fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/config.h>
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/string.h>
  61 #include <linux/seq_file.h>
  62 #include <linux/namei.h>
  63 #include <linux/namespace.h>
  64 #include <linux/mm.h>
  65 #include <linux/smp_lock.h>
  66 #include <linux/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/mount.h>
  69 #include <linux/security.h>
  70 #include <linux/ptrace.h>
  71 #include <linux/seccomp.h>
  72 #include <linux/cpuset.h>
  73 #include <linux/audit.h>
  74 #include <linux/poll.h>
  75 #include "internal.h"
  76
  77 /* NOTE:
  78  *      Implementing inode permission operations in /proc is almost
  79  *      certainly an error.  Permission checks need to happen during
  80  *      each system call not at open time.  The reason is that most of
  81  *      what we wish to check for permissions in /proc varies at runtime.
  82  *
  83  *      The classic example of a problem is opening file descriptors
  84  *      in /proc for a task before it execs a suid executable.
  85  */
  86
  87 /*
  88  * For hysterical raisins we keep the same inumbers as in the old procfs.
  89  * Feel free to change the macro below - just keep the range distinct from
  90  * inumbers of the rest of procfs (currently those are in 0x0000--0xffff).
  91  * As soon as we'll get a separate superblock we will be able to forget
  92  * about magical ranges too.
  93  */
  94
  95 #define fake_ino(pid,ino) (((pid)<<16)|(ino))
  96
  97 enum pid_directory_inos {
  98         PROC_TGID_INO = 2,
  99         PROC_TGID_TASK,
 100         PROC_TGID_STATUS,
 101         PROC_TGID_MEM,
 102 #ifdef CONFIG_SECCOMP
 103         PROC_TGID_SECCOMP,
 104 #endif
 105         PROC_TGID_CWD,
 106         PROC_TGID_ROOT,
 107         PROC_TGID_EXE,
 108         PROC_TGID_FD,
 109         PROC_TGID_ENVIRON,
 110         PROC_TGID_AUXV,
 111         PROC_TGID_CMDLINE,
 112         PROC_TGID_STAT,
 113         PROC_TGID_STATM,
 114         PROC_TGID_MAPS,
 115         PROC_TGID_NUMA_MAPS,
 116         PROC_TGID_MOUNTS,
 117         PROC_TGID_MOUNTSTATS,
 118         PROC_TGID_WCHAN,
 119 #ifdef CONFIG_MMU
 120         PROC_TGID_SMAPS,
 121 #endif
 122 #ifdef CONFIG_SCHEDSTATS
 123         PROC_TGID_SCHEDSTAT,
 124 #endif
 125 #ifdef CONFIG_CPUSETS
 126         PROC_TGID_CPUSET,
 127 #endif
 128 #ifdef CONFIG_SECURITY
 129         PROC_TGID_ATTR,
 130         PROC_TGID_ATTR_CURRENT,
 131         PROC_TGID_ATTR_PREV,
 132         PROC_TGID_ATTR_EXEC,
 133         PROC_TGID_ATTR_FSCREATE,
 134         PROC_TGID_ATTR_KEYCREATE,
 135         PROC_TGID_ATTR_SOCKCREATE,
 136 #endif
 137 #ifdef CONFIG_AUDITSYSCALL
 138         PROC_TGID_LOGINUID,
 139 #endif
 140         PROC_TGID_OOM_SCORE,
 141         PROC_TGID_OOM_ADJUST,
 142         PROC_TID_INO,
 143         PROC_TID_STATUS,
 144         PROC_TID_MEM,
 145 #ifdef CONFIG_SECCOMP
 146         PROC_TID_SECCOMP,
 147 #endif
 148         PROC_TID_CWD,
 149         PROC_TID_ROOT,
 150         PROC_TID_EXE,
 151         PROC_TID_FD,
 152         PROC_TID_ENVIRON,
 153         PROC_TID_AUXV,
 154         PROC_TID_CMDLINE,
 155         PROC_TID_STAT,
 156         PROC_TID_STATM,
 157         PROC_TID_MAPS,
 158         PROC_TID_NUMA_MAPS,
 159         PROC_TID_MOUNTS,
 160         PROC_TID_MOUNTSTATS,
 161         PROC_TID_WCHAN,
 162 #ifdef CONFIG_MMU
 163         PROC_TID_SMAPS,
 164 #endif
 165 #ifdef CONFIG_SCHEDSTATS
 166         PROC_TID_SCHEDSTAT,
 167 #endif
 168 #ifdef CONFIG_CPUSETS
 169         PROC_TID_CPUSET,
 170 #endif
 171 #ifdef CONFIG_SECURITY
 172         PROC_TID_ATTR,
 173         PROC_TID_ATTR_CURRENT,
 174         PROC_TID_ATTR_PREV,
 175         PROC_TID_ATTR_EXEC,
 176         PROC_TID_ATTR_FSCREATE,
 177         PROC_TID_ATTR_KEYCREATE,
 178         PROC_TID_ATTR_SOCKCREATE,
 179 #endif
 180 #ifdef CONFIG_AUDITSYSCALL
 181         PROC_TID_LOGINUID,
 182 #endif
 183         PROC_TID_OOM_SCORE,
 184         PROC_TID_OOM_ADJUST,
 185
 186         /* Add new entries before this */
 187         PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
 188 };
 189
 190 /* Worst case buffer size needed for holding an integer. */
 191 #define PROC_NUMBUF 10
 192
 193 struct pid_entry {
 194         int type;
 195         int len;
 196         char *name;
 197         mode_t mode;
 198 };
 199
 200 #define E(type,name,mode) {(type),sizeof(name)-1,(name),(mode)}
 201
 202 static struct pid_entry tgid_base_stuff[] = {
 203         E(PROC_TGID_TASK,      "task",    S_IFDIR|S_IRUGO|S_IXUGO),
 204         E(PROC_TGID_FD,        "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 205         E(PROC_TGID_ENVIRON,   "environ", S_IFREG|S_IRUSR),
 206         E(PROC_TGID_AUXV,      "auxv",    S_IFREG|S_IRUSR),
 207         E(PROC_TGID_STATUS,    "status",  S_IFREG|S_IRUGO),
 208         E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 209         E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 210         E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
 211         E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 212 #ifdef CONFIG_NUMA
 213         E(PROC_TGID_NUMA_MAPS, "numa_maps", S_IFREG|S_IRUGO),
 214 #endif
 215         E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 216 #ifdef CONFIG_SECCOMP
 217         E(PROC_TGID_SECCOMP,   "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 218 #endif
 219         E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
 220         E(PROC_TGID_ROOT,      "root",    S_IFLNK|S_IRWXUGO),
 221         E(PROC_TGID_EXE,       "exe",     S_IFLNK|S_IRWXUGO),
 222         E(PROC_TGID_MOUNTS,    "mounts",  S_IFREG|S_IRUGO),
 223         E(PROC_TGID_MOUNTSTATS, "mountstats", S_IFREG|S_IRUSR),
 224 #ifdef CONFIG_MMU
 225         E(PROC_TGID_SMAPS,     "smaps",   S_IFREG|S_IRUGO),
 226 #endif
 227 #ifdef CONFIG_SECURITY
 228         E(PROC_TGID_ATTR,      "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 229 #endif
 230 #ifdef CONFIG_KALLSYMS
 231         E(PROC_TGID_WCHAN,     "wchan",   S_IFREG|S_IRUGO),
 232 #endif
 233 #ifdef CONFIG_SCHEDSTATS
 234         E(PROC_TGID_SCHEDSTAT, "schedstat", S_IFREG|S_IRUGO),
 235 #endif
 236 #ifdef CONFIG_CPUSETS
 237         E(PROC_TGID_CPUSET,    "cpuset",  S_IFREG|S_IRUGO),
 238 #endif
 239         E(PROC_TGID_OOM_SCORE, "oom_score",S_IFREG|S_IRUGO),
 240         E(PROC_TGID_OOM_ADJUST,"oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 241 #ifdef CONFIG_AUDITSYSCALL
 242         E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 243 #endif
 244         {0,0,NULL,0}
 245 };
 246 static struct pid_entry tid_base_stuff[] = {
 247         E(PROC_TID_FD,         "fd",      S_IFDIR|S_IRUSR|S_IXUSR),
 248         E(PROC_TID_ENVIRON,    "environ", S_IFREG|S_IRUSR),
 249         E(PROC_TID_AUXV,       "auxv",    S_IFREG|S_IRUSR),
 250         E(PROC_TID_STATUS,     "status",  S_IFREG|S_IRUGO),
 251         E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 252         E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 253         E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
 254         E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 255 #ifdef CONFIG_NUMA
 256         E(PROC_TID_NUMA_MAPS,  "numa_maps",    S_IFREG|S_IRUGO),
 257 #endif
 258         E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 259 #ifdef CONFIG_SECCOMP
 260         E(PROC_TID_SECCOMP,    "seccomp", S_IFREG|S_IRUSR|S_IWUSR),
 261 #endif
 262         E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
 263         E(PROC_TID_ROOT,       "root",    S_IFLNK|S_IRWXUGO),
 264         E(PROC_TID_EXE,        "exe",     S_IFLNK|S_IRWXUGO),
 265         E(PROC_TID_MOUNTS,     "mounts",  S_IFREG|S_IRUGO),
 266 #ifdef CONFIG_MMU
 267         E(PROC_TID_SMAPS,      "smaps",   S_IFREG|S_IRUGO),
 268 #endif
 269 #ifdef CONFIG_SECURITY
 270         E(PROC_TID_ATTR,       "attr",    S_IFDIR|S_IRUGO|S_IXUGO),
 271 #endif
 272 #ifdef CONFIG_KALLSYMS
 273         E(PROC_TID_WCHAN,      "wchan",   S_IFREG|S_IRUGO),
 274 #endif
 275 #ifdef CONFIG_SCHEDSTATS
 276         E(PROC_TID_SCHEDSTAT, "schedstat",S_IFREG|S_IRUGO),
 277 #endif
 278 #ifdef CONFIG_CPUSETS
 279         E(PROC_TID_CPUSET,     "cpuset",  S_IFREG|S_IRUGO),
 280 #endif
 281         E(PROC_TID_OOM_SCORE,  "oom_score",S_IFREG|S_IRUGO),
 282         E(PROC_TID_OOM_ADJUST, "oom_adj", S_IFREG|S_IRUGO|S_IWUSR),
 283 #ifdef CONFIG_AUDITSYSCALL
 284         E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
 285 #endif
 286         {0,0,NULL,0}
 287 };
 288
 289 #ifdef CONFIG_SECURITY
 290 static struct pid_entry tgid_attr_stuff[] = {
 291         E(PROC_TGID_ATTR_CURRENT,  "current",  S_IFREG|S_IRUGO|S_IWUGO),
 292         E(PROC_TGID_ATTR_PREV,     "prev",     S_IFREG|S_IRUGO),
 293         E(PROC_TGID_ATTR_EXEC,     "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 294         E(PROC_TGID_ATTR_FSCREATE, "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 295         E(PROC_TGID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 296         E(PROC_TGID_ATTR_SOCKCREATE, "sockcreate", S_IFREG|S_IRUGO|S_IWUGO),
 297         {0,0,NULL,0}
 298 };
 299 static struct pid_entry tid_attr_stuff[] = {
 300         E(PROC_TID_ATTR_CURRENT,   "current",  S_IFREG|S_IRUGO|S_IWUGO),
 301         E(PROC_TID_ATTR_PREV,      "prev",     S_IFREG|S_IRUGO),
 302         E(PROC_TID_ATTR_EXEC,      "exec",     S_IFREG|S_IRUGO|S_IWUGO),
 303         E(PROC_TID_ATTR_FSCREATE,  "fscreate", S_IFREG|S_IRUGO|S_IWUGO),
 304         E(PROC_TID_ATTR_KEYCREATE, "keycreate", S_IFREG|S_IRUGO|S_IWUGO),
 305         E(PROC_TID_ATTR_SOCKCREATE, "sockcreate", S_IFREG|S_IRUGO|S_IWUGO),
 306         {0,0,NULL,0}
 307 };
 308 #endif
 309
 310 #undef E
 311
 312 static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 313 {
 314         struct task_struct *task = get_proc_task(inode);
 315         struct files_struct *files = NULL;
 316         struct file *file;
 317         int fd = proc_fd(inode);
 318
 319         if (task) {
 320                 files = get_files_struct(task);
 321                 put_task_struct(task);
 322         }
 323         if (files) {
 324                 /*
 325                  * We are not taking a ref to the file structure, so we must
 326                  * hold ->file_lock.
 327                  */
 328                 spin_lock(&files->file_lock);
 329                 file = fcheck_files(files, fd);
 330                 if (file) {
 331                         *mnt = mntget(file->f_vfsmnt);
 332                         *dentry = dget(file->f_dentry);
 333                         spin_unlock(&files->file_lock);
 334                         put_files_struct(files);
 335                         return 0;
 336                 }
 337                 spin_unlock(&files->file_lock);
 338                 put_files_struct(files);
 339         }
 340         return -ENOENT;
 341 }
 342
 343 static struct fs_struct *get_fs_struct(struct task_struct *task)
 344 {
 345         struct fs_struct *fs;
 346         task_lock(task);
 347         fs = task->fs;
 348         if(fs)
 349                 atomic_inc(&fs->count);
 350         task_unlock(task);
 351         return fs;
 352 }
 353
 354 static int get_nr_threads(struct task_struct *tsk)
 355 {
 356         /* Must be called with the rcu_read_lock held */
 357         unsigned long flags;
 358         int count = 0;
 359
 360         if (lock_task_sighand(tsk, &flags)) {
 361                 count = atomic_read(&tsk->signal->count);
 362                 unlock_task_sighand(tsk, &flags);
 363         }
 364         return count;
 365 }
 366
 367 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 368 {
 369         struct task_struct *task = get_proc_task(inode);
 370         struct fs_struct *fs = NULL;
 371         int result = -ENOENT;
 372
 373         if (task) {
 374                 fs = get_fs_struct(task);
 375                 put_task_struct(task);
 376         }
 377         if (fs) {
 378                 read_lock(&fs->lock);
 379                 *mnt = mntget(fs->pwdmnt);
 380                 *dentry = dget(fs->pwd);
 381                 read_unlock(&fs->lock);
 382                 result = 0;
 383                 put_fs_struct(fs);
 384         }
 385         return result;
 386 }
 387
 388 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
 389 {
 390         struct task_struct *task = get_proc_task(inode);
 391         struct fs_struct *fs = NULL;
 392         int result = -ENOENT;
 393
 394         if (task) {
 395                 fs = get_fs_struct(task);
 396                 put_task_struct(task);
 397         }
 398         if (fs) {
 399                 read_lock(&fs->lock);
 400                 *mnt = mntget(fs->rootmnt);
 401                 *dentry = dget(fs->root);
 402                 read_unlock(&fs->lock);
 403                 result = 0;
 404                 put_fs_struct(fs);
 405         }
 406         return result;
 407 }
 408
 409 #define MAY_PTRACE(task) \
 410         (task == current || \
 411         (task->parent == current && \
 412         (task->ptrace & PT_PTRACED) && \
 413          (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
 414          security_ptrace(current,task) == 0))
 415
 416 static int proc_pid_environ(struct task_struct *task, char * buffer)
 417 {
 418         int res = 0;
 419         struct mm_struct *mm = get_task_mm(task);
 420         if (mm) {
 421                 unsigned int len = mm->env_end - mm->env_start;
 422                 if (len > PAGE_SIZE)
 423                         len = PAGE_SIZE;
 424                 res = access_process_vm(task, mm->env_start, buffer, len, 0);
 425                 if (!ptrace_may_attach(task))
 426                         res = -ESRCH;
 427                 mmput(mm);
 428         }
 429         return res;
 430 }
 431
 432 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 433 {
 434         int res = 0;
 435         unsigned int len;
 436         struct mm_struct *mm = get_task_mm(task);
 437         if (!mm)
 438                 goto out;
 439         if (!mm->arg_end)
 440                 goto out_mm;    /* Shh! No looking before we're done */
 441
 442         len = mm->arg_end - mm->arg_start;
 443
 444         if (len > PAGE_SIZE)
 445                 len = PAGE_SIZE;
 446
 447         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 448
 449         // If the nul at the end of args has been overwritten, then
 450         // assume application is using setproctitle(3).
 451         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 452                 len = strnlen(buffer, res);
 453                 if (len < res) {
 454                     res = len;
 455                 } else {
 456                         len = mm->env_end - mm->env_start;
 457                         if (len > PAGE_SIZE - res)
 458                                 len = PAGE_SIZE - res;
 459                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 460                         res = strnlen(buffer, res);
 461                 }
 462         }
 463 out_mm:
 464         mmput(mm);
 465 out:
 466         return res;
 467 }
 468
 469 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 470 {
 471         int res = 0;
 472         struct mm_struct *mm = get_task_mm(task);
 473         if (mm) {
 474                 unsigned int nwords = 0;
 475                 do
 476                         nwords += 2;
 477                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 478                 res = nwords * sizeof(mm->saved_auxv[0]);
 479                 if (res > PAGE_SIZE)
 480                         res = PAGE_SIZE;
 481                 memcpy(buffer, mm->saved_auxv, res);
 482                 mmput(mm);
 483         }
 484         return res;
 485 }
 486
 487
 488 #ifdef CONFIG_KALLSYMS
 489 /*
 490  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 491  * Returns the resolved symbol.  If that fails, simply return the address.
 492  */
 493 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 494 {
 495         char *modname;
 496         const char *sym_name;
 497         unsigned long wchan, size, offset;
 498         char namebuf[KSYM_NAME_LEN+1];
 499
 500         wchan = get_wchan(task);
 501
 502         sym_name = kallsyms_lookup(wchan, &size, &offset, &modname, namebuf);
 503         if (sym_name)
 504                 return sprintf(buffer, "%s", sym_name);
 505         return sprintf(buffer, "%lu", wchan);
 506 }
 507 #endif /* CONFIG_KALLSYMS */
 508
 509 #ifdef CONFIG_SCHEDSTATS
 510 /*
 511  * Provides /proc/PID/schedstat
 512  */
 513 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 514 {
 515         return sprintf(buffer, "%lu %lu %lu\n",
 516                         task->sched_info.cpu_time,
 517                         task->sched_info.run_delay,
 518                         task->sched_info.pcnt);
 519 }
 520 #endif
 521
 522 /* The badness from the OOM killer */
 523 unsigned long badness(struct task_struct *p, unsigned long uptime);
 524 static int proc_oom_score(struct task_struct *task, char *buffer)
 525 {
 526         unsigned long points;
 527         struct timespec uptime;
 528
 529         do_posix_clock_monotonic_gettime(&uptime);
 530         points = badness(task, uptime.tv_sec);
 531         return sprintf(buffer, "%lu\n", points);
 532 }
 533
 534 /************************************************************************/
 535 /*                       Here the fs part begins                        */
 536 /************************************************************************/
 537
 538 /* permission checks */
 539 static int proc_fd_access_allowed(struct inode *inode)
 540 {
 541         struct task_struct *task;
 542         int allowed = 0;
 543         /* Allow access to a task's file descriptors if it is us or we
 544          * may use ptrace attach to the process and find out that
 545          * information.
 546          */
 547         task = get_proc_task(inode);
 548         if (task) {
 549                 allowed = ptrace_may_attach(task);
 550                 put_task_struct(task);
 551         }
 552         return allowed;
 553 }
 554
 555 extern struct seq_operations mounts_op;
 556 struct proc_mounts {
 557         struct seq_file m;
 558         int event;
 559 };
 560
 561 static int mounts_open(struct inode *inode, struct file *file)
 562 {
 563         struct task_struct *task = get_proc_task(inode);
 564         struct namespace *namespace = NULL;
 565         struct proc_mounts *p;
 566         int ret = -EINVAL;
 567
 568         if (task) {
 569                 task_lock(task);
 570                 namespace = task->namespace;
 571                 if (namespace)
 572                         get_namespace(namespace);
 573                 task_unlock(task);
 574                 put_task_struct(task);
 575         }
 576
 577         if (namespace) {
 578                 ret = -ENOMEM;
 579                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 580                 if (p) {
 581                         file->private_data = &p->m;
 582                         ret = seq_open(file, &mounts_op);
 583                         if (!ret) {
 584                                 p->m.private = namespace;
 585                                 p->event = namespace->event;
 586                                 return 0;
 587                         }
 588                         kfree(p);
 589                 }
 590                 put_namespace(namespace);
 591         }
 592         return ret;
 593 }
 594
 595 static int mounts_release(struct inode *inode, struct file *file)
 596 {
 597         struct seq_file *m = file->private_data;
 598         struct namespace *namespace = m->private;
 599         put_namespace(namespace);
 600         return seq_release(inode, file);
 601 }
 602
 603 static unsigned mounts_poll(struct file *file, poll_table *wait)
 604 {
 605         struct proc_mounts *p = file->private_data;
 606         struct namespace *ns = p->m.private;
 607         unsigned res = 0;
 608
 609         poll_wait(file, &ns->poll, wait);
 610
 611         spin_lock(&vfsmount_lock);
 612         if (p->event != ns->event) {
 613                 p->event = ns->event;
 614                 res = POLLERR;
 615         }
 616         spin_unlock(&vfsmount_lock);
 617
 618         return res;
 619 }
 620
 621 static struct file_operations proc_mounts_operations = {
 622         .open           = mounts_open,
 623         .read           = seq_read,
 624         .llseek         = seq_lseek,
 625         .release        = mounts_release,
 626         .poll           = mounts_poll,
 627 };
 628
 629 extern struct seq_operations mountstats_op;
 630 static int mountstats_open(struct inode *inode, struct file *file)
 631 {
 632         int ret = seq_open(file, &mountstats_op);
 633
 634         if (!ret) {
 635                 struct seq_file *m = file->private_data;
 636                 struct namespace *namespace = NULL;
 637                 struct task_struct *task = get_proc_task(inode);
 638
 639                 if (task) {
 640                         task_lock(task);
 641                         namespace = task->namespace;
 642                         if (namespace)
 643                                 get_namespace(namespace);
 644                         task_unlock(task);
 645                         put_task_struct(task);
 646                 }
 647
 648                 if (namespace)
 649                         m->private = namespace;
 650                 else {
 651                         seq_release(inode, file);
 652                         ret = -EINVAL;
 653                 }
 654         }
 655         return ret;
 656 }
 657
 658 static struct file_operations proc_mountstats_operations = {
 659         .open           = mountstats_open,
 660         .read           = seq_read,
 661         .llseek         = seq_lseek,
 662         .release        = mounts_release,
 663 };
 664
 665 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 666
 667 static ssize_t proc_info_read(struct file * file, char __user * buf,
 668                           size_t count, loff_t *ppos)
 669 {
 670         struct inode * inode = file->f_dentry->d_inode;
 671         unsigned long page;
 672         ssize_t length;
 673         struct task_struct *task = get_proc_task(inode);
 674
 675         length = -ESRCH;
 676         if (!task)
 677                 goto out_no_task;
 678
 679         if (count > PROC_BLOCK_SIZE)
 680                 count = PROC_BLOCK_SIZE;
 681
 682         length = -ENOMEM;
 683         if (!(page = __get_free_page(GFP_KERNEL)))
 684                 goto out;
 685
 686         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 687
 688         if (length >= 0)
 689                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 690         free_page(page);
 691 out:
 692         put_task_struct(task);
 693 out_no_task:
 694         return length;
 695 }
 696
 697 static struct file_operations proc_info_file_operations = {
 698         .read           = proc_info_read,
 699 };
 700
 701 static int mem_open(struct inode* inode, struct file* file)
 702 {
 703         file->private_data = (void*)((long)current->self_exec_id);
 704         return 0;
 705 }
 706
 707 static ssize_t mem_read(struct file * file, char __user * buf,
 708                         size_t count, loff_t *ppos)
 709 {
 710         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 711         char *page;
 712         unsigned long src = *ppos;
 713         int ret = -ESRCH;
 714         struct mm_struct *mm;
 715
 716         if (!task)
 717                 goto out_no_task;
 718
 719         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 720                 goto out;
 721
 722         ret = -ENOMEM;
 723         page = (char *)__get_free_page(GFP_USER);
 724         if (!page)
 725                 goto out;
 726
 727         ret = 0;
 728
 729         mm = get_task_mm(task);
 730         if (!mm)
 731                 goto out_free;
 732
 733         ret = -EIO;
 734
 735         if (file->private_data != (void*)((long)current->self_exec_id))
 736                 goto out_put;
 737
 738         ret = 0;
 739
 740         while (count > 0) {
 741                 int this_len, retval;
 742
 743                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 744                 retval = access_process_vm(task, src, page, this_len, 0);
 745                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 746                         if (!ret)
 747                                 ret = -EIO;
 748                         break;
 749                 }
 750
 751                 if (copy_to_user(buf, page, retval)) {
 752                         ret = -EFAULT;
 753                         break;
 754                 }
 755
 756                 ret += retval;
 757                 src += retval;
 758                 buf += retval;
 759                 count -= retval;
 760         }
 761         *ppos = src;
 762
 763 out_put:
 764         mmput(mm);
 765 out_free:
 766         free_page((unsigned long) page);
 767 out:
 768         put_task_struct(task);
 769 out_no_task:
 770         return ret;
 771 }
 772
 773 #define mem_write NULL
 774
 775 #ifndef mem_write
 776 /* This is a security hazard */
 777 static ssize_t mem_write(struct file * file, const char * buf,
 778                          size_t count, loff_t *ppos)
 779 {
 780         int copied = 0;
 781         char *page;
 782         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 783         unsigned long dst = *ppos;
 784
 785         copied = -ESRCH;
 786         if (!task)
 787                 goto out_no_task;
 788
 789         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 790                 goto out;
 791
 792         copied = -ENOMEM;
 793         page = (char *)__get_free_page(GFP_USER);
 794         if (!page)
 795                 goto out;
 796
 797         while (count > 0) {
 798                 int this_len, retval;
 799
 800                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 801                 if (copy_from_user(page, buf, this_len)) {
 802                         copied = -EFAULT;
 803                         break;
 804                 }
 805                 retval = access_process_vm(task, dst, page, this_len, 1);
 806                 if (!retval) {
 807                         if (!copied)
 808                                 copied = -EIO;
 809                         break;
 810                 }
 811                 copied += retval;
 812                 buf += retval;
 813                 dst += retval;
 814                 count -= retval;
 815         }
 816         *ppos = dst;
 817         free_page((unsigned long) page);
 818 out:
 819         put_task_struct(task);
 820 out_no_task:
 821         return copied;
 822 }
 823 #endif
 824
 825 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 826 {
 827         switch (orig) {
 828         case 0:
 829                 file->f_pos = offset;
 830                 break;
 831         case 1:
 832                 file->f_pos += offset;
 833                 break;
 834         default:
 835                 return -EINVAL;
 836         }
 837         force_successful_syscall_return();
 838         return file->f_pos;
 839 }
 840
 841 static struct file_operations proc_mem_operations = {
 842         .llseek         = mem_lseek,
 843         .read           = mem_read,
 844         .write          = mem_write,
 845         .open           = mem_open,
 846 };
 847
 848 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 849                                 size_t count, loff_t *ppos)
 850 {
 851         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 852         char buffer[PROC_NUMBUF];
 853         size_t len;
 854         int oom_adjust;
 855         loff_t __ppos = *ppos;
 856
 857         if (!task)
 858                 return -ESRCH;
 859         oom_adjust = task->oomkilladj;
 860         put_task_struct(task);
 861
 862         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 863         if (__ppos >= len)
 864                 return 0;
 865         if (count > len-__ppos)
 866                 count = len-__ppos;
 867         if (copy_to_user(buf, buffer + __ppos, count))
 868                 return -EFAULT;
 869         *ppos = __ppos + count;
 870         return count;
 871 }
 872
 873 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 874                                 size_t count, loff_t *ppos)
 875 {
 876         struct task_struct *task;
 877         char buffer[PROC_NUMBUF], *end;
 878         int oom_adjust;
 879
 880         if (!capable(CAP_SYS_RESOURCE))
 881                 return -EPERM;
 882         memset(buffer, 0, sizeof(buffer));
 883         if (count > sizeof(buffer) - 1)
 884                 count = sizeof(buffer) - 1;
 885         if (copy_from_user(buffer, buf, count))
 886                 return -EFAULT;
 887         oom_adjust = simple_strtol(buffer, &end, 0);
 888         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 889                 return -EINVAL;
 890         if (*end == '\n')
 891                 end++;
 892         task = get_proc_task(file->f_dentry->d_inode);
 893         if (!task)
 894                 return -ESRCH;
 895         task->oomkilladj = oom_adjust;
 896         put_task_struct(task);
 897         if (end - buffer == 0)
 898                 return -EIO;
 899         return end - buffer;
 900 }
 901
 902 static struct file_operations proc_oom_adjust_operations = {
 903         .read           = oom_adjust_read,
 904         .write          = oom_adjust_write,
 905 };
 906
 907 #ifdef CONFIG_AUDITSYSCALL
 908 #define TMPBUFLEN 21
 909 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 910                                   size_t count, loff_t *ppos)
 911 {
 912         struct inode * inode = file->f_dentry->d_inode;
 913         struct task_struct *task = get_proc_task(inode);
 914         ssize_t length;
 915         char tmpbuf[TMPBUFLEN];
 916
 917         if (!task)
 918                 return -ESRCH;
 919         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 920                                 audit_get_loginuid(task->audit_context));
 921         put_task_struct(task);
 922         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 923 }
 924
 925 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 926                                    size_t count, loff_t *ppos)
 927 {
 928         struct inode * inode = file->f_dentry->d_inode;
 929         char *page, *tmp;
 930         ssize_t length;
 931         uid_t loginuid;
 932
 933         if (!capable(CAP_AUDIT_CONTROL))
 934                 return -EPERM;
 935
 936         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
 937                 return -EPERM;
 938
 939         if (count >= PAGE_SIZE)
 940                 count = PAGE_SIZE - 1;
 941
 942         if (*ppos != 0) {
 943                 /* No partial writes. */
 944                 return -EINVAL;
 945         }
 946         page = (char*)__get_free_page(GFP_USER);
 947         if (!page)
 948                 return -ENOMEM;
 949         length = -EFAULT;
 950         if (copy_from_user(page, buf, count))
 951                 goto out_free_page;
 952
 953         page[count] = '\0';
 954         loginuid = simple_strtoul(page, &tmp, 10);
 955         if (tmp == page) {
 956                 length = -EINVAL;
 957                 goto out_free_page;
 958
 959         }
 960         length = audit_set_loginuid(current, loginuid);
 961         if (likely(length == 0))
 962                 length = count;
 963
 964 out_free_page:
 965         free_page((unsigned long) page);
 966         return length;
 967 }
 968
 969 static struct file_operations proc_loginuid_operations = {
 970         .read           = proc_loginuid_read,
 971         .write          = proc_loginuid_write,
 972 };
 973 #endif
 974
 975 #ifdef CONFIG_SECCOMP
 976 static ssize_t seccomp_read(struct file *file, char __user *buf,
 977                             size_t count, loff_t *ppos)
 978 {
 979         struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
 980         char __buf[20];
 981         loff_t __ppos = *ppos;
 982         size_t len;
 983
 984         if (!tsk)
 985                 return -ESRCH;
 986         /* no need to print the trailing zero, so use only len */
 987         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
 988         put_task_struct(tsk);
 989         if (__ppos >= len)
 990                 return 0;
 991         if (count > len - __ppos)
 992                 count = len - __ppos;
 993         if (copy_to_user(buf, __buf + __ppos, count))
 994                 return -EFAULT;
 995         *ppos = __ppos + count;
 996         return count;
 997 }
 998
 999 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1000                              size_t count, loff_t *ppos)
1001 {
1002         struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
1003         char __buf[20], *end;
1004         unsigned int seccomp_mode;
1005         ssize_t result;
1006
1007         result = -ESRCH;
1008         if (!tsk)
1009                 goto out_no_task;
1010
1011         /* can set it only once to be even more secure */
1012         result = -EPERM;
1013         if (unlikely(tsk->seccomp.mode))
1014                 goto out;
1015
1016         result = -EFAULT;
1017         memset(__buf, 0, sizeof(__buf));
1018         count = min(count, sizeof(__buf) - 1);
1019         if (copy_from_user(__buf, buf, count))
1020                 goto out;
1021
1022         seccomp_mode = simple_strtoul(__buf, &end, 0);
1023         if (*end == '\n')
1024                 end++;
1025         result = -EINVAL;
1026         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1027                 tsk->seccomp.mode = seccomp_mode;
1028                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1029         } else
1030                 goto out;
1031         result = -EIO;
1032         if (unlikely(!(end - __buf)))
1033                 goto out;
1034         result = end - __buf;
1035 out:
1036         put_task_struct(tsk);
1037 out_no_task:
1038         return result;
1039 }
1040
1041 static struct file_operations proc_seccomp_operations = {
1042         .read           = seccomp_read,
1043         .write          = seccomp_write,
1044 };
1045 #endif /* CONFIG_SECCOMP */
1046
1047 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1048 {
1049         struct inode *inode = dentry->d_inode;
1050         int error = -EACCES;
1051
1052         /* We don't need a base pointer in the /proc filesystem */
1053         path_release(nd);
1054
1055         /* Are we allowed to snoop on the tasks file descriptors? */
1056         if (!proc_fd_access_allowed(inode))
1057                 goto out;
1058
1059         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1060         nd->last_type = LAST_BIND;
1061 out:
1062         return ERR_PTR(error);
1063 }
1064
1065 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1066                             char __user *buffer, int buflen)
1067 {
1068         struct inode * inode;
1069         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1070         int len;
1071
1072         if (!tmp)
1073                 return -ENOMEM;
1074
1075         inode = dentry->d_inode;
1076         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1077         len = PTR_ERR(path);
1078         if (IS_ERR(path))
1079                 goto out;
1080         len = tmp + PAGE_SIZE - 1 - path;
1081
1082         if (len > buflen)
1083                 len = buflen;
1084         if (copy_to_user(buffer, path, len))
1085                 len = -EFAULT;
1086  out:
1087         free_page((unsigned long)tmp);
1088         return len;
1089 }
1090
1091 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1092 {
1093         int error = -EACCES;
1094         struct inode *inode = dentry->d_inode;
1095         struct dentry *de;
1096         struct vfsmount *mnt = NULL;
1097
1098         /* Are we allowed to snoop on the tasks file descriptors? */
1099         if (!proc_fd_access_allowed(inode))
1100                 goto out;
1101
1102         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1103         if (error)
1104                 goto out;
1105
1106         error = do_proc_readlink(de, mnt, buffer, buflen);
1107         dput(de);
1108         mntput(mnt);
1109 out:
1110         return error;
1111 }
1112
1113 static struct inode_operations proc_pid_link_inode_operations = {
1114         .readlink       = proc_pid_readlink,
1115         .follow_link    = proc_pid_follow_link
1116 };
1117
1118 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1119 {
1120         struct dentry *dentry = filp->f_dentry;
1121         struct inode *inode = dentry->d_inode;
1122         struct task_struct *p = get_proc_task(inode);
1123         unsigned int fd, tid, ino;
1124         int retval;
1125         char buf[PROC_NUMBUF];
1126         struct files_struct * files;
1127         struct fdtable *fdt;
1128
1129         retval = -ENOENT;
1130         if (!p)
1131                 goto out_no_task;
1132         retval = 0;
1133         tid = p->pid;
1134
1135         fd = filp->f_pos;
1136         switch (fd) {
1137                 case 0:
1138                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1139                                 goto out;
1140                         filp->f_pos++;
1141                 case 1:
1142                         ino = parent_ino(dentry);
1143                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1144                                 goto out;
1145                         filp->f_pos++;
1146                 default:
1147                         files = get_files_struct(p);
1148                         if (!files)
1149                                 goto out;
1150                         rcu_read_lock();
1151                         fdt = files_fdtable(files);
1152                         for (fd = filp->f_pos-2;
1153                              fd < fdt->max_fds;
1154                              fd++, filp->f_pos++) {
1155                                 unsigned int i,j;
1156
1157                                 if (!fcheck_files(files, fd))
1158                                         continue;
1159                                 rcu_read_unlock();
1160
1161                                 j = PROC_NUMBUF;
1162                                 i = fd;
1163                                 do {
1164                                         j--;
1165                                         buf[j] = '0' + (i % 10);
1166                                         i /= 10;
1167                                 } while (i);
1168
1169                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1170                                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1171                                         rcu_read_lock();
1172                                         break;
1173                                 }
1174                                 rcu_read_lock();
1175                         }
1176                         rcu_read_unlock();
1177                         put_files_struct(files);
1178         }
1179 out:
1180         put_task_struct(p);
1181 out_no_task:
1182         return retval;
1183 }
1184
1185 static int proc_pident_readdir(struct file *filp,
1186                 void *dirent, filldir_t filldir,
1187                 struct pid_entry *ents, unsigned int nents)
1188 {
1189         int i;
1190         int pid;
1191         struct dentry *dentry = filp->f_dentry;
1192         struct inode *inode = dentry->d_inode;
1193         struct task_struct *task = get_proc_task(inode);
1194         struct pid_entry *p;
1195         ino_t ino;
1196         int ret;
1197
1198         ret = -ENOENT;
1199         if (!task)
1200                 goto out;
1201
1202         ret = 0;
1203         pid = task->pid;
1204         put_task_struct(task);
1205         i = filp->f_pos;
1206         switch (i) {
1207         case 0:
1208                 ino = inode->i_ino;
1209                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1210                         goto out;
1211                 i++;
1212                 filp->f_pos++;
1213                 /* fall through */
1214         case 1:
1215                 ino = parent_ino(dentry);
1216                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1217                         goto out;
1218                 i++;
1219                 filp->f_pos++;
1220                 /* fall through */
1221         default:
1222                 i -= 2;
1223                 if (i >= nents) {
1224                         ret = 1;
1225                         goto out;
1226                 }
1227                 p = ents + i;
1228                 while (p->name) {
1229                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1230                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1231                                 goto out;
1232                         filp->f_pos++;
1233                         p++;
1234                 }
1235         }
1236
1237         ret = 1;
1238 out:
1239         return ret;
1240 }
1241
1242 static int proc_tgid_base_readdir(struct file * filp,
1243                              void * dirent, filldir_t filldir)
1244 {
1245         return proc_pident_readdir(filp,dirent,filldir,
1246                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1247 }
1248
1249 static int proc_tid_base_readdir(struct file * filp,
1250                              void * dirent, filldir_t filldir)
1251 {
1252         return proc_pident_readdir(filp,dirent,filldir,
1253                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1254 }
1255
1256 /* building an inode */
1257
1258 static int task_dumpable(struct task_struct *task)
1259 {
1260         int dumpable = 0;
1261         struct mm_struct *mm;
1262
1263         task_lock(task);
1264         mm = task->mm;
1265         if (mm)
1266                 dumpable = mm->dumpable;
1267         task_unlock(task);
1268         if(dumpable == 1)
1269                 return 1;
1270         return 0;
1271 }
1272
1273
1274 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1275 {
1276         struct inode * inode;
1277         struct proc_inode *ei;
1278
1279         /* We need a new inode */
1280
1281         inode = new_inode(sb);
1282         if (!inode)
1283                 goto out;
1284
1285         /* Common stuff */
1286         ei = PROC_I(inode);
1287         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1288         inode->i_ino = fake_ino(task->pid, ino);
1289
1290         /*
1291          * grab the reference to task.
1292          */
1293         ei->pid = get_pid(task->pids[PIDTYPE_PID].pid);
1294         if (!ei->pid)
1295                 goto out_unlock;
1296
1297         inode->i_uid = 0;
1298         inode->i_gid = 0;
1299         if (task_dumpable(task)) {
1300                 inode->i_uid = task->euid;
1301                 inode->i_gid = task->egid;
1302         }
1303         security_task_to_inode(task, inode);
1304
1305 out:
1306         return inode;
1307
1308 out_unlock:
1309         iput(inode);
1310         return NULL;
1311 }
1312
1313 /* dentry stuff */
1314
1315 /*
1316  *      Exceptional case: normally we are not allowed to unhash a busy
1317  * directory. In this case, however, we can do it - no aliasing problems
1318  * due to the way we treat inodes.
1319  *
1320  * Rewrite the inode's ownerships here because the owning task may have
1321  * performed a setuid(), etc.
1322  *
1323  * Before the /proc/pid/status file was created the only way to read
1324  * the effective uid of a /process was to stat /proc/pid.  Reading
1325  * /proc/pid/status is slow enough that procps and other packages
1326  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1327  * made this apply to all per process world readable and executable
1328  * directories.
1329  */
1330 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1331 {
1332         struct inode *inode = dentry->d_inode;
1333         struct task_struct *task = get_proc_task(inode);
1334         if (task) {
1335                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1336                     task_dumpable(task)) {
1337                         inode->i_uid = task->euid;
1338                         inode->i_gid = task->egid;
1339                 } else {
1340                         inode->i_uid = 0;
1341                         inode->i_gid = 0;
1342                 }
1343                 security_task_to_inode(task, inode);
1344                 put_task_struct(task);
1345                 return 1;
1346         }
1347         d_drop(dentry);
1348         return 0;
1349 }
1350
1351 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1352 {
1353         struct inode *inode = dentry->d_inode;
1354         struct task_struct *task;
1355         generic_fillattr(inode, stat);
1356
1357         rcu_read_lock();
1358         stat->uid = 0;
1359         stat->gid = 0;
1360         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1361         if (task) {
1362                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1363                     task_dumpable(task)) {
1364                         stat->uid = task->euid;
1365                         stat->gid = task->egid;
1366                 }
1367         }
1368         rcu_read_unlock();
1369         return 0;
1370 }
1371
1372 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1373 {
1374         struct inode *inode = dentry->d_inode;
1375         struct task_struct *task = get_proc_task(inode);
1376         int fd = proc_fd(inode);
1377         struct files_struct *files;
1378
1379         if (task) {
1380                 files = get_files_struct(task);
1381                 if (files) {
1382                         rcu_read_lock();
1383                         if (fcheck_files(files, fd)) {
1384                                 rcu_read_unlock();
1385                                 put_files_struct(files);
1386                                 if (task_dumpable(task)) {
1387                                         inode->i_uid = task->euid;
1388                                         inode->i_gid = task->egid;
1389                                 } else {
1390                                         inode->i_uid = 0;
1391                                         inode->i_gid = 0;
1392                                 }
1393                                 security_task_to_inode(task, inode);
1394                                 put_task_struct(task);
1395                                 return 1;
1396                         }
1397                         rcu_read_unlock();
1398                         put_files_struct(files);
1399                 }
1400                 put_task_struct(task);
1401         }
1402         d_drop(dentry);
1403         return 0;
1404 }
1405
1406 static int pid_delete_dentry(struct dentry * dentry)
1407 {
1408         /* Is the task we represent dead?
1409          * If so, then don't put the dentry on the lru list,
1410          * kill it immediately.
1411          */
1412         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1413 }
1414
1415 static struct dentry_operations tid_fd_dentry_operations =
1416 {
1417         .d_revalidate   = tid_fd_revalidate,
1418         .d_delete       = pid_delete_dentry,
1419 };
1420
1421 static struct dentry_operations pid_dentry_operations =
1422 {
1423         .d_revalidate   = pid_revalidate,
1424         .d_delete       = pid_delete_dentry,
1425 };
1426
1427 /* Lookups */
1428
1429 static unsigned name_to_int(struct dentry *dentry)
1430 {
1431         const char *name = dentry->d_name.name;
1432         int len = dentry->d_name.len;
1433         unsigned n = 0;
1434
1435         if (len > 1 && *name == '0')
1436                 goto out;
1437         while (len-- > 0) {
1438                 unsigned c = *name++ - '0';
1439                 if (c > 9)
1440                         goto out;
1441                 if (n >= (~0U-9)/10)
1442                         goto out;
1443                 n *= 10;
1444                 n += c;
1445         }
1446         return n;
1447 out:
1448         return ~0U;
1449 }
1450
1451 /* SMP-safe */
1452 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1453 {
1454         struct task_struct *task = get_proc_task(dir);
1455         unsigned fd = name_to_int(dentry);
1456         struct dentry *result = ERR_PTR(-ENOENT);
1457         struct file * file;
1458         struct files_struct * files;
1459         struct inode *inode;
1460         struct proc_inode *ei;
1461
1462         if (!task)
1463                 goto out_no_task;
1464         if (fd == ~0U)
1465                 goto out;
1466
1467         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1468         if (!inode)
1469                 goto out;
1470         ei = PROC_I(inode);
1471         ei->fd = fd;
1472         files = get_files_struct(task);
1473         if (!files)
1474                 goto out_unlock;
1475         inode->i_mode = S_IFLNK;
1476
1477         /*
1478          * We are not taking a ref to the file structure, so we must
1479          * hold ->file_lock.
1480          */
1481         spin_lock(&files->file_lock);
1482         file = fcheck_files(files, fd);
1483         if (!file)
1484                 goto out_unlock2;
1485         if (file->f_mode & 1)
1486                 inode->i_mode |= S_IRUSR | S_IXUSR;
1487         if (file->f_mode & 2)
1488                 inode->i_mode |= S_IWUSR | S_IXUSR;
1489         spin_unlock(&files->file_lock);
1490         put_files_struct(files);
1491         inode->i_op = &proc_pid_link_inode_operations;
1492         inode->i_size = 64;
1493         ei->op.proc_get_link = proc_fd_link;
1494         dentry->d_op = &tid_fd_dentry_operations;
1495         d_add(dentry, inode);
1496         /* Close the race of the process dying before we return the dentry */
1497         if (tid_fd_revalidate(dentry, NULL))
1498                 result = NULL;
1499 out:
1500         put_task_struct(task);
1501 out_no_task:
1502         return result;
1503
1504 out_unlock2:
1505         spin_unlock(&files->file_lock);
1506         put_files_struct(files);
1507 out_unlock:
1508         iput(inode);
1509         goto out;
1510 }
1511
1512 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1513 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1514 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat);
1515
1516 static struct file_operations proc_fd_operations = {
1517         .read           = generic_read_dir,
1518         .readdir        = proc_readfd,
1519 };
1520
1521 static struct file_operations proc_task_operations = {
1522         .read           = generic_read_dir,
1523         .readdir        = proc_task_readdir,
1524 };
1525
1526 /*
1527  * proc directories can do almost nothing..
1528  */
1529 static struct inode_operations proc_fd_inode_operations = {
1530         .lookup         = proc_lookupfd,
1531 };
1532
1533 static struct inode_operations proc_task_inode_operations = {
1534         .lookup         = proc_task_lookup,
1535         .getattr        = proc_task_getattr,
1536 };
1537
1538 #ifdef CONFIG_SECURITY
1539 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1540                                   size_t count, loff_t *ppos)
1541 {
1542         struct inode * inode = file->f_dentry->d_inode;
1543         unsigned long page;
1544         ssize_t length;
1545         struct task_struct *task = get_proc_task(inode);
1546
1547         length = -ESRCH;
1548         if (!task)
1549                 goto out_no_task;
1550
1551         if (count > PAGE_SIZE)
1552                 count = PAGE_SIZE;
1553         length = -ENOMEM;
1554         if (!(page = __get_free_page(GFP_KERNEL)))
1555                 goto out;
1556
1557         length = security_getprocattr(task,
1558                                       (char*)file->f_dentry->d_name.name,
1559                                       (void*)page, count);
1560         if (length >= 0)
1561                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1562         free_page(page);
1563 out:
1564         put_task_struct(task);
1565 out_no_task:
1566         return length;
1567 }
1568
1569 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1570                                    size_t count, loff_t *ppos)
1571 {
1572         struct inode * inode = file->f_dentry->d_inode;
1573         char *page;
1574         ssize_t length;
1575         struct task_struct *task = get_proc_task(inode);
1576
1577         length = -ESRCH;
1578         if (!task)
1579                 goto out_no_task;
1580         if (count > PAGE_SIZE)
1581                 count = PAGE_SIZE;
1582
1583         /* No partial writes. */
1584         length = -EINVAL;
1585         if (*ppos != 0)
1586                 goto out;
1587
1588         length = -ENOMEM;
1589         page = (char*)__get_free_page(GFP_USER);
1590         if (!page)
1591                 goto out;
1592
1593         length = -EFAULT;
1594         if (copy_from_user(page, buf, count))
1595                 goto out_free;
1596
1597         length = security_setprocattr(task,
1598                                       (char*)file->f_dentry->d_name.name,
1599                                       (void*)page, count);
1600 out_free:
1601         free_page((unsigned long) page);
1602 out:
1603         put_task_struct(task);
1604 out_no_task:
1605         return length;
1606 }
1607
1608 static struct file_operations proc_pid_attr_operations = {
1609         .read           = proc_pid_attr_read,
1610         .write          = proc_pid_attr_write,
1611 };
1612
1613 static struct file_operations proc_tid_attr_operations;
1614 static struct inode_operations proc_tid_attr_inode_operations;
1615 static struct file_operations proc_tgid_attr_operations;
1616 static struct inode_operations proc_tgid_attr_inode_operations;
1617 #endif
1618
1619 /* SMP-safe */
1620 static struct dentry *proc_pident_lookup(struct inode *dir,
1621                                          struct dentry *dentry,
1622                                          struct pid_entry *ents)
1623 {
1624         struct inode *inode;
1625         struct dentry *error;
1626         struct task_struct *task = get_proc_task(dir);
1627         struct pid_entry *p;
1628         struct proc_inode *ei;
1629
1630         error = ERR_PTR(-ENOENT);
1631         inode = NULL;
1632
1633         if (!task)
1634                 goto out_no_task;
1635
1636         for (p = ents; p->name; p++) {
1637                 if (p->len != dentry->d_name.len)
1638                         continue;
1639                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1640                         break;
1641         }
1642         if (!p->name)
1643                 goto out;
1644
1645         error = ERR_PTR(-EINVAL);
1646         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1647         if (!inode)
1648                 goto out;
1649
1650         ei = PROC_I(inode);
1651         inode->i_mode = p->mode;
1652         /*
1653          * Yes, it does not scale. And it should not. Don't add
1654          * new entries into /proc/<tgid>/ without very good reasons.
1655          */
1656         switch(p->type) {
1657                 case PROC_TGID_TASK:
1658                         inode->i_nlink = 2;
1659                         inode->i_op = &proc_task_inode_operations;
1660                         inode->i_fop = &proc_task_operations;
1661                         break;
1662                 case PROC_TID_FD:
1663                 case PROC_TGID_FD:
1664                         inode->i_nlink = 2;
1665                         inode->i_op = &proc_fd_inode_operations;
1666                         inode->i_fop = &proc_fd_operations;
1667                         break;
1668                 case PROC_TID_EXE:
1669                 case PROC_TGID_EXE:
1670                         inode->i_op = &proc_pid_link_inode_operations;
1671                         ei->op.proc_get_link = proc_exe_link;
1672                         break;
1673                 case PROC_TID_CWD:
1674                 case PROC_TGID_CWD:
1675                         inode->i_op = &proc_pid_link_inode_operations;
1676                         ei->op.proc_get_link = proc_cwd_link;
1677                         break;
1678                 case PROC_TID_ROOT:
1679                 case PROC_TGID_ROOT:
1680                         inode->i_op = &proc_pid_link_inode_operations;
1681                         ei->op.proc_get_link = proc_root_link;
1682                         break;
1683                 case PROC_TID_ENVIRON:
1684                 case PROC_TGID_ENVIRON:
1685                         inode->i_fop = &proc_info_file_operations;
1686                         ei->op.proc_read = proc_pid_environ;
1687                         break;
1688                 case PROC_TID_AUXV:
1689                 case PROC_TGID_AUXV:
1690                         inode->i_fop = &proc_info_file_operations;
1691                         ei->op.proc_read = proc_pid_auxv;
1692                         break;
1693                 case PROC_TID_STATUS:
1694                 case PROC_TGID_STATUS:
1695                         inode->i_fop = &proc_info_file_operations;
1696                         ei->op.proc_read = proc_pid_status;
1697                         break;
1698                 case PROC_TID_STAT:
1699                         inode->i_fop = &proc_info_file_operations;
1700                         ei->op.proc_read = proc_tid_stat;
1701                         break;
1702                 case PROC_TGID_STAT:
1703                         inode->i_fop = &proc_info_file_operations;
1704                         ei->op.proc_read = proc_tgid_stat;
1705                         break;
1706                 case PROC_TID_CMDLINE:
1707                 case PROC_TGID_CMDLINE:
1708                         inode->i_fop = &proc_info_file_operations;
1709                         ei->op.proc_read = proc_pid_cmdline;
1710                         break;
1711                 case PROC_TID_STATM:
1712                 case PROC_TGID_STATM:
1713                         inode->i_fop = &proc_info_file_operations;
1714                         ei->op.proc_read = proc_pid_statm;
1715                         break;
1716                 case PROC_TID_MAPS:
1717                 case PROC_TGID_MAPS:
1718                         inode->i_fop = &proc_maps_operations;
1719                         break;
1720 #ifdef CONFIG_NUMA
1721                 case PROC_TID_NUMA_MAPS:
1722                 case PROC_TGID_NUMA_MAPS:
1723                         inode->i_fop = &proc_numa_maps_operations;
1724                         break;
1725 #endif
1726                 case PROC_TID_MEM:
1727                 case PROC_TGID_MEM:
1728                         inode->i_fop = &proc_mem_operations;
1729                         break;
1730 #ifdef CONFIG_SECCOMP
1731                 case PROC_TID_SECCOMP:
1732                 case PROC_TGID_SECCOMP:
1733                         inode->i_fop = &proc_seccomp_operations;
1734                         break;
1735 #endif /* CONFIG_SECCOMP */
1736                 case PROC_TID_MOUNTS:
1737                 case PROC_TGID_MOUNTS:
1738                         inode->i_fop = &proc_mounts_operations;
1739                         break;
1740 #ifdef CONFIG_MMU
1741                 case PROC_TID_SMAPS:
1742                 case PROC_TGID_SMAPS:
1743                         inode->i_fop = &proc_smaps_operations;
1744                         break;
1745 #endif
1746                 case PROC_TID_MOUNTSTATS:
1747                 case PROC_TGID_MOUNTSTATS:
1748                         inode->i_fop = &proc_mountstats_operations;
1749                         break;
1750 #ifdef CONFIG_SECURITY
1751                 case PROC_TID_ATTR:
1752                         inode->i_nlink = 2;
1753                         inode->i_op = &proc_tid_attr_inode_operations;
1754                         inode->i_fop = &proc_tid_attr_operations;
1755                         break;
1756                 case PROC_TGID_ATTR:
1757                         inode->i_nlink = 2;
1758                         inode->i_op = &proc_tgid_attr_inode_operations;
1759                         inode->i_fop = &proc_tgid_attr_operations;
1760                         break;
1761                 case PROC_TID_ATTR_CURRENT:
1762                 case PROC_TGID_ATTR_CURRENT:
1763                 case PROC_TID_ATTR_PREV:
1764                 case PROC_TGID_ATTR_PREV:
1765                 case PROC_TID_ATTR_EXEC:
1766                 case PROC_TGID_ATTR_EXEC:
1767                 case PROC_TID_ATTR_FSCREATE:
1768                 case PROC_TGID_ATTR_FSCREATE:
1769                 case PROC_TID_ATTR_KEYCREATE:
1770                 case PROC_TGID_ATTR_KEYCREATE:
1771                 case PROC_TID_ATTR_SOCKCREATE:
1772                 case PROC_TGID_ATTR_SOCKCREATE:
1773                         inode->i_fop = &proc_pid_attr_operations;
1774                         break;
1775 #endif
1776 #ifdef CONFIG_KALLSYMS
1777                 case PROC_TID_WCHAN:
1778                 case PROC_TGID_WCHAN:
1779                         inode->i_fop = &proc_info_file_operations;
1780                         ei->op.proc_read = proc_pid_wchan;
1781                         break;
1782 #endif
1783 #ifdef CONFIG_SCHEDSTATS
1784                 case PROC_TID_SCHEDSTAT:
1785                 case PROC_TGID_SCHEDSTAT:
1786                         inode->i_fop = &proc_info_file_operations;
1787                         ei->op.proc_read = proc_pid_schedstat;
1788                         break;
1789 #endif
1790 #ifdef CONFIG_CPUSETS
1791                 case PROC_TID_CPUSET:
1792                 case PROC_TGID_CPUSET:
1793                         inode->i_fop = &proc_cpuset_operations;
1794                         break;
1795 #endif
1796                 case PROC_TID_OOM_SCORE:
1797                 case PROC_TGID_OOM_SCORE:
1798                         inode->i_fop = &proc_info_file_operations;
1799                         ei->op.proc_read = proc_oom_score;
1800                         break;
1801                 case PROC_TID_OOM_ADJUST:
1802                 case PROC_TGID_OOM_ADJUST:
1803                         inode->i_fop = &proc_oom_adjust_operations;
1804                         break;
1805 #ifdef CONFIG_AUDITSYSCALL
1806                 case PROC_TID_LOGINUID:
1807                 case PROC_TGID_LOGINUID:
1808                         inode->i_fop = &proc_loginuid_operations;
1809                         break;
1810 #endif
1811                 default:
1812                         printk("procfs: impossible type (%d)",p->type);
1813                         iput(inode);
1814                         error = ERR_PTR(-EINVAL);
1815                         goto out;
1816         }
1817         dentry->d_op = &pid_dentry_operations;
1818         d_add(dentry, inode);
1819         /* Close the race of the process dying before we return the dentry */
1820         if (pid_revalidate(dentry, NULL))
1821                 error = NULL;
1822 out:
1823         put_task_struct(task);
1824 out_no_task:
1825         return error;
1826 }
1827
1828 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1829         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1830 }
1831
1832 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1833         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1834 }
1835
1836 static struct file_operations proc_tgid_base_operations = {
1837         .read           = generic_read_dir,
1838         .readdir        = proc_tgid_base_readdir,
1839 };
1840
1841 static struct file_operations proc_tid_base_operations = {
1842         .read           = generic_read_dir,
1843         .readdir        = proc_tid_base_readdir,
1844 };
1845
1846 static struct inode_operations proc_tgid_base_inode_operations = {
1847         .lookup         = proc_tgid_base_lookup,
1848         .getattr        = pid_getattr,
1849 };
1850
1851 static struct inode_operations proc_tid_base_inode_operations = {
1852         .lookup         = proc_tid_base_lookup,
1853         .getattr        = pid_getattr,
1854 };
1855
1856 #ifdef CONFIG_SECURITY
1857 static int proc_tgid_attr_readdir(struct file * filp,
1858                              void * dirent, filldir_t filldir)
1859 {
1860         return proc_pident_readdir(filp,dirent,filldir,
1861                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1862 }
1863
1864 static int proc_tid_attr_readdir(struct file * filp,
1865                              void * dirent, filldir_t filldir)
1866 {
1867         return proc_pident_readdir(filp,dirent,filldir,
1868                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1869 }
1870
1871 static struct file_operations proc_tgid_attr_operations = {
1872         .read           = generic_read_dir,
1873         .readdir        = proc_tgid_attr_readdir,
1874 };
1875
1876 static struct file_operations proc_tid_attr_operations = {
1877         .read           = generic_read_dir,
1878         .readdir        = proc_tid_attr_readdir,
1879 };
1880
1881 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1882                                 struct dentry *dentry, struct nameidata *nd)
1883 {
1884         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1885 }
1886
1887 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1888                                 struct dentry *dentry, struct nameidata *nd)
1889 {
1890         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1891 }
1892
1893 static struct inode_operations proc_tgid_attr_inode_operations = {
1894         .lookup         = proc_tgid_attr_lookup,
1895         .getattr        = pid_getattr,
1896 };
1897
1898 static struct inode_operations proc_tid_attr_inode_operations = {
1899         .lookup         = proc_tid_attr_lookup,
1900         .getattr        = pid_getattr,
1901 };
1902 #endif
1903
1904 /*
1905  * /proc/self:
1906  */
1907 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1908                               int buflen)
1909 {
1910         char tmp[PROC_NUMBUF];
1911         sprintf(tmp, "%d", current->tgid);
1912         return vfs_readlink(dentry,buffer,buflen,tmp);
1913 }
1914
1915 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1916 {
1917         char tmp[PROC_NUMBUF];
1918         sprintf(tmp, "%d", current->tgid);
1919         return ERR_PTR(vfs_follow_link(nd,tmp));
1920 }
1921
1922 static struct inode_operations proc_self_inode_operations = {
1923         .readlink       = proc_self_readlink,
1924         .follow_link    = proc_self_follow_link,
1925 };
1926
1927 /**
1928  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
1929  *
1930  * @task: task that should be flushed.
1931  *
1932  * Looks in the dcache for
1933  * /proc/@pid
1934  * /proc/@tgid/task/@pid
1935  * if either directory is present flushes it and all of it'ts children
1936  * from the dcache.
1937  *
1938  * It is safe and reasonable to cache /proc entries for a task until
1939  * that task exits.  After that they just clog up the dcache with
1940  * useless entries, possibly causing useful dcache entries to be
1941  * flushed instead.  This routine is proved to flush those useless
1942  * dcache entries at process exit time.
1943  *
1944  * NOTE: This routine is just an optimization so it does not guarantee
1945  *       that no dcache entries will exist at process exit time it
1946  *       just makes it very unlikely that any will persist.
1947  */
1948 void proc_flush_task(struct task_struct *task)
1949 {
1950         struct dentry *dentry, *leader, *dir;
1951         char buf[PROC_NUMBUF];
1952         struct qstr name;
1953
1954         name.name = buf;
1955         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1956         dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1957         if (dentry) {
1958                 shrink_dcache_parent(dentry);
1959                 d_drop(dentry);
1960                 dput(dentry);
1961         }
1962
1963         if (thread_group_leader(task))
1964                 goto out;
1965
1966         name.name = buf;
1967         name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
1968         leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1969         if (!leader)
1970                 goto out;
1971
1972         name.name = "task";
1973         name.len = strlen(name.name);
1974         dir = d_hash_and_lookup(leader, &name);
1975         if (!dir)
1976                 goto out_put_leader;
1977
1978         name.name = buf;
1979         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1980         dentry = d_hash_and_lookup(dir, &name);
1981         if (dentry) {
1982                 shrink_dcache_parent(dentry);
1983                 d_drop(dentry);
1984                 dput(dentry);
1985         }
1986
1987         dput(dir);
1988 out_put_leader:
1989         dput(leader);
1990 out:
1991         return;
1992 }
1993
1994 /* SMP-safe */
1995 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1996 {
1997         struct dentry *result = ERR_PTR(-ENOENT);
1998         struct task_struct *task;
1999         struct inode *inode;
2000         struct proc_inode *ei;
2001         unsigned tgid;
2002
2003         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
2004                 inode = new_inode(dir->i_sb);
2005                 if (!inode)
2006                         return ERR_PTR(-ENOMEM);
2007                 ei = PROC_I(inode);
2008                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2009                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
2010                 ei->pde = NULL;
2011                 inode->i_mode = S_IFLNK|S_IRWXUGO;
2012                 inode->i_uid = inode->i_gid = 0;
2013                 inode->i_size = 64;
2014                 inode->i_op = &proc_self_inode_operations;
2015                 d_add(dentry, inode);
2016                 return NULL;
2017         }
2018         tgid = name_to_int(dentry);
2019         if (tgid == ~0U)
2020                 goto out;
2021
2022         rcu_read_lock();
2023         task = find_task_by_pid(tgid);
2024         if (task)
2025                 get_task_struct(task);
2026         rcu_read_unlock();
2027         if (!task)
2028                 goto out;
2029
2030         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2031         if (!inode)
2032                 goto out_put_task;
2033
2034         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2035         inode->i_op = &proc_tgid_base_inode_operations;
2036         inode->i_fop = &proc_tgid_base_operations;
2037         inode->i_flags|=S_IMMUTABLE;
2038 #ifdef CONFIG_SECURITY
2039         inode->i_nlink = 5;
2040 #else
2041         inode->i_nlink = 4;
2042 #endif
2043
2044         dentry->d_op = &pid_dentry_operations;
2045
2046         d_add(dentry, inode);
2047         /* Close the race of the process dying before we return the dentry */
2048         if (pid_revalidate(dentry, NULL))
2049                 result = NULL;
2050
2051 out_put_task:
2052         put_task_struct(task);
2053 out:
2054         return result;
2055 }
2056
2057 /* SMP-safe */
2058 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2059 {
2060         struct dentry *result = ERR_PTR(-ENOENT);
2061         struct task_struct *task;
2062         struct task_struct *leader = get_proc_task(dir);
2063         struct inode *inode;
2064         unsigned tid;
2065
2066         if (!leader)
2067                 goto out_no_task;
2068
2069         tid = name_to_int(dentry);
2070         if (tid == ~0U)
2071                 goto out;
2072
2073         rcu_read_lock();
2074         task = find_task_by_pid(tid);
2075         if (task)
2076                 get_task_struct(task);
2077         rcu_read_unlock();
2078         if (!task)
2079                 goto out;
2080         if (leader->tgid != task->tgid)
2081                 goto out_drop_task;
2082
2083         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2084
2085
2086         if (!inode)
2087                 goto out_drop_task;
2088         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2089         inode->i_op = &proc_tid_base_inode_operations;
2090         inode->i_fop = &proc_tid_base_operations;
2091         inode->i_flags|=S_IMMUTABLE;
2092 #ifdef CONFIG_SECURITY
2093         inode->i_nlink = 4;
2094 #else
2095         inode->i_nlink = 3;
2096 #endif
2097
2098         dentry->d_op = &pid_dentry_operations;
2099
2100         d_add(dentry, inode);
2101         /* Close the race of the process dying before we return the dentry */
2102         if (pid_revalidate(dentry, NULL))
2103                 result = NULL;
2104
2105 out_drop_task:
2106         put_task_struct(task);
2107 out:
2108         put_task_struct(leader);
2109 out_no_task:
2110         return result;
2111 }
2112
2113 /*
2114  * Find the first tgid to return to user space.
2115  *
2116  * Usually this is just whatever follows &init_task, but if the users
2117  * buffer was too small to hold the full list or there was a seek into
2118  * the middle of the directory we have more work to do.
2119  *
2120  * In the case of a short read we start with find_task_by_pid.
2121  *
2122  * In the case of a seek we start with &init_task and walk nr
2123  * threads past it.
2124  */
2125 static struct task_struct *first_tgid(int tgid, unsigned int nr)
2126 {
2127         struct task_struct *pos;
2128         rcu_read_lock();
2129         if (tgid && nr) {
2130                 pos = find_task_by_pid(tgid);
2131                 if (pos && thread_group_leader(pos))
2132                         goto found;
2133         }
2134         /* If nr exceeds the number of processes get out quickly */
2135         pos = NULL;
2136         if (nr && nr >= nr_processes())
2137                 goto done;
2138
2139         /* If we haven't found our starting place yet start with
2140          * the init_task and walk nr tasks forward.
2141          */
2142         for (pos = next_task(&init_task); nr > 0; --nr) {
2143                 pos = next_task(pos);
2144                 if (pos == &init_task) {
2145                         pos = NULL;
2146                         goto done;
2147                 }
2148         }
2149 found:
2150         get_task_struct(pos);
2151 done:
2152         rcu_read_unlock();
2153         return pos;
2154 }
2155
2156 /*
2157  * Find the next task in the task list.
2158  * Return NULL if we loop or there is any error.
2159  *
2160  * The reference to the input task_struct is released.
2161  */
2162 static struct task_struct *next_tgid(struct task_struct *start)
2163 {
2164         struct task_struct *pos;
2165         rcu_read_lock();
2166         pos = start;
2167         if (pid_alive(start))
2168                 pos = next_task(start);
2169         if (pid_alive(pos) && (pos != &init_task)) {
2170                 get_task_struct(pos);
2171                 goto done;
2172         }
2173         pos = NULL;
2174 done:
2175         rcu_read_unlock();
2176         put_task_struct(start);
2177         return pos;
2178 }
2179
2180 /* for the /proc/ directory itself, after non-process stuff has been done */
2181 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2182 {
2183         char buf[PROC_NUMBUF];
2184         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2185         struct task_struct *task;
2186         int tgid;
2187
2188         if (!nr) {
2189                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2190                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2191                         return 0;
2192                 filp->f_pos++;
2193                 nr++;
2194         }
2195         nr -= 1;
2196
2197         /* f_version caches the tgid value that the last readdir call couldn't
2198          * return. lseek aka telldir automagically resets f_version to 0.
2199          */
2200         tgid = filp->f_version;
2201         filp->f_version = 0;
2202         for (task = first_tgid(tgid, nr);
2203              task;
2204              task = next_tgid(task), filp->f_pos++) {
2205                 int len;
2206                 ino_t ino;
2207                 tgid = task->pid;
2208                 len = snprintf(buf, sizeof(buf), "%d", tgid);
2209                 ino = fake_ino(tgid, PROC_TGID_INO);
2210                 if (filldir(dirent, buf, len, filp->f_pos, ino, DT_DIR) < 0) {
2211                         /* returning this tgid failed, save it as the first
2212                          * pid for the next readir call */
2213                         filp->f_version = tgid;
2214                         put_task_struct(task);
2215                         break;
2216                 }
2217         }
2218         return 0;
2219 }
2220
2221 /*
2222  * Find the first tid of a thread group to return to user space.
2223  *
2224  * Usually this is just the thread group leader, but if the users
2225  * buffer was too small or there was a seek into the middle of the
2226  * directory we have more work todo.
2227  *
2228  * In the case of a short read we start with find_task_by_pid.
2229  *
2230  * In the case of a seek we start with the leader and walk nr
2231  * threads past it.
2232  */
2233 static struct task_struct *first_tid(struct task_struct *leader,
2234                                         int tid, int nr)
2235 {
2236         struct task_struct *pos;
2237
2238         rcu_read_lock();
2239         /* Attempt to start with the pid of a thread */
2240         if (tid && (nr > 0)) {
2241                 pos = find_task_by_pid(tid);
2242                 if (pos && (pos->group_leader == leader))
2243                         goto found;
2244         }
2245
2246         /* If nr exceeds the number of threads there is nothing todo */
2247         pos = NULL;
2248         if (nr && nr >= get_nr_threads(leader))
2249                 goto out;
2250
2251         /* If we haven't found our starting place yet start
2252          * with the leader and walk nr threads forward.
2253          */
2254         for (pos = leader; nr > 0; --nr) {
2255                 pos = next_thread(pos);
2256                 if (pos == leader) {
2257                         pos = NULL;
2258                         goto out;
2259                 }
2260         }
2261 found:
2262         get_task_struct(pos);
2263 out:
2264         rcu_read_unlock();
2265         return pos;
2266 }
2267
2268 /*
2269  * Find the next thread in the thread list.
2270  * Return NULL if there is an error or no next thread.
2271  *
2272  * The reference to the input task_struct is released.
2273  */
2274 static struct task_struct *next_tid(struct task_struct *start)
2275 {
2276         struct task_struct *pos = NULL;
2277         rcu_read_lock();
2278         if (pid_alive(start)) {
2279                 pos = next_thread(start);
2280                 if (thread_group_leader(pos))
2281                         pos = NULL;
2282                 else
2283                         get_task_struct(pos);
2284         }
2285         rcu_read_unlock();
2286         put_task_struct(start);
2287         return pos;
2288 }
2289
2290 /* for the /proc/TGID/task/ directories */
2291 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2292 {
2293         char buf[PROC_NUMBUF];
2294         struct dentry *dentry = filp->f_dentry;
2295         struct inode *inode = dentry->d_inode;
2296         struct task_struct *leader = get_proc_task(inode);
2297         struct task_struct *task;
2298         int retval = -ENOENT;
2299         ino_t ino;
2300         int tid;
2301         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2302
2303         if (!leader)
2304                 goto out_no_task;
2305         retval = 0;
2306
2307         switch (pos) {
2308         case 0:
2309                 ino = inode->i_ino;
2310                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2311                         goto out;
2312                 pos++;
2313                 /* fall through */
2314         case 1:
2315                 ino = parent_ino(dentry);
2316                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2317                         goto out;
2318                 pos++;
2319                 /* fall through */
2320         }
2321
2322         /* f_version caches the tgid value that the last readdir call couldn't
2323          * return. lseek aka telldir automagically resets f_version to 0.
2324          */
2325         tid = filp->f_version;
2326         filp->f_version = 0;
2327         for (task = first_tid(leader, tid, pos - 2);
2328              task;
2329              task = next_tid(task), pos++) {
2330                 int len;
2331                 tid = task->pid;
2332                 len = snprintf(buf, sizeof(buf), "%d", tid);
2333                 ino = fake_ino(tid, PROC_TID_INO);
2334                 if (filldir(dirent, buf, len, pos, ino, DT_DIR < 0)) {
2335                         /* returning this tgid failed, save it as the first
2336                          * pid for the next readir call */
2337                         filp->f_version = tid;
2338                         put_task_struct(task);
2339                         break;
2340                 }
2341         }
2342 out:
2343         filp->f_pos = pos;
2344         put_task_struct(leader);
2345 out_no_task:
2346         return retval;
2347 }
2348
2349 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2350 {
2351         struct inode *inode = dentry->d_inode;
2352         struct task_struct *p = get_proc_task(inode);
2353         generic_fillattr(inode, stat);
2354
2355         if (p) {
2356                 rcu_read_lock();
2357                 stat->nlink += get_nr_threads(p);
2358                 rcu_read_unlock();
2359                 put_task_struct(p);
2360         }
2361
2362         return 0;
2363 }