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