struct(task);
180                 put_task_struct(task);
181         }
182         if (fs) {
183                 read_lock(&fs->lock);
184                 *path = fs->pwd;
185                 path_get(&fs->pwd);
186                 read_unlock(&fs->lock);
187                 result = 0;
188                 put_fs_struct(fs);
189         }
190         return result;
191 }
192
193 static int proc_root_link(struct inode *inode, struct path *path)
194 {
195         struct task_struct *task = get_proc_task(inode);
196         struct fs_struct *fs = NULL;
197         int result = -ENOENT;
198
199         if (task) {
200                 fs = get_fs_struct(task);
201                 put_task_struct(task);
202         }
203         if (fs) {
204                 read_lock(&fs->lock);
205                 *path = fs->root;
206                 path_get(&fs->root);
207                 read_unlock(&fs->lock);
208                 result = 0;
209                 put_fs_struct(fs);
210         }
211         return result;
212 }
213
214 /*
215  * Return zero if current may access user memory in @task, -error if not.
216  */
217 static int check_mem_permission(struct task_struct *task)
218 {
219         /*
220          * A task can always look at itself, in case it chooses
221          * to use system calls instead of load instructions.
222          */
223         if (task == current)
224                 return 0;
225
226         /*
227          * If current is actively ptrace'ing, and would also be
228          * permitted to freshly attach with ptrace now, permit it.
229          */
230         if (task_is_stopped_or_traced(task)) {
231                 int match;
232                 rcu_read_lock();
233                 match = (tracehook_tracer_task(task) == current);
234                 rcu_read_unlock();
235                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
236                         return 0;
237         }
238
239         /*
240          * Noone else is allowed.
241          */
242         return -EPERM;
243 }
244
245 struct mm_struct *mm_for_maps(struct task_struct *task)
246 {
247         struct mm_struct *mm = get_task_mm(task);
248         if (!mm)
249                 return NULL;
250         down_read(&mm->mmap_sem);
251         task_lock(task);
252         if (task->mm != mm)
253                 goto out;
254         if (task->mm != current->mm &&
255             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
256                 goto out;
257         task_unlock(task);
258         return mm;
259 out:
260         task_unlock(task);
261         up_read(&mm->mmap_sem);
262         mmput(mm);
263         return NULL;
264 }
265
266 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
267 {
268         int res = 0;
269         unsigned int len;
270         struct mm_struct *mm = get_task_mm(task);
271         if (!mm)
272                 goto out;
273         if (!mm->arg_end)
274                 goto out_mm;    /* Shh! No looking before we're done */
275
276         len = mm->arg_end - mm->arg_start;
277  
278         if (len > PAGE_SIZE)
279                 len = PAGE_SIZE;
280  
281         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
282
283         // If the nul at the end of args has been overwritten, then
284         // assume application is using setproctitle(3).
285         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
286                 len = strnlen(buffer, res);
287                 if (len < res) {
288                     res = len;
289                 } else {
290                         len = mm->env_end - mm->env_start;
291                         if (len > PAGE_SIZE - res)
292                                 len = PAGE_SIZE - res;
293                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
294                         res = strnlen(buffer, res);
295                 }
296         }
297 out_mm:
298         mmput(mm);
299 out:
300         return res;
301 }
302
303 static int proc_pid_auxv(struct task_struct *task, char *buffer)
304 {
305         int res = 0;
306         struct mm_struct *mm = get_task_mm(task);
307         if (mm) {
308                 unsigned int nwords = 0;
309                 do {
310                         nwords += 2;
311                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
312                 res = nwords * sizeof(mm->saved_auxv[0]);
313                 if (res > PAGE_SIZE)
314                         res = PAGE_SIZE;
315                 memcpy(buffer, mm->saved_auxv, res);
316                 mmput(mm);
317         }
318         return res;
319 }
320
321
322 #ifdef CONFIG_KALLSYMS
323 /*
324  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
325  * Returns the resolved symbol.  If that fails, simply return the address.
326  */
327 static int proc_pid_wchan(struct task_struct *task, char *buffer)
328 {
329         unsigned long wchan;
330         char symname[KSYM_NAME_LEN];
331
332         wchan = get_wchan(task);
333
334         if (lookup_symbol_name(wchan, symname) < 0)
335                 return sprintf(buffer, "%lu", wchan);
336         else
337                 return sprintf(buffer, "%s", symname);
338 }
339 #endif /* CONFIG_KALLSYMS */
340
341 #ifdef CONFIG_STACKTRACE
342
343 #define MAX_STACK_TRACE_DEPTH   64
344
345 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
346                           struct pid *pid, struct task_struct *task)
347 {
348         struct stack_trace trace;
349         unsigned long *entries;
350         int i;
351
352         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
353         if (!entries)
354                 return -ENOMEM;
355
356         trace.nr_entries        = 0;
357         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
358         trace.entries           = entries;
359         trace.skip              = 0;
360         save_stack_trace_tsk(task, &trace);
361
362         for (i = 0; i < trace.nr_entries; i++) {
363                 seq_printf(m, "[<%p>] %pS\n",
364                            (void *)entries[i], (void *)entries[i]);
365         }
366         kfree(entries);
367
368         return 0;
369 }
370 #endif
371
372 #ifdef CONFIG_SCHEDSTATS
373 /*
374  * Provides /proc/PID/schedstat
375  */
376 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
377 {
378         return sprintf(buffer, "%llu %llu %lu\n",
379                         (unsigned long long)task->se.sum_exec_runtime,
380                         (unsigned long long)task->sched_info.run_delay,
381                         task->sched_info.pcount);
382 }
383 #endif
384
385 #ifdef CONFIG_LATENCYTOP
386 static int lstats_show_proc(struct seq_file *m, void *v)
387 {
388         int i;
389         struct inode *inode = m->private;
390         struct task_struct *task = get_proc_task(inode);
391
392         if (!task)
393                 return -ESRCH;
394         seq_puts(m, "Latency Top version : v0.1\n");
395         for (i = 0; i < 32; i++) {
396                 if (task->latency_record[i].backtrace[0]) {
397                         int q;
398                         seq_printf(m, "%i %li %li ",
399                                 task->latency_record[i].count,
400                                 task->latency_record[i].time,
401                                 task->latency_record[i].max);
402                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
403                                 char sym[KSYM_SYMBOL_LEN];
404                                 char *c;
405                                 if (!task->latency_record[i].backtrace[q])
406                                         break;
407                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
408                                         break;
409                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
410                                 c = strchr(sym, '+');
411                                 if (c)
412                                         *c = 0;
413                                 seq_printf(m, "%s ", sym);
414                         }
415                         seq_printf(m, "\n");
416                 }
417
418         }
419         put_task_struct(task);
420         return 0;
421 }
422
423 static int lstats_open(struct inode *inode, struct file *file)
424 {
425         return single_open(file, lstats_show_proc, inode);
426 }
427
428 static ssize_t lstats_write(struct file *file, const char __user *buf,
429                             size_t count, loff_t *offs)
430 {
431         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
432
433         if (!task)
434                 return -ESRCH;
435         clear_all_latency_tracing(task);
436         put_task_struct(task);
437
438         return count;
439 }
440
441 static const struct file_operations proc_lstats_operations = {
442         .open           = lstats_open,
443         .read           = seq_read,
444         .write          = lstats_write,
445         .llseek         = seq_lseek,
446         .release        = single_release,
447 };
448
449 #endif
450
451 /* The badness from the OOM killer */
452 unsigned long badness(struct task_struct *p, unsigned long uptime);
453 static int proc_oom_score(struct task_struct *task, char *buffer)
454 {
455         unsigned long points;
456         struct timespec uptime;
457
458         do_posix_clock_monotonic_gettime(&uptime);
459         read_lock(&tasklist_lock);
460         points = badness(task, uptime.tv_sec);
461         read_unlock(&tasklist_lock);
462         return sprintf(buffer, "%lu\n", points);
463 }
464
465 struct limit_names {
466         char *name;
467         char *unit;
468 };
469
470 static const struct limit_names lnames[RLIM_NLIMITS] = {
471         [RLIMIT_CPU] = {"Max cpu time", "ms"},
472         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
473         [RLIMIT_DATA] = {"Max data size", "bytes"},
474         [RLIMIT_STACK] = {"Max stack size", "bytes"},
475         [RLIMIT_CORE] = {"Max core file size", "bytes"},
476         [RLIMIT_RSS] = {"Max resident set", "bytes"},
477         [RLIMIT_NPROC] = {"Max processes", "processes"},
478         [RLIMIT_NOFILE] = {"Max open files", "files"},
479         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
480         [RLIMIT_AS] = {"Max address space", "bytes"},
481         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
482         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
483         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
484         [RLIMIT_NICE] = {"Max nice priority", NULL},
485         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
486         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
487 };
488
489 /* Display limits for a process */
490 static int proc_pid_limits(struct task_struct *task, char *buffer)
491 {
492         unsigned int i;
493         int count = 0;
494         unsigned long flags;
495         char *bufptr = buffer;
496
497         struct rlimit rlim[RLIM_NLIMITS];
498
499         if (!lock_task_sighand(task, &flags))
500                 return 0;
501         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
502         unlock_task_sighand(task, &flags);
503
504         /*
505          * print the file header
506          */
507         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
508                         "Limit", "Soft Limit", "Hard Limit", "Units");
509
510         for (i = 0; i < RLIM_NLIMITS; i++) {
511                 if (rlim[i].rlim_cur == RLIM_INFINITY)
512                         count += sprintf(&bufptr[count], "%-25s %-20s ",
513                                          lnames[i].name, "unlimited");
514                 else
515                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
516                                          lnames[i].name, rlim[i].rlim_cur);
517
518                 if (rlim[i].rlim_max == RLIM_INFINITY)
519                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
520                 else
521                         count += sprintf(&bufptr[count], "%-20lu ",
522                                          rlim[i].rlim_max);
523
524                 if (lnames[i].unit)
525                         count += sprintf(&bufptr[count], "%-10s\n",
526                                          lnames[i].unit);
527                 else
528                         count += sprintf(&bufptr[count], "\n");
529         }
530
531         return count;
532 }
533
534 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
535 static int proc_pid_syscall(struct task_struct *task, char *buffer)
536 {
537         long nr;
538         unsigned long args[6], sp, pc;
539
540         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
541                 return sprintf(buffer, "running\n");
542
543         if (nr < 0)
544                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
545
546         return sprintf(buffer,
547                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
548                        nr,
549                        args[0], args[1], args[2], args[3], args[4], args[5],
550                        sp, pc);
551 }
552 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
553
554 /************************************************************************/
555 /*                       Here the fs part begins                        */
556 /************************************************************************/
557
558 /* permission checks */
559 static int proc_fd_access_allowed(struct inode *inode)
560 {
561         struct task_struct *task;
562         int allowed = 0;
563         /* Allow access to a task's file descriptors if it is us or we
564          * may use ptrace attach to the process and find out that
565          * information.
566          */
567         task = get_proc_task(inode);
568         if (task) {
569                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
570                 put_task_struct(task);
571         }
572         return allowed;
573 }
574
575 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
576 {
577         int error;
578         struct inode *inode = dentry->d_inode;
579
580         if (attr->ia_valid & ATTR_MODE)
581                 return -EPERM;
582
583         error = inode_change_ok(inode, attr);
584         if (!error)
585                 error = inode_setattr(inode, attr);
586         return error;
587 }
588
589 static const struct inode_operations proc_def_inode_operations = {
590         .setattr        = proc_setattr,
591 };
592
593 static int mounts_open_common(struct inode *inode, struct file *file,
594                               const struct seq_operations *op)
595 {
596         struct task_struct *task = get_proc_task(inode);
597         struct nsproxy *nsp;
598         struct mnt_namespace *ns = NULL;
599         struct fs_struct *fs = NULL;
600         struct path root;
601         struct proc_mounts *p;
602         int ret = -EINVAL;
603
604         if (task) {
605                 rcu_read_lock();
606                 nsp = task_nsproxy(task);
607                 if (nsp) {
608                         ns = nsp->mnt_ns;
609                         if (ns)
610                                 get_mnt_ns(ns);
611                 }
612                 rcu_read_unlock();
613                 if (ns)
614                         fs = get_fs_struct(task);
615                 put_task_struct(task);
616         }
617
618         if (!ns)
619                 goto err;
620         if (!fs)
621                 goto err_put_ns;
622
623         read_lock(&fs->lock);
624         root = fs->root;
625         path_get(&root);
626         read_unlock(&fs->lock);
627         put_fs_struct(fs);
628
629         ret = -ENOMEM;
630         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
631         if (!p)
632                 goto err_put_path;
633
634         file->private_data = &p->m;
635         ret = seq_open(file, op);
636         if (ret)
637                 goto err_free;
638
639         p->m.private = p;
640         p->ns = ns;
641         p->root = root;
642         p->event = ns->event;
643
644         return 0;
645
646  err_free:
647         kfree(p);
648  err_put_path:
649         path_put(&root);
650  err_put_ns:
651         put_mnt_ns(ns);
652  err:
653         return ret;
654 }
655
656 static int mounts_release(struct inode *inode, struct file *file)
657 {
658         struct proc_mounts *p = file->private_data;
659         path_put(&p->root);
660         put_mnt_ns(p->ns);
661         return seq_release(inode, file);
662 }
663
664 static unsigned mounts_poll(struct file *file, poll_table *wait)
665 {
666         struct proc_mounts *p = file->private_data;
667         struct mnt_namespace *ns = p->ns;
668         unsigned res = 0;
669
670         poll_wait(file, &ns->poll, wait);
671
672         spin_lock(&vfsmount_lock);
673         if (p->event != ns->event) {
674                 p->event = ns->event;
675                 res = POLLERR;
676         }
677         spin_unlock(&vfsmount_lock);
678
679         return res;
680 }
681
682 static int mounts_open(struct inode *inode, struct file *file)
683 {
684         return mounts_open_common(inode, file, &mounts_op);
685 }
686
687 static const struct file_operations proc_mounts_operations = {
688         .open           = mounts_open,
689         .read           = seq_read,
690         .llseek         = seq_lseek,
691         .release        = mounts_release,
692         .poll           = mounts_poll,
693 };
694
695 static int mountinfo_open(struct inode *inode, struct file *file)
696 {
697         return mounts_open_common(inode, file, &mountinfo_op);
698 }
699
700 static const struct file_operations proc_mountinfo_operations = {
701         .open           = mountinfo_open,
702         .read           = seq_read,
703         .llseek         = seq_lseek,
704         .release        = mounts_release,
705         .poll           = mounts_poll,
706 };
707
708 static int mountstats_open(struct inode *inode, struct file *file)
709 {
710         return mounts_open_common(inode, file, &mountstats_op);
711 }
712
713 static const struct file_operations proc_mountstats_operations = {
714         .open           = mountstats_open,
715         .read           = seq_read,
716         .llseek         = seq_lseek,
717         .release        = mounts_release,
718 };
719
720 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
721
722 static ssize_t proc_info_read(struct file * file, char __user * buf,
723                           size_t count, loff_t *ppos)
724 {
725         struct inode * inode = file->f_path.dentry->d_inode;
726         unsigned long page;
727         ssize_t length;
728         struct task_struct *task = get_proc_task(inode);
729
730         length = -ESRCH;
731         if (!task)
732                 goto out_no_task;
733
734         if (count > PROC_BLOCK_SIZE)
735                 count = PROC_BLOCK_SIZE;
736
737         length = -ENOMEM;
738         if (!(page = __get_free_page(GFP_TEMPORARY)))
739                 goto out;
740
741         length = PROC_I(inode)->op.proc_read(task, (char*)page);
742
743         if (length >= 0)
744                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
745         free_page(page);
746 out:
747         put_task_struct(task);
748 out_no_task:
749         return length;
750 }
751
752 static const struct file_operations proc_info_file_operations = {
753         .read           = proc_info_read,
754 };
755
756 static int proc_single_show(struct seq_file *m, void *v)
757 {
758         struct inode *inode = m->private;
759         struct pid_namespace *ns;
760         struct pid *pid;
761         struct task_struct *task;
762         int ret;
763
764         ns = inode->i_sb->s_fs_info;
765         pid = proc_pid(inode);
766         task = get_pid_task(pid, PIDTYPE_PID);
767         if (!task)
768                 return -ESRCH;
769
770         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
771
772         put_task_struct(task);
773         return ret;
774 }
775
776 static int proc_single_open(struct inode *inode, struct file *filp)
777 {
778         int ret;
779         ret = single_open(filp, proc_single_show, NULL);
780         if (!ret) {
781                 struct seq_file *m = filp->private_data;
782
783                 m->private = inode;
784         }
785         return ret;
786 }
787
788 static const struct file_operations proc_single_file_operations = {
789         .open           = proc_single_open,
790         .read           = seq_read,
791         .llseek         = seq_lseek,
792         .release        = single_release,
793 };
794
795 static int mem_open(struct inode* inode, struct file* file)
796 {
797         file->private_data = (void*)((long)current->self_exec_id);
798         return 0;
799 }
800
801 static ssize_t mem_read(struct file * file, char __user * buf,
802                         size_t count, loff_t *ppos)
803 {
804         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
805         char *page;
806         unsigned long src = *ppos;
807         int ret = -ESRCH;
808         struct mm_struct *mm;
809
810         if (!task)
811                 goto out_no_task;
812
813         if (check_mem_permission(task))
814                 goto out;
815
816         ret = -ENOMEM;
817         page = (char *)__get_free_page(GFP_TEMPORARY);
818         if (!page)
819                 goto out;
820
821         ret = 0;
822  
823         mm = get_task_mm(task);
824         if (!mm)
825                 goto out_free;
826
827         ret = -EIO;
828  
829         if (file->private_data != (void*)((long)current->self_exec_id))
830                 goto out_put;
831
832         ret = 0;
833  
834         while (count > 0) {
835                 int this_len, retval;
836
837                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
838                 retval = access_process_vm(task, src, page, this_len, 0);
839                 if (!retval || check_mem_permission(task)) {
840                         if (!ret)
841                                 ret = -EIO;
842                         break;
843                 }
844
845                 if (copy_to_user(buf, page, retval)) {
846                         ret = -EFAULT;
847                         break;
848                 }
849  
850                 ret += retval;
851                 src += retval;
852                 buf += retval;
853                 count -= retval;
854         }
855         *ppos = src;
856
857 out_put:
858         mmput(mm);
859 out_free:
860         free_page((unsigned long) page);
861 out:
862         put_task_struct(task);
863 out_no_task:
864         return ret;
865 }
866
867 #define mem_write NULL
868
869 #ifndef mem_write
870 /* This is a security hazard */
871 static ssize_t mem_write(struct file * file, const char __user *buf,
872                          size_t count, loff_t *ppos)
873 {
874         int copied;
875         char *page;
876         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877         unsigned long dst = *ppos;
878
879         copied = -ESRCH;
880         if (!task)
881                 goto out_no_task;
882
883         if (check_mem_permission(task))
884                 goto out;
885
886         copied = -ENOMEM;
887         page = (char *)__get_free_page(GFP_TEMPORARY);
888         if (!page)
889                 goto out;
890
891         copied = 0;
892         while (count > 0) {
893                 int this_len, retval;
894
895                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
896                 if (copy_from_user(page, buf, this_len)) {
897                         copied = -EFAULT;
898                         break;
899                 }
900                 retval = access_process_vm(task, dst, page, this_len, 1);
901                 if (!retval) {
902                         if (!copied)
903                                 copied = -EIO;
904                         break;
905                 }
906                 copied += retval;
907                 buf += retval;
908                 dst += retval;
909                 count -= retval;                        
910         }
911         *ppos = dst;
912         free_page((unsigned long) page);
913 out:
914         put_task_struct(task);
915 out_no_task:
916         return copied;
917 }
918 #endif
919
920 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
921 {
922         switch (orig) {
923         case 0:
924                 file->f_pos = offset;
925                 break;
926         case 1:
927                 file->f_pos += offset;
928                 break;
929         default:
930                 return -EINVAL;
931         }
932         force_successful_syscall_return();
933         return file->f_pos;
934 }
935
936 static const struct file_operations proc_mem_operations = {
937         .llseek         = mem_lseek,
938         .read           = mem_read,
939         .write          = mem_write,
940         .open           = mem_open,
941 };
942
943 static ssize_t environ_read(struct file *file, char __user *buf,
944                         size_t count, loff_t *ppos)
945 {
946         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
947         char *page;
948         unsigned long src = *ppos;
949         int ret = -ESRCH;
950         struct mm_struct *mm;
951
952         if (!task)
953                 goto out_no_task;
954
955         if (!ptrace_may_access(task, PTRACE_MODE_READ))
956                 goto out;
957
958         ret = -ENOMEM;
959         page = (char *)__get_free_page(GFP_TEMPORARY);
960         if (!page)
961                 goto out;
962
963         ret = 0;
964
965         mm = get_task_mm(task);
966         if (!mm)
967                 goto out_free;
968
969         while (count > 0) {
970                 int this_len, retval, max_len;
971
972                 this_len = mm->env_end - (mm->env_start + src);
973
974                 if (this_len <= 0)
975                         break;
976
977                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
978                 this_len = (this_len > max_len) ? max_len : this_len;
979
980                 retval = access_process_vm(task, (mm->env_start + src),
981                         page, this_len, 0);
982
983                 if (retval <= 0) {
984                         ret = retval;
985                         break;
986                 }
987
988                 if (copy_to_user(buf, page, retval)) {
989                         ret = -EFAULT;
990                         break;
991                 }
992
993                 ret += retval;
994                 src += retval;
995                 buf += retval;
996                 count -= retval;
997         }
998         *ppos = src;
999
1000         mmput(mm);
1001 out_free:
1002         free_page((unsigned long) page);
1003 out:
1004         put_task_struct(task);
1005 out_no_task:
1006         return ret;
1007 }
1008
1009 static const struct file_operations proc_environ_operations = {
1010         .read           = environ_read,
1011 };
1012
1013 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1014                                 size_t count, loff_t *ppos)
1015 {
1016         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1017         char buffer[PROC_NUMBUF];
1018         size_t len;
1019         int oom_adjust;
1020
1021         if (!task)
1022                 return -ESRCH;
1023         oom_adjust = task->oomkilladj;
1024         put_task_struct(task);
1025
1026         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1027
1028         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1029 }
1030
1031 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1032                                 size_t count, loff_t *ppos)
1033 {
1034         struct task_struct *task;
1035         char buffer[PROC_NUMBUF], *end;
1036         int oom_adjust;
1037
1038         memset(buffer, 0, sizeof(buffer));
1039         if (count > sizeof(buffer) - 1)
1040                 count = sizeof(buffer) - 1;
1041         if (copy_from_user(buffer, buf, count))
1042                 return -EFAULT;
1043         oom_adjust = simple_strtol(buffer, &end, 0);
1044         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1045              oom_adjust != OOM_DISABLE)
1046                 return -EINVAL;
1047         if (*end == '\n')
1048                 end++;
1049         task = get_proc_task(file->f_path.dentry->d_inode);
1050         if (!task)
1051                 return -ESRCH;
1052         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1053                 put_task_struct(task);
1054                 return -EACCES;
1055         }
1056         task->oomkilladj = oom_adjust;
1057         put_task_struct(task);
1058         if (end - buffer == 0)
1059                 return -EIO;
1060         return end - buffer;
1061 }
1062
1063 static const struct file_operations proc_oom_adjust_operations = {
1064         .read           = oom_adjust_read,
1065         .write          = oom_adjust_write,
1066 };
1067
1068 #ifdef CONFIG_AUDITSYSCALL
1069 #define TMPBUFLEN 21
1070 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1071                                   size_t count, loff_t *ppos)
1072 {
1073         struct inode * inode = file->f_path.dentry->d_inode;
1074         struct task_struct *task = get_proc_task(inode);
1075         ssize_t length;
1076         char tmpbuf[TMPBUFLEN];
1077
1078         if (!task)
1079                 return -ESRCH;
1080         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1081                                 audit_get_loginuid(task));
1082         put_task_struct(task);
1083         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1084 }
1085
1086 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1087                                    size_t count, loff_t *ppos)
1088 {
1089         struct inode * inode = file->f_path.dentry->d_inode;
1090         char *page, *tmp;
1091         ssize_t length;
1092         uid_t loginuid;
1093
1094         if (!capable(CAP_AUDIT_CONTROL))
1095                 return -EPERM;
1096
1097         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1098                 return -EPERM;
1099
1100         if (count >= PAGE_SIZE)
1101                 count = PAGE_SIZE - 1;
1102
1103         if (*ppos != 0) {
1104                 /* No partial writes. */
1105                 return -EINVAL;
1106         }
1107         page = (char*)__get_free_page(GFP_TEMPORARY);
1108         if (!page)
1109                 return -ENOMEM;
1110         length = -EFAULT;
1111         if (copy_from_user(page, buf, count))
1112                 goto out_free_page;
1113
1114         page[count] = '\0';
1115         loginuid = simple_strtoul(page, &tmp, 10);
1116         if (tmp == page) {
1117                 length = -EINVAL;
1118                 goto out_free_page;
1119
1120         }
1121         length = audit_set_loginuid(current, loginuid);
1122         if (likely(length == 0))
1123                 length = count;
1124
1125 out_free_page:
1126         free_page((unsigned long) page);
1127         return length;
1128 }
1129
1130 static const struct file_operations proc_loginuid_operations = {
1131         .read           = proc_loginuid_read,
1132         .write          = proc_loginuid_write,
1133 };
1134
1135 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1136                                   size_t count, loff_t *ppos)
1137 {
1138         struct inode * inode = file->f_path.dentry->d_inode;
1139         struct task_struct *task = get_proc_task(inode);
1140         ssize_t length;
1141         char tmpbuf[TMPBUFLEN];
1142
1143         if (!task)
1144                 return -ESRCH;
1145         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1146                                 audit_get_sessionid(task));
1147         put_task_struct(task);
1148         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1149 }
1150
1151 static const struct file_operations proc_sessionid_operations = {
1152         .read           = proc_sessionid_read,
1153 };
1154 #endif
1155
1156 #ifdef CONFIG_FAULT_INJECTION
1157 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1158                                       size_t count, loff_t *ppos)
1159 {
1160         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1161         char buffer[PROC_NUMBUF];
1162         size_t len;
1163         int make_it_fail;
1164
1165         if (!task)
1166                 return -ESRCH;
1167         make_it_fail = task->make_it_fail;
1168         put_task_struct(task);
1169
1170         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1171
1172         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1173 }
1174
1175 static ssize_t proc_fault_inject_write(struct file * file,
1176                         const char __user * buf, size_t count, loff_t *ppos)
1177 {
1178         struct task_struct *task;
1179         char buffer[PROC_NUMBUF], *end;
1180         int make_it_fail;
1181
1182         if (!capable(CAP_SYS_RESOURCE))
1183                 return -EPERM;
1184         memset(buffer, 0, sizeof(buffer));
1185         if (count > sizeof(buffer) - 1)
1186                 count = sizeof(buffer) - 1;
1187         if (copy_from_user(buffer, buf, count))
1188                 return -EFAULT;
1189         make_it_fail = simple_strtol(buffer, &end, 0);
1190         if (*end == '\n')
1191                 end++;
1192         task = get_proc_task(file->f_dentry->d_inode);
1193         if (!task)
1194                 return -ESRCH;
1195         task->make_it_fail = make_it_fail;
1196         put_task_struct(task);
1197         if (end - buffer == 0)
1198                 return -EIO;
1199         return end - buffer;
1200 }
1201
1202 static const struct file_operations proc_fault_inject_operations = {
1203         .read           = proc_fault_inject_read,
1204         .write          = proc_fault_inject_write,
1205 };
1206 #endif
1207
1208
1209 #ifdef CONFIG_SCHED_DEBUG
1210 /*
1211  * Print out various scheduling related per-task fields:
1212  */
1213 static int sched_show(struct seq_file *m, void *v)
1214 {
1215         struct inode *inode = m->private;
1216         struct task_struct *p;
1217
1218         p = get_proc_task(inode);
1219         if (!p)
1220                 return -ESRCH;
1221         proc_sched_show_task(p, m);
1222
1223         put_task_struct(p);
1224
1225         return 0;
1226 }
1227
1228 static ssize_t
1229 sched_write(struct file *file, const char __user *buf,
1230             size_t count, loff_t *offset)
1231 {
1232         struct inode *inode = file->f_path.dentry->d_inode;
1233         struct task_struct *p;
1234
1235         p = get_proc_task(inode);
1236         if (!p)
1237                 return -ESRCH;
1238         proc_sched_set_task(p);
1239
1240         put_task_struct(p);
1241
1242         return count;
1243 }
1244
1245 static int sched_open(struct inode *inode, struct file *filp)
1246 {
1247         int ret;
1248
1249         ret = single_open(filp, sched_show, NULL);
1250         if (!ret) {
1251                 struct seq_file *m = filp->private_data;
1252
1253                 m->private = inode;
1254         }
1255         return ret;
1256 }
1257
1258 static const struct file_operations proc_pid_sched_operations = {
1259         .open           = sched_open,
1260         .read           = seq_read,
1261         .write          = sched_write,
1262         .llseek         = seq_lseek,
1263         .release        = single_release,
1264 };
1265
1266 #endif
1267
1268 /*
1269  * We added or removed a vma mapping the executable. The vmas are only mapped
1270  * during exec and are not mapped with the mmap system call.
1271  * Callers must hold down_write() on the mm's mmap_sem for these
1272  */
1273 void added_exe_file_vma(struct mm_struct *mm)
1274 {
1275         mm->num_exe_file_vmas++;
1276 }
1277
1278 void removed_exe_file_vma(struct mm_struct *mm)
1279 {
1280         mm->num_exe_file_vmas--;
1281         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1282                 fput(mm->exe_file);
1283                 mm->exe_file = NULL;
1284         }
1285
1286 }
1287
1288 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1289 {
1290         if (new_exe_file)
1291                 get_file(new_exe_file);
1292         if (mm->exe_file)
1293                 fput(mm->exe_file);
1294         mm->exe_file = new_exe_file;
1295         mm->num_exe_file_vmas = 0;
1296 }
1297
1298 struct file *get_mm_exe_file(struct mm_struct *mm)
1299 {
1300         struct file *exe_file;
1301
1302         /* We need mmap_sem to protect against races with removal of
1303          * VM_EXECUTABLE vmas */
1304         down_read(&mm->mmap_sem);
1305         exe_file = mm->exe_file;
1306         if (exe_file)
1307                 get_file(exe_file);
1308         up_read(&mm->mmap_sem);
1309         return exe_file;
1310 }
1311
1312 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1313 {
1314         /* It's safe to write the exe_file pointer without exe_file_lock because
1315          * this is called during fork when the task is not yet in /proc */
1316         newmm->exe_file = get_mm_exe_file(oldmm);
1317 }
1318
1319 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1320 {
1321         struct task_struct *task;
1322         struct mm_struct *mm;
1323         struct file *exe_file;
1324
1325         task = get_proc_task(inode);
1326         if (!task)
1327                 return -ENOENT;
1328         mm = get_task_mm(task);
1329         put_task_struct(task);
1330         if (!mm)
1331                 return -ENOENT;
1332         exe_file = get_mm_exe_file(mm);
1333         mmput(mm);
1334         if (exe_file) {
1335                 *exe_path = exe_file->f_path;
1336                 path_get(&exe_file->f_path);
1337                 fput(exe_file);
1338                 return 0;
1339         } else
1340                 return -ENOENT;
1341 }
1342
1343 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1344 {
1345         struct inode *inode = dentry->d_inode;
1346         int error = -EACCES;
1347
1348         /* We don't need a base pointer in the /proc filesystem */
1349         path_put(&nd->path);
1350
1351         /* Are we allowed to snoop on the tasks file descriptors? */
1352         if (!proc_fd_access_allowed(inode))
1353                 goto out;
1354
1355         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1356         nd->last_type = LAST_BIND;
1357 out:
1358         return ERR_PTR(error);
1359 }
1360
1361 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1362 {
1363         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1364         char *pathname;
1365         int len;
1366
1367         if (!tmp)
1368                 return -ENOMEM;
1369
1370         pathname = d_path(path, tmp, PAGE_SIZE);
1371         len = PTR_ERR(pathname);
1372         if (IS_ERR(pathname))
1373                 goto out;
1374         len = tmp + PAGE_SIZE - 1 - pathname;
1375
1376         if (len > buflen)
1377                 len = buflen;
1378         if (copy_to_user(buffer, pathname, len))
1379                 len = -EFAULT;
1380  out:
1381         free_page((unsigned long)tmp);
1382         return len;
1383 }
1384
1385 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1386 {
1387         int error = -EACCES;
1388         struct inode *inode = dentry->d_inode;
1389         struct path path;
1390
1391         /* Are we allowed to snoop on the tasks file descriptors? */
1392         if (!proc_fd_access_allowed(inode))
1393                 goto out;
1394
1395         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1396         if (error)
1397                 goto out;
1398
1399         error = do_proc_readlink(&path, buffer, buflen);
1400         path_put(&path);
1401 out:
1402         return error;
1403 }
1404
1405 static const struct inode_operations proc_pid_link_inode_operations = {
1406         .readlink       = proc_pid_readlink,
1407         .follow_link    = proc_pid_follow_link,
1408         .setattr        = proc_setattr,
1409 };
1410
1411
1412 /* building an inode */
1413
1414 static int task_dumpable(struct task_struct *task)
1415 {
1416         int dumpable = 0;
1417         struct mm_struct *mm;
1418
1419         task_lock(task);
1420         mm = task->mm;
1421         if (mm)
1422                 dumpable = get_dumpable(mm);
1423         task_unlock(task);
1424         if(dumpable == 1)
1425                 return 1;
1426         return 0;
1427 }
1428
1429
1430 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1431 {
1432         struct inode * inode;
1433         struct proc_inode *ei;
1434         const struct cred *cred;
1435
1436         /* We need a new inode */
1437
1438         inode = new_inode(sb);
1439         if (!inode)
1440                 goto out;
1441
1442         /* Common stuff */
1443         ei = PROC_I(inode);
1444         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1445         inode->i_op = &proc_def_inode_operations;
1446
1447         /*
1448          * grab the reference to task.
1449          */
1450         ei->pid = get_task_pid(task, PIDTYPE_PID);
1451         if (!ei->pid)
1452                 goto out_unlock;
1453
1454         if (task_dumpable(task)) {
1455                 rcu_read_lock();
1456                 cred = __task_cred(task);
1457                 inode->i_uid = cred->euid;
1458                 inode->i_gid = cred->egid;
1459                 rcu_read_unlock();
1460         }
1461         security_task_to_inode(task, inode);
1462
1463 out:
1464         return inode;
1465
1466 out_unlock:
1467         iput(inode);
1468         return NULL;
1469 }
1470
1471 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1472 {
1473         struct inode *inode = dentry->d_inode;
1474         struct task_struct *task;
1475         const struct cred *cred;
1476
1477         generic_fillattr(inode, stat);
1478
1479         rcu_read_lock();
1480         stat->uid = 0;
1481         stat->gid = 0;
1482         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1483         if (task) {
1484                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1485                     task_dumpable(task)) {
1486                         cred = __task_cred(task);
1487                         stat->uid = cred->euid;
1488                         stat->gid = cred->egid;
1489                 }
1490         }
1491         rcu_read_unlock();
1492         return 0;
1493 }
1494
1495 /* dentry stuff */
1496
1497 /*
1498  *      Exceptional case: normally we are not allowed to unhash a busy
1499  * directory. In this case, however, we can do it - no aliasing problems
1500  * due to the way we treat inodes.
1501  *
1502  * Rewrite the inode's ownerships here because the owning task may have
1503  * performed a setuid(), etc.
1504  *
1505  * Before the /proc/pid/status file was created the only way to read
1506  * the effective uid of a /process was to stat /proc/pid.  Reading
1507  * /proc/pid/status is slow enough that procps and other packages
1508  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1509  * made this apply to all per process world readable and executable
1510  * directories.
1511  */
1512 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1513 {
1514         struct inode *inode = dentry->d_inode;
1515         struct task_struct *task = get_proc_task(inode);
1516         const struct cred *cred;
1517
1518         if (task) {
1519                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1520                     task_dumpable(task)) {
1521                         rcu_read_lock();
1522                         cred = __task_cred(task);
1523                         inode->i_uid = cred->euid;
1524                         inode->i_gid = cred->egid;
1525                         rcu_read_unlock();
1526                 } else {
1527                         inode->i_uid = 0;
1528                         inode->i_gid = 0;
1529                 }
1530                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1531                 security_task_to_inode(task, inode);
1532                 put_task_struct(task);
1533                 return 1;
1534         }
1535         d_drop(dentry);
1536         return 0;
1537 }
1538
1539 static int pid_delete_dentry(struct dentry * dentry)
1540 {
1541         /* Is the task we represent dead?
1542          * If so, then don't put the dentry on the lru list,
1543          * kill it immediately.
1544          */
1545         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1546 }
1547
1548 static struct dentry_operations pid_dentry_operations =
1549 {
1550         .d_revalidate   = pid_revalidate,
1551         .d_delete       = pid_delete_dentry,
1552 };
1553
1554 /* Lookups */
1555
1556 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1557                                 struct task_struct *, const void *);
1558
1559 /*
1560  * Fill a directory entry.
1561  *
1562  * If possible create the dcache entry and derive our inode number and
1563  * file type from dcache entry.
1564  *
1565  * Since all of the proc inode numbers are dynamically generated, the inode
1566  * numbers do not exist until the inode is cache.  This means creating the
1567  * the dcache entry in readdir is necessary to keep the inode numbers
1568  * reported by readdir in sync with the inode numbers reported
1569  * by stat.
1570  */
1571 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1572         char *name, int len,
1573         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1574 {
1575         struct dentry *child, *dir = filp->f_path.dentry;
1576         struct inode *inode;
1577         struct qstr qname;
1578         ino_t ino = 0;
1579         unsigned type = DT_UNKNOWN;
1580
1581         qname.name = name;
1582         qname.len  = len;
1583         qname.hash = full_name_hash(name, len);
1584
1585         child = d_lookup(dir, &qname);
1586         if (!child) {
1587                 struct dentry *new;
1588                 new = d_alloc(dir, &qname);
1589                 if (new) {
1590                         child = instantiate(dir->d_inode, new, task, ptr);
1591                         if (child)
1592                                 dput(new);
1593                         else
1594                                 child = new;
1595                 }
1596         }
1597         if (!child || IS_ERR(child) || !child->d_inode)
1598                 goto end_instantiate;
1599         inode = child->d_inode;
1600         if (inode) {
1601                 ino = inode->i_ino;
1602                 type = inode->i_mode >> 12;
1603         }
1604         dput(child);
1605 end_instantiate:
1606         if (!ino)
1607                 ino = find_inode_number(dir, &qname);
1608         if (!ino)
1609                 ino = 1;
1610         return filldir(dirent, name, len, filp->f_pos, ino, type);
1611 }
1612
1613 static unsigned name_to_int(struct dentry *dentry)
1614 {
1615         const char *name = dentry->d_name.name;
1616         int len = dentry->d_name.len;
1617         unsigned n = 0;
1618
1619         if (len > 1 && *name == '0')
1620                 goto out;
1621         while (len-- > 0) {
1622                 unsigned c = *name++ - '0';
1623                 if (c > 9)
1624                         goto out;
1625                 if (n >= (~0U-9)/10)
1626                         goto out;
1627                 n *= 10;
1628                 n += c;
1629         }
1630         return n;
1631 out:
1632         return ~0U;
1633 }
1634
1635 #define PROC_FDINFO_MAX 64
1636
1637 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1638 {
1639         struct task_struct *task = get_proc_task(inode);
1640         struct files_struct *files = NULL;
1641         struct file *file;
1642         int fd = proc_fd(inode);
1643
1644         if (task) {
1645                 files = get_files_struct(task);
1646                 put_task_struct(task);
1647         }
1648         if (files) {
1649                 /*
1650                  * We are not taking a ref to the file structure, so we must
1651                  * hold ->file_lock.
1652                  */
1653                 spin_lock(&files->file_lock);
1654                 file = fcheck_files(files, fd);
1655                 if (file) {
1656                         if (path) {
1657                                 *path = file->f_path;
1658                                 path_get(&file->f_path);
1659                         }
1660                         if (info)
1661                                 snprintf(info, PROC_FDINFO_MAX,
1662                                          "pos:\t%lli\n"
1663                                          "flags:\t0%o\n",
1664                                          (long long) file->f_pos,
1665                                          file->f_flags);
1666                         spin_unlock(&files->file_lock);
1667                         put_files_struct(files);
1668                         return 0;
1669                 }
1670                 spin_unlock(&files->file_lock);
1671                 put_files_struct(files);
1672         }
1673         return -ENOENT;
1674 }
1675
1676 static int proc_fd_link(struct inode *inode, struct path *path)
1677 {
1678         return proc_fd_info(inode, path, NULL);
1679 }
1680
1681 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1682 {
1683         struct inode *inode = dentry->d_inode;
1684         struct task_struct *task = get_proc_task(inode);
1685         int fd = proc_fd(inode);
1686         struct files_struct *files;
1687         const struct cred *cred;
1688
1689         if (task) {
1690                 files = get_files_struct(task);
1691                 if (files) {
1692                         rcu_read_lock();
1693                         if (fcheck_files(files, fd)) {
1694                                 rcu_read_unlock();
1695                                 put_files_struct(files);
1696                                 if (task_dumpable(task)) {
1697                                         rcu_read_lock();
1698                                         cred = __task_cred(task);
1699                                         inode->i_uid = cred->euid;
1700                                         inode->i_gid = cred->egid;
1701                                         rcu_read_unlock();
1702                                 } else {
1703                                         inode->i_uid = 0;
1704                                         inode->i_gid = 0;
1705                                 }
1706                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1707                                 security_task_to_inode(task, inode);
1708                                 put_task_struct(task);
1709                                 return 1;
1710                         }
1711                         rcu_read_unlock();
1712                         put_files_struct(files);
1713                 }
1714                 put_task_struct(task);
1715         }
1716         d_drop(dentry);
1717         return 0;
1718 }
1719
1720 static struct dentry_operations tid_fd_dentry_operations =
1721 {
1722         .d_revalidate   = tid_fd_revalidate,
1723         .d_delete       = pid_delete_dentry,
1724 };
1725
1726 static struct dentry *proc_fd_instantiate(struct inode *dir,
1727         struct dentry *dentry, struct task_struct *task, const void *ptr)
1728 {
1729         unsigned fd = *(const unsigned *)ptr;
1730         struct file *file;
1731         struct files_struct *files;
1732         struct inode *inode;
1733         struct proc_inode *ei;
1734         struct dentry *error = ERR_PTR(-ENOENT);
1735
1736         inode = proc_pid_make_inode(dir->i_sb, task);
1737         if (!inode)
1738                 goto out;
1739         ei = PROC_I(inode);
1740         ei->fd = fd;
1741         files = get_files_struct(task);
1742         if (!files)
1743                 goto out_iput;
1744         inode->i_mode = S_IFLNK;
1745
1746         /*
1747          * We are not taking a ref to the file structure, so we must
1748          * hold ->file_lock.
1749          */
1750         spin_lock(&files->file_lock);
1751         file = fcheck_files(files, fd);
1752         if (!file)
1753                 goto out_unlock;
1754         if (file->f_mode & FMODE_READ)
1755                 inode->i_mode |= S_IRUSR | S_IXUSR;
1756         if (file->f_mode & FMODE_WRITE)
1757                 inode->i_mode |= S_IWUSR | S_IXUSR;
1758         spin_unlock(&files->file_lock);
1759         put_files_struct(files);
1760
1761         inode->i_op = &proc_pid_link_inode_operations;
1762         inode->i_size = 64;
1763         ei->op.proc_get_link = proc_fd_link;
1764         dentry->d_op = &tid_fd_dentry_operations;
1765         d_add(dentry, inode);
1766         /* Close the race of the process dying before we return the dentry */
1767         if (tid_fd_revalidate(dentry, NULL))
1768                 error = NULL;
1769
1770  out:
1771         return error;
1772 out_unlock:
1773         spin_unlock(&files->file_lock);
1774         put_files_struct(files);
1775 out_iput:
1776         iput(inode);
1777         goto out;
1778 }
1779
1780 static struct dentry *proc_lookupfd_common(struct inode *dir,
1781                                            struct dentry *dentry,
1782                                            instantiate_t instantiate)
1783 {
1784         struct task_struct *task = get_proc_task(dir);
1785         unsigned fd = name_to_int(dentry);
1786         struct dentry *result = ERR_PTR(-ENOENT);
1787
1788         if (!task)
1789                 goto out_no_task;
1790         if (fd == ~0U)
1791                 goto out;
1792
1793         result = instantiate(dir, dentry, task, &fd);
1794 out:
1795         put_task_struct(task);
1796 out_no_task:
1797         return result;
1798 }
1799
1800 static int proc_readfd_common(struct file * filp, void * dirent,
1801                               filldir_t filldir, instantiate_t instantiate)
1802 {
1803         struct dentry *dentry = filp->f_path.dentry;
1804         struct inode *inode = dentry->d_inode;
1805         struct task_struct *p = get_proc_task(inode);
1806         unsigned int fd, ino;
1807         int retval;
1808         struct files_struct * files;
1809
1810         retval = -ENOENT;
1811         if (!p)
1812                 goto out_no_task;
1813         retval = 0;
1814
1815         fd = filp->f_pos;
1816         switch (fd) {
1817                 case 0:
1818                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1819                                 goto out;
1820                         filp->f_pos++;
1821                 case 1:
1822                         ino = parent_ino(dentry);
1823                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1824                                 goto out;
1825                         filp->f_pos++;
1826                 default:
1827                         files = get_files_struct(p);
1828                         if (!files)
1829                                 goto out;
1830                         rcu_read_lock();
1831                         for (fd = filp->f_pos-2;
1832                              fd < files_fdtable(files)->max_fds;
1833                              fd++, filp->f_pos++) {
1834                                 char name[PROC_NUMBUF];
1835                                 int len;
1836
1837                                 if (!fcheck_files(files, fd))
1838                                         continue;
1839                                 rcu_read_unlock();
1840
1841                                 len = snprintf(name, sizeof(name), "%d", fd);
1842                                 if (proc_fill_cache(filp, dirent, filldir,
1843                                                     name, len, instantiate,
1844                                                     p, &fd) < 0) {
1845                                         rcu_read_lock();
1846                                         break;
1847                                 }
1848                                 rcu_read_lock();
1849                         }
1850                         rcu_read_unlock();
1851                         put_files_struct(files);
1852         }
1853 out:
1854         put_task_struct(p);
1855 out_no_task:
1856         return retval;
1857 }
1858
1859 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1860                                     struct nameidata *nd)
1861 {
1862         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1863 }
1864
1865 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1866 {
1867         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1868 }
1869
1870 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1871                                       size_t len, loff_t *ppos)
1872 {
1873         char tmp[PROC_FDINFO_MAX];
1874         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1875         if (!err)
1876                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1877         return err;
1878 }
1879
1880 static const struct file_operations proc_fdinfo_file_operations = {
1881         .open           = nonseekable_open,
1882         .read           = proc_fdinfo_read,
1883 };
1884
1885 static const struct file_operations proc_fd_operations = {
1886         .read           = generic_read_dir,
1887         .readdir        = proc_readfd,
1888 };
1889
1890 /*
1891  * /proc/pid/fd needs a special permission handler so that a process can still
1892  * access /proc/self/fd after it has executed a setuid().
1893  */
1894 static int proc_fd_permission(struct inode *inode, int mask)
1895 {
1896         int rv;
1897
1898         rv = generic_permission(inode, mask, NULL);
1899         if (rv == 0)
1900                 return 0;
1901         if (task_pid(current) == proc_pid(inode))
1902                 rv = 0;
1903         return rv;
1904 }
1905
1906 /*
1907  * proc directories can do almost nothing..
1908  */
1909 static const struct inode_operations proc_fd_inode_operations = {
1910         .lookup         = proc_lookupfd,
1911         .permission     = proc_fd_permission,
1912         .setattr        = proc_setattr,
1913 };
1914
1915 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1916         struct dentry *dentry, struct task_struct *task, const void *ptr)
1917 {
1918         unsigned fd = *(unsigned *)ptr;
1919         struct inode *inode;
1920         struct proc_inode *ei;
1921         struct dentry *error = ERR_PTR(-ENOENT);
1922
1923         inode = proc_pid_make_inode(dir->i_sb, task);
1924         if (!inode)
1925                 goto out;
1926         ei = PROC_I(inode);
1927         ei->fd = fd;
1928         inode->i_mode = S_IFREG | S_IRUSR;
1929         inode->i_fop = &proc_fdinfo_file_operations;
1930         dentry->d_op = &tid_fd_dentry_operations;
1931         d_add(dentry, inode);
1932         /* Close the race of the process dying before we return the dentry */
1933         if (tid_fd_revalidate(dentry, NULL))
1934                 error = NULL;
1935
1936  out:
1937         return error;
1938 }
1939
1940 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1941                                         struct dentry *dentry,
1942                                         struct nameidata *nd)
1943 {
1944         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1945 }
1946
1947 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1948 {
1949         return proc_readfd_common(filp, dirent, filldir,
1950                                   proc_fdinfo_instantiate);
1951 }
1952
1953 static const struct file_operations proc_fdinfo_operations = {
1954         .read           = generic_read_dir,
1955         .readdir        = proc_readfdinfo,
1956 };
1957
1958 /*
1959  * proc directories can do almost nothing..
1960  */
1961 static const struct inode_operations proc_fdinfo_inode_operations = {
1962         .lookup         = proc_lookupfdinfo,
1963         .setattr        = proc_setattr,
1964 };
1965
1966
1967 static struct dentry *proc_pident_instantiate(struct inode *dir,
1968         struct dentry *dentry, struct task_struct *task, const void *ptr)
1969 {
1970         const struct pid_entry *p = ptr;
1971         struct inode *inode;
1972         struct proc_inode *ei;
1973         struct dentry *error = ERR_PTR(-EINVAL);
1974
1975         inode = proc_pid_make_inode(dir->i_sb, task);
1976         if (!inode)
1977                 goto out;
1978
1979         ei = PROC_I(inode);
1980         inode->i_mode = p->mode;
1981         if (S_ISDIR(inode->i_mode))
1982                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1983         if (p->iop)
1984                 inode->i_op = p->iop;
1985         if (p->fop)
1986                 inode->i_fop = p->fop;
1987         ei->op = p->op;
1988         dentry->d_op = &pid_dentry_operations;
1989         d_add(dentry, inode);
1990         /* Close the race of the process dying before we return the dentry */
1991         if (pid_revalidate(dentry, NULL))
1992                 error = NULL;
1993 out:
1994         return error;
1995 }
1996
1997 static struct dentry *proc_pident_lookup(struct inode *dir, 
1998                                          struct dentry *dentry,
1999                                          const struct pid_entry *ents,
2000                                          unsigned int nents)
2001 {
2002         struct dentry *error;
2003         struct task_struct *task = get_proc_task(dir);
2004         const struct pid_entry *p, *last;
2005
2006         error = ERR_PTR(-ENOENT);
2007
2008         if (!task)
2009                 goto out_no_task;
2010
2011         /*
2012          * Yes, it does not scale. And it should not. Don't add
2013          * new entries into /proc/<tgid>/ without very good reasons.
2014          */
2015         last = &ents[nents - 1];
2016         for (p = ents; p <= last; p++) {
2017                 if (p->len != dentry->d_name.len)
2018                         continue;
2019                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2020                         break;
2021         }
2022         if (p > last)
2023                 goto out;
2024
2025         error = proc_pident_instantiate(dir, dentry, task, p);
2026 out:
2027         put_task_struct(task);
2028 out_no_task:
2029         return error;
2030 }
2031
2032 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2033         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2034 {
2035         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2036                                 proc_pident_instantiate, task, p);
2037 }
2038
2039 static int proc_pident_readdir(struct file *filp,
2040                 void *dirent, filldir_t filldir,
2041                 const struct pid_entry *ents, unsigned int nents)
2042 {
2043         int i;
2044         struct dentry *dentry = filp->f_path.dentry;
2045         struct inode *inode = dentry->d_inode;
2046         struct task_struct *task = get_proc_task(inode);
2047         const struct pid_entry *p, *last;
2048         ino_t ino;
2049         int ret;
2050
2051         ret = -ENOENT;
2052         if (!task)
2053                 goto out_no_task;
2054
2055         ret = 0;
2056         i = filp->f_pos;
2057         switch (i) {
2058         case 0:
2059                 ino = inode->i_ino;
2060                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2061                         goto out;
2062                 i++;
2063                 filp->f_pos++;
2064                 /* fall through */
2065         case 1:
2066                 ino = parent_ino(dentry);
2067                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2068                         goto out;
2069                 i++;
2070                 filp->f_pos++;
2071                 /* fall through */
2072         default:
2073                 i -= 2;
2074                 if (i >= nents) {
2075                         ret = 1;
2076                         goto out;
2077                 }
2078                 p = ents + i;
2079                 last = &ents[nents - 1];
2080                 while (p <= last) {
2081                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2082                                 goto out;
2083                         filp->f_pos++;
2084                         p++;
2085                 }
2086         }
2087
2088         ret = 1;
2089 out:
2090         put_task_struct(task);
2091 out_no_task:
2092         return ret;
2093 }
2094
2095 #ifdef CONFIG_SECURITY
2096 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2097                                   size_t count, loff_t *ppos)
2098 {
2099         struct inode * inode = file->f_path.dentry->d_inode;
2100         char *p = NULL;
2101         ssize_t length;
2102         struct task_struct *task = get_proc_task(inode);
2103
2104         if (!task)
2105                 return -ESRCH;
2106
2107         length = security_getprocattr(task,
2108                                       (char*)file->f_path.dentry->d_name.name,
2109                                       &p);
2110         put_task_struct(task);
2111         if (length > 0)
2112                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2113         kfree(p);
2114         return length;
2115 }
2116
2117 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2118                                    size_t count, loff_t *ppos)
2119 {
2120         struct inode * inode = file->f_path.dentry->d_inode;
2121         char *page;
2122         ssize_t length;
2123         struct task_struct *task = get_proc_task(inode);
2124
2125         length = -ESRCH;
2126         if (!task)
2127                 goto out_no_task;
2128         if (count > PAGE_SIZE)
2129                 count = PAGE_SIZE;
2130
2131         /* No partial writes. */
2132         length = -EINVAL;
2133         if (*ppos != 0)
2134                 goto out;
2135
2136         length = -ENOMEM;
2137         page = (char*)__get_free_page(GFP_TEMPORARY);
2138         if (!page)
2139                 goto out;
2140
2141         length = -EFAULT;
2142         if (copy_from_user(page, buf, count))
2143                 goto out_free;
2144
2145         length = security_setprocattr(task,
2146                                       (char*)file->f_path.dentry->d_name.name,
2147                                       (void*)page, count);
2148 out_free:
2149         free_page((unsigned long) page);
2150 out:
2151         put_task_struct(task);
2152 out_no_task:
2153         return length;
2154 }
2155
2156 static const struct file_operations proc_pid_attr_operations = {
2157         .read           = proc_pid_attr_read,
2158         .write          = proc_pid_attr_write,
2159 };
2160
2161 static const struct pid_entry attr_dir_stuff[] = {
2162         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2163         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2164         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2165         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2166         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2167         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2168 };
2169
2170 static int proc_attr_dir_readdir(struct file * filp,
2171                              void * dirent, filldir_t filldir)
2172 {
2173         return proc_pident_readdir(filp,dirent,filldir,
2174                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2175 }
2176
2177 static const struct file_operations proc_attr_dir_operations = {
2178         .read           = generic_read_dir,
2179         .readdir        = proc_attr_dir_readdir,
2180 };
2181
2182 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2183                                 struct dentry *dentry, struct nameidata *nd)
2184 {
2185         return proc_pident_lookup(dir, dentry,
2186                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2187 }
2188
2189 static const struct inode_operations proc_attr_dir_inode_operations = {
2190         .lookup         = proc_attr_dir_lookup,
2191         .getattr        = pid_getattr,
2192         .setattr        = proc_setattr,
2193 };
2194
2195 #endif
2196
2197 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2198 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2199                                          size_t count, loff_t *ppos)
2200 {
2201         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2202         struct mm_struct *mm;
2203         char buffer[PROC_NUMBUF];
2204         size_t len;
2205         int ret;
2206
2207         if (!task)
2208                 return -ESRCH;
2209
2210         ret = 0;
2211         mm = get_task_mm(task);
2212         if (mm) {
2213                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2214                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2215                                 MMF_DUMP_FILTER_SHIFT));
2216                 mmput(mm);
2217                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2218         }
2219
2220         put_task_struct(task);
2221
2222         return ret;
2223 }
2224
2225 static ssize_t proc_coredump_filter_write(struct file *file,
2226                                           const char __user *buf,
2227                                           size_t count,
2228                                           loff_t *ppos)
2229 {
2230         struct task_struct *task;
2231         struct mm_struct *mm;
2232         char buffer[PROC_NUMBUF], *end;
2233         unsigned int val;
2234         int ret;
2235         int i;
2236         unsigned long mask;
2237
2238         ret = -EFAULT;
2239         memset(buffer, 0, sizeof(buffer));
2240         if (count > sizeof(buffer) - 1)
2241                 count = sizeof(buffer) - 1;
2242         if (copy_from_user(buffer, buf, count))
2243                 goto out_no_task;
2244
2245         ret = -EINVAL;
2246         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2247         if (*end == '\n')
2248                 end++;
2249         if (end - buffer == 0)
2250                 goto out_no_task;
2251
2252         ret = -ESRCH;
2253         task = get_proc_task(file->f_dentry->d_inode);
2254         if (!task)
2255                 goto out_no_task;
2256
2257         ret = end - buffer;
2258         mm = get_task_mm(task);
2259         if (!mm)
2260                 goto out_no_mm;
2261
2262         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2263                 if (val & mask)
2264                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2265                 else
2266                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2267         }
2268
2269         mmput(mm);
2270  out_no_mm:
2271         put_task_struct(task);
2272  out_no_task:
2273         return ret;
2274 }
2275
2276 static const struct file_operations proc_coredump_filter_operations = {
2277         .read           = proc_coredump_filter_read,
2278         .write          = proc_coredump_filter_write,
2279 };
2280 #endif
2281
2282 /*
2283  * /proc/self:
2284  */
2285 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2286                               int buflen)
2287 {
2288         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2289         pid_t tgid = task_tgid_nr_ns(current, ns);
2290         char tmp[PROC_NUMBUF];
2291         if (!tgid)
2292                 return -ENOENT;
2293         sprintf(tmp, "%d", tgid);
2294         return vfs_readlink(dentry,buffer,buflen,tmp);
2295 }
2296
2297 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2298 {
2299         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2300         pid_t tgid = task_tgid_nr_ns(current, ns);
2301         char tmp[PROC_NUMBUF];
2302         if (!tgid)
2303                 return ERR_PTR(-ENOENT);
2304         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2305         return ERR_PTR(vfs_follow_link(nd,tmp));
2306 }
2307
2308 static const struct inode_operations proc_self_inode_operations = {
2309         .readlink       = proc_self_readlink,
2310         .follow_link    = proc_self_follow_link,
2311 };
2312
2313 /*
2314  * proc base
2315  *
2316  * These are the directory entries in the root directory of /proc
2317  * that properly belong to the /proc filesystem, as they describe
2318  * describe something that is process related.
2319  */
2320 static const struct pid_entry proc_base_stuff[] = {
2321         NOD("self", S_IFLNK|S_IRWXUGO,
2322                 &proc_self_inode_operations, NULL, {}),
2323 };
2324
2325 /*
2326  *      Exceptional case: normally we are not allowed to unhash a busy
2327  * directory. In this case, however, we can do it - no aliasing problems
2328  * due to the way we treat inodes.
2329  */
2330 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2331 {
2332         struct inode *inode = dentry->d_inode;
2333         struct task_struct *task = get_proc_task(inode);
2334         if (task) {
2335                 put_task_struct(task);
2336                 return 1;
2337         }
2338         d_drop(dentry);
2339         return 0;
2340 }
2341
2342 static struct dentry_operations proc_base_dentry_operations =
2343 {
2344         .d_revalidate   = proc_base_revalidate,
2345         .d_delete       = pid_delete_dentry,
2346 };
2347
2348 static struct dentry *proc_base_instantiate(struct inode *dir,
2349         struct dentry *dentry, struct task_struct *task, const void *ptr)
2350 {
2351         const struct pid_entry *p = ptr;
2352         struct inode *inode;
2353         struct proc_inode *ei;
2354         struct dentry *error = ERR_PTR(-EINVAL);
2355
2356         /* Allocate the inode */
2357         error = ERR_PTR(-ENOMEM);
2358         inode = new_inode(dir->i_sb);
2359         if (!inode)
2360                 goto out;
2361
2362         /* Initialize the inode */
2363         ei = PROC_I(inode);
2364         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2365
2366         /*
2367          * grab the reference to the task.
2368          */
2369         ei->pid = get_task_pid(task, PIDTYPE_PID);
2370         if (!ei->pid)
2371                 goto out_iput;
2372
2373         inode->i_mode = p->mode;
2374         if (S_ISDIR(inode->i_mode))
2375                 inode->i_nlink = 2;
2376         if (S_ISLNK(inode->i_mode))
2377                 inode->i_size = 64;
2378         if (p->iop)
2379                 inode->i_op = p->iop;
2380         if (p->fop)
2381                 inode->i_fop = p->fop;
2382         ei->op = p->op;
2383         dentry->d_op = &proc_base_dentry_operations;
2384         d_add(dentry, inode);
2385         error = NULL;
2386 out:
2387         return error;
2388 out_iput:
2389         iput(inode);
2390         goto out;
2391 }
2392
2393 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2394 {
2395         struct dentry *error;
2396         struct task_struct *task = get_proc_task(dir);
2397         const struct pid_entry *p, *last;
2398
2399         error = ERR_PTR(-ENOENT);
2400
2401         if (!task)
2402                 goto out_no_task;
2403
2404         /* Lookup the directory entry */
2405         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2406         for (p = proc_base_stuff; p <= last; p++) {
2407                 if (p->len != dentry->d_name.len)
2408                         continue;
2409                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2410                         break;
2411         }
2412         if (p > last)
2413                 goto out;
2414
2415         error = proc_base_instantiate(dir, dentry, task, p);
2416
2417 out:
2418         put_task_struct(task);
2419 out_no_task:
2420         return error;
2421 }
2422
2423 static int proc_base_fill_cache(struct file *filp, void *dirent,
2424         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2425 {
2426         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2427                                 proc_base_instantiate, task, p);
2428 }
2429
2430 #ifdef CONFIG_TASK_IO_ACCOUNTING
2431 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2432 {
2433         struct task_io_accounting acct = task->ioac;
2434         unsigned long flags;
2435
2436         if (whole && lock_task_sighand(task, &flags)) {
2437                 struct task_struct *t = task;
2438
2439                 task_io_accounting_add(&acct, &task->signal->ioac);
2440                 while_each_thread(task, t)
2441                         task_io_accounting_add(&acct, &t->ioac);
2442
2443                 unlock_task_sighand(task, &flags);
2444         }
2445         return sprintf(buffer,
2446                         "rchar: %llu\n"
2447                         "wchar: %llu\n"
2448                         "syscr: %llu\n"
2449                         "syscw: %llu\n"
2450                         "read_bytes: %llu\n"
2451                         "write_bytes: %llu\n"
2452                         "cancelled_write_bytes: %llu\n",
2453                         (unsigned long long)acct.rchar,
2454                         (unsigned long long)acct.wchar,
2455                         (unsigned long long)acct.syscr,
2456                         (unsigned long long)acct.syscw,
2457                         (unsigned long long)acct.read_bytes,
2458                         (unsigned long long)acct.write_bytes,
2459                         (unsigned long long)acct.cancelled_write_bytes);
2460 }
2461
2462 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2463 {
2464         return do_io_accounting(task, buffer, 0);
2465 }
2466
2467 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2468 {
2469         return do_io_accounting(task, buffer, 1);
2470 }
2471 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2472
2473 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2474                                 struct pid *pid, struct task_struct *task)
2475 {
2476         seq_printf(m, "%08x\n", task->personality);
2477         return 0;
2478 }
2479
2480 /*
2481  * Thread groups
2482  */
2483 static const struct file_operations proc_task_operations;
2484 static const struct inode_operations proc_task_inode_operations;
2485
2486 static const struct pid_entry tgid_base_stuff[] = {
2487         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2488         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2489         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2490 #ifdef CONFIG_NET
2491         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2492 #endif
2493         REG("environ",    S_IRUSR, proc_environ_operations),
2494         INF("auxv",       S_IRUSR, proc_pid_auxv),
2495         ONE("status",     S_IRUGO, proc_pid_status),
2496         ONE("personality", S_IRUSR, proc_pid_personality),
2497         INF("limits",     S_IRUSR, proc_pid_limits),
2498 #ifdef CONFIG_SCHED_DEBUG
2499         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2500 #endif
2501 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2502         INF("syscall",    S_IRUSR, proc_pid_syscall),
2503 #endif
2504         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2505         ONE("stat",       S_IRUGO, proc_tgid_stat),
2506         ONE("statm",      S_IRUGO, proc_pid_statm),
2507         REG("maps",       S_IRUGO, proc_maps_operations),
2508 #ifdef CONFIG_NUMA
2509         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2510 #endif
2511         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2512         LNK("cwd",        proc_cwd_link),
2513         LNK("root",       proc_root_link),
2514         LNK("exe",        proc_exe_link),
2515         REG("mounts",     S_IRUGO, proc_mounts_operations),
2516         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2517         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2518 #ifdef CONFIG_PROC_PAGE_MONITOR
2519         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2520         REG("smaps",      S_IRUGO, proc_smaps_operations),
2521         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2522 #endif
2523 #ifdef CONFIG_SECURITY
2524         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2525 #endif
2526 #ifdef CONFIG_KALLSYMS
2527         INF("wchan",      S_IRUGO, proc_pid_wchan),
2528 #endif
2529 #ifdef CONFIG_STACKTRACE
2530         ONE("stack",      S_IRUSR, proc_pid_stack),
2531 #endif
2532 #ifdef CONFIG_SCHEDSTATS
2533         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2534 #endif
2535 #ifdef CONFIG_LATENCYTOP
2536         REG("latency",  S_IRUGO, proc_lstats_operations),
2537 #endif
2538 #ifdef CONFIG_PROC_PID_CPUSET
2539         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2540 #endif
2541 #ifdef CONFIG_CGROUPS
2542         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2543 #endif
2544         INF("oom_score",  S_IRUGO, proc_oom_score),
2545         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2546 #ifdef CONFIG_AUDITSYSCALL
2547         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2548         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2549 #endif
2550 #ifdef CONFIG_FAULT_INJECTION
2551         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2552 #endif
2553 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2554         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2555 #endif
2556 #ifdef CONFIG_TASK_IO_ACCOUNTING
2557         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2558 #endif
2559 };
2560
2561 static int proc_tgid_base_readdir(struct file * filp,
2562                              void * dirent, filldir_t filldir)
2563 {
2564         return proc_pident_readdir(filp,dirent,filldir,
2565                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2566 }
2567
2568 static const struct file_operations proc_tgid_base_operations = {
2569         .read           = generic_read_dir,
2570         .readdir        = proc_tgid_base_readdir,
2571 };
2572
2573 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2574         return proc_pident_lookup(dir, dentry,
2575                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2576 }
2577
2578 static const struct inode_operations proc_tgid_base_inode_operations = {
2579         .lookup         = proc_tgid_base_lookup,
2580         .getattr        = pid_getattr,
2581         .setattr        = proc_setattr,
2582 };
2583
2584 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2585 {
2586         struct dentry *dentry, *leader, *dir;
2587         char buf[PROC_NUMBUF];
2588         struct qstr name;
2589
2590         name.name = buf;
2591         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2592         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2593         if (dentry) {
2594                 if (!(current->flags & PF_EXITING))
2595                         shrink_dcache_parent(dentry);
2596                 d_drop(dentry);
2597                 dput(dentry);
2598         }
2599
2600         if (tgid == 0)
2601                 goto out;
2602
2603         name.name = buf;
2604         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2605         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2606         if (!leader)
2607                 goto out;
2608
2609         name.name = "task";
2610         name.len = strlen(name.name);
2611         dir = d_hash_and_lookup(leader, &name);
2612         if (!dir)
2613                 goto out_put_leader;
2614
2615         name.name = buf;
2616         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2617         dentry = d_hash_and_lookup(dir, &name);
2618         if (dentry) {
2619                 shrink_dcache_parent(dentry);
2620                 d_drop(dentry);
2621                 dput(dentry);
2622         }
2623
2624         dput(dir);
2625 out_put_leader:
2626         dput(leader);
2627 out:
2628         return;
2629 }
2630
2631 /**
2632  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2633  * @task: task that should be flushed.
2634  *
2635  * When flushing dentries from proc, one needs to flush them from global
2636  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2637  * in. This call is supposed to do all of this job.
2638  *
2639  * Looks in the dcache for
2640  * /proc/@pid
2641  * /proc/@tgid/task/@pid
2642  * if either directory is present flushes it and all of it'ts children
2643  * from the dcache.
2644  *
2645  * It is safe and reasonable to cache /proc entries for a task until
2646  * that task exits.  After that they just clog up the dcache with
2647  * useless entries, possibly causing useful dcache entries to be
2648  * flushed instead.  This routine is proved to flush those useless
2649  * dcache entries at process exit time.
2650  *
2651  * NOTE: This routine is just an optimization so it does not guarantee
2652  *       that no dcache entries will exist at process exit time it
2653  *       just makes it very unlikely that any will persist.
2654  */
2655
2656 void proc_flush_task(struct task_struct *task)
2657 {
2658         int i;
2659         struct pid *pid, *tgid = NULL;
2660         struct upid *upid;
2661
2662         pid = task_pid(task);
2663         if (thread_group_leader(task))
2664                 tgid = task_tgid(task);
2665
2666         for (i = 0; i <= pid->level; i++) {
2667                 upid = &pid->numbers[i];
2668                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2669                         tgid ? tgid->numbers[i].nr : 0);
2670         }
2671
2672         upid = &pid->numbers[pid->level];
2673         if (upid->nr == 1)
2674                 pid_ns_release_proc(upid->ns);
2675 }
2676
2677 static struct dentry *proc_pid_instantiate(struct inode *dir,
2678                                            struct dentry * dentry,
2679                                            struct task_struct *task, const void *ptr)
2680 {
2681         struct dentry *error = ERR_PTR(-ENOENT);
2682         struct inode *inode;
2683
2684         inode = proc_pid_make_inode(dir->i_sb, task);
2685         if (!inode)
2686                 goto out;
2687
2688         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2689         inode->i_op = &proc_tgid_base_inode_operations;
2690         inode->i_fop = &proc_tgid_base_operations;
2691         inode->i_flags|=S_IMMUTABLE;
2692
2693         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2694                 ARRAY_SIZE(tgid_base_stuff));
2695
2696         dentry->d_op = &pid_dentry_operations;
2697
2698         d_add(dentry, inode);
2699         /* Close the race of the process dying before we return the dentry */
2700         if (pid_revalidate(dentry, NULL))
2701                 error = NULL;
2702 out:
2703         return error;
2704 }
2705
2706 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2707 {
2708         struct dentry *result = ERR_PTR(-ENOENT);
2709         struct task_struct *task;
2710         unsigned tgid;
2711         struct pid_namespace *ns;
2712
2713         result = proc_base_lookup(dir, dentry);
2714         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2715                 goto out;
2716
2717         tgid = name_to_int(dentry);
2718         if (tgid == ~0U)
2719                 goto out;
2720
2721         ns = dentry->d_sb->s_fs_info;
2722         rcu_read_lock();
2723         task = find_task_by_pid_ns(tgid, ns);
2724         if (task)
2725                 get_task_struct(task);
2726         rcu_read_unlock();
2727         if (!task)
2728                 goto out;
2729
2730         result = proc_pid_instantiate(dir, dentry, task, NULL);
2731         put_task_struct(task);
2732 out:
2733         return result;
2734 }
2735
2736 /*
2737  * Find the first task with tgid >= tgid
2738  *
2739  */
2740 struct tgid_iter {
2741         unsigned int tgid;
2742         struct task_struct *task;
2743 };
2744 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2745 {
2746         struct pid *pid;
2747
2748         if (iter.task)
2749                 put_task_struct(iter.task);
2750         rcu_read_lock();
2751 retry:
2752         iter.task = NULL;
2753         pid = find_ge_pid(iter.tgid, ns);
2754         if (pid) {
2755                 iter.tgid = pid_nr_ns(pid, ns);
2756                 iter.task = pid_task(pid, PIDTYPE_PID);
2757                 /* What we to know is if the pid we have find is the
2758                  * pid of a thread_group_leader.  Testing for task
2759                  * being a thread_group_leader is the obvious thing
2760                  * todo but there is a window when it fails, due to
2761                  * the pid transfer logic in de_thread.
2762                  *
2763                  * So we perform the straight forward test of seeing
2764                  * if the pid we have found is the pid of a thread
2765                  * group leader, and don't worry if the task we have
2766                  * found doesn't happen to be a thread group leader.
2767                  * As we don't care in the case of readdir.
2768                  */
2769                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2770                         iter.tgid += 1;
2771                         goto retry;
2772                 }
2773                 get_task_struct(iter.task);
2774         }
2775         rcu_read_unlock();
2776         return iter;
2777 }
2778
2779 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2780
2781 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2782         struct tgid_iter iter)
2783 {
2784         char name[PROC_NUMBUF];
2785         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2786         return proc_fill_cache(filp, dirent, filldir, name, len,
2787                                 proc_pid_instantiate, iter.task, NULL);
2788 }
2789
2790 /* for the /proc/ directory itself, after non-process stuff has been done */
2791 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2792 {
2793         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2794         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2795         struct tgid_iter iter;
2796         struct pid_namespace *ns;
2797
2798         if (!reaper)
2799                 goto out_no_task;
2800
2801         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2802                 const struct pid_entry *p = &proc_base_stuff[nr];
2803                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2804                         goto out;
2805         }
2806
2807         ns = filp->f_dentry->d_sb->s_fs_info;
2808         iter.task = NULL;
2809         iter.tgid = filp->f_pos - TGID_OFFSET;
2810         for (iter = next_tgid(ns, iter);
2811              iter.task;
2812              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2813                 filp->f_pos = iter.tgid + TGID_OFFSET;
2814                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2815                         put_task_struct(iter.task);
2816                         goto out;
2817                 }
2818         }
2819         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2820 out:
2821         put_task_struct(reaper);
2822 out_no_task:
2823         return 0;
2824 }
2825
2826 /*
2827  * Tasks
2828  */
2829 static const struct pid_entry tid_base_stuff[] = {
2830         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2831         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2832         REG("environ",   S_IRUSR, proc_environ_operations),
2833         INF("auxv",      S_IRUSR, proc_pid_auxv),
2834         ONE("status",    S_IRUGO, proc_pid_status),
2835         ONE("personality", S_IRUSR, proc_pid_personality),
2836         INF("limits",    S_IRUSR, proc_pid_limits),
2837 #ifdef CONFIG_SCHED_DEBUG
2838         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2839 #endif
2840 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2841         INF("syscall",   S_IRUSR, proc_pid_syscall),
2842 #endif
2843         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2844         ONE("stat",      S_IRUGO, proc_tid_stat),
2845         ONE("statm",     S_IRUGO, proc_pid_statm),
2846         REG("maps",      S_IRUGO, proc_maps_operations),
2847 #ifdef CONFIG_NUMA
2848         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2849 #endif
2850         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2851         LNK("cwd",       proc_cwd_link),
2852         LNK("root",      proc_root_link),
2853         LNK("exe",       proc_exe_link),
2854         REG("mounts",    S_IRUGO, proc_mounts_operations),
2855         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2856 #ifdef CONFIG_PROC_PAGE_MONITOR
2857         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2858         REG("smaps",     S_IRUGO, proc_smaps_operations),
2859         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2860 #endif
2861 #ifdef CONFIG_SECURITY
2862         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2863 #endif
2864 #ifdef CONFIG_KALLSYMS
2865         INF("wchan",     S_IRUGO, proc_pid_wchan),
2866 #endif
2867 #ifdef CONFIG_STACKTRACE
2868         ONE("stack",      S_IRUSR, proc_pid_stack),
2869 #endif
2870 #ifdef CONFIG_SCHEDSTATS
2871         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2872 #endif
2873 #ifdef CONFIG_LATENCYTOP
2874         REG("latency",  S_IRUGO, proc_lstats_operations),
2875 #endif
2876 #ifdef CONFIG_PROC_PID_CPUSET
2877         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2878 #endif
2879 #ifdef CONFIG_CGROUPS
2880         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2881 #endif
2882         INF("oom_score", S_IRUGO, proc_oom_score),
2883         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2884 #ifdef CONFIG_AUDITSYSCALL
2885         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2886         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2887 #endif
2888 #ifdef CONFIG_FAULT_INJECTION
2889         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2890 #endif
2891 #ifdef CONFIG_TASK_IO_ACCOUNTING
2892         INF("io",       S_IRUGO, proc_tid_io_accounting),
2893 #endif
2894 };
2895
2896 static int proc_tid_base_readdir(struct file * filp,
2897                              void * dirent, filldir_t filldir)
2898 {
2899         return proc_pident_readdir(filp,dirent,filldir,
2900                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2901 }
2902
2903 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2904         return proc_pident_lookup(dir, dentry,
2905                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2906 }
2907
2908 static const struct file_operations proc_tid_base_operations = {
2909         .read           = generic_read_dir,
2910         .readdir        = proc_tid_base_readdir,
2911 };
2912
2913 static const struct inode_operations proc_tid_base_inode_operations = {
2914         .lookup         = proc_tid_base_lookup,
2915         .getattr        = pid_getattr,
2916         .setattr        = proc_setattr,
2917 };
2918
2919 static struct dentry *proc_task_instantiate(struct inode *dir,
2920         struct dentry *dentry, struct task_struct *task, const void *ptr)
2921 {
2922         struct dentry *error = ERR_PTR(-ENOENT);
2923         struct inode *inode;
2924         inode = proc_pid_make_inode(dir->i_sb, task);
2925
2926         if (!inode)
2927                 goto out;
2928         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2929         inode->i_op = &proc_tid_base_inode_operations;
2930         inode->i_fop = &proc_tid_base_operations;
2931         inode->i_flags|=S_IMMUTABLE;
2932
2933         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2934                 ARRAY_SIZE(tid_base_stuff));
2935
2936         dentry->d_op = &pid_dentry_operations;
2937
2938         d_add(dentry, inode);
2939         /* Close the race of the process dying before we return the dentry */
2940         if (pid_revalidate(dentry, NULL))
2941                 error = NULL;
2942 out:
2943         return error;
2944 }
2945
2946 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2947 {
2948         struct dentry *result = ERR_PTR(-ENOENT);
2949         struct task_struct *task;
2950         struct task_struct *leader = get_proc_task(dir);
2951         unsigned tid;
2952         struct pid_namespace *ns;
2953
2954         if (!leader)
2955                 goto out_no_task;
2956
2957         tid = name_to_int(dentry);
2958         if (tid == ~0U)
2959                 goto out;
2960
2961         ns = dentry->d_sb->s_fs_info;
2962         rcu_read_lock();
2963         task = find_task_by_pid_ns(tid, ns);
2964         if (task)
2965                 get_task_struct(task);
2966         rcu_read_unlock();
2967         if (!task)
2968                 goto out;
2969         if (!same_thread_group(leader, task))
2970                 goto out_drop_task;
2971
2972         result = proc_task_instantiate(dir, dentry, task, NULL);
2973 out_drop_task:
2974         put_task_struct(task);
2975 out:
2976         put_task_struct(leader);
2977 out_no_task:
2978         return result;
2979 }
2980
2981 /*
2982  * Find the first tid of a thread group to return to user space.
2983  *
2984  * Usually this is just the thread group leader, but if the users
2985  * buffer was too small or there was a seek into the middle of the
2986  * directory we have more work todo.
2987  *
2988  * In the case of a short read we start with find_task_by_pid.
2989  *
2990  * In the case of a seek we start with the leader and walk nr
2991  * threads past it.
2992  */
2993 static struct task_struct *first_tid(struct task_struct *leader,
2994                 int tid, int nr, struct pid_namespace *ns)
2995 {
2996         struct task_struct *pos;
2997
2998         rcu_read_lock();
2999         /* Attempt to start with the pid of a thread */
3000         if (tid && (nr > 0)) {
3001                 pos = find_task_by_pid_ns(tid, ns);
3002                 if (pos && (pos->group_leader == leader))
3003                         goto found;
3004         }
3005
3006         /* If nr exceeds the number of threads there is nothing todo */
3007         pos = NULL;
3008         if (nr && nr >= get_nr_threads(leader))
3009                 goto out;
3010
3011         /* If we haven't found our starting place yet start
3012          * with the leader and walk nr threads forward.
3013          */
3014         for (pos = leader; nr > 0; --nr) {
3015                 pos = next_thread(pos);
3016                 if (pos == leader) {
3017                         pos = NULL;
3018                         goto out;
3019                 }
3020         }
3021 found:
3022         get_task_struct(pos);
3023 out:
3024         rcu_read_unlock();
3025         return pos;
3026 }
3027
3028 /*
3029  * Find the next thread in the thread list.
3030  * Return NULL if there is an error or no next thread.
3031  *
3032  * The reference to the input task_struct is released.
3033  */
3034 static struct task_struct *next_tid(struct task_struct *start)
3035 {
3036         struct task_struct *pos = NULL;
3037         rcu_read_lock();
3038         if (pid_alive(start)) {
3039                 pos = next_thread(start);
3040                 if (thread_group_leader(pos))
3041                         pos = NULL;
3042                 else
3043                         get_task_struct(pos);
3044         }
3045         rcu_read_unlock();
3046         put_task_struct(start);
3047         return pos;
3048 }
3049
3050 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3051         struct task_struct *task, int tid)
3052 {
3053         char name[PROC_NUMBUF];
3054         int len = snprintf(name, sizeof(name), "%d", tid);
3055         return proc_fill_cache(filp, dirent, filldir, name, len,
3056                                 proc_task_instantiate, task, NULL);
3057 }
3058
3059 /* for the /proc/TGID/task/ directories */
3060 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3061 {
3062         struct dentry *dentry = filp->f_path.dentry;
3063         struct inode *inode = dentry->d_inode;
3064         struct task_struct *leader = NULL;
3065         struct task_struct *task;
3066         int retval = -ENOENT;
3067         ino_t ino;
3068         int tid;
3069         struct pid_namespace *ns;
3070
3071         task = get_proc_task(inode);
3072         if (!task)
3073                 goto out_no_task;
3074         rcu_read_lock();
3075         if (pid_alive(task)) {
3076                 leader = task->group_leader;
3077                 get_task_struct(leader);
3078         }
3079         rcu_read_unlock();
3080         put_task_struct(task);
3081         if (!leader)
3082                 goto out_no_task;
3083         retval = 0;
3084
3085         switch (filp->f_pos) {
3086         case 0:
3087                 ino = inode->i_ino;
3088                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3089                         goto out;
3090                 filp->f_pos++;
3091                 /* fall through */
3092         case 1:
3093                 ino = parent_ino(dentry);
3094                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3095                         goto out;
3096                 filp->f_pos++;
3097                 /* fall through */
3098         }
3099
3100         /* f_version caches the tgid value that the last readdir call couldn't
3101          * return. lseek aka telldir automagically resets f_version to 0.
3102          */
3103         ns = filp->f_dentry->d_sb->s_fs_info;
3104         tid = (int)filp->f_version;
3105         filp->f_version = 0;
3106         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3107              task;
3108              task = next_tid(task), filp->f_pos++) {
3109                 tid = task_pid_nr_ns(task, ns);
3110                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3111                         /* returning this tgid failed, save it as the first
3112                          * pid for the next readir call */
3113                         filp->f_version = (u64)tid;
3114                         put_task_struct(task);
3115                         break;
3116                 }
3117         }
3118 out:
3119         put_task_struct(leader);
3120 out_no_task:
3121         return retval;
3122 }
3123
3124 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3125 {
3126         struct inode *inode = dentry->d_inode;
3127         struct task_struct *p = get_proc_task(inode);
3128         generic_fillattr(inode, stat);
3129
3130         if (p) {
3131                 stat->nlink += get_nr_threads(p);
3132                 put_task_struct(p);
3133         }
3134
3135         return 0;
3136 }
3137
3138 static const struct inode_operations proc_task_inode_operations = {
3139         .lookup         = proc_task_lookup,
3140         .getattr        = proc_task_getattr,
3141         .setattr        = proc_setattr,
3142 };
3143
3144 static const struct file_operations proc_task_operations = {
3145         .read           = generic_read_dir,
3146         .readdir        = proc_task_readdir,
3147 };