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 /*
  78  * For hysterical raisins we keep the same inumbers as in the old procfs.
  79  * Feel free to change the macro below - just keep the range distinct from
  80  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
  81  * As soon as we'll get a separate superblock we will be able to forget
  82  * about magical ranges too.
  83  */
  84
  85 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
  86
  87 enum pid_directory_inos {
  88         PROC_TGID_INO = 2,
  89         PROC_TGID_TASK,
  90         PROC_TGID_STATUS,
  91         PROC_TGID_MEM,
  92 #ifdef CONFIG_SECCOMP
  93         PROC_TGID_SECCOMP,
  94 #endif
  95         PROC_TGID_CWD,
  96         PROC_TGID_ROOT,
  97         PROC_TGID_EXE,
  98         PROC_TGID_FD,
  99         PROC_TGID_ENVIRON,
 100         PROC_TGID_AUXV,
 101         PROC_TGID_CMDLINE,
 102         PROC_TGID_STAT,
 103         PROC_TGID_STATM,
 104         PROC_TGID_MAPS,
 105         PROC_TGID_NUMA_MAPS,
 106         PROC_TGID_MOUNTS,
 107         PROC_TGID_MOUNTSTATS,
 108         PROC_TGID_WCHAN,
 109 #ifdef CONFIG_MMU
 110         PROC_TGID_SMAPS,
 111 #endif
 112 #ifdef CONFIG_SCHEDSTATS
 113         PROC_TGID_SCHEDSTAT,
 114 #endif
 115 #ifdef CONFIG_CPUSETS
 116         PROC_TGID_CPUSET,
 117 #endif
 118 #ifdef CONFIG_SECURITY
 119         PROC_TGID_ATTR,
 120         PROC_TGID_ATTR_CURRENT,
 121         PROC_TGID_ATTR_PREV,
 122         PROC_TGID_ATTR_EXEC,
 123         PROC_TGID_ATTR_FSCREATE,
 124 #endif
 125 #ifdef CONFIG_AUDITSYSCALL
 126         PROC_TGID_LOGINUID,
 127 #endif
 128         PROC_TGID_OOM_SCORE,
 129         PROC_TGID_OOM_ADJUST,
 130         PROC_TID_INO,
 131         PROC_TID_STATUS,
 132         PROC_TID_MEM,
 133 #ifdef CONFIG_SECCOMP
 134         PROC_TID_SECCOMP,
 135 #endif
 136         PROC_TID_CWD,
 137         PROC_TID_ROOT,
 138         PROC_TID_EXE,
 139         PROC_TID_FD,
 140         PROC_TID_ENVIRON,
 141         PROC_TID_AUXV,
 142         PROC_TID_CMDLINE,
 143         PROC_TID_STAT,
 144         PROC_TID_STATM,
 145         PROC_TID_MAPS,
 146         PROC_TID_NUMA_MAPS,
 147         PROC_TID_MOUNTS,
 148         PROC_TID_MOUNTSTATS,
 149         PROC_TID_WCHAN,
 150 #ifdef CONFIG_MMU
 151         PROC_TID_SMAPS,
 152 #endif
 153 #ifdef CONFIG_SCHEDSTATS
 154         PROC_TID_SCHEDSTAT,
 155 #endif
 156 #ifdef CONFIG_CPUSETS
 157         PROC_TID_CPUSET,
 158 #endif
 159 #ifdef CONFIG_SECURITY
 160         PROC_TID_ATTR,
 161         PROC_TID_ATTR_CURRENT,
 162         PROC_TID_ATTR_PREV,
 163         PROC_TID_ATTR_EXEC,
 164         PROC_TID_ATTR_FSCREATE,
 165 #endif
 166 #ifdef CONFIG_AUDITSYSCALL
 167         PROC_TID_LOGINUID,
 168 #endif
 169         PROC_TID_OOM_SCORE,
 170         PROC_TID_OOM_ADJUST,
 171
 172         /* Add new entries before this */
 173         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
 174 };
 175
 176 struct pid_entry {
 177         int type;
 178         int len;
 179         char *name;
 180         mode_t mode;
 181 };
 182
 183 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
 184
 185 static struct pid_entry tgid_base_stuff[] = {
 186         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
 187         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 188         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
 189         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
 190         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
 191         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 192         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 193         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
 194         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 195 #ifdef CONFIG_NUMA
 196         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
 197 #endif
 198         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 199 #ifdef CONFIG_SECCOMP
 200         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 201 #endif
 202         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
 203         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
 204         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
 205         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
 206         E(PROC_TGID_MOUNTSTATS, "mountstats", S_IFREG|S_IRUSR),
 207 #ifdef CONFIG_MMU
 208         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
 209 #endif
 210 #ifdef CONFIG_SECURITY
 211         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 212 #endif
 213 #ifdef CONFIG_KALLSYMS
 214         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
 215 #endif
 216 #ifdef CONFIG_SCHEDSTATS
 217         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
 218 #endif
 219 #ifdef CONFIG_CPUSETS
 220         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
 221 #endif
 222         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
 223         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 224 #ifdef CONFIG_AUDITSYSCALL
 225         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 226 #endif
 227         {0,0,NULL,0}
 228 };
 229 static struct pid_entry tid_base_stuff[] = {
 230         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 231         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
 232         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
 233         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
 234         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 235         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 236         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
 237         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 238 #ifdef CONFIG_NUMA
 239         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
 240 #endif
 241         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 242 #ifdef CONFIG_SECCOMP
 243         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 244 #endif
 245         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
 246         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
 247         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
 248         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
 249 #ifdef CONFIG_MMU
 250         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
 251 #endif
 252 #ifdef CONFIG_SECURITY
 253         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 254 #endif
 255 #ifdef CONFIG_KALLSYMS
 256         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
 257 #endif
 258 #ifdef CONFIG_SCHEDSTATS
 259         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
 260 #endif
 261 #ifdef CONFIG_CPUSETS
 262         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
 263 #endif
 264         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
 265         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 266 #ifdef CONFIG_AUDITSYSCALL
 267         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 268 #endif
 269         {0,0,NULL,0}
 270 };
 271
 272 #ifdef CONFIG_SECURITY
 273 static struct pid_entry tgid_attr_stuff[] = {
 274         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
 275         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
 276         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 277         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 278         {0,0,NULL,0}
 279 };
 280 static struct pid_entry tid_attr_stuff[] = {
 281         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
 282         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
 283         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 284         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 285         {0,0,NULL,0}
 286 };
 287 #endif
 288
 289 #undef E
 290
 291 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 292 {
 293         struct task_struct *task = proc_task(inode);
 294         struct files_struct *files;
 295         struct file *file;
 296         int fd = proc_type(inode) - PROC_TID_FD_DIR;
 297
 298         files = get_files_struct(task);
 299         if (files) {
 300                 /*
 301                  * We are not taking a ref to the file structure, so we must
 302                  * hold ->file_lock.
 303                  */
 304                 spin_lock(&files->file_lock);
 305                 file = fcheck_files(files, fd);
 306                 if (file) {
 307                         *mnt = mntget(file->f_vfsmnt);
 308                         *dentry = dget(file->f_dentry);
 309                         spin_unlock(&files->file_lock);
 310                         put_files_struct(files);
 311                         return 0;
 312                 }
 313                 spin_unlock(&files->file_lock);
 314                 put_files_struct(files);
 315         }
 316         return -ENOENT;
 317 }
 318
 319 static struct fs_struct *get_fs_struct(struct task_struct *task)
 320 {
 321         struct fs_struct *fs;
 322         task_lock(task);
 323         fs = task->fs;
 324         if(fs)
 325                 atomic_inc(&fs->count);
 326         task_unlock(task);
 327         return fs;
 328 }
 329
 330 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 331 {
 332         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 333         int result = -ENOENT;
 334         if (fs) {
 335                 read_lock(&fs->lock);
 336                 *mnt = mntget(fs->pwdmnt);
 337                 *dentry = dget(fs->pwd);
 338                 read_unlock(&fs->lock);
 339                 result = 0;
 340                 put_fs_struct(fs);
 341         }
 342         return result;
 343 }
 344
 345 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 346 {
 347         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 348         int result = -ENOENT;
 349         if (fs) {
 350                 read_lock(&fs->lock);
 351                 *mnt = mntget(fs->rootmnt);
 352                 *dentry = dget(fs->root);
 353                 read_unlock(&fs->lock);
 354                 result = 0;
 355                 put_fs_struct(fs);
 356         }
 357         return result;
 358 }
 359
 360
 361 /* Same as proc_root_link, but this addionally tries to get fs from other
 362  * threads in the group */
 363 static int proc_task_root_link(struct inode *inode, struct dentry **dentry,
 364                                 struct vfsmount **mnt)
 365 {
 366         struct fs_struct *fs;
 367         int result = -ENOENT;
 368         struct task_struct *leader = proc_task(inode);
 369
 370         task_lock(leader);
 371         fs = leader->fs;
 372         if (fs) {
 373                 atomic_inc(&fs->count);
 374                 task_unlock(leader);
 375         } else {
 376                 /* Try to get fs from other threads */
 377                 task_unlock(leader);
 378                 read_lock(&tasklist_lock);
 379                 if (pid_alive(leader)) {
 380                         struct task_struct *task = leader;
 381
 382                         while ((task = next_thread(task)) != leader) {
 383                                 task_lock(task);
 384                                 fs = task->fs;
 385                                 if (fs) {
 386                                         atomic_inc(&fs->count);
 387                                         task_unlock(task);
 388                                         break;
 389                                 }
 390                                 task_unlock(task);
 391                         }
 392                 }
 393                 read_unlock(&tasklist_lock);
 394         }
 395
 396         if (fs) {
 397                 read_lock(&fs->lock);
 398                 *mnt = mntget(fs->rootmnt);
 399                 *dentry = dget(fs->root);
 400                 read_unlock(&fs->lock);
 401                 result = 0;
 402                 put_fs_struct(fs);
 403         }
 404         return result;
 405 }
 406
 407
 408 #define MAY_PTRACE(task) \
 409         (task == current || \
 410         (task->parent == current && \
 411         (task->ptrace & PT_PTRACED) && \
 412          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
 413          security_ptrace(current,task) == 0))
 414
 415 static int proc_pid_environ(struct task_struct *task, char * buffer)
 416 {
 417         int res = 0;
 418         struct mm_struct *mm = get_task_mm(task);
 419         if (mm) {
 420                 unsigned int len = mm->env_end - mm->env_start;
 421                 if (len > PAGE_SIZE)
 422                         len = PAGE_SIZE;
 423                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
 424                 if (!ptrace_may_attach(task))
 425                         res = -ESRCH;
 426                 mmput(mm);
 427         }
 428         return res;
 429 }
 430
 431 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 432 {
 433         int res = 0;
 434         unsigned int len;
 435         struct mm_struct *mm = get_task_mm(task);
 436         if (!mm)
 437                 goto out;
 438         if (!mm->arg_end)
 439                 goto out_mm;    /* Shh! No looking before we're done */
 440
 441         len = mm->arg_end - mm->arg_start;
 442
 443         if (len > PAGE_SIZE)
 444                 len = PAGE_SIZE;
 445
 446         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 447
 448         // If the nul at the end of args has been overwritten, then
 449         // assume application is using setproctitle(3).
 450         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 451                 len = strnlen(buffer, res);
 452                 if (len < res) {
 453                     res = len;
 454                 } else {
 455                         len = mm->env_end - mm->env_start;
 456                         if (len > PAGE_SIZE - res)
 457                                 len = PAGE_SIZE - res;
 458                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 459                         res = strnlen(buffer, res);
 460                 }
 461         }
 462 out_mm:
 463         mmput(mm);
 464 out:
 465         return res;
 466 }
 467
 468 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 469 {
 470         int res = 0;
 471         struct mm_struct *mm = get_task_mm(task);
 472         if (mm) {
 473                 unsigned int nwords = 0;
 474                 do
 475                         nwords += 2;
 476                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 477                 res = nwords * sizeof(mm->saved_auxv[0]);
 478                 if (res > PAGE_SIZE)
 479                         res = PAGE_SIZE;
 480                 memcpy(buffer, mm->saved_auxv, res);
 481                 mmput(mm);
 482         }
 483         return res;
 484 }
 485
 486
 487 #ifdef CONFIG_KALLSYMS
 488 /*
 489  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 490  * Returns the resolved symbol.  If that fails, simply return the address.
 491  */
 492 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 493 {
 494         char *modname;
 495         const char *sym_name;
 496         unsigned long wchan, size, offset;
 497         char namebuf[KSYM_NAME_LEN+1];
 498
 499         wchan = get_wchan(task);
 500
 501         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
 502         if (sym_name)
 503                 return sprintf(buffer, "%s", sym_name);
 504         return sprintf(buffer, "%lu", wchan);
 505 }
 506 #endif /* CONFIG_KALLSYMS */
 507
 508 #ifdef CONFIG_SCHEDSTATS
 509 /*
 510  * Provides /proc/PID/schedstat
 511  */
 512 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 513 {
 514         return sprintf(buffer, "%lu %lu %lu\n",
 515                         task->sched_info.cpu_time,
 516                         task->sched_info.run_delay,
 517                         task->sched_info.pcnt);
 518 }
 519 #endif
 520
 521 /* The badness from the OOM killer */
 522 unsigned long badness(struct task_struct *p, unsigned long uptime);
 523 static int proc_oom_score(struct task_struct *task, char *buffer)
 524 {
 525         unsigned long points;
 526         struct timespec uptime;
 527
 528         do_posix_clock_monotonic_gettime(&uptime);
 529         points = badness(task, uptime.tv_sec);
 530         return sprintf(buffer, "%lu\n", points);
 531 }
 532
 533 /************************************************************************/
 534 /*                       Here the fs part begins                        */
 535 /************************************************************************/
 536
 537 /* permission checks */
 538
 539 /* If the process being read is separated by chroot from the reading process,
 540  * don't let the reader access the threads.
 541  *
 542  * note: this does dput(root) and mntput(vfsmnt) on exit.
 543  */
 544 static int proc_check_chroot(struct dentry *root, struct vfsmount *vfsmnt)
 545 {
 546         struct dentry *de, *base;
 547         struct vfsmount *our_vfsmnt, *mnt;
 548         int res = 0;
 549
 550         read_lock(&current->fs->lock);
 551         our_vfsmnt = mntget(current->fs->rootmnt);
 552         base = dget(current->fs->root);
 553         read_unlock(&current->fs->lock);
 554
 555         spin_lock(&vfsmount_lock);
 556         de = root;
 557         mnt = vfsmnt;
 558
 559         while (mnt != our_vfsmnt) {
 560                 if (mnt == mnt->mnt_parent)
 561                         goto out;
 562                 de = mnt->mnt_mountpoint;
 563                 mnt = mnt->mnt_parent;
 564         }
 565
 566         if (!is_subdir(de, base))
 567                 goto out;
 568         spin_unlock(&vfsmount_lock);
 569
 570 exit:
 571         dput(base);
 572         mntput(our_vfsmnt);
 573         dput(root);
 574         mntput(vfsmnt);
 575         return res;
 576 out:
 577         spin_unlock(&vfsmount_lock);
 578         res = -EACCES;
 579         goto exit;
 580 }
 581
 582 static int proc_check_root(struct inode *inode)
 583 {
 584         struct dentry *root;
 585         struct vfsmount *vfsmnt;
 586
 587         if (proc_root_link(inode, &root, &vfsmnt)) /* Ewww... */
 588                 return -ENOENT;
 589         return proc_check_chroot(root, vfsmnt);
 590 }
 591
 592 static int proc_permission(struct inode *inode, int mask, struct nameidata *nd)
 593 {
 594         if (generic_permission(inode, mask, NULL) != 0)
 595                 return -EACCES;
 596         return proc_check_root(inode);
 597 }
 598
 599 static int proc_task_permission(struct inode *inode, int mask, struct nameidata *nd)
 600 {
 601         struct dentry *root;
 602         struct vfsmount *vfsmnt;
 603
 604         if (generic_permission(inode, mask, NULL) != 0)
 605                 return -EACCES;
 606
 607         if (proc_task_root_link(inode, &root, &vfsmnt))
 608                 return -ENOENT;
 609
 610         return proc_check_chroot(root, vfsmnt);
 611 }
 612
 613 extern struct seq_operations proc_pid_maps_op;
 614 static int maps_open(struct inode *inode, struct file *file)
 615 {
 616         struct task_struct *task = proc_task(inode);
 617         int ret = seq_open(file, &proc_pid_maps_op);
 618         if (!ret) {
 619                 struct seq_file *m = file->private_data;
 620                 m->private = task;
 621         }
 622         return ret;
 623 }
 624
 625 static struct file_operations proc_maps_operations = {
 626         .open           = maps_open,
 627         .read           = seq_read,
 628         .llseek         = seq_lseek,
 629         .release        = seq_release,
 630 };
 631
 632 #ifdef CONFIG_NUMA
 633 extern struct seq_operations proc_pid_numa_maps_op;
 634 static int numa_maps_open(struct inode *inode, struct file *file)
 635 {
 636         struct task_struct *task = proc_task(inode);
 637         int ret = seq_open(file, &proc_pid_numa_maps_op);
 638         if (!ret) {
 639                 struct seq_file *m = file->private_data;
 640                 m->private = task;
 641         }
 642         return ret;
 643 }
 644
 645 static struct file_operations proc_numa_maps_operations = {
 646         .open           = numa_maps_open,
 647         .read           = seq_read,
 648         .llseek         = seq_lseek,
 649         .release        = seq_release,
 650 };
 651 #endif
 652
 653 #ifdef CONFIG_MMU
 654 extern struct seq_operations proc_pid_smaps_op;
 655 static int smaps_open(struct inode *inode, struct file *file)
 656 {
 657         struct task_struct *task = proc_task(inode);
 658         int ret = seq_open(file, &proc_pid_smaps_op);
 659         if (!ret) {
 660                 struct seq_file *m = file->private_data;
 661                 m->private = task;
 662         }
 663         return ret;
 664 }
 665
 666 static struct file_operations proc_smaps_operations = {
 667         .open           = smaps_open,
 668         .read           = seq_read,
 669         .llseek         = seq_lseek,
 670         .release        = seq_release,
 671 };
 672 #endif
 673
 674 extern struct seq_operations mounts_op;
 675 struct proc_mounts {
 676         struct seq_file m;
 677         int event;
 678 };
 679
 680 static int mounts_open(struct inode *inode, struct file *file)
 681 {
 682         struct task_struct *task = proc_task(inode);
 683         struct namespace *namespace;
 684         struct proc_mounts *p;
 685         int ret = -EINVAL;
 686
 687         task_lock(task);
 688         namespace = task->namespace;
 689         if (namespace)
 690                 get_namespace(namespace);
 691         task_unlock(task);
 692
 693         if (namespace) {
 694                 ret = -ENOMEM;
 695                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 696                 if (p) {
 697                         file->private_data = &p->m;
 698                         ret = seq_open(file, &mounts_op);
 699                         if (!ret) {
 700                                 p->m.private = namespace;
 701                                 p->event = namespace->event;
 702                                 return 0;
 703                         }
 704                         kfree(p);
 705                 }
 706                 put_namespace(namespace);
 707         }
 708         return ret;
 709 }
 710
 711 static int mounts_release(struct inode *inode, struct file *file)
 712 {
 713         struct seq_file *m = file->private_data;
 714         struct namespace *namespace = m->private;
 715         put_namespace(namespace);
 716         return seq_release(inode, file);
 717 }
 718
 719 static unsigned mounts_poll(struct file *file, poll_table *wait)
 720 {
 721         struct proc_mounts *p = file->private_data;
 722         struct namespace *ns = p->m.private;
 723         unsigned res = 0;
 724
 725         poll_wait(file, &ns->poll, wait);
 726
 727         spin_lock(&vfsmount_lock);
 728         if (p->event != ns->event) {
 729                 p->event = ns->event;
 730                 res = POLLERR;
 731         }
 732         spin_unlock(&vfsmount_lock);
 733
 734         return res;
 735 }
 736
 737 static struct file_operations proc_mounts_operations = {
 738         .open           = mounts_open,
 739         .read           = seq_read,
 740         .llseek         = seq_lseek,
 741         .release        = mounts_release,
 742         .poll           = mounts_poll,
 743 };
 744
 745 extern struct seq_operations mountstats_op;
 746 static int mountstats_open(struct inode *inode, struct file *file)
 747 {
 748         struct task_struct *task = proc_task(inode);
 749         int ret = seq_open(file, &mountstats_op);
 750
 751         if (!ret) {
 752                 struct seq_file *m = file->private_data;
 753                 struct namespace *namespace;
 754                 task_lock(task);
 755                 namespace = task->namespace;
 756                 if (namespace)
 757                         get_namespace(namespace);
 758                 task_unlock(task);
 759
 760                 if (namespace)
 761                         m->private = namespace;
 762                 else {
 763                         seq_release(inode, file);
 764                         ret = -EINVAL;
 765                 }
 766         }
 767         return ret;
 768 }
 769
 770 static struct file_operations proc_mountstats_operations = {
 771         .open           = mountstats_open,
 772         .read           = seq_read,
 773         .llseek         = seq_lseek,
 774         .release        = mounts_release,
 775 };
 776
 777 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 778
 779 static ssize_t proc_info_read(struct file * file, char __user * buf,
 780                           size_t count, loff_t *ppos)
 781 {
 782         struct inode * inode = file->f_dentry->d_inode;
 783         unsigned long page;
 784         ssize_t length;
 785         struct task_struct *task = proc_task(inode);
 786
 787         if (count > PROC_BLOCK_SIZE)
 788                 count = PROC_BLOCK_SIZE;
 789         if (!(page = __get_free_page(GFP_KERNEL)))
 790                 return -ENOMEM;
 791
 792         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 793
 794         if (length >= 0)
 795                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 796         free_page(page);
 797         return length;
 798 }
 799
 800 static struct file_operations proc_info_file_operations = {
 801         .read           = proc_info_read,
 802 };
 803
 804 static int mem_open(struct inode* inode, struct file* file)
 805 {
 806         file->private_data = (void*)((long)current->self_exec_id);
 807         return 0;
 808 }
 809
 810 static ssize_t mem_read(struct file * file, char __user * buf,
 811                         size_t count, loff_t *ppos)
 812 {
 813         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 814         char *page;
 815         unsigned long src = *ppos;
 816         int ret = -ESRCH;
 817         struct mm_struct *mm;
 818
 819         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 820                 goto out;
 821
 822         ret = -ENOMEM;
 823         page = (char *)__get_free_page(GFP_USER);
 824         if (!page)
 825                 goto out;
 826
 827         ret = 0;
 828
 829         mm = get_task_mm(task);
 830         if (!mm)
 831                 goto out_free;
 832
 833         ret = -EIO;
 834
 835         if (file->private_data != (void*)((long)current->self_exec_id))
 836                 goto out_put;
 837
 838         ret = 0;
 839
 840         while (count > 0) {
 841                 int this_len, retval;
 842
 843                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 844                 retval = access_process_vm(task, src, page, this_len, 0);
 845                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 846                         if (!ret)
 847                                 ret = -EIO;
 848                         break;
 849                 }
 850
 851                 if (copy_to_user(buf, page, retval)) {
 852                         ret = -EFAULT;
 853                         break;
 854                 }
 855
 856                 ret += retval;
 857                 src += retval;
 858                 buf += retval;
 859                 count -= retval;
 860         }
 861         *ppos = src;
 862
 863 out_put:
 864         mmput(mm);
 865 out_free:
 866         free_page((unsigned long) page);
 867 out:
 868         return ret;
 869 }
 870
 871 #define mem_write NULL
 872
 873 #ifndef mem_write
 874 /* This is a security hazard */
 875 static ssize_t mem_write(struct file * file, const char * buf,
 876                          size_t count, loff_t *ppos)
 877 {
 878         int copied = 0;
 879         char *page;
 880         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 881         unsigned long dst = *ppos;
 882
 883         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 884                 return -ESRCH;
 885
 886         page = (char *)__get_free_page(GFP_USER);
 887         if (!page)
 888                 return -ENOMEM;
 889
 890         while (count > 0) {
 891                 int this_len, retval;
 892
 893                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 894                 if (copy_from_user(page, buf, this_len)) {
 895                         copied = -EFAULT;
 896                         break;
 897                 }
 898                 retval = access_process_vm(task, dst, page, this_len, 1);
 899                 if (!retval) {
 900                         if (!copied)
 901                                 copied = -EIO;
 902                         break;
 903                 }
 904                 copied += retval;
 905                 buf += retval;
 906                 dst += retval;
 907                 count -= retval;
 908         }
 909         *ppos = dst;
 910         free_page((unsigned long) page);
 911         return copied;
 912 }
 913 #endif
 914
 915 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 916 {
 917         switch (orig) {
 918         case 0:
 919                 file->f_pos = offset;
 920                 break;
 921         case 1:
 922                 file->f_pos += offset;
 923                 break;
 924         default:
 925                 return -EINVAL;
 926         }
 927         force_successful_syscall_return();
 928         return file->f_pos;
 929 }
 930
 931 static struct file_operations proc_mem_operations = {
 932         .llseek         = mem_lseek,
 933         .read           = mem_read,
 934         .write          = mem_write,
 935         .open           = mem_open,
 936 };
 937
 938 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 939                                 size_t count, loff_t *ppos)
 940 {
 941         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 942         char buffer[8];
 943         size_t len;
 944         int oom_adjust = task->oomkilladj;
 945         loff_t __ppos = *ppos;
 946
 947         len = sprintf(buffer, "%i\n", oom_adjust);
 948         if (__ppos >= len)
 949                 return 0;
 950         if (count > len-__ppos)
 951                 count = len-__ppos;
 952         if (copy_to_user(buf, buffer + __ppos, count))
 953                 return -EFAULT;
 954         *ppos = __ppos + count;
 955         return count;
 956 }
 957
 958 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 959                                 size_t count, loff_t *ppos)
 960 {
 961         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 962         char buffer[8], *end;
 963         int oom_adjust;
 964
 965         if (!capable(CAP_SYS_RESOURCE))
 966                 return -EPERM;
 967         memset(buffer, 0, 8);
 968         if (count > 6)
 969                 count = 6;
 970         if (copy_from_user(buffer, buf, count))
 971                 return -EFAULT;
 972         oom_adjust = simple_strtol(buffer, &end, 0);
 973         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 974                 return -EINVAL;
 975         if (*end == '\n')
 976                 end++;
 977         task->oomkilladj = oom_adjust;
 978         if (end - buffer == 0)
 979                 return -EIO;
 980         return end - buffer;
 981 }
 982
 983 static struct file_operations proc_oom_adjust_operations = {
 984         .read           = oom_adjust_read,
 985         .write          = oom_adjust_write,
 986 };
 987
 988 static struct inode_operations proc_mem_inode_operations = {
 989         .permission     = proc_permission,
 990 };
 991
 992 #ifdef CONFIG_AUDITSYSCALL
 993 #define TMPBUFLEN 21
 994 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 995                                   size_t count, loff_t *ppos)
 996 {
 997         struct inode * inode = file->f_dentry->d_inode;
 998         struct task_struct *task = proc_task(inode);
 999         ssize_t length;
1000         char tmpbuf[TMPBUFLEN];
1001
1002         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1003                                 audit_get_loginuid(task->audit_context));
1004         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1005 }
1006
1007 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1008                                    size_t count, loff_t *ppos)
1009 {
1010         struct inode * inode = file->f_dentry->d_inode;
1011         char *page, *tmp;
1012         ssize_t length;
1013         struct task_struct *task = proc_task(inode);
1014         uid_t loginuid;
1015
1016         if (!capable(CAP_AUDIT_CONTROL))
1017                 return -EPERM;
1018
1019         if (current != task)
1020                 return -EPERM;
1021
1022         if (count >= PAGE_SIZE)
1023                 count = PAGE_SIZE - 1;
1024
1025         if (*ppos != 0) {
1026                 /* No partial writes. */
1027                 return -EINVAL;
1028         }
1029         page = (char*)__get_free_page(GFP_USER);
1030         if (!page)
1031                 return -ENOMEM;
1032         length = -EFAULT;
1033         if (copy_from_user(page, buf, count))
1034                 goto out_free_page;
1035
1036         page[count] = '\0';
1037         loginuid = simple_strtoul(page, &tmp, 10);
1038         if (tmp == page) {
1039                 length = -EINVAL;
1040                 goto out_free_page;
1041
1042         }
1043         length = audit_set_loginuid(task, loginuid);
1044         if (likely(length == 0))
1045                 length = count;
1046
1047 out_free_page:
1048         free_page((unsigned long) page);
1049         return length;
1050 }
1051
1052 static struct file_operations proc_loginuid_operations = {
1053         .read           = proc_loginuid_read,
1054         .write          = proc_loginuid_write,
1055 };
1056 #endif
1057
1058 #ifdef CONFIG_SECCOMP
1059 static ssize_t seccomp_read(struct file *file, char __user *buf,
1060                             size_t count, loff_t *ppos)
1061 {
1062         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1063         char __buf[20];
1064         loff_t __ppos = *ppos;
1065         size_t len;
1066
1067         /* no need to print the trailing zero, so use only len */
1068         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
1069         if (__ppos >= len)
1070                 return 0;
1071         if (count > len - __ppos)
1072                 count = len - __ppos;
1073         if (copy_to_user(buf, __buf + __ppos, count))
1074                 return -EFAULT;
1075         *ppos = __ppos + count;
1076         return count;
1077 }
1078
1079 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1080                              size_t count, loff_t *ppos)
1081 {
1082         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1083         char __buf[20], *end;
1084         unsigned int seccomp_mode;
1085
1086         /* can set it only once to be even more secure */
1087         if (unlikely(tsk->seccomp.mode))
1088                 return -EPERM;
1089
1090         memset(__buf, 0, sizeof(__buf));
1091         count = min(count, sizeof(__buf) - 1);
1092         if (copy_from_user(__buf, buf, count))
1093                 return -EFAULT;
1094         seccomp_mode = simple_strtoul(__buf, &end, 0);
1095         if (*end == '\n')
1096                 end++;
1097         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1098                 tsk->seccomp.mode = seccomp_mode;
1099                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1100         } else
1101                 return -EINVAL;
1102         if (unlikely(!(end - __buf)))
1103                 return -EIO;
1104         return end - __buf;
1105 }
1106
1107 static struct file_operations proc_seccomp_operations = {
1108         .read           = seccomp_read,
1109         .write          = seccomp_write,
1110 };
1111 #endif /* CONFIG_SECCOMP */
1112
1113 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1114 {
1115         struct inode *inode = dentry->d_inode;
1116         int error = -EACCES;
1117
1118         /* We don't need a base pointer in the /proc filesystem */
1119         path_release(nd);
1120
1121         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1122                 goto out;
1123         error = proc_check_root(inode);
1124         if (error)
1125                 goto out;
1126
1127         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1128         nd->last_type = LAST_BIND;
1129 out:
1130         return ERR_PTR(error);
1131 }
1132
1133 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1134                             char __user *buffer, int buflen)
1135 {
1136         struct inode * inode;
1137         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1138         int len;
1139
1140         if (!tmp)
1141                 return -ENOMEM;
1142
1143         inode = dentry->d_inode;
1144         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1145         len = PTR_ERR(path);
1146         if (IS_ERR(path))
1147                 goto out;
1148         len = tmp + PAGE_SIZE - 1 - path;
1149
1150         if (len > buflen)
1151                 len = buflen;
1152         if (copy_to_user(buffer, path, len))
1153                 len = -EFAULT;
1154  out:
1155         free_page((unsigned long)tmp);
1156         return len;
1157 }
1158
1159 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1160 {
1161         int error = -EACCES;
1162         struct inode *inode = dentry->d_inode;
1163         struct dentry *de;
1164         struct vfsmount *mnt = NULL;
1165
1166         lock_kernel();
1167
1168         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1169                 goto out;
1170         error = proc_check_root(inode);
1171         if (error)
1172                 goto out;
1173
1174         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1175         if (error)
1176                 goto out;
1177
1178         error = do_proc_readlink(de, mnt, buffer, buflen);
1179         dput(de);
1180         mntput(mnt);
1181 out:
1182         unlock_kernel();
1183         return error;
1184 }
1185
1186 static struct inode_operations proc_pid_link_inode_operations = {
1187         .readlink       = proc_pid_readlink,
1188         .follow_link    = proc_pid_follow_link
1189 };
1190
1191 #define NUMBUF 10
1192
1193 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1194 {
1195         struct inode *inode = filp->f_dentry->d_inode;
1196         struct task_struct *p = proc_task(inode);
1197         unsigned int fd, tid, ino;
1198         int retval;
1199         char buf[NUMBUF];
1200         struct files_struct * files;
1201         struct fdtable *fdt;
1202
1203         retval = -ENOENT;
1204         if (!pid_alive(p))
1205                 goto out;
1206         retval = 0;
1207         tid = p->pid;
1208
1209         fd = filp->f_pos;
1210         switch (fd) {
1211                 case 0:
1212                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1213                                 goto out;
1214                         filp->f_pos++;
1215                 case 1:
1216                         ino = fake_ino(tid, PROC_TID_INO);
1217                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1218                                 goto out;
1219                         filp->f_pos++;
1220                 default:
1221                         files = get_files_struct(p);
1222                         if (!files)
1223                                 goto out;
1224                         rcu_read_lock();
1225                         fdt = files_fdtable(files);
1226                         for (fd = filp->f_pos-2;
1227                              fd < fdt->max_fds;
1228                              fd++, filp->f_pos++) {
1229                                 unsigned int i,j;
1230
1231                                 if (!fcheck_files(files, fd))
1232                                         continue;
1233                                 rcu_read_unlock();
1234
1235                                 j = NUMBUF;
1236                                 i = fd;
1237                                 do {
1238                                         j--;
1239                                         buf[j] = '0' + (i % 10);
1240                                         i /= 10;
1241                                 } while (i);
1242
1243                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1244                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1245                                         rcu_read_lock();
1246                                         break;
1247                                 }
1248                                 rcu_read_lock();
1249                         }
1250                         rcu_read_unlock();
1251                         put_files_struct(files);
1252         }
1253 out:
1254         return retval;
1255 }
1256
1257 static int proc_pident_readdir(struct file *filp,
1258                 void *dirent, filldir_t filldir,
1259                 struct pid_entry *ents, unsigned int nents)
1260 {
1261         int i;
1262         int pid;
1263         struct dentry *dentry = filp->f_dentry;
1264         struct inode *inode = dentry->d_inode;
1265         struct pid_entry *p;
1266         ino_t ino;
1267         int ret;
1268
1269         ret = -ENOENT;
1270         if (!pid_alive(proc_task(inode)))
1271                 goto out;
1272
1273         ret = 0;
1274         pid = proc_task(inode)->pid;
1275         i = filp->f_pos;
1276         switch (i) {
1277         case 0:
1278                 ino = inode->i_ino;
1279                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1280                         goto out;
1281                 i++;
1282                 filp->f_pos++;
1283                 /* fall through */
1284         case 1:
1285                 ino = parent_ino(dentry);
1286                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1287                         goto out;
1288                 i++;
1289                 filp->f_pos++;
1290                 /* fall through */
1291         default:
1292                 i -= 2;
1293                 if (i >= nents) {
1294                         ret = 1;
1295                         goto out;
1296                 }
1297                 p = ents + i;
1298                 while (p->name) {
1299                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1300                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1301                                 goto out;
1302                         filp->f_pos++;
1303                         p++;
1304                 }
1305         }
1306
1307         ret = 1;
1308 out:
1309         return ret;
1310 }
1311
1312 static int proc_tgid_base_readdir(struct file * filp,
1313                              void * dirent, filldir_t filldir)
1314 {
1315         return proc_pident_readdir(filp,dirent,filldir,
1316                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1317 }
1318
1319 static int proc_tid_base_readdir(struct file * filp,
1320                              void * dirent, filldir_t filldir)
1321 {
1322         return proc_pident_readdir(filp,dirent,filldir,
1323                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1324 }
1325
1326 /* building an inode */
1327
1328 static int task_dumpable(struct task_struct *task)
1329 {
1330         int dumpable = 0;
1331         struct mm_struct *mm;
1332
1333         task_lock(task);
1334         mm = task->mm;
1335         if (mm)
1336                 dumpable = mm->dumpable;
1337         task_unlock(task);
1338         if(dumpable == 1)
1339                 return 1;
1340         return 0;
1341 }
1342
1343
1344 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1345 {
1346         struct inode * inode;
1347         struct proc_inode *ei;
1348
1349         /* We need a new inode */
1350
1351         inode = new_inode(sb);
1352         if (!inode)
1353                 goto out;
1354
1355         /* Common stuff */
1356         ei = PROC_I(inode);
1357         ei->task = NULL;
1358         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1359         inode->i_ino = fake_ino(task->pid, ino);
1360
1361         if (!pid_alive(task))
1362                 goto out_unlock;
1363
1364         /*
1365          * grab the reference to task.
1366          */
1367         get_task_struct(task);
1368         ei->task = task;
1369         ei->type = ino;
1370         inode->i_uid = 0;
1371         inode->i_gid = 0;
1372         if (ino == PROC_TGID_INO || ino == PROC_TID_INO || task_dumpable(task)) {
1373                 inode->i_uid = task->euid;
1374                 inode->i_gid = task->egid;
1375         }
1376         security_task_to_inode(task, inode);
1377
1378 out:
1379         return inode;
1380
1381 out_unlock:
1382         ei->pde = NULL;
1383         iput(inode);
1384         return NULL;
1385 }
1386
1387 /* dentry stuff */
1388
1389 /*
1390  *      Exceptional case: normally we are not allowed to unhash a busy
1391  * directory. In this case, however, we can do it - no aliasing problems
1392  * due to the way we treat inodes.
1393  *
1394  * Rewrite the inode's ownerships here because the owning task may have
1395  * performed a setuid(), etc.
1396  */
1397 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1398 {
1399         struct inode *inode = dentry->d_inode;
1400         struct task_struct *task = proc_task(inode);
1401         if (pid_alive(task)) {
1402                 if (proc_type(inode) == PROC_TGID_INO || proc_type(inode) == PROC_TID_INO || task_dumpable(task)) {
1403                         inode->i_uid = task->euid;
1404                         inode->i_gid = task->egid;
1405                 } else {
1406                         inode->i_uid = 0;
1407                         inode->i_gid = 0;
1408                 }
1409                 security_task_to_inode(task, inode);
1410                 return 1;
1411         }
1412         d_drop(dentry);
1413         return 0;
1414 }
1415
1416 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1417 {
1418         struct inode *inode = dentry->d_inode;
1419         struct task_struct *task = proc_task(inode);
1420         int fd = proc_type(inode) - PROC_TID_FD_DIR;
1421         struct files_struct *files;
1422
1423         files = get_files_struct(task);
1424         if (files) {
1425                 rcu_read_lock();
1426                 if (fcheck_files(files, fd)) {
1427                         rcu_read_unlock();
1428                         put_files_struct(files);
1429                         if (task_dumpable(task)) {
1430                                 inode->i_uid = task->euid;
1431                                 inode->i_gid = task->egid;
1432                         } else {
1433                                 inode->i_uid = 0;
1434                                 inode->i_gid = 0;
1435                         }
1436                         security_task_to_inode(task, inode);
1437                         return 1;
1438                 }
1439                 rcu_read_unlock();
1440                 put_files_struct(files);
1441         }
1442         d_drop(dentry);
1443         return 0;
1444 }
1445
1446 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1447 {
1448         struct task_struct *task = proc_task(inode);
1449         spin_lock(&task->proc_lock);
1450         if (task->proc_dentry == dentry)
1451                 task->proc_dentry = NULL;
1452         spin_unlock(&task->proc_lock);
1453         iput(inode);
1454 }
1455
1456 static int pid_delete_dentry(struct dentry * dentry)
1457 {
1458         /* Is the task we represent dead?
1459          * If so, then don't put the dentry on the lru list,
1460          * kill it immediately.
1461          */
1462         return !pid_alive(proc_task(dentry->d_inode));
1463 }
1464
1465 static struct dentry_operations tid_fd_dentry_operations =
1466 {
1467         .d_revalidate   = tid_fd_revalidate,
1468         .d_delete       = pid_delete_dentry,
1469 };
1470
1471 static struct dentry_operations pid_dentry_operations =
1472 {
1473         .d_revalidate   = pid_revalidate,
1474         .d_delete       = pid_delete_dentry,
1475 };
1476
1477 static struct dentry_operations pid_base_dentry_operations =
1478 {
1479         .d_revalidate   = pid_revalidate,
1480         .d_iput         = pid_base_iput,
1481         .d_delete       = pid_delete_dentry,
1482 };
1483
1484 /* Lookups */
1485
1486 static unsigned name_to_int(struct dentry *dentry)
1487 {
1488         const char *name = dentry->d_name.name;
1489         int len = dentry->d_name.len;
1490         unsigned n = 0;
1491
1492         if (len > 1 && *name == '0')
1493                 goto out;
1494         while (len-- > 0) {
1495                 unsigned c = *name++ - '0';
1496                 if (c > 9)
1497                         goto out;
1498                 if (n >= (~0U-9)/10)
1499                         goto out;
1500                 n *= 10;
1501                 n += c;
1502         }
1503         return n;
1504 out:
1505         return ~0U;
1506 }
1507
1508 /* SMP-safe */
1509 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1510 {
1511         struct task_struct *task = proc_task(dir);
1512         unsigned fd = name_to_int(dentry);
1513         struct file * file;
1514         struct files_struct * files;
1515         struct inode *inode;
1516         struct proc_inode *ei;
1517
1518         if (fd == ~0U)
1519                 goto out;
1520         if (!pid_alive(task))
1521                 goto out;
1522
1523         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1524         if (!inode)
1525                 goto out;
1526         ei = PROC_I(inode);
1527         files = get_files_struct(task);
1528         if (!files)
1529                 goto out_unlock;
1530         inode->i_mode = S_IFLNK;
1531
1532         /*
1533          * We are not taking a ref to the file structure, so we must
1534          * hold ->file_lock.
1535          */
1536         spin_lock(&files->file_lock);
1537         file = fcheck_files(files, fd);
1538         if (!file)
1539                 goto out_unlock2;
1540         if (file->f_mode & 1)
1541                 inode->i_mode |= S_IRUSR | S_IXUSR;
1542         if (file->f_mode & 2)
1543                 inode->i_mode |= S_IWUSR | S_IXUSR;
1544         spin_unlock(&files->file_lock);
1545         put_files_struct(files);
1546         inode->i_op = &proc_pid_link_inode_operations;
1547         inode->i_size = 64;
1548         ei->op.proc_get_link = proc_fd_link;
1549         dentry->d_op = &tid_fd_dentry_operations;
1550         d_add(dentry, inode);
1551         return NULL;
1552
1553 out_unlock2:
1554         spin_unlock(&files->file_lock);
1555         put_files_struct(files);
1556 out_unlock:
1557         iput(inode);
1558 out:
1559         return ERR_PTR(-ENOENT);
1560 }
1561
1562 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1563 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1564
1565 static struct file_operations proc_fd_operations = {
1566         .read           = generic_read_dir,
1567         .readdir        = proc_readfd,
1568 };
1569
1570 static struct file_operations proc_task_operations = {
1571         .read           = generic_read_dir,
1572         .readdir        = proc_task_readdir,
1573 };
1574
1575 /*
1576  * proc directories can do almost nothing..
1577  */
1578 static struct inode_operations proc_fd_inode_operations = {
1579         .lookup         = proc_lookupfd,
1580         .permission     = proc_permission,
1581 };
1582
1583 static struct inode_operations proc_task_inode_operations = {
1584         .lookup         = proc_task_lookup,
1585         .permission     = proc_task_permission,
1586 };
1587
1588 #ifdef CONFIG_SECURITY
1589 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1590                                   size_t count, loff_t *ppos)
1591 {
1592         struct inode * inode = file->f_dentry->d_inode;
1593         unsigned long page;
1594         ssize_t length;
1595         struct task_struct *task = proc_task(inode);
1596
1597         if (count > PAGE_SIZE)
1598                 count = PAGE_SIZE;
1599         if (!(page = __get_free_page(GFP_KERNEL)))
1600                 return -ENOMEM;
1601
1602         length = security_getprocattr(task,
1603                                       (char*)file->f_dentry->d_name.name,
1604                                       (void*)page, count);
1605         if (length >= 0)
1606                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1607         free_page(page);
1608         return length;
1609 }
1610
1611 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1612                                    size_t count, loff_t *ppos)
1613 {
1614         struct inode * inode = file->f_dentry->d_inode;
1615         char *page;
1616         ssize_t length;
1617         struct task_struct *task = proc_task(inode);
1618
1619         if (count > PAGE_SIZE)
1620                 count = PAGE_SIZE;
1621         if (*ppos != 0) {
1622                 /* No partial writes. */
1623                 return -EINVAL;
1624         }
1625         page = (char*)__get_free_page(GFP_USER);
1626         if (!page)
1627                 return -ENOMEM;
1628         length = -EFAULT;
1629         if (copy_from_user(page, buf, count))
1630                 goto out;
1631
1632         length = security_setprocattr(task,
1633                                       (char*)file->f_dentry->d_name.name,
1634                                       (void*)page, count);
1635 out:
1636         free_page((unsigned long) page);
1637         return length;
1638 }
1639
1640 static struct file_operations proc_pid_attr_operations = {
1641         .read           = proc_pid_attr_read,
1642         .write          = proc_pid_attr_write,
1643 };
1644
1645 static struct file_operations proc_tid_attr_operations;
1646 static struct inode_operations proc_tid_attr_inode_operations;
1647 static struct file_operations proc_tgid_attr_operations;
1648 static struct inode_operations proc_tgid_attr_inode_operations;
1649 #endif
1650
1651 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1652
1653 /* SMP-safe */
1654 static struct dentry *proc_pident_lookup(struct inode *dir,
1655                                          struct dentry *dentry,
1656                                          struct pid_entry *ents)
1657 {
1658         struct inode *inode;
1659         int error;
1660         struct task_struct *task = proc_task(dir);
1661         struct pid_entry *p;
1662         struct proc_inode *ei;
1663
1664         error = -ENOENT;
1665         inode = NULL;
1666
1667         if (!pid_alive(task))
1668                 goto out;
1669
1670         for (p = ents; p->name; p++) {
1671                 if (p->len != dentry->d_name.len)
1672                         continue;
1673                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1674                         break;
1675         }
1676         if (!p->name)
1677                 goto out;
1678
1679         error = -EINVAL;
1680         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1681         if (!inode)
1682                 goto out;
1683
1684         ei = PROC_I(inode);
1685         inode->i_mode = p->mode;
1686         /*
1687          * Yes, it does not scale. And it should not. Don't add
1688          * new entries into /proc/<tgid>/ without very good reasons.
1689          */
1690         switch(p->type) {
1691                 case PROC_TGID_TASK:
1692                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1693                         inode->i_op = &proc_task_inode_operations;
1694                         inode->i_fop = &proc_task_operations;
1695                         break;
1696                 case PROC_TID_FD:
1697                 case PROC_TGID_FD:
1698                         inode->i_nlink = 2;
1699                         inode->i_op = &proc_fd_inode_operations;
1700                         inode->i_fop = &proc_fd_operations;
1701                         break;
1702                 case PROC_TID_EXE:
1703                 case PROC_TGID_EXE:
1704                         inode->i_op = &proc_pid_link_inode_operations;
1705                         ei->op.proc_get_link = proc_exe_link;
1706                         break;
1707                 case PROC_TID_CWD:
1708                 case PROC_TGID_CWD:
1709                         inode->i_op = &proc_pid_link_inode_operations;
1710                         ei->op.proc_get_link = proc_cwd_link;
1711                         break;
1712                 case PROC_TID_ROOT:
1713                 case PROC_TGID_ROOT:
1714                         inode->i_op = &proc_pid_link_inode_operations;
1715                         ei->op.proc_get_link = proc_root_link;
1716                         break;
1717                 case PROC_TID_ENVIRON:
1718                 case PROC_TGID_ENVIRON:
1719                         inode->i_fop = &proc_info_file_operations;
1720                         ei->op.proc_read = proc_pid_environ;
1721                         break;
1722                 case PROC_TID_AUXV:
1723                 case PROC_TGID_AUXV:
1724                         inode->i_fop = &proc_info_file_operations;
1725                         ei->op.proc_read = proc_pid_auxv;
1726                         break;
1727                 case PROC_TID_STATUS:
1728                 case PROC_TGID_STATUS:
1729                         inode->i_fop = &proc_info_file_operations;
1730                         ei->op.proc_read = proc_pid_status;
1731                         break;
1732                 case PROC_TID_STAT:
1733                         inode->i_fop = &proc_info_file_operations;
1734                         ei->op.proc_read = proc_tid_stat;
1735                         break;
1736                 case PROC_TGID_STAT:
1737                         inode->i_fop = &proc_info_file_operations;
1738                         ei->op.proc_read = proc_tgid_stat;
1739                         break;
1740                 case PROC_TID_CMDLINE:
1741                 case PROC_TGID_CMDLINE:
1742                         inode->i_fop = &proc_info_file_operations;
1743                         ei->op.proc_read = proc_pid_cmdline;
1744                         break;
1745                 case PROC_TID_STATM:
1746                 case PROC_TGID_STATM:
1747                         inode->i_fop = &proc_info_file_operations;
1748                         ei->op.proc_read = proc_pid_statm;
1749                         break;
1750                 case PROC_TID_MAPS:
1751                 case PROC_TGID_MAPS:
1752                         inode->i_fop = &proc_maps_operations;
1753                         break;
1754 #ifdef CONFIG_NUMA
1755                 case PROC_TID_NUMA_MAPS:
1756                 case PROC_TGID_NUMA_MAPS:
1757                         inode->i_fop = &proc_numa_maps_operations;
1758                         break;
1759 #endif
1760                 case PROC_TID_MEM:
1761                 case PROC_TGID_MEM:
1762                         inode->i_op = &proc_mem_inode_operations;
1763                         inode->i_fop = &proc_mem_operations;
1764                         break;
1765 #ifdef CONFIG_SECCOMP
1766                 case PROC_TID_SECCOMP:
1767                 case PROC_TGID_SECCOMP:
1768                         inode->i_fop = &proc_seccomp_operations;
1769                         break;
1770 #endif /* CONFIG_SECCOMP */
1771                 case PROC_TID_MOUNTS:
1772                 case PROC_TGID_MOUNTS:
1773                         inode->i_fop = &proc_mounts_operations;
1774                         break;
1775 #ifdef CONFIG_MMU
1776                 case PROC_TID_SMAPS:
1777                 case PROC_TGID_SMAPS:
1778                         inode->i_fop = &proc_smaps_operations;
1779                         break;
1780 #endif
1781                 case PROC_TID_MOUNTSTATS:
1782                 case PROC_TGID_MOUNTSTATS:
1783                         inode->i_fop = &proc_mountstats_operations;
1784                         break;
1785 #ifdef CONFIG_SECURITY
1786                 case PROC_TID_ATTR:
1787                         inode->i_nlink = 2;
1788                         inode->i_op = &proc_tid_attr_inode_operations;
1789                         inode->i_fop = &proc_tid_attr_operations;
1790                         break;
1791                 case PROC_TGID_ATTR:
1792                         inode->i_nlink = 2;
1793                         inode->i_op = &proc_tgid_attr_inode_operations;
1794                         inode->i_fop = &proc_tgid_attr_operations;
1795                         break;
1796                 case PROC_TID_ATTR_CURRENT:
1797                 case PROC_TGID_ATTR_CURRENT:
1798                 case PROC_TID_ATTR_PREV:
1799                 case PROC_TGID_ATTR_PREV:
1800                 case PROC_TID_ATTR_EXEC:
1801                 case PROC_TGID_ATTR_EXEC:
1802                 case PROC_TID_ATTR_FSCREATE:
1803                 case PROC_TGID_ATTR_FSCREATE:
1804                         inode->i_fop = &proc_pid_attr_operations;
1805                         break;
1806 #endif
1807 #ifdef CONFIG_KALLSYMS
1808                 case PROC_TID_WCHAN:
1809                 case PROC_TGID_WCHAN:
1810                         inode->i_fop = &proc_info_file_operations;
1811                         ei->op.proc_read = proc_pid_wchan;
1812                         break;
1813 #endif
1814 #ifdef CONFIG_SCHEDSTATS
1815                 case PROC_TID_SCHEDSTAT:
1816                 case PROC_TGID_SCHEDSTAT:
1817                         inode->i_fop = &proc_info_file_operations;
1818                         ei->op.proc_read = proc_pid_schedstat;
1819                         break;
1820 #endif
1821 #ifdef CONFIG_CPUSETS
1822                 case PROC_TID_CPUSET:
1823                 case PROC_TGID_CPUSET:
1824                         inode->i_fop = &proc_cpuset_operations;
1825                         break;
1826 #endif
1827                 case PROC_TID_OOM_SCORE:
1828                 case PROC_TGID_OOM_SCORE:
1829                         inode->i_fop = &proc_info_file_operations;
1830                         ei->op.proc_read = proc_oom_score;
1831                         break;
1832                 case PROC_TID_OOM_ADJUST:
1833                 case PROC_TGID_OOM_ADJUST:
1834                         inode->i_fop = &proc_oom_adjust_operations;
1835                         break;
1836 #ifdef CONFIG_AUDITSYSCALL
1837                 case PROC_TID_LOGINUID:
1838                 case PROC_TGID_LOGINUID:
1839                         inode->i_fop = &proc_loginuid_operations;
1840                         break;
1841 #endif
1842                 default:
1843                         printk("procfs: impossible type (%d)",p->type);
1844                         iput(inode);
1845                         return ERR_PTR(-EINVAL);
1846         }
1847         dentry->d_op = &pid_dentry_operations;
1848         d_add(dentry, inode);
1849         return NULL;
1850
1851 out:
1852         return ERR_PTR(error);
1853 }
1854
1855 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1856         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1857 }
1858
1859 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1860         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1861 }
1862
1863 static struct file_operations proc_tgid_base_operations = {
1864         .read           = generic_read_dir,
1865         .readdir        = proc_tgid_base_readdir,
1866 };
1867
1868 static struct file_operations proc_tid_base_operations = {
1869         .read           = generic_read_dir,
1870         .readdir        = proc_tid_base_readdir,
1871 };
1872
1873 static struct inode_operations proc_tgid_base_inode_operations = {
1874         .lookup         = proc_tgid_base_lookup,
1875 };
1876
1877 static struct inode_operations proc_tid_base_inode_operations = {
1878         .lookup         = proc_tid_base_lookup,
1879 };
1880
1881 #ifdef CONFIG_SECURITY
1882 static int proc_tgid_attr_readdir(struct file * filp,
1883                              void * dirent, filldir_t filldir)
1884 {
1885         return proc_pident_readdir(filp,dirent,filldir,
1886                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1887 }
1888
1889 static int proc_tid_attr_readdir(struct file * filp,
1890                              void * dirent, filldir_t filldir)
1891 {
1892         return proc_pident_readdir(filp,dirent,filldir,
1893                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1894 }
1895
1896 static struct file_operations proc_tgid_attr_operations = {
1897         .read           = generic_read_dir,
1898         .readdir        = proc_tgid_attr_readdir,
1899 };
1900
1901 static struct file_operations proc_tid_attr_operations = {
1902         .read           = generic_read_dir,
1903         .readdir        = proc_tid_attr_readdir,
1904 };
1905
1906 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1907                                 struct dentry *dentry, struct nameidata *nd)
1908 {
1909         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1910 }
1911
1912 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1913                                 struct dentry *dentry, struct nameidata *nd)
1914 {
1915         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1916 }
1917
1918 static struct inode_operations proc_tgid_attr_inode_operations = {
1919         .lookup         = proc_tgid_attr_lookup,
1920 };
1921
1922 static struct inode_operations proc_tid_attr_inode_operations = {
1923         .lookup         = proc_tid_attr_lookup,
1924 };
1925 #endif
1926
1927 /*
1928  * /proc/self:
1929  */
1930 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1931                               int buflen)
1932 {
1933         char tmp[30];
1934         sprintf(tmp, "%d", current->tgid);
1935         return vfs_readlink(dentry,buffer,buflen,tmp);
1936 }
1937
1938 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1939 {
1940         char tmp[30];
1941         sprintf(tmp, "%d", current->tgid);
1942         return ERR_PTR(vfs_follow_link(nd,tmp));
1943 }
1944
1945 static struct inode_operations proc_self_inode_operations = {
1946         .readlink       = proc_self_readlink,
1947         .follow_link    = proc_self_follow_link,
1948 };
1949
1950 /**
1951  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1952  * @p: task that should be flushed.
1953  *
1954  * Drops the /proc/@pid dcache entry from the hash chains.
1955  *
1956  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1957  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1958  * if the pid value is immediately reused. This is enforced by
1959  * - caller must acquire spin_lock(p->proc_lock)
1960  * - must be called before detach_pid()
1961  * - proc_pid_lookup acquires proc_lock, and checks that
1962  *   the target is not dead by looking at the attach count
1963  *   of PIDTYPE_PID.
1964  */
1965
1966 struct dentry *proc_pid_unhash(struct task_struct *p)
1967 {
1968         struct dentry *proc_dentry;
1969
1970         proc_dentry = p->proc_dentry;
1971         if (proc_dentry != NULL) {
1972
1973                 spin_lock(&dcache_lock);
1974                 spin_lock(&proc_dentry->d_lock);
1975                 if (!d_unhashed(proc_dentry)) {
1976                         dget_locked(proc_dentry);
1977                         __d_drop(proc_dentry);
1978                         spin_unlock(&proc_dentry->d_lock);
1979                 } else {
1980                         spin_unlock(&proc_dentry->d_lock);
1981                         proc_dentry = NULL;
1982                 }
1983                 spin_unlock(&dcache_lock);
1984         }
1985         return proc_dentry;
1986 }
1987
1988 /**
1989  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1990  * @proc_dentry: directoy to prune.
1991  *
1992  * Shrink the /proc directory that was used by the just killed thread.
1993  */
1994
1995 void proc_pid_flush(struct dentry *proc_dentry)
1996 {
1997         might_sleep();
1998         if(proc_dentry != NULL) {
1999                 shrink_dcache_parent(proc_dentry);
2000                 dput(proc_dentry);
2001         }
2002 }
2003
2004 /* SMP-safe */
2005 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2006 {
2007         struct task_struct *task;
2008         struct inode *inode;
2009         struct proc_inode *ei;
2010         unsigned tgid;
2011         int died;
2012
2013         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
2014                 inode = new_inode(dir->i_sb);
2015                 if (!inode)
2016                         return ERR_PTR(-ENOMEM);
2017                 ei = PROC_I(inode);
2018                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2019                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
2020                 ei->pde = NULL;
2021                 inode->i_mode = S_IFLNK|S_IRWXUGO;
2022                 inode->i_uid = inode->i_gid = 0;
2023                 inode->i_size = 64;
2024                 inode->i_op = &proc_self_inode_operations;
2025                 d_add(dentry, inode);
2026                 return NULL;
2027         }
2028         tgid = name_to_int(dentry);
2029         if (tgid == ~0U)
2030                 goto out;
2031
2032         read_lock(&tasklist_lock);
2033         task = find_task_by_pid(tgid);
2034         if (task)
2035                 get_task_struct(task);
2036         read_unlock(&tasklist_lock);
2037         if (!task)
2038                 goto out;
2039
2040         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2041
2042
2043         if (!inode) {
2044                 put_task_struct(task);
2045                 goto out;
2046         }
2047         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2048         inode->i_op = &proc_tgid_base_inode_operations;
2049         inode->i_fop = &proc_tgid_base_operations;
2050         inode->i_flags|=S_IMMUTABLE;
2051 #ifdef CONFIG_SECURITY
2052         inode->i_nlink = 5;
2053 #else
2054         inode->i_nlink = 4;
2055 #endif
2056
2057         dentry->d_op = &pid_base_dentry_operations;
2058
2059         died = 0;
2060         d_add(dentry, inode);
2061         spin_lock(&task->proc_lock);
2062         task->proc_dentry = dentry;
2063         if (!pid_alive(task)) {
2064                 dentry = proc_pid_unhash(task);
2065                 died = 1;
2066         }
2067         spin_unlock(&task->proc_lock);
2068
2069         put_task_struct(task);
2070         if (died) {
2071                 proc_pid_flush(dentry);
2072                 goto out;
2073         }
2074         return NULL;
2075 out:
2076         return ERR_PTR(-ENOENT);
2077 }
2078
2079 /* SMP-safe */
2080 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2081 {
2082         struct task_struct *task;
2083         struct task_struct *leader = proc_task(dir);
2084         struct inode *inode;
2085         unsigned tid;
2086
2087         tid = name_to_int(dentry);
2088         if (tid == ~0U)
2089                 goto out;
2090
2091         read_lock(&tasklist_lock);
2092         task = find_task_by_pid(tid);
2093         if (task)
2094                 get_task_struct(task);
2095         read_unlock(&tasklist_lock);
2096         if (!task)
2097                 goto out;
2098         if (leader->tgid != task->tgid)
2099                 goto out_drop_task;
2100
2101         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2102
2103
2104         if (!inode)
2105                 goto out_drop_task;
2106         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2107         inode->i_op = &proc_tid_base_inode_operations;
2108         inode->i_fop = &proc_tid_base_operations;
2109         inode->i_flags|=S_IMMUTABLE;
2110 #ifdef CONFIG_SECURITY
2111         inode->i_nlink = 4;
2112 #else
2113         inode->i_nlink = 3;
2114 #endif
2115
2116         dentry->d_op = &pid_base_dentry_operations;
2117
2118         d_add(dentry, inode);
2119
2120         put_task_struct(task);
2121         return NULL;
2122 out_drop_task:
2123         put_task_struct(task);
2124 out:
2125         return ERR_PTR(-ENOENT);
2126 }
2127
2128 #define PROC_NUMBUF 10
2129 #define PROC_MAXPIDS 20
2130
2131 /*
2132  * Get a few tgid's to return for filldir - we need to hold the
2133  * tasklist lock while doing this, and we must release it before
2134  * we actually do the filldir itself, so we use a temp buffer..
2135  */
2136 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2137 {
2138         struct task_struct *p;
2139         int nr_tgids = 0;
2140
2141         index--;
2142         read_lock(&tasklist_lock);
2143         p = NULL;
2144         if (version) {
2145                 p = find_task_by_pid(version);
2146                 if (p && !thread_group_leader(p))
2147                         p = NULL;
2148         }
2149
2150         if (p)
2151                 index = 0;
2152         else
2153                 p = next_task(&init_task);
2154
2155         for ( ; p != &init_task; p = next_task(p)) {
2156                 int tgid = p->pid;
2157                 if (!pid_alive(p))
2158                         continue;
2159                 if (--index >= 0)
2160                         continue;
2161                 tgids[nr_tgids] = tgid;
2162                 nr_tgids++;
2163                 if (nr_tgids >= PROC_MAXPIDS)
2164                         break;
2165         }
2166         read_unlock(&tasklist_lock);
2167         return nr_tgids;
2168 }
2169
2170 /*
2171  * Get a few tid's to return for filldir - we need to hold the
2172  * tasklist lock while doing this, and we must release it before
2173  * we actually do the filldir itself, so we use a temp buffer..
2174  */
2175 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2176 {
2177         struct task_struct *leader_task = proc_task(dir);
2178         struct task_struct *task = leader_task;
2179         int nr_tids = 0;
2180
2181         index -= 2;
2182         read_lock(&tasklist_lock);
2183         /*
2184          * The starting point task (leader_task) might be an already
2185          * unlinked task, which cannot be used to access the task-list
2186          * via next_thread().
2187          */
2188         if (pid_alive(task)) do {
2189                 int tid = task->pid;
2190
2191                 if (--index >= 0)
2192                         continue;
2193                 if (tids != NULL)
2194                         tids[nr_tids] = tid;
2195                 nr_tids++;
2196                 if (nr_tids >= PROC_MAXPIDS)
2197                         break;
2198         } while ((task = next_thread(task)) != leader_task);
2199         read_unlock(&tasklist_lock);
2200         return nr_tids;
2201 }
2202
2203 /* for the /proc/ directory itself, after non-process stuff has been done */
2204 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2205 {
2206         unsigned int tgid_array[PROC_MAXPIDS];
2207         char buf[PROC_NUMBUF];
2208         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2209         unsigned int nr_tgids, i;
2210         int next_tgid;
2211
2212         if (!nr) {
2213                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2214                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2215                         return 0;
2216                 filp->f_pos++;
2217                 nr++;
2218         }
2219
2220         /* f_version caches the tgid value that the last readdir call couldn't
2221          * return. lseek aka telldir automagically resets f_version to 0.
2222          */
2223         next_tgid = filp->f_version;
2224         filp->f_version = 0;
2225         for (;;) {
2226                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2227                 if (!nr_tgids) {
2228                         /* no more entries ! */
2229                         break;
2230                 }
2231                 next_tgid = 0;
2232
2233                 /* do not use the last found pid, reserve it for next_tgid */
2234                 if (nr_tgids == PROC_MAXPIDS) {
2235                         nr_tgids--;
2236                         next_tgid = tgid_array[nr_tgids];
2237                 }
2238
2239                 for (i=0;i<nr_tgids;i++) {
2240                         int tgid = tgid_array[i];
2241                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2242                         unsigned long j = PROC_NUMBUF;
2243
2244                         do
2245                                 buf[--j] = '0' + (tgid % 10);
2246                         while ((tgid /= 10) != 0);
2247
2248                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2249                                 /* returning this tgid failed, save it as the first
2250                                  * pid for the next readir call */
2251                                 filp->f_version = tgid_array[i];
2252                                 goto out;
2253                         }
2254                         filp->f_pos++;
2255                         nr++;
2256                 }
2257         }
2258 out:
2259         return 0;
2260 }
2261
2262 /* for the /proc/TGID/task/ directories */
2263 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2264 {
2265         unsigned int tid_array[PROC_MAXPIDS];
2266         char buf[PROC_NUMBUF];
2267         unsigned int nr_tids, i;
2268         struct dentry *dentry = filp->f_dentry;
2269         struct inode *inode = dentry->d_inode;
2270         int retval = -ENOENT;
2271         ino_t ino;
2272         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2273
2274         if (!pid_alive(proc_task(inode)))
2275                 goto out;
2276         retval = 0;
2277
2278         switch (pos) {
2279         case 0:
2280                 ino = inode->i_ino;
2281                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2282                         goto out;
2283                 pos++;
2284                 /* fall through */
2285         case 1:
2286                 ino = parent_ino(dentry);
2287                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2288                         goto out;
2289                 pos++;
2290                 /* fall through */
2291         }
2292
2293         nr_tids = get_tid_list(pos, tid_array, inode);
2294         inode->i_nlink = pos + nr_tids;
2295
2296         for (i = 0; i < nr_tids; i++) {
2297                 unsigned long j = PROC_NUMBUF;
2298                 int tid = tid_array[i];
2299
2300                 ino = fake_ino(tid,PROC_TID_INO);
2301
2302                 do
2303                         buf[--j] = '0' + (tid % 10);
2304                 while ((tid /= 10) != 0);
2305
2306                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2307                         break;
2308                 pos++;
2309         }
2310 out:
2311         filp->f_pos = pos;
2312         return retval;
2313 }