fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/config.h>
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/init.h>
  58 #include <linux/file.h>
  59 #include <linux/string.h>
  60 #include <linux/seq_file.h>
  61 #include <linux/namei.h>
  62 #include <linux/namespace.h>
  63 #include <linux/mm.h>
  64 #include <linux/smp_lock.h>
  65 #include <linux/kallsyms.h>
  66 #include <linux/mount.h>
  67 #include <linux/security.h>
  68 #include <linux/ptrace.h>
  69 #include <linux/seccomp.h>
  70 #include <linux/cpuset.h>
  71 #include <linux/audit.h>
  72 #include "internal.h"
  73
  74 /*
  75  * For hysterical raisins we keep the same inumbers as in the old procfs.
  76  * Feel free to change the macro below - just keep the range distinct from
  77  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
  78  * As soon as we'll get a separate superblock we will be able to forget
  79  * about magical ranges too.
  80  */
  81
  82 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
  83
  84 enum pid_directory_inos {
  85         PROC_TGID_INO = 2,
  86         PROC_TGID_TASK,
  87         PROC_TGID_STATUS,
  88         PROC_TGID_MEM,
  89 #ifdef CONFIG_SECCOMP
  90         PROC_TGID_SECCOMP,
  91 #endif
  92         PROC_TGID_CWD,
  93         PROC_TGID_ROOT,
  94         PROC_TGID_EXE,
  95         PROC_TGID_FD,
  96         PROC_TGID_ENVIRON,
  97         PROC_TGID_AUXV,
  98         PROC_TGID_CMDLINE,
  99         PROC_TGID_STAT,
 100         PROC_TGID_STATM,
 101         PROC_TGID_MAPS,
 102         PROC_TGID_NUMA_MAPS,
 103         PROC_TGID_MOUNTS,
 104         PROC_TGID_WCHAN,
 105         PROC_TGID_SMAPS,
 106 #ifdef CONFIG_SCHEDSTATS
 107         PROC_TGID_SCHEDSTAT,
 108 #endif
 109 #ifdef CONFIG_CPUSETS
 110         PROC_TGID_CPUSET,
 111 #endif
 112 #ifdef CONFIG_SECURITY
 113         PROC_TGID_ATTR,
 114         PROC_TGID_ATTR_CURRENT,
 115         PROC_TGID_ATTR_PREV,
 116         PROC_TGID_ATTR_EXEC,
 117         PROC_TGID_ATTR_FSCREATE,
 118 #endif
 119 #ifdef CONFIG_AUDITSYSCALL
 120         PROC_TGID_LOGINUID,
 121 #endif
 122         PROC_TGID_OOM_SCORE,
 123         PROC_TGID_OOM_ADJUST,
 124         PROC_TID_INO,
 125         PROC_TID_STATUS,
 126         PROC_TID_MEM,
 127 #ifdef CONFIG_SECCOMP
 128         PROC_TID_SECCOMP,
 129 #endif
 130         PROC_TID_CWD,
 131         PROC_TID_ROOT,
 132         PROC_TID_EXE,
 133         PROC_TID_FD,
 134         PROC_TID_ENVIRON,
 135         PROC_TID_AUXV,
 136         PROC_TID_CMDLINE,
 137         PROC_TID_STAT,
 138         PROC_TID_STATM,
 139         PROC_TID_MAPS,
 140         PROC_TID_NUMA_MAPS,
 141         PROC_TID_MOUNTS,
 142         PROC_TID_WCHAN,
 143         PROC_TID_SMAPS,
 144 #ifdef CONFIG_SCHEDSTATS
 145         PROC_TID_SCHEDSTAT,
 146 #endif
 147 #ifdef CONFIG_CPUSETS
 148         PROC_TID_CPUSET,
 149 #endif
 150 #ifdef CONFIG_SECURITY
 151         PROC_TID_ATTR,
 152         PROC_TID_ATTR_CURRENT,
 153         PROC_TID_ATTR_PREV,
 154         PROC_TID_ATTR_EXEC,
 155         PROC_TID_ATTR_FSCREATE,
 156 #endif
 157 #ifdef CONFIG_AUDITSYSCALL
 158         PROC_TID_LOGINUID,
 159 #endif
 160         PROC_TID_OOM_SCORE,
 161         PROC_TID_OOM_ADJUST,
 162
 163         /* Add new entries before this */
 164         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
 165 };
 166
 167 struct pid_entry {
 168         int type;
 169         int len;
 170         char *name;
 171         mode_t mode;
 172 };
 173
 174 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
 175
 176 static struct pid_entry tgid_base_stuff[] = {
 177         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
 178         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 179         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
 180         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
 181         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
 182         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 183         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 184         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
 185         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 186 #ifdef CONFIG_NUMA
 187         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
 188 #endif
 189         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 190 #ifdef CONFIG_SECCOMP
 191         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 192 #endif
 193         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
 194         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
 195         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
 196         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
 197         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
 198 #ifdef CONFIG_SECURITY
 199         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 200 #endif
 201 #ifdef CONFIG_KALLSYMS
 202         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
 203 #endif
 204 #ifdef CONFIG_SCHEDSTATS
 205         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
 206 #endif
 207 #ifdef CONFIG_CPUSETS
 208         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
 209 #endif
 210         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
 211         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 212 #ifdef CONFIG_AUDITSYSCALL
 213         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 214 #endif
 215         {0,0,NULL,0}
 216 };
 217 static struct pid_entry tid_base_stuff[] = {
 218         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 219         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
 220         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
 221         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
 222         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 223         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 224         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
 225         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 226 #ifdef CONFIG_NUMA
 227         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
 228 #endif
 229         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 230 #ifdef CONFIG_SECCOMP
 231         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 232 #endif
 233         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
 234         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
 235         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
 236         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
 237         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
 238 #ifdef CONFIG_SECURITY
 239         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 240 #endif
 241 #ifdef CONFIG_KALLSYMS
 242         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
 243 #endif
 244 #ifdef CONFIG_SCHEDSTATS
 245         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
 246 #endif
 247 #ifdef CONFIG_CPUSETS
 248         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
 249 #endif
 250         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
 251         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 252 #ifdef CONFIG_AUDITSYSCALL
 253         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 254 #endif
 255         {0,0,NULL,0}
 256 };
 257
 258 #ifdef CONFIG_SECURITY
 259 static struct pid_entry tgid_attr_stuff[] = {
 260         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
 261         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
 262         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 263         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 264         {0,0,NULL,0}
 265 };
 266 static struct pid_entry tid_attr_stuff[] = {
 267         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
 268         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
 269         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 270         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 271         {0,0,NULL,0}
 272 };
 273 #endif
 274
 275 #undef E
 276
 277 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 278 {
 279         struct task_struct *task = proc_task(inode);
 280         struct files_struct *files;
 281         struct file *file;
 282         int fd = proc_type(inode) - PROC_TID_FD_DIR;
 283
 284         files = get_files_struct(task);
 285         if (files) {
 286                 spin_lock(&files->file_lock);
 287                 file = fcheck_files(files, fd);
 288                 if (file) {
 289                         *mnt = mntget(file->f_vfsmnt);
 290                         *dentry = dget(file->f_dentry);
 291                         spin_unlock(&files->file_lock);
 292                         put_files_struct(files);
 293                         return 0;
 294                 }
 295                 spin_unlock(&files->file_lock);
 296                 put_files_struct(files);
 297         }
 298         return -ENOENT;
 299 }
 300
 301 static struct fs_struct *get_fs_struct(struct task_struct *task)
 302 {
 303         struct fs_struct *fs;
 304         task_lock(task);
 305         fs = task->fs;
 306         if(fs)
 307                 atomic_inc(&fs->count);
 308         task_unlock(task);
 309         return fs;
 310 }
 311
 312 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 313 {
 314         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 315         int result = -ENOENT;
 316         if (fs) {
 317                 read_lock(&fs->lock);
 318                 *mnt = mntget(fs->pwdmnt);
 319                 *dentry = dget(fs->pwd);
 320                 read_unlock(&fs->lock);
 321                 result = 0;
 322                 put_fs_struct(fs);
 323         }
 324         return result;
 325 }
 326
 327 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 328 {
 329         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 330         int result = -ENOENT;
 331         if (fs) {
 332                 read_lock(&fs->lock);
 333                 *mnt = mntget(fs->rootmnt);
 334                 *dentry = dget(fs->root);
 335                 read_unlock(&fs->lock);
 336                 result = 0;
 337                 put_fs_struct(fs);
 338         }
 339         return result;
 340 }
 341
 342 #define MAY_PTRACE(task) \
 343         (task == current || \
 344         (task->parent == current && \
 345         (task->ptrace & PT_PTRACED) && \
 346          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
 347          security_ptrace(current,task) == 0))
 348
 349 static int proc_pid_environ(struct task_struct *task, char * buffer)
 350 {
 351         int res = 0;
 352         struct mm_struct *mm = get_task_mm(task);
 353         if (mm) {
 354                 unsigned int len = mm->env_end - mm->env_start;
 355                 if (len > PAGE_SIZE)
 356                         len = PAGE_SIZE;
 357                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
 358                 if (!ptrace_may_attach(task))
 359                         res = -ESRCH;
 360                 mmput(mm);
 361         }
 362         return res;
 363 }
 364
 365 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 366 {
 367         int res = 0;
 368         unsigned int len;
 369         struct mm_struct *mm = get_task_mm(task);
 370         if (!mm)
 371                 goto out;
 372         if (!mm->arg_end)
 373                 goto out_mm;    /* Shh! No looking before we're done */
 374
 375         len = mm->arg_end - mm->arg_start;
 376
 377         if (len > PAGE_SIZE)
 378                 len = PAGE_SIZE;
 379
 380         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 381
 382         // If the nul at the end of args has been overwritten, then
 383         // assume application is using setproctitle(3).
 384         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 385                 len = strnlen(buffer, res);
 386                 if (len < res) {
 387                     res = len;
 388                 } else {
 389                         len = mm->env_end - mm->env_start;
 390                         if (len > PAGE_SIZE - res)
 391                                 len = PAGE_SIZE - res;
 392                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 393                         res = strnlen(buffer, res);
 394                 }
 395         }
 396 out_mm:
 397         mmput(mm);
 398 out:
 399         return res;
 400 }
 401
 402 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 403 {
 404         int res = 0;
 405         struct mm_struct *mm = get_task_mm(task);
 406         if (mm) {
 407                 unsigned int nwords = 0;
 408                 do
 409                         nwords += 2;
 410                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 411                 res = nwords * sizeof(mm->saved_auxv[0]);
 412                 if (res > PAGE_SIZE)
 413                         res = PAGE_SIZE;
 414                 memcpy(buffer, mm->saved_auxv, res);
 415                 mmput(mm);
 416         }
 417         return res;
 418 }
 419
 420
 421 #ifdef CONFIG_KALLSYMS
 422 /*
 423  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 424  * Returns the resolved symbol.  If that fails, simply return the address.
 425  */
 426 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 427 {
 428         char *modname;
 429         const char *sym_name;
 430         unsigned long wchan, size, offset;
 431         char namebuf[KSYM_NAME_LEN+1];
 432
 433         wchan = get_wchan(task);
 434
 435         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
 436         if (sym_name)
 437                 return sprintf(buffer, "%s", sym_name);
 438         return sprintf(buffer, "%lu", wchan);
 439 }
 440 #endif /* CONFIG_KALLSYMS */
 441
 442 #ifdef CONFIG_SCHEDSTATS
 443 /*
 444  * Provides /proc/PID/schedstat
 445  */
 446 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 447 {
 448         return sprintf(buffer, "%lu %lu %lu\n",
 449                         task->sched_info.cpu_time,
 450                         task->sched_info.run_delay,
 451                         task->sched_info.pcnt);
 452 }
 453 #endif
 454
 455 /* The badness from the OOM killer */
 456 unsigned long badness(struct task_struct *p, unsigned long uptime);
 457 static int proc_oom_score(struct task_struct *task, char *buffer)
 458 {
 459         unsigned long points;
 460         struct timespec uptime;
 461
 462         do_posix_clock_monotonic_gettime(&uptime);
 463         points = badness(task, uptime.tv_sec);
 464         return sprintf(buffer, "%lu\n", points);
 465 }
 466
 467 /************************************************************************/
 468 /*                       Here the fs part begins                        */
 469 /************************************************************************/
 470
 471 /* permission checks */
 472
 473 static int proc_check_root(struct inode *inode)
 474 {
 475         struct dentry *de, *base, *root;
 476         struct vfsmount *our_vfsmnt, *vfsmnt, *mnt;
 477         int res = 0;
 478
 479         if (proc_root_link(inode, &root, &vfsmnt)) /* Ewww... */
 480                 return -ENOENT;
 481         read_lock(&current->fs->lock);
 482         our_vfsmnt = mntget(current->fs->rootmnt);
 483         base = dget(current->fs->root);
 484         read_unlock(&current->fs->lock);
 485
 486         spin_lock(&vfsmount_lock);
 487         de = root;
 488         mnt = vfsmnt;
 489
 490         while (vfsmnt != our_vfsmnt) {
 491                 if (vfsmnt == vfsmnt->mnt_parent)
 492                         goto out;
 493                 de = vfsmnt->mnt_mountpoint;
 494                 vfsmnt = vfsmnt->mnt_parent;
 495         }
 496
 497         if (!is_subdir(de, base))
 498                 goto out;
 499         spin_unlock(&vfsmount_lock);
 500
 501 exit:
 502         dput(base);
 503         mntput(our_vfsmnt);
 504         dput(root);
 505         mntput(mnt);
 506         return res;
 507 out:
 508         spin_unlock(&vfsmount_lock);
 509         res = -EACCES;
 510         goto exit;
 511 }
 512
 513 static int proc_permission(struct inode *inode, int mask, struct nameidata *nd)
 514 {
 515         if (generic_permission(inode, mask, NULL) != 0)
 516                 return -EACCES;
 517         return proc_check_root(inode);
 518 }
 519
 520 extern struct seq_operations proc_pid_maps_op;
 521 static int maps_open(struct inode *inode, struct file *file)
 522 {
 523         struct task_struct *task = proc_task(inode);
 524         int ret = seq_open(file, &proc_pid_maps_op);
 525         if (!ret) {
 526                 struct seq_file *m = file->private_data;
 527                 m->private = task;
 528         }
 529         return ret;
 530 }
 531
 532 static struct file_operations proc_maps_operations = {
 533         .open           = maps_open,
 534         .read           = seq_read,
 535         .llseek         = seq_lseek,
 536         .release        = seq_release,
 537 };
 538
 539 #ifdef CONFIG_NUMA
 540 extern struct seq_operations proc_pid_numa_maps_op;
 541 static int numa_maps_open(struct inode *inode, struct file *file)
 542 {
 543         struct task_struct *task = proc_task(inode);
 544         int ret = seq_open(file, &proc_pid_numa_maps_op);
 545         if (!ret) {
 546                 struct seq_file *m = file->private_data;
 547                 m->private = task;
 548         }
 549         return ret;
 550 }
 551
 552 static struct file_operations proc_numa_maps_operations = {
 553         .open           = numa_maps_open,
 554         .read           = seq_read,
 555         .llseek         = seq_lseek,
 556         .release        = seq_release,
 557 };
 558 #endif
 559
 560 extern struct seq_operations proc_pid_smaps_op;
 561 static int smaps_open(struct inode *inode, struct file *file)
 562 {
 563         struct task_struct *task = proc_task(inode);
 564         int ret = seq_open(file, &proc_pid_smaps_op);
 565         if (!ret) {
 566                 struct seq_file *m = file->private_data;
 567                 m->private = task;
 568         }
 569         return ret;
 570 }
 571
 572 static struct file_operations proc_smaps_operations = {
 573         .open           = smaps_open,
 574         .read           = seq_read,
 575         .llseek         = seq_lseek,
 576         .release        = seq_release,
 577 };
 578
 579 extern struct seq_operations mounts_op;
 580 static int mounts_open(struct inode *inode, struct file *file)
 581 {
 582         struct task_struct *task = proc_task(inode);
 583         int ret = seq_open(file, &mounts_op);
 584
 585         if (!ret) {
 586                 struct seq_file *m = file->private_data;
 587                 struct namespace *namespace;
 588                 task_lock(task);
 589                 namespace = task->namespace;
 590                 if (namespace)
 591                         get_namespace(namespace);
 592                 task_unlock(task);
 593
 594                 if (namespace)
 595                         m->private = namespace;
 596                 else {
 597                         seq_release(inode, file);
 598                         ret = -EINVAL;
 599                 }
 600         }
 601         return ret;
 602 }
 603
 604 static int mounts_release(struct inode *inode, struct file *file)
 605 {
 606         struct seq_file *m = file->private_data;
 607         struct namespace *namespace = m->private;
 608         put_namespace(namespace);
 609         return seq_release(inode, file);
 610 }
 611
 612 static struct file_operations proc_mounts_operations = {
 613         .open           = mounts_open,
 614         .read           = seq_read,
 615         .llseek         = seq_lseek,
 616         .release        = mounts_release,
 617 };
 618
 619 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 620
 621 static ssize_t proc_info_read(struct file * file, char __user * buf,
 622                           size_t count, loff_t *ppos)
 623 {
 624         struct inode * inode = file->f_dentry->d_inode;
 625         unsigned long page;
 626         ssize_t length;
 627         struct task_struct *task = proc_task(inode);
 628
 629         if (count > PROC_BLOCK_SIZE)
 630                 count = PROC_BLOCK_SIZE;
 631         if (!(page = __get_free_page(GFP_KERNEL)))
 632                 return -ENOMEM;
 633
 634         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 635
 636         if (length >= 0)
 637                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 638         free_page(page);
 639         return length;
 640 }
 641
 642 static struct file_operations proc_info_file_operations = {
 643         .read           = proc_info_read,
 644 };
 645
 646 static int mem_open(struct inode* inode, struct file* file)
 647 {
 648         file->private_data = (void*)((long)current->self_exec_id);
 649         return 0;
 650 }
 651
 652 static ssize_t mem_read(struct file * file, char __user * buf,
 653                         size_t count, loff_t *ppos)
 654 {
 655         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 656         char *page;
 657         unsigned long src = *ppos;
 658         int ret = -ESRCH;
 659         struct mm_struct *mm;
 660
 661         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 662                 goto out;
 663
 664         ret = -ENOMEM;
 665         page = (char *)__get_free_page(GFP_USER);
 666         if (!page)
 667                 goto out;
 668
 669         ret = 0;
 670
 671         mm = get_task_mm(task);
 672         if (!mm)
 673                 goto out_free;
 674
 675         ret = -EIO;
 676
 677         if (file->private_data != (void*)((long)current->self_exec_id))
 678                 goto out_put;
 679
 680         ret = 0;
 681
 682         while (count > 0) {
 683                 int this_len, retval;
 684
 685                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 686                 retval = access_process_vm(task, src, page, this_len, 0);
 687                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 688                         if (!ret)
 689                                 ret = -EIO;
 690                         break;
 691                 }
 692
 693                 if (copy_to_user(buf, page, retval)) {
 694                         ret = -EFAULT;
 695                         break;
 696                 }
 697
 698                 ret += retval;
 699                 src += retval;
 700                 buf += retval;
 701                 count -= retval;
 702         }
 703         *ppos = src;
 704
 705 out_put:
 706         mmput(mm);
 707 out_free:
 708         free_page((unsigned long) page);
 709 out:
 710         return ret;
 711 }
 712
 713 #define mem_write NULL
 714
 715 #ifndef mem_write
 716 /* This is a security hazard */
 717 static ssize_t mem_write(struct file * file, const char * buf,
 718                          size_t count, loff_t *ppos)
 719 {
 720         int copied = 0;
 721         char *page;
 722         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 723         unsigned long dst = *ppos;
 724
 725         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 726                 return -ESRCH;
 727
 728         page = (char *)__get_free_page(GFP_USER);
 729         if (!page)
 730                 return -ENOMEM;
 731
 732         while (count > 0) {
 733                 int this_len, retval;
 734
 735                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 736                 if (copy_from_user(page, buf, this_len)) {
 737                         copied = -EFAULT;
 738                         break;
 739                 }
 740                 retval = access_process_vm(task, dst, page, this_len, 1);
 741                 if (!retval) {
 742                         if (!copied)
 743                                 copied = -EIO;
 744                         break;
 745                 }
 746                 copied += retval;
 747                 buf += retval;
 748                 dst += retval;
 749                 count -= retval;
 750         }
 751         *ppos = dst;
 752         free_page((unsigned long) page);
 753         return copied;
 754 }
 755 #endif
 756
 757 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 758 {
 759         switch (orig) {
 760         case 0:
 761                 file->f_pos = offset;
 762                 break;
 763         case 1:
 764                 file->f_pos += offset;
 765                 break;
 766         default:
 767                 return -EINVAL;
 768         }
 769         force_successful_syscall_return();
 770         return file->f_pos;
 771 }
 772
 773 static struct file_operations proc_mem_operations = {
 774         .llseek         = mem_lseek,
 775         .read           = mem_read,
 776         .write          = mem_write,
 777         .open           = mem_open,
 778 };
 779
 780 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 781                                 size_t count, loff_t *ppos)
 782 {
 783         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 784         char buffer[8];
 785         size_t len;
 786         int oom_adjust = task->oomkilladj;
 787         loff_t __ppos = *ppos;
 788
 789         len = sprintf(buffer, "%i\n", oom_adjust);
 790         if (__ppos >= len)
 791                 return 0;
 792         if (count > len-__ppos)
 793                 count = len-__ppos;
 794         if (copy_to_user(buf, buffer + __ppos, count))
 795                 return -EFAULT;
 796         *ppos = __ppos + count;
 797         return count;
 798 }
 799
 800 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 801                                 size_t count, loff_t *ppos)
 802 {
 803         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 804         char buffer[8], *end;
 805         int oom_adjust;
 806
 807         if (!capable(CAP_SYS_RESOURCE))
 808                 return -EPERM;
 809         memset(buffer, 0, 8);
 810         if (count > 6)
 811                 count = 6;
 812         if (copy_from_user(buffer, buf, count))
 813                 return -EFAULT;
 814         oom_adjust = simple_strtol(buffer, &end, 0);
 815         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 816                 return -EINVAL;
 817         if (*end == '\n')
 818                 end++;
 819         task->oomkilladj = oom_adjust;
 820         if (end - buffer == 0)
 821                 return -EIO;
 822         return end - buffer;
 823 }
 824
 825 static struct file_operations proc_oom_adjust_operations = {
 826         .read           = oom_adjust_read,
 827         .write          = oom_adjust_write,
 828 };
 829
 830 static struct inode_operations proc_mem_inode_operations = {
 831         .permission     = proc_permission,
 832 };
 833
 834 #ifdef CONFIG_AUDITSYSCALL
 835 #define TMPBUFLEN 21
 836 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 837                                   size_t count, loff_t *ppos)
 838 {
 839         struct inode * inode = file->f_dentry->d_inode;
 840         struct task_struct *task = proc_task(inode);
 841         ssize_t length;
 842         char tmpbuf[TMPBUFLEN];
 843
 844         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 845                                 audit_get_loginuid(task->audit_context));
 846         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 847 }
 848
 849 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 850                                    size_t count, loff_t *ppos)
 851 {
 852         struct inode * inode = file->f_dentry->d_inode;
 853         char *page, *tmp;
 854         ssize_t length;
 855         struct task_struct *task = proc_task(inode);
 856         uid_t loginuid;
 857
 858         if (!capable(CAP_AUDIT_CONTROL))
 859                 return -EPERM;
 860
 861         if (current != task)
 862                 return -EPERM;
 863
 864         if (count > PAGE_SIZE)
 865                 count = PAGE_SIZE;
 866
 867         if (*ppos != 0) {
 868                 /* No partial writes. */
 869                 return -EINVAL;
 870         }
 871         page = (char*)__get_free_page(GFP_USER);
 872         if (!page)
 873                 return -ENOMEM;
 874         length = -EFAULT;
 875         if (copy_from_user(page, buf, count))
 876                 goto out_free_page;
 877
 878         loginuid = simple_strtoul(page, &tmp, 10);
 879         if (tmp == page) {
 880                 length = -EINVAL;
 881                 goto out_free_page;
 882
 883         }
 884         length = audit_set_loginuid(task, loginuid);
 885         if (likely(length == 0))
 886                 length = count;
 887
 888 out_free_page:
 889         free_page((unsigned long) page);
 890         return length;
 891 }
 892
 893 static struct file_operations proc_loginuid_operations = {
 894         .read           = proc_loginuid_read,
 895         .write          = proc_loginuid_write,
 896 };
 897 #endif
 898
 899 #ifdef CONFIG_SECCOMP
 900 static ssize_t seccomp_read(struct file *file, char __user *buf,
 901                             size_t count, loff_t *ppos)
 902 {
 903         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
 904         char __buf[20];
 905         loff_t __ppos = *ppos;
 906         size_t len;
 907
 908         /* no need to print the trailing zero, so use only len */
 909         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
 910         if (__ppos >= len)
 911                 return 0;
 912         if (count > len - __ppos)
 913                 count = len - __ppos;
 914         if (copy_to_user(buf, __buf + __ppos, count))
 915                 return -EFAULT;
 916         *ppos = __ppos + count;
 917         return count;
 918 }
 919
 920 static ssize_t seccomp_write(struct file *file, const char __user *buf,
 921                              size_t count, loff_t *ppos)
 922 {
 923         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
 924         char __buf[20], *end;
 925         unsigned int seccomp_mode;
 926
 927         /* can set it only once to be even more secure */
 928         if (unlikely(tsk->seccomp.mode))
 929                 return -EPERM;
 930
 931         memset(__buf, 0, sizeof(__buf));
 932         count = min(count, sizeof(__buf) - 1);
 933         if (copy_from_user(__buf, buf, count))
 934                 return -EFAULT;
 935         seccomp_mode = simple_strtoul(__buf, &end, 0);
 936         if (*end == '\n')
 937                 end++;
 938         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
 939                 tsk->seccomp.mode = seccomp_mode;
 940                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
 941         } else
 942                 return -EINVAL;
 943         if (unlikely(!(end - __buf)))
 944                 return -EIO;
 945         return end - __buf;
 946 }
 947
 948 static struct file_operations proc_seccomp_operations = {
 949         .read           = seccomp_read,
 950         .write          = seccomp_write,
 951 };
 952 #endif /* CONFIG_SECCOMP */
 953
 954 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
 955 {
 956         struct inode *inode = dentry->d_inode;
 957         int error = -EACCES;
 958
 959         /* We don't need a base pointer in the /proc filesystem */
 960         path_release(nd);
 961
 962         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
 963                 goto out;
 964         error = proc_check_root(inode);
 965         if (error)
 966                 goto out;
 967
 968         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
 969         nd->last_type = LAST_BIND;
 970 out:
 971         return ERR_PTR(error);
 972 }
 973
 974 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
 975                             char __user *buffer, int buflen)
 976 {
 977         struct inode * inode;
 978         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
 979         int len;
 980
 981         if (!tmp)
 982                 return -ENOMEM;
 983
 984         inode = dentry->d_inode;
 985         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
 986         len = PTR_ERR(path);
 987         if (IS_ERR(path))
 988                 goto out;
 989         len = tmp + PAGE_SIZE - 1 - path;
 990
 991         if (len > buflen)
 992                 len = buflen;
 993         if (copy_to_user(buffer, path, len))
 994                 len = -EFAULT;
 995  out:
 996         free_page((unsigned long)tmp);
 997         return len;
 998 }
 999
1000 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1001 {
1002         int error = -EACCES;
1003         struct inode *inode = dentry->d_inode;
1004         struct dentry *de;
1005         struct vfsmount *mnt = NULL;
1006
1007         lock_kernel();
1008
1009         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1010                 goto out;
1011         error = proc_check_root(inode);
1012         if (error)
1013                 goto out;
1014
1015         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1016         if (error)
1017                 goto out;
1018
1019         error = do_proc_readlink(de, mnt, buffer, buflen);
1020         dput(de);
1021         mntput(mnt);
1022 out:
1023         unlock_kernel();
1024         return error;
1025 }
1026
1027 static struct inode_operations proc_pid_link_inode_operations = {
1028         .readlink       = proc_pid_readlink,
1029         .follow_link    = proc_pid_follow_link
1030 };
1031
1032 #define NUMBUF 10
1033
1034 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1035 {
1036         struct inode *inode = filp->f_dentry->d_inode;
1037         struct task_struct *p = proc_task(inode);
1038         unsigned int fd, tid, ino;
1039         int retval;
1040         char buf[NUMBUF];
1041         struct files_struct * files;
1042
1043         retval = -ENOENT;
1044         if (!pid_alive(p))
1045                 goto out;
1046         retval = 0;
1047         tid = p->pid;
1048
1049         fd = filp->f_pos;
1050         switch (fd) {
1051                 case 0:
1052                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1053                                 goto out;
1054                         filp->f_pos++;
1055                 case 1:
1056                         ino = fake_ino(tid, PROC_TID_INO);
1057                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1058                                 goto out;
1059                         filp->f_pos++;
1060                 default:
1061                         files = get_files_struct(p);
1062                         if (!files)
1063                                 goto out;
1064                         spin_lock(&files->file_lock);
1065                         for (fd = filp->f_pos-2;
1066                              fd < files->max_fds;
1067                              fd++, filp->f_pos++) {
1068                                 unsigned int i,j;
1069
1070                                 if (!fcheck_files(files, fd))
1071                                         continue;
1072                                 spin_unlock(&files->file_lock);
1073
1074                                 j = NUMBUF;
1075                                 i = fd;
1076                                 do {
1077                                         j--;
1078                                         buf[j] = '0' + (i % 10);
1079                                         i /= 10;
1080                                 } while (i);
1081
1082                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1083                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1084                                         spin_lock(&files->file_lock);
1085                                         break;
1086                                 }
1087                                 spin_lock(&files->file_lock);
1088                         }
1089                         spin_unlock(&files->file_lock);
1090                         put_files_struct(files);
1091         }
1092 out:
1093         return retval;
1094 }
1095
1096 static int proc_pident_readdir(struct file *filp,
1097                 void *dirent, filldir_t filldir,
1098                 struct pid_entry *ents, unsigned int nents)
1099 {
1100         int i;
1101         int pid;
1102         struct dentry *dentry = filp->f_dentry;
1103         struct inode *inode = dentry->d_inode;
1104         struct pid_entry *p;
1105         ino_t ino;
1106         int ret;
1107
1108         ret = -ENOENT;
1109         if (!pid_alive(proc_task(inode)))
1110                 goto out;
1111
1112         ret = 0;
1113         pid = proc_task(inode)->pid;
1114         i = filp->f_pos;
1115         switch (i) {
1116         case 0:
1117                 ino = inode->i_ino;
1118                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1119                         goto out;
1120                 i++;
1121                 filp->f_pos++;
1122                 /* fall through */
1123         case 1:
1124                 ino = parent_ino(dentry);
1125                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1126                         goto out;
1127                 i++;
1128                 filp->f_pos++;
1129                 /* fall through */
1130         default:
1131                 i -= 2;
1132                 if (i >= nents) {
1133                         ret = 1;
1134                         goto out;
1135                 }
1136                 p = ents + i;
1137                 while (p->name) {
1138                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1139                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1140                                 goto out;
1141                         filp->f_pos++;
1142                         p++;
1143                 }
1144         }
1145
1146         ret = 1;
1147 out:
1148         return ret;
1149 }
1150
1151 static int proc_tgid_base_readdir(struct file * filp,
1152                              void * dirent, filldir_t filldir)
1153 {
1154         return proc_pident_readdir(filp,dirent,filldir,
1155                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1156 }
1157
1158 static int proc_tid_base_readdir(struct file * filp,
1159                              void * dirent, filldir_t filldir)
1160 {
1161         return proc_pident_readdir(filp,dirent,filldir,
1162                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1163 }
1164
1165 /* building an inode */
1166
1167 static int task_dumpable(struct task_struct *task)
1168 {
1169         int dumpable = 0;
1170         struct mm_struct *mm;
1171
1172         task_lock(task);
1173         mm = task->mm;
1174         if (mm)
1175                 dumpable = mm->dumpable;
1176         task_unlock(task);
1177         if(dumpable == 1)
1178                 return 1;
1179         return 0;
1180 }
1181
1182
1183 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1184 {
1185         struct inode * inode;
1186         struct proc_inode *ei;
1187
1188         /* We need a new inode */
1189
1190         inode = new_inode(sb);
1191         if (!inode)
1192                 goto out;
1193
1194         /* Common stuff */
1195         ei = PROC_I(inode);
1196         ei->task = NULL;
1197         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1198         inode->i_ino = fake_ino(task->pid, ino);
1199
1200         if (!pid_alive(task))
1201                 goto out_unlock;
1202
1203         /*
1204          * grab the reference to task.
1205          */
1206         get_task_struct(task);
1207         ei->task = task;
1208         ei->type = ino;
1209         inode->i_uid = 0;
1210         inode->i_gid = 0;
1211         if (ino == PROC_TGID_INO || ino == PROC_TID_INO || task_dumpable(task)) {
1212                 inode->i_uid = task->euid;
1213                 inode->i_gid = task->egid;
1214         }
1215         security_task_to_inode(task, inode);
1216
1217 out:
1218         return inode;
1219
1220 out_unlock:
1221         ei->pde = NULL;
1222         iput(inode);
1223         return NULL;
1224 }
1225
1226 /* dentry stuff */
1227
1228 /*
1229  *      Exceptional case: normally we are not allowed to unhash a busy
1230  * directory. In this case, however, we can do it - no aliasing problems
1231  * due to the way we treat inodes.
1232  *
1233  * Rewrite the inode's ownerships here because the owning task may have
1234  * performed a setuid(), etc.
1235  */
1236 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1237 {
1238         struct inode *inode = dentry->d_inode;
1239         struct task_struct *task = proc_task(inode);
1240         if (pid_alive(task)) {
1241                 if (proc_type(inode) == PROC_TGID_INO || proc_type(inode) == PROC_TID_INO || task_dumpable(task)) {
1242                         inode->i_uid = task->euid;
1243                         inode->i_gid = task->egid;
1244                 } else {
1245                         inode->i_uid = 0;
1246                         inode->i_gid = 0;
1247                 }
1248                 security_task_to_inode(task, inode);
1249                 return 1;
1250         }
1251         d_drop(dentry);
1252         return 0;
1253 }
1254
1255 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1256 {
1257         struct inode *inode = dentry->d_inode;
1258         struct task_struct *task = proc_task(inode);
1259         int fd = proc_type(inode) - PROC_TID_FD_DIR;
1260         struct files_struct *files;
1261
1262         files = get_files_struct(task);
1263         if (files) {
1264                 spin_lock(&files->file_lock);
1265                 if (fcheck_files(files, fd)) {
1266                         spin_unlock(&files->file_lock);
1267                         put_files_struct(files);
1268                         if (task_dumpable(task)) {
1269                                 inode->i_uid = task->euid;
1270                                 inode->i_gid = task->egid;
1271                         } else {
1272                                 inode->i_uid = 0;
1273                                 inode->i_gid = 0;
1274                         }
1275                         security_task_to_inode(task, inode);
1276                         return 1;
1277                 }
1278                 spin_unlock(&files->file_lock);
1279                 put_files_struct(files);
1280         }
1281         d_drop(dentry);
1282         return 0;
1283 }
1284
1285 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1286 {
1287         struct task_struct *task = proc_task(inode);
1288         spin_lock(&task->proc_lock);
1289         if (task->proc_dentry == dentry)
1290                 task->proc_dentry = NULL;
1291         spin_unlock(&task->proc_lock);
1292         iput(inode);
1293 }
1294
1295 static int pid_delete_dentry(struct dentry * dentry)
1296 {
1297         /* Is the task we represent dead?
1298          * If so, then don't put the dentry on the lru list,
1299          * kill it immediately.
1300          */
1301         return !pid_alive(proc_task(dentry->d_inode));
1302 }
1303
1304 static struct dentry_operations tid_fd_dentry_operations =
1305 {
1306         .d_revalidate   = tid_fd_revalidate,
1307         .d_delete       = pid_delete_dentry,
1308 };
1309
1310 static struct dentry_operations pid_dentry_operations =
1311 {
1312         .d_revalidate   = pid_revalidate,
1313         .d_delete       = pid_delete_dentry,
1314 };
1315
1316 static struct dentry_operations pid_base_dentry_operations =
1317 {
1318         .d_revalidate   = pid_revalidate,
1319         .d_iput         = pid_base_iput,
1320         .d_delete       = pid_delete_dentry,
1321 };
1322
1323 /* Lookups */
1324
1325 static unsigned name_to_int(struct dentry *dentry)
1326 {
1327         const char *name = dentry->d_name.name;
1328         int len = dentry->d_name.len;
1329         unsigned n = 0;
1330
1331         if (len > 1 && *name == '0')
1332                 goto out;
1333         while (len-- > 0) {
1334                 unsigned c = *name++ - '0';
1335                 if (c > 9)
1336                         goto out;
1337                 if (n >= (~0U-9)/10)
1338                         goto out;
1339                 n *= 10;
1340                 n += c;
1341         }
1342         return n;
1343 out:
1344         return ~0U;
1345 }
1346
1347 /* SMP-safe */
1348 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1349 {
1350         struct task_struct *task = proc_task(dir);
1351         unsigned fd = name_to_int(dentry);
1352         struct file * file;
1353         struct files_struct * files;
1354         struct inode *inode;
1355         struct proc_inode *ei;
1356
1357         if (fd == ~0U)
1358                 goto out;
1359         if (!pid_alive(task))
1360                 goto out;
1361
1362         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1363         if (!inode)
1364                 goto out;
1365         ei = PROC_I(inode);
1366         files = get_files_struct(task);
1367         if (!files)
1368                 goto out_unlock;
1369         inode->i_mode = S_IFLNK;
1370         spin_lock(&files->file_lock);
1371         file = fcheck_files(files, fd);
1372         if (!file)
1373                 goto out_unlock2;
1374         if (file->f_mode & 1)
1375                 inode->i_mode |= S_IRUSR | S_IXUSR;
1376         if (file->f_mode & 2)
1377                 inode->i_mode |= S_IWUSR | S_IXUSR;
1378         spin_unlock(&files->file_lock);
1379         put_files_struct(files);
1380         inode->i_op = &proc_pid_link_inode_operations;
1381         inode->i_size = 64;
1382         ei->op.proc_get_link = proc_fd_link;
1383         dentry->d_op = &tid_fd_dentry_operations;
1384         d_add(dentry, inode);
1385         return NULL;
1386
1387 out_unlock2:
1388         spin_unlock(&files->file_lock);
1389         put_files_struct(files);
1390 out_unlock:
1391         iput(inode);
1392 out:
1393         return ERR_PTR(-ENOENT);
1394 }
1395
1396 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1397 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1398
1399 static struct file_operations proc_fd_operations = {
1400         .read           = generic_read_dir,
1401         .readdir        = proc_readfd,
1402 };
1403
1404 static struct file_operations proc_task_operations = {
1405         .read           = generic_read_dir,
1406         .readdir        = proc_task_readdir,
1407 };
1408
1409 /*
1410  * proc directories can do almost nothing..
1411  */
1412 static struct inode_operations proc_fd_inode_operations = {
1413         .lookup         = proc_lookupfd,
1414         .permission     = proc_permission,
1415 };
1416
1417 static struct inode_operations proc_task_inode_operations = {
1418         .lookup         = proc_task_lookup,
1419         .permission     = proc_permission,
1420 };
1421
1422 #ifdef CONFIG_SECURITY
1423 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1424                                   size_t count, loff_t *ppos)
1425 {
1426         struct inode * inode = file->f_dentry->d_inode;
1427         unsigned long page;
1428         ssize_t length;
1429         struct task_struct *task = proc_task(inode);
1430
1431         if (count > PAGE_SIZE)
1432                 count = PAGE_SIZE;
1433         if (!(page = __get_free_page(GFP_KERNEL)))
1434                 return -ENOMEM;
1435
1436         length = security_getprocattr(task,
1437                                       (char*)file->f_dentry->d_name.name,
1438                                       (void*)page, count);
1439         if (length >= 0)
1440                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1441         free_page(page);
1442         return length;
1443 }
1444
1445 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1446                                    size_t count, loff_t *ppos)
1447 {
1448         struct inode * inode = file->f_dentry->d_inode;
1449         char *page;
1450         ssize_t length;
1451         struct task_struct *task = proc_task(inode);
1452
1453         if (count > PAGE_SIZE)
1454                 count = PAGE_SIZE;
1455         if (*ppos != 0) {
1456                 /* No partial writes. */
1457                 return -EINVAL;
1458         }
1459         page = (char*)__get_free_page(GFP_USER);
1460         if (!page)
1461                 return -ENOMEM;
1462         length = -EFAULT;
1463         if (copy_from_user(page, buf, count))
1464                 goto out;
1465
1466         length = security_setprocattr(task,
1467                                       (char*)file->f_dentry->d_name.name,
1468                                       (void*)page, count);
1469 out:
1470         free_page((unsigned long) page);
1471         return length;
1472 }
1473
1474 static struct file_operations proc_pid_attr_operations = {
1475         .read           = proc_pid_attr_read,
1476         .write          = proc_pid_attr_write,
1477 };
1478
1479 static struct file_operations proc_tid_attr_operations;
1480 static struct inode_operations proc_tid_attr_inode_operations;
1481 static struct file_operations proc_tgid_attr_operations;
1482 static struct inode_operations proc_tgid_attr_inode_operations;
1483 #endif
1484
1485 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1486
1487 /* SMP-safe */
1488 static struct dentry *proc_pident_lookup(struct inode *dir,
1489                                          struct dentry *dentry,
1490                                          struct pid_entry *ents)
1491 {
1492         struct inode *inode;
1493         int error;
1494         struct task_struct *task = proc_task(dir);
1495         struct pid_entry *p;
1496         struct proc_inode *ei;
1497
1498         error = -ENOENT;
1499         inode = NULL;
1500
1501         if (!pid_alive(task))
1502                 goto out;
1503
1504         for (p = ents; p->name; p++) {
1505                 if (p->len != dentry->d_name.len)
1506                         continue;
1507                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1508                         break;
1509         }
1510         if (!p->name)
1511                 goto out;
1512
1513         error = -EINVAL;
1514         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1515         if (!inode)
1516                 goto out;
1517
1518         ei = PROC_I(inode);
1519         inode->i_mode = p->mode;
1520         /*
1521          * Yes, it does not scale. And it should not. Don't add
1522          * new entries into /proc/<tgid>/ without very good reasons.
1523          */
1524         switch(p->type) {
1525                 case PROC_TGID_TASK:
1526                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1527                         inode->i_op = &proc_task_inode_operations;
1528                         inode->i_fop = &proc_task_operations;
1529                         break;
1530                 case PROC_TID_FD:
1531                 case PROC_TGID_FD:
1532                         inode->i_nlink = 2;
1533                         inode->i_op = &proc_fd_inode_operations;
1534                         inode->i_fop = &proc_fd_operations;
1535                         break;
1536                 case PROC_TID_EXE:
1537                 case PROC_TGID_EXE:
1538                         inode->i_op = &proc_pid_link_inode_operations;
1539                         ei->op.proc_get_link = proc_exe_link;
1540                         break;
1541                 case PROC_TID_CWD:
1542                 case PROC_TGID_CWD:
1543                         inode->i_op = &proc_pid_link_inode_operations;
1544                         ei->op.proc_get_link = proc_cwd_link;
1545                         break;
1546                 case PROC_TID_ROOT:
1547                 case PROC_TGID_ROOT:
1548                         inode->i_op = &proc_pid_link_inode_operations;
1549                         ei->op.proc_get_link = proc_root_link;
1550                         break;
1551                 case PROC_TID_ENVIRON:
1552                 case PROC_TGID_ENVIRON:
1553                         inode->i_fop = &proc_info_file_operations;
1554                         ei->op.proc_read = proc_pid_environ;
1555                         break;
1556                 case PROC_TID_AUXV:
1557                 case PROC_TGID_AUXV:
1558                         inode->i_fop = &proc_info_file_operations;
1559                         ei->op.proc_read = proc_pid_auxv;
1560                         break;
1561                 case PROC_TID_STATUS:
1562                 case PROC_TGID_STATUS:
1563                         inode->i_fop = &proc_info_file_operations;
1564                         ei->op.proc_read = proc_pid_status;
1565                         break;
1566                 case PROC_TID_STAT:
1567                         inode->i_fop = &proc_info_file_operations;
1568                         ei->op.proc_read = proc_tid_stat;
1569                         break;
1570                 case PROC_TGID_STAT:
1571                         inode->i_fop = &proc_info_file_operations;
1572                         ei->op.proc_read = proc_tgid_stat;
1573                         break;
1574                 case PROC_TID_CMDLINE:
1575                 case PROC_TGID_CMDLINE:
1576                         inode->i_fop = &proc_info_file_operations;
1577                         ei->op.proc_read = proc_pid_cmdline;
1578                         break;
1579                 case PROC_TID_STATM:
1580                 case PROC_TGID_STATM:
1581                         inode->i_fop = &proc_info_file_operations;
1582                         ei->op.proc_read = proc_pid_statm;
1583                         break;
1584                 case PROC_TID_MAPS:
1585                 case PROC_TGID_MAPS:
1586                         inode->i_fop = &proc_maps_operations;
1587                         break;
1588 #ifdef CONFIG_NUMA
1589                 case PROC_TID_NUMA_MAPS:
1590                 case PROC_TGID_NUMA_MAPS:
1591                         inode->i_fop = &proc_numa_maps_operations;
1592                         break;
1593 #endif
1594                 case PROC_TID_MEM:
1595                 case PROC_TGID_MEM:
1596                         inode->i_op = &proc_mem_inode_operations;
1597                         inode->i_fop = &proc_mem_operations;
1598                         break;
1599 #ifdef CONFIG_SECCOMP
1600                 case PROC_TID_SECCOMP:
1601                 case PROC_TGID_SECCOMP:
1602                         inode->i_fop = &proc_seccomp_operations;
1603                         break;
1604 #endif /* CONFIG_SECCOMP */
1605                 case PROC_TID_MOUNTS:
1606                 case PROC_TGID_MOUNTS:
1607                         inode->i_fop = &proc_mounts_operations;
1608                         break;
1609                 case PROC_TID_SMAPS:
1610                 case PROC_TGID_SMAPS:
1611                         inode->i_fop = &proc_smaps_operations;
1612                         break;
1613 #ifdef CONFIG_SECURITY
1614                 case PROC_TID_ATTR:
1615                         inode->i_nlink = 2;
1616                         inode->i_op = &proc_tid_attr_inode_operations;
1617                         inode->i_fop = &proc_tid_attr_operations;
1618                         break;
1619                 case PROC_TGID_ATTR:
1620                         inode->i_nlink = 2;
1621                         inode->i_op = &proc_tgid_attr_inode_operations;
1622                         inode->i_fop = &proc_tgid_attr_operations;
1623                         break;
1624                 case PROC_TID_ATTR_CURRENT:
1625                 case PROC_TGID_ATTR_CURRENT:
1626                 case PROC_TID_ATTR_PREV:
1627                 case PROC_TGID_ATTR_PREV:
1628                 case PROC_TID_ATTR_EXEC:
1629                 case PROC_TGID_ATTR_EXEC:
1630                 case PROC_TID_ATTR_FSCREATE:
1631                 case PROC_TGID_ATTR_FSCREATE:
1632                         inode->i_fop = &proc_pid_attr_operations;
1633                         break;
1634 #endif
1635 #ifdef CONFIG_KALLSYMS
1636                 case PROC_TID_WCHAN:
1637                 case PROC_TGID_WCHAN:
1638                         inode->i_fop = &proc_info_file_operations;
1639                         ei->op.proc_read = proc_pid_wchan;
1640                         break;
1641 #endif
1642 #ifdef CONFIG_SCHEDSTATS
1643                 case PROC_TID_SCHEDSTAT:
1644                 case PROC_TGID_SCHEDSTAT:
1645                         inode->i_fop = &proc_info_file_operations;
1646                         ei->op.proc_read = proc_pid_schedstat;
1647                         break;
1648 #endif
1649 #ifdef CONFIG_CPUSETS
1650                 case PROC_TID_CPUSET:
1651                 case PROC_TGID_CPUSET:
1652                         inode->i_fop = &proc_cpuset_operations;
1653                         break;
1654 #endif
1655                 case PROC_TID_OOM_SCORE:
1656                 case PROC_TGID_OOM_SCORE:
1657                         inode->i_fop = &proc_info_file_operations;
1658                         ei->op.proc_read = proc_oom_score;
1659                         break;
1660                 case PROC_TID_OOM_ADJUST:
1661                 case PROC_TGID_OOM_ADJUST:
1662                         inode->i_fop = &proc_oom_adjust_operations;
1663                         break;
1664 #ifdef CONFIG_AUDITSYSCALL
1665                 case PROC_TID_LOGINUID:
1666                 case PROC_TGID_LOGINUID:
1667                         inode->i_fop = &proc_loginuid_operations;
1668                         break;
1669 #endif
1670                 default:
1671                         printk("procfs: impossible type (%d)",p->type);
1672                         iput(inode);
1673                         return ERR_PTR(-EINVAL);
1674         }
1675         dentry->d_op = &pid_dentry_operations;
1676         d_add(dentry, inode);
1677         return NULL;
1678
1679 out:
1680         return ERR_PTR(error);
1681 }
1682
1683 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1684         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1685 }
1686
1687 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1688         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1689 }
1690
1691 static struct file_operations proc_tgid_base_operations = {
1692         .read           = generic_read_dir,
1693         .readdir        = proc_tgid_base_readdir,
1694 };
1695
1696 static struct file_operations proc_tid_base_operations = {
1697         .read           = generic_read_dir,
1698         .readdir        = proc_tid_base_readdir,
1699 };
1700
1701 static struct inode_operations proc_tgid_base_inode_operations = {
1702         .lookup         = proc_tgid_base_lookup,
1703 };
1704
1705 static struct inode_operations proc_tid_base_inode_operations = {
1706         .lookup         = proc_tid_base_lookup,
1707 };
1708
1709 #ifdef CONFIG_SECURITY
1710 static int proc_tgid_attr_readdir(struct file * filp,
1711                              void * dirent, filldir_t filldir)
1712 {
1713         return proc_pident_readdir(filp,dirent,filldir,
1714                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1715 }
1716
1717 static int proc_tid_attr_readdir(struct file * filp,
1718                              void * dirent, filldir_t filldir)
1719 {
1720         return proc_pident_readdir(filp,dirent,filldir,
1721                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1722 }
1723
1724 static struct file_operations proc_tgid_attr_operations = {
1725         .read           = generic_read_dir,
1726         .readdir        = proc_tgid_attr_readdir,
1727 };
1728
1729 static struct file_operations proc_tid_attr_operations = {
1730         .read           = generic_read_dir,
1731         .readdir        = proc_tid_attr_readdir,
1732 };
1733
1734 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1735                                 struct dentry *dentry, struct nameidata *nd)
1736 {
1737         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1738 }
1739
1740 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1741                                 struct dentry *dentry, struct nameidata *nd)
1742 {
1743         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1744 }
1745
1746 static struct inode_operations proc_tgid_attr_inode_operations = {
1747         .lookup         = proc_tgid_attr_lookup,
1748 };
1749
1750 static struct inode_operations proc_tid_attr_inode_operations = {
1751         .lookup         = proc_tid_attr_lookup,
1752 };
1753 #endif
1754
1755 /*
1756  * /proc/self:
1757  */
1758 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1759                               int buflen)
1760 {
1761         char tmp[30];
1762         sprintf(tmp, "%d", current->tgid);
1763         return vfs_readlink(dentry,buffer,buflen,tmp);
1764 }
1765
1766 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1767 {
1768         char tmp[30];
1769         sprintf(tmp, "%d", current->tgid);
1770         return ERR_PTR(vfs_follow_link(nd,tmp));
1771 }
1772
1773 static struct inode_operations proc_self_inode_operations = {
1774         .readlink       = proc_self_readlink,
1775         .follow_link    = proc_self_follow_link,
1776 };
1777
1778 /**
1779  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1780  * @p: task that should be flushed.
1781  *
1782  * Drops the /proc/@pid dcache entry from the hash chains.
1783  *
1784  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1785  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1786  * if the pid value is immediately reused. This is enforced by
1787  * - caller must acquire spin_lock(p->proc_lock)
1788  * - must be called before detach_pid()
1789  * - proc_pid_lookup acquires proc_lock, and checks that
1790  *   the target is not dead by looking at the attach count
1791  *   of PIDTYPE_PID.
1792  */
1793
1794 struct dentry *proc_pid_unhash(struct task_struct *p)
1795 {
1796         struct dentry *proc_dentry;
1797
1798         proc_dentry = p->proc_dentry;
1799         if (proc_dentry != NULL) {
1800
1801                 spin_lock(&dcache_lock);
1802                 spin_lock(&proc_dentry->d_lock);
1803                 if (!d_unhashed(proc_dentry)) {
1804                         dget_locked(proc_dentry);
1805                         __d_drop(proc_dentry);
1806                         spin_unlock(&proc_dentry->d_lock);
1807                 } else {
1808                         spin_unlock(&proc_dentry->d_lock);
1809                         proc_dentry = NULL;
1810                 }
1811                 spin_unlock(&dcache_lock);
1812         }
1813         return proc_dentry;
1814 }
1815
1816 /**
1817  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1818  * @proc_dentry: directoy to prune.
1819  *
1820  * Shrink the /proc directory that was used by the just killed thread.
1821  */
1822
1823 void proc_pid_flush(struct dentry *proc_dentry)
1824 {
1825         might_sleep();
1826         if(proc_dentry != NULL) {
1827                 shrink_dcache_parent(proc_dentry);
1828                 dput(proc_dentry);
1829         }
1830 }
1831
1832 /* SMP-safe */
1833 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1834 {
1835         struct task_struct *task;
1836         struct inode *inode;
1837         struct proc_inode *ei;
1838         unsigned tgid;
1839         int died;
1840
1841         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
1842                 inode = new_inode(dir->i_sb);
1843                 if (!inode)
1844                         return ERR_PTR(-ENOMEM);
1845                 ei = PROC_I(inode);
1846                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1847                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
1848                 ei->pde = NULL;
1849                 inode->i_mode = S_IFLNK|S_IRWXUGO;
1850                 inode->i_uid = inode->i_gid = 0;
1851                 inode->i_size = 64;
1852                 inode->i_op = &proc_self_inode_operations;
1853                 d_add(dentry, inode);
1854                 return NULL;
1855         }
1856         tgid = name_to_int(dentry);
1857         if (tgid == ~0U)
1858                 goto out;
1859
1860         read_lock(&tasklist_lock);
1861         task = find_task_by_pid(tgid);
1862         if (task)
1863                 get_task_struct(task);
1864         read_unlock(&tasklist_lock);
1865         if (!task)
1866                 goto out;
1867
1868         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
1869
1870
1871         if (!inode) {
1872                 put_task_struct(task);
1873                 goto out;
1874         }
1875         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1876         inode->i_op = &proc_tgid_base_inode_operations;
1877         inode->i_fop = &proc_tgid_base_operations;
1878         inode->i_flags|=S_IMMUTABLE;
1879 #ifdef CONFIG_SECURITY
1880         inode->i_nlink = 5;
1881 #else
1882         inode->i_nlink = 4;
1883 #endif
1884
1885         dentry->d_op = &pid_base_dentry_operations;
1886
1887         died = 0;
1888         d_add(dentry, inode);
1889         spin_lock(&task->proc_lock);
1890         task->proc_dentry = dentry;
1891         if (!pid_alive(task)) {
1892                 dentry = proc_pid_unhash(task);
1893                 died = 1;
1894         }
1895         spin_unlock(&task->proc_lock);
1896
1897         put_task_struct(task);
1898         if (died) {
1899                 proc_pid_flush(dentry);
1900                 goto out;
1901         }
1902         return NULL;
1903 out:
1904         return ERR_PTR(-ENOENT);
1905 }
1906
1907 /* SMP-safe */
1908 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1909 {
1910         struct task_struct *task;
1911         struct task_struct *leader = proc_task(dir);
1912         struct inode *inode;
1913         unsigned tid;
1914
1915         tid = name_to_int(dentry);
1916         if (tid == ~0U)
1917                 goto out;
1918
1919         read_lock(&tasklist_lock);
1920         task = find_task_by_pid(tid);
1921         if (task)
1922                 get_task_struct(task);
1923         read_unlock(&tasklist_lock);
1924         if (!task)
1925                 goto out;
1926         if (leader->tgid != task->tgid)
1927                 goto out_drop_task;
1928
1929         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
1930
1931
1932         if (!inode)
1933                 goto out_drop_task;
1934         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
1935         inode->i_op = &proc_tid_base_inode_operations;
1936         inode->i_fop = &proc_tid_base_operations;
1937         inode->i_flags|=S_IMMUTABLE;
1938 #ifdef CONFIG_SECURITY
1939         inode->i_nlink = 4;
1940 #else
1941         inode->i_nlink = 3;
1942 #endif
1943
1944         dentry->d_op = &pid_base_dentry_operations;
1945
1946         d_add(dentry, inode);
1947
1948         put_task_struct(task);
1949         return NULL;
1950 out_drop_task:
1951         put_task_struct(task);
1952 out:
1953         return ERR_PTR(-ENOENT);
1954 }
1955
1956 #define PROC_NUMBUF 10
1957 #define PROC_MAXPIDS 20
1958
1959 /*
1960  * Get a few tgid's to return for filldir - we need to hold the
1961  * tasklist lock while doing this, and we must release it before
1962  * we actually do the filldir itself, so we use a temp buffer..
1963  */
1964 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
1965 {
1966         struct task_struct *p;
1967         int nr_tgids = 0;
1968
1969         index--;
1970         read_lock(&tasklist_lock);
1971         p = NULL;
1972         if (version) {
1973                 p = find_task_by_pid(version);
1974                 if (p && !thread_group_leader(p))
1975                         p = NULL;
1976         }
1977
1978         if (p)
1979                 index = 0;
1980         else
1981                 p = next_task(&init_task);
1982
1983         for ( ; p != &init_task; p = next_task(p)) {
1984                 int tgid = p->pid;
1985                 if (!pid_alive(p))
1986                         continue;
1987                 if (--index >= 0)
1988                         continue;
1989                 tgids[nr_tgids] = tgid;
1990                 nr_tgids++;
1991                 if (nr_tgids >= PROC_MAXPIDS)
1992                         break;
1993         }
1994         read_unlock(&tasklist_lock);
1995         return nr_tgids;
1996 }
1997
1998 /*
1999  * Get a few tid's to return for filldir - we need to hold the
2000  * tasklist lock while doing this, and we must release it before
2001  * we actually do the filldir itself, so we use a temp buffer..
2002  */
2003 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2004 {
2005         struct task_struct *leader_task = proc_task(dir);
2006         struct task_struct *task = leader_task;
2007         int nr_tids = 0;
2008
2009         index -= 2;
2010         read_lock(&tasklist_lock);
2011         /*
2012          * The starting point task (leader_task) might be an already
2013          * unlinked task, which cannot be used to access the task-list
2014          * via next_thread().
2015          */
2016         if (pid_alive(task)) do {
2017                 int tid = task->pid;
2018
2019                 if (--index >= 0)
2020                         continue;
2021                 if (tids != NULL)
2022                         tids[nr_tids] = tid;
2023                 nr_tids++;
2024                 if (nr_tids >= PROC_MAXPIDS)
2025                         break;
2026         } while ((task = next_thread(task)) != leader_task);
2027         read_unlock(&tasklist_lock);
2028         return nr_tids;
2029 }
2030
2031 /* for the /proc/ directory itself, after non-process stuff has been done */
2032 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2033 {
2034         unsigned int tgid_array[PROC_MAXPIDS];
2035         char buf[PROC_NUMBUF];
2036         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2037         unsigned int nr_tgids, i;
2038         int next_tgid;
2039
2040         if (!nr) {
2041                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2042                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2043                         return 0;
2044                 filp->f_pos++;
2045                 nr++;
2046         }
2047
2048         /* f_version caches the tgid value that the last readdir call couldn't
2049          * return. lseek aka telldir automagically resets f_version to 0.
2050          */
2051         next_tgid = filp->f_version;
2052         filp->f_version = 0;
2053         for (;;) {
2054                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2055                 if (!nr_tgids) {
2056                         /* no more entries ! */
2057                         break;
2058                 }
2059                 next_tgid = 0;
2060
2061                 /* do not use the last found pid, reserve it for next_tgid */
2062                 if (nr_tgids == PROC_MAXPIDS) {
2063                         nr_tgids--;
2064                         next_tgid = tgid_array[nr_tgids];
2065                 }
2066
2067                 for (i=0;i<nr_tgids;i++) {
2068                         int tgid = tgid_array[i];
2069                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2070                         unsigned long j = PROC_NUMBUF;
2071
2072                         do
2073                                 buf[--j] = '0' + (tgid % 10);
2074                         while ((tgid /= 10) != 0);
2075
2076                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2077                                 /* returning this tgid failed, save it as the first
2078                                  * pid for the next readir call */
2079                                 filp->f_version = tgid_array[i];
2080                                 goto out;
2081                         }
2082                         filp->f_pos++;
2083                         nr++;
2084                 }
2085         }
2086 out:
2087         return 0;
2088 }
2089
2090 /* for the /proc/TGID/task/ directories */
2091 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2092 {
2093         unsigned int tid_array[PROC_MAXPIDS];
2094         char buf[PROC_NUMBUF];
2095         unsigned int nr_tids, i;
2096         struct dentry *dentry = filp->f_dentry;
2097         struct inode *inode = dentry->d_inode;
2098         int retval = -ENOENT;
2099         ino_t ino;
2100         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2101
2102         if (!pid_alive(proc_task(inode)))
2103                 goto out;
2104         retval = 0;
2105
2106         switch (pos) {
2107         case 0:
2108                 ino = inode->i_ino;
2109                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2110                         goto out;
2111                 pos++;
2112                 /* fall through */
2113         case 1:
2114                 ino = parent_ino(dentry);
2115                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2116                         goto out;
2117                 pos++;
2118                 /* fall through */
2119         }
2120
2121         nr_tids = get_tid_list(pos, tid_array, inode);
2122         inode->i_nlink = pos + nr_tids;
2123
2124         for (i = 0; i < nr_tids; i++) {
2125                 unsigned long j = PROC_NUMBUF;
2126                 int tid = tid_array[i];
2127
2128                 ino = fake_ino(tid,PROC_TID_INO);
2129
2130                 do
2131                         buf[--j] = '0' + (tid % 10);
2132                 while ((tid /= 10) != 0);
2133
2134                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2135                         break;
2136                 pos++;
2137         }
2138 out:
2139         filp->f_pos = pos;
2140         return retval;
2141 }