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 mounts_op;
 543 struct proc_mounts {
 544         struct seq_file m;
 545         int event;
 546 };
 547
 548 static int mounts_open(struct inode *inode, struct file *file)
 549 {
 550         struct task_struct *task = proc_task(inode);
 551         struct namespace *namespace;
 552         struct proc_mounts *p;
 553         int ret = -EINVAL;
 554
 555         task_lock(task);
 556         namespace = task->namespace;
 557         if (namespace)
 558                 get_namespace(namespace);
 559         task_unlock(task);
 560
 561         if (namespace) {
 562                 ret = -ENOMEM;
 563                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 564                 if (p) {
 565                         file->private_data = &p->m;
 566                         ret = seq_open(file, &mounts_op);
 567                         if (!ret) {
 568                                 p->m.private = namespace;
 569                                 p->event = namespace->event;
 570                                 return 0;
 571                         }
 572                         kfree(p);
 573                 }
 574                 put_namespace(namespace);
 575         }
 576         return ret;
 577 }
 578
 579 static int mounts_release(struct inode *inode, struct file *file)
 580 {
 581         struct seq_file *m = file->private_data;
 582         struct namespace *namespace = m->private;
 583         put_namespace(namespace);
 584         return seq_release(inode, file);
 585 }
 586
 587 static unsigned mounts_poll(struct file *file, poll_table *wait)
 588 {
 589         struct proc_mounts *p = file->private_data;
 590         struct namespace *ns = p->m.private;
 591         unsigned res = 0;
 592
 593         poll_wait(file, &ns->poll, wait);
 594
 595         spin_lock(&vfsmount_lock);
 596         if (p->event != ns->event) {
 597                 p->event = ns->event;
 598                 res = POLLERR;
 599         }
 600         spin_unlock(&vfsmount_lock);
 601
 602         return res;
 603 }
 604
 605 static struct file_operations proc_mounts_operations = {
 606         .open           = mounts_open,
 607         .read           = seq_read,
 608         .llseek         = seq_lseek,
 609         .release        = mounts_release,
 610         .poll           = mounts_poll,
 611 };
 612
 613 extern struct seq_operations mountstats_op;
 614 static int mountstats_open(struct inode *inode, struct file *file)
 615 {
 616         struct task_struct *task = proc_task(inode);
 617         int ret = seq_open(file, &mountstats_op);
 618
 619         if (!ret) {
 620                 struct seq_file *m = file->private_data;
 621                 struct namespace *namespace;
 622                 task_lock(task);
 623                 namespace = task->namespace;
 624                 if (namespace)
 625                         get_namespace(namespace);
 626                 task_unlock(task);
 627
 628                 if (namespace)
 629                         m->private = namespace;
 630                 else {
 631                         seq_release(inode, file);
 632                         ret = -EINVAL;
 633                 }
 634         }
 635         return ret;
 636 }
 637
 638 static struct file_operations proc_mountstats_operations = {
 639         .open           = mountstats_open,
 640         .read           = seq_read,
 641         .llseek         = seq_lseek,
 642         .release        = mounts_release,
 643 };
 644
 645 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 646
 647 static ssize_t proc_info_read(struct file * file, char __user * buf,
 648                           size_t count, loff_t *ppos)
 649 {
 650         struct inode * inode = file->f_dentry->d_inode;
 651         unsigned long page;
 652         ssize_t length;
 653         struct task_struct *task = proc_task(inode);
 654
 655         if (count > PROC_BLOCK_SIZE)
 656                 count = PROC_BLOCK_SIZE;
 657         if (!(page = __get_free_page(GFP_KERNEL)))
 658                 return -ENOMEM;
 659
 660         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 661
 662         if (length >= 0)
 663                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 664         free_page(page);
 665         return length;
 666 }
 667
 668 static struct file_operations proc_info_file_operations = {
 669         .read           = proc_info_read,
 670 };
 671
 672 static int mem_open(struct inode* inode, struct file* file)
 673 {
 674         file->private_data = (void*)((long)current->self_exec_id);
 675         return 0;
 676 }
 677
 678 static ssize_t mem_read(struct file * file, char __user * buf,
 679                         size_t count, loff_t *ppos)
 680 {
 681         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 682         char *page;
 683         unsigned long src = *ppos;
 684         int ret = -ESRCH;
 685         struct mm_struct *mm;
 686
 687         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 688                 goto out;
 689
 690         ret = -ENOMEM;
 691         page = (char *)__get_free_page(GFP_USER);
 692         if (!page)
 693                 goto out;
 694
 695         ret = 0;
 696
 697         mm = get_task_mm(task);
 698         if (!mm)
 699                 goto out_free;
 700
 701         ret = -EIO;
 702
 703         if (file->private_data != (void*)((long)current->self_exec_id))
 704                 goto out_put;
 705
 706         ret = 0;
 707
 708         while (count > 0) {
 709                 int this_len, retval;
 710
 711                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 712                 retval = access_process_vm(task, src, page, this_len, 0);
 713                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 714                         if (!ret)
 715                                 ret = -EIO;
 716                         break;
 717                 }
 718
 719                 if (copy_to_user(buf, page, retval)) {
 720                         ret = -EFAULT;
 721                         break;
 722                 }
 723
 724                 ret += retval;
 725                 src += retval;
 726                 buf += retval;
 727                 count -= retval;
 728         }
 729         *ppos = src;
 730
 731 out_put:
 732         mmput(mm);
 733 out_free:
 734         free_page((unsigned long) page);
 735 out:
 736         return ret;
 737 }
 738
 739 #define mem_write NULL
 740
 741 #ifndef mem_write
 742 /* This is a security hazard */
 743 static ssize_t mem_write(struct file * file, const char * buf,
 744                          size_t count, loff_t *ppos)
 745 {
 746         int copied = 0;
 747         char *page;
 748         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 749         unsigned long dst = *ppos;
 750
 751         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 752                 return -ESRCH;
 753
 754         page = (char *)__get_free_page(GFP_USER);
 755         if (!page)
 756                 return -ENOMEM;
 757
 758         while (count > 0) {
 759                 int this_len, retval;
 760
 761                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 762                 if (copy_from_user(page, buf, this_len)) {
 763                         copied = -EFAULT;
 764                         break;
 765                 }
 766                 retval = access_process_vm(task, dst, page, this_len, 1);
 767                 if (!retval) {
 768                         if (!copied)
 769                                 copied = -EIO;
 770                         break;
 771                 }
 772                 copied += retval;
 773                 buf += retval;
 774                 dst += retval;
 775                 count -= retval;
 776         }
 777         *ppos = dst;
 778         free_page((unsigned long) page);
 779         return copied;
 780 }
 781 #endif
 782
 783 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 784 {
 785         switch (orig) {
 786         case 0:
 787                 file->f_pos = offset;
 788                 break;
 789         case 1:
 790                 file->f_pos += offset;
 791                 break;
 792         default:
 793                 return -EINVAL;
 794         }
 795         force_successful_syscall_return();
 796         return file->f_pos;
 797 }
 798
 799 static struct file_operations proc_mem_operations = {
 800         .llseek         = mem_lseek,
 801         .read           = mem_read,
 802         .write          = mem_write,
 803         .open           = mem_open,
 804 };
 805
 806 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 807                                 size_t count, loff_t *ppos)
 808 {
 809         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 810         char buffer[8];
 811         size_t len;
 812         int oom_adjust = task->oomkilladj;
 813         loff_t __ppos = *ppos;
 814
 815         len = sprintf(buffer, "%i\n", oom_adjust);
 816         if (__ppos >= len)
 817                 return 0;
 818         if (count > len-__ppos)
 819                 count = len-__ppos;
 820         if (copy_to_user(buf, buffer + __ppos, count))
 821                 return -EFAULT;
 822         *ppos = __ppos + count;
 823         return count;
 824 }
 825
 826 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 827                                 size_t count, loff_t *ppos)
 828 {
 829         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 830         char buffer[8], *end;
 831         int oom_adjust;
 832
 833         if (!capable(CAP_SYS_RESOURCE))
 834                 return -EPERM;
 835         memset(buffer, 0, 8);
 836         if (count > 6)
 837                 count = 6;
 838         if (copy_from_user(buffer, buf, count))
 839                 return -EFAULT;
 840         oom_adjust = simple_strtol(buffer, &end, 0);
 841         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 842                 return -EINVAL;
 843         if (*end == '\n')
 844                 end++;
 845         task->oomkilladj = oom_adjust;
 846         if (end - buffer == 0)
 847                 return -EIO;
 848         return end - buffer;
 849 }
 850
 851 static struct file_operations proc_oom_adjust_operations = {
 852         .read           = oom_adjust_read,
 853         .write          = oom_adjust_write,
 854 };
 855
 856 #ifdef CONFIG_AUDITSYSCALL
 857 #define TMPBUFLEN 21
 858 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 859                                   size_t count, loff_t *ppos)
 860 {
 861         struct inode * inode = file->f_dentry->d_inode;
 862         struct task_struct *task = proc_task(inode);
 863         ssize_t length;
 864         char tmpbuf[TMPBUFLEN];
 865
 866         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 867                                 audit_get_loginuid(task->audit_context));
 868         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 869 }
 870
 871 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 872                                    size_t count, loff_t *ppos)
 873 {
 874         struct inode * inode = file->f_dentry->d_inode;
 875         char *page, *tmp;
 876         ssize_t length;
 877         struct task_struct *task = proc_task(inode);
 878         uid_t loginuid;
 879
 880         if (!capable(CAP_AUDIT_CONTROL))
 881                 return -EPERM;
 882
 883         if (current != task)
 884                 return -EPERM;
 885
 886         if (count >= PAGE_SIZE)
 887                 count = PAGE_SIZE - 1;
 888
 889         if (*ppos != 0) {
 890                 /* No partial writes. */
 891                 return -EINVAL;
 892         }
 893         page = (char*)__get_free_page(GFP_USER);
 894         if (!page)
 895                 return -ENOMEM;
 896         length = -EFAULT;
 897         if (copy_from_user(page, buf, count))
 898                 goto out_free_page;
 899
 900         page[count] = '\0';
 901         loginuid = simple_strtoul(page, &tmp, 10);
 902         if (tmp == page) {
 903                 length = -EINVAL;
 904                 goto out_free_page;
 905
 906         }
 907         length = audit_set_loginuid(task, loginuid);
 908         if (likely(length == 0))
 909                 length = count;
 910
 911 out_free_page:
 912         free_page((unsigned long) page);
 913         return length;
 914 }
 915
 916 static struct file_operations proc_loginuid_operations = {
 917         .read           = proc_loginuid_read,
 918         .write          = proc_loginuid_write,
 919 };
 920 #endif
 921
 922 #ifdef CONFIG_SECCOMP
 923 static ssize_t seccomp_read(struct file *file, char __user *buf,
 924                             size_t count, loff_t *ppos)
 925 {
 926         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
 927         char __buf[20];
 928         loff_t __ppos = *ppos;
 929         size_t len;
 930
 931         /* no need to print the trailing zero, so use only len */
 932         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
 933         if (__ppos >= len)
 934                 return 0;
 935         if (count > len - __ppos)
 936                 count = len - __ppos;
 937         if (copy_to_user(buf, __buf + __ppos, count))
 938                 return -EFAULT;
 939         *ppos = __ppos + count;
 940         return count;
 941 }
 942
 943 static ssize_t seccomp_write(struct file *file, const char __user *buf,
 944                              size_t count, loff_t *ppos)
 945 {
 946         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
 947         char __buf[20], *end;
 948         unsigned int seccomp_mode;
 949
 950         /* can set it only once to be even more secure */
 951         if (unlikely(tsk->seccomp.mode))
 952                 return -EPERM;
 953
 954         memset(__buf, 0, sizeof(__buf));
 955         count = min(count, sizeof(__buf) - 1);
 956         if (copy_from_user(__buf, buf, count))
 957                 return -EFAULT;
 958         seccomp_mode = simple_strtoul(__buf, &end, 0);
 959         if (*end == '\n')
 960                 end++;
 961         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
 962                 tsk->seccomp.mode = seccomp_mode;
 963                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
 964         } else
 965                 return -EINVAL;
 966         if (unlikely(!(end - __buf)))
 967                 return -EIO;
 968         return end - __buf;
 969 }
 970
 971 static struct file_operations proc_seccomp_operations = {
 972         .read           = seccomp_read,
 973         .write          = seccomp_write,
 974 };
 975 #endif /* CONFIG_SECCOMP */
 976
 977 static int proc_check_dentry_visible(struct inode *inode,
 978         struct dentry *dentry, struct vfsmount *mnt)
 979 {
 980         /* Verify that the current process can already see the
 981          * file pointed at by the file descriptor.
 982          * This prevents /proc from being an accidental information leak.
 983          *
 984          * This prevents access to files that are not visible do to
 985          * being on the otherside of a chroot, in a different
 986          * namespace, or are simply process local (like pipes).
 987          */
 988         struct task_struct *task;
 989         struct files_struct *task_files, *files;
 990         int error = -EACCES;
 991
 992         /* See if the the two tasks share a commone set of
 993          * file descriptors.  If so everything is visible.
 994          */
 995         task = proc_task(inode);
 996         if (!task)
 997                 goto out;
 998         files = get_files_struct(current);
 999         task_files = get_files_struct(task);
1000         if (files && task_files && (files == task_files))
1001                 error = 0;
1002         if (task_files)
1003                 put_files_struct(task_files);
1004         if (files)
1005                 put_files_struct(files);
1006         if (!error)
1007                 goto out;
1008
1009         /* If the two tasks don't share a common set of file
1010          * descriptors see if the destination dentry is already
1011          * visible in the current tasks filesystem namespace.
1012          */
1013         error = proc_check_chroot(dentry, mnt);
1014 out:
1015         return error;
1016
1017 }
1018
1019 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1020 {
1021         struct inode *inode = dentry->d_inode;
1022         int error = -EACCES;
1023
1024         /* We don't need a base pointer in the /proc filesystem */
1025         path_release(nd);
1026
1027         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1028                 goto out;
1029
1030         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1031         nd->last_type = LAST_BIND;
1032         if (error)
1033                 goto out;
1034
1035         /* Only return files this task can already see */
1036         error = proc_check_dentry_visible(inode, nd->dentry, nd->mnt);
1037         if (error)
1038                 path_release(nd);
1039 out:
1040         return ERR_PTR(error);
1041 }
1042
1043 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1044                             char __user *buffer, int buflen)
1045 {
1046         struct inode * inode;
1047         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1048         int len;
1049
1050         if (!tmp)
1051                 return -ENOMEM;
1052
1053         inode = dentry->d_inode;
1054         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1055         len = PTR_ERR(path);
1056         if (IS_ERR(path))
1057                 goto out;
1058         len = tmp + PAGE_SIZE - 1 - path;
1059
1060         if (len > buflen)
1061                 len = buflen;
1062         if (copy_to_user(buffer, path, len))
1063                 len = -EFAULT;
1064  out:
1065         free_page((unsigned long)tmp);
1066         return len;
1067 }
1068
1069 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1070 {
1071         int error = -EACCES;
1072         struct inode *inode = dentry->d_inode;
1073         struct dentry *de;
1074         struct vfsmount *mnt = NULL;
1075
1076
1077         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1078                 goto out;
1079
1080         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1081         if (error)
1082                 goto out;
1083
1084         /* Only return files this task can already see */
1085         error = proc_check_dentry_visible(inode, de, mnt);
1086         if (error)
1087                 goto out_put;
1088
1089         error = do_proc_readlink(de, mnt, buffer, buflen);
1090 out_put:
1091         dput(de);
1092         mntput(mnt);
1093 out:
1094         return error;
1095 }
1096
1097 static struct inode_operations proc_pid_link_inode_operations = {
1098         .readlink       = proc_pid_readlink,
1099         .follow_link    = proc_pid_follow_link
1100 };
1101
1102 #define NUMBUF 10
1103
1104 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1105 {
1106         struct dentry *dentry = filp->f_dentry;
1107         struct inode *inode = dentry->d_inode;
1108         struct task_struct *p = proc_task(inode);
1109         unsigned int fd, tid, ino;
1110         int retval;
1111         char buf[NUMBUF];
1112         struct files_struct * files;
1113         struct fdtable *fdt;
1114
1115         retval = -ENOENT;
1116         if (!pid_alive(p))
1117                 goto out;
1118         retval = 0;
1119         tid = p->pid;
1120
1121         fd = filp->f_pos;
1122         switch (fd) {
1123                 case 0:
1124                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1125                                 goto out;
1126                         filp->f_pos++;
1127                 case 1:
1128                         ino = parent_ino(dentry);
1129                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1130                                 goto out;
1131                         filp->f_pos++;
1132                 default:
1133                         files = get_files_struct(p);
1134                         if (!files)
1135                                 goto out;
1136                         rcu_read_lock();
1137                         fdt = files_fdtable(files);
1138                         for (fd = filp->f_pos-2;
1139                              fd < fdt->max_fds;
1140                              fd++, filp->f_pos++) {
1141                                 unsigned int i,j;
1142
1143                                 if (!fcheck_files(files, fd))
1144                                         continue;
1145                                 rcu_read_unlock();
1146
1147                                 j = NUMBUF;
1148                                 i = fd;
1149                                 do {
1150                                         j--;
1151                                         buf[j] = '0' + (i % 10);
1152                                         i /= 10;
1153                                 } while (i);
1154
1155                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1156                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1157                                         rcu_read_lock();
1158                                         break;
1159                                 }
1160                                 rcu_read_lock();
1161                         }
1162                         rcu_read_unlock();
1163                         put_files_struct(files);
1164         }
1165 out:
1166         return retval;
1167 }
1168
1169 static int proc_pident_readdir(struct file *filp,
1170                 void *dirent, filldir_t filldir,
1171                 struct pid_entry *ents, unsigned int nents)
1172 {
1173         int i;
1174         int pid;
1175         struct dentry *dentry = filp->f_dentry;
1176         struct inode *inode = dentry->d_inode;
1177         struct pid_entry *p;
1178         ino_t ino;
1179         int ret;
1180
1181         ret = -ENOENT;
1182         if (!pid_alive(proc_task(inode)))
1183                 goto out;
1184
1185         ret = 0;
1186         pid = proc_task(inode)->pid;
1187         i = filp->f_pos;
1188         switch (i) {
1189         case 0:
1190                 ino = inode->i_ino;
1191                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1192                         goto out;
1193                 i++;
1194                 filp->f_pos++;
1195                 /* fall through */
1196         case 1:
1197                 ino = parent_ino(dentry);
1198                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1199                         goto out;
1200                 i++;
1201                 filp->f_pos++;
1202                 /* fall through */
1203         default:
1204                 i -= 2;
1205                 if (i >= nents) {
1206                         ret = 1;
1207                         goto out;
1208                 }
1209                 p = ents + i;
1210                 while (p->name) {
1211                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1212                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1213                                 goto out;
1214                         filp->f_pos++;
1215                         p++;
1216                 }
1217         }
1218
1219         ret = 1;
1220 out:
1221         return ret;
1222 }
1223
1224 static int proc_tgid_base_readdir(struct file * filp,
1225                              void * dirent, filldir_t filldir)
1226 {
1227         return proc_pident_readdir(filp,dirent,filldir,
1228                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1229 }
1230
1231 static int proc_tid_base_readdir(struct file * filp,
1232                              void * dirent, filldir_t filldir)
1233 {
1234         return proc_pident_readdir(filp,dirent,filldir,
1235                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1236 }
1237
1238 /* building an inode */
1239
1240 static int task_dumpable(struct task_struct *task)
1241 {
1242         int dumpable = 0;
1243         struct mm_struct *mm;
1244
1245         task_lock(task);
1246         mm = task->mm;
1247         if (mm)
1248                 dumpable = mm->dumpable;
1249         task_unlock(task);
1250         if(dumpable == 1)
1251                 return 1;
1252         return 0;
1253 }
1254
1255
1256 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1257 {
1258         struct inode * inode;
1259         struct proc_inode *ei;
1260
1261         /* We need a new inode */
1262
1263         inode = new_inode(sb);
1264         if (!inode)
1265                 goto out;
1266
1267         /* Common stuff */
1268         ei = PROC_I(inode);
1269         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1270         inode->i_ino = fake_ino(task->pid, ino);
1271
1272         if (!pid_alive(task))
1273                 goto out_unlock;
1274
1275         /*
1276          * grab the reference to task.
1277          */
1278         get_task_struct(task);
1279         ei->task = task;
1280         inode->i_uid = 0;
1281         inode->i_gid = 0;
1282         if (task_dumpable(task)) {
1283                 inode->i_uid = task->euid;
1284                 inode->i_gid = task->egid;
1285         }
1286         security_task_to_inode(task, inode);
1287
1288 out:
1289         return inode;
1290
1291 out_unlock:
1292         iput(inode);
1293         return NULL;
1294 }
1295
1296 /* dentry stuff */
1297
1298 /*
1299  *      Exceptional case: normally we are not allowed to unhash a busy
1300  * directory. In this case, however, we can do it - no aliasing problems
1301  * due to the way we treat inodes.
1302  *
1303  * Rewrite the inode's ownerships here because the owning task may have
1304  * performed a setuid(), etc.
1305  */
1306 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1307 {
1308         struct inode *inode = dentry->d_inode;
1309         struct task_struct *task = proc_task(inode);
1310         if (pid_alive(task)) {
1311                 if (task_dumpable(task)) {
1312                         inode->i_uid = task->euid;
1313                         inode->i_gid = task->egid;
1314                 } else {
1315                         inode->i_uid = 0;
1316                         inode->i_gid = 0;
1317                 }
1318                 security_task_to_inode(task, inode);
1319                 return 1;
1320         }
1321         d_drop(dentry);
1322         return 0;
1323 }
1324
1325 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1326 {
1327         struct inode *inode = dentry->d_inode;
1328         struct task_struct *task = proc_task(inode);
1329         int fd = proc_fd(inode);
1330         struct files_struct *files;
1331
1332         files = get_files_struct(task);
1333         if (files) {
1334                 rcu_read_lock();
1335                 if (fcheck_files(files, fd)) {
1336                         rcu_read_unlock();
1337                         put_files_struct(files);
1338                         if (task_dumpable(task)) {
1339                                 inode->i_uid = task->euid;
1340                                 inode->i_gid = task->egid;
1341                         } else {
1342                                 inode->i_uid = 0;
1343                                 inode->i_gid = 0;
1344                         }
1345                         security_task_to_inode(task, inode);
1346                         return 1;
1347                 }
1348                 rcu_read_unlock();
1349                 put_files_struct(files);
1350         }
1351         d_drop(dentry);
1352         return 0;
1353 }
1354
1355 static int pid_delete_dentry(struct dentry * dentry)
1356 {
1357         /* Is the task we represent dead?
1358          * If so, then don't put the dentry on the lru list,
1359          * kill it immediately.
1360          */
1361         return !pid_alive(proc_task(dentry->d_inode));
1362 }
1363
1364 static struct dentry_operations tid_fd_dentry_operations =
1365 {
1366         .d_revalidate   = tid_fd_revalidate,
1367         .d_delete       = pid_delete_dentry,
1368 };
1369
1370 static struct dentry_operations pid_dentry_operations =
1371 {
1372         .d_revalidate   = pid_revalidate,
1373         .d_delete       = pid_delete_dentry,
1374 };
1375
1376 /* Lookups */
1377
1378 static unsigned name_to_int(struct dentry *dentry)
1379 {
1380         const char *name = dentry->d_name.name;
1381         int len = dentry->d_name.len;
1382         unsigned n = 0;
1383
1384         if (len > 1 && *name == '0')
1385                 goto out;
1386         while (len-- > 0) {
1387                 unsigned c = *name++ - '0';
1388                 if (c > 9)
1389                         goto out;
1390                 if (n >= (~0U-9)/10)
1391                         goto out;
1392                 n *= 10;
1393                 n += c;
1394         }
1395         return n;
1396 out:
1397         return ~0U;
1398 }
1399
1400 /* SMP-safe */
1401 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1402 {
1403         struct task_struct *task = proc_task(dir);
1404         unsigned fd = name_to_int(dentry);
1405         struct dentry *result = ERR_PTR(-ENOENT);
1406         struct file * file;
1407         struct files_struct * files;
1408         struct inode *inode;
1409         struct proc_inode *ei;
1410
1411         if (fd == ~0U)
1412                 goto out;
1413         if (!pid_alive(task))
1414                 goto out;
1415
1416         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1417         if (!inode)
1418                 goto out;
1419         ei = PROC_I(inode);
1420         ei->fd = fd;
1421         files = get_files_struct(task);
1422         if (!files)
1423                 goto out_unlock;
1424         inode->i_mode = S_IFLNK;
1425
1426         /*
1427          * We are not taking a ref to the file structure, so we must
1428          * hold ->file_lock.
1429          */
1430         spin_lock(&files->file_lock);
1431         file = fcheck_files(files, fd);
1432         if (!file)
1433                 goto out_unlock2;
1434         if (file->f_mode & 1)
1435                 inode->i_mode |= S_IRUSR | S_IXUSR;
1436         if (file->f_mode & 2)
1437                 inode->i_mode |= S_IWUSR | S_IXUSR;
1438         spin_unlock(&files->file_lock);
1439         put_files_struct(files);
1440         inode->i_op = &proc_pid_link_inode_operations;
1441         inode->i_size = 64;
1442         ei->op.proc_get_link = proc_fd_link;
1443         dentry->d_op = &tid_fd_dentry_operations;
1444         d_add(dentry, inode);
1445         /* Close the race of the process dying before we return the dentry */
1446         if (tid_fd_revalidate(dentry, NULL))
1447                 result = NULL;
1448 out:
1449         return result;
1450
1451 out_unlock2:
1452         spin_unlock(&files->file_lock);
1453         put_files_struct(files);
1454 out_unlock:
1455         iput(inode);
1456         goto out;
1457 }
1458
1459 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1460 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1461 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat);
1462
1463 static struct file_operations proc_fd_operations = {
1464         .read           = generic_read_dir,
1465         .readdir        = proc_readfd,
1466 };
1467
1468 static struct file_operations proc_task_operations = {
1469         .read           = generic_read_dir,
1470         .readdir        = proc_task_readdir,
1471 };
1472
1473 /*
1474  * proc directories can do almost nothing..
1475  */
1476 static struct inode_operations proc_fd_inode_operations = {
1477         .lookup         = proc_lookupfd,
1478 };
1479
1480 static struct inode_operations proc_task_inode_operations = {
1481         .lookup         = proc_task_lookup,
1482         .getattr        = proc_task_getattr,
1483 };
1484
1485 #ifdef CONFIG_SECURITY
1486 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1487                                   size_t count, loff_t *ppos)
1488 {
1489         struct inode * inode = file->f_dentry->d_inode;
1490         unsigned long page;
1491         ssize_t length;
1492         struct task_struct *task = proc_task(inode);
1493
1494         if (count > PAGE_SIZE)
1495                 count = PAGE_SIZE;
1496         if (!(page = __get_free_page(GFP_KERNEL)))
1497                 return -ENOMEM;
1498
1499         length = security_getprocattr(task,
1500                                       (char*)file->f_dentry->d_name.name,
1501                                       (void*)page, count);
1502         if (length >= 0)
1503                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1504         free_page(page);
1505         return length;
1506 }
1507
1508 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1509                                    size_t count, loff_t *ppos)
1510 {
1511         struct inode * inode = file->f_dentry->d_inode;
1512         char *page;
1513         ssize_t length;
1514         struct task_struct *task = proc_task(inode);
1515
1516         if (count > PAGE_SIZE)
1517                 count = PAGE_SIZE;
1518         if (*ppos != 0) {
1519                 /* No partial writes. */
1520                 return -EINVAL;
1521         }
1522         page = (char*)__get_free_page(GFP_USER);
1523         if (!page)
1524                 return -ENOMEM;
1525         length = -EFAULT;
1526         if (copy_from_user(page, buf, count))
1527                 goto out;
1528
1529         length = security_setprocattr(task,
1530                                       (char*)file->f_dentry->d_name.name,
1531                                       (void*)page, count);
1532 out:
1533         free_page((unsigned long) page);
1534         return length;
1535 }
1536
1537 static struct file_operations proc_pid_attr_operations = {
1538         .read           = proc_pid_attr_read,
1539         .write          = proc_pid_attr_write,
1540 };
1541
1542 static struct file_operations proc_tid_attr_operations;
1543 static struct inode_operations proc_tid_attr_inode_operations;
1544 static struct file_operations proc_tgid_attr_operations;
1545 static struct inode_operations proc_tgid_attr_inode_operations;
1546 #endif
1547
1548 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1549
1550 /* SMP-safe */
1551 static struct dentry *proc_pident_lookup(struct inode *dir,
1552                                          struct dentry *dentry,
1553                                          struct pid_entry *ents)
1554 {
1555         struct inode *inode;
1556         struct dentry *error;
1557         struct task_struct *task = proc_task(dir);
1558         struct pid_entry *p;
1559         struct proc_inode *ei;
1560
1561         error = ERR_PTR(-ENOENT);
1562         inode = NULL;
1563
1564         if (!pid_alive(task))
1565                 goto out;
1566
1567         for (p = ents; p->name; p++) {
1568                 if (p->len != dentry->d_name.len)
1569                         continue;
1570                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1571                         break;
1572         }
1573         if (!p->name)
1574                 goto out;
1575
1576         error = ERR_PTR(-EINVAL);
1577         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1578         if (!inode)
1579                 goto out;
1580
1581         ei = PROC_I(inode);
1582         inode->i_mode = p->mode;
1583         /*
1584          * Yes, it does not scale. And it should not. Don't add
1585          * new entries into /proc/<tgid>/ without very good reasons.
1586          */
1587         switch(p->type) {
1588                 case PROC_TGID_TASK:
1589                         inode->i_nlink = 2;
1590                         inode->i_op = &proc_task_inode_operations;
1591                         inode->i_fop = &proc_task_operations;
1592                         break;
1593                 case PROC_TID_FD:
1594                 case PROC_TGID_FD:
1595                         inode->i_nlink = 2;
1596                         inode->i_op = &proc_fd_inode_operations;
1597                         inode->i_fop = &proc_fd_operations;
1598                         break;
1599                 case PROC_TID_EXE:
1600                 case PROC_TGID_EXE:
1601                         inode->i_op = &proc_pid_link_inode_operations;
1602                         ei->op.proc_get_link = proc_exe_link;
1603                         break;
1604                 case PROC_TID_CWD:
1605                 case PROC_TGID_CWD:
1606                         inode->i_op = &proc_pid_link_inode_operations;
1607                         ei->op.proc_get_link = proc_cwd_link;
1608                         break;
1609                 case PROC_TID_ROOT:
1610                 case PROC_TGID_ROOT:
1611                         inode->i_op = &proc_pid_link_inode_operations;
1612                         ei->op.proc_get_link = proc_root_link;
1613                         break;
1614                 case PROC_TID_ENVIRON:
1615                 case PROC_TGID_ENVIRON:
1616                         inode->i_fop = &proc_info_file_operations;
1617                         ei->op.proc_read = proc_pid_environ;
1618                         break;
1619                 case PROC_TID_AUXV:
1620                 case PROC_TGID_AUXV:
1621                         inode->i_fop = &proc_info_file_operations;
1622                         ei->op.proc_read = proc_pid_auxv;
1623                         break;
1624                 case PROC_TID_STATUS:
1625                 case PROC_TGID_STATUS:
1626                         inode->i_fop = &proc_info_file_operations;
1627                         ei->op.proc_read = proc_pid_status;
1628                         break;
1629                 case PROC_TID_STAT:
1630                         inode->i_fop = &proc_info_file_operations;
1631                         ei->op.proc_read = proc_tid_stat;
1632                         break;
1633                 case PROC_TGID_STAT:
1634                         inode->i_fop = &proc_info_file_operations;
1635                         ei->op.proc_read = proc_tgid_stat;
1636                         break;
1637                 case PROC_TID_CMDLINE:
1638                 case PROC_TGID_CMDLINE:
1639                         inode->i_fop = &proc_info_file_operations;
1640                         ei->op.proc_read = proc_pid_cmdline;
1641                         break;
1642                 case PROC_TID_STATM:
1643                 case PROC_TGID_STATM:
1644                         inode->i_fop = &proc_info_file_operations;
1645                         ei->op.proc_read = proc_pid_statm;
1646                         break;
1647                 case PROC_TID_MAPS:
1648                 case PROC_TGID_MAPS:
1649                         inode->i_fop = &proc_maps_operations;
1650                         break;
1651 #ifdef CONFIG_NUMA
1652                 case PROC_TID_NUMA_MAPS:
1653                 case PROC_TGID_NUMA_MAPS:
1654                         inode->i_fop = &proc_numa_maps_operations;
1655                         break;
1656 #endif
1657                 case PROC_TID_MEM:
1658                 case PROC_TGID_MEM:
1659                         inode->i_fop = &proc_mem_operations;
1660                         break;
1661 #ifdef CONFIG_SECCOMP
1662                 case PROC_TID_SECCOMP:
1663                 case PROC_TGID_SECCOMP:
1664                         inode->i_fop = &proc_seccomp_operations;
1665                         break;
1666 #endif /* CONFIG_SECCOMP */
1667                 case PROC_TID_MOUNTS:
1668                 case PROC_TGID_MOUNTS:
1669                         inode->i_fop = &proc_mounts_operations;
1670                         break;
1671 #ifdef CONFIG_MMU
1672                 case PROC_TID_SMAPS:
1673                 case PROC_TGID_SMAPS:
1674                         inode->i_fop = &proc_smaps_operations;
1675                         break;
1676 #endif
1677                 case PROC_TID_MOUNTSTATS:
1678                 case PROC_TGID_MOUNTSTATS:
1679                         inode->i_fop = &proc_mountstats_operations;
1680                         break;
1681 #ifdef CONFIG_SECURITY
1682                 case PROC_TID_ATTR:
1683                         inode->i_nlink = 2;
1684                         inode->i_op = &proc_tid_attr_inode_operations;
1685                         inode->i_fop = &proc_tid_attr_operations;
1686                         break;
1687                 case PROC_TGID_ATTR:
1688                         inode->i_nlink = 2;
1689                         inode->i_op = &proc_tgid_attr_inode_operations;
1690                         inode->i_fop = &proc_tgid_attr_operations;
1691                         break;
1692                 case PROC_TID_ATTR_CURRENT:
1693                 case PROC_TGID_ATTR_CURRENT:
1694                 case PROC_TID_ATTR_PREV:
1695                 case PROC_TGID_ATTR_PREV:
1696                 case PROC_TID_ATTR_EXEC:
1697                 case PROC_TGID_ATTR_EXEC:
1698                 case PROC_TID_ATTR_FSCREATE:
1699                 case PROC_TGID_ATTR_FSCREATE:
1700                 case PROC_TID_ATTR_KEYCREATE:
1701                 case PROC_TGID_ATTR_KEYCREATE:
1702                         inode->i_fop = &proc_pid_attr_operations;
1703                         break;
1704 #endif
1705 #ifdef CONFIG_KALLSYMS
1706                 case PROC_TID_WCHAN:
1707                 case PROC_TGID_WCHAN:
1708                         inode->i_fop = &proc_info_file_operations;
1709                         ei->op.proc_read = proc_pid_wchan;
1710                         break;
1711 #endif
1712 #ifdef CONFIG_SCHEDSTATS
1713                 case PROC_TID_SCHEDSTAT:
1714                 case PROC_TGID_SCHEDSTAT:
1715                         inode->i_fop = &proc_info_file_operations;
1716                         ei->op.proc_read = proc_pid_schedstat;
1717                         break;
1718 #endif
1719 #ifdef CONFIG_CPUSETS
1720                 case PROC_TID_CPUSET:
1721                 case PROC_TGID_CPUSET:
1722                         inode->i_fop = &proc_cpuset_operations;
1723                         break;
1724 #endif
1725                 case PROC_TID_OOM_SCORE:
1726                 case PROC_TGID_OOM_SCORE:
1727                         inode->i_fop = &proc_info_file_operations;
1728                         ei->op.proc_read = proc_oom_score;
1729                         break;
1730                 case PROC_TID_OOM_ADJUST:
1731                 case PROC_TGID_OOM_ADJUST:
1732                         inode->i_fop = &proc_oom_adjust_operations;
1733                         break;
1734 #ifdef CONFIG_AUDITSYSCALL
1735                 case PROC_TID_LOGINUID:
1736                 case PROC_TGID_LOGINUID:
1737                         inode->i_fop = &proc_loginuid_operations;
1738                         break;
1739 #endif
1740                 default:
1741                         printk("procfs: impossible type (%d)",p->type);
1742                         iput(inode);
1743                         error = ERR_PTR(-EINVAL);
1744                         goto out;
1745         }
1746         dentry->d_op = &pid_dentry_operations;
1747         d_add(dentry, inode);
1748         /* Close the race of the process dying before we return the dentry */
1749         if (pid_revalidate(dentry, NULL))
1750                 error = NULL;
1751 out:
1752         return error;
1753 }
1754
1755 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1756         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1757 }
1758
1759 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1760         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1761 }
1762
1763 static struct file_operations proc_tgid_base_operations = {
1764         .read           = generic_read_dir,
1765         .readdir        = proc_tgid_base_readdir,
1766 };
1767
1768 static struct file_operations proc_tid_base_operations = {
1769         .read           = generic_read_dir,
1770         .readdir        = proc_tid_base_readdir,
1771 };
1772
1773 static struct inode_operations proc_tgid_base_inode_operations = {
1774         .lookup         = proc_tgid_base_lookup,
1775 };
1776
1777 static struct inode_operations proc_tid_base_inode_operations = {
1778         .lookup         = proc_tid_base_lookup,
1779 };
1780
1781 #ifdef CONFIG_SECURITY
1782 static int proc_tgid_attr_readdir(struct file * filp,
1783                              void * dirent, filldir_t filldir)
1784 {
1785         return proc_pident_readdir(filp,dirent,filldir,
1786                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1787 }
1788
1789 static int proc_tid_attr_readdir(struct file * filp,
1790                              void * dirent, filldir_t filldir)
1791 {
1792         return proc_pident_readdir(filp,dirent,filldir,
1793                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1794 }
1795
1796 static struct file_operations proc_tgid_attr_operations = {
1797         .read           = generic_read_dir,
1798         .readdir        = proc_tgid_attr_readdir,
1799 };
1800
1801 static struct file_operations proc_tid_attr_operations = {
1802         .read           = generic_read_dir,
1803         .readdir        = proc_tid_attr_readdir,
1804 };
1805
1806 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1807                                 struct dentry *dentry, struct nameidata *nd)
1808 {
1809         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1810 }
1811
1812 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1813                                 struct dentry *dentry, struct nameidata *nd)
1814 {
1815         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1816 }
1817
1818 static struct inode_operations proc_tgid_attr_inode_operations = {
1819         .lookup         = proc_tgid_attr_lookup,
1820 };
1821
1822 static struct inode_operations proc_tid_attr_inode_operations = {
1823         .lookup         = proc_tid_attr_lookup,
1824 };
1825 #endif
1826
1827 /*
1828  * /proc/self:
1829  */
1830 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1831                               int buflen)
1832 {
1833         char tmp[30];
1834         sprintf(tmp, "%d", current->tgid);
1835         return vfs_readlink(dentry,buffer,buflen,tmp);
1836 }
1837
1838 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1839 {
1840         char tmp[30];
1841         sprintf(tmp, "%d", current->tgid);
1842         return ERR_PTR(vfs_follow_link(nd,tmp));
1843 }
1844
1845 static struct inode_operations proc_self_inode_operations = {
1846         .readlink       = proc_self_readlink,
1847         .follow_link    = proc_self_follow_link,
1848 };
1849
1850 /**
1851  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
1852  *
1853  * @task: task that should be flushed.
1854  *
1855  * Looks in the dcache for
1856  * /proc/@pid
1857  * /proc/@tgid/task/@pid
1858  * if either directory is present flushes it and all of it'ts children
1859  * from the dcache.
1860  *
1861  * It is safe and reasonable to cache /proc entries for a task until
1862  * that task exits.  After that they just clog up the dcache with
1863  * useless entries, possibly causing useful dcache entries to be
1864  * flushed instead.  This routine is proved to flush those useless
1865  * dcache entries at process exit time.
1866  *
1867  * NOTE: This routine is just an optimization so it does not guarantee
1868  *       that no dcache entries will exist at process exit time it
1869  *       just makes it very unlikely that any will persist.
1870  */
1871 void proc_flush_task(struct task_struct *task)
1872 {
1873         struct dentry *dentry, *leader, *dir;
1874         char buf[30];
1875         struct qstr name;
1876
1877         name.name = buf;
1878         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1879         dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1880         if (dentry) {
1881                 shrink_dcache_parent(dentry);
1882                 d_drop(dentry);
1883                 dput(dentry);
1884         }
1885
1886         if (thread_group_leader(task))
1887                 goto out;
1888
1889         name.name = buf;
1890         name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
1891         leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1892         if (!leader)
1893                 goto out;
1894
1895         name.name = "task";
1896         name.len = strlen(name.name);
1897         dir = d_hash_and_lookup(leader, &name);
1898         if (!dir)
1899                 goto out_put_leader;
1900
1901         name.name = buf;
1902         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1903         dentry = d_hash_and_lookup(dir, &name);
1904         if (dentry) {
1905                 shrink_dcache_parent(dentry);
1906                 d_drop(dentry);
1907                 dput(dentry);
1908         }
1909
1910         dput(dir);
1911 out_put_leader:
1912         dput(leader);
1913 out:
1914         return;
1915 }
1916
1917 /* SMP-safe */
1918 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1919 {
1920         struct dentry *result = ERR_PTR(-ENOENT);
1921         struct task_struct *task;
1922         struct inode *inode;
1923         struct proc_inode *ei;
1924         unsigned tgid;
1925
1926         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
1927                 inode = new_inode(dir->i_sb);
1928                 if (!inode)
1929                         return ERR_PTR(-ENOMEM);
1930                 ei = PROC_I(inode);
1931                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1932                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
1933                 ei->pde = NULL;
1934                 inode->i_mode = S_IFLNK|S_IRWXUGO;
1935                 inode->i_uid = inode->i_gid = 0;
1936                 inode->i_size = 64;
1937                 inode->i_op = &proc_self_inode_operations;
1938                 d_add(dentry, inode);
1939                 return NULL;
1940         }
1941         tgid = name_to_int(dentry);
1942         if (tgid == ~0U)
1943                 goto out;
1944
1945         read_lock(&tasklist_lock);
1946         task = find_task_by_pid(tgid);
1947         if (task)
1948                 get_task_struct(task);
1949         read_unlock(&tasklist_lock);
1950         if (!task)
1951                 goto out;
1952
1953         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
1954         if (!inode)
1955                 goto out_put_task;
1956
1957         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1958         inode->i_op = &proc_tgid_base_inode_operations;
1959         inode->i_fop = &proc_tgid_base_operations;
1960         inode->i_flags|=S_IMMUTABLE;
1961 #ifdef CONFIG_SECURITY
1962         inode->i_nlink = 5;
1963 #else
1964         inode->i_nlink = 4;
1965 #endif
1966
1967         dentry->d_op = &pid_dentry_operations;
1968
1969         d_add(dentry, inode);
1970         /* Close the race of the process dying before we return the dentry */
1971         if (pid_revalidate(dentry, NULL))
1972                 result = NULL;
1973
1974 out_put_task:
1975         put_task_struct(task);
1976 out:
1977         return result;
1978 }
1979
1980 /* SMP-safe */
1981 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1982 {
1983         struct dentry *result = ERR_PTR(-ENOENT);
1984         struct task_struct *task;
1985         struct task_struct *leader = proc_task(dir);
1986         struct inode *inode;
1987         unsigned tid;
1988
1989         tid = name_to_int(dentry);
1990         if (tid == ~0U)
1991                 goto out;
1992
1993         read_lock(&tasklist_lock);
1994         task = find_task_by_pid(tid);
1995         if (task)
1996                 get_task_struct(task);
1997         read_unlock(&tasklist_lock);
1998         if (!task)
1999                 goto out;
2000         if (leader->tgid != task->tgid)
2001                 goto out_drop_task;
2002
2003         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2004
2005
2006         if (!inode)
2007                 goto out_drop_task;
2008         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2009         inode->i_op = &proc_tid_base_inode_operations;
2010         inode->i_fop = &proc_tid_base_operations;
2011         inode->i_flags|=S_IMMUTABLE;
2012 #ifdef CONFIG_SECURITY
2013         inode->i_nlink = 4;
2014 #else
2015         inode->i_nlink = 3;
2016 #endif
2017
2018         dentry->d_op = &pid_dentry_operations;
2019
2020         d_add(dentry, inode);
2021         /* Close the race of the process dying before we return the dentry */
2022         if (pid_revalidate(dentry, NULL))
2023                 result = NULL;
2024
2025 out_drop_task:
2026         put_task_struct(task);
2027 out:
2028         return result;
2029 }
2030
2031 #define PROC_NUMBUF 10
2032 #define PROC_MAXPIDS 20
2033
2034 /*
2035  * Get a few tgid's to return for filldir - we need to hold the
2036  * tasklist lock while doing this, and we must release it before
2037  * we actually do the filldir itself, so we use a temp buffer..
2038  */
2039 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2040 {
2041         struct task_struct *p;
2042         int nr_tgids = 0;
2043
2044         index--;
2045         read_lock(&tasklist_lock);
2046         p = NULL;
2047         if (version) {
2048                 p = find_task_by_pid(version);
2049                 if (p && !thread_group_leader(p))
2050                         p = NULL;
2051         }
2052
2053         if (p)
2054                 index = 0;
2055         else
2056                 p = next_task(&init_task);
2057
2058         for ( ; p != &init_task; p = next_task(p)) {
2059                 int tgid = p->pid;
2060                 if (!pid_alive(p))
2061                         continue;
2062                 if (--index >= 0)
2063                         continue;
2064                 tgids[nr_tgids] = tgid;
2065                 nr_tgids++;
2066                 if (nr_tgids >= PROC_MAXPIDS)
2067                         break;
2068         }
2069         read_unlock(&tasklist_lock);
2070         return nr_tgids;
2071 }
2072
2073 /*
2074  * Get a few tid's to return for filldir - we need to hold the
2075  * tasklist lock while doing this, and we must release it before
2076  * we actually do the filldir itself, so we use a temp buffer..
2077  */
2078 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2079 {
2080         struct task_struct *leader_task = proc_task(dir);
2081         struct task_struct *task = leader_task;
2082         int nr_tids = 0;
2083
2084         index -= 2;
2085         read_lock(&tasklist_lock);
2086         /*
2087          * The starting point task (leader_task) might be an already
2088          * unlinked task, which cannot be used to access the task-list
2089          * via next_thread().
2090          */
2091         if (pid_alive(task)) do {
2092                 int tid = task->pid;
2093
2094                 if (--index >= 0)
2095                         continue;
2096                 if (tids != NULL)
2097                         tids[nr_tids] = tid;
2098                 nr_tids++;
2099                 if (nr_tids >= PROC_MAXPIDS)
2100                         break;
2101         } while ((task = next_thread(task)) != leader_task);
2102         read_unlock(&tasklist_lock);
2103         return nr_tids;
2104 }
2105
2106 /* for the /proc/ directory itself, after non-process stuff has been done */
2107 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2108 {
2109         unsigned int tgid_array[PROC_MAXPIDS];
2110         char buf[PROC_NUMBUF];
2111         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2112         unsigned int nr_tgids, i;
2113         int next_tgid;
2114
2115         if (!nr) {
2116                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2117                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2118                         return 0;
2119                 filp->f_pos++;
2120                 nr++;
2121         }
2122
2123         /* f_version caches the tgid value that the last readdir call couldn't
2124          * return. lseek aka telldir automagically resets f_version to 0.
2125          */
2126         next_tgid = filp->f_version;
2127         filp->f_version = 0;
2128         for (;;) {
2129                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2130                 if (!nr_tgids) {
2131                         /* no more entries ! */
2132                         break;
2133                 }
2134                 next_tgid = 0;
2135
2136                 /* do not use the last found pid, reserve it for next_tgid */
2137                 if (nr_tgids == PROC_MAXPIDS) {
2138                         nr_tgids--;
2139                         next_tgid = tgid_array[nr_tgids];
2140                 }
2141
2142                 for (i=0;i<nr_tgids;i++) {
2143                         int tgid = tgid_array[i];
2144                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2145                         unsigned long j = PROC_NUMBUF;
2146
2147                         do
2148                                 buf[--j] = '0' + (tgid % 10);
2149                         while ((tgid /= 10) != 0);
2150
2151                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2152                                 /* returning this tgid failed, save it as the first
2153                                  * pid for the next readir call */
2154                                 filp->f_version = tgid_array[i];
2155                                 goto out;
2156                         }
2157                         filp->f_pos++;
2158                         nr++;
2159                 }
2160         }
2161 out:
2162         return 0;
2163 }
2164
2165 /* for the /proc/TGID/task/ directories */
2166 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2167 {
2168         unsigned int tid_array[PROC_MAXPIDS];
2169         char buf[PROC_NUMBUF];
2170         unsigned int nr_tids, i;
2171         struct dentry *dentry = filp->f_dentry;
2172         struct inode *inode = dentry->d_inode;
2173         int retval = -ENOENT;
2174         ino_t ino;
2175         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2176
2177         if (!pid_alive(proc_task(inode)))
2178                 goto out;
2179         retval = 0;
2180
2181         switch (pos) {
2182         case 0:
2183                 ino = inode->i_ino;
2184                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2185                         goto out;
2186                 pos++;
2187                 /* fall through */
2188         case 1:
2189                 ino = parent_ino(dentry);
2190                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2191                         goto out;
2192                 pos++;
2193                 /* fall through */
2194         }
2195
2196         nr_tids = get_tid_list(pos, tid_array, inode);
2197
2198         for (i = 0; i < nr_tids; i++) {
2199                 unsigned long j = PROC_NUMBUF;
2200                 int tid = tid_array[i];
2201
2202                 ino = fake_ino(tid,PROC_TID_INO);
2203
2204                 do
2205                         buf[--j] = '0' + (tid % 10);
2206                 while ((tid /= 10) != 0);
2207
2208                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2209                         break;
2210                 pos++;
2211         }
2212 out:
2213         filp->f_pos = pos;
2214         return retval;
2215 }
2216
2217 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2218 {
2219         struct inode *inode = dentry->d_inode;
2220         struct task_struct *p = proc_task(inode);
2221         generic_fillattr(inode, stat);
2222
2223         if (pid_alive(p)) {
2224                 task_lock(p);
2225                 if (p->signal)
2226                         stat->nlink += atomic_read(&p->signal->count);
2227                 task_unlock(p);
2228         }
2229
2230         return 0;
2231 }