[PATCH] files: break up files struct
[linux-2.6] / 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/kallsyms.h>
66 #include <linux/mount.h>
67 #include <linux/security.h>
68 #include <linux/ptrace.h>
69 #include <linux/seccomp.h>
70 #include <linux/cpuset.h>
71 #include <linux/audit.h>
72 #include "internal.h"
73
74 /*
75  * For hysterical raisins we keep the same inumbers as in the old procfs.
76  * Feel free to change the macro below - just keep the range distinct from
77  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
78  * As soon as we'll get a separate superblock we will be able to forget
79  * about magical ranges too.
80  */
81
82 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
83
84 enum pid_directory_inos {
85         PROC_TGID_INO = 2,
86         PROC_TGID_TASK,
87         PROC_TGID_STATUS,
88         PROC_TGID_MEM,
89 #ifdef CONFIG_SECCOMP
90         PROC_TGID_SECCOMP,
91 #endif
92         PROC_TGID_CWD,
93         PROC_TGID_ROOT,
94         PROC_TGID_EXE,
95         PROC_TGID_FD,
96         PROC_TGID_ENVIRON,
97         PROC_TGID_AUXV,
98         PROC_TGID_CMDLINE,
99         PROC_TGID_STAT,
100         PROC_TGID_STATM,
101         PROC_TGID_MAPS,
102         PROC_TGID_NUMA_MAPS,
103         PROC_TGID_MOUNTS,
104         PROC_TGID_WCHAN,
105         PROC_TGID_SMAPS,
106 #ifdef CONFIG_SCHEDSTATS
107         PROC_TGID_SCHEDSTAT,
108 #endif
109 #ifdef CONFIG_CPUSETS
110         PROC_TGID_CPUSET,
111 #endif
112 #ifdef CONFIG_SECURITY
113         PROC_TGID_ATTR,
114         PROC_TGID_ATTR_CURRENT,
115         PROC_TGID_ATTR_PREV,
116         PROC_TGID_ATTR_EXEC,
117         PROC_TGID_ATTR_FSCREATE,
118 #endif
119 #ifdef CONFIG_AUDITSYSCALL
120         PROC_TGID_LOGINUID,
121 #endif
122         PROC_TGID_OOM_SCORE,
123         PROC_TGID_OOM_ADJUST,
124         PROC_TID_INO,
125         PROC_TID_STATUS,
126         PROC_TID_MEM,
127 #ifdef CONFIG_SECCOMP
128         PROC_TID_SECCOMP,
129 #endif
130         PROC_TID_CWD,
131         PROC_TID_ROOT,
132         PROC_TID_EXE,
133         PROC_TID_FD,
134         PROC_TID_ENVIRON,
135         PROC_TID_AUXV,
136         PROC_TID_CMDLINE,
137         PROC_TID_STAT,
138         PROC_TID_STATM,
139         PROC_TID_MAPS,
140         PROC_TID_NUMA_MAPS,
141         PROC_TID_MOUNTS,
142         PROC_TID_WCHAN,
143         PROC_TID_SMAPS,
144 #ifdef CONFIG_SCHEDSTATS
145         PROC_TID_SCHEDSTAT,
146 #endif
147 #ifdef CONFIG_CPUSETS
148         PROC_TID_CPUSET,
149 #endif
150 #ifdef CONFIG_SECURITY
151         PROC_TID_ATTR,
152         PROC_TID_ATTR_CURRENT,
153         PROC_TID_ATTR_PREV,
154         PROC_TID_ATTR_EXEC,
155         PROC_TID_ATTR_FSCREATE,
156 #endif
157 #ifdef CONFIG_AUDITSYSCALL
158         PROC_TID_LOGINUID,
159 #endif
160         PROC_TID_OOM_SCORE,
161         PROC_TID_OOM_ADJUST,
162
163         /* Add new entries before this */
164         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
165 };
166
167 struct pid_entry {
168         int type;
169         int len;
170         char *name;
171         mode_t mode;
172 };
173
174 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
175
176 static struct pid_entry tgid_base_stuff[] = {
177         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
178         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
179         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
180         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
181         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
182         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
183         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
184         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
185         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
186 #ifdef CONFIG_NUMA
187         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
188 #endif
189         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
190 #ifdef CONFIG_SECCOMP
191         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
192 #endif
193         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
194         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
195         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
196         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
197         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
198 #ifdef CONFIG_SECURITY
199         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
200 #endif
201 #ifdef CONFIG_KALLSYMS
202         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
203 #endif
204 #ifdef CONFIG_SCHEDSTATS
205         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
206 #endif
207 #ifdef CONFIG_CPUSETS
208         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
209 #endif
210         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
211         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
212 #ifdef CONFIG_AUDITSYSCALL
213         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
214 #endif
215         {0,0,NULL,0}
216 };
217 static struct pid_entry tid_base_stuff[] = {
218         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
219         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
220         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
221         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
222         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
223         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
224         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
225         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
226 #ifdef CONFIG_NUMA
227         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
228 #endif
229         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
230 #ifdef CONFIG_SECCOMP
231         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
232 #endif
233         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
234         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
235         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
236         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
237         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
238 #ifdef CONFIG_SECURITY
239         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
240 #endif
241 #ifdef CONFIG_KALLSYMS
242         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
243 #endif
244 #ifdef CONFIG_SCHEDSTATS
245         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
246 #endif
247 #ifdef CONFIG_CPUSETS
248         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
249 #endif
250         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
251         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
252 #ifdef CONFIG_AUDITSYSCALL
253         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
254 #endif
255         {0,0,NULL,0}
256 };
257
258 #ifdef CONFIG_SECURITY
259 static struct pid_entry tgid_attr_stuff[] = {
260         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
261         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
262         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
263         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
264         {0,0,NULL,0}
265 };
266 static struct pid_entry tid_attr_stuff[] = {
267         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
268         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
269         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
270         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
271         {0,0,NULL,0}
272 };
273 #endif
274
275 #undef E
276
277 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
278 {
279         struct task_struct *task = proc_task(inode);
280         struct files_struct *files;
281         struct file *file;
282         int fd = proc_type(inode) - PROC_TID_FD_DIR;
283
284         files = get_files_struct(task);
285         if (files) {
286                 spin_lock(&files->file_lock);
287                 file = fcheck_files(files, fd);
288                 if (file) {
289                         *mnt = mntget(file->f_vfsmnt);
290                         *dentry = dget(file->f_dentry);
291                         spin_unlock(&files->file_lock);
292                         put_files_struct(files);
293                         return 0;
294                 }
295                 spin_unlock(&files->file_lock);
296                 put_files_struct(files);
297         }
298         return -ENOENT;
299 }
300
301 static struct fs_struct *get_fs_struct(struct task_struct *task)
302 {
303         struct fs_struct *fs;
304         task_lock(task);
305         fs = task->fs;
306         if(fs)
307                 atomic_inc(&fs->count);
308         task_unlock(task);
309         return fs;
310 }
311
312 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
313 {
314         struct fs_struct *fs = get_fs_struct(proc_task(inode));
315         int result = -ENOENT;
316         if (fs) {
317                 read_lock(&fs->lock);
318                 *mnt = mntget(fs->pwdmnt);
319                 *dentry = dget(fs->pwd);
320                 read_unlock(&fs->lock);
321                 result = 0;
322                 put_fs_struct(fs);
323         }
324         return result;
325 }
326
327 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
328 {
329         struct fs_struct *fs = get_fs_struct(proc_task(inode));
330         int result = -ENOENT;
331         if (fs) {
332                 read_lock(&fs->lock);
333                 *mnt = mntget(fs->rootmnt);
334                 *dentry = dget(fs->root);
335                 read_unlock(&fs->lock);
336                 result = 0;
337                 put_fs_struct(fs);
338         }
339         return result;
340 }
341
342 #define MAY_PTRACE(task) \
343         (task == current || \
344         (task->parent == current && \
345         (task->ptrace & PT_PTRACED) && \
346          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
347          security_ptrace(current,task) == 0))
348
349 static int proc_pid_environ(struct task_struct *task, char * buffer)
350 {
351         int res = 0;
352         struct mm_struct *mm = get_task_mm(task);
353         if (mm) {
354                 unsigned int len = mm->env_end - mm->env_start;
355                 if (len > PAGE_SIZE)
356                         len = PAGE_SIZE;
357                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
358                 if (!ptrace_may_attach(task))
359                         res = -ESRCH;
360                 mmput(mm);
361         }
362         return res;
363 }
364
365 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
366 {
367         int res = 0;
368         unsigned int len;
369         struct mm_struct *mm = get_task_mm(task);
370         if (!mm)
371                 goto out;
372         if (!mm->arg_end)
373                 goto out_mm;    /* Shh! No looking before we're done */
374
375         len = mm->arg_end - mm->arg_start;
376  
377         if (len > PAGE_SIZE)
378                 len = PAGE_SIZE;
379  
380         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
381
382         // If the nul at the end of args has been overwritten, then
383         // assume application is using setproctitle(3).
384         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
385                 len = strnlen(buffer, res);
386                 if (len < res) {
387                     res = len;
388                 } else {
389                         len = mm->env_end - mm->env_start;
390                         if (len > PAGE_SIZE - res)
391                                 len = PAGE_SIZE - res;
392                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
393                         res = strnlen(buffer, res);
394                 }
395         }
396 out_mm:
397         mmput(mm);
398 out:
399         return res;
400 }
401
402 static int proc_pid_auxv(struct task_struct *task, char *buffer)
403 {
404         int res = 0;
405         struct mm_struct *mm = get_task_mm(task);
406         if (mm) {
407                 unsigned int nwords = 0;
408                 do
409                         nwords += 2;
410                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
411                 res = nwords * sizeof(mm->saved_auxv[0]);
412                 if (res > PAGE_SIZE)
413                         res = PAGE_SIZE;
414                 memcpy(buffer, mm->saved_auxv, res);
415                 mmput(mm);
416         }
417         return res;
418 }
419
420
421 #ifdef CONFIG_KALLSYMS
422 /*
423  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
424  * Returns the resolved symbol.  If that fails, simply return the address.
425  */
426 static int proc_pid_wchan(struct task_struct *task, char *buffer)
427 {
428         char *modname;
429         const char *sym_name;
430         unsigned long wchan, size, offset;
431         char namebuf[KSYM_NAME_LEN+1];
432
433         wchan = get_wchan(task);
434
435         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
436         if (sym_name)
437                 return sprintf(buffer, "%s", sym_name);
438         return sprintf(buffer, "%lu", wchan);
439 }
440 #endif /* CONFIG_KALLSYMS */
441
442 #ifdef CONFIG_SCHEDSTATS
443 /*
444  * Provides /proc/PID/schedstat
445  */
446 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
447 {
448         return sprintf(buffer, "%lu %lu %lu\n",
449                         task->sched_info.cpu_time,
450                         task->sched_info.run_delay,
451                         task->sched_info.pcnt);
452 }
453 #endif
454
455 /* The badness from the OOM killer */
456 unsigned long badness(struct task_struct *p, unsigned long uptime);
457 static int proc_oom_score(struct task_struct *task, char *buffer)
458 {
459         unsigned long points;
460         struct timespec uptime;
461
462         do_posix_clock_monotonic_gettime(&uptime);
463         points = badness(task, uptime.tv_sec);
464         return sprintf(buffer, "%lu\n", points);
465 }
466
467 /************************************************************************/
468 /*                       Here the fs part begins                        */
469 /************************************************************************/
470
471 /* permission checks */
472
473 static int proc_check_root(struct inode *inode)
474 {
475         struct dentry *de, *base, *root;
476         struct vfsmount *our_vfsmnt, *vfsmnt, *mnt;
477         int res = 0;
478
479         if (proc_root_link(inode, &root, &vfsmnt)) /* Ewww... */
480                 return -ENOENT;
481         read_lock(&current->fs->lock);
482         our_vfsmnt = mntget(current->fs->rootmnt);
483         base = dget(current->fs->root);
484         read_unlock(&current->fs->lock);
485
486         spin_lock(&vfsmount_lock);
487         de = root;
488         mnt = vfsmnt;
489
490         while (vfsmnt != our_vfsmnt) {
491                 if (vfsmnt == vfsmnt->mnt_parent)
492                         goto out;
493                 de = vfsmnt->mnt_mountpoint;
494                 vfsmnt = vfsmnt->mnt_parent;
495         }
496
497         if (!is_subdir(de, base))
498                 goto out;
499         spin_unlock(&vfsmount_lock);
500
501 exit:
502         dput(base);
503         mntput(our_vfsmnt);
504         dput(root);
505         mntput(mnt);
506         return res;
507 out:
508         spin_unlock(&vfsmount_lock);
509         res = -EACCES;
510         goto exit;
511 }
512
513 static int proc_permission(struct inode *inode, int mask, struct nameidata *nd)
514 {
515         if (generic_permission(inode, mask, NULL) != 0)
516                 return -EACCES;
517         return proc_check_root(inode);
518 }
519
520 extern struct seq_operations proc_pid_maps_op;
521 static int maps_open(struct inode *inode, struct file *file)
522 {
523         struct task_struct *task = proc_task(inode);
524         int ret = seq_open(file, &proc_pid_maps_op);
525         if (!ret) {
526                 struct seq_file *m = file->private_data;
527                 m->private = task;
528         }
529         return ret;
530 }
531
532 static struct file_operations proc_maps_operations = {
533         .open           = maps_open,
534         .read           = seq_read,
535         .llseek         = seq_lseek,
536         .release        = seq_release,
537 };
538
539 #ifdef CONFIG_NUMA
540 extern struct seq_operations proc_pid_numa_maps_op;
541 static int numa_maps_open(struct inode *inode, struct file *file)
542 {
543         struct task_struct *task = proc_task(inode);
544         int ret = seq_open(file, &proc_pid_numa_maps_op);
545         if (!ret) {
546                 struct seq_file *m = file->private_data;
547                 m->private = task;
548         }
549         return ret;
550 }
551
552 static struct file_operations proc_numa_maps_operations = {
553         .open           = numa_maps_open,
554         .read           = seq_read,
555         .llseek         = seq_lseek,
556         .release        = seq_release,
557 };
558 #endif
559
560 extern struct seq_operations proc_pid_smaps_op;
561 static int smaps_open(struct inode *inode, struct file *file)
562 {
563         struct task_struct *task = proc_task(inode);
564         int ret = seq_open(file, &proc_pid_smaps_op);
565         if (!ret) {
566                 struct seq_file *m = file->private_data;
567                 m->private = task;
568         }
569         return ret;
570 }
571
572 static struct file_operations proc_smaps_operations = {
573         .open           = smaps_open,
574         .read           = seq_read,
575         .llseek         = seq_lseek,
576         .release        = seq_release,
577 };
578
579 extern struct seq_operations mounts_op;
580 static int mounts_open(struct inode *inode, struct file *file)
581 {
582         struct task_struct *task = proc_task(inode);
583         int ret = seq_open(file, &mounts_op);
584
585         if (!ret) {
586                 struct seq_file *m = file->private_data;
587                 struct namespace *namespace;
588                 task_lock(task);
589                 namespace = task->namespace;
590                 if (namespace)
591                         get_namespace(namespace);
592                 task_unlock(task);
593
594                 if (namespace)
595                         m->private = namespace;
596                 else {
597                         seq_release(inode, file);
598                         ret = -EINVAL;
599                 }
600         }
601         return ret;
602 }
603
604 static int mounts_release(struct inode *inode, struct file *file)
605 {
606         struct seq_file *m = file->private_data;
607         struct namespace *namespace = m->private;
608         put_namespace(namespace);
609         return seq_release(inode, file);
610 }
611
612 static struct file_operations proc_mounts_operations = {
613         .open           = mounts_open,
614         .read           = seq_read,
615         .llseek         = seq_lseek,
616         .release        = mounts_release,
617 };
618
619 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
620
621 static ssize_t proc_info_read(struct file * file, char __user * buf,
622                           size_t count, loff_t *ppos)
623 {
624         struct inode * inode = file->f_dentry->d_inode;
625         unsigned long page;
626         ssize_t length;
627         struct task_struct *task = proc_task(inode);
628
629         if (count > PROC_BLOCK_SIZE)
630                 count = PROC_BLOCK_SIZE;
631         if (!(page = __get_free_page(GFP_KERNEL)))
632                 return -ENOMEM;
633
634         length = PROC_I(inode)->op.proc_read(task, (char*)page);
635
636         if (length >= 0)
637                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
638         free_page(page);
639         return length;
640 }
641
642 static struct file_operations proc_info_file_operations = {
643         .read           = proc_info_read,
644 };
645
646 static int mem_open(struct inode* inode, struct file* file)
647 {
648         file->private_data = (void*)((long)current->self_exec_id);
649         return 0;
650 }
651
652 static ssize_t mem_read(struct file * file, char __user * buf,
653                         size_t count, loff_t *ppos)
654 {
655         struct task_struct *task = proc_task(file->f_dentry->d_inode);
656         char *page;
657         unsigned long src = *ppos;
658         int ret = -ESRCH;
659         struct mm_struct *mm;
660
661         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
662                 goto out;
663
664         ret = -ENOMEM;
665         page = (char *)__get_free_page(GFP_USER);
666         if (!page)
667                 goto out;
668
669         ret = 0;
670  
671         mm = get_task_mm(task);
672         if (!mm)
673                 goto out_free;
674
675         ret = -EIO;
676  
677         if (file->private_data != (void*)((long)current->self_exec_id))
678                 goto out_put;
679
680         ret = 0;
681  
682         while (count > 0) {
683                 int this_len, retval;
684
685                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
686                 retval = access_process_vm(task, src, page, this_len, 0);
687                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
688                         if (!ret)
689                                 ret = -EIO;
690                         break;
691                 }
692
693                 if (copy_to_user(buf, page, retval)) {
694                         ret = -EFAULT;
695                         break;
696                 }
697  
698                 ret += retval;
699                 src += retval;
700                 buf += retval;
701                 count -= retval;
702         }
703         *ppos = src;
704
705 out_put:
706         mmput(mm);
707 out_free:
708         free_page((unsigned long) page);
709 out:
710         return ret;
711 }
712
713 #define mem_write NULL
714
715 #ifndef mem_write
716 /* This is a security hazard */
717 static ssize_t mem_write(struct file * file, const char * buf,
718                          size_t count, loff_t *ppos)
719 {
720         int copied = 0;
721         char *page;
722         struct task_struct *task = proc_task(file->f_dentry->d_inode);
723         unsigned long dst = *ppos;
724
725         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
726                 return -ESRCH;
727
728         page = (char *)__get_free_page(GFP_USER);
729         if (!page)
730                 return -ENOMEM;
731
732         while (count > 0) {
733                 int this_len, retval;
734
735                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
736                 if (copy_from_user(page, buf, this_len)) {
737                         copied = -EFAULT;
738                         break;
739                 }
740                 retval = access_process_vm(task, dst, page, this_len, 1);
741                 if (!retval) {
742                         if (!copied)
743                                 copied = -EIO;
744                         break;
745                 }
746                 copied += retval;
747                 buf += retval;
748                 dst += retval;
749                 count -= retval;                        
750         }
751         *ppos = dst;
752         free_page((unsigned long) page);
753         return copied;
754 }
755 #endif
756
757 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
758 {
759         switch (orig) {
760         case 0:
761                 file->f_pos = offset;
762                 break;
763         case 1:
764                 file->f_pos += offset;
765                 break;
766         default:
767                 return -EINVAL;
768         }
769         force_successful_syscall_return();
770         return file->f_pos;
771 }
772
773 static struct file_operations proc_mem_operations = {
774         .llseek         = mem_lseek,
775         .read           = mem_read,
776         .write          = mem_write,
777         .open           = mem_open,
778 };
779
780 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
781                                 size_t count, loff_t *ppos)
782 {
783         struct task_struct *task = proc_task(file->f_dentry->d_inode);
784         char buffer[8];
785         size_t len;
786         int oom_adjust = task->oomkilladj;
787         loff_t __ppos = *ppos;
788
789         len = sprintf(buffer, "%i\n", oom_adjust);
790         if (__ppos >= len)
791                 return 0;
792         if (count > len-__ppos)
793                 count = len-__ppos;
794         if (copy_to_user(buf, buffer + __ppos, count))
795                 return -EFAULT;
796         *ppos = __ppos + count;
797         return count;
798 }
799
800 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
801                                 size_t count, loff_t *ppos)
802 {
803         struct task_struct *task = proc_task(file->f_dentry->d_inode);
804         char buffer[8], *end;
805         int oom_adjust;
806
807         if (!capable(CAP_SYS_RESOURCE))
808                 return -EPERM;
809         memset(buffer, 0, 8);
810         if (count > 6)
811                 count = 6;
812         if (copy_from_user(buffer, buf, count))
813                 return -EFAULT;
814         oom_adjust = simple_strtol(buffer, &end, 0);
815         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
816                 return -EINVAL;
817         if (*end == '\n')
818                 end++;
819         task->oomkilladj = oom_adjust;
820         if (end - buffer == 0)
821                 return -EIO;
822         return end - buffer;
823 }
824
825 static struct file_operations proc_oom_adjust_operations = {
826         .read           = oom_adjust_read,
827         .write          = oom_adjust_write,
828 };
829
830 static struct inode_operations proc_mem_inode_operations = {
831         .permission     = proc_permission,
832 };
833
834 #ifdef CONFIG_AUDITSYSCALL
835 #define TMPBUFLEN 21
836 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
837                                   size_t count, loff_t *ppos)
838 {
839         struct inode * inode = file->f_dentry->d_inode;
840         struct task_struct *task = proc_task(inode);
841         ssize_t length;
842         char tmpbuf[TMPBUFLEN];
843
844         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
845                                 audit_get_loginuid(task->audit_context));
846         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
847 }
848
849 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
850                                    size_t count, loff_t *ppos)
851 {
852         struct inode * inode = file->f_dentry->d_inode;
853         char *page, *tmp;
854         ssize_t length;
855         struct task_struct *task = proc_task(inode);
856         uid_t loginuid;
857
858         if (!capable(CAP_AUDIT_CONTROL))
859                 return -EPERM;
860
861         if (current != task)
862                 return -EPERM;
863
864         if (count > PAGE_SIZE)
865                 count = PAGE_SIZE;
866
867         if (*ppos != 0) {
868                 /* No partial writes. */
869                 return -EINVAL;
870         }
871         page = (char*)__get_free_page(GFP_USER);
872         if (!page)
873                 return -ENOMEM;
874         length = -EFAULT;
875         if (copy_from_user(page, buf, count))
876                 goto out_free_page;
877
878         loginuid = simple_strtoul(page, &tmp, 10);
879         if (tmp == page) {
880                 length = -EINVAL;
881                 goto out_free_page;
882
883         }
884         length = audit_set_loginuid(task, loginuid);
885         if (likely(length == 0))
886                 length = count;
887
888 out_free_page:
889         free_page((unsigned long) page);
890         return length;
891 }
892
893 static struct file_operations proc_loginuid_operations = {
894         .read           = proc_loginuid_read,
895         .write          = proc_loginuid_write,
896 };
897 #endif
898
899 #ifdef CONFIG_SECCOMP
900 static ssize_t seccomp_read(struct file *file, char __user *buf,
901                             size_t count, loff_t *ppos)
902 {
903         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
904         char __buf[20];
905         loff_t __ppos = *ppos;
906         size_t len;
907
908         /* no need to print the trailing zero, so use only len */
909         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
910         if (__ppos >= len)
911                 return 0;
912         if (count > len - __ppos)
913                 count = len - __ppos;
914         if (copy_to_user(buf, __buf + __ppos, count))
915                 return -EFAULT;
916         *ppos = __ppos + count;
917         return count;
918 }
919
920 static ssize_t seccomp_write(struct file *file, const char __user *buf,
921                              size_t count, loff_t *ppos)
922 {
923         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
924         char __buf[20], *end;
925         unsigned int seccomp_mode;
926
927         /* can set it only once to be even more secure */
928         if (unlikely(tsk->seccomp.mode))
929                 return -EPERM;
930
931         memset(__buf, 0, sizeof(__buf));
932         count = min(count, sizeof(__buf) - 1);
933         if (copy_from_user(__buf, buf, count))
934                 return -EFAULT;
935         seccomp_mode = simple_strtoul(__buf, &end, 0);
936         if (*end == '\n')
937                 end++;
938         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
939                 tsk->seccomp.mode = seccomp_mode;
940                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
941         } else
942                 return -EINVAL;
943         if (unlikely(!(end - __buf)))
944                 return -EIO;
945         return end - __buf;
946 }
947
948 static struct file_operations proc_seccomp_operations = {
949         .read           = seccomp_read,
950         .write          = seccomp_write,
951 };
952 #endif /* CONFIG_SECCOMP */
953
954 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
955 {
956         struct inode *inode = dentry->d_inode;
957         int error = -EACCES;
958
959         /* We don't need a base pointer in the /proc filesystem */
960         path_release(nd);
961
962         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
963                 goto out;
964         error = proc_check_root(inode);
965         if (error)
966                 goto out;
967
968         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
969         nd->last_type = LAST_BIND;
970 out:
971         return ERR_PTR(error);
972 }
973
974 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
975                             char __user *buffer, int buflen)
976 {
977         struct inode * inode;
978         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
979         int len;
980
981         if (!tmp)
982                 return -ENOMEM;
983                 
984         inode = dentry->d_inode;
985         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
986         len = PTR_ERR(path);
987         if (IS_ERR(path))
988                 goto out;
989         len = tmp + PAGE_SIZE - 1 - path;
990
991         if (len > buflen)
992                 len = buflen;
993         if (copy_to_user(buffer, path, len))
994                 len = -EFAULT;
995  out:
996         free_page((unsigned long)tmp);
997         return len;
998 }
999
1000 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1001 {
1002         int error = -EACCES;
1003         struct inode *inode = dentry->d_inode;
1004         struct dentry *de;
1005         struct vfsmount *mnt = NULL;
1006
1007         lock_kernel();
1008
1009         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1010                 goto out;
1011         error = proc_check_root(inode);
1012         if (error)
1013                 goto out;
1014
1015         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1016         if (error)
1017                 goto out;
1018
1019         error = do_proc_readlink(de, mnt, buffer, buflen);
1020         dput(de);
1021         mntput(mnt);
1022 out:
1023         unlock_kernel();
1024         return error;
1025 }
1026
1027 static struct inode_operations proc_pid_link_inode_operations = {
1028         .readlink       = proc_pid_readlink,
1029         .follow_link    = proc_pid_follow_link
1030 };
1031
1032 #define NUMBUF 10
1033
1034 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1035 {
1036         struct inode *inode = filp->f_dentry->d_inode;
1037         struct task_struct *p = proc_task(inode);
1038         unsigned int fd, tid, ino;
1039         int retval;
1040         char buf[NUMBUF];
1041         struct files_struct * files;
1042         struct fdtable *fdt;
1043
1044         retval = -ENOENT;
1045         if (!pid_alive(p))
1046                 goto out;
1047         retval = 0;
1048         tid = p->pid;
1049
1050         fd = filp->f_pos;
1051         switch (fd) {
1052                 case 0:
1053                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1054                                 goto out;
1055                         filp->f_pos++;
1056                 case 1:
1057                         ino = fake_ino(tid, PROC_TID_INO);
1058                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1059                                 goto out;
1060                         filp->f_pos++;
1061                 default:
1062                         files = get_files_struct(p);
1063                         if (!files)
1064                                 goto out;
1065                         spin_lock(&files->file_lock);
1066                         fdt = files_fdtable(files);
1067                         for (fd = filp->f_pos-2;
1068                              fd < fdt->max_fds;
1069                              fd++, filp->f_pos++) {
1070                                 unsigned int i,j;
1071
1072                                 if (!fcheck_files(files, fd))
1073                                         continue;
1074                                 spin_unlock(&files->file_lock);
1075
1076                                 j = NUMBUF;
1077                                 i = fd;
1078                                 do {
1079                                         j--;
1080                                         buf[j] = '0' + (i % 10);
1081                                         i /= 10;
1082                                 } while (i);
1083
1084                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1085                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1086                                         spin_lock(&files->file_lock);
1087                                         break;
1088                                 }
1089                                 spin_lock(&files->file_lock);
1090                         }
1091                         spin_unlock(&files->file_lock);
1092                         put_files_struct(files);
1093         }
1094 out:
1095         return retval;
1096 }
1097
1098 static int proc_pident_readdir(struct file *filp,
1099                 void *dirent, filldir_t filldir,
1100                 struct pid_entry *ents, unsigned int nents)
1101 {
1102         int i;
1103         int pid;
1104         struct dentry *dentry = filp->f_dentry;
1105         struct inode *inode = dentry->d_inode;
1106         struct pid_entry *p;
1107         ino_t ino;
1108         int ret;
1109
1110         ret = -ENOENT;
1111         if (!pid_alive(proc_task(inode)))
1112                 goto out;
1113
1114         ret = 0;
1115         pid = proc_task(inode)->pid;
1116         i = filp->f_pos;
1117         switch (i) {
1118         case 0:
1119                 ino = inode->i_ino;
1120                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1121                         goto out;
1122                 i++;
1123                 filp->f_pos++;
1124                 /* fall through */
1125         case 1:
1126                 ino = parent_ino(dentry);
1127                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1128                         goto out;
1129                 i++;
1130                 filp->f_pos++;
1131                 /* fall through */
1132         default:
1133                 i -= 2;
1134                 if (i >= nents) {
1135                         ret = 1;
1136                         goto out;
1137                 }
1138                 p = ents + i;
1139                 while (p->name) {
1140                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1141                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1142                                 goto out;
1143                         filp->f_pos++;
1144                         p++;
1145                 }
1146         }
1147
1148         ret = 1;
1149 out:
1150         return ret;
1151 }
1152
1153 static int proc_tgid_base_readdir(struct file * filp,
1154                              void * dirent, filldir_t filldir)
1155 {
1156         return proc_pident_readdir(filp,dirent,filldir,
1157                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1158 }
1159
1160 static int proc_tid_base_readdir(struct file * filp,
1161                              void * dirent, filldir_t filldir)
1162 {
1163         return proc_pident_readdir(filp,dirent,filldir,
1164                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1165 }
1166
1167 /* building an inode */
1168
1169 static int task_dumpable(struct task_struct *task)
1170 {
1171         int dumpable = 0;
1172         struct mm_struct *mm;
1173
1174         task_lock(task);
1175         mm = task->mm;
1176         if (mm)
1177                 dumpable = mm->dumpable;
1178         task_unlock(task);
1179         if(dumpable == 1)
1180                 return 1;
1181         return 0;
1182 }
1183
1184
1185 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1186 {
1187         struct inode * inode;
1188         struct proc_inode *ei;
1189
1190         /* We need a new inode */
1191         
1192         inode = new_inode(sb);
1193         if (!inode)
1194                 goto out;
1195
1196         /* Common stuff */
1197         ei = PROC_I(inode);
1198         ei->task = NULL;
1199         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1200         inode->i_ino = fake_ino(task->pid, ino);
1201
1202         if (!pid_alive(task))
1203                 goto out_unlock;
1204
1205         /*
1206          * grab the reference to task.
1207          */
1208         get_task_struct(task);
1209         ei->task = task;
1210         ei->type = ino;
1211         inode->i_uid = 0;
1212         inode->i_gid = 0;
1213         if (ino == PROC_TGID_INO || ino == PROC_TID_INO || task_dumpable(task)) {
1214                 inode->i_uid = task->euid;
1215                 inode->i_gid = task->egid;
1216         }
1217         security_task_to_inode(task, inode);
1218
1219 out:
1220         return inode;
1221
1222 out_unlock:
1223         ei->pde = NULL;
1224         iput(inode);
1225         return NULL;
1226 }
1227
1228 /* dentry stuff */
1229
1230 /*
1231  *      Exceptional case: normally we are not allowed to unhash a busy
1232  * directory. In this case, however, we can do it - no aliasing problems
1233  * due to the way we treat inodes.
1234  *
1235  * Rewrite the inode's ownerships here because the owning task may have
1236  * performed a setuid(), etc.
1237  */
1238 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1239 {
1240         struct inode *inode = dentry->d_inode;
1241         struct task_struct *task = proc_task(inode);
1242         if (pid_alive(task)) {
1243                 if (proc_type(inode) == PROC_TGID_INO || proc_type(inode) == PROC_TID_INO || task_dumpable(task)) {
1244                         inode->i_uid = task->euid;
1245                         inode->i_gid = task->egid;
1246                 } else {
1247                         inode->i_uid = 0;
1248                         inode->i_gid = 0;
1249                 }
1250                 security_task_to_inode(task, inode);
1251                 return 1;
1252         }
1253         d_drop(dentry);
1254         return 0;
1255 }
1256
1257 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1258 {
1259         struct inode *inode = dentry->d_inode;
1260         struct task_struct *task = proc_task(inode);
1261         int fd = proc_type(inode) - PROC_TID_FD_DIR;
1262         struct files_struct *files;
1263
1264         files = get_files_struct(task);
1265         if (files) {
1266                 spin_lock(&files->file_lock);
1267                 if (fcheck_files(files, fd)) {
1268                         spin_unlock(&files->file_lock);
1269                         put_files_struct(files);
1270                         if (task_dumpable(task)) {
1271                                 inode->i_uid = task->euid;
1272                                 inode->i_gid = task->egid;
1273                         } else {
1274                                 inode->i_uid = 0;
1275                                 inode->i_gid = 0;
1276                         }
1277                         security_task_to_inode(task, inode);
1278                         return 1;
1279                 }
1280                 spin_unlock(&files->file_lock);
1281                 put_files_struct(files);
1282         }
1283         d_drop(dentry);
1284         return 0;
1285 }
1286
1287 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1288 {
1289         struct task_struct *task = proc_task(inode);
1290         spin_lock(&task->proc_lock);
1291         if (task->proc_dentry == dentry)
1292                 task->proc_dentry = NULL;
1293         spin_unlock(&task->proc_lock);
1294         iput(inode);
1295 }
1296
1297 static int pid_delete_dentry(struct dentry * dentry)
1298 {
1299         /* Is the task we represent dead?
1300          * If so, then don't put the dentry on the lru list,
1301          * kill it immediately.
1302          */
1303         return !pid_alive(proc_task(dentry->d_inode));
1304 }
1305
1306 static struct dentry_operations tid_fd_dentry_operations =
1307 {
1308         .d_revalidate   = tid_fd_revalidate,
1309         .d_delete       = pid_delete_dentry,
1310 };
1311
1312 static struct dentry_operations pid_dentry_operations =
1313 {
1314         .d_revalidate   = pid_revalidate,
1315         .d_delete       = pid_delete_dentry,
1316 };
1317
1318 static struct dentry_operations pid_base_dentry_operations =
1319 {
1320         .d_revalidate   = pid_revalidate,
1321         .d_iput         = pid_base_iput,
1322         .d_delete       = pid_delete_dentry,
1323 };
1324
1325 /* Lookups */
1326
1327 static unsigned name_to_int(struct dentry *dentry)
1328 {
1329         const char *name = dentry->d_name.name;
1330         int len = dentry->d_name.len;
1331         unsigned n = 0;
1332
1333         if (len > 1 && *name == '0')
1334                 goto out;
1335         while (len-- > 0) {
1336                 unsigned c = *name++ - '0';
1337                 if (c > 9)
1338                         goto out;
1339                 if (n >= (~0U-9)/10)
1340                         goto out;
1341                 n *= 10;
1342                 n += c;
1343         }
1344         return n;
1345 out:
1346         return ~0U;
1347 }
1348
1349 /* SMP-safe */
1350 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1351 {
1352         struct task_struct *task = proc_task(dir);
1353         unsigned fd = name_to_int(dentry);
1354         struct file * file;
1355         struct files_struct * files;
1356         struct inode *inode;
1357         struct proc_inode *ei;
1358
1359         if (fd == ~0U)
1360                 goto out;
1361         if (!pid_alive(task))
1362                 goto out;
1363
1364         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1365         if (!inode)
1366                 goto out;
1367         ei = PROC_I(inode);
1368         files = get_files_struct(task);
1369         if (!files)
1370                 goto out_unlock;
1371         inode->i_mode = S_IFLNK;
1372         spin_lock(&files->file_lock);
1373         file = fcheck_files(files, fd);
1374         if (!file)
1375                 goto out_unlock2;
1376         if (file->f_mode & 1)
1377                 inode->i_mode |= S_IRUSR | S_IXUSR;
1378         if (file->f_mode & 2)
1379                 inode->i_mode |= S_IWUSR | S_IXUSR;
1380         spin_unlock(&files->file_lock);
1381         put_files_struct(files);
1382         inode->i_op = &proc_pid_link_inode_operations;
1383         inode->i_size = 64;
1384         ei->op.proc_get_link = proc_fd_link;
1385         dentry->d_op = &tid_fd_dentry_operations;
1386         d_add(dentry, inode);
1387         return NULL;
1388
1389 out_unlock2:
1390         spin_unlock(&files->file_lock);
1391         put_files_struct(files);
1392 out_unlock:
1393         iput(inode);
1394 out:
1395         return ERR_PTR(-ENOENT);
1396 }
1397
1398 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1399 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1400
1401 static struct file_operations proc_fd_operations = {
1402         .read           = generic_read_dir,
1403         .readdir        = proc_readfd,
1404 };
1405
1406 static struct file_operations proc_task_operations = {
1407         .read           = generic_read_dir,
1408         .readdir        = proc_task_readdir,
1409 };
1410
1411 /*
1412  * proc directories can do almost nothing..
1413  */
1414 static struct inode_operations proc_fd_inode_operations = {
1415         .lookup         = proc_lookupfd,
1416         .permission     = proc_permission,
1417 };
1418
1419 static struct inode_operations proc_task_inode_operations = {
1420         .lookup         = proc_task_lookup,
1421         .permission     = proc_permission,
1422 };
1423
1424 #ifdef CONFIG_SECURITY
1425 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1426                                   size_t count, loff_t *ppos)
1427 {
1428         struct inode * inode = file->f_dentry->d_inode;
1429         unsigned long page;
1430         ssize_t length;
1431         struct task_struct *task = proc_task(inode);
1432
1433         if (count > PAGE_SIZE)
1434                 count = PAGE_SIZE;
1435         if (!(page = __get_free_page(GFP_KERNEL)))
1436                 return -ENOMEM;
1437
1438         length = security_getprocattr(task, 
1439                                       (char*)file->f_dentry->d_name.name, 
1440                                       (void*)page, count);
1441         if (length >= 0)
1442                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1443         free_page(page);
1444         return length;
1445 }
1446
1447 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1448                                    size_t count, loff_t *ppos)
1449
1450         struct inode * inode = file->f_dentry->d_inode;
1451         char *page; 
1452         ssize_t length; 
1453         struct task_struct *task = proc_task(inode); 
1454
1455         if (count > PAGE_SIZE) 
1456                 count = PAGE_SIZE; 
1457         if (*ppos != 0) {
1458                 /* No partial writes. */
1459                 return -EINVAL;
1460         }
1461         page = (char*)__get_free_page(GFP_USER); 
1462         if (!page) 
1463                 return -ENOMEM;
1464         length = -EFAULT; 
1465         if (copy_from_user(page, buf, count)) 
1466                 goto out;
1467
1468         length = security_setprocattr(task, 
1469                                       (char*)file->f_dentry->d_name.name, 
1470                                       (void*)page, count);
1471 out:
1472         free_page((unsigned long) page);
1473         return length;
1474
1475
1476 static struct file_operations proc_pid_attr_operations = {
1477         .read           = proc_pid_attr_read,
1478         .write          = proc_pid_attr_write,
1479 };
1480
1481 static struct file_operations proc_tid_attr_operations;
1482 static struct inode_operations proc_tid_attr_inode_operations;
1483 static struct file_operations proc_tgid_attr_operations;
1484 static struct inode_operations proc_tgid_attr_inode_operations;
1485 #endif
1486
1487 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1488
1489 /* SMP-safe */
1490 static struct dentry *proc_pident_lookup(struct inode *dir, 
1491                                          struct dentry *dentry,
1492                                          struct pid_entry *ents)
1493 {
1494         struct inode *inode;
1495         int error;
1496         struct task_struct *task = proc_task(dir);
1497         struct pid_entry *p;
1498         struct proc_inode *ei;
1499
1500         error = -ENOENT;
1501         inode = NULL;
1502
1503         if (!pid_alive(task))
1504                 goto out;
1505
1506         for (p = ents; p->name; p++) {
1507                 if (p->len != dentry->d_name.len)
1508                         continue;
1509                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1510                         break;
1511         }
1512         if (!p->name)
1513                 goto out;
1514
1515         error = -EINVAL;
1516         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1517         if (!inode)
1518                 goto out;
1519
1520         ei = PROC_I(inode);
1521         inode->i_mode = p->mode;
1522         /*
1523          * Yes, it does not scale. And it should not. Don't add
1524          * new entries into /proc/<tgid>/ without very good reasons.
1525          */
1526         switch(p->type) {
1527                 case PROC_TGID_TASK:
1528                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1529                         inode->i_op = &proc_task_inode_operations;
1530                         inode->i_fop = &proc_task_operations;
1531                         break;
1532                 case PROC_TID_FD:
1533                 case PROC_TGID_FD:
1534                         inode->i_nlink = 2;
1535                         inode->i_op = &proc_fd_inode_operations;
1536                         inode->i_fop = &proc_fd_operations;
1537                         break;
1538                 case PROC_TID_EXE:
1539                 case PROC_TGID_EXE:
1540                         inode->i_op = &proc_pid_link_inode_operations;
1541                         ei->op.proc_get_link = proc_exe_link;
1542                         break;
1543                 case PROC_TID_CWD:
1544                 case PROC_TGID_CWD:
1545                         inode->i_op = &proc_pid_link_inode_operations;
1546                         ei->op.proc_get_link = proc_cwd_link;
1547                         break;
1548                 case PROC_TID_ROOT:
1549                 case PROC_TGID_ROOT:
1550                         inode->i_op = &proc_pid_link_inode_operations;
1551                         ei->op.proc_get_link = proc_root_link;
1552                         break;
1553                 case PROC_TID_ENVIRON:
1554                 case PROC_TGID_ENVIRON:
1555                         inode->i_fop = &proc_info_file_operations;
1556                         ei->op.proc_read = proc_pid_environ;
1557                         break;
1558                 case PROC_TID_AUXV:
1559                 case PROC_TGID_AUXV:
1560                         inode->i_fop = &proc_info_file_operations;
1561                         ei->op.proc_read = proc_pid_auxv;
1562                         break;
1563                 case PROC_TID_STATUS:
1564                 case PROC_TGID_STATUS:
1565                         inode->i_fop = &proc_info_file_operations;
1566                         ei->op.proc_read = proc_pid_status;
1567                         break;
1568                 case PROC_TID_STAT:
1569                         inode->i_fop = &proc_info_file_operations;
1570                         ei->op.proc_read = proc_tid_stat;
1571                         break;
1572                 case PROC_TGID_STAT:
1573                         inode->i_fop = &proc_info_file_operations;
1574                         ei->op.proc_read = proc_tgid_stat;
1575                         break;
1576                 case PROC_TID_CMDLINE:
1577                 case PROC_TGID_CMDLINE:
1578                         inode->i_fop = &proc_info_file_operations;
1579                         ei->op.proc_read = proc_pid_cmdline;
1580                         break;
1581                 case PROC_TID_STATM:
1582                 case PROC_TGID_STATM:
1583                         inode->i_fop = &proc_info_file_operations;
1584                         ei->op.proc_read = proc_pid_statm;
1585                         break;
1586                 case PROC_TID_MAPS:
1587                 case PROC_TGID_MAPS:
1588                         inode->i_fop = &proc_maps_operations;
1589                         break;
1590 #ifdef CONFIG_NUMA
1591                 case PROC_TID_NUMA_MAPS:
1592                 case PROC_TGID_NUMA_MAPS:
1593                         inode->i_fop = &proc_numa_maps_operations;
1594                         break;
1595 #endif
1596                 case PROC_TID_MEM:
1597                 case PROC_TGID_MEM:
1598                         inode->i_op = &proc_mem_inode_operations;
1599                         inode->i_fop = &proc_mem_operations;
1600                         break;
1601 #ifdef CONFIG_SECCOMP
1602                 case PROC_TID_SECCOMP:
1603                 case PROC_TGID_SECCOMP:
1604                         inode->i_fop = &proc_seccomp_operations;
1605                         break;
1606 #endif /* CONFIG_SECCOMP */
1607                 case PROC_TID_MOUNTS:
1608                 case PROC_TGID_MOUNTS:
1609                         inode->i_fop = &proc_mounts_operations;
1610                         break;
1611                 case PROC_TID_SMAPS:
1612                 case PROC_TGID_SMAPS:
1613                         inode->i_fop = &proc_smaps_operations;
1614                         break;
1615 #ifdef CONFIG_SECURITY
1616                 case PROC_TID_ATTR:
1617                         inode->i_nlink = 2;
1618                         inode->i_op = &proc_tid_attr_inode_operations;
1619                         inode->i_fop = &proc_tid_attr_operations;
1620                         break;
1621                 case PROC_TGID_ATTR:
1622                         inode->i_nlink = 2;
1623                         inode->i_op = &proc_tgid_attr_inode_operations;
1624                         inode->i_fop = &proc_tgid_attr_operations;
1625                         break;
1626                 case PROC_TID_ATTR_CURRENT:
1627                 case PROC_TGID_ATTR_CURRENT:
1628                 case PROC_TID_ATTR_PREV:
1629                 case PROC_TGID_ATTR_PREV:
1630                 case PROC_TID_ATTR_EXEC:
1631                 case PROC_TGID_ATTR_EXEC:
1632                 case PROC_TID_ATTR_FSCREATE:
1633                 case PROC_TGID_ATTR_FSCREATE:
1634                         inode->i_fop = &proc_pid_attr_operations;
1635                         break;
1636 #endif
1637 #ifdef CONFIG_KALLSYMS
1638                 case PROC_TID_WCHAN:
1639                 case PROC_TGID_WCHAN:
1640                         inode->i_fop = &proc_info_file_operations;
1641                         ei->op.proc_read = proc_pid_wchan;
1642                         break;
1643 #endif
1644 #ifdef CONFIG_SCHEDSTATS
1645                 case PROC_TID_SCHEDSTAT:
1646                 case PROC_TGID_SCHEDSTAT:
1647                         inode->i_fop = &proc_info_file_operations;
1648                         ei->op.proc_read = proc_pid_schedstat;
1649                         break;
1650 #endif
1651 #ifdef CONFIG_CPUSETS
1652                 case PROC_TID_CPUSET:
1653                 case PROC_TGID_CPUSET:
1654                         inode->i_fop = &proc_cpuset_operations;
1655                         break;
1656 #endif
1657                 case PROC_TID_OOM_SCORE:
1658                 case PROC_TGID_OOM_SCORE:
1659                         inode->i_fop = &proc_info_file_operations;
1660                         ei->op.proc_read = proc_oom_score;
1661                         break;
1662                 case PROC_TID_OOM_ADJUST:
1663                 case PROC_TGID_OOM_ADJUST:
1664                         inode->i_fop = &proc_oom_adjust_operations;
1665                         break;
1666 #ifdef CONFIG_AUDITSYSCALL
1667                 case PROC_TID_LOGINUID:
1668                 case PROC_TGID_LOGINUID:
1669                         inode->i_fop = &proc_loginuid_operations;
1670                         break;
1671 #endif
1672                 default:
1673                         printk("procfs: impossible type (%d)",p->type);
1674                         iput(inode);
1675                         return ERR_PTR(-EINVAL);
1676         }
1677         dentry->d_op = &pid_dentry_operations;
1678         d_add(dentry, inode);
1679         return NULL;
1680
1681 out:
1682         return ERR_PTR(error);
1683 }
1684
1685 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1686         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1687 }
1688
1689 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1690         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1691 }
1692
1693 static struct file_operations proc_tgid_base_operations = {
1694         .read           = generic_read_dir,
1695         .readdir        = proc_tgid_base_readdir,
1696 };
1697
1698 static struct file_operations proc_tid_base_operations = {
1699         .read           = generic_read_dir,
1700         .readdir        = proc_tid_base_readdir,
1701 };
1702
1703 static struct inode_operations proc_tgid_base_inode_operations = {
1704         .lookup         = proc_tgid_base_lookup,
1705 };
1706
1707 static struct inode_operations proc_tid_base_inode_operations = {
1708         .lookup         = proc_tid_base_lookup,
1709 };
1710
1711 #ifdef CONFIG_SECURITY
1712 static int proc_tgid_attr_readdir(struct file * filp,
1713                              void * dirent, filldir_t filldir)
1714 {
1715         return proc_pident_readdir(filp,dirent,filldir,
1716                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1717 }
1718
1719 static int proc_tid_attr_readdir(struct file * filp,
1720                              void * dirent, filldir_t filldir)
1721 {
1722         return proc_pident_readdir(filp,dirent,filldir,
1723                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1724 }
1725
1726 static struct file_operations proc_tgid_attr_operations = {
1727         .read           = generic_read_dir,
1728         .readdir        = proc_tgid_attr_readdir,
1729 };
1730
1731 static struct file_operations proc_tid_attr_operations = {
1732         .read           = generic_read_dir,
1733         .readdir        = proc_tid_attr_readdir,
1734 };
1735
1736 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1737                                 struct dentry *dentry, struct nameidata *nd)
1738 {
1739         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1740 }
1741
1742 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1743                                 struct dentry *dentry, struct nameidata *nd)
1744 {
1745         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1746 }
1747
1748 static struct inode_operations proc_tgid_attr_inode_operations = {
1749         .lookup         = proc_tgid_attr_lookup,
1750 };
1751
1752 static struct inode_operations proc_tid_attr_inode_operations = {
1753         .lookup         = proc_tid_attr_lookup,
1754 };
1755 #endif
1756
1757 /*
1758  * /proc/self:
1759  */
1760 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1761                               int buflen)
1762 {
1763         char tmp[30];
1764         sprintf(tmp, "%d", current->tgid);
1765         return vfs_readlink(dentry,buffer,buflen,tmp);
1766 }
1767
1768 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1769 {
1770         char tmp[30];
1771         sprintf(tmp, "%d", current->tgid);
1772         return ERR_PTR(vfs_follow_link(nd,tmp));
1773 }       
1774
1775 static struct inode_operations proc_self_inode_operations = {
1776         .readlink       = proc_self_readlink,
1777         .follow_link    = proc_self_follow_link,
1778 };
1779
1780 /**
1781  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1782  * @p: task that should be flushed.
1783  *
1784  * Drops the /proc/@pid dcache entry from the hash chains.
1785  *
1786  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1787  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1788  * if the pid value is immediately reused. This is enforced by
1789  * - caller must acquire spin_lock(p->proc_lock)
1790  * - must be called before detach_pid()
1791  * - proc_pid_lookup acquires proc_lock, and checks that
1792  *   the target is not dead by looking at the attach count
1793  *   of PIDTYPE_PID.
1794  */
1795
1796 struct dentry *proc_pid_unhash(struct task_struct *p)
1797 {
1798         struct dentry *proc_dentry;
1799
1800         proc_dentry = p->proc_dentry;
1801         if (proc_dentry != NULL) {
1802
1803                 spin_lock(&dcache_lock);
1804                 spin_lock(&proc_dentry->d_lock);
1805                 if (!d_unhashed(proc_dentry)) {
1806                         dget_locked(proc_dentry);
1807                         __d_drop(proc_dentry);
1808                         spin_unlock(&proc_dentry->d_lock);
1809                 } else {
1810                         spin_unlock(&proc_dentry->d_lock);
1811                         proc_dentry = NULL;
1812                 }
1813                 spin_unlock(&dcache_lock);
1814         }
1815         return proc_dentry;
1816 }
1817
1818 /**
1819  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1820  * @proc_dentry: directoy to prune.
1821  *
1822  * Shrink the /proc directory that was used by the just killed thread.
1823  */
1824         
1825 void proc_pid_flush(struct dentry *proc_dentry)
1826 {
1827         might_sleep();
1828         if(proc_dentry != NULL) {
1829                 shrink_dcache_parent(proc_dentry);
1830                 dput(proc_dentry);
1831         }
1832 }
1833
1834 /* SMP-safe */
1835 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1836 {
1837         struct task_struct *task;
1838         struct inode *inode;
1839         struct proc_inode *ei;
1840         unsigned tgid;
1841         int died;
1842
1843         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
1844                 inode = new_inode(dir->i_sb);
1845                 if (!inode)
1846                         return ERR_PTR(-ENOMEM);
1847                 ei = PROC_I(inode);
1848                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1849                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
1850                 ei->pde = NULL;
1851                 inode->i_mode = S_IFLNK|S_IRWXUGO;
1852                 inode->i_uid = inode->i_gid = 0;
1853                 inode->i_size = 64;
1854                 inode->i_op = &proc_self_inode_operations;
1855                 d_add(dentry, inode);
1856                 return NULL;
1857         }
1858         tgid = name_to_int(dentry);
1859         if (tgid == ~0U)
1860                 goto out;
1861
1862         read_lock(&tasklist_lock);
1863         task = find_task_by_pid(tgid);
1864         if (task)
1865                 get_task_struct(task);
1866         read_unlock(&tasklist_lock);
1867         if (!task)
1868                 goto out;
1869
1870         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
1871
1872
1873         if (!inode) {
1874                 put_task_struct(task);
1875                 goto out;
1876         }
1877         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1878         inode->i_op = &proc_tgid_base_inode_operations;
1879         inode->i_fop = &proc_tgid_base_operations;
1880         inode->i_flags|=S_IMMUTABLE;
1881 #ifdef CONFIG_SECURITY
1882         inode->i_nlink = 5;
1883 #else
1884         inode->i_nlink = 4;
1885 #endif
1886
1887         dentry->d_op = &pid_base_dentry_operations;
1888
1889         died = 0;
1890         d_add(dentry, inode);
1891         spin_lock(&task->proc_lock);
1892         task->proc_dentry = dentry;
1893         if (!pid_alive(task)) {
1894                 dentry = proc_pid_unhash(task);
1895                 died = 1;
1896         }
1897         spin_unlock(&task->proc_lock);
1898
1899         put_task_struct(task);
1900         if (died) {
1901                 proc_pid_flush(dentry);
1902                 goto out;
1903         }
1904         return NULL;
1905 out:
1906         return ERR_PTR(-ENOENT);
1907 }
1908
1909 /* SMP-safe */
1910 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1911 {
1912         struct task_struct *task;
1913         struct task_struct *leader = proc_task(dir);
1914         struct inode *inode;
1915         unsigned tid;
1916
1917         tid = name_to_int(dentry);
1918         if (tid == ~0U)
1919                 goto out;
1920
1921         read_lock(&tasklist_lock);
1922         task = find_task_by_pid(tid);
1923         if (task)
1924                 get_task_struct(task);
1925         read_unlock(&tasklist_lock);
1926         if (!task)
1927                 goto out;
1928         if (leader->tgid != task->tgid)
1929                 goto out_drop_task;
1930
1931         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
1932
1933
1934         if (!inode)
1935                 goto out_drop_task;
1936         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1937         inode->i_op = &proc_tid_base_inode_operations;
1938         inode->i_fop = &proc_tid_base_operations;
1939         inode->i_flags|=S_IMMUTABLE;
1940 #ifdef CONFIG_SECURITY
1941         inode->i_nlink = 4;
1942 #else
1943         inode->i_nlink = 3;
1944 #endif
1945
1946         dentry->d_op = &pid_base_dentry_operations;
1947
1948         d_add(dentry, inode);
1949
1950         put_task_struct(task);
1951         return NULL;
1952 out_drop_task:
1953         put_task_struct(task);
1954 out:
1955         return ERR_PTR(-ENOENT);
1956 }
1957
1958 #define PROC_NUMBUF 10
1959 #define PROC_MAXPIDS 20
1960
1961 /*
1962  * Get a few tgid's to return for filldir - we need to hold the
1963  * tasklist lock while doing this, and we must release it before
1964  * we actually do the filldir itself, so we use a temp buffer..
1965  */
1966 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
1967 {
1968         struct task_struct *p;
1969         int nr_tgids = 0;
1970
1971         index--;
1972         read_lock(&tasklist_lock);
1973         p = NULL;
1974         if (version) {
1975                 p = find_task_by_pid(version);
1976                 if (p && !thread_group_leader(p))
1977                         p = NULL;
1978         }
1979
1980         if (p)
1981                 index = 0;
1982         else
1983                 p = next_task(&init_task);
1984
1985         for ( ; p != &init_task; p = next_task(p)) {
1986                 int tgid = p->pid;
1987                 if (!pid_alive(p))
1988                         continue;
1989                 if (--index >= 0)
1990                         continue;
1991                 tgids[nr_tgids] = tgid;
1992                 nr_tgids++;
1993                 if (nr_tgids >= PROC_MAXPIDS)
1994                         break;
1995         }
1996         read_unlock(&tasklist_lock);
1997         return nr_tgids;
1998 }
1999
2000 /*
2001  * Get a few tid's to return for filldir - we need to hold the
2002  * tasklist lock while doing this, and we must release it before
2003  * we actually do the filldir itself, so we use a temp buffer..
2004  */
2005 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2006 {
2007         struct task_struct *leader_task = proc_task(dir);
2008         struct task_struct *task = leader_task;
2009         int nr_tids = 0;
2010
2011         index -= 2;
2012         read_lock(&tasklist_lock);
2013         /*
2014          * The starting point task (leader_task) might be an already
2015          * unlinked task, which cannot be used to access the task-list
2016          * via next_thread().
2017          */
2018         if (pid_alive(task)) do {
2019                 int tid = task->pid;
2020
2021                 if (--index >= 0)
2022                         continue;
2023                 if (tids != NULL)
2024                         tids[nr_tids] = tid;
2025                 nr_tids++;
2026                 if (nr_tids >= PROC_MAXPIDS)
2027                         break;
2028         } while ((task = next_thread(task)) != leader_task);
2029         read_unlock(&tasklist_lock);
2030         return nr_tids;
2031 }
2032
2033 /* for the /proc/ directory itself, after non-process stuff has been done */
2034 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2035 {
2036         unsigned int tgid_array[PROC_MAXPIDS];
2037         char buf[PROC_NUMBUF];
2038         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2039         unsigned int nr_tgids, i;
2040         int next_tgid;
2041
2042         if (!nr) {
2043                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2044                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2045                         return 0;
2046                 filp->f_pos++;
2047                 nr++;
2048         }
2049
2050         /* f_version caches the tgid value that the last readdir call couldn't
2051          * return. lseek aka telldir automagically resets f_version to 0.
2052          */
2053         next_tgid = filp->f_version;
2054         filp->f_version = 0;
2055         for (;;) {
2056                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2057                 if (!nr_tgids) {
2058                         /* no more entries ! */
2059                         break;
2060                 }
2061                 next_tgid = 0;
2062
2063                 /* do not use the last found pid, reserve it for next_tgid */
2064                 if (nr_tgids == PROC_MAXPIDS) {
2065                         nr_tgids--;
2066                         next_tgid = tgid_array[nr_tgids];
2067                 }
2068
2069                 for (i=0;i<nr_tgids;i++) {
2070                         int tgid = tgid_array[i];
2071                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2072                         unsigned long j = PROC_NUMBUF;
2073
2074                         do
2075                                 buf[--j] = '0' + (tgid % 10);
2076                         while ((tgid /= 10) != 0);
2077
2078                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2079                                 /* returning this tgid failed, save it as the first
2080                                  * pid for the next readir call */
2081                                 filp->f_version = tgid_array[i];
2082                                 goto out;
2083                         }
2084                         filp->f_pos++;
2085                         nr++;
2086                 }
2087         }
2088 out:
2089         return 0;
2090 }
2091
2092 /* for the /proc/TGID/task/ directories */
2093 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2094 {
2095         unsigned int tid_array[PROC_MAXPIDS];
2096         char buf[PROC_NUMBUF];
2097         unsigned int nr_tids, i;
2098         struct dentry *dentry = filp->f_dentry;
2099         struct inode *inode = dentry->d_inode;
2100         int retval = -ENOENT;
2101         ino_t ino;
2102         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2103
2104         if (!pid_alive(proc_task(inode)))
2105                 goto out;
2106         retval = 0;
2107
2108         switch (pos) {
2109         case 0:
2110                 ino = inode->i_ino;
2111                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2112                         goto out;
2113                 pos++;
2114                 /* fall through */
2115         case 1:
2116                 ino = parent_ino(dentry);
2117                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2118                         goto out;
2119                 pos++;
2120                 /* fall through */
2121         }
2122
2123         nr_tids = get_tid_list(pos, tid_array, inode);
2124         inode->i_nlink = pos + nr_tids;
2125
2126         for (i = 0; i < nr_tids; i++) {
2127                 unsigned long j = PROC_NUMBUF;
2128                 int tid = tid_array[i];
2129
2130                 ino = fake_ino(tid,PROC_TID_INO);
2131
2132                 do
2133                         buf[--j] = '0' + (tid % 10);
2134                 while ((tid /= 10) != 0);
2135
2136                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2137                         break;
2138                 pos++;
2139         }
2140 out:
2141         filp->f_pos = pos;
2142         return retval;
2143 }