[PATCH] Fix invisible threads problem
[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/config.h>
53 #include <linux/errno.h>
54 #include <linux/time.h>
55 #include <linux/proc_fs.h>
56 #include <linux/stat.h>
57 #include <linux/init.h>
58 #include <linux/file.h>
59 #include <linux/string.h>
60 #include <linux/seq_file.h>
61 #include <linux/namei.h>
62 #include <linux/namespace.h>
63 #include <linux/mm.h>
64 #include <linux/smp_lock.h>
65 #include <linux/rcupdate.h>
66 #include <linux/kallsyms.h>
67 #include <linux/mount.h>
68 #include <linux/security.h>
69 #include <linux/ptrace.h>
70 #include <linux/seccomp.h>
71 #include <linux/cpuset.h>
72 #include <linux/audit.h>
73 #include "internal.h"
74
75 /*
76  * For hysterical raisins we keep the same inumbers as in the old procfs.
77  * Feel free to change the macro below - just keep the range distinct from
78  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
79  * As soon as we'll get a separate superblock we will be able to forget
80  * about magical ranges too.
81  */
82
83 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
84
85 enum pid_directory_inos {
86         PROC_TGID_INO = 2,
87         PROC_TGID_TASK,
88         PROC_TGID_STATUS,
89         PROC_TGID_MEM,
90 #ifdef CONFIG_SECCOMP
91         PROC_TGID_SECCOMP,
92 #endif
93         PROC_TGID_CWD,
94         PROC_TGID_ROOT,
95         PROC_TGID_EXE,
96         PROC_TGID_FD,
97         PROC_TGID_ENVIRON,
98         PROC_TGID_AUXV,
99         PROC_TGID_CMDLINE,
100         PROC_TGID_STAT,
101         PROC_TGID_STATM,
102         PROC_TGID_MAPS,
103         PROC_TGID_NUMA_MAPS,
104         PROC_TGID_MOUNTS,
105         PROC_TGID_WCHAN,
106         PROC_TGID_SMAPS,
107 #ifdef CONFIG_SCHEDSTATS
108         PROC_TGID_SCHEDSTAT,
109 #endif
110 #ifdef CONFIG_CPUSETS
111         PROC_TGID_CPUSET,
112 #endif
113 #ifdef CONFIG_SECURITY
114         PROC_TGID_ATTR,
115         PROC_TGID_ATTR_CURRENT,
116         PROC_TGID_ATTR_PREV,
117         PROC_TGID_ATTR_EXEC,
118         PROC_TGID_ATTR_FSCREATE,
119 #endif
120 #ifdef CONFIG_AUDITSYSCALL
121         PROC_TGID_LOGINUID,
122 #endif
123         PROC_TGID_OOM_SCORE,
124         PROC_TGID_OOM_ADJUST,
125         PROC_TID_INO,
126         PROC_TID_STATUS,
127         PROC_TID_MEM,
128 #ifdef CONFIG_SECCOMP
129         PROC_TID_SECCOMP,
130 #endif
131         PROC_TID_CWD,
132         PROC_TID_ROOT,
133         PROC_TID_EXE,
134         PROC_TID_FD,
135         PROC_TID_ENVIRON,
136         PROC_TID_AUXV,
137         PROC_TID_CMDLINE,
138         PROC_TID_STAT,
139         PROC_TID_STATM,
140         PROC_TID_MAPS,
141         PROC_TID_NUMA_MAPS,
142         PROC_TID_MOUNTS,
143         PROC_TID_WCHAN,
144         PROC_TID_SMAPS,
145 #ifdef CONFIG_SCHEDSTATS
146         PROC_TID_SCHEDSTAT,
147 #endif
148 #ifdef CONFIG_CPUSETS
149         PROC_TID_CPUSET,
150 #endif
151 #ifdef CONFIG_SECURITY
152         PROC_TID_ATTR,
153         PROC_TID_ATTR_CURRENT,
154         PROC_TID_ATTR_PREV,
155         PROC_TID_ATTR_EXEC,
156         PROC_TID_ATTR_FSCREATE,
157 #endif
158 #ifdef CONFIG_AUDITSYSCALL
159         PROC_TID_LOGINUID,
160 #endif
161         PROC_TID_OOM_SCORE,
162         PROC_TID_OOM_ADJUST,
163
164         /* Add new entries before this */
165         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
166 };
167
168 struct pid_entry {
169         int type;
170         int len;
171         char *name;
172         mode_t mode;
173 };
174
175 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
176
177 static struct pid_entry tgid_base_stuff[] = {
178         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
179         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
180         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
181         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
182         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
183         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
184         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
185         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
186         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
187 #ifdef CONFIG_NUMA
188         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
189 #endif
190         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
191 #ifdef CONFIG_SECCOMP
192         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
193 #endif
194         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
195         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
196         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
197         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
198         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
199 #ifdef CONFIG_SECURITY
200         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
201 #endif
202 #ifdef CONFIG_KALLSYMS
203         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
204 #endif
205 #ifdef CONFIG_SCHEDSTATS
206         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
207 #endif
208 #ifdef CONFIG_CPUSETS
209         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
210 #endif
211         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
212         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
213 #ifdef CONFIG_AUDITSYSCALL
214         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
215 #endif
216         {0,0,NULL,0}
217 };
218 static struct pid_entry tid_base_stuff[] = {
219         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
220         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
221         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
222         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
223         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
224         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
225         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
226         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
227 #ifdef CONFIG_NUMA
228         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
229 #endif
230         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
231 #ifdef CONFIG_SECCOMP
232         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
233 #endif
234         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
235         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
236         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
237         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
238         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
239 #ifdef CONFIG_SECURITY
240         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
241 #endif
242 #ifdef CONFIG_KALLSYMS
243         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
244 #endif
245 #ifdef CONFIG_SCHEDSTATS
246         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
247 #endif
248 #ifdef CONFIG_CPUSETS
249         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
250 #endif
251         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
252         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
253 #ifdef CONFIG_AUDITSYSCALL
254         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
255 #endif
256         {0,0,NULL,0}
257 };
258
259 #ifdef CONFIG_SECURITY
260 static struct pid_entry tgid_attr_stuff[] = {
261         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
262         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
263         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
264         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
265         {0,0,NULL,0}
266 };
267 static struct pid_entry tid_attr_stuff[] = {
268         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
269         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
270         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
271         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
272         {0,0,NULL,0}
273 };
274 #endif
275
276 #undef E
277
278 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
279 {
280         struct task_struct *task = proc_task(inode);
281         struct files_struct *files;
282         struct file *file;
283         int fd = proc_type(inode) - PROC_TID_FD_DIR;
284
285         files = get_files_struct(task);
286         if (files) {
287                 rcu_read_lock();
288                 file = fcheck_files(files, fd);
289                 if (file) {
290                         *mnt = mntget(file->f_vfsmnt);
291                         *dentry = dget(file->f_dentry);
292                         rcu_read_unlock();
293                         put_files_struct(files);
294                         return 0;
295                 }
296                 rcu_read_unlock();
297                 put_files_struct(files);
298         }
299         return -ENOENT;
300 }
301
302 static struct fs_struct *get_fs_struct(struct task_struct *task)
303 {
304         struct fs_struct *fs;
305         task_lock(task);
306         fs = task->fs;
307         if(fs)
308                 atomic_inc(&fs->count);
309         task_unlock(task);
310         return fs;
311 }
312
313 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
314 {
315         struct fs_struct *fs = get_fs_struct(proc_task(inode));
316         int result = -ENOENT;
317         if (fs) {
318                 read_lock(&fs->lock);
319                 *mnt = mntget(fs->pwdmnt);
320                 *dentry = dget(fs->pwd);
321                 read_unlock(&fs->lock);
322                 result = 0;
323                 put_fs_struct(fs);
324         }
325         return result;
326 }
327
328 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
329 {
330         struct fs_struct *fs = get_fs_struct(proc_task(inode));
331         int result = -ENOENT;
332         if (fs) {
333                 read_lock(&fs->lock);
334                 *mnt = mntget(fs->rootmnt);
335                 *dentry = dget(fs->root);
336                 read_unlock(&fs->lock);
337                 result = 0;
338                 put_fs_struct(fs);
339         }
340         return result;
341 }
342
343
344 /* Same as proc_root_link, but this addionally tries to get fs from other
345  * threads in the group */
346 static int proc_task_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
347 {
348         struct fs_struct *fs;
349         int result = -ENOENT;
350         struct task_struct *leader = proc_task(inode);
351
352         task_lock(leader);
353         fs = leader->fs;
354         if (fs) {
355                 atomic_inc(&fs->count);
356                 task_unlock(leader);
357         } else {
358                 /* Try to get fs from other threads */
359                 task_unlock(leader);
360                 struct task_struct *task = leader;
361                 read_lock(&tasklist_lock);
362                 if (pid_alive(task)) {
363                         while ((task = next_thread(task)) != leader) {
364                                 task_lock(task);
365                                 fs = task->fs;
366                                 if (fs) {
367                                         atomic_inc(&fs->count);
368                                         task_unlock(task);
369                                         break;
370                                 }
371                                 task_unlock(task);
372                         }
373                 }
374                 read_unlock(&tasklist_lock);
375         }
376
377         if (fs) {
378                 read_lock(&fs->lock);
379                 *mnt = mntget(fs->rootmnt);
380                 *dentry = dget(fs->root);
381                 read_unlock(&fs->lock);
382                 result = 0;
383                 put_fs_struct(fs);
384         }
385         return result;
386 }
387
388
389 #define MAY_PTRACE(task) \
390         (task == current || \
391         (task->parent == current && \
392         (task->ptrace & PT_PTRACED) && \
393          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
394          security_ptrace(current,task) == 0))
395
396 static int proc_pid_environ(struct task_struct *task, char * buffer)
397 {
398         int res = 0;
399         struct mm_struct *mm = get_task_mm(task);
400         if (mm) {
401                 unsigned int len = mm->env_end - mm->env_start;
402                 if (len > PAGE_SIZE)
403                         len = PAGE_SIZE;
404                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
405                 if (!ptrace_may_attach(task))
406                         res = -ESRCH;
407                 mmput(mm);
408         }
409         return res;
410 }
411
412 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
413 {
414         int res = 0;
415         unsigned int len;
416         struct mm_struct *mm = get_task_mm(task);
417         if (!mm)
418                 goto out;
419         if (!mm->arg_end)
420                 goto out_mm;    /* Shh! No looking before we're done */
421
422         len = mm->arg_end - mm->arg_start;
423  
424         if (len > PAGE_SIZE)
425                 len = PAGE_SIZE;
426  
427         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
428
429         // If the nul at the end of args has been overwritten, then
430         // assume application is using setproctitle(3).
431         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
432                 len = strnlen(buffer, res);
433                 if (len < res) {
434                     res = len;
435                 } else {
436                         len = mm->env_end - mm->env_start;
437                         if (len > PAGE_SIZE - res)
438                                 len = PAGE_SIZE - res;
439                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
440                         res = strnlen(buffer, res);
441                 }
442         }
443 out_mm:
444         mmput(mm);
445 out:
446         return res;
447 }
448
449 static int proc_pid_auxv(struct task_struct *task, char *buffer)
450 {
451         int res = 0;
452         struct mm_struct *mm = get_task_mm(task);
453         if (mm) {
454                 unsigned int nwords = 0;
455                 do
456                         nwords += 2;
457                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
458                 res = nwords * sizeof(mm->saved_auxv[0]);
459                 if (res > PAGE_SIZE)
460                         res = PAGE_SIZE;
461                 memcpy(buffer, mm->saved_auxv, res);
462                 mmput(mm);
463         }
464         return res;
465 }
466
467
468 #ifdef CONFIG_KALLSYMS
469 /*
470  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
471  * Returns the resolved symbol.  If that fails, simply return the address.
472  */
473 static int proc_pid_wchan(struct task_struct *task, char *buffer)
474 {
475         char *modname;
476         const char *sym_name;
477         unsigned long wchan, size, offset;
478         char namebuf[KSYM_NAME_LEN+1];
479
480         wchan = get_wchan(task);
481
482         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
483         if (sym_name)
484                 return sprintf(buffer, "%s", sym_name);
485         return sprintf(buffer, "%lu", wchan);
486 }
487 #endif /* CONFIG_KALLSYMS */
488
489 #ifdef CONFIG_SCHEDSTATS
490 /*
491  * Provides /proc/PID/schedstat
492  */
493 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
494 {
495         return sprintf(buffer, "%lu %lu %lu\n",
496                         task->sched_info.cpu_time,
497                         task->sched_info.run_delay,
498                         task->sched_info.pcnt);
499 }
500 #endif
501
502 /* The badness from the OOM killer */
503 unsigned long badness(struct task_struct *p, unsigned long uptime);
504 static int proc_oom_score(struct task_struct *task, char *buffer)
505 {
506         unsigned long points;
507         struct timespec uptime;
508
509         do_posix_clock_monotonic_gettime(&uptime);
510         points = badness(task, uptime.tv_sec);
511         return sprintf(buffer, "%lu\n", points);
512 }
513
514 /************************************************************************/
515 /*                       Here the fs part begins                        */
516 /************************************************************************/
517
518 /* permission checks */
519
520 /* If the process being read is separated by chroot from the reading process,
521  * don't let the reader access the threads.
522  */
523 static int proc_check_chroot(struct dentry *root, struct vfsmount *vfsmnt)
524 {
525         struct dentry *de, *base;
526         struct vfsmount *our_vfsmnt, *mnt;
527         int res = 0;
528         read_lock(&current->fs->lock);
529         our_vfsmnt = mntget(current->fs->rootmnt);
530         base = dget(current->fs->root);
531         read_unlock(&current->fs->lock);
532
533         spin_lock(&vfsmount_lock);
534         de = root;
535         mnt = vfsmnt;
536
537         while (vfsmnt != our_vfsmnt) {
538                 if (vfsmnt == vfsmnt->mnt_parent)
539                         goto out;
540                 de = vfsmnt->mnt_mountpoint;
541                 vfsmnt = vfsmnt->mnt_parent;
542         }
543
544         if (!is_subdir(de, base))
545                 goto out;
546         spin_unlock(&vfsmount_lock);
547
548 exit:
549         dput(base);
550         mntput(our_vfsmnt);
551         dput(root);
552         mntput(mnt);
553         return res;
554 out:
555         spin_unlock(&vfsmount_lock);
556         res = -EACCES;
557         goto exit;
558 }
559
560 static int proc_check_root(struct inode *inode)
561 {
562         struct dentry *root;
563         struct vfsmount *vfsmnt;
564
565         if (proc_root_link(inode, &root, &vfsmnt)) /* Ewww... */
566                 return -ENOENT;
567         return proc_check_chroot(root, vfsmnt);
568 }
569
570 static int proc_permission(struct inode *inode, int mask, struct nameidata *nd)
571 {
572         if (generic_permission(inode, mask, NULL) != 0)
573                 return -EACCES;
574         return proc_check_root(inode);
575 }
576
577 static int proc_task_permission(struct inode *inode, int mask, struct nameidata *nd)
578 {
579         struct dentry *root;
580         struct vfsmount *vfsmnt;
581
582         if (generic_permission(inode, mask, NULL) != 0)
583                 return -EACCES;
584
585         if (proc_task_root_link(inode, &root, &vfsmnt))
586                 return -ENOENT;
587
588         return proc_check_chroot(root, vfsmnt);
589 }
590
591 extern struct seq_operations proc_pid_maps_op;
592 static int maps_open(struct inode *inode, struct file *file)
593 {
594         struct task_struct *task = proc_task(inode);
595         int ret = seq_open(file, &proc_pid_maps_op);
596         if (!ret) {
597                 struct seq_file *m = file->private_data;
598                 m->private = task;
599         }
600         return ret;
601 }
602
603 static struct file_operations proc_maps_operations = {
604         .open           = maps_open,
605         .read           = seq_read,
606         .llseek         = seq_lseek,
607         .release        = seq_release,
608 };
609
610 #ifdef CONFIG_NUMA
611 extern struct seq_operations proc_pid_numa_maps_op;
612 static int numa_maps_open(struct inode *inode, struct file *file)
613 {
614         struct task_struct *task = proc_task(inode);
615         int ret = seq_open(file, &proc_pid_numa_maps_op);
616         if (!ret) {
617                 struct seq_file *m = file->private_data;
618                 m->private = task;
619         }
620         return ret;
621 }
622
623 static struct file_operations proc_numa_maps_operations = {
624         .open           = numa_maps_open,
625         .read           = seq_read,
626         .llseek         = seq_lseek,
627         .release        = seq_release,
628 };
629 #endif
630
631 extern struct seq_operations proc_pid_smaps_op;
632 static int smaps_open(struct inode *inode, struct file *file)
633 {
634         struct task_struct *task = proc_task(inode);
635         int ret = seq_open(file, &proc_pid_smaps_op);
636         if (!ret) {
637                 struct seq_file *m = file->private_data;
638                 m->private = task;
639         }
640         return ret;
641 }
642
643 static struct file_operations proc_smaps_operations = {
644         .open           = smaps_open,
645         .read           = seq_read,
646         .llseek         = seq_lseek,
647         .release        = seq_release,
648 };
649
650 extern struct seq_operations mounts_op;
651 static int mounts_open(struct inode *inode, struct file *file)
652 {
653         struct task_struct *task = proc_task(inode);
654         int ret = seq_open(file, &mounts_op);
655
656         if (!ret) {
657                 struct seq_file *m = file->private_data;
658                 struct namespace *namespace;
659                 task_lock(task);
660                 namespace = task->namespace;
661                 if (namespace)
662                         get_namespace(namespace);
663                 task_unlock(task);
664
665                 if (namespace)
666                         m->private = namespace;
667                 else {
668                         seq_release(inode, file);
669                         ret = -EINVAL;
670                 }
671         }
672         return ret;
673 }
674
675 static int mounts_release(struct inode *inode, struct file *file)
676 {
677         struct seq_file *m = file->private_data;
678         struct namespace *namespace = m->private;
679         put_namespace(namespace);
680         return seq_release(inode, file);
681 }
682
683 static struct file_operations proc_mounts_operations = {
684         .open           = mounts_open,
685         .read           = seq_read,
686         .llseek         = seq_lseek,
687         .release        = mounts_release,
688 };
689
690 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
691
692 static ssize_t proc_info_read(struct file * file, char __user * buf,
693                           size_t count, loff_t *ppos)
694 {
695         struct inode * inode = file->f_dentry->d_inode;
696         unsigned long page;
697         ssize_t length;
698         struct task_struct *task = proc_task(inode);
699
700         if (count > PROC_BLOCK_SIZE)
701                 count = PROC_BLOCK_SIZE;
702         if (!(page = __get_free_page(GFP_KERNEL)))
703                 return -ENOMEM;
704
705         length = PROC_I(inode)->op.proc_read(task, (char*)page);
706
707         if (length >= 0)
708                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
709         free_page(page);
710         return length;
711 }
712
713 static struct file_operations proc_info_file_operations = {
714         .read           = proc_info_read,
715 };
716
717 static int mem_open(struct inode* inode, struct file* file)
718 {
719         file->private_data = (void*)((long)current->self_exec_id);
720         return 0;
721 }
722
723 static ssize_t mem_read(struct file * file, char __user * buf,
724                         size_t count, loff_t *ppos)
725 {
726         struct task_struct *task = proc_task(file->f_dentry->d_inode);
727         char *page;
728         unsigned long src = *ppos;
729         int ret = -ESRCH;
730         struct mm_struct *mm;
731
732         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
733                 goto out;
734
735         ret = -ENOMEM;
736         page = (char *)__get_free_page(GFP_USER);
737         if (!page)
738                 goto out;
739
740         ret = 0;
741  
742         mm = get_task_mm(task);
743         if (!mm)
744                 goto out_free;
745
746         ret = -EIO;
747  
748         if (file->private_data != (void*)((long)current->self_exec_id))
749                 goto out_put;
750
751         ret = 0;
752  
753         while (count > 0) {
754                 int this_len, retval;
755
756                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
757                 retval = access_process_vm(task, src, page, this_len, 0);
758                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
759                         if (!ret)
760                                 ret = -EIO;
761                         break;
762                 }
763
764                 if (copy_to_user(buf, page, retval)) {
765                         ret = -EFAULT;
766                         break;
767                 }
768  
769                 ret += retval;
770                 src += retval;
771                 buf += retval;
772                 count -= retval;
773         }
774         *ppos = src;
775
776 out_put:
777         mmput(mm);
778 out_free:
779         free_page((unsigned long) page);
780 out:
781         return ret;
782 }
783
784 #define mem_write NULL
785
786 #ifndef mem_write
787 /* This is a security hazard */
788 static ssize_t mem_write(struct file * file, const char * buf,
789                          size_t count, loff_t *ppos)
790 {
791         int copied = 0;
792         char *page;
793         struct task_struct *task = proc_task(file->f_dentry->d_inode);
794         unsigned long dst = *ppos;
795
796         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
797                 return -ESRCH;
798
799         page = (char *)__get_free_page(GFP_USER);
800         if (!page)
801                 return -ENOMEM;
802
803         while (count > 0) {
804                 int this_len, retval;
805
806                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
807                 if (copy_from_user(page, buf, this_len)) {
808                         copied = -EFAULT;
809                         break;
810                 }
811                 retval = access_process_vm(task, dst, page, this_len, 1);
812                 if (!retval) {
813                         if (!copied)
814                                 copied = -EIO;
815                         break;
816                 }
817                 copied += retval;
818                 buf += retval;
819                 dst += retval;
820                 count -= retval;                        
821         }
822         *ppos = dst;
823         free_page((unsigned long) page);
824         return copied;
825 }
826 #endif
827
828 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
829 {
830         switch (orig) {
831         case 0:
832                 file->f_pos = offset;
833                 break;
834         case 1:
835                 file->f_pos += offset;
836                 break;
837         default:
838                 return -EINVAL;
839         }
840         force_successful_syscall_return();
841         return file->f_pos;
842 }
843
844 static struct file_operations proc_mem_operations = {
845         .llseek         = mem_lseek,
846         .read           = mem_read,
847         .write          = mem_write,
848         .open           = mem_open,
849 };
850
851 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
852                                 size_t count, loff_t *ppos)
853 {
854         struct task_struct *task = proc_task(file->f_dentry->d_inode);
855         char buffer[8];
856         size_t len;
857         int oom_adjust = task->oomkilladj;
858         loff_t __ppos = *ppos;
859
860         len = sprintf(buffer, "%i\n", oom_adjust);
861         if (__ppos >= len)
862                 return 0;
863         if (count > len-__ppos)
864                 count = len-__ppos;
865         if (copy_to_user(buf, buffer + __ppos, count))
866                 return -EFAULT;
867         *ppos = __ppos + count;
868         return count;
869 }
870
871 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
872                                 size_t count, loff_t *ppos)
873 {
874         struct task_struct *task = proc_task(file->f_dentry->d_inode);
875         char buffer[8], *end;
876         int oom_adjust;
877
878         if (!capable(CAP_SYS_RESOURCE))
879                 return -EPERM;
880         memset(buffer, 0, 8);
881         if (count > 6)
882                 count = 6;
883         if (copy_from_user(buffer, buf, count))
884                 return -EFAULT;
885         oom_adjust = simple_strtol(buffer, &end, 0);
886         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
887                 return -EINVAL;
888         if (*end == '\n')
889                 end++;
890         task->oomkilladj = oom_adjust;
891         if (end - buffer == 0)
892                 return -EIO;
893         return end - buffer;
894 }
895
896 static struct file_operations proc_oom_adjust_operations = {
897         .read           = oom_adjust_read,
898         .write          = oom_adjust_write,
899 };
900
901 static struct inode_operations proc_mem_inode_operations = {
902         .permission     = proc_permission,
903 };
904
905 #ifdef CONFIG_AUDITSYSCALL
906 #define TMPBUFLEN 21
907 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
908                                   size_t count, loff_t *ppos)
909 {
910         struct inode * inode = file->f_dentry->d_inode;
911         struct task_struct *task = proc_task(inode);
912         ssize_t length;
913         char tmpbuf[TMPBUFLEN];
914
915         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
916                                 audit_get_loginuid(task->audit_context));
917         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
918 }
919
920 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
921                                    size_t count, loff_t *ppos)
922 {
923         struct inode * inode = file->f_dentry->d_inode;
924         char *page, *tmp;
925         ssize_t length;
926         struct task_struct *task = proc_task(inode);
927         uid_t loginuid;
928
929         if (!capable(CAP_AUDIT_CONTROL))
930                 return -EPERM;
931
932         if (current != task)
933                 return -EPERM;
934
935         if (count > PAGE_SIZE)
936                 count = PAGE_SIZE;
937
938         if (*ppos != 0) {
939                 /* No partial writes. */
940                 return -EINVAL;
941         }
942         page = (char*)__get_free_page(GFP_USER);
943         if (!page)
944                 return -ENOMEM;
945         length = -EFAULT;
946         if (copy_from_user(page, buf, count))
947                 goto out_free_page;
948
949         loginuid = simple_strtoul(page, &tmp, 10);
950         if (tmp == page) {
951                 length = -EINVAL;
952                 goto out_free_page;
953
954         }
955         length = audit_set_loginuid(task, loginuid);
956         if (likely(length == 0))
957                 length = count;
958
959 out_free_page:
960         free_page((unsigned long) page);
961         return length;
962 }
963
964 static struct file_operations proc_loginuid_operations = {
965         .read           = proc_loginuid_read,
966         .write          = proc_loginuid_write,
967 };
968 #endif
969
970 #ifdef CONFIG_SECCOMP
971 static ssize_t seccomp_read(struct file *file, char __user *buf,
972                             size_t count, loff_t *ppos)
973 {
974         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
975         char __buf[20];
976         loff_t __ppos = *ppos;
977         size_t len;
978
979         /* no need to print the trailing zero, so use only len */
980         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
981         if (__ppos >= len)
982                 return 0;
983         if (count > len - __ppos)
984                 count = len - __ppos;
985         if (copy_to_user(buf, __buf + __ppos, count))
986                 return -EFAULT;
987         *ppos = __ppos + count;
988         return count;
989 }
990
991 static ssize_t seccomp_write(struct file *file, const char __user *buf,
992                              size_t count, loff_t *ppos)
993 {
994         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
995         char __buf[20], *end;
996         unsigned int seccomp_mode;
997
998         /* can set it only once to be even more secure */
999         if (unlikely(tsk->seccomp.mode))
1000                 return -EPERM;
1001
1002         memset(__buf, 0, sizeof(__buf));
1003         count = min(count, sizeof(__buf) - 1);
1004         if (copy_from_user(__buf, buf, count))
1005                 return -EFAULT;
1006         seccomp_mode = simple_strtoul(__buf, &end, 0);
1007         if (*end == '\n')
1008                 end++;
1009         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1010                 tsk->seccomp.mode = seccomp_mode;
1011                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1012         } else
1013                 return -EINVAL;
1014         if (unlikely(!(end - __buf)))
1015                 return -EIO;
1016         return end - __buf;
1017 }
1018
1019 static struct file_operations proc_seccomp_operations = {
1020         .read           = seccomp_read,
1021         .write          = seccomp_write,
1022 };
1023 #endif /* CONFIG_SECCOMP */
1024
1025 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1026 {
1027         struct inode *inode = dentry->d_inode;
1028         int error = -EACCES;
1029
1030         /* We don't need a base pointer in the /proc filesystem */
1031         path_release(nd);
1032
1033         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1034                 goto out;
1035         error = proc_check_root(inode);
1036         if (error)
1037                 goto out;
1038
1039         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1040         nd->last_type = LAST_BIND;
1041 out:
1042         return ERR_PTR(error);
1043 }
1044
1045 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1046                             char __user *buffer, int buflen)
1047 {
1048         struct inode * inode;
1049         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1050         int len;
1051
1052         if (!tmp)
1053                 return -ENOMEM;
1054                 
1055         inode = dentry->d_inode;
1056         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1057         len = PTR_ERR(path);
1058         if (IS_ERR(path))
1059                 goto out;
1060         len = tmp + PAGE_SIZE - 1 - path;
1061
1062         if (len > buflen)
1063                 len = buflen;
1064         if (copy_to_user(buffer, path, len))
1065                 len = -EFAULT;
1066  out:
1067         free_page((unsigned long)tmp);
1068         return len;
1069 }
1070
1071 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1072 {
1073         int error = -EACCES;
1074         struct inode *inode = dentry->d_inode;
1075         struct dentry *de;
1076         struct vfsmount *mnt = NULL;
1077
1078         lock_kernel();
1079
1080         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1081                 goto out;
1082         error = proc_check_root(inode);
1083         if (error)
1084                 goto out;
1085
1086         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1087         if (error)
1088                 goto out;
1089
1090         error = do_proc_readlink(de, mnt, buffer, buflen);
1091         dput(de);
1092         mntput(mnt);
1093 out:
1094         unlock_kernel();
1095         return error;
1096 }
1097
1098 static struct inode_operations proc_pid_link_inode_operations = {
1099         .readlink       = proc_pid_readlink,
1100         .follow_link    = proc_pid_follow_link
1101 };
1102
1103 #define NUMBUF 10
1104
1105 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1106 {
1107         struct inode *inode = filp->f_dentry->d_inode;
1108         struct task_struct *p = proc_task(inode);
1109         unsigned int fd, tid, ino;
1110         int retval;
1111         char buf[NUMBUF];
1112         struct files_struct * files;
1113         struct fdtable *fdt;
1114
1115         retval = -ENOENT;
1116         if (!pid_alive(p))
1117                 goto out;
1118         retval = 0;
1119         tid = p->pid;
1120
1121         fd = filp->f_pos;
1122         switch (fd) {
1123                 case 0:
1124                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1125                                 goto out;
1126                         filp->f_pos++;
1127                 case 1:
1128                         ino = fake_ino(tid, PROC_TID_INO);
1129                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1130                                 goto out;
1131                         filp->f_pos++;
1132                 default:
1133                         files = get_files_struct(p);
1134                         if (!files)
1135                                 goto out;
1136                         rcu_read_lock();
1137                         fdt = files_fdtable(files);
1138                         for (fd = filp->f_pos-2;
1139                              fd < fdt->max_fds;
1140                              fd++, filp->f_pos++) {
1141                                 unsigned int i,j;
1142
1143                                 if (!fcheck_files(files, fd))
1144                                         continue;
1145                                 rcu_read_unlock();
1146
1147                                 j = NUMBUF;
1148                                 i = fd;
1149                                 do {
1150                                         j--;
1151                                         buf[j] = '0' + (i % 10);
1152                                         i /= 10;
1153                                 } while (i);
1154
1155                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1156                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1157                                         rcu_read_lock();
1158                                         break;
1159                                 }
1160                                 rcu_read_lock();
1161                         }
1162                         rcu_read_unlock();
1163                         put_files_struct(files);
1164         }
1165 out:
1166         return retval;
1167 }
1168
1169 static int proc_pident_readdir(struct file *filp,
1170                 void *dirent, filldir_t filldir,
1171                 struct pid_entry *ents, unsigned int nents)
1172 {
1173         int i;
1174         int pid;
1175         struct dentry *dentry = filp->f_dentry;
1176         struct inode *inode = dentry->d_inode;
1177         struct pid_entry *p;
1178         ino_t ino;
1179         int ret;
1180
1181         ret = -ENOENT;
1182         if (!pid_alive(proc_task(inode)))
1183                 goto out;
1184
1185         ret = 0;
1186         pid = proc_task(inode)->pid;
1187         i = filp->f_pos;
1188         switch (i) {
1189         case 0:
1190                 ino = inode->i_ino;
1191                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1192                         goto out;
1193                 i++;
1194                 filp->f_pos++;
1195                 /* fall through */
1196         case 1:
1197                 ino = parent_ino(dentry);
1198                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1199                         goto out;
1200                 i++;
1201                 filp->f_pos++;
1202                 /* fall through */
1203         default:
1204                 i -= 2;
1205                 if (i >= nents) {
1206                         ret = 1;
1207                         goto out;
1208                 }
1209                 p = ents + i;
1210                 while (p->name) {
1211                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1212                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1213                                 goto out;
1214                         filp->f_pos++;
1215                         p++;
1216                 }
1217         }
1218
1219         ret = 1;
1220 out:
1221         return ret;
1222 }
1223
1224 static int proc_tgid_base_readdir(struct file * filp,
1225                              void * dirent, filldir_t filldir)
1226 {
1227         return proc_pident_readdir(filp,dirent,filldir,
1228                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1229 }
1230
1231 static int proc_tid_base_readdir(struct file * filp,
1232                              void * dirent, filldir_t filldir)
1233 {
1234         return proc_pident_readdir(filp,dirent,filldir,
1235                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1236 }
1237
1238 /* building an inode */
1239
1240 static int task_dumpable(struct task_struct *task)
1241 {
1242         int dumpable = 0;
1243         struct mm_struct *mm;
1244
1245         task_lock(task);
1246         mm = task->mm;
1247         if (mm)
1248                 dumpable = mm->dumpable;
1249         task_unlock(task);
1250         if(dumpable == 1)
1251                 return 1;
1252         return 0;
1253 }
1254
1255
1256 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1257 {
1258         struct inode * inode;
1259         struct proc_inode *ei;
1260
1261         /* We need a new inode */
1262         
1263         inode = new_inode(sb);
1264         if (!inode)
1265                 goto out;
1266
1267         /* Common stuff */
1268         ei = PROC_I(inode);
1269         ei->task = NULL;
1270         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1271         inode->i_ino = fake_ino(task->pid, ino);
1272
1273         if (!pid_alive(task))
1274                 goto out_unlock;
1275
1276         /*
1277          * grab the reference to task.
1278          */
1279         get_task_struct(task);
1280         ei->task = task;
1281         ei->type = ino;
1282         inode->i_uid = 0;
1283         inode->i_gid = 0;
1284         if (ino == PROC_TGID_INO || ino == PROC_TID_INO || task_dumpable(task)) {
1285                 inode->i_uid = task->euid;
1286                 inode->i_gid = task->egid;
1287         }
1288         security_task_to_inode(task, inode);
1289
1290 out:
1291         return inode;
1292
1293 out_unlock:
1294         ei->pde = NULL;
1295         iput(inode);
1296         return NULL;
1297 }
1298
1299 /* dentry stuff */
1300
1301 /*
1302  *      Exceptional case: normally we are not allowed to unhash a busy
1303  * directory. In this case, however, we can do it - no aliasing problems
1304  * due to the way we treat inodes.
1305  *
1306  * Rewrite the inode's ownerships here because the owning task may have
1307  * performed a setuid(), etc.
1308  */
1309 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1310 {
1311         struct inode *inode = dentry->d_inode;
1312         struct task_struct *task = proc_task(inode);
1313         if (pid_alive(task)) {
1314                 if (proc_type(inode) == PROC_TGID_INO || proc_type(inode) == PROC_TID_INO || task_dumpable(task)) {
1315                         inode->i_uid = task->euid;
1316                         inode->i_gid = task->egid;
1317                 } else {
1318                         inode->i_uid = 0;
1319                         inode->i_gid = 0;
1320                 }
1321                 security_task_to_inode(task, inode);
1322                 return 1;
1323         }
1324         d_drop(dentry);
1325         return 0;
1326 }
1327
1328 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1329 {
1330         struct inode *inode = dentry->d_inode;
1331         struct task_struct *task = proc_task(inode);
1332         int fd = proc_type(inode) - PROC_TID_FD_DIR;
1333         struct files_struct *files;
1334
1335         files = get_files_struct(task);
1336         if (files) {
1337                 rcu_read_lock();
1338                 if (fcheck_files(files, fd)) {
1339                         rcu_read_unlock();
1340                         put_files_struct(files);
1341                         if (task_dumpable(task)) {
1342                                 inode->i_uid = task->euid;
1343                                 inode->i_gid = task->egid;
1344                         } else {
1345                                 inode->i_uid = 0;
1346                                 inode->i_gid = 0;
1347                         }
1348                         security_task_to_inode(task, inode);
1349                         return 1;
1350                 }
1351                 rcu_read_unlock();
1352                 put_files_struct(files);
1353         }
1354         d_drop(dentry);
1355         return 0;
1356 }
1357
1358 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1359 {
1360         struct task_struct *task = proc_task(inode);
1361         spin_lock(&task->proc_lock);
1362         if (task->proc_dentry == dentry)
1363                 task->proc_dentry = NULL;
1364         spin_unlock(&task->proc_lock);
1365         iput(inode);
1366 }
1367
1368 static int pid_delete_dentry(struct dentry * dentry)
1369 {
1370         /* Is the task we represent dead?
1371          * If so, then don't put the dentry on the lru list,
1372          * kill it immediately.
1373          */
1374         return !pid_alive(proc_task(dentry->d_inode));
1375 }
1376
1377 static struct dentry_operations tid_fd_dentry_operations =
1378 {
1379         .d_revalidate   = tid_fd_revalidate,
1380         .d_delete       = pid_delete_dentry,
1381 };
1382
1383 static struct dentry_operations pid_dentry_operations =
1384 {
1385         .d_revalidate   = pid_revalidate,
1386         .d_delete       = pid_delete_dentry,
1387 };
1388
1389 static struct dentry_operations pid_base_dentry_operations =
1390 {
1391         .d_revalidate   = pid_revalidate,
1392         .d_iput         = pid_base_iput,
1393         .d_delete       = pid_delete_dentry,
1394 };
1395
1396 /* Lookups */
1397
1398 static unsigned name_to_int(struct dentry *dentry)
1399 {
1400         const char *name = dentry->d_name.name;
1401         int len = dentry->d_name.len;
1402         unsigned n = 0;
1403
1404         if (len > 1 && *name == '0')
1405                 goto out;
1406         while (len-- > 0) {
1407                 unsigned c = *name++ - '0';
1408                 if (c > 9)
1409                         goto out;
1410                 if (n >= (~0U-9)/10)
1411                         goto out;
1412                 n *= 10;
1413                 n += c;
1414         }
1415         return n;
1416 out:
1417         return ~0U;
1418 }
1419
1420 /* SMP-safe */
1421 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1422 {
1423         struct task_struct *task = proc_task(dir);
1424         unsigned fd = name_to_int(dentry);
1425         struct file * file;
1426         struct files_struct * files;
1427         struct inode *inode;
1428         struct proc_inode *ei;
1429
1430         if (fd == ~0U)
1431                 goto out;
1432         if (!pid_alive(task))
1433                 goto out;
1434
1435         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1436         if (!inode)
1437                 goto out;
1438         ei = PROC_I(inode);
1439         files = get_files_struct(task);
1440         if (!files)
1441                 goto out_unlock;
1442         inode->i_mode = S_IFLNK;
1443         rcu_read_lock();
1444         file = fcheck_files(files, fd);
1445         if (!file)
1446                 goto out_unlock2;
1447         if (file->f_mode & 1)
1448                 inode->i_mode |= S_IRUSR | S_IXUSR;
1449         if (file->f_mode & 2)
1450                 inode->i_mode |= S_IWUSR | S_IXUSR;
1451         rcu_read_unlock();
1452         put_files_struct(files);
1453         inode->i_op = &proc_pid_link_inode_operations;
1454         inode->i_size = 64;
1455         ei->op.proc_get_link = proc_fd_link;
1456         dentry->d_op = &tid_fd_dentry_operations;
1457         d_add(dentry, inode);
1458         return NULL;
1459
1460 out_unlock2:
1461         rcu_read_unlock();
1462         put_files_struct(files);
1463 out_unlock:
1464         iput(inode);
1465 out:
1466         return ERR_PTR(-ENOENT);
1467 }
1468
1469 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1470 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1471
1472 static struct file_operations proc_fd_operations = {
1473         .read           = generic_read_dir,
1474         .readdir        = proc_readfd,
1475 };
1476
1477 static struct file_operations proc_task_operations = {
1478         .read           = generic_read_dir,
1479         .readdir        = proc_task_readdir,
1480 };
1481
1482 /*
1483  * proc directories can do almost nothing..
1484  */
1485 static struct inode_operations proc_fd_inode_operations = {
1486         .lookup         = proc_lookupfd,
1487         .permission     = proc_permission,
1488 };
1489
1490 static struct inode_operations proc_task_inode_operations = {
1491         .lookup         = proc_task_lookup,
1492         .permission     = proc_task_permission,
1493 };
1494
1495 #ifdef CONFIG_SECURITY
1496 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1497                                   size_t count, loff_t *ppos)
1498 {
1499         struct inode * inode = file->f_dentry->d_inode;
1500         unsigned long page;
1501         ssize_t length;
1502         struct task_struct *task = proc_task(inode);
1503
1504         if (count > PAGE_SIZE)
1505                 count = PAGE_SIZE;
1506         if (!(page = __get_free_page(GFP_KERNEL)))
1507                 return -ENOMEM;
1508
1509         length = security_getprocattr(task, 
1510                                       (char*)file->f_dentry->d_name.name, 
1511                                       (void*)page, count);
1512         if (length >= 0)
1513                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1514         free_page(page);
1515         return length;
1516 }
1517
1518 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1519                                    size_t count, loff_t *ppos)
1520
1521         struct inode * inode = file->f_dentry->d_inode;
1522         char *page; 
1523         ssize_t length; 
1524         struct task_struct *task = proc_task(inode); 
1525
1526         if (count > PAGE_SIZE) 
1527                 count = PAGE_SIZE; 
1528         if (*ppos != 0) {
1529                 /* No partial writes. */
1530                 return -EINVAL;
1531         }
1532         page = (char*)__get_free_page(GFP_USER); 
1533         if (!page) 
1534                 return -ENOMEM;
1535         length = -EFAULT; 
1536         if (copy_from_user(page, buf, count)) 
1537                 goto out;
1538
1539         length = security_setprocattr(task, 
1540                                       (char*)file->f_dentry->d_name.name, 
1541                                       (void*)page, count);
1542 out:
1543         free_page((unsigned long) page);
1544         return length;
1545
1546
1547 static struct file_operations proc_pid_attr_operations = {
1548         .read           = proc_pid_attr_read,
1549         .write          = proc_pid_attr_write,
1550 };
1551
1552 static struct file_operations proc_tid_attr_operations;
1553 static struct inode_operations proc_tid_attr_inode_operations;
1554 static struct file_operations proc_tgid_attr_operations;
1555 static struct inode_operations proc_tgid_attr_inode_operations;
1556 #endif
1557
1558 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1559
1560 /* SMP-safe */
1561 static struct dentry *proc_pident_lookup(struct inode *dir, 
1562                                          struct dentry *dentry,
1563                                          struct pid_entry *ents)
1564 {
1565         struct inode *inode;
1566         int error;
1567         struct task_struct *task = proc_task(dir);
1568         struct pid_entry *p;
1569         struct proc_inode *ei;
1570
1571         error = -ENOENT;
1572         inode = NULL;
1573
1574         if (!pid_alive(task))
1575                 goto out;
1576
1577         for (p = ents; p->name; p++) {
1578                 if (p->len != dentry->d_name.len)
1579                         continue;
1580                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1581                         break;
1582         }
1583         if (!p->name)
1584                 goto out;
1585
1586         error = -EINVAL;
1587         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1588         if (!inode)
1589                 goto out;
1590
1591         ei = PROC_I(inode);
1592         inode->i_mode = p->mode;
1593         /*
1594          * Yes, it does not scale. And it should not. Don't add
1595          * new entries into /proc/<tgid>/ without very good reasons.
1596          */
1597         switch(p->type) {
1598                 case PROC_TGID_TASK:
1599                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1600                         inode->i_op = &proc_task_inode_operations;
1601                         inode->i_fop = &proc_task_operations;
1602                         break;
1603                 case PROC_TID_FD:
1604                 case PROC_TGID_FD:
1605                         inode->i_nlink = 2;
1606                         inode->i_op = &proc_fd_inode_operations;
1607                         inode->i_fop = &proc_fd_operations;
1608                         break;
1609                 case PROC_TID_EXE:
1610                 case PROC_TGID_EXE:
1611                         inode->i_op = &proc_pid_link_inode_operations;
1612                         ei->op.proc_get_link = proc_exe_link;
1613                         break;
1614                 case PROC_TID_CWD:
1615                 case PROC_TGID_CWD:
1616                         inode->i_op = &proc_pid_link_inode_operations;
1617                         ei->op.proc_get_link = proc_cwd_link;
1618                         break;
1619                 case PROC_TID_ROOT:
1620                 case PROC_TGID_ROOT:
1621                         inode->i_op = &proc_pid_link_inode_operations;
1622                         ei->op.proc_get_link = proc_root_link;
1623                         break;
1624                 case PROC_TID_ENVIRON:
1625                 case PROC_TGID_ENVIRON:
1626                         inode->i_fop = &proc_info_file_operations;
1627                         ei->op.proc_read = proc_pid_environ;
1628                         break;
1629                 case PROC_TID_AUXV:
1630                 case PROC_TGID_AUXV:
1631                         inode->i_fop = &proc_info_file_operations;
1632                         ei->op.proc_read = proc_pid_auxv;
1633                         break;
1634                 case PROC_TID_STATUS:
1635                 case PROC_TGID_STATUS:
1636                         inode->i_fop = &proc_info_file_operations;
1637                         ei->op.proc_read = proc_pid_status;
1638                         break;
1639                 case PROC_TID_STAT:
1640                         inode->i_fop = &proc_info_file_operations;
1641                         ei->op.proc_read = proc_tid_stat;
1642                         break;
1643                 case PROC_TGID_STAT:
1644                         inode->i_fop = &proc_info_file_operations;
1645                         ei->op.proc_read = proc_tgid_stat;
1646                         break;
1647                 case PROC_TID_CMDLINE:
1648                 case PROC_TGID_CMDLINE:
1649                         inode->i_fop = &proc_info_file_operations;
1650                         ei->op.proc_read = proc_pid_cmdline;
1651                         break;
1652                 case PROC_TID_STATM:
1653                 case PROC_TGID_STATM:
1654                         inode->i_fop = &proc_info_file_operations;
1655                         ei->op.proc_read = proc_pid_statm;
1656                         break;
1657                 case PROC_TID_MAPS:
1658                 case PROC_TGID_MAPS:
1659                         inode->i_fop = &proc_maps_operations;
1660                         break;
1661 #ifdef CONFIG_NUMA
1662                 case PROC_TID_NUMA_MAPS:
1663                 case PROC_TGID_NUMA_MAPS:
1664                         inode->i_fop = &proc_numa_maps_operations;
1665                         break;
1666 #endif
1667                 case PROC_TID_MEM:
1668                 case PROC_TGID_MEM:
1669                         inode->i_op = &proc_mem_inode_operations;
1670                         inode->i_fop = &proc_mem_operations;
1671                         break;
1672 #ifdef CONFIG_SECCOMP
1673                 case PROC_TID_SECCOMP:
1674                 case PROC_TGID_SECCOMP:
1675                         inode->i_fop = &proc_seccomp_operations;
1676                         break;
1677 #endif /* CONFIG_SECCOMP */
1678                 case PROC_TID_MOUNTS:
1679                 case PROC_TGID_MOUNTS:
1680                         inode->i_fop = &proc_mounts_operations;
1681                         break;
1682                 case PROC_TID_SMAPS:
1683                 case PROC_TGID_SMAPS:
1684                         inode->i_fop = &proc_smaps_operations;
1685                         break;
1686 #ifdef CONFIG_SECURITY
1687                 case PROC_TID_ATTR:
1688                         inode->i_nlink = 2;
1689                         inode->i_op = &proc_tid_attr_inode_operations;
1690                         inode->i_fop = &proc_tid_attr_operations;
1691                         break;
1692                 case PROC_TGID_ATTR:
1693                         inode->i_nlink = 2;
1694                         inode->i_op = &proc_tgid_attr_inode_operations;
1695                         inode->i_fop = &proc_tgid_attr_operations;
1696                         break;
1697                 case PROC_TID_ATTR_CURRENT:
1698                 case PROC_TGID_ATTR_CURRENT:
1699                 case PROC_TID_ATTR_PREV:
1700                 case PROC_TGID_ATTR_PREV:
1701                 case PROC_TID_ATTR_EXEC:
1702                 case PROC_TGID_ATTR_EXEC:
1703                 case PROC_TID_ATTR_FSCREATE:
1704                 case PROC_TGID_ATTR_FSCREATE:
1705                         inode->i_fop = &proc_pid_attr_operations;
1706                         break;
1707 #endif
1708 #ifdef CONFIG_KALLSYMS
1709                 case PROC_TID_WCHAN:
1710                 case PROC_TGID_WCHAN:
1711                         inode->i_fop = &proc_info_file_operations;
1712                         ei->op.proc_read = proc_pid_wchan;
1713                         break;
1714 #endif
1715 #ifdef CONFIG_SCHEDSTATS
1716                 case PROC_TID_SCHEDSTAT:
1717                 case PROC_TGID_SCHEDSTAT:
1718                         inode->i_fop = &proc_info_file_operations;
1719                         ei->op.proc_read = proc_pid_schedstat;
1720                         break;
1721 #endif
1722 #ifdef CONFIG_CPUSETS
1723                 case PROC_TID_CPUSET:
1724                 case PROC_TGID_CPUSET:
1725                         inode->i_fop = &proc_cpuset_operations;
1726                         break;
1727 #endif
1728                 case PROC_TID_OOM_SCORE:
1729                 case PROC_TGID_OOM_SCORE:
1730                         inode->i_fop = &proc_info_file_operations;
1731                         ei->op.proc_read = proc_oom_score;
1732                         break;
1733                 case PROC_TID_OOM_ADJUST:
1734                 case PROC_TGID_OOM_ADJUST:
1735                         inode->i_fop = &proc_oom_adjust_operations;
1736                         break;
1737 #ifdef CONFIG_AUDITSYSCALL
1738                 case PROC_TID_LOGINUID:
1739                 case PROC_TGID_LOGINUID:
1740                         inode->i_fop = &proc_loginuid_operations;
1741                         break;
1742 #endif
1743                 default:
1744                         printk("procfs: impossible type (%d)",p->type);
1745                         iput(inode);
1746                         return ERR_PTR(-EINVAL);
1747         }
1748         dentry->d_op = &pid_dentry_operations;
1749         d_add(dentry, inode);
1750         return NULL;
1751
1752 out:
1753         return ERR_PTR(error);
1754 }
1755
1756 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1757         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1758 }
1759
1760 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1761         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1762 }
1763
1764 static struct file_operations proc_tgid_base_operations = {
1765         .read           = generic_read_dir,
1766         .readdir        = proc_tgid_base_readdir,
1767 };
1768
1769 static struct file_operations proc_tid_base_operations = {
1770         .read           = generic_read_dir,
1771         .readdir        = proc_tid_base_readdir,
1772 };
1773
1774 static struct inode_operations proc_tgid_base_inode_operations = {
1775         .lookup         = proc_tgid_base_lookup,
1776 };
1777
1778 static struct inode_operations proc_tid_base_inode_operations = {
1779         .lookup         = proc_tid_base_lookup,
1780 };
1781
1782 #ifdef CONFIG_SECURITY
1783 static int proc_tgid_attr_readdir(struct file * filp,
1784                              void * dirent, filldir_t filldir)
1785 {
1786         return proc_pident_readdir(filp,dirent,filldir,
1787                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1788 }
1789
1790 static int proc_tid_attr_readdir(struct file * filp,
1791                              void * dirent, filldir_t filldir)
1792 {
1793         return proc_pident_readdir(filp,dirent,filldir,
1794                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1795 }
1796
1797 static struct file_operations proc_tgid_attr_operations = {
1798         .read           = generic_read_dir,
1799         .readdir        = proc_tgid_attr_readdir,
1800 };
1801
1802 static struct file_operations proc_tid_attr_operations = {
1803         .read           = generic_read_dir,
1804         .readdir        = proc_tid_attr_readdir,
1805 };
1806
1807 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1808                                 struct dentry *dentry, struct nameidata *nd)
1809 {
1810         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1811 }
1812
1813 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1814                                 struct dentry *dentry, struct nameidata *nd)
1815 {
1816         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1817 }
1818
1819 static struct inode_operations proc_tgid_attr_inode_operations = {
1820         .lookup         = proc_tgid_attr_lookup,
1821 };
1822
1823 static struct inode_operations proc_tid_attr_inode_operations = {
1824         .lookup         = proc_tid_attr_lookup,
1825 };
1826 #endif
1827
1828 /*
1829  * /proc/self:
1830  */
1831 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1832                               int buflen)
1833 {
1834         char tmp[30];
1835         sprintf(tmp, "%d", current->tgid);
1836         return vfs_readlink(dentry,buffer,buflen,tmp);
1837 }
1838
1839 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1840 {
1841         char tmp[30];
1842         sprintf(tmp, "%d", current->tgid);
1843         return ERR_PTR(vfs_follow_link(nd,tmp));
1844 }       
1845
1846 static struct inode_operations proc_self_inode_operations = {
1847         .readlink       = proc_self_readlink,
1848         .follow_link    = proc_self_follow_link,
1849 };
1850
1851 /**
1852  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1853  * @p: task that should be flushed.
1854  *
1855  * Drops the /proc/@pid dcache entry from the hash chains.
1856  *
1857  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1858  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1859  * if the pid value is immediately reused. This is enforced by
1860  * - caller must acquire spin_lock(p->proc_lock)
1861  * - must be called before detach_pid()
1862  * - proc_pid_lookup acquires proc_lock, and checks that
1863  *   the target is not dead by looking at the attach count
1864  *   of PIDTYPE_PID.
1865  */
1866
1867 struct dentry *proc_pid_unhash(struct task_struct *p)
1868 {
1869         struct dentry *proc_dentry;
1870
1871         proc_dentry = p->proc_dentry;
1872         if (proc_dentry != NULL) {
1873
1874                 spin_lock(&dcache_lock);
1875                 spin_lock(&proc_dentry->d_lock);
1876                 if (!d_unhashed(proc_dentry)) {
1877                         dget_locked(proc_dentry);
1878                         __d_drop(proc_dentry);
1879                         spin_unlock(&proc_dentry->d_lock);
1880                 } else {
1881                         spin_unlock(&proc_dentry->d_lock);
1882                         proc_dentry = NULL;
1883                 }
1884                 spin_unlock(&dcache_lock);
1885         }
1886         return proc_dentry;
1887 }
1888
1889 /**
1890  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1891  * @proc_dentry: directoy to prune.
1892  *
1893  * Shrink the /proc directory that was used by the just killed thread.
1894  */
1895         
1896 void proc_pid_flush(struct dentry *proc_dentry)
1897 {
1898         might_sleep();
1899         if(proc_dentry != NULL) {
1900                 shrink_dcache_parent(proc_dentry);
1901                 dput(proc_dentry);
1902         }
1903 }
1904
1905 /* SMP-safe */
1906 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1907 {
1908         struct task_struct *task;
1909         struct inode *inode;
1910         struct proc_inode *ei;
1911         unsigned tgid;
1912         int died;
1913
1914         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
1915                 inode = new_inode(dir->i_sb);
1916                 if (!inode)
1917                         return ERR_PTR(-ENOMEM);
1918                 ei = PROC_I(inode);
1919                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1920                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
1921                 ei->pde = NULL;
1922                 inode->i_mode = S_IFLNK|S_IRWXUGO;
1923                 inode->i_uid = inode->i_gid = 0;
1924                 inode->i_size = 64;
1925                 inode->i_op = &proc_self_inode_operations;
1926                 d_add(dentry, inode);
1927                 return NULL;
1928         }
1929         tgid = name_to_int(dentry);
1930         if (tgid == ~0U)
1931                 goto out;
1932
1933         read_lock(&tasklist_lock);
1934         task = find_task_by_pid(tgid);
1935         if (task)
1936                 get_task_struct(task);
1937         read_unlock(&tasklist_lock);
1938         if (!task)
1939                 goto out;
1940
1941         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
1942
1943
1944         if (!inode) {
1945                 put_task_struct(task);
1946                 goto out;
1947         }
1948         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1949         inode->i_op = &proc_tgid_base_inode_operations;
1950         inode->i_fop = &proc_tgid_base_operations;
1951         inode->i_flags|=S_IMMUTABLE;
1952 #ifdef CONFIG_SECURITY
1953         inode->i_nlink = 5;
1954 #else
1955         inode->i_nlink = 4;
1956 #endif
1957
1958         dentry->d_op = &pid_base_dentry_operations;
1959
1960         died = 0;
1961         d_add(dentry, inode);
1962         spin_lock(&task->proc_lock);
1963         task->proc_dentry = dentry;
1964         if (!pid_alive(task)) {
1965                 dentry = proc_pid_unhash(task);
1966                 died = 1;
1967         }
1968         spin_unlock(&task->proc_lock);
1969
1970         put_task_struct(task);
1971         if (died) {
1972                 proc_pid_flush(dentry);
1973                 goto out;
1974         }
1975         return NULL;
1976 out:
1977         return ERR_PTR(-ENOENT);
1978 }
1979
1980 /* SMP-safe */
1981 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1982 {
1983         struct task_struct *task;
1984         struct task_struct *leader = proc_task(dir);
1985         struct inode *inode;
1986         unsigned tid;
1987
1988         tid = name_to_int(dentry);
1989         if (tid == ~0U)
1990                 goto out;
1991
1992         read_lock(&tasklist_lock);
1993         task = find_task_by_pid(tid);
1994         if (task)
1995                 get_task_struct(task);
1996         read_unlock(&tasklist_lock);
1997         if (!task)
1998                 goto out;
1999         if (leader->tgid != task->tgid)
2000                 goto out_drop_task;
2001
2002         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2003
2004
2005         if (!inode)
2006                 goto out_drop_task;
2007         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2008         inode->i_op = &proc_tid_base_inode_operations;
2009         inode->i_fop = &proc_tid_base_operations;
2010         inode->i_flags|=S_IMMUTABLE;
2011 #ifdef CONFIG_SECURITY
2012         inode->i_nlink = 4;
2013 #else
2014         inode->i_nlink = 3;
2015 #endif
2016
2017         dentry->d_op = &pid_base_dentry_operations;
2018
2019         d_add(dentry, inode);
2020
2021         put_task_struct(task);
2022         return NULL;
2023 out_drop_task:
2024         put_task_struct(task);
2025 out:
2026         return ERR_PTR(-ENOENT);
2027 }
2028
2029 #define PROC_NUMBUF 10
2030 #define PROC_MAXPIDS 20
2031
2032 /*
2033  * Get a few tgid's to return for filldir - we need to hold the
2034  * tasklist lock while doing this, and we must release it before
2035  * we actually do the filldir itself, so we use a temp buffer..
2036  */
2037 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2038 {
2039         struct task_struct *p;
2040         int nr_tgids = 0;
2041
2042         index--;
2043         read_lock(&tasklist_lock);
2044         p = NULL;
2045         if (version) {
2046                 p = find_task_by_pid(version);
2047                 if (p && !thread_group_leader(p))
2048                         p = NULL;
2049         }
2050
2051         if (p)
2052                 index = 0;
2053         else
2054                 p = next_task(&init_task);
2055
2056         for ( ; p != &init_task; p = next_task(p)) {
2057                 int tgid = p->pid;
2058                 if (!pid_alive(p))
2059                         continue;
2060                 if (--index >= 0)
2061                         continue;
2062                 tgids[nr_tgids] = tgid;
2063                 nr_tgids++;
2064                 if (nr_tgids >= PROC_MAXPIDS)
2065                         break;
2066         }
2067         read_unlock(&tasklist_lock);
2068         return nr_tgids;
2069 }
2070
2071 /*
2072  * Get a few tid's to return for filldir - we need to hold the
2073  * tasklist lock while doing this, and we must release it before
2074  * we actually do the filldir itself, so we use a temp buffer..
2075  */
2076 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2077 {
2078         struct task_struct *leader_task = proc_task(dir);
2079         struct task_struct *task = leader_task;
2080         int nr_tids = 0;
2081
2082         index -= 2;
2083         read_lock(&tasklist_lock);
2084         /*
2085          * The starting point task (leader_task) might be an already
2086          * unlinked task, which cannot be used to access the task-list
2087          * via next_thread().
2088          */
2089         if (pid_alive(task)) do {
2090                 int tid = task->pid;
2091
2092                 if (--index >= 0)
2093                         continue;
2094                 if (tids != NULL)
2095                         tids[nr_tids] = tid;
2096                 nr_tids++;
2097                 if (nr_tids >= PROC_MAXPIDS)
2098                         break;
2099         } while ((task = next_thread(task)) != leader_task);
2100         read_unlock(&tasklist_lock);
2101         return nr_tids;
2102 }
2103
2104 /* for the /proc/ directory itself, after non-process stuff has been done */
2105 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2106 {
2107         unsigned int tgid_array[PROC_MAXPIDS];
2108         char buf[PROC_NUMBUF];
2109         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2110         unsigned int nr_tgids, i;
2111         int next_tgid;
2112
2113         if (!nr) {
2114                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2115                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2116                         return 0;
2117                 filp->f_pos++;
2118                 nr++;
2119         }
2120
2121         /* f_version caches the tgid value that the last readdir call couldn't
2122          * return. lseek aka telldir automagically resets f_version to 0.
2123          */
2124         next_tgid = filp->f_version;
2125         filp->f_version = 0;
2126         for (;;) {
2127                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2128                 if (!nr_tgids) {
2129                         /* no more entries ! */
2130                         break;
2131                 }
2132                 next_tgid = 0;
2133
2134                 /* do not use the last found pid, reserve it for next_tgid */
2135                 if (nr_tgids == PROC_MAXPIDS) {
2136                         nr_tgids--;
2137                         next_tgid = tgid_array[nr_tgids];
2138                 }
2139
2140                 for (i=0;i<nr_tgids;i++) {
2141                         int tgid = tgid_array[i];
2142                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2143                         unsigned long j = PROC_NUMBUF;
2144
2145                         do
2146                                 buf[--j] = '0' + (tgid % 10);
2147                         while ((tgid /= 10) != 0);
2148
2149                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2150                                 /* returning this tgid failed, save it as the first
2151                                  * pid for the next readir call */
2152                                 filp->f_version = tgid_array[i];
2153                                 goto out;
2154                         }
2155                         filp->f_pos++;
2156                         nr++;
2157                 }
2158         }
2159 out:
2160         return 0;
2161 }
2162
2163 /* for the /proc/TGID/task/ directories */
2164 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2165 {
2166         unsigned int tid_array[PROC_MAXPIDS];
2167         char buf[PROC_NUMBUF];
2168         unsigned int nr_tids, i;
2169         struct dentry *dentry = filp->f_dentry;
2170         struct inode *inode = dentry->d_inode;
2171         int retval = -ENOENT;
2172         ino_t ino;
2173         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2174
2175         if (!pid_alive(proc_task(inode)))
2176                 goto out;
2177         retval = 0;
2178
2179         switch (pos) {
2180         case 0:
2181                 ino = inode->i_ino;
2182                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2183                         goto out;
2184                 pos++;
2185                 /* fall through */
2186         case 1:
2187                 ino = parent_ino(dentry);
2188                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2189                         goto out;
2190                 pos++;
2191                 /* fall through */
2192         }
2193
2194         nr_tids = get_tid_list(pos, tid_array, inode);
2195         inode->i_nlink = pos + nr_tids;
2196
2197         for (i = 0; i < nr_tids; i++) {
2198                 unsigned long j = PROC_NUMBUF;
2199                 int tid = tid_array[i];
2200
2201                 ino = fake_ino(tid,PROC_TID_INO);
2202
2203                 do
2204                         buf[--j] = '0' + (tid % 10);
2205                 while ((tid /= 10) != 0);
2206
2207                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2208                         break;
2209                 pos++;
2210         }
2211 out:
2212         filp->f_pos = pos;
2213         return retval;
2214 }