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/mnt_namespace.h>
  63 #include <linux/mm.h>
  64 #include <linux/rcupdate.h>
  65 #include <linux/kallsyms.h>
  66 #include <linux/resource.h>
  67 #include <linux/module.h>
  68 #include <linux/mount.h>
  69 #include <linux/security.h>
  70 #include <linux/ptrace.h>
  71 #include <linux/cgroup.h>
  72 #include <linux/cpuset.h>
  73 #include <linux/audit.h>
  74 #include <linux/poll.h>
  75 #include <linux/nsproxy.h>
  76 #include <linux/oom.h>
  77 #include <linux/elf.h>
  78 #include <linux/pid_namespace.h>
  79 #include "internal.h"
  80
  81 /* NOTE:
  82  *      Implementing inode permission operations in /proc is almost
  83  *      certainly an error.  Permission checks need to happen during
  84  *      each system call not at open time.  The reason is that most of
  85  *      what we wish to check for permissions in /proc varies at runtime.
  86  *
  87  *      The classic example of a problem is opening file descriptors
  88  *      in /proc for a task before it execs a suid executable.
  89  */
  90
  91 struct pid_entry {
  92         char *name;
  93         int len;
  94         mode_t mode;
  95         const struct inode_operations *iop;
  96         const struct file_operations *fop;
  97         union proc_op op;
  98 };
  99
 100 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 101         .name = (NAME),                                 \
 102         .len  = sizeof(NAME) - 1,                       \
 103         .mode = MODE,                                   \
 104         .iop  = IOP,                                    \
 105         .fop  = FOP,                                    \
 106         .op   = OP,                                     \
 107 }
 108
 109 #define DIR(NAME, MODE, OTYPE)                                                  \
 110         NOD(NAME, (S_IFDIR|(MODE)),                                             \
 111                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
 112                 {} )
 113 #define LNK(NAME, OTYPE)                                        \
 114         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 115                 &proc_pid_link_inode_operations, NULL,          \
 116                 { .proc_get_link = &proc_##OTYPE##_link } )
 117 #define REG(NAME, MODE, OTYPE)                          \
 118         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
 119                 &proc_##OTYPE##_operations, {})
 120 #define INF(NAME, MODE, OTYPE)                          \
 121         NOD(NAME, (S_IFREG|(MODE)),                     \
 122                 NULL, &proc_info_file_operations,       \
 123                 { .proc_read = &proc_##OTYPE } )
 124 #define ONE(NAME, MODE, OTYPE)                          \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_single_file_operations,     \
 127                 { .proc_show = &proc_##OTYPE } )
 128
 129 int maps_protect;
 130 EXPORT_SYMBOL(maps_protect);
 131
 132 static struct fs_struct *get_fs_struct(struct task_struct *task)
 133 {
 134         struct fs_struct *fs;
 135         task_lock(task);
 136         fs = task->fs;
 137         if(fs)
 138                 atomic_inc(&fs->count);
 139         task_unlock(task);
 140         return fs;
 141 }
 142
 143 static int get_nr_threads(struct task_struct *tsk)
 144 {
 145         /* Must be called with the rcu_read_lock held */
 146         unsigned long flags;
 147         int count = 0;
 148
 149         if (lock_task_sighand(tsk, &flags)) {
 150                 count = atomic_read(&tsk->signal->count);
 151                 unlock_task_sighand(tsk, &flags);
 152         }
 153         return count;
 154 }
 155
 156 static int proc_cwd_link(struct inode *inode, struct path *path)
 157 {
 158         struct task_struct *task = get_proc_task(inode);
 159         struct fs_struct *fs = NULL;
 160         int result = -ENOENT;
 161
 162         if (task) {
 163                 fs = get_fs_struct(task);
 164                 put_task_struct(task);
 165         }
 166         if (fs) {
 167                 read_lock(&fs->lock);
 168                 *path = fs->pwd;
 169                 path_get(&fs->pwd);
 170                 read_unlock(&fs->lock);
 171                 result = 0;
 172                 put_fs_struct(fs);
 173         }
 174         return result;
 175 }
 176
 177 static int proc_root_link(struct inode *inode, struct path *path)
 178 {
 179         struct task_struct *task = get_proc_task(inode);
 180         struct fs_struct *fs = NULL;
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 fs = get_fs_struct(task);
 185                 put_task_struct(task);
 186         }
 187         if (fs) {
 188                 read_lock(&fs->lock);
 189                 *path = fs->root;
 190                 path_get(&fs->root);
 191                 read_unlock(&fs->lock);
 192                 result = 0;
 193                 put_fs_struct(fs);
 194         }
 195         return result;
 196 }
 197
 198 #define MAY_PTRACE(task) \
 199         (task == current || \
 200         (task->parent == current && \
 201         (task->ptrace & PT_PTRACED) && \
 202          (task_is_stopped_or_traced(task)) && \
 203          security_ptrace(current,task) == 0))
 204
 205 struct mm_struct *mm_for_maps(struct task_struct *task)
 206 {
 207         struct mm_struct *mm = get_task_mm(task);
 208         if (!mm)
 209                 return NULL;
 210         down_read(&mm->mmap_sem);
 211         task_lock(task);
 212         if (task->mm != mm)
 213                 goto out;
 214         if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
 215                 goto out;
 216         task_unlock(task);
 217         return mm;
 218 out:
 219         task_unlock(task);
 220         up_read(&mm->mmap_sem);
 221         mmput(mm);
 222         return NULL;
 223 }
 224
 225 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 226 {
 227         int res = 0;
 228         unsigned int len;
 229         struct mm_struct *mm = get_task_mm(task);
 230         if (!mm)
 231                 goto out;
 232         if (!mm->arg_end)
 233                 goto out_mm;    /* Shh! No looking before we're done */
 234
 235         len = mm->arg_end - mm->arg_start;
 236
 237         if (len > PAGE_SIZE)
 238                 len = PAGE_SIZE;
 239
 240         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 241
 242         // If the nul at the end of args has been overwritten, then
 243         // assume application is using setproctitle(3).
 244         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 245                 len = strnlen(buffer, res);
 246                 if (len < res) {
 247                     res = len;
 248                 } else {
 249                         len = mm->env_end - mm->env_start;
 250                         if (len > PAGE_SIZE - res)
 251                                 len = PAGE_SIZE - res;
 252                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 253                         res = strnlen(buffer, res);
 254                 }
 255         }
 256 out_mm:
 257         mmput(mm);
 258 out:
 259         return res;
 260 }
 261
 262 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 263 {
 264         int res = 0;
 265         struct mm_struct *mm = get_task_mm(task);
 266         if (mm) {
 267                 unsigned int nwords = 0;
 268                 do
 269                         nwords += 2;
 270                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 271                 res = nwords * sizeof(mm->saved_auxv[0]);
 272                 if (res > PAGE_SIZE)
 273                         res = PAGE_SIZE;
 274                 memcpy(buffer, mm->saved_auxv, res);
 275                 mmput(mm);
 276         }
 277         return res;
 278 }
 279
 280
 281 #ifdef CONFIG_KALLSYMS
 282 /*
 283  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 284  * Returns the resolved symbol.  If that fails, simply return the address.
 285  */
 286 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 287 {
 288         unsigned long wchan;
 289         char symname[KSYM_NAME_LEN];
 290
 291         wchan = get_wchan(task);
 292
 293         if (lookup_symbol_name(wchan, symname) < 0)
 294                 return sprintf(buffer, "%lu", wchan);
 295         else
 296                 return sprintf(buffer, "%s", symname);
 297 }
 298 #endif /* CONFIG_KALLSYMS */
 299
 300 #ifdef CONFIG_SCHEDSTATS
 301 /*
 302  * Provides /proc/PID/schedstat
 303  */
 304 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 305 {
 306         return sprintf(buffer, "%llu %llu %lu\n",
 307                         task->sched_info.cpu_time,
 308                         task->sched_info.run_delay,
 309                         task->sched_info.pcount);
 310 }
 311 #endif
 312
 313 #ifdef CONFIG_LATENCYTOP
 314 static int lstats_show_proc(struct seq_file *m, void *v)
 315 {
 316         int i;
 317         struct task_struct *task = m->private;
 318         seq_puts(m, "Latency Top version : v0.1\n");
 319
 320         for (i = 0; i < 32; i++) {
 321                 if (task->latency_record[i].backtrace[0]) {
 322                         int q;
 323                         seq_printf(m, "%i %li %li ",
 324                                 task->latency_record[i].count,
 325                                 task->latency_record[i].time,
 326                                 task->latency_record[i].max);
 327                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 328                                 char sym[KSYM_NAME_LEN];
 329                                 char *c;
 330                                 if (!task->latency_record[i].backtrace[q])
 331                                         break;
 332                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 333                                         break;
 334                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 335                                 c = strchr(sym, '+');
 336                                 if (c)
 337                                         *c = 0;
 338                                 seq_printf(m, "%s ", sym);
 339                         }
 340                         seq_printf(m, "\n");
 341                 }
 342
 343         }
 344         return 0;
 345 }
 346
 347 static int lstats_open(struct inode *inode, struct file *file)
 348 {
 349         int ret;
 350         struct seq_file *m;
 351         struct task_struct *task = get_proc_task(inode);
 352
 353         ret = single_open(file, lstats_show_proc, NULL);
 354         if (!ret) {
 355                 m = file->private_data;
 356                 m->private = task;
 357         }
 358         return ret;
 359 }
 360
 361 static ssize_t lstats_write(struct file *file, const char __user *buf,
 362                             size_t count, loff_t *offs)
 363 {
 364         struct seq_file *m;
 365         struct task_struct *task;
 366
 367         m = file->private_data;
 368         task = m->private;
 369         clear_all_latency_tracing(task);
 370
 371         return count;
 372 }
 373
 374 static const struct file_operations proc_lstats_operations = {
 375         .open           = lstats_open,
 376         .read           = seq_read,
 377         .write          = lstats_write,
 378         .llseek         = seq_lseek,
 379         .release        = single_release,
 380 };
 381
 382 #endif
 383
 384 /* The badness from the OOM killer */
 385 unsigned long badness(struct task_struct *p, unsigned long uptime);
 386 static int proc_oom_score(struct task_struct *task, char *buffer)
 387 {
 388         unsigned long points;
 389         struct timespec uptime;
 390
 391         do_posix_clock_monotonic_gettime(&uptime);
 392         read_lock(&tasklist_lock);
 393         points = badness(task, uptime.tv_sec);
 394         read_unlock(&tasklist_lock);
 395         return sprintf(buffer, "%lu\n", points);
 396 }
 397
 398 struct limit_names {
 399         char *name;
 400         char *unit;
 401 };
 402
 403 static const struct limit_names lnames[RLIM_NLIMITS] = {
 404         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 405         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 406         [RLIMIT_DATA] = {"Max data size", "bytes"},
 407         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 408         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 409         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 410         [RLIMIT_NPROC] = {"Max processes", "processes"},
 411         [RLIMIT_NOFILE] = {"Max open files", "files"},
 412         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 413         [RLIMIT_AS] = {"Max address space", "bytes"},
 414         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 415         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 416         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 417         [RLIMIT_NICE] = {"Max nice priority", NULL},
 418         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 419         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 420 };
 421
 422 /* Display limits for a process */
 423 static int proc_pid_limits(struct task_struct *task, char *buffer)
 424 {
 425         unsigned int i;
 426         int count = 0;
 427         unsigned long flags;
 428         char *bufptr = buffer;
 429
 430         struct rlimit rlim[RLIM_NLIMITS];
 431
 432         rcu_read_lock();
 433         if (!lock_task_sighand(task,&flags)) {
 434                 rcu_read_unlock();
 435                 return 0;
 436         }
 437         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 438         unlock_task_sighand(task, &flags);
 439         rcu_read_unlock();
 440
 441         /*
 442          * print the file header
 443          */
 444         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 445                         "Limit", "Soft Limit", "Hard Limit", "Units");
 446
 447         for (i = 0; i < RLIM_NLIMITS; i++) {
 448                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 449                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 450                                          lnames[i].name, "unlimited");
 451                 else
 452                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 453                                          lnames[i].name, rlim[i].rlim_cur);
 454
 455                 if (rlim[i].rlim_max == RLIM_INFINITY)
 456                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 457                 else
 458                         count += sprintf(&bufptr[count], "%-20lu ",
 459                                          rlim[i].rlim_max);
 460
 461                 if (lnames[i].unit)
 462                         count += sprintf(&bufptr[count], "%-10s\n",
 463                                          lnames[i].unit);
 464                 else
 465                         count += sprintf(&bufptr[count], "\n");
 466         }
 467
 468         return count;
 469 }
 470
 471 /************************************************************************/
 472 /*                       Here the fs part begins                        */
 473 /************************************************************************/
 474
 475 /* permission checks */
 476 static int proc_fd_access_allowed(struct inode *inode)
 477 {
 478         struct task_struct *task;
 479         int allowed = 0;
 480         /* Allow access to a task's file descriptors if it is us or we
 481          * may use ptrace attach to the process and find out that
 482          * information.
 483          */
 484         task = get_proc_task(inode);
 485         if (task) {
 486                 allowed = ptrace_may_attach(task);
 487                 put_task_struct(task);
 488         }
 489         return allowed;
 490 }
 491
 492 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 493 {
 494         int error;
 495         struct inode *inode = dentry->d_inode;
 496
 497         if (attr->ia_valid & ATTR_MODE)
 498                 return -EPERM;
 499
 500         error = inode_change_ok(inode, attr);
 501         if (!error)
 502                 error = inode_setattr(inode, attr);
 503         return error;
 504 }
 505
 506 static const struct inode_operations proc_def_inode_operations = {
 507         .setattr        = proc_setattr,
 508 };
 509
 510 extern const struct seq_operations mounts_op;
 511 struct proc_mounts {
 512         struct seq_file m;
 513         int event;
 514 };
 515
 516 static int mounts_open(struct inode *inode, struct file *file)
 517 {
 518         struct task_struct *task = get_proc_task(inode);
 519         struct nsproxy *nsp;
 520         struct mnt_namespace *ns = NULL;
 521         struct proc_mounts *p;
 522         int ret = -EINVAL;
 523
 524         if (task) {
 525                 rcu_read_lock();
 526                 nsp = task_nsproxy(task);
 527                 if (nsp) {
 528                         ns = nsp->mnt_ns;
 529                         if (ns)
 530                                 get_mnt_ns(ns);
 531                 }
 532                 rcu_read_unlock();
 533
 534                 put_task_struct(task);
 535         }
 536
 537         if (ns) {
 538                 ret = -ENOMEM;
 539                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 540                 if (p) {
 541                         file->private_data = &p->m;
 542                         ret = seq_open(file, &mounts_op);
 543                         if (!ret) {
 544                                 p->m.private = ns;
 545                                 p->event = ns->event;
 546                                 return 0;
 547                         }
 548                         kfree(p);
 549                 }
 550                 put_mnt_ns(ns);
 551         }
 552         return ret;
 553 }
 554
 555 static int mounts_release(struct inode *inode, struct file *file)
 556 {
 557         struct seq_file *m = file->private_data;
 558         struct mnt_namespace *ns = m->private;
 559         put_mnt_ns(ns);
 560         return seq_release(inode, file);
 561 }
 562
 563 static unsigned mounts_poll(struct file *file, poll_table *wait)
 564 {
 565         struct proc_mounts *p = file->private_data;
 566         struct mnt_namespace *ns = p->m.private;
 567         unsigned res = 0;
 568
 569         poll_wait(file, &ns->poll, wait);
 570
 571         spin_lock(&vfsmount_lock);
 572         if (p->event != ns->event) {
 573                 p->event = ns->event;
 574                 res = POLLERR;
 575         }
 576         spin_unlock(&vfsmount_lock);
 577
 578         return res;
 579 }
 580
 581 static const struct file_operations proc_mounts_operations = {
 582         .open           = mounts_open,
 583         .read           = seq_read,
 584         .llseek         = seq_lseek,
 585         .release        = mounts_release,
 586         .poll           = mounts_poll,
 587 };
 588
 589 extern const struct seq_operations mountstats_op;
 590 static int mountstats_open(struct inode *inode, struct file *file)
 591 {
 592         int ret = seq_open(file, &mountstats_op);
 593
 594         if (!ret) {
 595                 struct seq_file *m = file->private_data;
 596                 struct nsproxy *nsp;
 597                 struct mnt_namespace *mnt_ns = NULL;
 598                 struct task_struct *task = get_proc_task(inode);
 599
 600                 if (task) {
 601                         rcu_read_lock();
 602                         nsp = task_nsproxy(task);
 603                         if (nsp) {
 604                                 mnt_ns = nsp->mnt_ns;
 605                                 if (mnt_ns)
 606                                         get_mnt_ns(mnt_ns);
 607                         }
 608                         rcu_read_unlock();
 609
 610                         put_task_struct(task);
 611                 }
 612
 613                 if (mnt_ns)
 614                         m->private = mnt_ns;
 615                 else {
 616                         seq_release(inode, file);
 617                         ret = -EINVAL;
 618                 }
 619         }
 620         return ret;
 621 }
 622
 623 static const struct file_operations proc_mountstats_operations = {
 624         .open           = mountstats_open,
 625         .read           = seq_read,
 626         .llseek         = seq_lseek,
 627         .release        = mounts_release,
 628 };
 629
 630 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 631
 632 static ssize_t proc_info_read(struct file * file, char __user * buf,
 633                           size_t count, loff_t *ppos)
 634 {
 635         struct inode * inode = file->f_path.dentry->d_inode;
 636         unsigned long page;
 637         ssize_t length;
 638         struct task_struct *task = get_proc_task(inode);
 639
 640         length = -ESRCH;
 641         if (!task)
 642                 goto out_no_task;
 643
 644         if (count > PROC_BLOCK_SIZE)
 645                 count = PROC_BLOCK_SIZE;
 646
 647         length = -ENOMEM;
 648         if (!(page = __get_free_page(GFP_TEMPORARY)))
 649                 goto out;
 650
 651         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 652
 653         if (length >= 0)
 654                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 655         free_page(page);
 656 out:
 657         put_task_struct(task);
 658 out_no_task:
 659         return length;
 660 }
 661
 662 static const struct file_operations proc_info_file_operations = {
 663         .read           = proc_info_read,
 664 };
 665
 666 static int proc_single_show(struct seq_file *m, void *v)
 667 {
 668         struct inode *inode = m->private;
 669         struct pid_namespace *ns;
 670         struct pid *pid;
 671         struct task_struct *task;
 672         int ret;
 673
 674         ns = inode->i_sb->s_fs_info;
 675         pid = proc_pid(inode);
 676         task = get_pid_task(pid, PIDTYPE_PID);
 677         if (!task)
 678                 return -ESRCH;
 679
 680         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 681
 682         put_task_struct(task);
 683         return ret;
 684 }
 685
 686 static int proc_single_open(struct inode *inode, struct file *filp)
 687 {
 688         int ret;
 689         ret = single_open(filp, proc_single_show, NULL);
 690         if (!ret) {
 691                 struct seq_file *m = filp->private_data;
 692
 693                 m->private = inode;
 694         }
 695         return ret;
 696 }
 697
 698 static const struct file_operations proc_single_file_operations = {
 699         .open           = proc_single_open,
 700         .read           = seq_read,
 701         .llseek         = seq_lseek,
 702         .release        = single_release,
 703 };
 704
 705 static int mem_open(struct inode* inode, struct file* file)
 706 {
 707         file->private_data = (void*)((long)current->self_exec_id);
 708         return 0;
 709 }
 710
 711 static ssize_t mem_read(struct file * file, char __user * buf,
 712                         size_t count, loff_t *ppos)
 713 {
 714         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 715         char *page;
 716         unsigned long src = *ppos;
 717         int ret = -ESRCH;
 718         struct mm_struct *mm;
 719
 720         if (!task)
 721                 goto out_no_task;
 722
 723         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 724                 goto out;
 725
 726         ret = -ENOMEM;
 727         page = (char *)__get_free_page(GFP_TEMPORARY);
 728         if (!page)
 729                 goto out;
 730
 731         ret = 0;
 732
 733         mm = get_task_mm(task);
 734         if (!mm)
 735                 goto out_free;
 736
 737         ret = -EIO;
 738
 739         if (file->private_data != (void*)((long)current->self_exec_id))
 740                 goto out_put;
 741
 742         ret = 0;
 743
 744         while (count > 0) {
 745                 int this_len, retval;
 746
 747                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 748                 retval = access_process_vm(task, src, page, this_len, 0);
 749                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 750                         if (!ret)
 751                                 ret = -EIO;
 752                         break;
 753                 }
 754
 755                 if (copy_to_user(buf, page, retval)) {
 756                         ret = -EFAULT;
 757                         break;
 758                 }
 759
 760                 ret += retval;
 761                 src += retval;
 762                 buf += retval;
 763                 count -= retval;
 764         }
 765         *ppos = src;
 766
 767 out_put:
 768         mmput(mm);
 769 out_free:
 770         free_page((unsigned long) page);
 771 out:
 772         put_task_struct(task);
 773 out_no_task:
 774         return ret;
 775 }
 776
 777 #define mem_write NULL
 778
 779 #ifndef mem_write
 780 /* This is a security hazard */
 781 static ssize_t mem_write(struct file * file, const char __user *buf,
 782                          size_t count, loff_t *ppos)
 783 {
 784         int copied;
 785         char *page;
 786         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 787         unsigned long dst = *ppos;
 788
 789         copied = -ESRCH;
 790         if (!task)
 791                 goto out_no_task;
 792
 793         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 794                 goto out;
 795
 796         copied = -ENOMEM;
 797         page = (char *)__get_free_page(GFP_TEMPORARY);
 798         if (!page)
 799                 goto out;
 800
 801         copied = 0;
 802         while (count > 0) {
 803                 int this_len, retval;
 804
 805                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 806                 if (copy_from_user(page, buf, this_len)) {
 807                         copied = -EFAULT;
 808                         break;
 809                 }
 810                 retval = access_process_vm(task, dst, page, this_len, 1);
 811                 if (!retval) {
 812                         if (!copied)
 813                                 copied = -EIO;
 814                         break;
 815                 }
 816                 copied += retval;
 817                 buf += retval;
 818                 dst += retval;
 819                 count -= retval;
 820         }
 821         *ppos = dst;
 822         free_page((unsigned long) page);
 823 out:
 824         put_task_struct(task);
 825 out_no_task:
 826         return copied;
 827 }
 828 #endif
 829
 830 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 831 {
 832         switch (orig) {
 833         case 0:
 834                 file->f_pos = offset;
 835                 break;
 836         case 1:
 837                 file->f_pos += offset;
 838                 break;
 839         default:
 840                 return -EINVAL;
 841         }
 842         force_successful_syscall_return();
 843         return file->f_pos;
 844 }
 845
 846 static const struct file_operations proc_mem_operations = {
 847         .llseek         = mem_lseek,
 848         .read           = mem_read,
 849         .write          = mem_write,
 850         .open           = mem_open,
 851 };
 852
 853 static ssize_t environ_read(struct file *file, char __user *buf,
 854                         size_t count, loff_t *ppos)
 855 {
 856         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 857         char *page;
 858         unsigned long src = *ppos;
 859         int ret = -ESRCH;
 860         struct mm_struct *mm;
 861
 862         if (!task)
 863                 goto out_no_task;
 864
 865         if (!ptrace_may_attach(task))
 866                 goto out;
 867
 868         ret = -ENOMEM;
 869         page = (char *)__get_free_page(GFP_TEMPORARY);
 870         if (!page)
 871                 goto out;
 872
 873         ret = 0;
 874
 875         mm = get_task_mm(task);
 876         if (!mm)
 877                 goto out_free;
 878
 879         while (count > 0) {
 880                 int this_len, retval, max_len;
 881
 882                 this_len = mm->env_end - (mm->env_start + src);
 883
 884                 if (this_len <= 0)
 885                         break;
 886
 887                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 888                 this_len = (this_len > max_len) ? max_len : this_len;
 889
 890                 retval = access_process_vm(task, (mm->env_start + src),
 891                         page, this_len, 0);
 892
 893                 if (retval <= 0) {
 894                         ret = retval;
 895                         break;
 896                 }
 897
 898                 if (copy_to_user(buf, page, retval)) {
 899                         ret = -EFAULT;
 900                         break;
 901                 }
 902
 903                 ret += retval;
 904                 src += retval;
 905                 buf += retval;
 906                 count -= retval;
 907         }
 908         *ppos = src;
 909
 910         mmput(mm);
 911 out_free:
 912         free_page((unsigned long) page);
 913 out:
 914         put_task_struct(task);
 915 out_no_task:
 916         return ret;
 917 }
 918
 919 static const struct file_operations proc_environ_operations = {
 920         .read           = environ_read,
 921 };
 922
 923 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 924                                 size_t count, loff_t *ppos)
 925 {
 926         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 927         char buffer[PROC_NUMBUF];
 928         size_t len;
 929         int oom_adjust;
 930
 931         if (!task)
 932                 return -ESRCH;
 933         oom_adjust = task->oomkilladj;
 934         put_task_struct(task);
 935
 936         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 937
 938         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 939 }
 940
 941 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 942                                 size_t count, loff_t *ppos)
 943 {
 944         struct task_struct *task;
 945         char buffer[PROC_NUMBUF], *end;
 946         int oom_adjust;
 947
 948         memset(buffer, 0, sizeof(buffer));
 949         if (count > sizeof(buffer) - 1)
 950                 count = sizeof(buffer) - 1;
 951         if (copy_from_user(buffer, buf, count))
 952                 return -EFAULT;
 953         oom_adjust = simple_strtol(buffer, &end, 0);
 954         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 955              oom_adjust != OOM_DISABLE)
 956                 return -EINVAL;
 957         if (*end == '\n')
 958                 end++;
 959         task = get_proc_task(file->f_path.dentry->d_inode);
 960         if (!task)
 961                 return -ESRCH;
 962         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
 963                 put_task_struct(task);
 964                 return -EACCES;
 965         }
 966         task->oomkilladj = oom_adjust;
 967         put_task_struct(task);
 968         if (end - buffer == 0)
 969                 return -EIO;
 970         return end - buffer;
 971 }
 972
 973 static const struct file_operations proc_oom_adjust_operations = {
 974         .read           = oom_adjust_read,
 975         .write          = oom_adjust_write,
 976 };
 977
 978 #ifdef CONFIG_AUDITSYSCALL
 979 #define TMPBUFLEN 21
 980 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 981                                   size_t count, loff_t *ppos)
 982 {
 983         struct inode * inode = file->f_path.dentry->d_inode;
 984         struct task_struct *task = get_proc_task(inode);
 985         ssize_t length;
 986         char tmpbuf[TMPBUFLEN];
 987
 988         if (!task)
 989                 return -ESRCH;
 990         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 991                                 audit_get_loginuid(task));
 992         put_task_struct(task);
 993         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 994 }
 995
 996 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 997                                    size_t count, loff_t *ppos)
 998 {
 999         struct inode * inode = file->f_path.dentry->d_inode;
1000         char *page, *tmp;
1001         ssize_t length;
1002         uid_t loginuid;
1003
1004         if (!capable(CAP_AUDIT_CONTROL))
1005                 return -EPERM;
1006
1007         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1008                 return -EPERM;
1009
1010         if (count >= PAGE_SIZE)
1011                 count = PAGE_SIZE - 1;
1012
1013         if (*ppos != 0) {
1014                 /* No partial writes. */
1015                 return -EINVAL;
1016         }
1017         page = (char*)__get_free_page(GFP_TEMPORARY);
1018         if (!page)
1019                 return -ENOMEM;
1020         length = -EFAULT;
1021         if (copy_from_user(page, buf, count))
1022                 goto out_free_page;
1023
1024         page[count] = '\0';
1025         loginuid = simple_strtoul(page, &tmp, 10);
1026         if (tmp == page) {
1027                 length = -EINVAL;
1028                 goto out_free_page;
1029
1030         }
1031         length = audit_set_loginuid(current, loginuid);
1032         if (likely(length == 0))
1033                 length = count;
1034
1035 out_free_page:
1036         free_page((unsigned long) page);
1037         return length;
1038 }
1039
1040 static const struct file_operations proc_loginuid_operations = {
1041         .read           = proc_loginuid_read,
1042         .write          = proc_loginuid_write,
1043 };
1044 #endif
1045
1046 #ifdef CONFIG_FAULT_INJECTION
1047 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1048                                       size_t count, loff_t *ppos)
1049 {
1050         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1051         char buffer[PROC_NUMBUF];
1052         size_t len;
1053         int make_it_fail;
1054
1055         if (!task)
1056                 return -ESRCH;
1057         make_it_fail = task->make_it_fail;
1058         put_task_struct(task);
1059
1060         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1061
1062         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1063 }
1064
1065 static ssize_t proc_fault_inject_write(struct file * file,
1066                         const char __user * buf, size_t count, loff_t *ppos)
1067 {
1068         struct task_struct *task;
1069         char buffer[PROC_NUMBUF], *end;
1070         int make_it_fail;
1071
1072         if (!capable(CAP_SYS_RESOURCE))
1073                 return -EPERM;
1074         memset(buffer, 0, sizeof(buffer));
1075         if (count > sizeof(buffer) - 1)
1076                 count = sizeof(buffer) - 1;
1077         if (copy_from_user(buffer, buf, count))
1078                 return -EFAULT;
1079         make_it_fail = simple_strtol(buffer, &end, 0);
1080         if (*end == '\n')
1081                 end++;
1082         task = get_proc_task(file->f_dentry->d_inode);
1083         if (!task)
1084                 return -ESRCH;
1085         task->make_it_fail = make_it_fail;
1086         put_task_struct(task);
1087         if (end - buffer == 0)
1088                 return -EIO;
1089         return end - buffer;
1090 }
1091
1092 static const struct file_operations proc_fault_inject_operations = {
1093         .read           = proc_fault_inject_read,
1094         .write          = proc_fault_inject_write,
1095 };
1096 #endif
1097
1098
1099 #ifdef CONFIG_SCHED_DEBUG
1100 /*
1101  * Print out various scheduling related per-task fields:
1102  */
1103 static int sched_show(struct seq_file *m, void *v)
1104 {
1105         struct inode *inode = m->private;
1106         struct task_struct *p;
1107
1108         WARN_ON(!inode);
1109
1110         p = get_proc_task(inode);
1111         if (!p)
1112                 return -ESRCH;
1113         proc_sched_show_task(p, m);
1114
1115         put_task_struct(p);
1116
1117         return 0;
1118 }
1119
1120 static ssize_t
1121 sched_write(struct file *file, const char __user *buf,
1122             size_t count, loff_t *offset)
1123 {
1124         struct inode *inode = file->f_path.dentry->d_inode;
1125         struct task_struct *p;
1126
1127         WARN_ON(!inode);
1128
1129         p = get_proc_task(inode);
1130         if (!p)
1131                 return -ESRCH;
1132         proc_sched_set_task(p);
1133
1134         put_task_struct(p);
1135
1136         return count;
1137 }
1138
1139 static int sched_open(struct inode *inode, struct file *filp)
1140 {
1141         int ret;
1142
1143         ret = single_open(filp, sched_show, NULL);
1144         if (!ret) {
1145                 struct seq_file *m = filp->private_data;
1146
1147                 m->private = inode;
1148         }
1149         return ret;
1150 }
1151
1152 static const struct file_operations proc_pid_sched_operations = {
1153         .open           = sched_open,
1154         .read           = seq_read,
1155         .write          = sched_write,
1156         .llseek         = seq_lseek,
1157         .release        = single_release,
1158 };
1159
1160 #endif
1161
1162 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1163 {
1164         struct inode *inode = dentry->d_inode;
1165         int error = -EACCES;
1166
1167         /* We don't need a base pointer in the /proc filesystem */
1168         path_put(&nd->path);
1169
1170         /* Are we allowed to snoop on the tasks file descriptors? */
1171         if (!proc_fd_access_allowed(inode))
1172                 goto out;
1173
1174         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1175         nd->last_type = LAST_BIND;
1176 out:
1177         return ERR_PTR(error);
1178 }
1179
1180 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1181 {
1182         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1183         char *pathname;
1184         int len;
1185
1186         if (!tmp)
1187                 return -ENOMEM;
1188
1189         pathname = d_path(path, tmp, PAGE_SIZE);
1190         len = PTR_ERR(pathname);
1191         if (IS_ERR(pathname))
1192                 goto out;
1193         len = tmp + PAGE_SIZE - 1 - pathname;
1194
1195         if (len > buflen)
1196                 len = buflen;
1197         if (copy_to_user(buffer, pathname, len))
1198                 len = -EFAULT;
1199  out:
1200         free_page((unsigned long)tmp);
1201         return len;
1202 }
1203
1204 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1205 {
1206         int error = -EACCES;
1207         struct inode *inode = dentry->d_inode;
1208         struct path path;
1209
1210         /* Are we allowed to snoop on the tasks file descriptors? */
1211         if (!proc_fd_access_allowed(inode))
1212                 goto out;
1213
1214         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1215         if (error)
1216                 goto out;
1217
1218         error = do_proc_readlink(&path, buffer, buflen);
1219         path_put(&path);
1220 out:
1221         return error;
1222 }
1223
1224 static const struct inode_operations proc_pid_link_inode_operations = {
1225         .readlink       = proc_pid_readlink,
1226         .follow_link    = proc_pid_follow_link,
1227         .setattr        = proc_setattr,
1228 };
1229
1230
1231 /* building an inode */
1232
1233 static int task_dumpable(struct task_struct *task)
1234 {
1235         int dumpable = 0;
1236         struct mm_struct *mm;
1237
1238         task_lock(task);
1239         mm = task->mm;
1240         if (mm)
1241                 dumpable = get_dumpable(mm);
1242         task_unlock(task);
1243         if(dumpable == 1)
1244                 return 1;
1245         return 0;
1246 }
1247
1248
1249 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1250 {
1251         struct inode * inode;
1252         struct proc_inode *ei;
1253
1254         /* We need a new inode */
1255
1256         inode = new_inode(sb);
1257         if (!inode)
1258                 goto out;
1259
1260         /* Common stuff */
1261         ei = PROC_I(inode);
1262         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1263         inode->i_op = &proc_def_inode_operations;
1264
1265         /*
1266          * grab the reference to task.
1267          */
1268         ei->pid = get_task_pid(task, PIDTYPE_PID);
1269         if (!ei->pid)
1270                 goto out_unlock;
1271
1272         inode->i_uid = 0;
1273         inode->i_gid = 0;
1274         if (task_dumpable(task)) {
1275                 inode->i_uid = task->euid;
1276                 inode->i_gid = task->egid;
1277         }
1278         security_task_to_inode(task, inode);
1279
1280 out:
1281         return inode;
1282
1283 out_unlock:
1284         iput(inode);
1285         return NULL;
1286 }
1287
1288 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1289 {
1290         struct inode *inode = dentry->d_inode;
1291         struct task_struct *task;
1292         generic_fillattr(inode, stat);
1293
1294         rcu_read_lock();
1295         stat->uid = 0;
1296         stat->gid = 0;
1297         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1298         if (task) {
1299                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1300                     task_dumpable(task)) {
1301                         stat->uid = task->euid;
1302                         stat->gid = task->egid;
1303                 }
1304         }
1305         rcu_read_unlock();
1306         return 0;
1307 }
1308
1309 /* dentry stuff */
1310
1311 /*
1312  *      Exceptional case: normally we are not allowed to unhash a busy
1313  * directory. In this case, however, we can do it - no aliasing problems
1314  * due to the way we treat inodes.
1315  *
1316  * Rewrite the inode's ownerships here because the owning task may have
1317  * performed a setuid(), etc.
1318  *
1319  * Before the /proc/pid/status file was created the only way to read
1320  * the effective uid of a /process was to stat /proc/pid.  Reading
1321  * /proc/pid/status is slow enough that procps and other packages
1322  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1323  * made this apply to all per process world readable and executable
1324  * directories.
1325  */
1326 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1327 {
1328         struct inode *inode = dentry->d_inode;
1329         struct task_struct *task = get_proc_task(inode);
1330         if (task) {
1331                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1332                     task_dumpable(task)) {
1333                         inode->i_uid = task->euid;
1334                         inode->i_gid = task->egid;
1335                 } else {
1336                         inode->i_uid = 0;
1337                         inode->i_gid = 0;
1338                 }
1339                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1340                 security_task_to_inode(task, inode);
1341                 put_task_struct(task);
1342                 return 1;
1343         }
1344         d_drop(dentry);
1345         return 0;
1346 }
1347
1348 static int pid_delete_dentry(struct dentry * dentry)
1349 {
1350         /* Is the task we represent dead?
1351          * If so, then don't put the dentry on the lru list,
1352          * kill it immediately.
1353          */
1354         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1355 }
1356
1357 static struct dentry_operations pid_dentry_operations =
1358 {
1359         .d_revalidate   = pid_revalidate,
1360         .d_delete       = pid_delete_dentry,
1361 };
1362
1363 /* Lookups */
1364
1365 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1366                                 struct task_struct *, const void *);
1367
1368 /*
1369  * Fill a directory entry.
1370  *
1371  * If possible create the dcache entry and derive our inode number and
1372  * file type from dcache entry.
1373  *
1374  * Since all of the proc inode numbers are dynamically generated, the inode
1375  * numbers do not exist until the inode is cache.  This means creating the
1376  * the dcache entry in readdir is necessary to keep the inode numbers
1377  * reported by readdir in sync with the inode numbers reported
1378  * by stat.
1379  */
1380 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1381         char *name, int len,
1382         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1383 {
1384         struct dentry *child, *dir = filp->f_path.dentry;
1385         struct inode *inode;
1386         struct qstr qname;
1387         ino_t ino = 0;
1388         unsigned type = DT_UNKNOWN;
1389
1390         qname.name = name;
1391         qname.len  = len;
1392         qname.hash = full_name_hash(name, len);
1393
1394         child = d_lookup(dir, &qname);
1395         if (!child) {
1396                 struct dentry *new;
1397                 new = d_alloc(dir, &qname);
1398                 if (new) {
1399                         child = instantiate(dir->d_inode, new, task, ptr);
1400                         if (child)
1401                                 dput(new);
1402                         else
1403                                 child = new;
1404                 }
1405         }
1406         if (!child || IS_ERR(child) || !child->d_inode)
1407                 goto end_instantiate;
1408         inode = child->d_inode;
1409         if (inode) {
1410                 ino = inode->i_ino;
1411                 type = inode->i_mode >> 12;
1412         }
1413         dput(child);
1414 end_instantiate:
1415         if (!ino)
1416                 ino = find_inode_number(dir, &qname);
1417         if (!ino)
1418                 ino = 1;
1419         return filldir(dirent, name, len, filp->f_pos, ino, type);
1420 }
1421
1422 static unsigned name_to_int(struct dentry *dentry)
1423 {
1424         const char *name = dentry->d_name.name;
1425         int len = dentry->d_name.len;
1426         unsigned n = 0;
1427
1428         if (len > 1 && *name == '0')
1429                 goto out;
1430         while (len-- > 0) {
1431                 unsigned c = *name++ - '0';
1432                 if (c > 9)
1433                         goto out;
1434                 if (n >= (~0U-9)/10)
1435                         goto out;
1436                 n *= 10;
1437                 n += c;
1438         }
1439         return n;
1440 out:
1441         return ~0U;
1442 }
1443
1444 #define PROC_FDINFO_MAX 64
1445
1446 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1447 {
1448         struct task_struct *task = get_proc_task(inode);
1449         struct files_struct *files = NULL;
1450         struct file *file;
1451         int fd = proc_fd(inode);
1452
1453         if (task) {
1454                 files = get_files_struct(task);
1455                 put_task_struct(task);
1456         }
1457         if (files) {
1458                 /*
1459                  * We are not taking a ref to the file structure, so we must
1460                  * hold ->file_lock.
1461                  */
1462                 spin_lock(&files->file_lock);
1463                 file = fcheck_files(files, fd);
1464                 if (file) {
1465                         if (path) {
1466                                 *path = file->f_path;
1467                                 path_get(&file->f_path);
1468                         }
1469                         if (info)
1470                                 snprintf(info, PROC_FDINFO_MAX,
1471                                          "pos:\t%lli\n"
1472                                          "flags:\t0%o\n",
1473                                          (long long) file->f_pos,
1474                                          file->f_flags);
1475                         spin_unlock(&files->file_lock);
1476                         put_files_struct(files);
1477                         return 0;
1478                 }
1479                 spin_unlock(&files->file_lock);
1480                 put_files_struct(files);
1481         }
1482         return -ENOENT;
1483 }
1484
1485 static int proc_fd_link(struct inode *inode, struct path *path)
1486 {
1487         return proc_fd_info(inode, path, NULL);
1488 }
1489
1490 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1491 {
1492         struct inode *inode = dentry->d_inode;
1493         struct task_struct *task = get_proc_task(inode);
1494         int fd = proc_fd(inode);
1495         struct files_struct *files;
1496
1497         if (task) {
1498                 files = get_files_struct(task);
1499                 if (files) {
1500                         rcu_read_lock();
1501                         if (fcheck_files(files, fd)) {
1502                                 rcu_read_unlock();
1503                                 put_files_struct(files);
1504                                 if (task_dumpable(task)) {
1505                                         inode->i_uid = task->euid;
1506                                         inode->i_gid = task->egid;
1507                                 } else {
1508                                         inode->i_uid = 0;
1509                                         inode->i_gid = 0;
1510                                 }
1511                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1512                                 security_task_to_inode(task, inode);
1513                                 put_task_struct(task);
1514                                 return 1;
1515                         }
1516                         rcu_read_unlock();
1517                         put_files_struct(files);
1518                 }
1519                 put_task_struct(task);
1520         }
1521         d_drop(dentry);
1522         return 0;
1523 }
1524
1525 static struct dentry_operations tid_fd_dentry_operations =
1526 {
1527         .d_revalidate   = tid_fd_revalidate,
1528         .d_delete       = pid_delete_dentry,
1529 };
1530
1531 static struct dentry *proc_fd_instantiate(struct inode *dir,
1532         struct dentry *dentry, struct task_struct *task, const void *ptr)
1533 {
1534         unsigned fd = *(const unsigned *)ptr;
1535         struct file *file;
1536         struct files_struct *files;
1537         struct inode *inode;
1538         struct proc_inode *ei;
1539         struct dentry *error = ERR_PTR(-ENOENT);
1540
1541         inode = proc_pid_make_inode(dir->i_sb, task);
1542         if (!inode)
1543                 goto out;
1544         ei = PROC_I(inode);
1545         ei->fd = fd;
1546         files = get_files_struct(task);
1547         if (!files)
1548                 goto out_iput;
1549         inode->i_mode = S_IFLNK;
1550
1551         /*
1552          * We are not taking a ref to the file structure, so we must
1553          * hold ->file_lock.
1554          */
1555         spin_lock(&files->file_lock);
1556         file = fcheck_files(files, fd);
1557         if (!file)
1558                 goto out_unlock;
1559         if (file->f_mode & 1)
1560                 inode->i_mode |= S_IRUSR | S_IXUSR;
1561         if (file->f_mode & 2)
1562                 inode->i_mode |= S_IWUSR | S_IXUSR;
1563         spin_unlock(&files->file_lock);
1564         put_files_struct(files);
1565
1566         inode->i_op = &proc_pid_link_inode_operations;
1567         inode->i_size = 64;
1568         ei->op.proc_get_link = proc_fd_link;
1569         dentry->d_op = &tid_fd_dentry_operations;
1570         d_add(dentry, inode);
1571         /* Close the race of the process dying before we return the dentry */
1572         if (tid_fd_revalidate(dentry, NULL))
1573                 error = NULL;
1574
1575  out:
1576         return error;
1577 out_unlock:
1578         spin_unlock(&files->file_lock);
1579         put_files_struct(files);
1580 out_iput:
1581         iput(inode);
1582         goto out;
1583 }
1584
1585 static struct dentry *proc_lookupfd_common(struct inode *dir,
1586                                            struct dentry *dentry,
1587                                            instantiate_t instantiate)
1588 {
1589         struct task_struct *task = get_proc_task(dir);
1590         unsigned fd = name_to_int(dentry);
1591         struct dentry *result = ERR_PTR(-ENOENT);
1592
1593         if (!task)
1594                 goto out_no_task;
1595         if (fd == ~0U)
1596                 goto out;
1597
1598         result = instantiate(dir, dentry, task, &fd);
1599 out:
1600         put_task_struct(task);
1601 out_no_task:
1602         return result;
1603 }
1604
1605 static int proc_readfd_common(struct file * filp, void * dirent,
1606                               filldir_t filldir, instantiate_t instantiate)
1607 {
1608         struct dentry *dentry = filp->f_path.dentry;
1609         struct inode *inode = dentry->d_inode;
1610         struct task_struct *p = get_proc_task(inode);
1611         unsigned int fd, ino;
1612         int retval;
1613         struct files_struct * files;
1614         struct fdtable *fdt;
1615
1616         retval = -ENOENT;
1617         if (!p)
1618                 goto out_no_task;
1619         retval = 0;
1620
1621         fd = filp->f_pos;
1622         switch (fd) {
1623                 case 0:
1624                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1625                                 goto out;
1626                         filp->f_pos++;
1627                 case 1:
1628                         ino = parent_ino(dentry);
1629                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1630                                 goto out;
1631                         filp->f_pos++;
1632                 default:
1633                         files = get_files_struct(p);
1634                         if (!files)
1635                                 goto out;
1636                         rcu_read_lock();
1637                         fdt = files_fdtable(files);
1638                         for (fd = filp->f_pos-2;
1639                              fd < fdt->max_fds;
1640                              fd++, filp->f_pos++) {
1641                                 char name[PROC_NUMBUF];
1642                                 int len;
1643
1644                                 if (!fcheck_files(files, fd))
1645                                         continue;
1646                                 rcu_read_unlock();
1647
1648                                 len = snprintf(name, sizeof(name), "%d", fd);
1649                                 if (proc_fill_cache(filp, dirent, filldir,
1650                                                     name, len, instantiate,
1651                                                     p, &fd) < 0) {
1652                                         rcu_read_lock();
1653                                         break;
1654                                 }
1655                                 rcu_read_lock();
1656                         }
1657                         rcu_read_unlock();
1658                         put_files_struct(files);
1659         }
1660 out:
1661         put_task_struct(p);
1662 out_no_task:
1663         return retval;
1664 }
1665
1666 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1667                                     struct nameidata *nd)
1668 {
1669         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1670 }
1671
1672 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1673 {
1674         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1675 }
1676
1677 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1678                                       size_t len, loff_t *ppos)
1679 {
1680         char tmp[PROC_FDINFO_MAX];
1681         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1682         if (!err)
1683                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1684         return err;
1685 }
1686
1687 static const struct file_operations proc_fdinfo_file_operations = {
1688         .open           = nonseekable_open,
1689         .read           = proc_fdinfo_read,
1690 };
1691
1692 static const struct file_operations proc_fd_operations = {
1693         .read           = generic_read_dir,
1694         .readdir        = proc_readfd,
1695 };
1696
1697 /*
1698  * /proc/pid/fd needs a special permission handler so that a process can still
1699  * access /proc/self/fd after it has executed a setuid().
1700  */
1701 static int proc_fd_permission(struct inode *inode, int mask,
1702                                 struct nameidata *nd)
1703 {
1704         int rv;
1705
1706         rv = generic_permission(inode, mask, NULL);
1707         if (rv == 0)
1708                 return 0;
1709         if (task_pid(current) == proc_pid(inode))
1710                 rv = 0;
1711         return rv;
1712 }
1713
1714 /*
1715  * proc directories can do almost nothing..
1716  */
1717 static const struct inode_operations proc_fd_inode_operations = {
1718         .lookup         = proc_lookupfd,
1719         .permission     = proc_fd_permission,
1720         .setattr        = proc_setattr,
1721 };
1722
1723 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1724         struct dentry *dentry, struct task_struct *task, const void *ptr)
1725 {
1726         unsigned fd = *(unsigned *)ptr;
1727         struct inode *inode;
1728         struct proc_inode *ei;
1729         struct dentry *error = ERR_PTR(-ENOENT);
1730
1731         inode = proc_pid_make_inode(dir->i_sb, task);
1732         if (!inode)
1733                 goto out;
1734         ei = PROC_I(inode);
1735         ei->fd = fd;
1736         inode->i_mode = S_IFREG | S_IRUSR;
1737         inode->i_fop = &proc_fdinfo_file_operations;
1738         dentry->d_op = &tid_fd_dentry_operations;
1739         d_add(dentry, inode);
1740         /* Close the race of the process dying before we return the dentry */
1741         if (tid_fd_revalidate(dentry, NULL))
1742                 error = NULL;
1743
1744  out:
1745         return error;
1746 }
1747
1748 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1749                                         struct dentry *dentry,
1750                                         struct nameidata *nd)
1751 {
1752         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1753 }
1754
1755 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1756 {
1757         return proc_readfd_common(filp, dirent, filldir,
1758                                   proc_fdinfo_instantiate);
1759 }
1760
1761 static const struct file_operations proc_fdinfo_operations = {
1762         .read           = generic_read_dir,
1763         .readdir        = proc_readfdinfo,
1764 };
1765
1766 /*
1767  * proc directories can do almost nothing..
1768  */
1769 static const struct inode_operations proc_fdinfo_inode_operations = {
1770         .lookup         = proc_lookupfdinfo,
1771         .setattr        = proc_setattr,
1772 };
1773
1774
1775 static struct dentry *proc_pident_instantiate(struct inode *dir,
1776         struct dentry *dentry, struct task_struct *task, const void *ptr)
1777 {
1778         const struct pid_entry *p = ptr;
1779         struct inode *inode;
1780         struct proc_inode *ei;
1781         struct dentry *error = ERR_PTR(-EINVAL);
1782
1783         inode = proc_pid_make_inode(dir->i_sb, task);
1784         if (!inode)
1785                 goto out;
1786
1787         ei = PROC_I(inode);
1788         inode->i_mode = p->mode;
1789         if (S_ISDIR(inode->i_mode))
1790                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1791         if (p->iop)
1792                 inode->i_op = p->iop;
1793         if (p->fop)
1794                 inode->i_fop = p->fop;
1795         ei->op = p->op;
1796         dentry->d_op = &pid_dentry_operations;
1797         d_add(dentry, inode);
1798         /* Close the race of the process dying before we return the dentry */
1799         if (pid_revalidate(dentry, NULL))
1800                 error = NULL;
1801 out:
1802         return error;
1803 }
1804
1805 static struct dentry *proc_pident_lookup(struct inode *dir,
1806                                          struct dentry *dentry,
1807                                          const struct pid_entry *ents,
1808                                          unsigned int nents)
1809 {
1810         struct inode *inode;
1811         struct dentry *error;
1812         struct task_struct *task = get_proc_task(dir);
1813         const struct pid_entry *p, *last;
1814
1815         error = ERR_PTR(-ENOENT);
1816         inode = NULL;
1817
1818         if (!task)
1819                 goto out_no_task;
1820
1821         /*
1822          * Yes, it does not scale. And it should not. Don't add
1823          * new entries into /proc/<tgid>/ without very good reasons.
1824          */
1825         last = &ents[nents - 1];
1826         for (p = ents; p <= last; p++) {
1827                 if (p->len != dentry->d_name.len)
1828                         continue;
1829                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1830                         break;
1831         }
1832         if (p > last)
1833                 goto out;
1834
1835         error = proc_pident_instantiate(dir, dentry, task, p);
1836 out:
1837         put_task_struct(task);
1838 out_no_task:
1839         return error;
1840 }
1841
1842 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1843         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1844 {
1845         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1846                                 proc_pident_instantiate, task, p);
1847 }
1848
1849 static int proc_pident_readdir(struct file *filp,
1850                 void *dirent, filldir_t filldir,
1851                 const struct pid_entry *ents, unsigned int nents)
1852 {
1853         int i;
1854         struct dentry *dentry = filp->f_path.dentry;
1855         struct inode *inode = dentry->d_inode;
1856         struct task_struct *task = get_proc_task(inode);
1857         const struct pid_entry *p, *last;
1858         ino_t ino;
1859         int ret;
1860
1861         ret = -ENOENT;
1862         if (!task)
1863                 goto out_no_task;
1864
1865         ret = 0;
1866         i = filp->f_pos;
1867         switch (i) {
1868         case 0:
1869                 ino = inode->i_ino;
1870                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1871                         goto out;
1872                 i++;
1873                 filp->f_pos++;
1874                 /* fall through */
1875         case 1:
1876                 ino = parent_ino(dentry);
1877                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1878                         goto out;
1879                 i++;
1880                 filp->f_pos++;
1881                 /* fall through */
1882         default:
1883                 i -= 2;
1884                 if (i >= nents) {
1885                         ret = 1;
1886                         goto out;
1887                 }
1888                 p = ents + i;
1889                 last = &ents[nents - 1];
1890                 while (p <= last) {
1891                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1892                                 goto out;
1893                         filp->f_pos++;
1894                         p++;
1895                 }
1896         }
1897
1898         ret = 1;
1899 out:
1900         put_task_struct(task);
1901 out_no_task:
1902         return ret;
1903 }
1904
1905 #ifdef CONFIG_SECURITY
1906 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1907                                   size_t count, loff_t *ppos)
1908 {
1909         struct inode * inode = file->f_path.dentry->d_inode;
1910         char *p = NULL;
1911         ssize_t length;
1912         struct task_struct *task = get_proc_task(inode);
1913
1914         if (!task)
1915                 return -ESRCH;
1916
1917         length = security_getprocattr(task,
1918                                       (char*)file->f_path.dentry->d_name.name,
1919                                       &p);
1920         put_task_struct(task);
1921         if (length > 0)
1922                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1923         kfree(p);
1924         return length;
1925 }
1926
1927 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1928                                    size_t count, loff_t *ppos)
1929 {
1930         struct inode * inode = file->f_path.dentry->d_inode;
1931         char *page;
1932         ssize_t length;
1933         struct task_struct *task = get_proc_task(inode);
1934
1935         length = -ESRCH;
1936         if (!task)
1937                 goto out_no_task;
1938         if (count > PAGE_SIZE)
1939                 count = PAGE_SIZE;
1940
1941         /* No partial writes. */
1942         length = -EINVAL;
1943         if (*ppos != 0)
1944                 goto out;
1945
1946         length = -ENOMEM;
1947         page = (char*)__get_free_page(GFP_TEMPORARY);
1948         if (!page)
1949                 goto out;
1950
1951         length = -EFAULT;
1952         if (copy_from_user(page, buf, count))
1953                 goto out_free;
1954
1955         length = security_setprocattr(task,
1956                                       (char*)file->f_path.dentry->d_name.name,
1957                                       (void*)page, count);
1958 out_free:
1959         free_page((unsigned long) page);
1960 out:
1961         put_task_struct(task);
1962 out_no_task:
1963         return length;
1964 }
1965
1966 static const struct file_operations proc_pid_attr_operations = {
1967         .read           = proc_pid_attr_read,
1968         .write          = proc_pid_attr_write,
1969 };
1970
1971 static const struct pid_entry attr_dir_stuff[] = {
1972         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1973         REG("prev",       S_IRUGO,         pid_attr),
1974         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1975         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1976         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1977         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1978 };
1979
1980 static int proc_attr_dir_readdir(struct file * filp,
1981                              void * dirent, filldir_t filldir)
1982 {
1983         return proc_pident_readdir(filp,dirent,filldir,
1984                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1985 }
1986
1987 static const struct file_operations proc_attr_dir_operations = {
1988         .read           = generic_read_dir,
1989         .readdir        = proc_attr_dir_readdir,
1990 };
1991
1992 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1993                                 struct dentry *dentry, struct nameidata *nd)
1994 {
1995         return proc_pident_lookup(dir, dentry,
1996                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
1997 }
1998
1999 static const struct inode_operations proc_attr_dir_inode_operations = {
2000         .lookup         = proc_attr_dir_lookup,
2001         .getattr        = pid_getattr,
2002         .setattr        = proc_setattr,
2003 };
2004
2005 #endif
2006
2007 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2008 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2009                                          size_t count, loff_t *ppos)
2010 {
2011         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2012         struct mm_struct *mm;
2013         char buffer[PROC_NUMBUF];
2014         size_t len;
2015         int ret;
2016
2017         if (!task)
2018                 return -ESRCH;
2019
2020         ret = 0;
2021         mm = get_task_mm(task);
2022         if (mm) {
2023                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2024                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2025                                 MMF_DUMP_FILTER_SHIFT));
2026                 mmput(mm);
2027                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2028         }
2029
2030         put_task_struct(task);
2031
2032         return ret;
2033 }
2034
2035 static ssize_t proc_coredump_filter_write(struct file *file,
2036                                           const char __user *buf,
2037                                           size_t count,
2038                                           loff_t *ppos)
2039 {
2040         struct task_struct *task;
2041         struct mm_struct *mm;
2042         char buffer[PROC_NUMBUF], *end;
2043         unsigned int val;
2044         int ret;
2045         int i;
2046         unsigned long mask;
2047
2048         ret = -EFAULT;
2049         memset(buffer, 0, sizeof(buffer));
2050         if (count > sizeof(buffer) - 1)
2051                 count = sizeof(buffer) - 1;
2052         if (copy_from_user(buffer, buf, count))
2053                 goto out_no_task;
2054
2055         ret = -EINVAL;
2056         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2057         if (*end == '\n')
2058                 end++;
2059         if (end - buffer == 0)
2060                 goto out_no_task;
2061
2062         ret = -ESRCH;
2063         task = get_proc_task(file->f_dentry->d_inode);
2064         if (!task)
2065                 goto out_no_task;
2066
2067         ret = end - buffer;
2068         mm = get_task_mm(task);
2069         if (!mm)
2070                 goto out_no_mm;
2071
2072         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2073                 if (val & mask)
2074                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2075                 else
2076                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2077         }
2078
2079         mmput(mm);
2080  out_no_mm:
2081         put_task_struct(task);
2082  out_no_task:
2083         return ret;
2084 }
2085
2086 static const struct file_operations proc_coredump_filter_operations = {
2087         .read           = proc_coredump_filter_read,
2088         .write          = proc_coredump_filter_write,
2089 };
2090 #endif
2091
2092 /*
2093  * /proc/self:
2094  */
2095 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2096                               int buflen)
2097 {
2098         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2099         pid_t tgid = task_tgid_nr_ns(current, ns);
2100         char tmp[PROC_NUMBUF];
2101         if (!tgid)
2102                 return -ENOENT;
2103         sprintf(tmp, "%d", tgid);
2104         return vfs_readlink(dentry,buffer,buflen,tmp);
2105 }
2106
2107 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2108 {
2109         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2110         pid_t tgid = task_tgid_nr_ns(current, ns);
2111         char tmp[PROC_NUMBUF];
2112         if (!tgid)
2113                 return ERR_PTR(-ENOENT);
2114         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2115         return ERR_PTR(vfs_follow_link(nd,tmp));
2116 }
2117
2118 static const struct inode_operations proc_self_inode_operations = {
2119         .readlink       = proc_self_readlink,
2120         .follow_link    = proc_self_follow_link,
2121 };
2122
2123 /*
2124  * proc base
2125  *
2126  * These are the directory entries in the root directory of /proc
2127  * that properly belong to the /proc filesystem, as they describe
2128  * describe something that is process related.
2129  */
2130 static const struct pid_entry proc_base_stuff[] = {
2131         NOD("self", S_IFLNK|S_IRWXUGO,
2132                 &proc_self_inode_operations, NULL, {}),
2133 };
2134
2135 /*
2136  *      Exceptional case: normally we are not allowed to unhash a busy
2137  * directory. In this case, however, we can do it - no aliasing problems
2138  * due to the way we treat inodes.
2139  */
2140 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2141 {
2142         struct inode *inode = dentry->d_inode;
2143         struct task_struct *task = get_proc_task(inode);
2144         if (task) {
2145                 put_task_struct(task);
2146                 return 1;
2147         }
2148         d_drop(dentry);
2149         return 0;
2150 }
2151
2152 static struct dentry_operations proc_base_dentry_operations =
2153 {
2154         .d_revalidate   = proc_base_revalidate,
2155         .d_delete       = pid_delete_dentry,
2156 };
2157
2158 static struct dentry *proc_base_instantiate(struct inode *dir,
2159         struct dentry *dentry, struct task_struct *task, const void *ptr)
2160 {
2161         const struct pid_entry *p = ptr;
2162         struct inode *inode;
2163         struct proc_inode *ei;
2164         struct dentry *error = ERR_PTR(-EINVAL);
2165
2166         /* Allocate the inode */
2167         error = ERR_PTR(-ENOMEM);
2168         inode = new_inode(dir->i_sb);
2169         if (!inode)
2170                 goto out;
2171
2172         /* Initialize the inode */
2173         ei = PROC_I(inode);
2174         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2175
2176         /*
2177          * grab the reference to the task.
2178          */
2179         ei->pid = get_task_pid(task, PIDTYPE_PID);
2180         if (!ei->pid)
2181                 goto out_iput;
2182
2183         inode->i_uid = 0;
2184         inode->i_gid = 0;
2185         inode->i_mode = p->mode;
2186         if (S_ISDIR(inode->i_mode))
2187                 inode->i_nlink = 2;
2188         if (S_ISLNK(inode->i_mode))
2189                 inode->i_size = 64;
2190         if (p->iop)
2191                 inode->i_op = p->iop;
2192         if (p->fop)
2193                 inode->i_fop = p->fop;
2194         ei->op = p->op;
2195         dentry->d_op = &proc_base_dentry_operations;
2196         d_add(dentry, inode);
2197         error = NULL;
2198 out:
2199         return error;
2200 out_iput:
2201         iput(inode);
2202         goto out;
2203 }
2204
2205 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2206 {
2207         struct dentry *error;
2208         struct task_struct *task = get_proc_task(dir);
2209         const struct pid_entry *p, *last;
2210
2211         error = ERR_PTR(-ENOENT);
2212
2213         if (!task)
2214                 goto out_no_task;
2215
2216         /* Lookup the directory entry */
2217         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2218         for (p = proc_base_stuff; p <= last; p++) {
2219                 if (p->len != dentry->d_name.len)
2220                         continue;
2221                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2222                         break;
2223         }
2224         if (p > last)
2225                 goto out;
2226
2227         error = proc_base_instantiate(dir, dentry, task, p);
2228
2229 out:
2230         put_task_struct(task);
2231 out_no_task:
2232         return error;
2233 }
2234
2235 static int proc_base_fill_cache(struct file *filp, void *dirent,
2236         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2237 {
2238         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2239                                 proc_base_instantiate, task, p);
2240 }
2241
2242 #ifdef CONFIG_TASK_IO_ACCOUNTING
2243 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2244 {
2245         return sprintf(buffer,
2246 #ifdef CONFIG_TASK_XACCT
2247                         "rchar: %llu\n"
2248                         "wchar: %llu\n"
2249                         "syscr: %llu\n"
2250                         "syscw: %llu\n"
2251 #endif
2252                         "read_bytes: %llu\n"
2253                         "write_bytes: %llu\n"
2254                         "cancelled_write_bytes: %llu\n",
2255 #ifdef CONFIG_TASK_XACCT
2256                         (unsigned long long)task->rchar,
2257                         (unsigned long long)task->wchar,
2258                         (unsigned long long)task->syscr,
2259                         (unsigned long long)task->syscw,
2260 #endif
2261                         (unsigned long long)task->ioac.read_bytes,
2262                         (unsigned long long)task->ioac.write_bytes,
2263                         (unsigned long long)task->ioac.cancelled_write_bytes);
2264 }
2265 #endif
2266
2267 /*
2268  * Thread groups
2269  */
2270 static const struct file_operations proc_task_operations;
2271 static const struct inode_operations proc_task_inode_operations;
2272
2273 static const struct pid_entry tgid_base_stuff[] = {
2274         DIR("task",       S_IRUGO|S_IXUGO, task),
2275         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2276         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2277         REG("environ",    S_IRUSR, environ),
2278         INF("auxv",       S_IRUSR, pid_auxv),
2279         ONE("status",     S_IRUGO, pid_status),
2280         INF("limits",     S_IRUSR, pid_limits),
2281 #ifdef CONFIG_SCHED_DEBUG
2282         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2283 #endif
2284         INF("cmdline",    S_IRUGO, pid_cmdline),
2285         ONE("stat",       S_IRUGO, tgid_stat),
2286         ONE("statm",      S_IRUGO, pid_statm),
2287         REG("maps",       S_IRUGO, maps),
2288 #ifdef CONFIG_NUMA
2289         REG("numa_maps",  S_IRUGO, numa_maps),
2290 #endif
2291         REG("mem",        S_IRUSR|S_IWUSR, mem),
2292         LNK("cwd",        cwd),
2293         LNK("root",       root),
2294         LNK("exe",        exe),
2295         REG("mounts",     S_IRUGO, mounts),
2296         REG("mountstats", S_IRUSR, mountstats),
2297 #ifdef CONFIG_PROC_PAGE_MONITOR
2298         REG("clear_refs", S_IWUSR, clear_refs),
2299         REG("smaps",      S_IRUGO, smaps),
2300         REG("pagemap",    S_IRUSR, pagemap),
2301 #endif
2302 #ifdef CONFIG_SECURITY
2303         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2304 #endif
2305 #ifdef CONFIG_KALLSYMS
2306         INF("wchan",      S_IRUGO, pid_wchan),
2307 #endif
2308 #ifdef CONFIG_SCHEDSTATS
2309         INF("schedstat",  S_IRUGO, pid_schedstat),
2310 #endif
2311 #ifdef CONFIG_LATENCYTOP
2312         REG("latency",  S_IRUGO, lstats),
2313 #endif
2314 #ifdef CONFIG_PROC_PID_CPUSET
2315         REG("cpuset",     S_IRUGO, cpuset),
2316 #endif
2317 #ifdef CONFIG_CGROUPS
2318         REG("cgroup",  S_IRUGO, cgroup),
2319 #endif
2320         INF("oom_score",  S_IRUGO, oom_score),
2321         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2322 #ifdef CONFIG_AUDITSYSCALL
2323         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2324 #endif
2325 #ifdef CONFIG_FAULT_INJECTION
2326         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2327 #endif
2328 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2329         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2330 #endif
2331 #ifdef CONFIG_TASK_IO_ACCOUNTING
2332         INF("io",       S_IRUGO, pid_io_accounting),
2333 #endif
2334 };
2335
2336 static int proc_tgid_base_readdir(struct file * filp,
2337                              void * dirent, filldir_t filldir)
2338 {
2339         return proc_pident_readdir(filp,dirent,filldir,
2340                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2341 }
2342
2343 static const struct file_operations proc_tgid_base_operations = {
2344         .read           = generic_read_dir,
2345         .readdir        = proc_tgid_base_readdir,
2346 };
2347
2348 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2349         return proc_pident_lookup(dir, dentry,
2350                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2351 }
2352
2353 static const struct inode_operations proc_tgid_base_inode_operations = {
2354         .lookup         = proc_tgid_base_lookup,
2355         .getattr        = pid_getattr,
2356         .setattr        = proc_setattr,
2357 };
2358
2359 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2360 {
2361         struct dentry *dentry, *leader, *dir;
2362         char buf[PROC_NUMBUF];
2363         struct qstr name;
2364
2365         name.name = buf;
2366         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2367         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2368         if (dentry) {
2369                 if (!(current->flags & PF_EXITING))
2370                         shrink_dcache_parent(dentry);
2371                 d_drop(dentry);
2372                 dput(dentry);
2373         }
2374
2375         if (tgid == 0)
2376                 goto out;
2377
2378         name.name = buf;
2379         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2380         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2381         if (!leader)
2382                 goto out;
2383
2384         name.name = "task";
2385         name.len = strlen(name.name);
2386         dir = d_hash_and_lookup(leader, &name);
2387         if (!dir)
2388                 goto out_put_leader;
2389
2390         name.name = buf;
2391         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2392         dentry = d_hash_and_lookup(dir, &name);
2393         if (dentry) {
2394                 shrink_dcache_parent(dentry);
2395                 d_drop(dentry);
2396                 dput(dentry);
2397         }
2398
2399         dput(dir);
2400 out_put_leader:
2401         dput(leader);
2402 out:
2403         return;
2404 }
2405
2406 /**
2407  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2408  * @task: task that should be flushed.
2409  *
2410  * When flushing dentries from proc, one needs to flush them from global
2411  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2412  * in. This call is supposed to do all of this job.
2413  *
2414  * Looks in the dcache for
2415  * /proc/@pid
2416  * /proc/@tgid/task/@pid
2417  * if either directory is present flushes it and all of it'ts children
2418  * from the dcache.
2419  *
2420  * It is safe and reasonable to cache /proc entries for a task until
2421  * that task exits.  After that they just clog up the dcache with
2422  * useless entries, possibly causing useful dcache entries to be
2423  * flushed instead.  This routine is proved to flush those useless
2424  * dcache entries at process exit time.
2425  *
2426  * NOTE: This routine is just an optimization so it does not guarantee
2427  *       that no dcache entries will exist at process exit time it
2428  *       just makes it very unlikely that any will persist.
2429  */
2430
2431 void proc_flush_task(struct task_struct *task)
2432 {
2433         int i;
2434         struct pid *pid, *tgid = NULL;
2435         struct upid *upid;
2436
2437         pid = task_pid(task);
2438         if (thread_group_leader(task))
2439                 tgid = task_tgid(task);
2440
2441         for (i = 0; i <= pid->level; i++) {
2442                 upid = &pid->numbers[i];
2443                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2444                         tgid ? tgid->numbers[i].nr : 0);
2445         }
2446
2447         upid = &pid->numbers[pid->level];
2448         if (upid->nr == 1)
2449                 pid_ns_release_proc(upid->ns);
2450 }
2451
2452 static struct dentry *proc_pid_instantiate(struct inode *dir,
2453                                            struct dentry * dentry,
2454                                            struct task_struct *task, const void *ptr)
2455 {
2456         struct dentry *error = ERR_PTR(-ENOENT);
2457         struct inode *inode;
2458
2459         inode = proc_pid_make_inode(dir->i_sb, task);
2460         if (!inode)
2461                 goto out;
2462
2463         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2464         inode->i_op = &proc_tgid_base_inode_operations;
2465         inode->i_fop = &proc_tgid_base_operations;
2466         inode->i_flags|=S_IMMUTABLE;
2467         inode->i_nlink = 5;
2468 #ifdef CONFIG_SECURITY
2469         inode->i_nlink += 1;
2470 #endif
2471
2472         dentry->d_op = &pid_dentry_operations;
2473
2474         d_add(dentry, inode);
2475         /* Close the race of the process dying before we return the dentry */
2476         if (pid_revalidate(dentry, NULL))
2477                 error = NULL;
2478 out:
2479         return error;
2480 }
2481
2482 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2483 {
2484         struct dentry *result = ERR_PTR(-ENOENT);
2485         struct task_struct *task;
2486         unsigned tgid;
2487         struct pid_namespace *ns;
2488
2489         result = proc_base_lookup(dir, dentry);
2490         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2491                 goto out;
2492
2493         tgid = name_to_int(dentry);
2494         if (tgid == ~0U)
2495                 goto out;
2496
2497         ns = dentry->d_sb->s_fs_info;
2498         rcu_read_lock();
2499         task = find_task_by_pid_ns(tgid, ns);
2500         if (task)
2501                 get_task_struct(task);
2502         rcu_read_unlock();
2503         if (!task)
2504                 goto out;
2505
2506         result = proc_pid_instantiate(dir, dentry, task, NULL);
2507         put_task_struct(task);
2508 out:
2509         return result;
2510 }
2511
2512 /*
2513  * Find the first task with tgid >= tgid
2514  *
2515  */
2516 struct tgid_iter {
2517         unsigned int tgid;
2518         struct task_struct *task;
2519 };
2520 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2521 {
2522         struct pid *pid;
2523
2524         if (iter.task)
2525                 put_task_struct(iter.task);
2526         rcu_read_lock();
2527 retry:
2528         iter.task = NULL;
2529         pid = find_ge_pid(iter.tgid, ns);
2530         if (pid) {
2531                 iter.tgid = pid_nr_ns(pid, ns);
2532                 iter.task = pid_task(pid, PIDTYPE_PID);
2533                 /* What we to know is if the pid we have find is the
2534                  * pid of a thread_group_leader.  Testing for task
2535                  * being a thread_group_leader is the obvious thing
2536                  * todo but there is a window when it fails, due to
2537                  * the pid transfer logic in de_thread.
2538                  *
2539                  * So we perform the straight forward test of seeing
2540                  * if the pid we have found is the pid of a thread
2541                  * group leader, and don't worry if the task we have
2542                  * found doesn't happen to be a thread group leader.
2543                  * As we don't care in the case of readdir.
2544                  */
2545                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2546                         iter.tgid += 1;
2547                         goto retry;
2548                 }
2549                 get_task_struct(iter.task);
2550         }
2551         rcu_read_unlock();
2552         return iter;
2553 }
2554
2555 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2556
2557 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2558         struct tgid_iter iter)
2559 {
2560         char name[PROC_NUMBUF];
2561         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2562         return proc_fill_cache(filp, dirent, filldir, name, len,
2563                                 proc_pid_instantiate, iter.task, NULL);
2564 }
2565
2566 /* for the /proc/ directory itself, after non-process stuff has been done */
2567 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2568 {
2569         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2570         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2571         struct tgid_iter iter;
2572         struct pid_namespace *ns;
2573
2574         if (!reaper)
2575                 goto out_no_task;
2576
2577         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2578                 const struct pid_entry *p = &proc_base_stuff[nr];
2579                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2580                         goto out;
2581         }
2582
2583         ns = filp->f_dentry->d_sb->s_fs_info;
2584         iter.task = NULL;
2585         iter.tgid = filp->f_pos - TGID_OFFSET;
2586         for (iter = next_tgid(ns, iter);
2587              iter.task;
2588              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2589                 filp->f_pos = iter.tgid + TGID_OFFSET;
2590                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2591                         put_task_struct(iter.task);
2592                         goto out;
2593                 }
2594         }
2595         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2596 out:
2597         put_task_struct(reaper);
2598 out_no_task:
2599         return 0;
2600 }
2601
2602 /*
2603  * Tasks
2604  */
2605 static const struct pid_entry tid_base_stuff[] = {
2606         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2607         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2608         REG("environ",   S_IRUSR, environ),
2609         INF("auxv",      S_IRUSR, pid_auxv),
2610         ONE("status",    S_IRUGO, pid_status),
2611         INF("limits",    S_IRUSR, pid_limits),
2612 #ifdef CONFIG_SCHED_DEBUG
2613         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2614 #endif
2615         INF("cmdline",   S_IRUGO, pid_cmdline),
2616         ONE("stat",      S_IRUGO, tid_stat),
2617         ONE("statm",     S_IRUGO, pid_statm),
2618         REG("maps",      S_IRUGO, maps),
2619 #ifdef CONFIG_NUMA
2620         REG("numa_maps", S_IRUGO, numa_maps),
2621 #endif
2622         REG("mem",       S_IRUSR|S_IWUSR, mem),
2623         LNK("cwd",       cwd),
2624         LNK("root",      root),
2625         LNK("exe",       exe),
2626         REG("mounts",    S_IRUGO, mounts),
2627 #ifdef CONFIG_PROC_PAGE_MONITOR
2628         REG("clear_refs", S_IWUSR, clear_refs),
2629         REG("smaps",     S_IRUGO, smaps),
2630         REG("pagemap",    S_IRUSR, pagemap),
2631 #endif
2632 #ifdef CONFIG_SECURITY
2633         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2634 #endif
2635 #ifdef CONFIG_KALLSYMS
2636         INF("wchan",     S_IRUGO, pid_wchan),
2637 #endif
2638 #ifdef CONFIG_SCHEDSTATS
2639         INF("schedstat", S_IRUGO, pid_schedstat),
2640 #endif
2641 #ifdef CONFIG_LATENCYTOP
2642         REG("latency",  S_IRUGO, lstats),
2643 #endif
2644 #ifdef CONFIG_PROC_PID_CPUSET
2645         REG("cpuset",    S_IRUGO, cpuset),
2646 #endif
2647 #ifdef CONFIG_CGROUPS
2648         REG("cgroup",  S_IRUGO, cgroup),
2649 #endif
2650         INF("oom_score", S_IRUGO, oom_score),
2651         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2652 #ifdef CONFIG_AUDITSYSCALL
2653         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2654 #endif
2655 #ifdef CONFIG_FAULT_INJECTION
2656         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2657 #endif
2658 };
2659
2660 static int proc_tid_base_readdir(struct file * filp,
2661                              void * dirent, filldir_t filldir)
2662 {
2663         return proc_pident_readdir(filp,dirent,filldir,
2664                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2665 }
2666
2667 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2668         return proc_pident_lookup(dir, dentry,
2669                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2670 }
2671
2672 static const struct file_operations proc_tid_base_operations = {
2673         .read           = generic_read_dir,
2674         .readdir        = proc_tid_base_readdir,
2675 };
2676
2677 static const struct inode_operations proc_tid_base_inode_operations = {
2678         .lookup         = proc_tid_base_lookup,
2679         .getattr        = pid_getattr,
2680         .setattr        = proc_setattr,
2681 };
2682
2683 static struct dentry *proc_task_instantiate(struct inode *dir,
2684         struct dentry *dentry, struct task_struct *task, const void *ptr)
2685 {
2686         struct dentry *error = ERR_PTR(-ENOENT);
2687         struct inode *inode;
2688         inode = proc_pid_make_inode(dir->i_sb, task);
2689
2690         if (!inode)
2691                 goto out;
2692         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2693         inode->i_op = &proc_tid_base_inode_operations;
2694         inode->i_fop = &proc_tid_base_operations;
2695         inode->i_flags|=S_IMMUTABLE;
2696         inode->i_nlink = 4;
2697 #ifdef CONFIG_SECURITY
2698         inode->i_nlink += 1;
2699 #endif
2700
2701         dentry->d_op = &pid_dentry_operations;
2702
2703         d_add(dentry, inode);
2704         /* Close the race of the process dying before we return the dentry */
2705         if (pid_revalidate(dentry, NULL))
2706                 error = NULL;
2707 out:
2708         return error;
2709 }
2710
2711 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2712 {
2713         struct dentry *result = ERR_PTR(-ENOENT);
2714         struct task_struct *task;
2715         struct task_struct *leader = get_proc_task(dir);
2716         unsigned tid;
2717         struct pid_namespace *ns;
2718
2719         if (!leader)
2720                 goto out_no_task;
2721
2722         tid = name_to_int(dentry);
2723         if (tid == ~0U)
2724                 goto out;
2725
2726         ns = dentry->d_sb->s_fs_info;
2727         rcu_read_lock();
2728         task = find_task_by_pid_ns(tid, ns);
2729         if (task)
2730                 get_task_struct(task);
2731         rcu_read_unlock();
2732         if (!task)
2733                 goto out;
2734         if (!same_thread_group(leader, task))
2735                 goto out_drop_task;
2736
2737         result = proc_task_instantiate(dir, dentry, task, NULL);
2738 out_drop_task:
2739         put_task_struct(task);
2740 out:
2741         put_task_struct(leader);
2742 out_no_task:
2743         return result;
2744 }
2745
2746 /*
2747  * Find the first tid of a thread group to return to user space.
2748  *
2749  * Usually this is just the thread group leader, but if the users
2750  * buffer was too small or there was a seek into the middle of the
2751  * directory we have more work todo.
2752  *
2753  * In the case of a short read we start with find_task_by_pid.
2754  *
2755  * In the case of a seek we start with the leader and walk nr
2756  * threads past it.
2757  */
2758 static struct task_struct *first_tid(struct task_struct *leader,
2759                 int tid, int nr, struct pid_namespace *ns)
2760 {
2761         struct task_struct *pos;
2762
2763         rcu_read_lock();
2764         /* Attempt to start with the pid of a thread */
2765         if (tid && (nr > 0)) {
2766                 pos = find_task_by_pid_ns(tid, ns);
2767                 if (pos && (pos->group_leader == leader))
2768                         goto found;
2769         }
2770
2771         /* If nr exceeds the number of threads there is nothing todo */
2772         pos = NULL;
2773         if (nr && nr >= get_nr_threads(leader))
2774                 goto out;
2775
2776         /* If we haven't found our starting place yet start
2777          * with the leader and walk nr threads forward.
2778          */
2779         for (pos = leader; nr > 0; --nr) {
2780                 pos = next_thread(pos);
2781                 if (pos == leader) {
2782                         pos = NULL;
2783                         goto out;
2784                 }
2785         }
2786 found:
2787         get_task_struct(pos);
2788 out:
2789         rcu_read_unlock();
2790         return pos;
2791 }
2792
2793 /*
2794  * Find the next thread in the thread list.
2795  * Return NULL if there is an error or no next thread.
2796  *
2797  * The reference to the input task_struct is released.
2798  */
2799 static struct task_struct *next_tid(struct task_struct *start)
2800 {
2801         struct task_struct *pos = NULL;
2802         rcu_read_lock();
2803         if (pid_alive(start)) {
2804                 pos = next_thread(start);
2805                 if (thread_group_leader(pos))
2806                         pos = NULL;
2807                 else
2808                         get_task_struct(pos);
2809         }
2810         rcu_read_unlock();
2811         put_task_struct(start);
2812         return pos;
2813 }
2814
2815 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2816         struct task_struct *task, int tid)
2817 {
2818         char name[PROC_NUMBUF];
2819         int len = snprintf(name, sizeof(name), "%d", tid);
2820         return proc_fill_cache(filp, dirent, filldir, name, len,
2821                                 proc_task_instantiate, task, NULL);
2822 }
2823
2824 /* for the /proc/TGID/task/ directories */
2825 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2826 {
2827         struct dentry *dentry = filp->f_path.dentry;
2828         struct inode *inode = dentry->d_inode;
2829         struct task_struct *leader = NULL;
2830         struct task_struct *task;
2831         int retval = -ENOENT;
2832         ino_t ino;
2833         int tid;
2834         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2835         struct pid_namespace *ns;
2836
2837         task = get_proc_task(inode);
2838         if (!task)
2839                 goto out_no_task;
2840         rcu_read_lock();
2841         if (pid_alive(task)) {
2842                 leader = task->group_leader;
2843                 get_task_struct(leader);
2844         }
2845         rcu_read_unlock();
2846         put_task_struct(task);
2847         if (!leader)
2848                 goto out_no_task;
2849         retval = 0;
2850
2851         switch (pos) {
2852         case 0:
2853                 ino = inode->i_ino;
2854                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2855                         goto out;
2856                 pos++;
2857                 /* fall through */
2858         case 1:
2859                 ino = parent_ino(dentry);
2860                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2861                         goto out;
2862                 pos++;
2863                 /* fall through */
2864         }
2865
2866         /* f_version caches the tgid value that the last readdir call couldn't
2867          * return. lseek aka telldir automagically resets f_version to 0.
2868          */
2869         ns = filp->f_dentry->d_sb->s_fs_info;
2870         tid = (int)filp->f_version;
2871         filp->f_version = 0;
2872         for (task = first_tid(leader, tid, pos - 2, ns);
2873              task;
2874              task = next_tid(task), pos++) {
2875                 tid = task_pid_nr_ns(task, ns);
2876                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2877                         /* returning this tgid failed, save it as the first
2878                          * pid for the next readir call */
2879                         filp->f_version = (u64)tid;
2880                         put_task_struct(task);
2881                         break;
2882                 }
2883         }
2884 out:
2885         filp->f_pos = pos;
2886         put_task_struct(leader);
2887 out_no_task:
2888         return retval;
2889 }
2890
2891 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2892 {
2893         struct inode *inode = dentry->d_inode;
2894         struct task_struct *p = get_proc_task(inode);
2895         generic_fillattr(inode, stat);
2896
2897         if (p) {
2898                 rcu_read_lock();
2899                 stat->nlink += get_nr_threads(p);
2900                 rcu_read_unlock();
2901                 put_task_struct(p);
2902         }
2903
2904         return 0;
2905 }
2906
2907 static const struct inode_operations proc_task_inode_operations = {
2908         .lookup         = proc_task_lookup,
2909         .getattr        = proc_task_getattr,
2910         .setattr        = proc_setattr,
2911 };
2912
2913 static const struct file_operations proc_task_operations = {
2914         .read           = generic_read_dir,
2915         .readdir        = proc_task_readdir,
2916 };