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;
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         loginuid = simple_strtoul(page, &tmp, 10);
1037         if (tmp == page) {
1038                 length = -EINVAL;
1039                 goto out_free_page;
1040
1041         }
1042         length = audit_set_loginuid(task, loginuid);
1043         if (likely(length == 0))
1044                 length = count;
1045
1046 out_free_page:
1047         free_page((unsigned long) page);
1048         return length;
1049 }
1050
1051 static struct file_operations proc_loginuid_operations = {
1052         .read           = proc_loginuid_read,
1053         .write          = proc_loginuid_write,
1054 };
1055 #endif
1056
1057 #ifdef CONFIG_SECCOMP
1058 static ssize_t seccomp_read(struct file *file, char __user *buf,
1059                             size_t count, loff_t *ppos)
1060 {
1061         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1062         char __buf[20];
1063         loff_t __ppos = *ppos;
1064         size_t len;
1065
1066         /* no need to print the trailing zero, so use only len */
1067         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
1068         if (__ppos >= len)
1069                 return 0;
1070         if (count > len - __ppos)
1071                 count = len - __ppos;
1072         if (copy_to_user(buf, __buf + __ppos, count))
1073                 return -EFAULT;
1074         *ppos = __ppos + count;
1075         return count;
1076 }
1077
1078 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1079                              size_t count, loff_t *ppos)
1080 {
1081         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1082         char __buf[20], *end;
1083         unsigned int seccomp_mode;
1084
1085         /* can set it only once to be even more secure */
1086         if (unlikely(tsk->seccomp.mode))
1087                 return -EPERM;
1088
1089         memset(__buf, 0, sizeof(__buf));
1090         count = min(count, sizeof(__buf) - 1);
1091         if (copy_from_user(__buf, buf, count))
1092                 return -EFAULT;
1093         seccomp_mode = simple_strtoul(__buf, &end, 0);
1094         if (*end == '\n')
1095                 end++;
1096         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1097                 tsk->seccomp.mode = seccomp_mode;
1098                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1099         } else
1100                 return -EINVAL;
1101         if (unlikely(!(end - __buf)))
1102                 return -EIO;
1103         return end - __buf;
1104 }
1105
1106 static struct file_operations proc_seccomp_operations = {
1107         .read           = seccomp_read,
1108         .write          = seccomp_write,
1109 };
1110 #endif /* CONFIG_SECCOMP */
1111
1112 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1113 {
1114         struct inode *inode = dentry->d_inode;
1115         int error = -EACCES;
1116
1117         /* We don't need a base pointer in the /proc filesystem */
1118         path_release(nd);
1119
1120         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1121                 goto out;
1122         error = proc_check_root(inode);
1123         if (error)
1124                 goto out;
1125
1126         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1127         nd->last_type = LAST_BIND;
1128 out:
1129         return ERR_PTR(error);
1130 }
1131
1132 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1133                             char __user *buffer, int buflen)
1134 {
1135         struct inode * inode;
1136         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1137         int len;
1138
1139         if (!tmp)
1140                 return -ENOMEM;
1141
1142         inode = dentry->d_inode;
1143         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1144         len = PTR_ERR(path);
1145         if (IS_ERR(path))
1146                 goto out;
1147         len = tmp + PAGE_SIZE - 1 - path;
1148
1149         if (len > buflen)
1150                 len = buflen;
1151         if (copy_to_user(buffer, path, len))
1152                 len = -EFAULT;
1153  out:
1154         free_page((unsigned long)tmp);
1155         return len;
1156 }
1157
1158 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1159 {
1160         int error = -EACCES;
1161         struct inode *inode = dentry->d_inode;
1162         struct dentry *de;
1163         struct vfsmount *mnt = NULL;
1164
1165         lock_kernel();
1166
1167         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1168                 goto out;
1169         error = proc_check_root(inode);
1170         if (error)
1171                 goto out;
1172
1173         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1174         if (error)
1175                 goto out;
1176
1177         error = do_proc_readlink(de, mnt, buffer, buflen);
1178         dput(de);
1179         mntput(mnt);
1180 out:
1181         unlock_kernel();
1182         return error;
1183 }
1184
1185 static struct inode_operations proc_pid_link_inode_operations = {
1186         .readlink       = proc_pid_readlink,
1187         .follow_link    = proc_pid_follow_link
1188 };
1189
1190 #define NUMBUF 10
1191
1192 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1193 {
1194         struct inode *inode = filp->f_dentry->d_inode;
1195         struct task_struct *p = proc_task(inode);
1196         unsigned int fd, tid, ino;
1197         int retval;
1198         char buf[NUMBUF];
1199         struct files_struct * files;
1200         struct fdtable *fdt;
1201
1202         retval = -ENOENT;
1203         if (!pid_alive(p))
1204                 goto out;
1205         retval = 0;
1206         tid = p->pid;
1207
1208         fd = filp->f_pos;
1209         switch (fd) {
1210                 case 0:
1211                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1212                                 goto out;
1213                         filp->f_pos++;
1214                 case 1:
1215                         ino = fake_ino(tid, PROC_TID_INO);
1216                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1217                                 goto out;
1218                         filp->f_pos++;
1219                 default:
1220                         files = get_files_struct(p);
1221                         if (!files)
1222                                 goto out;
1223                         rcu_read_lock();
1224                         fdt = files_fdtable(files);
1225                         for (fd = filp->f_pos-2;
1226                              fd < fdt->max_fds;
1227                              fd++, filp->f_pos++) {
1228                                 unsigned int i,j;
1229
1230                                 if (!fcheck_files(files, fd))
1231                                         continue;
1232                                 rcu_read_unlock();
1233
1234                                 j = NUMBUF;
1235                                 i = fd;
1236                                 do {
1237                                         j--;
1238                                         buf[j] = '0' + (i % 10);
1239                                         i /= 10;
1240                                 } while (i);
1241
1242                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1243                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1244                                         rcu_read_lock();
1245                                         break;
1246                                 }
1247                                 rcu_read_lock();
1248                         }
1249                         rcu_read_unlock();
1250                         put_files_struct(files);
1251         }
1252 out:
1253         return retval;
1254 }
1255
1256 static int proc_pident_readdir(struct file *filp,
1257                 void *dirent, filldir_t filldir,
1258                 struct pid_entry *ents, unsigned int nents)
1259 {
1260         int i;
1261         int pid;
1262         struct dentry *dentry = filp->f_dentry;
1263         struct inode *inode = dentry->d_inode;
1264         struct pid_entry *p;
1265         ino_t ino;
1266         int ret;
1267
1268         ret = -ENOENT;
1269         if (!pid_alive(proc_task(inode)))
1270                 goto out;
1271
1272         ret = 0;
1273         pid = proc_task(inode)->pid;
1274         i = filp->f_pos;
1275         switch (i) {
1276         case 0:
1277                 ino = inode->i_ino;
1278                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1279                         goto out;
1280                 i++;
1281                 filp->f_pos++;
1282                 /* fall through */
1283         case 1:
1284                 ino = parent_ino(dentry);
1285                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1286                         goto out;
1287                 i++;
1288                 filp->f_pos++;
1289                 /* fall through */
1290         default:
1291                 i -= 2;
1292                 if (i >= nents) {
1293                         ret = 1;
1294                         goto out;
1295                 }
1296                 p = ents + i;
1297                 while (p->name) {
1298                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1299                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1300                                 goto out;
1301                         filp->f_pos++;
1302                         p++;
1303                 }
1304         }
1305
1306         ret = 1;
1307 out:
1308         return ret;
1309 }
1310
1311 static int proc_tgid_base_readdir(struct file * filp,
1312                              void * dirent, filldir_t filldir)
1313 {
1314         return proc_pident_readdir(filp,dirent,filldir,
1315                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1316 }
1317
1318 static int proc_tid_base_readdir(struct file * filp,
1319                              void * dirent, filldir_t filldir)
1320 {
1321         return proc_pident_readdir(filp,dirent,filldir,
1322                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1323 }
1324
1325 /* building an inode */
1326
1327 static int task_dumpable(struct task_struct *task)
1328 {
1329         int dumpable = 0;
1330         struct mm_struct *mm;
1331
1332         task_lock(task);
1333         mm = task->mm;
1334         if (mm)
1335                 dumpable = mm->dumpable;
1336         task_unlock(task);
1337         if(dumpable == 1)
1338                 return 1;
1339         return 0;
1340 }
1341
1342
1343 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1344 {
1345         struct inode * inode;
1346         struct proc_inode *ei;
1347
1348         /* We need a new inode */
1349
1350         inode = new_inode(sb);
1351         if (!inode)
1352                 goto out;
1353
1354         /* Common stuff */
1355         ei = PROC_I(inode);
1356         ei->task = NULL;
1357         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1358         inode->i_ino = fake_ino(task->pid, ino);
1359
1360         if (!pid_alive(task))
1361                 goto out_unlock;
1362
1363         /*
1364          * grab the reference to task.
1365          */
1366         get_task_struct(task);
1367         ei->task = task;
1368         ei->type = ino;
1369         inode->i_uid = 0;
1370         inode->i_gid = 0;
1371         if (ino == PROC_TGID_INO || ino == PROC_TID_INO || task_dumpable(task)) {
1372                 inode->i_uid = task->euid;
1373                 inode->i_gid = task->egid;
1374         }
1375         security_task_to_inode(task, inode);
1376
1377 out:
1378         return inode;
1379
1380 out_unlock:
1381         ei->pde = NULL;
1382         iput(inode);
1383         return NULL;
1384 }
1385
1386 /* dentry stuff */
1387
1388 /*
1389  *      Exceptional case: normally we are not allowed to unhash a busy
1390  * directory. In this case, however, we can do it - no aliasing problems
1391  * due to the way we treat inodes.
1392  *
1393  * Rewrite the inode's ownerships here because the owning task may have
1394  * performed a setuid(), etc.
1395  */
1396 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1397 {
1398         struct inode *inode = dentry->d_inode;
1399         struct task_struct *task = proc_task(inode);
1400         if (pid_alive(task)) {
1401                 if (proc_type(inode) == PROC_TGID_INO || proc_type(inode) == PROC_TID_INO || task_dumpable(task)) {
1402                         inode->i_uid = task->euid;
1403                         inode->i_gid = task->egid;
1404                 } else {
1405                         inode->i_uid = 0;
1406                         inode->i_gid = 0;
1407                 }
1408                 security_task_to_inode(task, inode);
1409                 return 1;
1410         }
1411         d_drop(dentry);
1412         return 0;
1413 }
1414
1415 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1416 {
1417         struct inode *inode = dentry->d_inode;
1418         struct task_struct *task = proc_task(inode);
1419         int fd = proc_type(inode) - PROC_TID_FD_DIR;
1420         struct files_struct *files;
1421
1422         files = get_files_struct(task);
1423         if (files) {
1424                 rcu_read_lock();
1425                 if (fcheck_files(files, fd)) {
1426                         rcu_read_unlock();
1427                         put_files_struct(files);
1428                         if (task_dumpable(task)) {
1429                                 inode->i_uid = task->euid;
1430                                 inode->i_gid = task->egid;
1431                         } else {
1432                                 inode->i_uid = 0;
1433                                 inode->i_gid = 0;
1434                         }
1435                         security_task_to_inode(task, inode);
1436                         return 1;
1437                 }
1438                 rcu_read_unlock();
1439                 put_files_struct(files);
1440         }
1441         d_drop(dentry);
1442         return 0;
1443 }
1444
1445 static void pid_base_iput(struct dentry *dentry, struct inode *inode)
1446 {
1447         struct task_struct *task = proc_task(inode);
1448         spin_lock(&task->proc_lock);
1449         if (task->proc_dentry == dentry)
1450                 task->proc_dentry = NULL;
1451         spin_unlock(&task->proc_lock);
1452         iput(inode);
1453 }
1454
1455 static int pid_delete_dentry(struct dentry * dentry)
1456 {
1457         /* Is the task we represent dead?
1458          * If so, then don't put the dentry on the lru list,
1459          * kill it immediately.
1460          */
1461         return !pid_alive(proc_task(dentry->d_inode));
1462 }
1463
1464 static struct dentry_operations tid_fd_dentry_operations =
1465 {
1466         .d_revalidate   = tid_fd_revalidate,
1467         .d_delete       = pid_delete_dentry,
1468 };
1469
1470 static struct dentry_operations pid_dentry_operations =
1471 {
1472         .d_revalidate   = pid_revalidate,
1473         .d_delete       = pid_delete_dentry,
1474 };
1475
1476 static struct dentry_operations pid_base_dentry_operations =
1477 {
1478         .d_revalidate   = pid_revalidate,
1479         .d_iput         = pid_base_iput,
1480         .d_delete       = pid_delete_dentry,
1481 };
1482
1483 /* Lookups */
1484
1485 static unsigned name_to_int(struct dentry *dentry)
1486 {
1487         const char *name = dentry->d_name.name;
1488         int len = dentry->d_name.len;
1489         unsigned n = 0;
1490
1491         if (len > 1 && *name == '0')
1492                 goto out;
1493         while (len-- > 0) {
1494                 unsigned c = *name++ - '0';
1495                 if (c > 9)
1496                         goto out;
1497                 if (n >= (~0U-9)/10)
1498                         goto out;
1499                 n *= 10;
1500                 n += c;
1501         }
1502         return n;
1503 out:
1504         return ~0U;
1505 }
1506
1507 /* SMP-safe */
1508 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1509 {
1510         struct task_struct *task = proc_task(dir);
1511         unsigned fd = name_to_int(dentry);
1512         struct file * file;
1513         struct files_struct * files;
1514         struct inode *inode;
1515         struct proc_inode *ei;
1516
1517         if (fd == ~0U)
1518                 goto out;
1519         if (!pid_alive(task))
1520                 goto out;
1521
1522         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1523         if (!inode)
1524                 goto out;
1525         ei = PROC_I(inode);
1526         files = get_files_struct(task);
1527         if (!files)
1528                 goto out_unlock;
1529         inode->i_mode = S_IFLNK;
1530
1531         /*
1532          * We are not taking a ref to the file structure, so we must
1533          * hold ->file_lock.
1534          */
1535         spin_lock(&files->file_lock);
1536         file = fcheck_files(files, fd);
1537         if (!file)
1538                 goto out_unlock2;
1539         if (file->f_mode & 1)
1540                 inode->i_mode |= S_IRUSR | S_IXUSR;
1541         if (file->f_mode & 2)
1542                 inode->i_mode |= S_IWUSR | S_IXUSR;
1543         spin_unlock(&files->file_lock);
1544         put_files_struct(files);
1545         inode->i_op = &proc_pid_link_inode_operations;
1546         inode->i_size = 64;
1547         ei->op.proc_get_link = proc_fd_link;
1548         dentry->d_op = &tid_fd_dentry_operations;
1549         d_add(dentry, inode);
1550         return NULL;
1551
1552 out_unlock2:
1553         spin_unlock(&files->file_lock);
1554         put_files_struct(files);
1555 out_unlock:
1556         iput(inode);
1557 out:
1558         return ERR_PTR(-ENOENT);
1559 }
1560
1561 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1562 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1563
1564 static struct file_operations proc_fd_operations = {
1565         .read           = generic_read_dir,
1566         .readdir        = proc_readfd,
1567 };
1568
1569 static struct file_operations proc_task_operations = {
1570         .read           = generic_read_dir,
1571         .readdir        = proc_task_readdir,
1572 };
1573
1574 /*
1575  * proc directories can do almost nothing..
1576  */
1577 static struct inode_operations proc_fd_inode_operations = {
1578         .lookup         = proc_lookupfd,
1579         .permission     = proc_permission,
1580 };
1581
1582 static struct inode_operations proc_task_inode_operations = {
1583         .lookup         = proc_task_lookup,
1584         .permission     = proc_task_permission,
1585 };
1586
1587 #ifdef CONFIG_SECURITY
1588 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1589                                   size_t count, loff_t *ppos)
1590 {
1591         struct inode * inode = file->f_dentry->d_inode;
1592         unsigned long page;
1593         ssize_t length;
1594         struct task_struct *task = proc_task(inode);
1595
1596         if (count > PAGE_SIZE)
1597                 count = PAGE_SIZE;
1598         if (!(page = __get_free_page(GFP_KERNEL)))
1599                 return -ENOMEM;
1600
1601         length = security_getprocattr(task,
1602                                       (char*)file->f_dentry->d_name.name,
1603                                       (void*)page, count);
1604         if (length >= 0)
1605                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1606         free_page(page);
1607         return length;
1608 }
1609
1610 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1611                                    size_t count, loff_t *ppos)
1612 {
1613         struct inode * inode = file->f_dentry->d_inode;
1614         char *page;
1615         ssize_t length;
1616         struct task_struct *task = proc_task(inode);
1617
1618         if (count > PAGE_SIZE)
1619                 count = PAGE_SIZE;
1620         if (*ppos != 0) {
1621                 /* No partial writes. */
1622                 return -EINVAL;
1623         }
1624         page = (char*)__get_free_page(GFP_USER);
1625         if (!page)
1626                 return -ENOMEM;
1627         length = -EFAULT;
1628         if (copy_from_user(page, buf, count))
1629                 goto out;
1630
1631         length = security_setprocattr(task,
1632                                       (char*)file->f_dentry->d_name.name,
1633                                       (void*)page, count);
1634 out:
1635         free_page((unsigned long) page);
1636         return length;
1637 }
1638
1639 static struct file_operations proc_pid_attr_operations = {
1640         .read           = proc_pid_attr_read,
1641         .write          = proc_pid_attr_write,
1642 };
1643
1644 static struct file_operations proc_tid_attr_operations;
1645 static struct inode_operations proc_tid_attr_inode_operations;
1646 static struct file_operations proc_tgid_attr_operations;
1647 static struct inode_operations proc_tgid_attr_inode_operations;
1648 #endif
1649
1650 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1651
1652 /* SMP-safe */
1653 static struct dentry *proc_pident_lookup(struct inode *dir,
1654                                          struct dentry *dentry,
1655                                          struct pid_entry *ents)
1656 {
1657         struct inode *inode;
1658         int error;
1659         struct task_struct *task = proc_task(dir);
1660         struct pid_entry *p;
1661         struct proc_inode *ei;
1662
1663         error = -ENOENT;
1664         inode = NULL;
1665
1666         if (!pid_alive(task))
1667                 goto out;
1668
1669         for (p = ents; p->name; p++) {
1670                 if (p->len != dentry->d_name.len)
1671                         continue;
1672                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1673                         break;
1674         }
1675         if (!p->name)
1676                 goto out;
1677
1678         error = -EINVAL;
1679         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1680         if (!inode)
1681                 goto out;
1682
1683         ei = PROC_I(inode);
1684         inode->i_mode = p->mode;
1685         /*
1686          * Yes, it does not scale. And it should not. Don't add
1687          * new entries into /proc/<tgid>/ without very good reasons.
1688          */
1689         switch(p->type) {
1690                 case PROC_TGID_TASK:
1691                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1692                         inode->i_op = &proc_task_inode_operations;
1693                         inode->i_fop = &proc_task_operations;
1694                         break;
1695                 case PROC_TID_FD:
1696                 case PROC_TGID_FD:
1697                         inode->i_nlink = 2;
1698                         inode->i_op = &proc_fd_inode_operations;
1699                         inode->i_fop = &proc_fd_operations;
1700                         break;
1701                 case PROC_TID_EXE:
1702                 case PROC_TGID_EXE:
1703                         inode->i_op = &proc_pid_link_inode_operations;
1704                         ei->op.proc_get_link = proc_exe_link;
1705                         break;
1706                 case PROC_TID_CWD:
1707                 case PROC_TGID_CWD:
1708                         inode->i_op = &proc_pid_link_inode_operations;
1709                         ei->op.proc_get_link = proc_cwd_link;
1710                         break;
1711                 case PROC_TID_ROOT:
1712                 case PROC_TGID_ROOT:
1713                         inode->i_op = &proc_pid_link_inode_operations;
1714                         ei->op.proc_get_link = proc_root_link;
1715                         break;
1716                 case PROC_TID_ENVIRON:
1717                 case PROC_TGID_ENVIRON:
1718                         inode->i_fop = &proc_info_file_operations;
1719                         ei->op.proc_read = proc_pid_environ;
1720                         break;
1721                 case PROC_TID_AUXV:
1722                 case PROC_TGID_AUXV:
1723                         inode->i_fop = &proc_info_file_operations;
1724                         ei->op.proc_read = proc_pid_auxv;
1725                         break;
1726                 case PROC_TID_STATUS:
1727                 case PROC_TGID_STATUS:
1728                         inode->i_fop = &proc_info_file_operations;
1729                         ei->op.proc_read = proc_pid_status;
1730                         break;
1731                 case PROC_TID_STAT:
1732                         inode->i_fop = &proc_info_file_operations;
1733                         ei->op.proc_read = proc_tid_stat;
1734                         break;
1735                 case PROC_TGID_STAT:
1736                         inode->i_fop = &proc_info_file_operations;
1737                         ei->op.proc_read = proc_tgid_stat;
1738                         break;
1739                 case PROC_TID_CMDLINE:
1740                 case PROC_TGID_CMDLINE:
1741                         inode->i_fop = &proc_info_file_operations;
1742                         ei->op.proc_read = proc_pid_cmdline;
1743                         break;
1744                 case PROC_TID_STATM:
1745                 case PROC_TGID_STATM:
1746                         inode->i_fop = &proc_info_file_operations;
1747                         ei->op.proc_read = proc_pid_statm;
1748                         break;
1749                 case PROC_TID_MAPS:
1750                 case PROC_TGID_MAPS:
1751                         inode->i_fop = &proc_maps_operations;
1752                         break;
1753 #ifdef CONFIG_NUMA
1754                 case PROC_TID_NUMA_MAPS:
1755                 case PROC_TGID_NUMA_MAPS:
1756                         inode->i_fop = &proc_numa_maps_operations;
1757                         break;
1758 #endif
1759                 case PROC_TID_MEM:
1760                 case PROC_TGID_MEM:
1761                         inode->i_op = &proc_mem_inode_operations;
1762                         inode->i_fop = &proc_mem_operations;
1763                         break;
1764 #ifdef CONFIG_SECCOMP
1765                 case PROC_TID_SECCOMP:
1766                 case PROC_TGID_SECCOMP:
1767                         inode->i_fop = &proc_seccomp_operations;
1768                         break;
1769 #endif /* CONFIG_SECCOMP */
1770                 case PROC_TID_MOUNTS:
1771                 case PROC_TGID_MOUNTS:
1772                         inode->i_fop = &proc_mounts_operations;
1773                         break;
1774 #ifdef CONFIG_MMU
1775                 case PROC_TID_SMAPS:
1776                 case PROC_TGID_SMAPS:
1777                         inode->i_fop = &proc_smaps_operations;
1778                         break;
1779 #endif
1780                 case PROC_TID_MOUNTSTATS:
1781                 case PROC_TGID_MOUNTSTATS:
1782                         inode->i_fop = &proc_mountstats_operations;
1783                         break;
1784 #ifdef CONFIG_SECURITY
1785                 case PROC_TID_ATTR:
1786                         inode->i_nlink = 2;
1787                         inode->i_op = &proc_tid_attr_inode_operations;
1788                         inode->i_fop = &proc_tid_attr_operations;
1789                         break;
1790                 case PROC_TGID_ATTR:
1791                         inode->i_nlink = 2;
1792                         inode->i_op = &proc_tgid_attr_inode_operations;
1793                         inode->i_fop = &proc_tgid_attr_operations;
1794                         break;
1795                 case PROC_TID_ATTR_CURRENT:
1796                 case PROC_TGID_ATTR_CURRENT:
1797                 case PROC_TID_ATTR_PREV:
1798                 case PROC_TGID_ATTR_PREV:
1799                 case PROC_TID_ATTR_EXEC:
1800                 case PROC_TGID_ATTR_EXEC:
1801                 case PROC_TID_ATTR_FSCREATE:
1802                 case PROC_TGID_ATTR_FSCREATE:
1803                         inode->i_fop = &proc_pid_attr_operations;
1804                         break;
1805 #endif
1806 #ifdef CONFIG_KALLSYMS
1807                 case PROC_TID_WCHAN:
1808                 case PROC_TGID_WCHAN:
1809                         inode->i_fop = &proc_info_file_operations;
1810                         ei->op.proc_read = proc_pid_wchan;
1811                         break;
1812 #endif
1813 #ifdef CONFIG_SCHEDSTATS
1814                 case PROC_TID_SCHEDSTAT:
1815                 case PROC_TGID_SCHEDSTAT:
1816                         inode->i_fop = &proc_info_file_operations;
1817                         ei->op.proc_read = proc_pid_schedstat;
1818                         break;
1819 #endif
1820 #ifdef CONFIG_CPUSETS
1821                 case PROC_TID_CPUSET:
1822                 case PROC_TGID_CPUSET:
1823                         inode->i_fop = &proc_cpuset_operations;
1824                         break;
1825 #endif
1826                 case PROC_TID_OOM_SCORE:
1827                 case PROC_TGID_OOM_SCORE:
1828                         inode->i_fop = &proc_info_file_operations;
1829                         ei->op.proc_read = proc_oom_score;
1830                         break;
1831                 case PROC_TID_OOM_ADJUST:
1832                 case PROC_TGID_OOM_ADJUST:
1833                         inode->i_fop = &proc_oom_adjust_operations;
1834                         break;
1835 #ifdef CONFIG_AUDITSYSCALL
1836                 case PROC_TID_LOGINUID:
1837                 case PROC_TGID_LOGINUID:
1838                         inode->i_fop = &proc_loginuid_operations;
1839                         break;
1840 #endif
1841                 default:
1842                         printk("procfs: impossible type (%d)",p->type);
1843                         iput(inode);
1844                         return ERR_PTR(-EINVAL);
1845         }
1846         dentry->d_op = &pid_dentry_operations;
1847         d_add(dentry, inode);
1848         return NULL;
1849
1850 out:
1851         return ERR_PTR(error);
1852 }
1853
1854 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1855         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1856 }
1857
1858 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1859         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1860 }
1861
1862 static struct file_operations proc_tgid_base_operations = {
1863         .read           = generic_read_dir,
1864         .readdir        = proc_tgid_base_readdir,
1865 };
1866
1867 static struct file_operations proc_tid_base_operations = {
1868         .read           = generic_read_dir,
1869         .readdir        = proc_tid_base_readdir,
1870 };
1871
1872 static struct inode_operations proc_tgid_base_inode_operations = {
1873         .lookup         = proc_tgid_base_lookup,
1874 };
1875
1876 static struct inode_operations proc_tid_base_inode_operations = {
1877         .lookup         = proc_tid_base_lookup,
1878 };
1879
1880 #ifdef CONFIG_SECURITY
1881 static int proc_tgid_attr_readdir(struct file * filp,
1882                              void * dirent, filldir_t filldir)
1883 {
1884         return proc_pident_readdir(filp,dirent,filldir,
1885                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1886 }
1887
1888 static int proc_tid_attr_readdir(struct file * filp,
1889                              void * dirent, filldir_t filldir)
1890 {
1891         return proc_pident_readdir(filp,dirent,filldir,
1892                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1893 }
1894
1895 static struct file_operations proc_tgid_attr_operations = {
1896         .read           = generic_read_dir,
1897         .readdir        = proc_tgid_attr_readdir,
1898 };
1899
1900 static struct file_operations proc_tid_attr_operations = {
1901         .read           = generic_read_dir,
1902         .readdir        = proc_tid_attr_readdir,
1903 };
1904
1905 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1906                                 struct dentry *dentry, struct nameidata *nd)
1907 {
1908         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1909 }
1910
1911 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1912                                 struct dentry *dentry, struct nameidata *nd)
1913 {
1914         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1915 }
1916
1917 static struct inode_operations proc_tgid_attr_inode_operations = {
1918         .lookup         = proc_tgid_attr_lookup,
1919 };
1920
1921 static struct inode_operations proc_tid_attr_inode_operations = {
1922         .lookup         = proc_tid_attr_lookup,
1923 };
1924 #endif
1925
1926 /*
1927  * /proc/self:
1928  */
1929 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1930                               int buflen)
1931 {
1932         char tmp[30];
1933         sprintf(tmp, "%d", current->tgid);
1934         return vfs_readlink(dentry,buffer,buflen,tmp);
1935 }
1936
1937 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1938 {
1939         char tmp[30];
1940         sprintf(tmp, "%d", current->tgid);
1941         return ERR_PTR(vfs_follow_link(nd,tmp));
1942 }
1943
1944 static struct inode_operations proc_self_inode_operations = {
1945         .readlink       = proc_self_readlink,
1946         .follow_link    = proc_self_follow_link,
1947 };
1948
1949 /**
1950  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1951  * @p: task that should be flushed.
1952  *
1953  * Drops the /proc/@pid dcache entry from the hash chains.
1954  *
1955  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1956  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1957  * if the pid value is immediately reused. This is enforced by
1958  * - caller must acquire spin_lock(p->proc_lock)
1959  * - must be called before detach_pid()
1960  * - proc_pid_lookup acquires proc_lock, and checks that
1961  *   the target is not dead by looking at the attach count
1962  *   of PIDTYPE_PID.
1963  */
1964
1965 struct dentry *proc_pid_unhash(struct task_struct *p)
1966 {
1967         struct dentry *proc_dentry;
1968
1969         proc_dentry = p->proc_dentry;
1970         if (proc_dentry != NULL) {
1971
1972                 spin_lock(&dcache_lock);
1973                 spin_lock(&proc_dentry->d_lock);
1974                 if (!d_unhashed(proc_dentry)) {
1975                         dget_locked(proc_dentry);
1976                         __d_drop(proc_dentry);
1977                         spin_unlock(&proc_dentry->d_lock);
1978                 } else {
1979                         spin_unlock(&proc_dentry->d_lock);
1980                         proc_dentry = NULL;
1981                 }
1982                 spin_unlock(&dcache_lock);
1983         }
1984         return proc_dentry;
1985 }
1986
1987 /**
1988  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1989  * @proc_dentry: directoy to prune.
1990  *
1991  * Shrink the /proc directory that was used by the just killed thread.
1992  */
1993
1994 void proc_pid_flush(struct dentry *proc_dentry)
1995 {
1996         might_sleep();
1997         if(proc_dentry != NULL) {
1998                 shrink_dcache_parent(proc_dentry);
1999                 dput(proc_dentry);
2000         }
2001 }
2002
2003 /* SMP-safe */
2004 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2005 {
2006         struct task_struct *task;
2007         struct inode *inode;
2008         struct proc_inode *ei;
2009         unsigned tgid;
2010         int died;
2011
2012         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
2013                 inode = new_inode(dir->i_sb);
2014                 if (!inode)
2015                         return ERR_PTR(-ENOMEM);
2016                 ei = PROC_I(inode);
2017                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2018                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
2019                 ei->pde = NULL;
2020                 inode->i_mode = S_IFLNK|S_IRWXUGO;
2021                 inode->i_uid = inode->i_gid = 0;
2022                 inode->i_size = 64;
2023                 inode->i_op = &proc_self_inode_operations;
2024                 d_add(dentry, inode);
2025                 return NULL;
2026         }
2027         tgid = name_to_int(dentry);
2028         if (tgid == ~0U)
2029                 goto out;
2030
2031         read_lock(&tasklist_lock);
2032         task = find_task_by_pid(tgid);
2033         if (task)
2034                 get_task_struct(task);
2035         read_unlock(&tasklist_lock);
2036         if (!task)
2037                 goto out;
2038
2039         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2040
2041
2042         if (!inode) {
2043                 put_task_struct(task);
2044                 goto out;
2045         }
2046         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2047         inode->i_op = &proc_tgid_base_inode_operations;
2048         inode->i_fop = &proc_tgid_base_operations;
2049         inode->i_flags|=S_IMMUTABLE;
2050 #ifdef CONFIG_SECURITY
2051         inode->i_nlink = 5;
2052 #else
2053         inode->i_nlink = 4;
2054 #endif
2055
2056         dentry->d_op = &pid_base_dentry_operations;
2057
2058         died = 0;
2059         d_add(dentry, inode);
2060         spin_lock(&task->proc_lock);
2061         task->proc_dentry = dentry;
2062         if (!pid_alive(task)) {
2063                 dentry = proc_pid_unhash(task);
2064                 died = 1;
2065         }
2066         spin_unlock(&task->proc_lock);
2067
2068         put_task_struct(task);
2069         if (died) {
2070                 proc_pid_flush(dentry);
2071                 goto out;
2072         }
2073         return NULL;
2074 out:
2075         return ERR_PTR(-ENOENT);
2076 }
2077
2078 /* SMP-safe */
2079 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2080 {
2081         struct task_struct *task;
2082         struct task_struct *leader = proc_task(dir);
2083         struct inode *inode;
2084         unsigned tid;
2085
2086         tid = name_to_int(dentry);
2087         if (tid == ~0U)
2088                 goto out;
2089
2090         read_lock(&tasklist_lock);
2091         task = find_task_by_pid(tid);
2092         if (task)
2093                 get_task_struct(task);
2094         read_unlock(&tasklist_lock);
2095         if (!task)
2096                 goto out;
2097         if (leader->tgid != task->tgid)
2098                 goto out_drop_task;
2099
2100         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2101
2102
2103         if (!inode)
2104                 goto out_drop_task;
2105         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2106         inode->i_op = &proc_tid_base_inode_operations;
2107         inode->i_fop = &proc_tid_base_operations;
2108         inode->i_flags|=S_IMMUTABLE;
2109 #ifdef CONFIG_SECURITY
2110         inode->i_nlink = 4;
2111 #else
2112         inode->i_nlink = 3;
2113 #endif
2114
2115         dentry->d_op = &pid_base_dentry_operations;
2116
2117         d_add(dentry, inode);
2118
2119         put_task_struct(task);
2120         return NULL;
2121 out_drop_task:
2122         put_task_struct(task);
2123 out:
2124         return ERR_PTR(-ENOENT);
2125 }
2126
2127 #define PROC_NUMBUF 10
2128 #define PROC_MAXPIDS 20
2129
2130 /*
2131  * Get a few tgid's to return for filldir - we need to hold the
2132  * tasklist lock while doing this, and we must release it before
2133  * we actually do the filldir itself, so we use a temp buffer..
2134  */
2135 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2136 {
2137         struct task_struct *p;
2138         int nr_tgids = 0;
2139
2140         index--;
2141         read_lock(&tasklist_lock);
2142         p = NULL;
2143         if (version) {
2144                 p = find_task_by_pid(version);
2145                 if (p && !thread_group_leader(p))
2146                         p = NULL;
2147         }
2148
2149         if (p)
2150                 index = 0;
2151         else
2152                 p = next_task(&init_task);
2153
2154         for ( ; p != &init_task; p = next_task(p)) {
2155                 int tgid = p->pid;
2156                 if (!pid_alive(p))
2157                         continue;
2158                 if (--index >= 0)
2159                         continue;
2160                 tgids[nr_tgids] = tgid;
2161                 nr_tgids++;
2162                 if (nr_tgids >= PROC_MAXPIDS)
2163                         break;
2164         }
2165         read_unlock(&tasklist_lock);
2166         return nr_tgids;
2167 }
2168
2169 /*
2170  * Get a few tid's to return for filldir - we need to hold the
2171  * tasklist lock while doing this, and we must release it before
2172  * we actually do the filldir itself, so we use a temp buffer..
2173  */
2174 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2175 {
2176         struct task_struct *leader_task = proc_task(dir);
2177         struct task_struct *task = leader_task;
2178         int nr_tids = 0;
2179
2180         index -= 2;
2181         read_lock(&tasklist_lock);
2182         /*
2183          * The starting point task (leader_task) might be an already
2184          * unlinked task, which cannot be used to access the task-list
2185          * via next_thread().
2186          */
2187         if (pid_alive(task)) do {
2188                 int tid = task->pid;
2189
2190                 if (--index >= 0)
2191                         continue;
2192                 if (tids != NULL)
2193                         tids[nr_tids] = tid;
2194                 nr_tids++;
2195                 if (nr_tids >= PROC_MAXPIDS)
2196                         break;
2197         } while ((task = next_thread(task)) != leader_task);
2198         read_unlock(&tasklist_lock);
2199         return nr_tids;
2200 }
2201
2202 /* for the /proc/ directory itself, after non-process stuff has been done */
2203 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2204 {
2205         unsigned int tgid_array[PROC_MAXPIDS];
2206         char buf[PROC_NUMBUF];
2207         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2208         unsigned int nr_tgids, i;
2209         int next_tgid;
2210
2211         if (!nr) {
2212                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2213                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2214                         return 0;
2215                 filp->f_pos++;
2216                 nr++;
2217         }
2218
2219         /* f_version caches the tgid value that the last readdir call couldn't
2220          * return. lseek aka telldir automagically resets f_version to 0.
2221          */
2222         next_tgid = filp->f_version;
2223         filp->f_version = 0;
2224         for (;;) {
2225                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2226                 if (!nr_tgids) {
2227                         /* no more entries ! */
2228                         break;
2229                 }
2230                 next_tgid = 0;
2231
2232                 /* do not use the last found pid, reserve it for next_tgid */
2233                 if (nr_tgids == PROC_MAXPIDS) {
2234                         nr_tgids--;
2235                         next_tgid = tgid_array[nr_tgids];
2236                 }
2237
2238                 for (i=0;i<nr_tgids;i++) {
2239                         int tgid = tgid_array[i];
2240                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2241                         unsigned long j = PROC_NUMBUF;
2242
2243                         do
2244                                 buf[--j] = '0' + (tgid % 10);
2245                         while ((tgid /= 10) != 0);
2246
2247                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2248                                 /* returning this tgid failed, save it as the first
2249                                  * pid for the next readir call */
2250                                 filp->f_version = tgid_array[i];
2251                                 goto out;
2252                         }
2253                         filp->f_pos++;
2254                         nr++;
2255                 }
2256         }
2257 out:
2258         return 0;
2259 }
2260
2261 /* for the /proc/TGID/task/ directories */
2262 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2263 {
2264         unsigned int tid_array[PROC_MAXPIDS];
2265         char buf[PROC_NUMBUF];
2266         unsigned int nr_tids, i;
2267         struct dentry *dentry = filp->f_dentry;
2268         struct inode *inode = dentry->d_inode;
2269         int retval = -ENOENT;
2270         ino_t ino;
2271         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2272
2273         if (!pid_alive(proc_task(inode)))
2274                 goto out;
2275         retval = 0;
2276
2277         switch (pos) {
2278         case 0:
2279                 ino = inode->i_ino;
2280                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2281                         goto out;
2282                 pos++;
2283                 /* fall through */
2284         case 1:
2285                 ino = parent_ino(dentry);
2286                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2287                         goto out;
2288                 pos++;
2289                 /* fall through */
2290         }
2291
2292         nr_tids = get_tid_list(pos, tid_array, inode);
2293         inode->i_nlink = pos + nr_tids;
2294
2295         for (i = 0; i < nr_tids; i++) {
2296                 unsigned long j = PROC_NUMBUF;
2297                 int tid = tid_array[i];
2298
2299                 ino = fake_ino(tid,PROC_TID_INO);
2300
2301                 do
2302                         buf[--j] = '0' + (tid % 10);
2303                 while ((tid /= 10) != 0);
2304
2305                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2306                         break;
2307                 pos++;
2308         }
2309 out:
2310         filp->f_pos = pos;
2311         return retval;
2312 }