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 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 555 {
 556         int error;
 557         struct inode *inode = dentry->d_inode;
 558
 559         if (attr->ia_valid & ATTR_MODE)
 560                 return -EPERM;
 561
 562         error = inode_change_ok(inode, attr);
 563         if (!error) {
 564                 error = security_inode_setattr(dentry, attr);
 565                 if (!error)
 566                         error = inode_setattr(inode, attr);
 567         }
 568         return error;
 569 }
 570
 571 static struct inode_operations proc_def_inode_operations = {
 572         .setattr        = proc_setattr,
 573 };
 574
 575 extern struct seq_operations mounts_op;
 576 struct proc_mounts {
 577         struct seq_file m;
 578         int event;
 579 };
 580
 581 static int mounts_open(struct inode *inode, struct file *file)
 582 {
 583         struct task_struct *task = get_proc_task(inode);
 584         struct namespace *namespace = NULL;
 585         struct proc_mounts *p;
 586         int ret = -EINVAL;
 587
 588         if (task) {
 589                 task_lock(task);
 590                 namespace = task->namespace;
 591                 if (namespace)
 592                         get_namespace(namespace);
 593                 task_unlock(task);
 594                 put_task_struct(task);
 595         }
 596
 597         if (namespace) {
 598                 ret = -ENOMEM;
 599                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 600                 if (p) {
 601                         file->private_data = &p->m;
 602                         ret = seq_open(file, &mounts_op);
 603                         if (!ret) {
 604                                 p->m.private = namespace;
 605                                 p->event = namespace->event;
 606                                 return 0;
 607                         }
 608                         kfree(p);
 609                 }
 610                 put_namespace(namespace);
 611         }
 612         return ret;
 613 }
 614
 615 static int mounts_release(struct inode *inode, struct file *file)
 616 {
 617         struct seq_file *m = file->private_data;
 618         struct namespace *namespace = m->private;
 619         put_namespace(namespace);
 620         return seq_release(inode, file);
 621 }
 622
 623 static unsigned mounts_poll(struct file *file, poll_table *wait)
 624 {
 625         struct proc_mounts *p = file->private_data;
 626         struct namespace *ns = p->m.private;
 627         unsigned res = 0;
 628
 629         poll_wait(file, &ns->poll, wait);
 630
 631         spin_lock(&vfsmount_lock);
 632         if (p->event != ns->event) {
 633                 p->event = ns->event;
 634                 res = POLLERR;
 635         }
 636         spin_unlock(&vfsmount_lock);
 637
 638         return res;
 639 }
 640
 641 static struct file_operations proc_mounts_operations = {
 642         .open           = mounts_open,
 643         .read           = seq_read,
 644         .llseek         = seq_lseek,
 645         .release        = mounts_release,
 646         .poll           = mounts_poll,
 647 };
 648
 649 extern struct seq_operations mountstats_op;
 650 static int mountstats_open(struct inode *inode, struct file *file)
 651 {
 652         int ret = seq_open(file, &mountstats_op);
 653
 654         if (!ret) {
 655                 struct seq_file *m = file->private_data;
 656                 struct namespace *namespace = NULL;
 657                 struct task_struct *task = get_proc_task(inode);
 658
 659                 if (task) {
 660                         task_lock(task);
 661                         namespace = task->namespace;
 662                         if (namespace)
 663                                 get_namespace(namespace);
 664                         task_unlock(task);
 665                         put_task_struct(task);
 666                 }
 667
 668                 if (namespace)
 669                         m->private = namespace;
 670                 else {
 671                         seq_release(inode, file);
 672                         ret = -EINVAL;
 673                 }
 674         }
 675         return ret;
 676 }
 677
 678 static struct file_operations proc_mountstats_operations = {
 679         .open           = mountstats_open,
 680         .read           = seq_read,
 681         .llseek         = seq_lseek,
 682         .release        = mounts_release,
 683 };
 684
 685 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 686
 687 static ssize_t proc_info_read(struct file * file, char __user * buf,
 688                           size_t count, loff_t *ppos)
 689 {
 690         struct inode * inode = file->f_dentry->d_inode;
 691         unsigned long page;
 692         ssize_t length;
 693         struct task_struct *task = get_proc_task(inode);
 694
 695         length = -ESRCH;
 696         if (!task)
 697                 goto out_no_task;
 698
 699         if (count > PROC_BLOCK_SIZE)
 700                 count = PROC_BLOCK_SIZE;
 701
 702         length = -ENOMEM;
 703         if (!(page = __get_free_page(GFP_KERNEL)))
 704                 goto out;
 705
 706         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 707
 708         if (length >= 0)
 709                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 710         free_page(page);
 711 out:
 712         put_task_struct(task);
 713 out_no_task:
 714         return length;
 715 }
 716
 717 static struct file_operations proc_info_file_operations = {
 718         .read           = proc_info_read,
 719 };
 720
 721 static int mem_open(struct inode* inode, struct file* file)
 722 {
 723         file->private_data = (void*)((long)current->self_exec_id);
 724         return 0;
 725 }
 726
 727 static ssize_t mem_read(struct file * file, char __user * buf,
 728                         size_t count, loff_t *ppos)
 729 {
 730         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 731         char *page;
 732         unsigned long src = *ppos;
 733         int ret = -ESRCH;
 734         struct mm_struct *mm;
 735
 736         if (!task)
 737                 goto out_no_task;
 738
 739         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 740                 goto out;
 741
 742         ret = -ENOMEM;
 743         page = (char *)__get_free_page(GFP_USER);
 744         if (!page)
 745                 goto out;
 746
 747         ret = 0;
 748
 749         mm = get_task_mm(task);
 750         if (!mm)
 751                 goto out_free;
 752
 753         ret = -EIO;
 754
 755         if (file->private_data != (void*)((long)current->self_exec_id))
 756                 goto out_put;
 757
 758         ret = 0;
 759
 760         while (count > 0) {
 761                 int this_len, retval;
 762
 763                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 764                 retval = access_process_vm(task, src, page, this_len, 0);
 765                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 766                         if (!ret)
 767                                 ret = -EIO;
 768                         break;
 769                 }
 770
 771                 if (copy_to_user(buf, page, retval)) {
 772                         ret = -EFAULT;
 773                         break;
 774                 }
 775
 776                 ret += retval;
 777                 src += retval;
 778                 buf += retval;
 779                 count -= retval;
 780         }
 781         *ppos = src;
 782
 783 out_put:
 784         mmput(mm);
 785 out_free:
 786         free_page((unsigned long) page);
 787 out:
 788         put_task_struct(task);
 789 out_no_task:
 790         return ret;
 791 }
 792
 793 #define mem_write NULL
 794
 795 #ifndef mem_write
 796 /* This is a security hazard */
 797 static ssize_t mem_write(struct file * file, const char * buf,
 798                          size_t count, loff_t *ppos)
 799 {
 800         int copied;
 801         char *page;
 802         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 803         unsigned long dst = *ppos;
 804
 805         copied = -ESRCH;
 806         if (!task)
 807                 goto out_no_task;
 808
 809         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 810                 goto out;
 811
 812         copied = -ENOMEM;
 813         page = (char *)__get_free_page(GFP_USER);
 814         if (!page)
 815                 goto out;
 816
 817         copied = 0;
 818         while (count > 0) {
 819                 int this_len, retval;
 820
 821                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 822                 if (copy_from_user(page, buf, this_len)) {
 823                         copied = -EFAULT;
 824                         break;
 825                 }
 826                 retval = access_process_vm(task, dst, page, this_len, 1);
 827                 if (!retval) {
 828                         if (!copied)
 829                                 copied = -EIO;
 830                         break;
 831                 }
 832                 copied += retval;
 833                 buf += retval;
 834                 dst += retval;
 835                 count -= retval;
 836         }
 837         *ppos = dst;
 838         free_page((unsigned long) page);
 839 out:
 840         put_task_struct(task);
 841 out_no_task:
 842         return copied;
 843 }
 844 #endif
 845
 846 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 847 {
 848         switch (orig) {
 849         case 0:
 850                 file->f_pos = offset;
 851                 break;
 852         case 1:
 853                 file->f_pos += offset;
 854                 break;
 855         default:
 856                 return -EINVAL;
 857         }
 858         force_successful_syscall_return();
 859         return file->f_pos;
 860 }
 861
 862 static struct file_operations proc_mem_operations = {
 863         .llseek         = mem_lseek,
 864         .read           = mem_read,
 865         .write          = mem_write,
 866         .open           = mem_open,
 867 };
 868
 869 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 870                                 size_t count, loff_t *ppos)
 871 {
 872         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 873         char buffer[PROC_NUMBUF];
 874         size_t len;
 875         int oom_adjust;
 876         loff_t __ppos = *ppos;
 877
 878         if (!task)
 879                 return -ESRCH;
 880         oom_adjust = task->oomkilladj;
 881         put_task_struct(task);
 882
 883         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 884         if (__ppos >= len)
 885                 return 0;
 886         if (count > len-__ppos)
 887                 count = len-__ppos;
 888         if (copy_to_user(buf, buffer + __ppos, count))
 889                 return -EFAULT;
 890         *ppos = __ppos + count;
 891         return count;
 892 }
 893
 894 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 895                                 size_t count, loff_t *ppos)
 896 {
 897         struct task_struct *task;
 898         char buffer[PROC_NUMBUF], *end;
 899         int oom_adjust;
 900
 901         if (!capable(CAP_SYS_RESOURCE))
 902                 return -EPERM;
 903         memset(buffer, 0, sizeof(buffer));
 904         if (count > sizeof(buffer) - 1)
 905                 count = sizeof(buffer) - 1;
 906         if (copy_from_user(buffer, buf, count))
 907                 return -EFAULT;
 908         oom_adjust = simple_strtol(buffer, &end, 0);
 909         if ((oom_adjust < -16 || oom_adjust > 15) && oom_adjust != OOM_DISABLE)
 910                 return -EINVAL;
 911         if (*end == '\n')
 912                 end++;
 913         task = get_proc_task(file->f_dentry->d_inode);
 914         if (!task)
 915                 return -ESRCH;
 916         task->oomkilladj = oom_adjust;
 917         put_task_struct(task);
 918         if (end - buffer == 0)
 919                 return -EIO;
 920         return end - buffer;
 921 }
 922
 923 static struct file_operations proc_oom_adjust_operations = {
 924         .read           = oom_adjust_read,
 925         .write          = oom_adjust_write,
 926 };
 927
 928 #ifdef CONFIG_AUDITSYSCALL
 929 #define TMPBUFLEN 21
 930 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 931                                   size_t count, loff_t *ppos)
 932 {
 933         struct inode * inode = file->f_dentry->d_inode;
 934         struct task_struct *task = get_proc_task(inode);
 935         ssize_t length;
 936         char tmpbuf[TMPBUFLEN];
 937
 938         if (!task)
 939                 return -ESRCH;
 940         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 941                                 audit_get_loginuid(task->audit_context));
 942         put_task_struct(task);
 943         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 944 }
 945
 946 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 947                                    size_t count, loff_t *ppos)
 948 {
 949         struct inode * inode = file->f_dentry->d_inode;
 950         char *page, *tmp;
 951         ssize_t length;
 952         uid_t loginuid;
 953
 954         if (!capable(CAP_AUDIT_CONTROL))
 955                 return -EPERM;
 956
 957         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
 958                 return -EPERM;
 959
 960         if (count >= PAGE_SIZE)
 961                 count = PAGE_SIZE - 1;
 962
 963         if (*ppos != 0) {
 964                 /* No partial writes. */
 965                 return -EINVAL;
 966         }
 967         page = (char*)__get_free_page(GFP_USER);
 968         if (!page)
 969                 return -ENOMEM;
 970         length = -EFAULT;
 971         if (copy_from_user(page, buf, count))
 972                 goto out_free_page;
 973
 974         page[count] = '\0';
 975         loginuid = simple_strtoul(page, &tmp, 10);
 976         if (tmp == page) {
 977                 length = -EINVAL;
 978                 goto out_free_page;
 979
 980         }
 981         length = audit_set_loginuid(current, loginuid);
 982         if (likely(length == 0))
 983                 length = count;
 984
 985 out_free_page:
 986         free_page((unsigned long) page);
 987         return length;
 988 }
 989
 990 static struct file_operations proc_loginuid_operations = {
 991         .read           = proc_loginuid_read,
 992         .write          = proc_loginuid_write,
 993 };
 994 #endif
 995
 996 #ifdef CONFIG_SECCOMP
 997 static ssize_t seccomp_read(struct file *file, char __user *buf,
 998                             size_t count, loff_t *ppos)
 999 {
1000         struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
1001         char __buf[20];
1002         loff_t __ppos = *ppos;
1003         size_t len;
1004
1005         if (!tsk)
1006                 return -ESRCH;
1007         /* no need to print the trailing zero, so use only len */
1008         len = sprintf(__buf, "%u\n", tsk->seccomp.mode);
1009         put_task_struct(tsk);
1010         if (__ppos >= len)
1011                 return 0;
1012         if (count > len - __ppos)
1013                 count = len - __ppos;
1014         if (copy_to_user(buf, __buf + __ppos, count))
1015                 return -EFAULT;
1016         *ppos = __ppos + count;
1017         return count;
1018 }
1019
1020 static ssize_t seccomp_write(struct file *file, const char __user *buf,
1021                              size_t count, loff_t *ppos)
1022 {
1023         struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode);
1024         char __buf[20], *end;
1025         unsigned int seccomp_mode;
1026         ssize_t result;
1027
1028         result = -ESRCH;
1029         if (!tsk)
1030                 goto out_no_task;
1031
1032         /* can set it only once to be even more secure */
1033         result = -EPERM;
1034         if (unlikely(tsk->seccomp.mode))
1035                 goto out;
1036
1037         result = -EFAULT;
1038         memset(__buf, 0, sizeof(__buf));
1039         count = min(count, sizeof(__buf) - 1);
1040         if (copy_from_user(__buf, buf, count))
1041                 goto out;
1042
1043         seccomp_mode = simple_strtoul(__buf, &end, 0);
1044         if (*end == '\n')
1045                 end++;
1046         result = -EINVAL;
1047         if (seccomp_mode && seccomp_mode <= NR_SECCOMP_MODES) {
1048                 tsk->seccomp.mode = seccomp_mode;
1049                 set_tsk_thread_flag(tsk, TIF_SECCOMP);
1050         } else
1051                 goto out;
1052         result = -EIO;
1053         if (unlikely(!(end - __buf)))
1054                 goto out;
1055         result = end - __buf;
1056 out:
1057         put_task_struct(tsk);
1058 out_no_task:
1059         return result;
1060 }
1061
1062 static struct file_operations proc_seccomp_operations = {
1063         .read           = seccomp_read,
1064         .write          = seccomp_write,
1065 };
1066 #endif /* CONFIG_SECCOMP */
1067
1068 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1069 {
1070         struct inode *inode = dentry->d_inode;
1071         int error = -EACCES;
1072
1073         /* We don't need a base pointer in the /proc filesystem */
1074         path_release(nd);
1075
1076         /* Are we allowed to snoop on the tasks file descriptors? */
1077         if (!proc_fd_access_allowed(inode))
1078                 goto out;
1079
1080         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1081         nd->last_type = LAST_BIND;
1082 out:
1083         return ERR_PTR(error);
1084 }
1085
1086 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1087                             char __user *buffer, int buflen)
1088 {
1089         struct inode * inode;
1090         char *tmp = (char*)__get_free_page(GFP_KERNEL), *path;
1091         int len;
1092
1093         if (!tmp)
1094                 return -ENOMEM;
1095
1096         inode = dentry->d_inode;
1097         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1098         len = PTR_ERR(path);
1099         if (IS_ERR(path))
1100                 goto out;
1101         len = tmp + PAGE_SIZE - 1 - path;
1102
1103         if (len > buflen)
1104                 len = buflen;
1105         if (copy_to_user(buffer, path, len))
1106                 len = -EFAULT;
1107  out:
1108         free_page((unsigned long)tmp);
1109         return len;
1110 }
1111
1112 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1113 {
1114         int error = -EACCES;
1115         struct inode *inode = dentry->d_inode;
1116         struct dentry *de;
1117         struct vfsmount *mnt = NULL;
1118
1119         /* Are we allowed to snoop on the tasks file descriptors? */
1120         if (!proc_fd_access_allowed(inode))
1121                 goto out;
1122
1123         error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
1124         if (error)
1125                 goto out;
1126
1127         error = do_proc_readlink(de, mnt, buffer, buflen);
1128         dput(de);
1129         mntput(mnt);
1130 out:
1131         return error;
1132 }
1133
1134 static struct inode_operations proc_pid_link_inode_operations = {
1135         .readlink       = proc_pid_readlink,
1136         .follow_link    = proc_pid_follow_link,
1137         .setattr        = proc_setattr,
1138 };
1139
1140 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
1141 {
1142         struct dentry *dentry = filp->f_dentry;
1143         struct inode *inode = dentry->d_inode;
1144         struct task_struct *p = get_proc_task(inode);
1145         unsigned int fd, tid, ino;
1146         int retval;
1147         char buf[PROC_NUMBUF];
1148         struct files_struct * files;
1149         struct fdtable *fdt;
1150
1151         retval = -ENOENT;
1152         if (!p)
1153                 goto out_no_task;
1154         retval = 0;
1155         tid = p->pid;
1156
1157         fd = filp->f_pos;
1158         switch (fd) {
1159                 case 0:
1160                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1161                                 goto out;
1162                         filp->f_pos++;
1163                 case 1:
1164                         ino = parent_ino(dentry);
1165                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1166                                 goto out;
1167                         filp->f_pos++;
1168                 default:
1169                         files = get_files_struct(p);
1170                         if (!files)
1171                                 goto out;
1172                         rcu_read_lock();
1173                         fdt = files_fdtable(files);
1174                         for (fd = filp->f_pos-2;
1175                              fd < fdt->max_fds;
1176                              fd++, filp->f_pos++) {
1177                                 unsigned int i,j;
1178
1179                                 if (!fcheck_files(files, fd))
1180                                         continue;
1181                                 rcu_read_unlock();
1182
1183                                 j = PROC_NUMBUF;
1184                                 i = fd;
1185                                 do {
1186                                         j--;
1187                                         buf[j] = '0' + (i % 10);
1188                                         i /= 10;
1189                                 } while (i);
1190
1191                                 ino = fake_ino(tid, PROC_TID_FD_DIR + fd);
1192                                 if (filldir(dirent, buf+j, PROC_NUMBUF-j, fd+2, ino, DT_LNK) < 0) {
1193                                         rcu_read_lock();
1194                                         break;
1195                                 }
1196                                 rcu_read_lock();
1197                         }
1198                         rcu_read_unlock();
1199                         put_files_struct(files);
1200         }
1201 out:
1202         put_task_struct(p);
1203 out_no_task:
1204         return retval;
1205 }
1206
1207 static int proc_pident_readdir(struct file *filp,
1208                 void *dirent, filldir_t filldir,
1209                 struct pid_entry *ents, unsigned int nents)
1210 {
1211         int i;
1212         int pid;
1213         struct dentry *dentry = filp->f_dentry;
1214         struct inode *inode = dentry->d_inode;
1215         struct task_struct *task = get_proc_task(inode);
1216         struct pid_entry *p;
1217         ino_t ino;
1218         int ret;
1219
1220         ret = -ENOENT;
1221         if (!task)
1222                 goto out;
1223
1224         ret = 0;
1225         pid = task->pid;
1226         put_task_struct(task);
1227         i = filp->f_pos;
1228         switch (i) {
1229         case 0:
1230                 ino = inode->i_ino;
1231                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1232                         goto out;
1233                 i++;
1234                 filp->f_pos++;
1235                 /* fall through */
1236         case 1:
1237                 ino = parent_ino(dentry);
1238                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1239                         goto out;
1240                 i++;
1241                 filp->f_pos++;
1242                 /* fall through */
1243         default:
1244                 i -= 2;
1245                 if (i >= nents) {
1246                         ret = 1;
1247                         goto out;
1248                 }
1249                 p = ents + i;
1250                 while (p->name) {
1251                         if (filldir(dirent, p->name, p->len, filp->f_pos,
1252                                     fake_ino(pid, p->type), p->mode >> 12) < 0)
1253                                 goto out;
1254                         filp->f_pos++;
1255                         p++;
1256                 }
1257         }
1258
1259         ret = 1;
1260 out:
1261         return ret;
1262 }
1263
1264 static int proc_tgid_base_readdir(struct file * filp,
1265                              void * dirent, filldir_t filldir)
1266 {
1267         return proc_pident_readdir(filp,dirent,filldir,
1268                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
1269 }
1270
1271 static int proc_tid_base_readdir(struct file * filp,
1272                              void * dirent, filldir_t filldir)
1273 {
1274         return proc_pident_readdir(filp,dirent,filldir,
1275                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
1276 }
1277
1278 /* building an inode */
1279
1280 static int task_dumpable(struct task_struct *task)
1281 {
1282         int dumpable = 0;
1283         struct mm_struct *mm;
1284
1285         task_lock(task);
1286         mm = task->mm;
1287         if (mm)
1288                 dumpable = mm->dumpable;
1289         task_unlock(task);
1290         if(dumpable == 1)
1291                 return 1;
1292         return 0;
1293 }
1294
1295
1296 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task, int ino)
1297 {
1298         struct inode * inode;
1299         struct proc_inode *ei;
1300
1301         /* We need a new inode */
1302
1303         inode = new_inode(sb);
1304         if (!inode)
1305                 goto out;
1306
1307         /* Common stuff */
1308         ei = PROC_I(inode);
1309         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1310         inode->i_ino = fake_ino(task->pid, ino);
1311         inode->i_op = &proc_def_inode_operations;
1312
1313         /*
1314          * grab the reference to task.
1315          */
1316         ei->pid = get_pid(task->pids[PIDTYPE_PID].pid);
1317         if (!ei->pid)
1318                 goto out_unlock;
1319
1320         inode->i_uid = 0;
1321         inode->i_gid = 0;
1322         if (task_dumpable(task)) {
1323                 inode->i_uid = task->euid;
1324                 inode->i_gid = task->egid;
1325         }
1326         security_task_to_inode(task, inode);
1327
1328 out:
1329         return inode;
1330
1331 out_unlock:
1332         iput(inode);
1333         return NULL;
1334 }
1335
1336 /* dentry stuff */
1337
1338 /*
1339  *      Exceptional case: normally we are not allowed to unhash a busy
1340  * directory. In this case, however, we can do it - no aliasing problems
1341  * due to the way we treat inodes.
1342  *
1343  * Rewrite the inode's ownerships here because the owning task may have
1344  * performed a setuid(), etc.
1345  *
1346  * Before the /proc/pid/status file was created the only way to read
1347  * the effective uid of a /process was to stat /proc/pid.  Reading
1348  * /proc/pid/status is slow enough that procps and other packages
1349  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1350  * made this apply to all per process world readable and executable
1351  * directories.
1352  */
1353 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1354 {
1355         struct inode *inode = dentry->d_inode;
1356         struct task_struct *task = get_proc_task(inode);
1357         if (task) {
1358                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1359                     task_dumpable(task)) {
1360                         inode->i_uid = task->euid;
1361                         inode->i_gid = task->egid;
1362                 } else {
1363                         inode->i_uid = 0;
1364                         inode->i_gid = 0;
1365                 }
1366                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1367                 security_task_to_inode(task, inode);
1368                 put_task_struct(task);
1369                 return 1;
1370         }
1371         d_drop(dentry);
1372         return 0;
1373 }
1374
1375 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1376 {
1377         struct inode *inode = dentry->d_inode;
1378         struct task_struct *task;
1379         generic_fillattr(inode, stat);
1380
1381         rcu_read_lock();
1382         stat->uid = 0;
1383         stat->gid = 0;
1384         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1385         if (task) {
1386                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1387                     task_dumpable(task)) {
1388                         stat->uid = task->euid;
1389                         stat->gid = task->egid;
1390                 }
1391         }
1392         rcu_read_unlock();
1393         return 0;
1394 }
1395
1396 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1397 {
1398         struct inode *inode = dentry->d_inode;
1399         struct task_struct *task = get_proc_task(inode);
1400         int fd = proc_fd(inode);
1401         struct files_struct *files;
1402
1403         if (task) {
1404                 files = get_files_struct(task);
1405                 if (files) {
1406                         rcu_read_lock();
1407                         if (fcheck_files(files, fd)) {
1408                                 rcu_read_unlock();
1409                                 put_files_struct(files);
1410                                 if (task_dumpable(task)) {
1411                                         inode->i_uid = task->euid;
1412                                         inode->i_gid = task->egid;
1413                                 } else {
1414                                         inode->i_uid = 0;
1415                                         inode->i_gid = 0;
1416                                 }
1417                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1418                                 security_task_to_inode(task, inode);
1419                                 put_task_struct(task);
1420                                 return 1;
1421                         }
1422                         rcu_read_unlock();
1423                         put_files_struct(files);
1424                 }
1425                 put_task_struct(task);
1426         }
1427         d_drop(dentry);
1428         return 0;
1429 }
1430
1431 static int pid_delete_dentry(struct dentry * dentry)
1432 {
1433         /* Is the task we represent dead?
1434          * If so, then don't put the dentry on the lru list,
1435          * kill it immediately.
1436          */
1437         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1438 }
1439
1440 static struct dentry_operations tid_fd_dentry_operations =
1441 {
1442         .d_revalidate   = tid_fd_revalidate,
1443         .d_delete       = pid_delete_dentry,
1444 };
1445
1446 static struct dentry_operations pid_dentry_operations =
1447 {
1448         .d_revalidate   = pid_revalidate,
1449         .d_delete       = pid_delete_dentry,
1450 };
1451
1452 /* Lookups */
1453
1454 static unsigned name_to_int(struct dentry *dentry)
1455 {
1456         const char *name = dentry->d_name.name;
1457         int len = dentry->d_name.len;
1458         unsigned n = 0;
1459
1460         if (len > 1 && *name == '0')
1461                 goto out;
1462         while (len-- > 0) {
1463                 unsigned c = *name++ - '0';
1464                 if (c > 9)
1465                         goto out;
1466                 if (n >= (~0U-9)/10)
1467                         goto out;
1468                 n *= 10;
1469                 n += c;
1470         }
1471         return n;
1472 out:
1473         return ~0U;
1474 }
1475
1476 /* SMP-safe */
1477 static struct dentry *proc_lookupfd(struct inode * dir, struct dentry * dentry, struct nameidata *nd)
1478 {
1479         struct task_struct *task = get_proc_task(dir);
1480         unsigned fd = name_to_int(dentry);
1481         struct dentry *result = ERR_PTR(-ENOENT);
1482         struct file * file;
1483         struct files_struct * files;
1484         struct inode *inode;
1485         struct proc_inode *ei;
1486
1487         if (!task)
1488                 goto out_no_task;
1489         if (fd == ~0U)
1490                 goto out;
1491
1492         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_FD_DIR+fd);
1493         if (!inode)
1494                 goto out;
1495         ei = PROC_I(inode);
1496         ei->fd = fd;
1497         files = get_files_struct(task);
1498         if (!files)
1499                 goto out_unlock;
1500         inode->i_mode = S_IFLNK;
1501
1502         /*
1503          * We are not taking a ref to the file structure, so we must
1504          * hold ->file_lock.
1505          */
1506         spin_lock(&files->file_lock);
1507         file = fcheck_files(files, fd);
1508         if (!file)
1509                 goto out_unlock2;
1510         if (file->f_mode & 1)
1511                 inode->i_mode |= S_IRUSR | S_IXUSR;
1512         if (file->f_mode & 2)
1513                 inode->i_mode |= S_IWUSR | S_IXUSR;
1514         spin_unlock(&files->file_lock);
1515         put_files_struct(files);
1516         inode->i_op = &proc_pid_link_inode_operations;
1517         inode->i_size = 64;
1518         ei->op.proc_get_link = proc_fd_link;
1519         dentry->d_op = &tid_fd_dentry_operations;
1520         d_add(dentry, inode);
1521         /* Close the race of the process dying before we return the dentry */
1522         if (tid_fd_revalidate(dentry, NULL))
1523                 result = NULL;
1524 out:
1525         put_task_struct(task);
1526 out_no_task:
1527         return result;
1528
1529 out_unlock2:
1530         spin_unlock(&files->file_lock);
1531         put_files_struct(files);
1532 out_unlock:
1533         iput(inode);
1534         goto out;
1535 }
1536
1537 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir);
1538 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd);
1539 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat);
1540
1541 static struct file_operations proc_fd_operations = {
1542         .read           = generic_read_dir,
1543         .readdir        = proc_readfd,
1544 };
1545
1546 static struct file_operations proc_task_operations = {
1547         .read           = generic_read_dir,
1548         .readdir        = proc_task_readdir,
1549 };
1550
1551 /*
1552  * proc directories can do almost nothing..
1553  */
1554 static struct inode_operations proc_fd_inode_operations = {
1555         .lookup         = proc_lookupfd,
1556         .setattr        = proc_setattr,
1557 };
1558
1559 static struct inode_operations proc_task_inode_operations = {
1560         .lookup         = proc_task_lookup,
1561         .getattr        = proc_task_getattr,
1562         .setattr        = proc_setattr,
1563 };
1564
1565 #ifdef CONFIG_SECURITY
1566 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1567                                   size_t count, loff_t *ppos)
1568 {
1569         struct inode * inode = file->f_dentry->d_inode;
1570         unsigned long page;
1571         ssize_t length;
1572         struct task_struct *task = get_proc_task(inode);
1573
1574         length = -ESRCH;
1575         if (!task)
1576                 goto out_no_task;
1577
1578         if (count > PAGE_SIZE)
1579                 count = PAGE_SIZE;
1580         length = -ENOMEM;
1581         if (!(page = __get_free_page(GFP_KERNEL)))
1582                 goto out;
1583
1584         length = security_getprocattr(task,
1585                                       (char*)file->f_dentry->d_name.name,
1586                                       (void*)page, count);
1587         if (length >= 0)
1588                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
1589         free_page(page);
1590 out:
1591         put_task_struct(task);
1592 out_no_task:
1593         return length;
1594 }
1595
1596 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1597                                    size_t count, loff_t *ppos)
1598 {
1599         struct inode * inode = file->f_dentry->d_inode;
1600         char *page;
1601         ssize_t length;
1602         struct task_struct *task = get_proc_task(inode);
1603
1604         length = -ESRCH;
1605         if (!task)
1606                 goto out_no_task;
1607         if (count > PAGE_SIZE)
1608                 count = PAGE_SIZE;
1609
1610         /* No partial writes. */
1611         length = -EINVAL;
1612         if (*ppos != 0)
1613                 goto out;
1614
1615         length = -ENOMEM;
1616         page = (char*)__get_free_page(GFP_USER);
1617         if (!page)
1618                 goto out;
1619
1620         length = -EFAULT;
1621         if (copy_from_user(page, buf, count))
1622                 goto out_free;
1623
1624         length = security_setprocattr(task,
1625                                       (char*)file->f_dentry->d_name.name,
1626                                       (void*)page, count);
1627 out_free:
1628         free_page((unsigned long) page);
1629 out:
1630         put_task_struct(task);
1631 out_no_task:
1632         return length;
1633 }
1634
1635 static struct file_operations proc_pid_attr_operations = {
1636         .read           = proc_pid_attr_read,
1637         .write          = proc_pid_attr_write,
1638 };
1639
1640 static struct file_operations proc_tid_attr_operations;
1641 static struct inode_operations proc_tid_attr_inode_operations;
1642 static struct file_operations proc_tgid_attr_operations;
1643 static struct inode_operations proc_tgid_attr_inode_operations;
1644 #endif
1645
1646 /* SMP-safe */
1647 static struct dentry *proc_pident_lookup(struct inode *dir,
1648                                          struct dentry *dentry,
1649                                          struct pid_entry *ents)
1650 {
1651         struct inode *inode;
1652         struct dentry *error;
1653         struct task_struct *task = get_proc_task(dir);
1654         struct pid_entry *p;
1655         struct proc_inode *ei;
1656
1657         error = ERR_PTR(-ENOENT);
1658         inode = NULL;
1659
1660         if (!task)
1661                 goto out_no_task;
1662
1663         for (p = ents; p->name; p++) {
1664                 if (p->len != dentry->d_name.len)
1665                         continue;
1666                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1667                         break;
1668         }
1669         if (!p->name)
1670                 goto out;
1671
1672         error = ERR_PTR(-EINVAL);
1673         inode = proc_pid_make_inode(dir->i_sb, task, p->type);
1674         if (!inode)
1675                 goto out;
1676
1677         ei = PROC_I(inode);
1678         inode->i_mode = p->mode;
1679         /*
1680          * Yes, it does not scale. And it should not. Don't add
1681          * new entries into /proc/<tgid>/ without very good reasons.
1682          */
1683         switch(p->type) {
1684                 case PROC_TGID_TASK:
1685                         inode->i_nlink = 2;
1686                         inode->i_op = &proc_task_inode_operations;
1687                         inode->i_fop = &proc_task_operations;
1688                         break;
1689                 case PROC_TID_FD:
1690                 case PROC_TGID_FD:
1691                         inode->i_nlink = 2;
1692                         inode->i_op = &proc_fd_inode_operations;
1693                         inode->i_fop = &proc_fd_operations;
1694                         break;
1695                 case PROC_TID_EXE:
1696                 case PROC_TGID_EXE:
1697                         inode->i_op = &proc_pid_link_inode_operations;
1698                         ei->op.proc_get_link = proc_exe_link;
1699                         break;
1700                 case PROC_TID_CWD:
1701                 case PROC_TGID_CWD:
1702                         inode->i_op = &proc_pid_link_inode_operations;
1703                         ei->op.proc_get_link = proc_cwd_link;
1704                         break;
1705                 case PROC_TID_ROOT:
1706                 case PROC_TGID_ROOT:
1707                         inode->i_op = &proc_pid_link_inode_operations;
1708                         ei->op.proc_get_link = proc_root_link;
1709                         break;
1710                 case PROC_TID_ENVIRON:
1711                 case PROC_TGID_ENVIRON:
1712                         inode->i_fop = &proc_info_file_operations;
1713                         ei->op.proc_read = proc_pid_environ;
1714                         break;
1715                 case PROC_TID_AUXV:
1716                 case PROC_TGID_AUXV:
1717                         inode->i_fop = &proc_info_file_operations;
1718                         ei->op.proc_read = proc_pid_auxv;
1719                         break;
1720                 case PROC_TID_STATUS:
1721                 case PROC_TGID_STATUS:
1722                         inode->i_fop = &proc_info_file_operations;
1723                         ei->op.proc_read = proc_pid_status;
1724                         break;
1725                 case PROC_TID_STAT:
1726                         inode->i_fop = &proc_info_file_operations;
1727                         ei->op.proc_read = proc_tid_stat;
1728                         break;
1729                 case PROC_TGID_STAT:
1730                         inode->i_fop = &proc_info_file_operations;
1731                         ei->op.proc_read = proc_tgid_stat;
1732                         break;
1733                 case PROC_TID_CMDLINE:
1734                 case PROC_TGID_CMDLINE:
1735                         inode->i_fop = &proc_info_file_operations;
1736                         ei->op.proc_read = proc_pid_cmdline;
1737                         break;
1738                 case PROC_TID_STATM:
1739                 case PROC_TGID_STATM:
1740                         inode->i_fop = &proc_info_file_operations;
1741                         ei->op.proc_read = proc_pid_statm;
1742                         break;
1743                 case PROC_TID_MAPS:
1744                 case PROC_TGID_MAPS:
1745                         inode->i_fop = &proc_maps_operations;
1746                         break;
1747 #ifdef CONFIG_NUMA
1748                 case PROC_TID_NUMA_MAPS:
1749                 case PROC_TGID_NUMA_MAPS:
1750                         inode->i_fop = &proc_numa_maps_operations;
1751                         break;
1752 #endif
1753                 case PROC_TID_MEM:
1754                 case PROC_TGID_MEM:
1755                         inode->i_fop = &proc_mem_operations;
1756                         break;
1757 #ifdef CONFIG_SECCOMP
1758                 case PROC_TID_SECCOMP:
1759                 case PROC_TGID_SECCOMP:
1760                         inode->i_fop = &proc_seccomp_operations;
1761                         break;
1762 #endif /* CONFIG_SECCOMP */
1763                 case PROC_TID_MOUNTS:
1764                 case PROC_TGID_MOUNTS:
1765                         inode->i_fop = &proc_mounts_operations;
1766                         break;
1767 #ifdef CONFIG_MMU
1768                 case PROC_TID_SMAPS:
1769                 case PROC_TGID_SMAPS:
1770                         inode->i_fop = &proc_smaps_operations;
1771                         break;
1772 #endif
1773                 case PROC_TID_MOUNTSTATS:
1774                 case PROC_TGID_MOUNTSTATS:
1775                         inode->i_fop = &proc_mountstats_operations;
1776                         break;
1777 #ifdef CONFIG_SECURITY
1778                 case PROC_TID_ATTR:
1779                         inode->i_nlink = 2;
1780                         inode->i_op = &proc_tid_attr_inode_operations;
1781                         inode->i_fop = &proc_tid_attr_operations;
1782                         break;
1783                 case PROC_TGID_ATTR:
1784                         inode->i_nlink = 2;
1785                         inode->i_op = &proc_tgid_attr_inode_operations;
1786                         inode->i_fop = &proc_tgid_attr_operations;
1787                         break;
1788                 case PROC_TID_ATTR_CURRENT:
1789                 case PROC_TGID_ATTR_CURRENT:
1790                 case PROC_TID_ATTR_PREV:
1791                 case PROC_TGID_ATTR_PREV:
1792                 case PROC_TID_ATTR_EXEC:
1793                 case PROC_TGID_ATTR_EXEC:
1794                 case PROC_TID_ATTR_FSCREATE:
1795                 case PROC_TGID_ATTR_FSCREATE:
1796                 case PROC_TID_ATTR_KEYCREATE:
1797                 case PROC_TGID_ATTR_KEYCREATE:
1798                 case PROC_TID_ATTR_SOCKCREATE:
1799                 case PROC_TGID_ATTR_SOCKCREATE:
1800                         inode->i_fop = &proc_pid_attr_operations;
1801                         break;
1802 #endif
1803 #ifdef CONFIG_KALLSYMS
1804                 case PROC_TID_WCHAN:
1805                 case PROC_TGID_WCHAN:
1806                         inode->i_fop = &proc_info_file_operations;
1807                         ei->op.proc_read = proc_pid_wchan;
1808                         break;
1809 #endif
1810 #ifdef CONFIG_SCHEDSTATS
1811                 case PROC_TID_SCHEDSTAT:
1812                 case PROC_TGID_SCHEDSTAT:
1813                         inode->i_fop = &proc_info_file_operations;
1814                         ei->op.proc_read = proc_pid_schedstat;
1815                         break;
1816 #endif
1817 #ifdef CONFIG_CPUSETS
1818                 case PROC_TID_CPUSET:
1819                 case PROC_TGID_CPUSET:
1820                         inode->i_fop = &proc_cpuset_operations;
1821                         break;
1822 #endif
1823                 case PROC_TID_OOM_SCORE:
1824                 case PROC_TGID_OOM_SCORE:
1825                         inode->i_fop = &proc_info_file_operations;
1826                         ei->op.proc_read = proc_oom_score;
1827                         break;
1828                 case PROC_TID_OOM_ADJUST:
1829                 case PROC_TGID_OOM_ADJUST:
1830                         inode->i_fop = &proc_oom_adjust_operations;
1831                         break;
1832 #ifdef CONFIG_AUDITSYSCALL
1833                 case PROC_TID_LOGINUID:
1834                 case PROC_TGID_LOGINUID:
1835                         inode->i_fop = &proc_loginuid_operations;
1836                         break;
1837 #endif
1838                 default:
1839                         printk("procfs: impossible type (%d)",p->type);
1840                         iput(inode);
1841                         error = ERR_PTR(-EINVAL);
1842                         goto out;
1843         }
1844         dentry->d_op = &pid_dentry_operations;
1845         d_add(dentry, inode);
1846         /* Close the race of the process dying before we return the dentry */
1847         if (pid_revalidate(dentry, NULL))
1848                 error = NULL;
1849 out:
1850         put_task_struct(task);
1851 out_no_task:
1852         return error;
1853 }
1854
1855 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1856         return proc_pident_lookup(dir, dentry, tgid_base_stuff);
1857 }
1858
1859 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
1860         return proc_pident_lookup(dir, dentry, tid_base_stuff);
1861 }
1862
1863 static struct file_operations proc_tgid_base_operations = {
1864         .read           = generic_read_dir,
1865         .readdir        = proc_tgid_base_readdir,
1866 };
1867
1868 static struct file_operations proc_tid_base_operations = {
1869         .read           = generic_read_dir,
1870         .readdir        = proc_tid_base_readdir,
1871 };
1872
1873 static struct inode_operations proc_tgid_base_inode_operations = {
1874         .lookup         = proc_tgid_base_lookup,
1875         .getattr        = pid_getattr,
1876         .setattr        = proc_setattr,
1877 };
1878
1879 static struct inode_operations proc_tid_base_inode_operations = {
1880         .lookup         = proc_tid_base_lookup,
1881         .getattr        = pid_getattr,
1882         .setattr        = proc_setattr,
1883 };
1884
1885 #ifdef CONFIG_SECURITY
1886 static int proc_tgid_attr_readdir(struct file * filp,
1887                              void * dirent, filldir_t filldir)
1888 {
1889         return proc_pident_readdir(filp,dirent,filldir,
1890                                    tgid_attr_stuff,ARRAY_SIZE(tgid_attr_stuff));
1891 }
1892
1893 static int proc_tid_attr_readdir(struct file * filp,
1894                              void * dirent, filldir_t filldir)
1895 {
1896         return proc_pident_readdir(filp,dirent,filldir,
1897                                    tid_attr_stuff,ARRAY_SIZE(tid_attr_stuff));
1898 }
1899
1900 static struct file_operations proc_tgid_attr_operations = {
1901         .read           = generic_read_dir,
1902         .readdir        = proc_tgid_attr_readdir,
1903 };
1904
1905 static struct file_operations proc_tid_attr_operations = {
1906         .read           = generic_read_dir,
1907         .readdir        = proc_tid_attr_readdir,
1908 };
1909
1910 static struct dentry *proc_tgid_attr_lookup(struct inode *dir,
1911                                 struct dentry *dentry, struct nameidata *nd)
1912 {
1913         return proc_pident_lookup(dir, dentry, tgid_attr_stuff);
1914 }
1915
1916 static struct dentry *proc_tid_attr_lookup(struct inode *dir,
1917                                 struct dentry *dentry, struct nameidata *nd)
1918 {
1919         return proc_pident_lookup(dir, dentry, tid_attr_stuff);
1920 }
1921
1922 static struct inode_operations proc_tgid_attr_inode_operations = {
1923         .lookup         = proc_tgid_attr_lookup,
1924         .getattr        = pid_getattr,
1925         .setattr        = proc_setattr,
1926 };
1927
1928 static struct inode_operations proc_tid_attr_inode_operations = {
1929         .lookup         = proc_tid_attr_lookup,
1930         .getattr        = pid_getattr,
1931         .setattr        = proc_setattr,
1932 };
1933 #endif
1934
1935 /*
1936  * /proc/self:
1937  */
1938 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
1939                               int buflen)
1940 {
1941         char tmp[PROC_NUMBUF];
1942         sprintf(tmp, "%d", current->tgid);
1943         return vfs_readlink(dentry,buffer,buflen,tmp);
1944 }
1945
1946 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
1947 {
1948         char tmp[PROC_NUMBUF];
1949         sprintf(tmp, "%d", current->tgid);
1950         return ERR_PTR(vfs_follow_link(nd,tmp));
1951 }
1952
1953 static struct inode_operations proc_self_inode_operations = {
1954         .readlink       = proc_self_readlink,
1955         .follow_link    = proc_self_follow_link,
1956 };
1957
1958 /**
1959  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
1960  *
1961  * @task: task that should be flushed.
1962  *
1963  * Looks in the dcache for
1964  * /proc/@pid
1965  * /proc/@tgid/task/@pid
1966  * if either directory is present flushes it and all of it'ts children
1967  * from the dcache.
1968  *
1969  * It is safe and reasonable to cache /proc entries for a task until
1970  * that task exits.  After that they just clog up the dcache with
1971  * useless entries, possibly causing useful dcache entries to be
1972  * flushed instead.  This routine is proved to flush those useless
1973  * dcache entries at process exit time.
1974  *
1975  * NOTE: This routine is just an optimization so it does not guarantee
1976  *       that no dcache entries will exist at process exit time it
1977  *       just makes it very unlikely that any will persist.
1978  */
1979 void proc_flush_task(struct task_struct *task)
1980 {
1981         struct dentry *dentry, *leader, *dir;
1982         char buf[PROC_NUMBUF];
1983         struct qstr name;
1984
1985         name.name = buf;
1986         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
1987         dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
1988         if (dentry) {
1989                 shrink_dcache_parent(dentry);
1990                 d_drop(dentry);
1991                 dput(dentry);
1992         }
1993
1994         if (thread_group_leader(task))
1995                 goto out;
1996
1997         name.name = buf;
1998         name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
1999         leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
2000         if (!leader)
2001                 goto out;
2002
2003         name.name = "task";
2004         name.len = strlen(name.name);
2005         dir = d_hash_and_lookup(leader, &name);
2006         if (!dir)
2007                 goto out_put_leader;
2008
2009         name.name = buf;
2010         name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
2011         dentry = d_hash_and_lookup(dir, &name);
2012         if (dentry) {
2013                 shrink_dcache_parent(dentry);
2014                 d_drop(dentry);
2015                 dput(dentry);
2016         }
2017
2018         dput(dir);
2019 out_put_leader:
2020         dput(leader);
2021 out:
2022         return;
2023 }
2024
2025 /* SMP-safe */
2026 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2027 {
2028         struct dentry *result = ERR_PTR(-ENOENT);
2029         struct task_struct *task;
2030         struct inode *inode;
2031         struct proc_inode *ei;
2032         unsigned tgid;
2033
2034         if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
2035                 inode = new_inode(dir->i_sb);
2036                 if (!inode)
2037                         return ERR_PTR(-ENOMEM);
2038                 ei = PROC_I(inode);
2039                 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2040                 inode->i_ino = fake_ino(0, PROC_TGID_INO);
2041                 ei->pde = NULL;
2042                 inode->i_mode = S_IFLNK|S_IRWXUGO;
2043                 inode->i_uid = inode->i_gid = 0;
2044                 inode->i_size = 64;
2045                 inode->i_op = &proc_self_inode_operations;
2046                 d_add(dentry, inode);
2047                 return NULL;
2048         }
2049         tgid = name_to_int(dentry);
2050         if (tgid == ~0U)
2051                 goto out;
2052
2053         rcu_read_lock();
2054         task = find_task_by_pid(tgid);
2055         if (task)
2056                 get_task_struct(task);
2057         rcu_read_unlock();
2058         if (!task)
2059                 goto out;
2060
2061         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TGID_INO);
2062         if (!inode)
2063                 goto out_put_task;
2064
2065         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2066         inode->i_op = &proc_tgid_base_inode_operations;
2067         inode->i_fop = &proc_tgid_base_operations;
2068         inode->i_flags|=S_IMMUTABLE;
2069 #ifdef CONFIG_SECURITY
2070         inode->i_nlink = 5;
2071 #else
2072         inode->i_nlink = 4;
2073 #endif
2074
2075         dentry->d_op = &pid_dentry_operations;
2076
2077         d_add(dentry, inode);
2078         /* Close the race of the process dying before we return the dentry */
2079         if (pid_revalidate(dentry, NULL))
2080                 result = NULL;
2081
2082 out_put_task:
2083         put_task_struct(task);
2084 out:
2085         return result;
2086 }
2087
2088 /* SMP-safe */
2089 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2090 {
2091         struct dentry *result = ERR_PTR(-ENOENT);
2092         struct task_struct *task;
2093         struct task_struct *leader = get_proc_task(dir);
2094         struct inode *inode;
2095         unsigned tid;
2096
2097         if (!leader)
2098                 goto out_no_task;
2099
2100         tid = name_to_int(dentry);
2101         if (tid == ~0U)
2102                 goto out;
2103
2104         rcu_read_lock();
2105         task = find_task_by_pid(tid);
2106         if (task)
2107                 get_task_struct(task);
2108         rcu_read_unlock();
2109         if (!task)
2110                 goto out;
2111         if (leader->tgid != task->tgid)
2112                 goto out_drop_task;
2113
2114         inode = proc_pid_make_inode(dir->i_sb, task, PROC_TID_INO);
2115
2116
2117         if (!inode)
2118                 goto out_drop_task;
2119         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2120         inode->i_op = &proc_tid_base_inode_operations;
2121         inode->i_fop = &proc_tid_base_operations;
2122         inode->i_flags|=S_IMMUTABLE;
2123 #ifdef CONFIG_SECURITY
2124         inode->i_nlink = 4;
2125 #else
2126         inode->i_nlink = 3;
2127 #endif
2128
2129         dentry->d_op = &pid_dentry_operations;
2130
2131         d_add(dentry, inode);
2132         /* Close the race of the process dying before we return the dentry */
2133         if (pid_revalidate(dentry, NULL))
2134                 result = NULL;
2135
2136 out_drop_task:
2137         put_task_struct(task);
2138 out:
2139         put_task_struct(leader);
2140 out_no_task:
2141         return result;
2142 }
2143
2144 /*
2145  * Find the first tgid to return to user space.
2146  *
2147  * Usually this is just whatever follows &init_task, but if the users
2148  * buffer was too small to hold the full list or there was a seek into
2149  * the middle of the directory we have more work to do.
2150  *
2151  * In the case of a short read we start with find_task_by_pid.
2152  *
2153  * In the case of a seek we start with &init_task and walk nr
2154  * threads past it.
2155  */
2156 static struct task_struct *first_tgid(int tgid, unsigned int nr)
2157 {
2158         struct task_struct *pos;
2159         rcu_read_lock();
2160         if (tgid && nr) {
2161                 pos = find_task_by_pid(tgid);
2162                 if (pos && thread_group_leader(pos))
2163                         goto found;
2164         }
2165         /* If nr exceeds the number of processes get out quickly */
2166         pos = NULL;
2167         if (nr && nr >= nr_processes())
2168                 goto done;
2169
2170         /* If we haven't found our starting place yet start with
2171          * the init_task and walk nr tasks forward.
2172          */
2173         for (pos = next_task(&init_task); nr > 0; --nr) {
2174                 pos = next_task(pos);
2175                 if (pos == &init_task) {
2176                         pos = NULL;
2177                         goto done;
2178                 }
2179         }
2180 found:
2181         get_task_struct(pos);
2182 done:
2183         rcu_read_unlock();
2184         return pos;
2185 }
2186
2187 /*
2188  * Find the next task in the task list.
2189  * Return NULL if we loop or there is any error.
2190  *
2191  * The reference to the input task_struct is released.
2192  */
2193 static struct task_struct *next_tgid(struct task_struct *start)
2194 {
2195         struct task_struct *pos;
2196         rcu_read_lock();
2197         pos = start;
2198         if (pid_alive(start))
2199                 pos = next_task(start);
2200         if (pid_alive(pos) && (pos != &init_task)) {
2201                 get_task_struct(pos);
2202                 goto done;
2203         }
2204         pos = NULL;
2205 done:
2206         rcu_read_unlock();
2207         put_task_struct(start);
2208         return pos;
2209 }
2210
2211 /* for the /proc/ directory itself, after non-process stuff has been done */
2212 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2213 {
2214         char buf[PROC_NUMBUF];
2215         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2216         struct task_struct *task;
2217         int tgid;
2218
2219         if (!nr) {
2220                 ino_t ino = fake_ino(0,PROC_TGID_INO);
2221                 if (filldir(dirent, "self", 4, filp->f_pos, ino, DT_LNK) < 0)
2222                         return 0;
2223                 filp->f_pos++;
2224                 nr++;
2225         }
2226         nr -= 1;
2227
2228         /* f_version caches the tgid value that the last readdir call couldn't
2229          * return. lseek aka telldir automagically resets f_version to 0.
2230          */
2231         tgid = filp->f_version;
2232         filp->f_version = 0;
2233         for (task = first_tgid(tgid, nr);
2234              task;
2235              task = next_tgid(task), filp->f_pos++) {
2236                 int len;
2237                 ino_t ino;
2238                 tgid = task->pid;
2239                 len = snprintf(buf, sizeof(buf), "%d", tgid);
2240                 ino = fake_ino(tgid, PROC_TGID_INO);
2241                 if (filldir(dirent, buf, len, filp->f_pos, ino, DT_DIR) < 0) {
2242                         /* returning this tgid failed, save it as the first
2243                          * pid for the next readir call */
2244                         filp->f_version = tgid;
2245                         put_task_struct(task);
2246                         break;
2247                 }
2248         }
2249         return 0;
2250 }
2251
2252 /*
2253  * Find the first tid of a thread group to return to user space.
2254  *
2255  * Usually this is just the thread group leader, but if the users
2256  * buffer was too small or there was a seek into the middle of the
2257  * directory we have more work todo.
2258  *
2259  * In the case of a short read we start with find_task_by_pid.
2260  *
2261  * In the case of a seek we start with the leader and walk nr
2262  * threads past it.
2263  */
2264 static struct task_struct *first_tid(struct task_struct *leader,
2265                                         int tid, int nr)
2266 {
2267         struct task_struct *pos;
2268
2269         rcu_read_lock();
2270         /* Attempt to start with the pid of a thread */
2271         if (tid && (nr > 0)) {
2272                 pos = find_task_by_pid(tid);
2273                 if (pos && (pos->group_leader == leader))
2274                         goto found;
2275         }
2276
2277         /* If nr exceeds the number of threads there is nothing todo */
2278         pos = NULL;
2279         if (nr && nr >= get_nr_threads(leader))
2280                 goto out;
2281
2282         /* If we haven't found our starting place yet start
2283          * with the leader and walk nr threads forward.
2284          */
2285         for (pos = leader; nr > 0; --nr) {
2286                 pos = next_thread(pos);
2287                 if (pos == leader) {
2288                         pos = NULL;
2289                         goto out;
2290                 }
2291         }
2292 found:
2293         get_task_struct(pos);
2294 out:
2295         rcu_read_unlock();
2296         return pos;
2297 }
2298
2299 /*
2300  * Find the next thread in the thread list.
2301  * Return NULL if there is an error or no next thread.
2302  *
2303  * The reference to the input task_struct is released.
2304  */
2305 static struct task_struct *next_tid(struct task_struct *start)
2306 {
2307         struct task_struct *pos = NULL;
2308         rcu_read_lock();
2309         if (pid_alive(start)) {
2310                 pos = next_thread(start);
2311                 if (thread_group_leader(pos))
2312                         pos = NULL;
2313                 else
2314                         get_task_struct(pos);
2315         }
2316         rcu_read_unlock();
2317         put_task_struct(start);
2318         return pos;
2319 }
2320
2321 /* for the /proc/TGID/task/ directories */
2322 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2323 {
2324         char buf[PROC_NUMBUF];
2325         struct dentry *dentry = filp->f_dentry;
2326         struct inode *inode = dentry->d_inode;
2327         struct task_struct *leader = get_proc_task(inode);
2328         struct task_struct *task;
2329         int retval = -ENOENT;
2330         ino_t ino;
2331         int tid;
2332         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2333
2334         if (!leader)
2335                 goto out_no_task;
2336         retval = 0;
2337
2338         switch (pos) {
2339         case 0:
2340                 ino = inode->i_ino;
2341                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2342                         goto out;
2343                 pos++;
2344                 /* fall through */
2345         case 1:
2346                 ino = parent_ino(dentry);
2347                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2348                         goto out;
2349                 pos++;
2350                 /* fall through */
2351         }
2352
2353         /* f_version caches the tgid value that the last readdir call couldn't
2354          * return. lseek aka telldir automagically resets f_version to 0.
2355          */
2356         tid = filp->f_version;
2357         filp->f_version = 0;
2358         for (task = first_tid(leader, tid, pos - 2);
2359              task;
2360              task = next_tid(task), pos++) {
2361                 int len;
2362                 tid = task->pid;
2363                 len = snprintf(buf, sizeof(buf), "%d", tid);
2364                 ino = fake_ino(tid, PROC_TID_INO);
2365                 if (filldir(dirent, buf, len, pos, ino, DT_DIR < 0)) {
2366                         /* returning this tgid failed, save it as the first
2367                          * pid for the next readir call */
2368                         filp->f_version = tid;
2369                         put_task_struct(task);
2370                         break;
2371                 }
2372         }
2373 out:
2374         filp->f_pos = pos;
2375         put_task_struct(leader);
2376 out_no_task:
2377         return retval;
2378 }
2379
2380 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2381 {
2382         struct inode *inode = dentry->d_inode;
2383         struct task_struct *p = get_proc_task(inode);
2384         generic_fillattr(inode, stat);
2385
2386         if (p) {
2387                 rcu_read_lock();
2388                 stat->nlink += get_nr_threads(p);
2389                 rcu_read_unlock();
2390                 put_task_struct(p);
2391         }
2392
2393         return 0;
2394 }