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/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/string.h>
  61 #include <linux/seq_file.h>
  62 #include <linux/namei.h>
  63 #include <linux/namespace.h>
  64 #include <linux/mm.h>
  65 #include <linux/smp_lock.h>
  66 #include <linux/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/mount.h>
  69 #include <linux/security.h>
  70 #include <linux/ptrace.h>
  71 #include <linux/seccomp.h>
  72 #include <linux/cpuset.h>
  73 #include <linux/audit.h>
  74 #include <linux/poll.h>
  75 #include "internal.h"
  76
  77 /* NOTE:
  78  *      Implementing inode permission operations in /proc is almost
  79  *      certainly an error.  Permission checks need to happen during
  80  *      each system call not at open time.  The reason is that most of
  81  *      what we wish to check for permissions in /proc varies at runtime.
  82  *
  83  *      The classic example of a problem is opening file descriptors
  84  *      in /proc for a task before it execs a suid executable.
  85  */
  86
  87 /*
  88  * For hysterical raisins we keep the same inumbers as in the old procfs.
  89  * Feel free to change the macro below - just keep the range distinct from
  90  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
  91  * As soon as we'll get a separate superblock we will be able to forget
  92  * about magical ranges too.
  93  */
  94
  95 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
  96
  97 enum pid_directory_inos {
  98         PROC_TGID_INO = 2,
  99         PROC_TGID_TASK,
 100         PROC_TGID_STATUS,
 101         PROC_TGID_MEM,
 102 #ifdef CONFIG_SECCOMP
 103         PROC_TGID_SECCOMP,
 104 #endif
 105         PROC_TGID_CWD,
 106         PROC_TGID_ROOT,
 107         PROC_TGID_EXE,
 108         PROC_TGID_FD,
 109         PROC_TGID_ENVIRON,
 110         PROC_TGID_AUXV,
 111         PROC_TGID_CMDLINE,
 112         PROC_TGID_STAT,
 113         PROC_TGID_STATM,
 114         PROC_TGID_MAPS,
 115         PROC_TGID_NUMA_MAPS,
 116         PROC_TGID_MOUNTS,
 117         PROC_TGID_MOUNTSTATS,
 118         PROC_TGID_WCHAN,
 119 #ifdef CONFIG_MMU
 120         PROC_TGID_SMAPS,
 121 #endif
 122 #ifdef CONFIG_SCHEDSTATS
 123         PROC_TGID_SCHEDSTAT,
 124 #endif
 125 #ifdef CONFIG_CPUSETS
 126         PROC_TGID_CPUSET,
 127 #endif
 128 #ifdef CONFIG_SECURITY
 129         PROC_TGID_ATTR,
 130         PROC_TGID_ATTR_CURRENT,
 131         PROC_TGID_ATTR_PREV,
 132         PROC_TGID_ATTR_EXEC,
 133         PROC_TGID_ATTR_FSCREATE,
 134         PROC_TGID_ATTR_KEYCREATE,
 135 #endif
 136 #ifdef CONFIG_AUDITSYSCALL
 137         PROC_TGID_LOGINUID,
 138 #endif
 139         PROC_TGID_OOM_SCORE,
 140         PROC_TGID_OOM_ADJUST,
 141         PROC_TID_INO,
 142         PROC_TID_STATUS,
 143         PROC_TID_MEM,
 144 #ifdef CONFIG_SECCOMP
 145         PROC_TID_SECCOMP,
 146 #endif
 147         PROC_TID_CWD,
 148         PROC_TID_ROOT,
 149         PROC_TID_EXE,
 150         PROC_TID_FD,
 151         PROC_TID_ENVIRON,
 152         PROC_TID_AUXV,
 153         PROC_TID_CMDLINE,
 154         PROC_TID_STAT,
 155         PROC_TID_STATM,
 156         PROC_TID_MAPS,
 157         PROC_TID_NUMA_MAPS,
 158         PROC_TID_MOUNTS,
 159         PROC_TID_MOUNTSTATS,
 160         PROC_TID_WCHAN,
 161 #ifdef CONFIG_MMU
 162         PROC_TID_SMAPS,
 163 #endif
 164 #ifdef CONFIG_SCHEDSTATS
 165         PROC_TID_SCHEDSTAT,
 166 #endif
 167 #ifdef CONFIG_CPUSETS
 168         PROC_TID_CPUSET,
 169 #endif
 170 #ifdef CONFIG_SECURITY
 171         PROC_TID_ATTR,
 172         PROC_TID_ATTR_CURRENT,
 173         PROC_TID_ATTR_PREV,
 174         PROC_TID_ATTR_EXEC,
 175         PROC_TID_ATTR_FSCREATE,
 176         PROC_TID_ATTR_KEYCREATE,
 177 #endif
 178 #ifdef CONFIG_AUDITSYSCALL
 179         PROC_TID_LOGINUID,
 180 #endif
 181         PROC_TID_OOM_SCORE,
 182         PROC_TID_OOM_ADJUST,
 183
 184         /* Add new entries before this */
 185         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
 186 };
 187
 188 struct pid_entry {
 189         int type;
 190         int len;
 191         char *name;
 192         mode_t mode;
 193 };
 194
 195 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
 196
 197 static struct pid_entry tgid_base_stuff[] = {
 198         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
 199         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 200         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
 201         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
 202         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
 203         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 204         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 205         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
 206         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 207 #ifdef CONFIG_NUMA
 208         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
 209 #endif
 210         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 211 #ifdef CONFIG_SECCOMP
 212         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 213 #endif
 214         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
 215         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
 216         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
 217         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
 218         E(PROC_TGID_MOUNTSTATS, "mountstats", S_IFREG|S_IRUSR),
 219 #ifdef CONFIG_MMU
 220         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
 221 #endif
 222 #ifdef CONFIG_SECURITY
 223         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 224 #endif
 225 #ifdef CONFIG_KALLSYMS
 226         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
 227 #endif
 228 #ifdef CONFIG_SCHEDSTATS
 229         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
 230 #endif
 231 #ifdef CONFIG_CPUSETS
 232         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
 233 #endif
 234         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
 235         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 236 #ifdef CONFIG_AUDITSYSCALL
 237         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 238 #endif
 239         {0,0,NULL,0}
 240 };
 241 static struct pid_entry tid_base_stuff[] = {
 242         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 243         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
 244         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
 245         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
 246         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 247         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 248         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
 249         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 250 #ifdef CONFIG_NUMA
 251         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
 252 #endif
 253         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 254 #ifdef CONFIG_SECCOMP
 255         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 256 #endif
 257         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
 258         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
 259         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
 260         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
 261 #ifdef CONFIG_MMU
 262         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
 263 #endif
 264 #ifdef CONFIG_SECURITY
 265         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 266 #endif
 267 #ifdef CONFIG_KALLSYMS
 268         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
 269 #endif
 270 #ifdef CONFIG_SCHEDSTATS
 271         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
 272 #endif
 273 #ifdef CONFIG_CPUSETS
 274         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
 275 #endif
 276         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
 277         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 278 #ifdef CONFIG_AUDITSYSCALL
 279         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 280 #endif
 281         {0,0,NULL,0}
 282 };
 283
 284 #ifdef CONFIG_SECURITY
 285 static struct pid_entry tgid_attr_stuff[] = {
 286         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
 287         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
 288         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 289         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 290         E(PROC_TGID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 291         {0,0,NULL,0}
 292 };
 293 static struct pid_entry tid_attr_stuff[] = {
 294         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
 295         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
 296         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 297         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 298         E(PROC_TID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 299         {0,0,NULL,0}
 300 };
 301 #endif
 302
 303 #undef E
 304
 305 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 306 {
 307         struct task_struct *task = proc_task(inode);
 308         struct files_struct *files;
 309         struct file *file;
 310         int fd = proc_fd(inode);
 311
 312         files = get_files_struct(task);
 313         if (files) {
 314                 /*
 315                  * We are not taking a ref to the file structure, so we must
 316                  * hold ->file_lock.
 317                  */
 318                 spin_lock(&files->file_lock);
 319                 file = fcheck_files(files, fd);
 320                 if (file) {
 321                         *mnt = mntget(file->f_vfsmnt);
 322                         *dentry = dget(file->f_dentry);
 323                         spin_unlock(&files->file_lock);
 324                         put_files_struct(files);
 325                         return 0;
 326                 }
 327                 spin_unlock(&files->file_lock);
 328                 put_files_struct(files);
 329         }
 330         return -ENOENT;
 331 }
 332
 333 static struct fs_struct *get_fs_struct(struct task_struct *task)
 334 {
 335         struct fs_struct *fs;
 336         task_lock(task);
 337         fs = task->fs;
 338         if(fs)
 339                 atomic_inc(&fs->count);
 340         task_unlock(task);
 341         return fs;
 342 }
 343
 344 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 345 {
 346         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 347         int result = -ENOENT;
 348         if (fs) {
 349                 read_lock(&fs->lock);
 350                 *mnt = mntget(fs->pwdmnt);
 351                 *dentry = dget(fs->pwd);
 352                 read_unlock(&fs->lock);
 353                 result = 0;
 354                 put_fs_struct(fs);
 355         }
 356         return result;
 357 }
 358
 359 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 360 {
 361         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 362         int result = -ENOENT;
 363         if (fs) {
 364                 read_lock(&fs->lock);
 365                 *mnt = mntget(fs->rootmnt);
 366                 *dentry = dget(fs->root);
 367                 read_unlock(&fs->lock);
 368                 result = 0;
 369                 put_fs_struct(fs);
 370         }
 371         return result;
 372 }
 373
 374 #define MAY_PTRACE(task) \
 375         (task == current || \
 376         (task->parent == current && \
 377         (task->ptrace & PT_PTRACED) && \
 378          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
 379          security_ptrace(current,task) == 0))
 380
 381 static int proc_pid_environ(struct task_struct *task, char * buffer)
 382 {
 383         int res = 0;
 384         struct mm_struct *mm = get_task_mm(task);
 385         if (mm) {
 386                 unsigned int len = mm->env_end - mm->env_start;
 387                 if (len > PAGE_SIZE)
 388                         len = PAGE_SIZE;
 389                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
 390                 if (!ptrace_may_attach(task))
 391                         res = -ESRCH;
 392                 mmput(mm);
 393         }
 394         return res;
 395 }
 396
 397 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 398 {
 399         int res = 0;
 400         unsigned int len;
 401         struct mm_struct *mm = get_task_mm(task);
 402         if (!mm)
 403                 goto out;
 404         if (!mm->arg_end)
 405                 goto out_mm;    /* Shh! No looking before we're done */
 406
 407         len = mm->arg_end - mm->arg_start;
 408
 409         if (len > PAGE_SIZE)
 410                 len = PAGE_SIZE;
 411
 412         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 413
 414         // If the nul at the end of args has been overwritten, then
 415         // assume application is using setproctitle(3).
 416         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 417                 len = strnlen(buffer, res);
 418                 if (len < res) {
 419                     res = len;
 420                 } else {
 421                         len = mm->env_end - mm->env_start;
 422                         if (len > PAGE_SIZE - res)
 423                                 len = PAGE_SIZE - res;
 424                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 425                         res = strnlen(buffer, res);
 426                 }
 427         }
 428 out_mm:
 429         mmput(mm);
 430 out:
 431         return res;
 432 }
 433
 434 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 435 {
 436         int res = 0;
 437         struct mm_struct *mm = get_task_mm(task);
 438         if (mm) {
 439                 unsigned int nwords = 0;
 440                 do
 441                         nwords += 2;
 442                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 443                 res = nwords * sizeof(mm->saved_auxv[0]);
 444                 if (res > PAGE_SIZE)
 445                         res = PAGE_SIZE;
 446                 memcpy(buffer, mm->saved_auxv, res);
 447                 mmput(mm);
 448         }
 449         return res;
 450 }
 451
 452
 453 #ifdef CONFIG_KALLSYMS
 454 /*
 455  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 456  * Returns the resolved symbol.  If that fails, simply return the address.
 457  */
 458 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 459 {
 460         char *modname;
 461         const char *sym_name;
 462         unsigned long wchan, size, offset;
 463         char namebuf[KSYM_NAME_LEN+1];
 464
 465         wchan = get_wchan(task);
 466
 467         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
 468         if (sym_name)
 469                 return sprintf(buffer, "%s", sym_name);
 470         return sprintf(buffer, "%lu", wchan);
 471 }
 472 #endif /* CONFIG_KALLSYMS */
 473
 474 #ifdef CONFIG_SCHEDSTATS
 475 /*
 476  * Provides /proc/PID/schedstat
 477  */
 478 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 479 {
 480         return sprintf(buffer, "%lu %lu %lu\n",
 481                         task->sched_info.cpu_time,
 482                         task->sched_info.run_delay,
 483                         task->sched_info.pcnt);
 484 }
 485 #endif
 486
 487 /* The badness from the OOM killer */
 488 unsigned long badness(struct task_struct *p, unsigned long uptime);
 489 static int proc_oom_score(struct task_struct *task, char *buffer)
 490 {
 491         unsigned long points;
 492         struct timespec uptime;
 493
 494         do_posix_clock_monotonic_gettime(&uptime);
 495         points = badness(task, uptime.tv_sec);
 496         return sprintf(buffer, "%lu\n", points);
 497 }
 498
 499 /************************************************************************/
 500 /*                       Here the fs part begins                        */
 501 /************************************************************************/
 502
 503 /* permission checks */
 504
 505 /* If the process being read is separated by chroot from the reading process,
 506  * don't let the reader access the threads.
 507  */
 508 static int proc_check_chroot(struct dentry *de, struct vfsmount *mnt)
 509 {
 510         struct dentry *base;
 511         struct vfsmount *our_vfsmnt;
 512         int res = 0;
 513
 514         read_lock(&current->fs->lock);
 515         our_vfsmnt = mntget(current->fs->rootmnt);
 516         base = dget(current->fs->root);
 517         read_unlock(&current->fs->lock);
 518
 519         spin_lock(&vfsmount_lock);
 520
 521         while (mnt != our_vfsmnt) {
 522                 if (mnt == mnt->mnt_parent)
 523                         goto out;
 524                 de = mnt->mnt_mountpoint;
 525                 mnt = mnt->mnt_parent;
 526         }
 527
 528         if (!is_subdir(de, base))
 529                 goto out;
 530         spin_unlock(&vfsmount_lock);
 531
 532 exit:
 533         dput(base);
 534         mntput(our_vfsmnt);
 535         return res;
 536 out:
 537         spin_unlock(&vfsmount_lock);
 538         res = -EACCES;
 539         goto exit;
 540 }
 541
 542 extern struct seq_operations proc_pid_maps_op;
 543 static int maps_open(struct inode *inode, struct file *file)
 544 {
 545         struct task_struct *task = proc_task(inode);
 546         int ret = seq_open(file, &proc_pid_maps_op);
 547         if (!ret) {
 548                 struct seq_file *m = file->private_data;
 549                 m->private = task;
 550         }
 551         return ret;
 552 }
 553
 554 static struct file_operations proc_maps_operations = {
 555         .open           = maps_open,
 556         .read           = seq_read,
 557         .llseek         = seq_lseek,
 558         .release        = seq_release,
 559 };
 560
 561 #ifdef CONFIG_NUMA
 562 extern struct seq_operations proc_pid_numa_maps_op;
 563 static int numa_maps_open(struct inode *inode, struct file *file)
 564 {
 565         struct task_struct *task = proc_task(inode);
 566         int ret = seq_open(file, &proc_pid_numa_maps_op);
 567         if (!ret) {
 568                 struct seq_file *m = file->private_data;
 569                 m->private = task;
 570         }
 571         return ret;
 572 }
 573
 574 static struct file_operations proc_numa_maps_operations = {
 575         .open           = numa_maps_open,
 576         .read           = seq_read,
 577         .llseek         = seq_lseek,
 578         .release        = seq_release,
 579 };
 580 #endif
 581
 582 #ifdef CONFIG_MMU
 583 extern struct seq_operations proc_pid_smaps_op;
 584 static int smaps_open(struct inode *inode, struct file *file)
 585 {
 586         struct task_struct *task = proc_task(inode);
 587         int ret = seq_open(file, &proc_pid_smaps_op);
 588         if (!ret) {
 589                 struct seq_file *m = file->private_data;
 590                 m->private = task;
 591         }
 592         return ret;
 593 }
 594
 595 static struct file_operations proc_smaps_operations = {
 596         .open           = smaps_open,
 597         .read           = seq_read,
 598         .llseek         = seq_lseek,
 599         .release        = seq_release,
 600 };
 601 #endif
 602
 603 extern struct seq_operations mounts_op;
 604 struct proc_mounts {
 605         struct seq_file m;
 606         int event;
 607 };
 608
 609 static int mounts_open(struct inode *inode, struct file *file)
 610 {
 611         struct task_struct *task = proc_task(inode);
 612         struct namespace *namespace;
 613         struct proc_mounts *p;
 614         int ret = -EINVAL;
 615
 616         task_lock(task);
 617         namespace = task->namespace;
 618         if (namespace)
 619                 get_namespace(namespace);
 620         task_unlock(task);
 621
 622         if (namespace) {
 623                 ret = -ENOMEM;
 624                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 625                 if (p) {
 626                         file->private_data = &p->m;
 627                         ret = seq_open(file, &mounts_op);
 628                         if (!ret) {
 629                                 p->m.private = namespace;
 630                                 p->event = namespace->event;
 631                                 return 0;
 632                         }
 633                         kfree(p);
 634                 }
 635                 put_namespace(namespace);
 636         }
 637         return ret;
 638 }
 639
 640 static int mounts_release(struct inode *inode, struct file *file)
 641 {
 642         struct seq_file *m = file->private_data;
 643         struct namespace *namespace = m->private;
 644         put_namespace(namespace);
 645         return seq_release(inode, file);
 646 }
 647
 648 static unsigned mounts_poll(struct file *file, poll_table *wait)
 649 {
 650         struct proc_mounts *p = file->private_data;
 651         struct namespace *ns = p->m.private;
 652         unsigned res = 0;
 653
 654         poll_wait(file, &ns->poll, wait);
 655
 656         spin_lock(&vfsmount_lock);
 657         if (p->event != ns->event) {
 658                 p->event = ns->event;
 659                 res = POLLERR;
 660         }
 661         spin_unlock(&vfsmount_lock);
 662
 663         return res;
 664 }
 665
 666 static struct file_operations proc_mounts_operations = {
 667         .open           = mounts_open,
 668         .read           = seq_read,
 669         .llseek         = seq_lseek,
 670         .release        = mounts_release,
 671         .poll           = mounts_poll,
 672 };
 673
 674 extern struct seq_operations mountstats_op;
 675 static int mountstats_open(struct inode *inode, struct file *file)
 676 {
 677         struct task_struct *task = proc_task(inode);
 678         int ret = seq_open(file, &mountstats_op);
 679
 680         if (!ret) {
 681                 struct seq_file *m = file->private_data;
 682                 struct namespace *namespace;
 683                 task_lock(task);
 684                 namespace = task->namespace;
 685                 if (namespace)
 686                         get_namespace(namespace);
 687                 task_unlock(task);
 688
 689                 if (namespace)
 690                         m->private = namespace;
 691                 else {
 692                         seq_release(inode, file);
 693                         ret = -EINVAL;
 694                 }
 695         }
 696         return ret;
 697 }
 698
 699 static struct file_operations proc_mountstats_operations = {
 700         .open           = mountstats_open,
 701         .read           = seq_read,
 702         .llseek         = seq_lseek,
 703         .release        = mounts_release,
 704 };
 705
 706 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 707
 708 static ssize_t proc_info_read(struct file * file, char __user * buf,
 709                           size_t count, loff_t *ppos)
 710 {
 711         struct inode * inode = file->f_dentry->d_inode;
 712         unsigned long page;
 713         ssize_t length;
 714         struct task_struct *task = proc_task(inode);
 715
 716         if (count > PROC_BLOCK_SIZE)
 717                 count = PROC_BLOCK_SIZE;
 718         if (!(page = __get_free_page(GFP_KERNEL)))
 719                 return -ENOMEM;
 720
 721         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 722
 723         if (length >= 0)
 724                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 725         free_page(page);
 726         return length;
 727 }
 728
 729 static struct file_operations proc_info_file_operations = {
 730         .read           = proc_info_read,
 731 };
 732
 733 static int mem_open(struct inode* inode, struct file* file)
 734 {
 735         file->private_data = (void*)((long)current->self_exec_id);
 736         return 0;
 737 }
 738
 739 static ssize_t mem_read(struct file * file, char __user * buf,
 740                         size_t count, loff_t *ppos)
 741 {
 742         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 743         char *page;
 744         unsigned long src = *ppos;
 745         int ret = -ESRCH;
 746         struct mm_struct *mm;
 747
 748         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 749                 goto out;
 750
 751         ret = -ENOMEM;
 752         page = (char *)__get_free_page(GFP_USER);
 753         if (!page)
 754                 goto out;
 755
 756         ret = 0;
 757
 758         mm = get_task_mm(task);
 759         if (!mm)
 760                 goto out_free;
 761
 762         ret = -EIO;
 763
 764         if (file->private_data != (void*)((long)current->self_exec_id))
 765                 goto out_put;
 766
 767         ret = 0;
 768
 769         while (count > 0) {
 770                 int this_len, retval;
 771
 772                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 773                 retval = access_process_vm(task, src, page, this_len, 0);
 774                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 775                         if (!ret)
 776                                 ret = -EIO;
 777                         break;
 778                 }
 779
 780                 if (copy_to_user(buf, page, retval)) {
 781                         ret = -EFAULT;
 782                         break;
 783                 }
 784
 785                 ret += retval;
 786                 src += retval;
 787                 buf += retval;
 788                 count -= retval;
 789         }
 790         *ppos = src;
 791
 792 out_put:
 793         mmput(mm);
 794 out_free:
 795         free_page((unsigned long) page);
 796 out:
 797         return ret;
 798 }
 799
 800 #define mem_write NULL
 801
 802 #ifndef mem_write
 803 /* This is a security hazard */
 804 static ssize_t mem_write(struct file * file, const char * buf,
 805                          size_t count, loff_t *ppos)
 806 {
 807         int copied = 0;
 808         char *page;
 809         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 810         unsigned long dst = *ppos;
 811
 812         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 813                 return -ESRCH;
 814
 815         page = (char *)__get_free_page(GFP_USER);
 816         if (!page)
 817                 return -ENOMEM;
 818
 819         while (count > 0) {
 820                 int this_len, retval;
 821
 822                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 823                 if (copy_from_user(page, buf, this_len)) {
 824                         copied = -EFAULT;
 825                         break;
 826                 }
 827                 retval = access_process_vm(task, dst, page, this_len, 1);
 828                 if (!retval) {
 829                         if (!copied)
 830                                 copied = -EIO;
 831                         break;
 832                 }
 833                 copied += retval;
 834                 buf += retval;
 835                 dst += retval;
 836                 count -= retval;
 837         }
 838         *ppos = dst;
 839         free_page((unsigned long) page);
 840         return copied;
 841 }
 842 #endif
 843
 844 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 845 {
 846         switch (orig) {
 847         case 0:
 848                 file->f_pos = offset;
 849                 break;
 850         case 1:
 851                 file->f_pos += offset;
 852                 break;
 853         default:
 854                 return -EINVAL;
 855         }
 856         force_successful_syscall_return();
 857         return file->f_pos;
 858 }
 859
 860 static struct file_operations proc_mem_operations = {
 861         .llseek         = mem_lseek,
 862         .read           = mem_read,
 863         .write          = mem_write,
 864         .open           = mem_open,
 865 };
 866
 867 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 868                                 size_t count, loff_t *ppos)
 869 {
 870         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 871         char buffer[8];
 872         size_t len;
 873         int oom_adjust = task->oomkilladj;
 874         loff_t __ppos = *ppos;
 875
 876         len = sprintf(buffer, "%i\n", oom_adjust);
 877         if (__ppos >= len)
 878                 return 0;
 879         if (count > len-__ppos)
 880                 count = len-__ppos;
 881         if (copy_to_user(buf, buffer + __ppos, count))
 882                 return -EFAULT;
 883         *ppos = __ppos + count;
 884         return count;
 885 }
 886
 887 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 888                                 size_t count, loff_t *ppos)
 889 {
 890         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 891         char buffer[8], *end;
 892         int oom_adjust;
 893
 894         if (!capable(CAP_SYS_RESOURCE))
 895                 return -EPERM;
 896         memset(buffer, 0, 8);
 897         if (count > 6)
 898                 count = 6;
 899         if (copy_from_user(buffer, buf, count))
 900                 return -EFAULT;
 901         oom_adjust = simple_strtol(buffer, &end, 0);
 902         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 903                 return -EINVAL;
 904         if (*end == '\n')
 905                 end++;
 906         task->oomkilladj = oom_adjust;
 907         if (end - buffer == 0)
 908                 return -EIO;
 909         return end - buffer;
 910 }
 911
 912 static struct file_operations proc_oom_adjust_operations = {
 913         .read           = oom_adjust_read,
 914         .write          = oom_adjust_write,
 915 };
 916
 917 #ifdef CONFIG_AUDITSYSCALL
 918 #define TMPBUFLEN 21
 919 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 920                                   size_t count, loff_t *ppos)
 921 {
 922         struct inode * inode = file->f_dentry->d_inode;
 923         struct task_struct *task = proc_task(inode);
 924         ssize_t length;
 925         char tmpbuf[TMPBUFLEN];
 926
 927         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 928                                 audit_get_loginuid(task->audit_context));
 929         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 930 }
 931
 932 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 933                                    size_t count, loff_t *ppos)
 934 {
 935         struct inode * inode = file->f_dentry->d_inode;
 936         char *page, *tmp;
 937         ssize_t length;
 938         struct task_struct *task = proc_task(inode);
 939         uid_t loginuid;
 940
 941         if (!capable(CAP_AUDIT_CONTROL))
 942                 return -EPERM;
 943
 944         if (current != task)
 945                 return -EPERM;
 946
 947         if (count >= PAGE_SIZE)
 948                 count = PAGE_SIZE - 1;
 949
 950         if (*ppos != 0) {
 951                 /* No partial writes. */
 952                 return -EINVAL;
 953         }
 954         page = (char*)__get_free_page(GFP_USER);
 955         if (!page)
 956                 return -ENOMEM;
 957         length = -EFAULT;
 958         if (copy_from_user(page, buf, count))
 959                 goto out_free_page;
 960
 961         page[count] = '\0';
 962         loginuid = simple_strtoul(page, &tmp, 10);
 963         if (tmp == page) {
 964                 length = -EINVAL;
 965                 goto out_free_page;
 966
 967         }
 968         length = audit_set_loginuid(task, loginuid);
 969         if (likely(length == 0))
 970                 length = count;
 971
 972 out_free_page:
 973         free_page((unsigned long) page);
 974         return length;
 975 }
 976
 977 static struct file_operations proc_loginuid_operations = {
 978         .read           = proc_loginuid_read,
 979         .write          = proc_loginuid_write,
 980 };
 981 #endif
 982
 983 #ifdef CONFIG_SECCOMP
 984 static ssize_t seccomp_read(struct file *file, char __user *buf,
 985                             size_t count, loff_t *ppos)
 986 {
 987         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
 988         char __buf[20];
 989         loff_t __ppos = *ppos;
 990         size_t len;
 991
 992         /* no need to print the trailing zero, so use only len */
 993         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
 994         if (__ppos >= len)
 995                 return 0;
 996         if (count > len - __ppos)
 997                 count = len - __ppos;
 998         if (copy_to_user(buf, __buf + __ppos, count))
 999                 return -EFAULT;
1000         *ppos = __ppos + count;
1001         return count;
1002 }
1003
1004 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1005                              size_t count, loff_t *ppos)
1006 {
1007         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1008         char __buf[20], *end;
1009         unsigned int seccomp_mode;
1010
1011         /* can set it only once to be even more secure */
1012         if (unlikely(tsk->seccomp.mode))
1013                 return -EPERM;
1014
1015         memset(__buf, 0, sizeof(__buf));
1016         count = min(count, sizeof(__buf) - 1);
1017         if (copy_from_user(__buf, buf, count))
1018                 return -EFAULT;
1019         seccomp_mode = simple_strtoul(__buf, &end, 0);
1020         if (*end == '\n')
1021                 end++;
1022         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1023                 tsk->seccomp.mode = seccomp_mode;
1024                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1025         } else
1026                 return -EINVAL;
1027         if (unlikely(!(end - __buf)))
1028                 return -EIO;
1029         return end - __buf;
1030 }
1031
1032 static struct file_operations proc_seccomp_operations = {
1033         .read           = seccomp_read,
1034         .write          = seccomp_write,
1035 };
1036 #endif /* CONFIG_SECCOMP */
1037
1038 static int proc_check_dentry_visible(struct inode *inode,
1039         struct dentry *dentry, struct vfsmount *mnt)
1040 {
1041         /* Verify that the current process can already see the
1042          * file pointed at by the file descriptor.
1043          * This prevents /proc from being an accidental information leak.
1044          *
1045          * This prevents access to files that are not visible do to
1046          * being on the otherside of a chroot, in a different
1047          * namespace, or are simply process local (like pipes).
1048          */
1049         struct task_struct *task;
1050         struct files_struct *task_files, *files;
1051         int error = -EACCES;
1052
1053         /* See if the the two tasks share a commone set of
1054          * file descriptors.  If so everything is visible.
1055          */
1056         task = proc_task(inode);
1057         if (!task)
1058                 goto out;
1059         files = get_files_struct(current);
1060         task_files = get_files_struct(task);
1061         if (files && task_files && (files == task_files))
1062                 error = 0;
1063         if (task_files)
1064                 put_files_struct(task_files);
1065         if (files)
1066                 put_files_struct(files);
1067         if (!error)
1068                 goto out;
1069
1070         /* If the two tasks don't share a common set of file
1071          * descriptors see if the destination dentry is already
1072          * visible in the current tasks filesystem namespace.
1073          */
1074         error = proc_check_chroot(dentry, mnt);
1075 out:
1076         return error;
1077
1078 }
1079
1080 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1081 {
1082         struct inode *inode = dentry->d_inode;
1083         int error = -EACCES;
1084
1085         /* We don't need a base pointer in the /proc filesystem */
1086         path_release(nd);
1087
1088         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1089                 goto out;
1090
1091         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1092         nd->last_type = LAST_BIND;
1093         if (error)
1094                 goto out;
1095
1096         /* Only return files this task can already see */
1097         error = proc_check_dentry_visible(inode, nd->dentry, nd->mnt);
1098         if (error)
1099                 path_release(nd);
1100 out:
1101         return ERR_PTR(error);
1102 }
1103
1104 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1105                             char __user *buffer, int buflen)
1106 {
1107         struct inode * inode;
1108         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1109         int len;
1110
1111         if (!tmp)
1112                 return -ENOMEM;
1113
1114         inode = dentry->d_inode;
1115         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1116         len = PTR_ERR(path);
1117         if (IS_ERR(path))
1118                 goto out;
1119         len = tmp + PAGE_SIZE - 1 - path;
1120
1121         if (len > buflen)
1122                 len = buflen;
1123         if (copy_to_user(buffer, path, len))
1124                 len = -EFAULT;
1125  out:
1126         free_page((unsigned long)tmp);
1127         return len;
1128 }
1129
1130 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1131 {
1132         int error = -EACCES;
1133         struct inode *inode = dentry->d_inode;
1134         struct dentry *de;
1135         struct vfsmount *mnt = NULL;
1136
1137
1138         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1139                 goto out;
1140
1141         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1142         if (error)
1143                 goto out;
1144
1145         /* Only return files this task can already see */
1146         error = proc_check_dentry_visible(inode, de, mnt);
1147         if (error)
1148                 goto out_put;
1149
1150         error = do_proc_readlink(de, mnt, buffer, buflen);
1151 out_put:
1152         dput(de);
1153         mntput(mnt);
1154 out:
1155         return error;
1156 }
1157
1158 static struct inode_operations proc_pid_link_inode_operations = {
1159         .readlink       = proc_pid_readlink,
1160         .follow_link    = proc_pid_follow_link
1161 };
1162
1163 #define NUMBUF 10
1164
1165 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1166 {
1167         struct dentry *dentry = filp->f_dentry;
1168         struct inode *inode = dentry->d_inode;
1169         struct task_struct *p = proc_task(inode);
1170         unsigned int fd, tid, ino;
1171         int retval;
1172         char buf[NUMBUF];
1173         struct files_struct * files;
1174         struct fdtable *fdt;
1175
1176         retval = -ENOENT;
1177         if (!pid_alive(p))
1178                 goto out;
1179         retval = 0;
1180         tid = p->pid;
1181
1182         fd = filp->f_pos;
1183         switch (fd) {
1184                 case 0:
1185                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1186                                 goto out;
1187                         filp->f_pos++;
1188                 case 1:
1189                         ino = parent_ino(dentry);
1190                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1191                                 goto out;
1192                         filp->f_pos++;
1193                 default:
1194                         files = get_files_struct(p);
1195                         if (!files)
1196                                 goto out;
1197                         rcu_read_lock();
1198                         fdt = files_fdtable(files);
1199                         for (fd = filp->f_pos-2;
1200                              fd < fdt->max_fds;
1201                              fd++, filp->f_pos++) {
1202                                 unsigned int i,j;
1203
1204                                 if (!fcheck_files(files, fd))
1205                                         continue;
1206                                 rcu_read_unlock();
1207
1208                                 j = NUMBUF;
1209                                 i = fd;
1210                                 do {
1211                                         j--;
1212                                         buf[j] = '0' + (i % 10);
1213                                         i /= 10;
1214                                 } while (i);
1215
1216                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1217                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1218                                         rcu_read_lock();
1219                                         break;
1220                                 }
1221                                 rcu_read_lock();
1222                         }
1223                         rcu_read_unlock();
1224                         put_files_struct(files);
1225         }
1226 out:
1227         return retval;
1228 }
1229
1230 static int proc_pident_readdir(struct file *filp,
1231                 void *dirent, filldir_t filldir,
1232                 struct pid_entry *ents, unsigned int nents)
1233 {
1234         int i;
1235         int pid;
1236         struct dentry *dentry = filp->f_dentry;
1237         struct inode *inode = dentry->d_inode;
1238         struct pid_entry *p;
1239         ino_t ino;
1240         int ret;
1241
1242         ret = -ENOENT;
1243         if (!pid_alive(proc_task(inode)))
1244                 goto out;
1245
1246         ret = 0;
1247         pid = proc_task(inode)->pid;
1248         i = filp->f_pos;
1249         switch (i) {
1250         case 0:
1251                 ino = inode->i_ino;
1252                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1253                         goto out;
1254                 i++;
1255                 filp->f_pos++;
1256                 /* fall through */
1257         case 1:
1258                 ino = parent_ino(dentry);
1259                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1260                         goto out;
1261                 i++;
1262                 filp->f_pos++;
1263                 /* fall through */
1264         default:
1265                 i -= 2;
1266                 if (i >= nents) {
1267                         ret = 1;
1268                         goto out;
1269                 }
1270                 p = ents + i;
1271                 while (p->name) {
1272                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1273                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1274                                 goto out;
1275                         filp->f_pos++;
1276                         p++;
1277                 }
1278         }
1279
1280         ret = 1;
1281 out:
1282         return ret;
1283 }
1284
1285 static int proc_tgid_base_readdir(struct file * filp,
1286                              void * dirent, filldir_t filldir)
1287 {
1288         return proc_pident_readdir(filp,dirent,filldir,
1289                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1290 }
1291
1292 static int proc_tid_base_readdir(struct file * filp,
1293                              void * dirent, filldir_t filldir)
1294 {
1295         return proc_pident_readdir(filp,dirent,filldir,
1296                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1297 }
1298
1299 /* building an inode */
1300
1301 static int task_dumpable(struct task_struct *task)
1302 {
1303         int dumpable = 0;
1304         struct mm_struct *mm;
1305
1306         task_lock(task);
1307         mm = task->mm;
1308         if (mm)
1309                 dumpable = mm->dumpable;
1310         task_unlock(task);
1311         if(dumpable == 1)
1312                 return 1;
1313         return 0;
1314 }
1315
1316
1317 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1318 {
1319         struct inode * inode;
1320         struct proc_inode *ei;
1321
1322         /* We need a new inode */
1323
1324         inode = new_inode(sb);
1325         if (!inode)
1326                 goto out;
1327
1328         /* Common stuff */
1329         ei = PROC_I(inode);
1330         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1331         inode->i_ino = fake_ino(task->pid, ino);
1332
1333         if (!pid_alive(task))
1334                 goto out_unlock;
1335
1336         /*
1337          * grab the reference to task.
1338          */
1339         get_task_struct(task);
1340         ei->task = task;
1341         inode->i_uid = 0;
1342         inode->i_gid = 0;
1343         if (task_dumpable(task)) {
1344                 inode->i_uid = task->euid;
1345                 inode->i_gid = task->egid;
1346         }
1347         security_task_to_inode(task, inode);
1348
1349 out:
1350         return inode;
1351
1352 out_unlock:
1353         iput(inode);
1354         return NULL;
1355 }
1356
1357 /* dentry stuff */
1358
1359 /*
1360  *      Exceptional case: normally we are not allowed to unhash a busy
1361  * directory. In this case, however, we can do it - no aliasing problems
1362  * due to the way we treat inodes.
1363  *
1364  * Rewrite the inode's ownerships here because the owning task may have
1365  * performed a setuid(), etc.
1366  */
1367 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1368 {
1369         struct inode *inode = dentry->d_inode;
1370         struct task_struct *task = proc_task(inode);
1371         if (pid_alive(task)) {
1372                 if (task_dumpable(task)) {
1373                         inode->i_uid = task->euid;
1374                         inode->i_gid = task->egid;
1375                 } else {
1376                         inode->i_uid = 0;
1377                         inode->i_gid = 0;
1378                 }
1379                 security_task_to_inode(task, inode);
1380                 return 1;
1381         }
1382         d_drop(dentry);
1383         return 0;
1384 }
1385
1386 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1387 {
1388         struct inode *inode = dentry->d_inode;
1389         struct task_struct *task = proc_task(inode);
1390         int fd = proc_fd(inode);
1391         struct files_struct *files;
1392
1393         files = get_files_struct(task);
1394         if (files) {
1395                 rcu_read_lock();
1396                 if (fcheck_files(files, fd)) {
1397                         rcu_read_unlock();
1398                         put_files_struct(files);
1399                         if (task_dumpable(task)) {
1400                                 inode->i_uid = task->euid;
1401                                 inode->i_gid = task->egid;
1402                         } else {
1403                                 inode->i_uid = 0;
1404                                 inode->i_gid = 0;
1405                         }
1406                         security_task_to_inode(task, inode);
1407                         return 1;
1408                 }
1409                 rcu_read_unlock();
1410                 put_files_struct(files);
1411         }
1412         d_drop(dentry);
1413         return 0;
1414 }
1415
1416 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1417 {
1418         struct task_struct *task = proc_task(inode);
1419         spin_lock(&task->proc_lock);
1420         if (task->proc_dentry == dentry)
1421                 task->proc_dentry = NULL;
1422         spin_unlock(&task->proc_lock);
1423         iput(inode);
1424 }
1425
1426 static int pid_delete_dentry(struct dentry * dentry)
1427 {
1428         /* Is the task we represent dead?
1429          * If so, then don't put the dentry on the lru list,
1430          * kill it immediately.
1431          */
1432         return !pid_alive(proc_task(dentry->d_inode));
1433 }
1434
1435 static struct dentry_operations tid_fd_dentry_operations =
1436 {
1437         .d_revalidate   = tid_fd_revalidate,
1438         .d_delete       = pid_delete_dentry,
1439 };
1440
1441 static struct dentry_operations pid_dentry_operations =
1442 {
1443         .d_revalidate   = pid_revalidate,
1444         .d_delete       = pid_delete_dentry,
1445 };
1446
1447 static struct dentry_operations pid_base_dentry_operations =
1448 {
1449         .d_revalidate   = pid_revalidate,
1450         .d_iput         = pid_base_iput,
1451         .d_delete       = pid_delete_dentry,
1452 };
1453
1454 /* Lookups */
1455
1456 static unsigned name_to_int(struct dentry *dentry)
1457 {
1458         const char *name = dentry->d_name.name;
1459         int len = dentry->d_name.len;
1460         unsigned n = 0;
1461
1462         if (len > 1 && *name == '0')
1463                 goto out;
1464         while (len-- > 0) {
1465                 unsigned c = *name++ - '0';
1466                 if (c > 9)
1467                         goto out;
1468                 if (n >= (~0U-9)/10)
1469                         goto out;
1470                 n *= 10;
1471                 n += c;
1472         }
1473         return n;
1474 out:
1475         return ~0U;
1476 }
1477
1478 /* SMP-safe */
1479 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1480 {
1481         struct task_struct *task = proc_task(dir);
1482         unsigned fd = name_to_int(dentry);
1483         struct file * file;
1484         struct files_struct * files;
1485         struct inode *inode;
1486         struct proc_inode *ei;
1487
1488         if (fd == ~0U)
1489                 goto out;
1490         if (!pid_alive(task))
1491                 goto out;
1492
1493         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1494         if (!inode)
1495                 goto out;
1496         ei = PROC_I(inode);
1497         ei->fd = fd;
1498         files = get_files_struct(task);
1499         if (!files)
1500                 goto out_unlock;
1501         inode->i_mode = S_IFLNK;
1502
1503         /*
1504          * We are not taking a ref to the file structure, so we must
1505          * hold ->file_lock.
1506          */
1507         spin_lock(&files->file_lock);
1508         file = fcheck_files(files, fd);
1509         if (!file)
1510                 goto out_unlock2;
1511         if (file->f_mode & 1)
1512                 inode->i_mode |= S_IRUSR | S_IXUSR;
1513         if (file->f_mode & 2)
1514                 inode->i_mode |= S_IWUSR | S_IXUSR;
1515         spin_unlock(&files->file_lock);
1516         put_files_struct(files);
1517         inode->i_op = &proc_pid_link_inode_operations;
1518         inode->i_size = 64;
1519         ei->op.proc_get_link = proc_fd_link;
1520         dentry->d_op = &tid_fd_dentry_operations;
1521         d_add(dentry, inode);
1522         return NULL;
1523
1524 out_unlock2:
1525         spin_unlock(&files->file_lock);
1526         put_files_struct(files);
1527 out_unlock:
1528         iput(inode);
1529 out:
1530         return ERR_PTR(-ENOENT);
1531 }
1532
1533 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1534 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1535
1536 static struct file_operations proc_fd_operations = {
1537         .read           = generic_read_dir,
1538         .readdir        = proc_readfd,
1539 };
1540
1541 static struct file_operations proc_task_operations = {
1542         .read           = generic_read_dir,
1543         .readdir        = proc_task_readdir,
1544 };
1545
1546 /*
1547  * proc directories can do almost nothing..
1548  */
1549 static struct inode_operations proc_fd_inode_operations = {
1550         .lookup         = proc_lookupfd,
1551 };
1552
1553 static struct inode_operations proc_task_inode_operations = {
1554         .lookup         = proc_task_lookup,
1555 };
1556
1557 #ifdef CONFIG_SECURITY
1558 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1559                                   size_t count, loff_t *ppos)
1560 {
1561         struct inode * inode = file->f_dentry->d_inode;
1562         unsigned long page;
1563         ssize_t length;
1564         struct task_struct *task = proc_task(inode);
1565
1566         if (count > PAGE_SIZE)
1567                 count = PAGE_SIZE;
1568         if (!(page = __get_free_page(GFP_KERNEL)))
1569                 return -ENOMEM;
1570
1571         length = security_getprocattr(task,
1572                                       (char*)file->f_dentry->d_name.name,
1573                                       (void*)page, count);
1574         if (length >= 0)
1575                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1576         free_page(page);
1577         return length;
1578 }
1579
1580 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1581                                    size_t count, loff_t *ppos)
1582 {
1583         struct inode * inode = file->f_dentry->d_inode;
1584         char *page;
1585         ssize_t length;
1586         struct task_struct *task = proc_task(inode);
1587
1588         if (count > PAGE_SIZE)
1589                 count = PAGE_SIZE;
1590         if (*ppos != 0) {
1591                 /* No partial writes. */
1592                 return -EINVAL;
1593         }
1594         page = (char*)__get_free_page(GFP_USER);
1595         if (!page)
1596                 return -ENOMEM;
1597         length = -EFAULT;
1598         if (copy_from_user(page, buf, count))
1599                 goto out;
1600
1601         length = security_setprocattr(task,
1602                                       (char*)file->f_dentry->d_name.name,
1603                                       (void*)page, count);
1604 out:
1605         free_page((unsigned long) page);
1606         return length;
1607 }
1608
1609 static struct file_operations proc_pid_attr_operations = {
1610         .read           = proc_pid_attr_read,
1611         .write          = proc_pid_attr_write,
1612 };
1613
1614 static struct file_operations proc_tid_attr_operations;
1615 static struct inode_operations proc_tid_attr_inode_operations;
1616 static struct file_operations proc_tgid_attr_operations;
1617 static struct inode_operations proc_tgid_attr_inode_operations;
1618 #endif
1619
1620 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1621
1622 /* SMP-safe */
1623 static struct dentry *proc_pident_lookup(struct inode *dir,
1624                                          struct dentry *dentry,
1625                                          struct pid_entry *ents)
1626 {
1627         struct inode *inode;
1628         int error;
1629         struct task_struct *task = proc_task(dir);
1630         struct pid_entry *p;
1631         struct proc_inode *ei;
1632
1633         error = -ENOENT;
1634         inode = NULL;
1635
1636         if (!pid_alive(task))
1637                 goto out;
1638
1639         for (p = ents; p->name; p++) {
1640                 if (p->len != dentry->d_name.len)
1641                         continue;
1642                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1643                         break;
1644         }
1645         if (!p->name)
1646                 goto out;
1647
1648         error = -EINVAL;
1649         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1650         if (!inode)
1651                 goto out;
1652
1653         ei = PROC_I(inode);
1654         inode->i_mode = p->mode;
1655         /*
1656          * Yes, it does not scale. And it should not. Don't add
1657          * new entries into /proc/<tgid>/ without very good reasons.
1658          */
1659         switch(p->type) {
1660                 case PROC_TGID_TASK:
1661                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1662                         inode->i_op = &proc_task_inode_operations;
1663                         inode->i_fop = &proc_task_operations;
1664                         break;
1665                 case PROC_TID_FD:
1666                 case PROC_TGID_FD:
1667                         inode->i_nlink = 2;
1668                         inode->i_op = &proc_fd_inode_operations;
1669                         inode->i_fop = &proc_fd_operations;
1670                         break;
1671                 case PROC_TID_EXE:
1672                 case PROC_TGID_EXE:
1673                         inode->i_op = &proc_pid_link_inode_operations;
1674                         ei->op.proc_get_link = proc_exe_link;
1675                         break;
1676                 case PROC_TID_CWD:
1677                 case PROC_TGID_CWD:
1678                         inode->i_op = &proc_pid_link_inode_operations;
1679                         ei->op.proc_get_link = proc_cwd_link;
1680                         break;
1681                 case PROC_TID_ROOT:
1682                 case PROC_TGID_ROOT:
1683                         inode->i_op = &proc_pid_link_inode_operations;
1684                         ei->op.proc_get_link = proc_root_link;
1685                         break;
1686                 case PROC_TID_ENVIRON:
1687                 case PROC_TGID_ENVIRON:
1688                         inode->i_fop = &proc_info_file_operations;
1689                         ei->op.proc_read = proc_pid_environ;
1690                         break;
1691                 case PROC_TID_AUXV:
1692                 case PROC_TGID_AUXV:
1693                         inode->i_fop = &proc_info_file_operations;
1694                         ei->op.proc_read = proc_pid_auxv;
1695                         break;
1696                 case PROC_TID_STATUS:
1697                 case PROC_TGID_STATUS:
1698                         inode->i_fop = &proc_info_file_operations;
1699                         ei->op.proc_read = proc_pid_status;
1700                         break;
1701                 case PROC_TID_STAT:
1702                         inode->i_fop = &proc_info_file_operations;
1703                         ei->op.proc_read = proc_tid_stat;
1704                         break;
1705                 case PROC_TGID_STAT:
1706                         inode->i_fop = &proc_info_file_operations;
1707                         ei->op.proc_read = proc_tgid_stat;
1708                         break;
1709                 case PROC_TID_CMDLINE:
1710                 case PROC_TGID_CMDLINE:
1711                         inode->i_fop = &proc_info_file_operations;
1712                         ei->op.proc_read = proc_pid_cmdline;
1713                         break;
1714                 case PROC_TID_STATM:
1715                 case PROC_TGID_STATM:
1716                         inode->i_fop = &proc_info_file_operations;
1717                         ei->op.proc_read = proc_pid_statm;
1718                         break;
1719                 case PROC_TID_MAPS:
1720                 case PROC_TGID_MAPS:
1721                         inode->i_fop = &proc_maps_operations;
1722                         break;
1723 #ifdef CONFIG_NUMA
1724                 case PROC_TID_NUMA_MAPS:
1725                 case PROC_TGID_NUMA_MAPS:
1726                         inode->i_fop = &proc_numa_maps_operations;
1727                         break;
1728 #endif
1729                 case PROC_TID_MEM:
1730                 case PROC_TGID_MEM:
1731                         inode->i_fop = &proc_mem_operations;
1732                         break;
1733 #ifdef CONFIG_SECCOMP
1734                 case PROC_TID_SECCOMP:
1735                 case PROC_TGID_SECCOMP:
1736                         inode->i_fop = &proc_seccomp_operations;
1737                         break;
1738 #endif /* CONFIG_SECCOMP */
1739                 case PROC_TID_MOUNTS:
1740                 case PROC_TGID_MOUNTS:
1741                         inode->i_fop = &proc_mounts_operations;
1742                         break;
1743 #ifdef CONFIG_MMU
1744                 case PROC_TID_SMAPS:
1745                 case PROC_TGID_SMAPS:
1746                         inode->i_fop = &proc_smaps_operations;
1747                         break;
1748 #endif
1749                 case PROC_TID_MOUNTSTATS:
1750                 case PROC_TGID_MOUNTSTATS:
1751                         inode->i_fop = &proc_mountstats_operations;
1752                         break;
1753 #ifdef CONFIG_SECURITY
1754                 case PROC_TID_ATTR:
1755                         inode->i_nlink = 2;
1756                         inode->i_op = &proc_tid_attr_inode_operations;
1757                         inode->i_fop = &proc_tid_attr_operations;
1758                         break;
1759                 case PROC_TGID_ATTR:
1760                         inode->i_nlink = 2;
1761                         inode->i_op = &proc_tgid_attr_inode_operations;
1762                         inode->i_fop = &proc_tgid_attr_operations;
1763                         break;
1764                 case PROC_TID_ATTR_CURRENT:
1765                 case PROC_TGID_ATTR_CURRENT:
1766                 case PROC_TID_ATTR_PREV:
1767                 case PROC_TGID_ATTR_PREV:
1768                 case PROC_TID_ATTR_EXEC:
1769                 case PROC_TGID_ATTR_EXEC:
1770                 case PROC_TID_ATTR_FSCREATE:
1771                 case PROC_TGID_ATTR_FSCREATE:
1772                 case PROC_TID_ATTR_KEYCREATE:
1773                 case PROC_TGID_ATTR_KEYCREATE:
1774                         inode->i_fop = &proc_pid_attr_operations;
1775                         break;
1776 #endif
1777 #ifdef CONFIG_KALLSYMS
1778                 case PROC_TID_WCHAN:
1779                 case PROC_TGID_WCHAN:
1780                         inode->i_fop = &proc_info_file_operations;
1781                         ei->op.proc_read = proc_pid_wchan;
1782                         break;
1783 #endif
1784 #ifdef CONFIG_SCHEDSTATS
1785                 case PROC_TID_SCHEDSTAT:
1786                 case PROC_TGID_SCHEDSTAT:
1787                         inode->i_fop = &proc_info_file_operations;
1788                         ei->op.proc_read = proc_pid_schedstat;
1789                         break;
1790 #endif
1791 #ifdef CONFIG_CPUSETS
1792                 case PROC_TID_CPUSET:
1793                 case PROC_TGID_CPUSET:
1794                         inode->i_fop = &proc_cpuset_operations;
1795                         break;
1796 #endif
1797                 case PROC_TID_OOM_SCORE:
1798                 case PROC_TGID_OOM_SCORE:
1799                         inode->i_fop = &proc_info_file_operations;
1800                         ei->op.proc_read = proc_oom_score;
1801                         break;
1802                 case PROC_TID_OOM_ADJUST:
1803                 case PROC_TGID_OOM_ADJUST:
1804                         inode->i_fop = &proc_oom_adjust_operations;
1805                         break;
1806 #ifdef CONFIG_AUDITSYSCALL
1807                 case PROC_TID_LOGINUID:
1808                 case PROC_TGID_LOGINUID:
1809                         inode->i_fop = &proc_loginuid_operations;
1810                         break;
1811 #endif
1812                 default:
1813                         printk("procfs: impossible type (%d)",p->type);
1814                         iput(inode);
1815                         return ERR_PTR(-EINVAL);
1816         }
1817         dentry->d_op = &pid_dentry_operations;
1818         d_add(dentry, inode);
1819         return NULL;
1820
1821 out:
1822         return ERR_PTR(error);
1823 }
1824
1825 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1826         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1827 }
1828
1829 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1830         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1831 }
1832
1833 static struct file_operations proc_tgid_base_operations = {
1834         .read           = generic_read_dir,
1835         .readdir        = proc_tgid_base_readdir,
1836 };
1837
1838 static struct file_operations proc_tid_base_operations = {
1839         .read           = generic_read_dir,
1840         .readdir        = proc_tid_base_readdir,
1841 };
1842
1843 static struct inode_operations proc_tgid_base_inode_operations = {
1844         .lookup         = proc_tgid_base_lookup,
1845 };
1846
1847 static struct inode_operations proc_tid_base_inode_operations = {
1848         .lookup         = proc_tid_base_lookup,
1849 };
1850
1851 #ifdef CONFIG_SECURITY
1852 static int proc_tgid_attr_readdir(struct file * filp,
1853                              void * dirent, filldir_t filldir)
1854 {
1855         return proc_pident_readdir(filp,dirent,filldir,
1856                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1857 }
1858
1859 static int proc_tid_attr_readdir(struct file * filp,
1860                              void * dirent, filldir_t filldir)
1861 {
1862         return proc_pident_readdir(filp,dirent,filldir,
1863                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1864 }
1865
1866 static struct file_operations proc_tgid_attr_operations = {
1867         .read           = generic_read_dir,
1868         .readdir        = proc_tgid_attr_readdir,
1869 };
1870
1871 static struct file_operations proc_tid_attr_operations = {
1872         .read           = generic_read_dir,
1873         .readdir        = proc_tid_attr_readdir,
1874 };
1875
1876 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1877                                 struct dentry *dentry, struct nameidata *nd)
1878 {
1879         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1880 }
1881
1882 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1883                                 struct dentry *dentry, struct nameidata *nd)
1884 {
1885         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1886 }
1887
1888 static struct inode_operations proc_tgid_attr_inode_operations = {
1889         .lookup         = proc_tgid_attr_lookup,
1890 };
1891
1892 static struct inode_operations proc_tid_attr_inode_operations = {
1893         .lookup         = proc_tid_attr_lookup,
1894 };
1895 #endif
1896
1897 /*
1898  * /proc/self:
1899  */
1900 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1901                               int buflen)
1902 {
1903         char tmp[30];
1904         sprintf(tmp, "%d", current->tgid);
1905         return vfs_readlink(dentry,buffer,buflen,tmp);
1906 }
1907
1908 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1909 {
1910         char tmp[30];
1911         sprintf(tmp, "%d", current->tgid);
1912         return ERR_PTR(vfs_follow_link(nd,tmp));
1913 }
1914
1915 static struct inode_operations proc_self_inode_operations = {
1916         .readlink       = proc_self_readlink,
1917         .follow_link    = proc_self_follow_link,
1918 };
1919
1920 /**
1921  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1922  * @p: task that should be flushed.
1923  *
1924  * Drops the /proc/@pid dcache entry from the hash chains.
1925  *
1926  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1927  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1928  * if the pid value is immediately reused. This is enforced by
1929  * - caller must acquire spin_lock(p->proc_lock)
1930  * - must be called before detach_pid()
1931  * - proc_pid_lookup acquires proc_lock, and checks that
1932  *   the target is not dead by looking at the attach count
1933  *   of PIDTYPE_PID.
1934  */
1935
1936 struct dentry *proc_pid_unhash(struct task_struct *p)
1937 {
1938         struct dentry *proc_dentry;
1939
1940         proc_dentry = p->proc_dentry;
1941         if (proc_dentry != NULL) {
1942
1943                 spin_lock(&dcache_lock);
1944                 spin_lock(&proc_dentry->d_lock);
1945                 if (!d_unhashed(proc_dentry)) {
1946                         dget_locked(proc_dentry);
1947                         __d_drop(proc_dentry);
1948                         spin_unlock(&proc_dentry->d_lock);
1949                 } else {
1950                         spin_unlock(&proc_dentry->d_lock);
1951                         proc_dentry = NULL;
1952                 }
1953                 spin_unlock(&dcache_lock);
1954         }
1955         return proc_dentry;
1956 }
1957
1958 /**
1959  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1960  * @proc_dentry: directoy to prune.
1961  *
1962  * Shrink the /proc directory that was used by the just killed thread.
1963  */
1964
1965 void proc_pid_flush(struct dentry *proc_dentry)
1966 {
1967         might_sleep();
1968         if(proc_dentry != NULL) {
1969                 shrink_dcache_parent(proc_dentry);
1970                 dput(proc_dentry);
1971         }
1972 }
1973
1974 /* SMP-safe */
1975 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1976 {
1977         struct task_struct *task;
1978         struct inode *inode;
1979         struct proc_inode *ei;
1980         unsigned tgid;
1981         int died;
1982
1983         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
1984                 inode = new_inode(dir->i_sb);
1985                 if (!inode)
1986                         return ERR_PTR(-ENOMEM);
1987                 ei = PROC_I(inode);
1988                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1989                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
1990                 ei->pde = NULL;
1991                 inode->i_mode = S_IFLNK|S_IRWXUGO;
1992                 inode->i_uid = inode->i_gid = 0;
1993                 inode->i_size = 64;
1994                 inode->i_op = &proc_self_inode_operations;
1995                 d_add(dentry, inode);
1996                 return NULL;
1997         }
1998         tgid = name_to_int(dentry);
1999         if (tgid == ~0U)
2000                 goto out;
2001
2002         read_lock(&tasklist_lock);
2003         task = find_task_by_pid(tgid);
2004         if (task)
2005                 get_task_struct(task);
2006         read_unlock(&tasklist_lock);
2007         if (!task)
2008                 goto out;
2009
2010         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2011
2012
2013         if (!inode) {
2014                 put_task_struct(task);
2015                 goto out;
2016         }
2017         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2018         inode->i_op = &proc_tgid_base_inode_operations;
2019         inode->i_fop = &proc_tgid_base_operations;
2020         inode->i_flags|=S_IMMUTABLE;
2021 #ifdef CONFIG_SECURITY
2022         inode->i_nlink = 5;
2023 #else
2024         inode->i_nlink = 4;
2025 #endif
2026
2027         dentry->d_op = &pid_base_dentry_operations;
2028
2029         died = 0;
2030         d_add(dentry, inode);
2031         spin_lock(&task->proc_lock);
2032         task->proc_dentry = dentry;
2033         if (!pid_alive(task)) {
2034                 dentry = proc_pid_unhash(task);
2035                 died = 1;
2036         }
2037         spin_unlock(&task->proc_lock);
2038
2039         put_task_struct(task);
2040         if (died) {
2041                 proc_pid_flush(dentry);
2042                 goto out;
2043         }
2044         return NULL;
2045 out:
2046         return ERR_PTR(-ENOENT);
2047 }
2048
2049 /* SMP-safe */
2050 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2051 {
2052         struct task_struct *task;
2053         struct task_struct *leader = proc_task(dir);
2054         struct inode *inode;
2055         unsigned tid;
2056
2057         tid = name_to_int(dentry);
2058         if (tid == ~0U)
2059                 goto out;
2060
2061         read_lock(&tasklist_lock);
2062         task = find_task_by_pid(tid);
2063         if (task)
2064                 get_task_struct(task);
2065         read_unlock(&tasklist_lock);
2066         if (!task)
2067                 goto out;
2068         if (leader->tgid != task->tgid)
2069                 goto out_drop_task;
2070
2071         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2072
2073
2074         if (!inode)
2075                 goto out_drop_task;
2076         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2077         inode->i_op = &proc_tid_base_inode_operations;
2078         inode->i_fop = &proc_tid_base_operations;
2079         inode->i_flags|=S_IMMUTABLE;
2080 #ifdef CONFIG_SECURITY
2081         inode->i_nlink = 4;
2082 #else
2083         inode->i_nlink = 3;
2084 #endif
2085
2086         dentry->d_op = &pid_base_dentry_operations;
2087
2088         d_add(dentry, inode);
2089
2090         put_task_struct(task);
2091         return NULL;
2092 out_drop_task:
2093         put_task_struct(task);
2094 out:
2095         return ERR_PTR(-ENOENT);
2096 }
2097
2098 #define PROC_NUMBUF 10
2099 #define PROC_MAXPIDS 20
2100
2101 /*
2102  * Get a few tgid's to return for filldir - we need to hold the
2103  * tasklist lock while doing this, and we must release it before
2104  * we actually do the filldir itself, so we use a temp buffer..
2105  */
2106 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2107 {
2108         struct task_struct *p;
2109         int nr_tgids = 0;
2110
2111         index--;
2112         read_lock(&tasklist_lock);
2113         p = NULL;
2114         if (version) {
2115                 p = find_task_by_pid(version);
2116                 if (p && !thread_group_leader(p))
2117                         p = NULL;
2118         }
2119
2120         if (p)
2121                 index = 0;
2122         else
2123                 p = next_task(&init_task);
2124
2125         for ( ; p != &init_task; p = next_task(p)) {
2126                 int tgid = p->pid;
2127                 if (!pid_alive(p))
2128                         continue;
2129                 if (--index >= 0)
2130                         continue;
2131                 tgids[nr_tgids] = tgid;
2132                 nr_tgids++;
2133                 if (nr_tgids >= PROC_MAXPIDS)
2134                         break;
2135         }
2136         read_unlock(&tasklist_lock);
2137         return nr_tgids;
2138 }
2139
2140 /*
2141  * Get a few tid's to return for filldir - we need to hold the
2142  * tasklist lock while doing this, and we must release it before
2143  * we actually do the filldir itself, so we use a temp buffer..
2144  */
2145 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2146 {
2147         struct task_struct *leader_task = proc_task(dir);
2148         struct task_struct *task = leader_task;
2149         int nr_tids = 0;
2150
2151         index -= 2;
2152         read_lock(&tasklist_lock);
2153         /*
2154          * The starting point task (leader_task) might be an already
2155          * unlinked task, which cannot be used to access the task-list
2156          * via next_thread().
2157          */
2158         if (pid_alive(task)) do {
2159                 int tid = task->pid;
2160
2161                 if (--index >= 0)
2162                         continue;
2163                 if (tids != NULL)
2164                         tids[nr_tids] = tid;
2165                 nr_tids++;
2166                 if (nr_tids >= PROC_MAXPIDS)
2167                         break;
2168         } while ((task = next_thread(task)) != leader_task);
2169         read_unlock(&tasklist_lock);
2170         return nr_tids;
2171 }
2172
2173 /* for the /proc/ directory itself, after non-process stuff has been done */
2174 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2175 {
2176         unsigned int tgid_array[PROC_MAXPIDS];
2177         char buf[PROC_NUMBUF];
2178         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2179         unsigned int nr_tgids, i;
2180         int next_tgid;
2181
2182         if (!nr) {
2183                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2184                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2185                         return 0;
2186                 filp->f_pos++;
2187                 nr++;
2188         }
2189
2190         /* f_version caches the tgid value that the last readdir call couldn't
2191          * return. lseek aka telldir automagically resets f_version to 0.
2192          */
2193         next_tgid = filp->f_version;
2194         filp->f_version = 0;
2195         for (;;) {
2196                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2197                 if (!nr_tgids) {
2198                         /* no more entries ! */
2199                         break;
2200                 }
2201                 next_tgid = 0;
2202
2203                 /* do not use the last found pid, reserve it for next_tgid */
2204                 if (nr_tgids == PROC_MAXPIDS) {
2205                         nr_tgids--;
2206                         next_tgid = tgid_array[nr_tgids];
2207                 }
2208
2209                 for (i=0;i<nr_tgids;i++) {
2210                         int tgid = tgid_array[i];
2211                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2212                         unsigned long j = PROC_NUMBUF;
2213
2214                         do
2215                                 buf[--j] = '0' + (tgid % 10);
2216                         while ((tgid /= 10) != 0);
2217
2218                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2219                                 /* returning this tgid failed, save it as the first
2220                                  * pid for the next readir call */
2221                                 filp->f_version = tgid_array[i];
2222                                 goto out;
2223                         }
2224                         filp->f_pos++;
2225                         nr++;
2226                 }
2227         }
2228 out:
2229         return 0;
2230 }
2231
2232 /* for the /proc/TGID/task/ directories */
2233 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2234 {
2235         unsigned int tid_array[PROC_MAXPIDS];
2236         char buf[PROC_NUMBUF];
2237         unsigned int nr_tids, i;
2238         struct dentry *dentry = filp->f_dentry;
2239         struct inode *inode = dentry->d_inode;
2240         int retval = -ENOENT;
2241         ino_t ino;
2242         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2243
2244         if (!pid_alive(proc_task(inode)))
2245                 goto out;
2246         retval = 0;
2247
2248         switch (pos) {
2249         case 0:
2250                 ino = inode->i_ino;
2251                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2252                         goto out;
2253                 pos++;
2254                 /* fall through */
2255         case 1:
2256                 ino = parent_ino(dentry);
2257                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2258                         goto out;
2259                 pos++;
2260                 /* fall through */
2261         }
2262
2263         nr_tids = get_tid_list(pos, tid_array, inode);
2264         inode->i_nlink = pos + nr_tids;
2265
2266         for (i = 0; i < nr_tids; i++) {
2267                 unsigned long j = PROC_NUMBUF;
2268                 int tid = tid_array[i];
2269
2270                 ino = fake_ino(tid,PROC_TID_INO);
2271
2272                 do
2273                         buf[--j] = '0' + (tid % 10);
2274                 while ((tid /= 10) != 0);
2275
2276                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2277                         break;
2278                 pos++;
2279         }
2280 out:
2281         filp->f_pos = pos;
2282         return retval;
2283 }