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         PROC_TGID_ATTR_KEYCREATE,
 125 #endif
 126 #ifdef CONFIG_AUDITSYSCALL
 127         PROC_TGID_LOGINUID,
 128 #endif
 129         PROC_TGID_OOM_SCORE,
 130         PROC_TGID_OOM_ADJUST,
 131         PROC_TID_INO,
 132         PROC_TID_STATUS,
 133         PROC_TID_MEM,
 134 #ifdef CONFIG_SECCOMP
 135         PROC_TID_SECCOMP,
 136 #endif
 137         PROC_TID_CWD,
 138         PROC_TID_ROOT,
 139         PROC_TID_EXE,
 140         PROC_TID_FD,
 141         PROC_TID_ENVIRON,
 142         PROC_TID_AUXV,
 143         PROC_TID_CMDLINE,
 144         PROC_TID_STAT,
 145         PROC_TID_STATM,
 146         PROC_TID_MAPS,
 147         PROC_TID_NUMA_MAPS,
 148         PROC_TID_MOUNTS,
 149         PROC_TID_MOUNTSTATS,
 150         PROC_TID_WCHAN,
 151 #ifdef CONFIG_MMU
 152         PROC_TID_SMAPS,
 153 #endif
 154 #ifdef CONFIG_SCHEDSTATS
 155         PROC_TID_SCHEDSTAT,
 156 #endif
 157 #ifdef CONFIG_CPUSETS
 158         PROC_TID_CPUSET,
 159 #endif
 160 #ifdef CONFIG_SECURITY
 161         PROC_TID_ATTR,
 162         PROC_TID_ATTR_CURRENT,
 163         PROC_TID_ATTR_PREV,
 164         PROC_TID_ATTR_EXEC,
 165         PROC_TID_ATTR_FSCREATE,
 166         PROC_TID_ATTR_KEYCREATE,
 167 #endif
 168 #ifdef CONFIG_AUDITSYSCALL
 169         PROC_TID_LOGINUID,
 170 #endif
 171         PROC_TID_OOM_SCORE,
 172         PROC_TID_OOM_ADJUST,
 173
 174         /* Add new entries before this */
 175         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
 176 };
 177
 178 struct pid_entry {
 179         int type;
 180         int len;
 181         char *name;
 182         mode_t mode;
 183 };
 184
 185 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
 186
 187 static struct pid_entry tgid_base_stuff[] = {
 188         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
 189         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 190         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
 191         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
 192         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
 193         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 194         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 195         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
 196         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 197 #ifdef CONFIG_NUMA
 198         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
 199 #endif
 200         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 201 #ifdef CONFIG_SECCOMP
 202         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 203 #endif
 204         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
 205         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
 206         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
 207         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
 208         E(PROC_TGID_MOUNTSTATS, "mountstats", S_IFREG|S_IRUSR),
 209 #ifdef CONFIG_MMU
 210         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
 211 #endif
 212 #ifdef CONFIG_SECURITY
 213         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 214 #endif
 215 #ifdef CONFIG_KALLSYMS
 216         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
 217 #endif
 218 #ifdef CONFIG_SCHEDSTATS
 219         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
 220 #endif
 221 #ifdef CONFIG_CPUSETS
 222         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
 223 #endif
 224         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
 225         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 226 #ifdef CONFIG_AUDITSYSCALL
 227         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 228 #endif
 229         {0,0,NULL,0}
 230 };
 231 static struct pid_entry tid_base_stuff[] = {
 232         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 233         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
 234         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
 235         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
 236         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 237         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 238         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
 239         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 240 #ifdef CONFIG_NUMA
 241         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
 242 #endif
 243         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 244 #ifdef CONFIG_SECCOMP
 245         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 246 #endif
 247         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
 248         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
 249         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
 250         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
 251 #ifdef CONFIG_MMU
 252         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
 253 #endif
 254 #ifdef CONFIG_SECURITY
 255         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 256 #endif
 257 #ifdef CONFIG_KALLSYMS
 258         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
 259 #endif
 260 #ifdef CONFIG_SCHEDSTATS
 261         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
 262 #endif
 263 #ifdef CONFIG_CPUSETS
 264         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
 265 #endif
 266         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
 267         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 268 #ifdef CONFIG_AUDITSYSCALL
 269         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 270 #endif
 271         {0,0,NULL,0}
 272 };
 273
 274 #ifdef CONFIG_SECURITY
 275 static struct pid_entry tgid_attr_stuff[] = {
 276         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
 277         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
 278         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 279         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 280         E(PROC_TGID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 281         {0,0,NULL,0}
 282 };
 283 static struct pid_entry tid_attr_stuff[] = {
 284         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
 285         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
 286         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 287         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 288         E(PROC_TID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 289         {0,0,NULL,0}
 290 };
 291 #endif
 292
 293 #undef E
 294
 295 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 296 {
 297         struct task_struct *task = proc_task(inode);
 298         struct files_struct *files;
 299         struct file *file;
 300         int fd = proc_fd(inode);
 301
 302         files = get_files_struct(task);
 303         if (files) {
 304                 /*
 305                  * We are not taking a ref to the file structure, so we must
 306                  * hold ->file_lock.
 307                  */
 308                 spin_lock(&files->file_lock);
 309                 file = fcheck_files(files, fd);
 310                 if (file) {
 311                         *mnt = mntget(file->f_vfsmnt);
 312                         *dentry = dget(file->f_dentry);
 313                         spin_unlock(&files->file_lock);
 314                         put_files_struct(files);
 315                         return 0;
 316                 }
 317                 spin_unlock(&files->file_lock);
 318                 put_files_struct(files);
 319         }
 320         return -ENOENT;
 321 }
 322
 323 static struct fs_struct *get_fs_struct(struct task_struct *task)
 324 {
 325         struct fs_struct *fs;
 326         task_lock(task);
 327         fs = task->fs;
 328         if(fs)
 329                 atomic_inc(&fs->count);
 330         task_unlock(task);
 331         return fs;
 332 }
 333
 334 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 335 {
 336         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 337         int result = -ENOENT;
 338         if (fs) {
 339                 read_lock(&fs->lock);
 340                 *mnt = mntget(fs->pwdmnt);
 341                 *dentry = dget(fs->pwd);
 342                 read_unlock(&fs->lock);
 343                 result = 0;
 344                 put_fs_struct(fs);
 345         }
 346         return result;
 347 }
 348
 349 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 350 {
 351         struct fs_struct *fs = get_fs_struct(proc_task(inode));
 352         int result = -ENOENT;
 353         if (fs) {
 354                 read_lock(&fs->lock);
 355                 *mnt = mntget(fs->rootmnt);
 356                 *dentry = dget(fs->root);
 357                 read_unlock(&fs->lock);
 358                 result = 0;
 359                 put_fs_struct(fs);
 360         }
 361         return result;
 362 }
 363
 364
 365 /* Same as proc_root_link, but this addionally tries to get fs from other
 366  * threads in the group */
 367 static int proc_task_root_link(struct inode *inode, struct dentry **dentry,
 368                                 struct vfsmount **mnt)
 369 {
 370         struct fs_struct *fs;
 371         int result = -ENOENT;
 372         struct task_struct *leader = proc_task(inode);
 373
 374         task_lock(leader);
 375         fs = leader->fs;
 376         if (fs) {
 377                 atomic_inc(&fs->count);
 378                 task_unlock(leader);
 379         } else {
 380                 /* Try to get fs from other threads */
 381                 task_unlock(leader);
 382                 read_lock(&tasklist_lock);
 383                 if (pid_alive(leader)) {
 384                         struct task_struct *task = leader;
 385
 386                         while ((task = next_thread(task)) != leader) {
 387                                 task_lock(task);
 388                                 fs = task->fs;
 389                                 if (fs) {
 390                                         atomic_inc(&fs->count);
 391                                         task_unlock(task);
 392                                         break;
 393                                 }
 394                                 task_unlock(task);
 395                         }
 396                 }
 397                 read_unlock(&tasklist_lock);
 398         }
 399
 400         if (fs) {
 401                 read_lock(&fs->lock);
 402                 *mnt = mntget(fs->rootmnt);
 403                 *dentry = dget(fs->root);
 404                 read_unlock(&fs->lock);
 405                 result = 0;
 406                 put_fs_struct(fs);
 407         }
 408         return result;
 409 }
 410
 411
 412 #define MAY_PTRACE(task) \
 413         (task == current || \
 414         (task->parent == current && \
 415         (task->ptrace & PT_PTRACED) && \
 416          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
 417          security_ptrace(current,task) == 0))
 418
 419 static int proc_pid_environ(struct task_struct *task, char * buffer)
 420 {
 421         int res = 0;
 422         struct mm_struct *mm = get_task_mm(task);
 423         if (mm) {
 424                 unsigned int len = mm->env_end - mm->env_start;
 425                 if (len > PAGE_SIZE)
 426                         len = PAGE_SIZE;
 427                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
 428                 if (!ptrace_may_attach(task))
 429                         res = -ESRCH;
 430                 mmput(mm);
 431         }
 432         return res;
 433 }
 434
 435 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 436 {
 437         int res = 0;
 438         unsigned int len;
 439         struct mm_struct *mm = get_task_mm(task);
 440         if (!mm)
 441                 goto out;
 442         if (!mm->arg_end)
 443                 goto out_mm;    /* Shh! No looking before we're done */
 444
 445         len = mm->arg_end - mm->arg_start;
 446
 447         if (len > PAGE_SIZE)
 448                 len = PAGE_SIZE;
 449
 450         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 451
 452         // If the nul at the end of args has been overwritten, then
 453         // assume application is using setproctitle(3).
 454         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 455                 len = strnlen(buffer, res);
 456                 if (len < res) {
 457                     res = len;
 458                 } else {
 459                         len = mm->env_end - mm->env_start;
 460                         if (len > PAGE_SIZE - res)
 461                                 len = PAGE_SIZE - res;
 462                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 463                         res = strnlen(buffer, res);
 464                 }
 465         }
 466 out_mm:
 467         mmput(mm);
 468 out:
 469         return res;
 470 }
 471
 472 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 473 {
 474         int res = 0;
 475         struct mm_struct *mm = get_task_mm(task);
 476         if (mm) {
 477                 unsigned int nwords = 0;
 478                 do
 479                         nwords += 2;
 480                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 481                 res = nwords * sizeof(mm->saved_auxv[0]);
 482                 if (res > PAGE_SIZE)
 483                         res = PAGE_SIZE;
 484                 memcpy(buffer, mm->saved_auxv, res);
 485                 mmput(mm);
 486         }
 487         return res;
 488 }
 489
 490
 491 #ifdef CONFIG_KALLSYMS
 492 /*
 493  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 494  * Returns the resolved symbol.  If that fails, simply return the address.
 495  */
 496 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 497 {
 498         char *modname;
 499         const char *sym_name;
 500         unsigned long wchan, size, offset;
 501         char namebuf[KSYM_NAME_LEN+1];
 502
 503         wchan = get_wchan(task);
 504
 505         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
 506         if (sym_name)
 507                 return sprintf(buffer, "%s", sym_name);
 508         return sprintf(buffer, "%lu", wchan);
 509 }
 510 #endif /* CONFIG_KALLSYMS */
 511
 512 #ifdef CONFIG_SCHEDSTATS
 513 /*
 514  * Provides /proc/PID/schedstat
 515  */
 516 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 517 {
 518         return sprintf(buffer, "%lu %lu %lu\n",
 519                         task->sched_info.cpu_time,
 520                         task->sched_info.run_delay,
 521                         task->sched_info.pcnt);
 522 }
 523 #endif
 524
 525 /* The badness from the OOM killer */
 526 unsigned long badness(struct task_struct *p, unsigned long uptime);
 527 static int proc_oom_score(struct task_struct *task, char *buffer)
 528 {
 529         unsigned long points;
 530         struct timespec uptime;
 531
 532         do_posix_clock_monotonic_gettime(&uptime);
 533         points = badness(task, uptime.tv_sec);
 534         return sprintf(buffer, "%lu\n", points);
 535 }
 536
 537 /************************************************************************/
 538 /*                       Here the fs part begins                        */
 539 /************************************************************************/
 540
 541 /* permission checks */
 542
 543 /* If the process being read is separated by chroot from the reading process,
 544  * don't let the reader access the threads.
 545  *
 546  * note: this does dput(root) and mntput(vfsmnt) on exit.
 547  */
 548 static int proc_check_chroot(struct dentry *root, struct vfsmount *vfsmnt)
 549 {
 550         struct dentry *de, *base;
 551         struct vfsmount *our_vfsmnt, *mnt;
 552         int res = 0;
 553
 554         read_lock(&current->fs->lock);
 555         our_vfsmnt = mntget(current->fs->rootmnt);
 556         base = dget(current->fs->root);
 557         read_unlock(&current->fs->lock);
 558
 559         spin_lock(&vfsmount_lock);
 560         de = root;
 561         mnt = vfsmnt;
 562
 563         while (mnt != our_vfsmnt) {
 564                 if (mnt == mnt->mnt_parent)
 565                         goto out;
 566                 de = mnt->mnt_mountpoint;
 567                 mnt = mnt->mnt_parent;
 568         }
 569
 570         if (!is_subdir(de, base))
 571                 goto out;
 572         spin_unlock(&vfsmount_lock);
 573
 574 exit:
 575         dput(base);
 576         mntput(our_vfsmnt);
 577         dput(root);
 578         mntput(vfsmnt);
 579         return res;
 580 out:
 581         spin_unlock(&vfsmount_lock);
 582         res = -EACCES;
 583         goto exit;
 584 }
 585
 586 static int proc_check_root(struct inode *inode)
 587 {
 588         struct dentry *root;
 589         struct vfsmount *vfsmnt;
 590
 591         if (proc_root_link(inode, &root, &vfsmnt)) /* Ewww... */
 592                 return -ENOENT;
 593         return proc_check_chroot(root, vfsmnt);
 594 }
 595
 596 static int proc_permission(struct inode *inode, int mask, struct nameidata *nd)
 597 {
 598         if (generic_permission(inode, mask, NULL) != 0)
 599                 return -EACCES;
 600         return proc_check_root(inode);
 601 }
 602
 603 static int proc_task_permission(struct inode *inode, int mask, struct nameidata *nd)
 604 {
 605         struct dentry *root;
 606         struct vfsmount *vfsmnt;
 607
 608         if (generic_permission(inode, mask, NULL) != 0)
 609                 return -EACCES;
 610
 611         if (proc_task_root_link(inode, &root, &vfsmnt))
 612                 return -ENOENT;
 613
 614         return proc_check_chroot(root, vfsmnt);
 615 }
 616
 617 extern struct seq_operations proc_pid_maps_op;
 618 static int maps_open(struct inode *inode, struct file *file)
 619 {
 620         struct task_struct *task = proc_task(inode);
 621         int ret = seq_open(file, &proc_pid_maps_op);
 622         if (!ret) {
 623                 struct seq_file *m = file->private_data;
 624                 m->private = task;
 625         }
 626         return ret;
 627 }
 628
 629 static struct file_operations proc_maps_operations = {
 630         .open           = maps_open,
 631         .read           = seq_read,
 632         .llseek         = seq_lseek,
 633         .release        = seq_release,
 634 };
 635
 636 #ifdef CONFIG_NUMA
 637 extern struct seq_operations proc_pid_numa_maps_op;
 638 static int numa_maps_open(struct inode *inode, struct file *file)
 639 {
 640         struct task_struct *task = proc_task(inode);
 641         int ret = seq_open(file, &proc_pid_numa_maps_op);
 642         if (!ret) {
 643                 struct seq_file *m = file->private_data;
 644                 m->private = task;
 645         }
 646         return ret;
 647 }
 648
 649 static struct file_operations proc_numa_maps_operations = {
 650         .open           = numa_maps_open,
 651         .read           = seq_read,
 652         .llseek         = seq_lseek,
 653         .release        = seq_release,
 654 };
 655 #endif
 656
 657 #ifdef CONFIG_MMU
 658 extern struct seq_operations proc_pid_smaps_op;
 659 static int smaps_open(struct inode *inode, struct file *file)
 660 {
 661         struct task_struct *task = proc_task(inode);
 662         int ret = seq_open(file, &proc_pid_smaps_op);
 663         if (!ret) {
 664                 struct seq_file *m = file->private_data;
 665                 m->private = task;
 666         }
 667         return ret;
 668 }
 669
 670 static struct file_operations proc_smaps_operations = {
 671         .open           = smaps_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = seq_release,
 675 };
 676 #endif
 677
 678 extern struct seq_operations mounts_op;
 679 struct proc_mounts {
 680         struct seq_file m;
 681         int event;
 682 };
 683
 684 static int mounts_open(struct inode *inode, struct file *file)
 685 {
 686         struct task_struct *task = proc_task(inode);
 687         struct namespace *namespace;
 688         struct proc_mounts *p;
 689         int ret = -EINVAL;
 690
 691         task_lock(task);
 692         namespace = task->namespace;
 693         if (namespace)
 694                 get_namespace(namespace);
 695         task_unlock(task);
 696
 697         if (namespace) {
 698                 ret = -ENOMEM;
 699                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 700                 if (p) {
 701                         file->private_data = &p->m;
 702                         ret = seq_open(file, &mounts_op);
 703                         if (!ret) {
 704                                 p->m.private = namespace;
 705                                 p->event = namespace->event;
 706                                 return 0;
 707                         }
 708                         kfree(p);
 709                 }
 710                 put_namespace(namespace);
 711         }
 712         return ret;
 713 }
 714
 715 static int mounts_release(struct inode *inode, struct file *file)
 716 {
 717         struct seq_file *m = file->private_data;
 718         struct namespace *namespace = m->private;
 719         put_namespace(namespace);
 720         return seq_release(inode, file);
 721 }
 722
 723 static unsigned mounts_poll(struct file *file, poll_table *wait)
 724 {
 725         struct proc_mounts *p = file->private_data;
 726         struct namespace *ns = p->m.private;
 727         unsigned res = 0;
 728
 729         poll_wait(file, &ns->poll, wait);
 730
 731         spin_lock(&vfsmount_lock);
 732         if (p->event != ns->event) {
 733                 p->event = ns->event;
 734                 res = POLLERR;
 735         }
 736         spin_unlock(&vfsmount_lock);
 737
 738         return res;
 739 }
 740
 741 static struct file_operations proc_mounts_operations = {
 742         .open           = mounts_open,
 743         .read           = seq_read,
 744         .llseek         = seq_lseek,
 745         .release        = mounts_release,
 746         .poll           = mounts_poll,
 747 };
 748
 749 extern struct seq_operations mountstats_op;
 750 static int mountstats_open(struct inode *inode, struct file *file)
 751 {
 752         struct task_struct *task = proc_task(inode);
 753         int ret = seq_open(file, &mountstats_op);
 754
 755         if (!ret) {
 756                 struct seq_file *m = file->private_data;
 757                 struct namespace *namespace;
 758                 task_lock(task);
 759                 namespace = task->namespace;
 760                 if (namespace)
 761                         get_namespace(namespace);
 762                 task_unlock(task);
 763
 764                 if (namespace)
 765                         m->private = namespace;
 766                 else {
 767                         seq_release(inode, file);
 768                         ret = -EINVAL;
 769                 }
 770         }
 771         return ret;
 772 }
 773
 774 static struct file_operations proc_mountstats_operations = {
 775         .open           = mountstats_open,
 776         .read           = seq_read,
 777         .llseek         = seq_lseek,
 778         .release        = mounts_release,
 779 };
 780
 781 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 782
 783 static ssize_t proc_info_read(struct file * file, char __user * buf,
 784                           size_t count, loff_t *ppos)
 785 {
 786         struct inode * inode = file->f_dentry->d_inode;
 787         unsigned long page;
 788         ssize_t length;
 789         struct task_struct *task = proc_task(inode);
 790
 791         if (count > PROC_BLOCK_SIZE)
 792                 count = PROC_BLOCK_SIZE;
 793         if (!(page = __get_free_page(GFP_KERNEL)))
 794                 return -ENOMEM;
 795
 796         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 797
 798         if (length >= 0)
 799                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 800         free_page(page);
 801         return length;
 802 }
 803
 804 static struct file_operations proc_info_file_operations = {
 805         .read           = proc_info_read,
 806 };
 807
 808 static int mem_open(struct inode* inode, struct file* file)
 809 {
 810         file->private_data = (void*)((long)current->self_exec_id);
 811         return 0;
 812 }
 813
 814 static ssize_t mem_read(struct file * file, char __user * buf,
 815                         size_t count, loff_t *ppos)
 816 {
 817         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 818         char *page;
 819         unsigned long src = *ppos;
 820         int ret = -ESRCH;
 821         struct mm_struct *mm;
 822
 823         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 824                 goto out;
 825
 826         ret = -ENOMEM;
 827         page = (char *)__get_free_page(GFP_USER);
 828         if (!page)
 829                 goto out;
 830
 831         ret = 0;
 832
 833         mm = get_task_mm(task);
 834         if (!mm)
 835                 goto out_free;
 836
 837         ret = -EIO;
 838
 839         if (file->private_data != (void*)((long)current->self_exec_id))
 840                 goto out_put;
 841
 842         ret = 0;
 843
 844         while (count > 0) {
 845                 int this_len, retval;
 846
 847                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 848                 retval = access_process_vm(task, src, page, this_len, 0);
 849                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 850                         if (!ret)
 851                                 ret = -EIO;
 852                         break;
 853                 }
 854
 855                 if (copy_to_user(buf, page, retval)) {
 856                         ret = -EFAULT;
 857                         break;
 858                 }
 859
 860                 ret += retval;
 861                 src += retval;
 862                 buf += retval;
 863                 count -= retval;
 864         }
 865         *ppos = src;
 866
 867 out_put:
 868         mmput(mm);
 869 out_free:
 870         free_page((unsigned long) page);
 871 out:
 872         return ret;
 873 }
 874
 875 #define mem_write NULL
 876
 877 #ifndef mem_write
 878 /* This is a security hazard */
 879 static ssize_t mem_write(struct file * file, const char * buf,
 880                          size_t count, loff_t *ppos)
 881 {
 882         int copied = 0;
 883         char *page;
 884         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 885         unsigned long dst = *ppos;
 886
 887         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 888                 return -ESRCH;
 889
 890         page = (char *)__get_free_page(GFP_USER);
 891         if (!page)
 892                 return -ENOMEM;
 893
 894         while (count > 0) {
 895                 int this_len, retval;
 896
 897                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 898                 if (copy_from_user(page, buf, this_len)) {
 899                         copied = -EFAULT;
 900                         break;
 901                 }
 902                 retval = access_process_vm(task, dst, page, this_len, 1);
 903                 if (!retval) {
 904                         if (!copied)
 905                                 copied = -EIO;
 906                         break;
 907                 }
 908                 copied += retval;
 909                 buf += retval;
 910                 dst += retval;
 911                 count -= retval;
 912         }
 913         *ppos = dst;
 914         free_page((unsigned long) page);
 915         return copied;
 916 }
 917 #endif
 918
 919 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 920 {
 921         switch (orig) {
 922         case 0:
 923                 file->f_pos = offset;
 924                 break;
 925         case 1:
 926                 file->f_pos += offset;
 927                 break;
 928         default:
 929                 return -EINVAL;
 930         }
 931         force_successful_syscall_return();
 932         return file->f_pos;
 933 }
 934
 935 static struct file_operations proc_mem_operations = {
 936         .llseek         = mem_lseek,
 937         .read           = mem_read,
 938         .write          = mem_write,
 939         .open           = mem_open,
 940 };
 941
 942 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 943                                 size_t count, loff_t *ppos)
 944 {
 945         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 946         char buffer[8];
 947         size_t len;
 948         int oom_adjust = task->oomkilladj;
 949         loff_t __ppos = *ppos;
 950
 951         len = sprintf(buffer, "%i\n", oom_adjust);
 952         if (__ppos >= len)
 953                 return 0;
 954         if (count > len-__ppos)
 955                 count = len-__ppos;
 956         if (copy_to_user(buf, buffer + __ppos, count))
 957                 return -EFAULT;
 958         *ppos = __ppos + count;
 959         return count;
 960 }
 961
 962 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 963                                 size_t count, loff_t *ppos)
 964 {
 965         struct task_struct *task = proc_task(file->f_dentry->d_inode);
 966         char buffer[8], *end;
 967         int oom_adjust;
 968
 969         if (!capable(CAP_SYS_RESOURCE))
 970                 return -EPERM;
 971         memset(buffer, 0, 8);
 972         if (count > 6)
 973                 count = 6;
 974         if (copy_from_user(buffer, buf, count))
 975                 return -EFAULT;
 976         oom_adjust = simple_strtol(buffer, &end, 0);
 977         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 978                 return -EINVAL;
 979         if (*end == '\n')
 980                 end++;
 981         task->oomkilladj = oom_adjust;
 982         if (end - buffer == 0)
 983                 return -EIO;
 984         return end - buffer;
 985 }
 986
 987 static struct file_operations proc_oom_adjust_operations = {
 988         .read           = oom_adjust_read,
 989         .write          = oom_adjust_write,
 990 };
 991
 992 static struct inode_operations proc_mem_inode_operations = {
 993         .permission     = proc_permission,
 994 };
 995
 996 #ifdef CONFIG_AUDITSYSCALL
 997 #define TMPBUFLEN 21
 998 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 999                                   size_t count, loff_t *ppos)
1000 {
1001         struct inode * inode = file->f_dentry->d_inode;
1002         struct task_struct *task = proc_task(inode);
1003         ssize_t length;
1004         char tmpbuf[TMPBUFLEN];
1005
1006         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1007                                 audit_get_loginuid(task->audit_context));
1008         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1009 }
1010
1011 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1012                                    size_t count, loff_t *ppos)
1013 {
1014         struct inode * inode = file->f_dentry->d_inode;
1015         char *page, *tmp;
1016         ssize_t length;
1017         struct task_struct *task = proc_task(inode);
1018         uid_t loginuid;
1019
1020         if (!capable(CAP_AUDIT_CONTROL))
1021                 return -EPERM;
1022
1023         if (current != task)
1024                 return -EPERM;
1025
1026         if (count >= PAGE_SIZE)
1027                 count = PAGE_SIZE - 1;
1028
1029         if (*ppos != 0) {
1030                 /* No partial writes. */
1031                 return -EINVAL;
1032         }
1033         page = (char*)__get_free_page(GFP_USER);
1034         if (!page)
1035                 return -ENOMEM;
1036         length = -EFAULT;
1037         if (copy_from_user(page, buf, count))
1038                 goto out_free_page;
1039
1040         page[count] = '\0';
1041         loginuid = simple_strtoul(page, &tmp, 10);
1042         if (tmp == page) {
1043                 length = -EINVAL;
1044                 goto out_free_page;
1045
1046         }
1047         length = audit_set_loginuid(task, loginuid);
1048         if (likely(length == 0))
1049                 length = count;
1050
1051 out_free_page:
1052         free_page((unsigned long) page);
1053         return length;
1054 }
1055
1056 static struct file_operations proc_loginuid_operations = {
1057         .read           = proc_loginuid_read,
1058         .write          = proc_loginuid_write,
1059 };
1060 #endif
1061
1062 #ifdef CONFIG_SECCOMP
1063 static ssize_t seccomp_read(struct file *file, char __user *buf,
1064                             size_t count, loff_t *ppos)
1065 {
1066         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1067         char __buf[20];
1068         loff_t __ppos = *ppos;
1069         size_t len;
1070
1071         /* no need to print the trailing zero, so use only len */
1072         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
1073         if (__ppos >= len)
1074                 return 0;
1075         if (count > len - __ppos)
1076                 count = len - __ppos;
1077         if (copy_to_user(buf, __buf + __ppos, count))
1078                 return -EFAULT;
1079         *ppos = __ppos + count;
1080         return count;
1081 }
1082
1083 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1084                              size_t count, loff_t *ppos)
1085 {
1086         struct task_struct *tsk = proc_task(file->f_dentry->d_inode);
1087         char __buf[20], *end;
1088         unsigned int seccomp_mode;
1089
1090         /* can set it only once to be even more secure */
1091         if (unlikely(tsk->seccomp.mode))
1092                 return -EPERM;
1093
1094         memset(__buf, 0, sizeof(__buf));
1095         count = min(count, sizeof(__buf) - 1);
1096         if (copy_from_user(__buf, buf, count))
1097                 return -EFAULT;
1098         seccomp_mode = simple_strtoul(__buf, &end, 0);
1099         if (*end == '\n')
1100                 end++;
1101         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1102                 tsk->seccomp.mode = seccomp_mode;
1103                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1104         } else
1105                 return -EINVAL;
1106         if (unlikely(!(end - __buf)))
1107                 return -EIO;
1108         return end - __buf;
1109 }
1110
1111 static struct file_operations proc_seccomp_operations = {
1112         .read           = seccomp_read,
1113         .write          = seccomp_write,
1114 };
1115 #endif /* CONFIG_SECCOMP */
1116
1117 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1118 {
1119         struct inode *inode = dentry->d_inode;
1120         int error = -EACCES;
1121
1122         /* We don't need a base pointer in the /proc filesystem */
1123         path_release(nd);
1124
1125         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1126                 goto out;
1127         error = proc_check_root(inode);
1128         if (error)
1129                 goto out;
1130
1131         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1132         nd->last_type = LAST_BIND;
1133 out:
1134         return ERR_PTR(error);
1135 }
1136
1137 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1138                             char __user *buffer, int buflen)
1139 {
1140         struct inode * inode;
1141         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1142         int len;
1143
1144         if (!tmp)
1145                 return -ENOMEM;
1146
1147         inode = dentry->d_inode;
1148         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1149         len = PTR_ERR(path);
1150         if (IS_ERR(path))
1151                 goto out;
1152         len = tmp + PAGE_SIZE - 1 - path;
1153
1154         if (len > buflen)
1155                 len = buflen;
1156         if (copy_to_user(buffer, path, len))
1157                 len = -EFAULT;
1158  out:
1159         free_page((unsigned long)tmp);
1160         return len;
1161 }
1162
1163 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1164 {
1165         int error = -EACCES;
1166         struct inode *inode = dentry->d_inode;
1167         struct dentry *de;
1168         struct vfsmount *mnt = NULL;
1169
1170
1171         if (current->fsuid != inode->i_uid && !capable(CAP_DAC_OVERRIDE))
1172                 goto out;
1173         error = proc_check_root(inode);
1174         if (error)
1175                 goto out;
1176
1177         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1178         if (error)
1179                 goto out;
1180
1181         error = do_proc_readlink(de, mnt, buffer, buflen);
1182         dput(de);
1183         mntput(mnt);
1184 out:
1185         return error;
1186 }
1187
1188 static struct inode_operations proc_pid_link_inode_operations = {
1189         .readlink       = proc_pid_readlink,
1190         .follow_link    = proc_pid_follow_link
1191 };
1192
1193 #define NUMBUF 10
1194
1195 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1196 {
1197         struct dentry *dentry = filp->f_dentry;
1198         struct inode *inode = dentry->d_inode;
1199         struct task_struct *p = proc_task(inode);
1200         unsigned int fd, tid, ino;
1201         int retval;
1202         char buf[NUMBUF];
1203         struct files_struct * files;
1204         struct fdtable *fdt;
1205
1206         retval = -ENOENT;
1207         if (!pid_alive(p))
1208                 goto out;
1209         retval = 0;
1210         tid = p->pid;
1211
1212         fd = filp->f_pos;
1213         switch (fd) {
1214                 case 0:
1215                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1216                                 goto out;
1217                         filp->f_pos++;
1218                 case 1:
1219                         ino = parent_ino(dentry);
1220                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1221                                 goto out;
1222                         filp->f_pos++;
1223                 default:
1224                         files = get_files_struct(p);
1225                         if (!files)
1226                                 goto out;
1227                         rcu_read_lock();
1228                         fdt = files_fdtable(files);
1229                         for (fd = filp->f_pos-2;
1230                              fd < fdt->max_fds;
1231                              fd++, filp->f_pos++) {
1232                                 unsigned int i,j;
1233
1234                                 if (!fcheck_files(files, fd))
1235                                         continue;
1236                                 rcu_read_unlock();
1237
1238                                 j = NUMBUF;
1239                                 i = fd;
1240                                 do {
1241                                         j--;
1242                                         buf[j] = '0' + (i % 10);
1243                                         i /= 10;
1244                                 } while (i);
1245
1246                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1247                                 if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1248                                         rcu_read_lock();
1249                                         break;
1250                                 }
1251                                 rcu_read_lock();
1252                         }
1253                         rcu_read_unlock();
1254                         put_files_struct(files);
1255         }
1256 out:
1257         return retval;
1258 }
1259
1260 static int proc_pident_readdir(struct file *filp,
1261                 void *dirent, filldir_t filldir,
1262                 struct pid_entry *ents, unsigned int nents)
1263 {
1264         int i;
1265         int pid;
1266         struct dentry *dentry = filp->f_dentry;
1267         struct inode *inode = dentry->d_inode;
1268         struct pid_entry *p;
1269         ino_t ino;
1270         int ret;
1271
1272         ret = -ENOENT;
1273         if (!pid_alive(proc_task(inode)))
1274                 goto out;
1275
1276         ret = 0;
1277         pid = proc_task(inode)->pid;
1278         i = filp->f_pos;
1279         switch (i) {
1280         case 0:
1281                 ino = inode->i_ino;
1282                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1283                         goto out;
1284                 i++;
1285                 filp->f_pos++;
1286                 /* fall through */
1287         case 1:
1288                 ino = parent_ino(dentry);
1289                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1290                         goto out;
1291                 i++;
1292                 filp->f_pos++;
1293                 /* fall through */
1294         default:
1295                 i -= 2;
1296                 if (i >= nents) {
1297                         ret = 1;
1298                         goto out;
1299                 }
1300                 p = ents + i;
1301                 while (p->name) {
1302                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1303                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1304                                 goto out;
1305                         filp->f_pos++;
1306                         p++;
1307                 }
1308         }
1309
1310         ret = 1;
1311 out:
1312         return ret;
1313 }
1314
1315 static int proc_tgid_base_readdir(struct file * filp,
1316                              void * dirent, filldir_t filldir)
1317 {
1318         return proc_pident_readdir(filp,dirent,filldir,
1319                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1320 }
1321
1322 static int proc_tid_base_readdir(struct file * filp,
1323                              void * dirent, filldir_t filldir)
1324 {
1325         return proc_pident_readdir(filp,dirent,filldir,
1326                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1327 }
1328
1329 /* building an inode */
1330
1331 static int task_dumpable(struct task_struct *task)
1332 {
1333         int dumpable = 0;
1334         struct mm_struct *mm;
1335
1336         task_lock(task);
1337         mm = task->mm;
1338         if (mm)
1339                 dumpable = mm->dumpable;
1340         task_unlock(task);
1341         if(dumpable == 1)
1342                 return 1;
1343         return 0;
1344 }
1345
1346
1347 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1348 {
1349         struct inode * inode;
1350         struct proc_inode *ei;
1351
1352         /* We need a new inode */
1353
1354         inode = new_inode(sb);
1355         if (!inode)
1356                 goto out;
1357
1358         /* Common stuff */
1359         ei = PROC_I(inode);
1360         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1361         inode->i_ino = fake_ino(task->pid, ino);
1362
1363         if (!pid_alive(task))
1364                 goto out_unlock;
1365
1366         /*
1367          * grab the reference to task.
1368          */
1369         get_task_struct(task);
1370         ei->task = task;
1371         inode->i_uid = 0;
1372         inode->i_gid = 0;
1373         if (task_dumpable(task)) {
1374                 inode->i_uid = task->euid;
1375                 inode->i_gid = task->egid;
1376         }
1377         security_task_to_inode(task, inode);
1378
1379 out:
1380         return inode;
1381
1382 out_unlock:
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 (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_fd(inode);
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         ei->fd = fd;
1528         files = get_files_struct(task);
1529         if (!files)
1530                 goto out_unlock;
1531         inode->i_mode = S_IFLNK;
1532
1533         /*
1534          * We are not taking a ref to the file structure, so we must
1535          * hold ->file_lock.
1536          */
1537         spin_lock(&files->file_lock);
1538         file = fcheck_files(files, fd);
1539         if (!file)
1540                 goto out_unlock2;
1541         if (file->f_mode & 1)
1542                 inode->i_mode |= S_IRUSR | S_IXUSR;
1543         if (file->f_mode & 2)
1544                 inode->i_mode |= S_IWUSR | S_IXUSR;
1545         spin_unlock(&files->file_lock);
1546         put_files_struct(files);
1547         inode->i_op = &proc_pid_link_inode_operations;
1548         inode->i_size = 64;
1549         ei->op.proc_get_link = proc_fd_link;
1550         dentry->d_op = &tid_fd_dentry_operations;
1551         d_add(dentry, inode);
1552         return NULL;
1553
1554 out_unlock2:
1555         spin_unlock(&files->file_lock);
1556         put_files_struct(files);
1557 out_unlock:
1558         iput(inode);
1559 out:
1560         return ERR_PTR(-ENOENT);
1561 }
1562
1563 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1564 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1565
1566 static struct file_operations proc_fd_operations = {
1567         .read           = generic_read_dir,
1568         .readdir        = proc_readfd,
1569 };
1570
1571 static struct file_operations proc_task_operations = {
1572         .read           = generic_read_dir,
1573         .readdir        = proc_task_readdir,
1574 };
1575
1576 /*
1577  * proc directories can do almost nothing..
1578  */
1579 static struct inode_operations proc_fd_inode_operations = {
1580         .lookup         = proc_lookupfd,
1581         .permission     = proc_permission,
1582 };
1583
1584 static struct inode_operations proc_task_inode_operations = {
1585         .lookup         = proc_task_lookup,
1586         .permission     = proc_task_permission,
1587 };
1588
1589 #ifdef CONFIG_SECURITY
1590 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1591                                   size_t count, loff_t *ppos)
1592 {
1593         struct inode * inode = file->f_dentry->d_inode;
1594         unsigned long page;
1595         ssize_t length;
1596         struct task_struct *task = proc_task(inode);
1597
1598         if (count > PAGE_SIZE)
1599                 count = PAGE_SIZE;
1600         if (!(page = __get_free_page(GFP_KERNEL)))
1601                 return -ENOMEM;
1602
1603         length = security_getprocattr(task,
1604                                       (char*)file->f_dentry->d_name.name,
1605                                       (void*)page, count);
1606         if (length >= 0)
1607                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1608         free_page(page);
1609         return length;
1610 }
1611
1612 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1613                                    size_t count, loff_t *ppos)
1614 {
1615         struct inode * inode = file->f_dentry->d_inode;
1616         char *page;
1617         ssize_t length;
1618         struct task_struct *task = proc_task(inode);
1619
1620         if (count > PAGE_SIZE)
1621                 count = PAGE_SIZE;
1622         if (*ppos != 0) {
1623                 /* No partial writes. */
1624                 return -EINVAL;
1625         }
1626         page = (char*)__get_free_page(GFP_USER);
1627         if (!page)
1628                 return -ENOMEM;
1629         length = -EFAULT;
1630         if (copy_from_user(page, buf, count))
1631                 goto out;
1632
1633         length = security_setprocattr(task,
1634                                       (char*)file->f_dentry->d_name.name,
1635                                       (void*)page, count);
1636 out:
1637         free_page((unsigned long) page);
1638         return length;
1639 }
1640
1641 static struct file_operations proc_pid_attr_operations = {
1642         .read           = proc_pid_attr_read,
1643         .write          = proc_pid_attr_write,
1644 };
1645
1646 static struct file_operations proc_tid_attr_operations;
1647 static struct inode_operations proc_tid_attr_inode_operations;
1648 static struct file_operations proc_tgid_attr_operations;
1649 static struct inode_operations proc_tgid_attr_inode_operations;
1650 #endif
1651
1652 static int get_tid_list(int index, unsigned int *tids, struct inode *dir);
1653
1654 /* SMP-safe */
1655 static struct dentry *proc_pident_lookup(struct inode *dir,
1656                                          struct dentry *dentry,
1657                                          struct pid_entry *ents)
1658 {
1659         struct inode *inode;
1660         int error;
1661         struct task_struct *task = proc_task(dir);
1662         struct pid_entry *p;
1663         struct proc_inode *ei;
1664
1665         error = -ENOENT;
1666         inode = NULL;
1667
1668         if (!pid_alive(task))
1669                 goto out;
1670
1671         for (p = ents; p->name; p++) {
1672                 if (p->len != dentry->d_name.len)
1673                         continue;
1674                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1675                         break;
1676         }
1677         if (!p->name)
1678                 goto out;
1679
1680         error = -EINVAL;
1681         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1682         if (!inode)
1683                 goto out;
1684
1685         ei = PROC_I(inode);
1686         inode->i_mode = p->mode;
1687         /*
1688          * Yes, it does not scale. And it should not. Don't add
1689          * new entries into /proc/<tgid>/ without very good reasons.
1690          */
1691         switch(p->type) {
1692                 case PROC_TGID_TASK:
1693                         inode->i_nlink = 2 + get_tid_list(2, NULL, dir);
1694                         inode->i_op = &proc_task_inode_operations;
1695                         inode->i_fop = &proc_task_operations;
1696                         break;
1697                 case PROC_TID_FD:
1698                 case PROC_TGID_FD:
1699                         inode->i_nlink = 2;
1700                         inode->i_op = &proc_fd_inode_operations;
1701                         inode->i_fop = &proc_fd_operations;
1702                         break;
1703                 case PROC_TID_EXE:
1704                 case PROC_TGID_EXE:
1705                         inode->i_op = &proc_pid_link_inode_operations;
1706                         ei->op.proc_get_link = proc_exe_link;
1707                         break;
1708                 case PROC_TID_CWD:
1709                 case PROC_TGID_CWD:
1710                         inode->i_op = &proc_pid_link_inode_operations;
1711                         ei->op.proc_get_link = proc_cwd_link;
1712                         break;
1713                 case PROC_TID_ROOT:
1714                 case PROC_TGID_ROOT:
1715                         inode->i_op = &proc_pid_link_inode_operations;
1716                         ei->op.proc_get_link = proc_root_link;
1717                         break;
1718                 case PROC_TID_ENVIRON:
1719                 case PROC_TGID_ENVIRON:
1720                         inode->i_fop = &proc_info_file_operations;
1721                         ei->op.proc_read = proc_pid_environ;
1722                         break;
1723                 case PROC_TID_AUXV:
1724                 case PROC_TGID_AUXV:
1725                         inode->i_fop = &proc_info_file_operations;
1726                         ei->op.proc_read = proc_pid_auxv;
1727                         break;
1728                 case PROC_TID_STATUS:
1729                 case PROC_TGID_STATUS:
1730                         inode->i_fop = &proc_info_file_operations;
1731                         ei->op.proc_read = proc_pid_status;
1732                         break;
1733                 case PROC_TID_STAT:
1734                         inode->i_fop = &proc_info_file_operations;
1735                         ei->op.proc_read = proc_tid_stat;
1736                         break;
1737                 case PROC_TGID_STAT:
1738                         inode->i_fop = &proc_info_file_operations;
1739                         ei->op.proc_read = proc_tgid_stat;
1740                         break;
1741                 case PROC_TID_CMDLINE:
1742                 case PROC_TGID_CMDLINE:
1743                         inode->i_fop = &proc_info_file_operations;
1744                         ei->op.proc_read = proc_pid_cmdline;
1745                         break;
1746                 case PROC_TID_STATM:
1747                 case PROC_TGID_STATM:
1748                         inode->i_fop = &proc_info_file_operations;
1749                         ei->op.proc_read = proc_pid_statm;
1750                         break;
1751                 case PROC_TID_MAPS:
1752                 case PROC_TGID_MAPS:
1753                         inode->i_fop = &proc_maps_operations;
1754                         break;
1755 #ifdef CONFIG_NUMA
1756                 case PROC_TID_NUMA_MAPS:
1757                 case PROC_TGID_NUMA_MAPS:
1758                         inode->i_fop = &proc_numa_maps_operations;
1759                         break;
1760 #endif
1761                 case PROC_TID_MEM:
1762                 case PROC_TGID_MEM:
1763                         inode->i_op = &proc_mem_inode_operations;
1764                         inode->i_fop = &proc_mem_operations;
1765                         break;
1766 #ifdef CONFIG_SECCOMP
1767                 case PROC_TID_SECCOMP:
1768                 case PROC_TGID_SECCOMP:
1769                         inode->i_fop = &proc_seccomp_operations;
1770                         break;
1771 #endif /* CONFIG_SECCOMP */
1772                 case PROC_TID_MOUNTS:
1773                 case PROC_TGID_MOUNTS:
1774                         inode->i_fop = &proc_mounts_operations;
1775                         break;
1776 #ifdef CONFIG_MMU
1777                 case PROC_TID_SMAPS:
1778                 case PROC_TGID_SMAPS:
1779                         inode->i_fop = &proc_smaps_operations;
1780                         break;
1781 #endif
1782                 case PROC_TID_MOUNTSTATS:
1783                 case PROC_TGID_MOUNTSTATS:
1784                         inode->i_fop = &proc_mountstats_operations;
1785                         break;
1786 #ifdef CONFIG_SECURITY
1787                 case PROC_TID_ATTR:
1788                         inode->i_nlink = 2;
1789                         inode->i_op = &proc_tid_attr_inode_operations;
1790                         inode->i_fop = &proc_tid_attr_operations;
1791                         break;
1792                 case PROC_TGID_ATTR:
1793                         inode->i_nlink = 2;
1794                         inode->i_op = &proc_tgid_attr_inode_operations;
1795                         inode->i_fop = &proc_tgid_attr_operations;
1796                         break;
1797                 case PROC_TID_ATTR_CURRENT:
1798                 case PROC_TGID_ATTR_CURRENT:
1799                 case PROC_TID_ATTR_PREV:
1800                 case PROC_TGID_ATTR_PREV:
1801                 case PROC_TID_ATTR_EXEC:
1802                 case PROC_TGID_ATTR_EXEC:
1803                 case PROC_TID_ATTR_FSCREATE:
1804                 case PROC_TGID_ATTR_FSCREATE:
1805                 case PROC_TID_ATTR_KEYCREATE:
1806                 case PROC_TGID_ATTR_KEYCREATE:
1807                         inode->i_fop = &proc_pid_attr_operations;
1808                         break;
1809 #endif
1810 #ifdef CONFIG_KALLSYMS
1811                 case PROC_TID_WCHAN:
1812                 case PROC_TGID_WCHAN:
1813                         inode->i_fop = &proc_info_file_operations;
1814                         ei->op.proc_read = proc_pid_wchan;
1815                         break;
1816 #endif
1817 #ifdef CONFIG_SCHEDSTATS
1818                 case PROC_TID_SCHEDSTAT:
1819                 case PROC_TGID_SCHEDSTAT:
1820                         inode->i_fop = &proc_info_file_operations;
1821                         ei->op.proc_read = proc_pid_schedstat;
1822                         break;
1823 #endif
1824 #ifdef CONFIG_CPUSETS
1825                 case PROC_TID_CPUSET:
1826                 case PROC_TGID_CPUSET:
1827                         inode->i_fop = &proc_cpuset_operations;
1828                         break;
1829 #endif
1830                 case PROC_TID_OOM_SCORE:
1831                 case PROC_TGID_OOM_SCORE:
1832                         inode->i_fop = &proc_info_file_operations;
1833                         ei->op.proc_read = proc_oom_score;
1834                         break;
1835                 case PROC_TID_OOM_ADJUST:
1836                 case PROC_TGID_OOM_ADJUST:
1837                         inode->i_fop = &proc_oom_adjust_operations;
1838                         break;
1839 #ifdef CONFIG_AUDITSYSCALL
1840                 case PROC_TID_LOGINUID:
1841                 case PROC_TGID_LOGINUID:
1842                         inode->i_fop = &proc_loginuid_operations;
1843                         break;
1844 #endif
1845                 default:
1846                         printk("procfs: impossible type (%d)",p->type);
1847                         iput(inode);
1848                         return ERR_PTR(-EINVAL);
1849         }
1850         dentry->d_op = &pid_dentry_operations;
1851         d_add(dentry, inode);
1852         return NULL;
1853
1854 out:
1855         return ERR_PTR(error);
1856 }
1857
1858 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1859         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1860 }
1861
1862 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1863         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1864 }
1865
1866 static struct file_operations proc_tgid_base_operations = {
1867         .read           = generic_read_dir,
1868         .readdir        = proc_tgid_base_readdir,
1869 };
1870
1871 static struct file_operations proc_tid_base_operations = {
1872         .read           = generic_read_dir,
1873         .readdir        = proc_tid_base_readdir,
1874 };
1875
1876 static struct inode_operations proc_tgid_base_inode_operations = {
1877         .lookup         = proc_tgid_base_lookup,
1878 };
1879
1880 static struct inode_operations proc_tid_base_inode_operations = {
1881         .lookup         = proc_tid_base_lookup,
1882 };
1883
1884 #ifdef CONFIG_SECURITY
1885 static int proc_tgid_attr_readdir(struct file * filp,
1886                              void * dirent, filldir_t filldir)
1887 {
1888         return proc_pident_readdir(filp,dirent,filldir,
1889                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1890 }
1891
1892 static int proc_tid_attr_readdir(struct file * filp,
1893                              void * dirent, filldir_t filldir)
1894 {
1895         return proc_pident_readdir(filp,dirent,filldir,
1896                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1897 }
1898
1899 static struct file_operations proc_tgid_attr_operations = {
1900         .read           = generic_read_dir,
1901         .readdir        = proc_tgid_attr_readdir,
1902 };
1903
1904 static struct file_operations proc_tid_attr_operations = {
1905         .read           = generic_read_dir,
1906         .readdir        = proc_tid_attr_readdir,
1907 };
1908
1909 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1910                                 struct dentry *dentry, struct nameidata *nd)
1911 {
1912         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1913 }
1914
1915 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1916                                 struct dentry *dentry, struct nameidata *nd)
1917 {
1918         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1919 }
1920
1921 static struct inode_operations proc_tgid_attr_inode_operations = {
1922         .lookup         = proc_tgid_attr_lookup,
1923 };
1924
1925 static struct inode_operations proc_tid_attr_inode_operations = {
1926         .lookup         = proc_tid_attr_lookup,
1927 };
1928 #endif
1929
1930 /*
1931  * /proc/self:
1932  */
1933 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1934                               int buflen)
1935 {
1936         char tmp[30];
1937         sprintf(tmp, "%d", current->tgid);
1938         return vfs_readlink(dentry,buffer,buflen,tmp);
1939 }
1940
1941 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1942 {
1943         char tmp[30];
1944         sprintf(tmp, "%d", current->tgid);
1945         return ERR_PTR(vfs_follow_link(nd,tmp));
1946 }
1947
1948 static struct inode_operations proc_self_inode_operations = {
1949         .readlink       = proc_self_readlink,
1950         .follow_link    = proc_self_follow_link,
1951 };
1952
1953 /**
1954  * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
1955  * @p: task that should be flushed.
1956  *
1957  * Drops the /proc/@pid dcache entry from the hash chains.
1958  *
1959  * Dropping /proc/@pid entries and detach_pid must be synchroneous,
1960  * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
1961  * if the pid value is immediately reused. This is enforced by
1962  * - caller must acquire spin_lock(p->proc_lock)
1963  * - must be called before detach_pid()
1964  * - proc_pid_lookup acquires proc_lock, and checks that
1965  *   the target is not dead by looking at the attach count
1966  *   of PIDTYPE_PID.
1967  */
1968
1969 struct dentry *proc_pid_unhash(struct task_struct *p)
1970 {
1971         struct dentry *proc_dentry;
1972
1973         proc_dentry = p->proc_dentry;
1974         if (proc_dentry != NULL) {
1975
1976                 spin_lock(&dcache_lock);
1977                 spin_lock(&proc_dentry->d_lock);
1978                 if (!d_unhashed(proc_dentry)) {
1979                         dget_locked(proc_dentry);
1980                         __d_drop(proc_dentry);
1981                         spin_unlock(&proc_dentry->d_lock);
1982                 } else {
1983                         spin_unlock(&proc_dentry->d_lock);
1984                         proc_dentry = NULL;
1985                 }
1986                 spin_unlock(&dcache_lock);
1987         }
1988         return proc_dentry;
1989 }
1990
1991 /**
1992  * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
1993  * @proc_dentry: directoy to prune.
1994  *
1995  * Shrink the /proc directory that was used by the just killed thread.
1996  */
1997
1998 void proc_pid_flush(struct dentry *proc_dentry)
1999 {
2000         might_sleep();
2001         if(proc_dentry != NULL) {
2002                 shrink_dcache_parent(proc_dentry);
2003                 dput(proc_dentry);
2004         }
2005 }
2006
2007 /* SMP-safe */
2008 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2009 {
2010         struct task_struct *task;
2011         struct inode *inode;
2012         struct proc_inode *ei;
2013         unsigned tgid;
2014         int died;
2015
2016         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
2017                 inode = new_inode(dir->i_sb);
2018                 if (!inode)
2019                         return ERR_PTR(-ENOMEM);
2020                 ei = PROC_I(inode);
2021                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2022                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
2023                 ei->pde = NULL;
2024                 inode->i_mode = S_IFLNK|S_IRWXUGO;
2025                 inode->i_uid = inode->i_gid = 0;
2026                 inode->i_size = 64;
2027                 inode->i_op = &proc_self_inode_operations;
2028                 d_add(dentry, inode);
2029                 return NULL;
2030         }
2031         tgid = name_to_int(dentry);
2032         if (tgid == ~0U)
2033                 goto out;
2034
2035         read_lock(&tasklist_lock);
2036         task = find_task_by_pid(tgid);
2037         if (task)
2038                 get_task_struct(task);
2039         read_unlock(&tasklist_lock);
2040         if (!task)
2041                 goto out;
2042
2043         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2044
2045
2046         if (!inode) {
2047                 put_task_struct(task);
2048                 goto out;
2049         }
2050         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2051         inode->i_op = &proc_tgid_base_inode_operations;
2052         inode->i_fop = &proc_tgid_base_operations;
2053         inode->i_flags|=S_IMMUTABLE;
2054 #ifdef CONFIG_SECURITY
2055         inode->i_nlink = 5;
2056 #else
2057         inode->i_nlink = 4;
2058 #endif
2059
2060         dentry->d_op = &pid_base_dentry_operations;
2061
2062         died = 0;
2063         d_add(dentry, inode);
2064         spin_lock(&task->proc_lock);
2065         task->proc_dentry = dentry;
2066         if (!pid_alive(task)) {
2067                 dentry = proc_pid_unhash(task);
2068                 died = 1;
2069         }
2070         spin_unlock(&task->proc_lock);
2071
2072         put_task_struct(task);
2073         if (died) {
2074                 proc_pid_flush(dentry);
2075                 goto out;
2076         }
2077         return NULL;
2078 out:
2079         return ERR_PTR(-ENOENT);
2080 }
2081
2082 /* SMP-safe */
2083 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2084 {
2085         struct task_struct *task;
2086         struct task_struct *leader = proc_task(dir);
2087         struct inode *inode;
2088         unsigned tid;
2089
2090         tid = name_to_int(dentry);
2091         if (tid == ~0U)
2092                 goto out;
2093
2094         read_lock(&tasklist_lock);
2095         task = find_task_by_pid(tid);
2096         if (task)
2097                 get_task_struct(task);
2098         read_unlock(&tasklist_lock);
2099         if (!task)
2100                 goto out;
2101         if (leader->tgid != task->tgid)
2102                 goto out_drop_task;
2103
2104         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2105
2106
2107         if (!inode)
2108                 goto out_drop_task;
2109         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2110         inode->i_op = &proc_tid_base_inode_operations;
2111         inode->i_fop = &proc_tid_base_operations;
2112         inode->i_flags|=S_IMMUTABLE;
2113 #ifdef CONFIG_SECURITY
2114         inode->i_nlink = 4;
2115 #else
2116         inode->i_nlink = 3;
2117 #endif
2118
2119         dentry->d_op = &pid_base_dentry_operations;
2120
2121         d_add(dentry, inode);
2122
2123         put_task_struct(task);
2124         return NULL;
2125 out_drop_task:
2126         put_task_struct(task);
2127 out:
2128         return ERR_PTR(-ENOENT);
2129 }
2130
2131 #define PROC_NUMBUF 10
2132 #define PROC_MAXPIDS 20
2133
2134 /*
2135  * Get a few tgid's to return for filldir - we need to hold the
2136  * tasklist lock while doing this, and we must release it before
2137  * we actually do the filldir itself, so we use a temp buffer..
2138  */
2139 static int get_tgid_list(int index, unsigned long version, unsigned int *tgids)
2140 {
2141         struct task_struct *p;
2142         int nr_tgids = 0;
2143
2144         index--;
2145         read_lock(&tasklist_lock);
2146         p = NULL;
2147         if (version) {
2148                 p = find_task_by_pid(version);
2149                 if (p && !thread_group_leader(p))
2150                         p = NULL;
2151         }
2152
2153         if (p)
2154                 index = 0;
2155         else
2156                 p = next_task(&init_task);
2157
2158         for ( ; p != &init_task; p = next_task(p)) {
2159                 int tgid = p->pid;
2160                 if (!pid_alive(p))
2161                         continue;
2162                 if (--index >= 0)
2163                         continue;
2164                 tgids[nr_tgids] = tgid;
2165                 nr_tgids++;
2166                 if (nr_tgids >= PROC_MAXPIDS)
2167                         break;
2168         }
2169         read_unlock(&tasklist_lock);
2170         return nr_tgids;
2171 }
2172
2173 /*
2174  * Get a few tid's to return for filldir - we need to hold the
2175  * tasklist lock while doing this, and we must release it before
2176  * we actually do the filldir itself, so we use a temp buffer..
2177  */
2178 static int get_tid_list(int index, unsigned int *tids, struct inode *dir)
2179 {
2180         struct task_struct *leader_task = proc_task(dir);
2181         struct task_struct *task = leader_task;
2182         int nr_tids = 0;
2183
2184         index -= 2;
2185         read_lock(&tasklist_lock);
2186         /*
2187          * The starting point task (leader_task) might be an already
2188          * unlinked task, which cannot be used to access the task-list
2189          * via next_thread().
2190          */
2191         if (pid_alive(task)) do {
2192                 int tid = task->pid;
2193
2194                 if (--index >= 0)
2195                         continue;
2196                 if (tids != NULL)
2197                         tids[nr_tids] = tid;
2198                 nr_tids++;
2199                 if (nr_tids >= PROC_MAXPIDS)
2200                         break;
2201         } while ((task = next_thread(task)) != leader_task);
2202         read_unlock(&tasklist_lock);
2203         return nr_tids;
2204 }
2205
2206 /* for the /proc/ directory itself, after non-process stuff has been done */
2207 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2208 {
2209         unsigned int tgid_array[PROC_MAXPIDS];
2210         char buf[PROC_NUMBUF];
2211         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2212         unsigned int nr_tgids, i;
2213         int next_tgid;
2214
2215         if (!nr) {
2216                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2217                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2218                         return 0;
2219                 filp->f_pos++;
2220                 nr++;
2221         }
2222
2223         /* f_version caches the tgid value that the last readdir call couldn't
2224          * return. lseek aka telldir automagically resets f_version to 0.
2225          */
2226         next_tgid = filp->f_version;
2227         filp->f_version = 0;
2228         for (;;) {
2229                 nr_tgids = get_tgid_list(nr, next_tgid, tgid_array);
2230                 if (!nr_tgids) {
2231                         /* no more entries ! */
2232                         break;
2233                 }
2234                 next_tgid = 0;
2235
2236                 /* do not use the last found pid, reserve it for next_tgid */
2237                 if (nr_tgids == PROC_MAXPIDS) {
2238                         nr_tgids--;
2239                         next_tgid = tgid_array[nr_tgids];
2240                 }
2241
2242                 for (i=0;i<nr_tgids;i++) {
2243                         int tgid = tgid_array[i];
2244                         ino_t ino = fake_ino(tgid,PROC_TGID_INO);
2245                         unsigned long j = PROC_NUMBUF;
2246
2247                         do
2248                                 buf[--j] = '0' + (tgid % 10);
2249                         while ((tgid /= 10) != 0);
2250
2251                         if (filldir(dirent, buf+j, PROC_NUMBUF-j, filp->f_pos, ino, DT_DIR) < 0) {
2252                                 /* returning this tgid failed, save it as the first
2253                                  * pid for the next readir call */
2254                                 filp->f_version = tgid_array[i];
2255                                 goto out;
2256                         }
2257                         filp->f_pos++;
2258                         nr++;
2259                 }
2260         }
2261 out:
2262         return 0;
2263 }
2264
2265 /* for the /proc/TGID/task/ directories */
2266 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2267 {
2268         unsigned int tid_array[PROC_MAXPIDS];
2269         char buf[PROC_NUMBUF];
2270         unsigned int nr_tids, i;
2271         struct dentry *dentry = filp->f_dentry;
2272         struct inode *inode = dentry->d_inode;
2273         int retval = -ENOENT;
2274         ino_t ino;
2275         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2276
2277         if (!pid_alive(proc_task(inode)))
2278                 goto out;
2279         retval = 0;
2280
2281         switch (pos) {
2282         case 0:
2283                 ino = inode->i_ino;
2284                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2285                         goto out;
2286                 pos++;
2287                 /* fall through */
2288         case 1:
2289                 ino = parent_ino(dentry);
2290                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2291                         goto out;
2292                 pos++;
2293                 /* fall through */
2294         }
2295
2296         nr_tids = get_tid_list(pos, tid_array, inode);
2297         inode->i_nlink = pos + nr_tids;
2298
2299         for (i = 0; i < nr_tids; i++) {
2300                 unsigned long j = PROC_NUMBUF;
2301                 int tid = tid_array[i];
2302
2303                 ino = fake_ino(tid,PROC_TID_INO);
2304
2305                 do
2306                         buf[--j] = '0' + (tid % 10);
2307                 while ((tid /= 10) != 0);
2308
2309                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, pos, ino, DT_DIR) < 0)
2310                         break;
2311                 pos++;
2312         }
2313 out:
2314         filp->f_pos = pos;
2315         return retval;
2316 }