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
  92 /* Worst case buffer size needed for holding an integer. */
  93 #define PROC_NUMBUF 13
  94
  95 struct pid_entry {
  96         char *name;
  97         int len;
  98         mode_t mode;
  99         const struct inode_operations *iop;
 100         const struct file_operations *fop;
 101         union proc_op op;
 102 };
 103
 104 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 105         .name = (NAME),                                 \
 106         .len  = sizeof(NAME) - 1,                       \
 107         .mode = MODE,                                   \
 108         .iop  = IOP,                                    \
 109         .fop  = FOP,                                    \
 110         .op   = OP,                                     \
 111 }
 112
 113 #define DIR(NAME, MODE, OTYPE)                                                  \
 114         NOD(NAME, (S_IFDIR|(MODE)),                                             \
 115                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
 116                 {} )
 117 #define LNK(NAME, OTYPE)                                        \
 118         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 119                 &proc_pid_link_inode_operations, NULL,          \
 120                 { .proc_get_link = &proc_##OTYPE##_link } )
 121 #define REG(NAME, MODE, OTYPE)                          \
 122         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
 123                 &proc_##OTYPE##_operations, {})
 124 #define INF(NAME, MODE, OTYPE)                          \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_info_file_operations,       \
 127                 { .proc_read = &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 dentry **dentry, struct vfsmount **mnt)
 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                 *mnt = mntget(fs->pwdmnt);
 169                 *dentry = dget(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 dentry **dentry, struct vfsmount **mnt)
 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                 *mnt = mntget(fs->rootmnt);
 190                 *dentry = dget(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 };
 420
 421 /* Display limits for a process */
 422 static int proc_pid_limits(struct task_struct *task, char *buffer)
 423 {
 424         unsigned int i;
 425         int count = 0;
 426         unsigned long flags;
 427         char *bufptr = buffer;
 428
 429         struct rlimit rlim[RLIM_NLIMITS];
 430
 431         rcu_read_lock();
 432         if (!lock_task_sighand(task,&flags)) {
 433                 rcu_read_unlock();
 434                 return 0;
 435         }
 436         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 437         unlock_task_sighand(task, &flags);
 438         rcu_read_unlock();
 439
 440         /*
 441          * print the file header
 442          */
 443         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 444                         "Limit", "Soft Limit", "Hard Limit", "Units");
 445
 446         for (i = 0; i < RLIM_NLIMITS; i++) {
 447                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 448                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 449                                          lnames[i].name, "unlimited");
 450                 else
 451                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 452                                          lnames[i].name, rlim[i].rlim_cur);
 453
 454                 if (rlim[i].rlim_max == RLIM_INFINITY)
 455                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 456                 else
 457                         count += sprintf(&bufptr[count], "%-20lu ",
 458                                          rlim[i].rlim_max);
 459
 460                 if (lnames[i].unit)
 461                         count += sprintf(&bufptr[count], "%-10s\n",
 462                                          lnames[i].unit);
 463                 else
 464                         count += sprintf(&bufptr[count], "\n");
 465         }
 466
 467         return count;
 468 }
 469
 470 /************************************************************************/
 471 /*                       Here the fs part begins                        */
 472 /************************************************************************/
 473
 474 /* permission checks */
 475 static int proc_fd_access_allowed(struct inode *inode)
 476 {
 477         struct task_struct *task;
 478         int allowed = 0;
 479         /* Allow access to a task's file descriptors if it is us or we
 480          * may use ptrace attach to the process and find out that
 481          * information.
 482          */
 483         task = get_proc_task(inode);
 484         if (task) {
 485                 allowed = ptrace_may_attach(task);
 486                 put_task_struct(task);
 487         }
 488         return allowed;
 489 }
 490
 491 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 492 {
 493         int error;
 494         struct inode *inode = dentry->d_inode;
 495
 496         if (attr->ia_valid & ATTR_MODE)
 497                 return -EPERM;
 498
 499         error = inode_change_ok(inode, attr);
 500         if (!error)
 501                 error = inode_setattr(inode, attr);
 502         return error;
 503 }
 504
 505 static const struct inode_operations proc_def_inode_operations = {
 506         .setattr        = proc_setattr,
 507 };
 508
 509 extern struct seq_operations mounts_op;
 510 struct proc_mounts {
 511         struct seq_file m;
 512         int event;
 513 };
 514
 515 static int mounts_open(struct inode *inode, struct file *file)
 516 {
 517         struct task_struct *task = get_proc_task(inode);
 518         struct nsproxy *nsp;
 519         struct mnt_namespace *ns = NULL;
 520         struct proc_mounts *p;
 521         int ret = -EINVAL;
 522
 523         if (task) {
 524                 rcu_read_lock();
 525                 nsp = task_nsproxy(task);
 526                 if (nsp) {
 527                         ns = nsp->mnt_ns;
 528                         if (ns)
 529                                 get_mnt_ns(ns);
 530                 }
 531                 rcu_read_unlock();
 532
 533                 put_task_struct(task);
 534         }
 535
 536         if (ns) {
 537                 ret = -ENOMEM;
 538                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 539                 if (p) {
 540                         file->private_data = &p->m;
 541                         ret = seq_open(file, &mounts_op);
 542                         if (!ret) {
 543                                 p->m.private = ns;
 544                                 p->event = ns->event;
 545                                 return 0;
 546                         }
 547                         kfree(p);
 548                 }
 549                 put_mnt_ns(ns);
 550         }
 551         return ret;
 552 }
 553
 554 static int mounts_release(struct inode *inode, struct file *file)
 555 {
 556         struct seq_file *m = file->private_data;
 557         struct mnt_namespace *ns = m->private;
 558         put_mnt_ns(ns);
 559         return seq_release(inode, file);
 560 }
 561
 562 static unsigned mounts_poll(struct file *file, poll_table *wait)
 563 {
 564         struct proc_mounts *p = file->private_data;
 565         struct mnt_namespace *ns = p->m.private;
 566         unsigned res = 0;
 567
 568         poll_wait(file, &ns->poll, wait);
 569
 570         spin_lock(&vfsmount_lock);
 571         if (p->event != ns->event) {
 572                 p->event = ns->event;
 573                 res = POLLERR;
 574         }
 575         spin_unlock(&vfsmount_lock);
 576
 577         return res;
 578 }
 579
 580 static const struct file_operations proc_mounts_operations = {
 581         .open           = mounts_open,
 582         .read           = seq_read,
 583         .llseek         = seq_lseek,
 584         .release        = mounts_release,
 585         .poll           = mounts_poll,
 586 };
 587
 588 extern struct seq_operations mountstats_op;
 589 static int mountstats_open(struct inode *inode, struct file *file)
 590 {
 591         int ret = seq_open(file, &mountstats_op);
 592
 593         if (!ret) {
 594                 struct seq_file *m = file->private_data;
 595                 struct nsproxy *nsp;
 596                 struct mnt_namespace *mnt_ns = NULL;
 597                 struct task_struct *task = get_proc_task(inode);
 598
 599                 if (task) {
 600                         rcu_read_lock();
 601                         nsp = task_nsproxy(task);
 602                         if (nsp) {
 603                                 mnt_ns = nsp->mnt_ns;
 604                                 if (mnt_ns)
 605                                         get_mnt_ns(mnt_ns);
 606                         }
 607                         rcu_read_unlock();
 608
 609                         put_task_struct(task);
 610                 }
 611
 612                 if (mnt_ns)
 613                         m->private = mnt_ns;
 614                 else {
 615                         seq_release(inode, file);
 616                         ret = -EINVAL;
 617                 }
 618         }
 619         return ret;
 620 }
 621
 622 static const struct file_operations proc_mountstats_operations = {
 623         .open           = mountstats_open,
 624         .read           = seq_read,
 625         .llseek         = seq_lseek,
 626         .release        = mounts_release,
 627 };
 628
 629 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 630
 631 static ssize_t proc_info_read(struct file * file, char __user * buf,
 632                           size_t count, loff_t *ppos)
 633 {
 634         struct inode * inode = file->f_path.dentry->d_inode;
 635         unsigned long page;
 636         ssize_t length;
 637         struct task_struct *task = get_proc_task(inode);
 638
 639         length = -ESRCH;
 640         if (!task)
 641                 goto out_no_task;
 642
 643         if (count > PROC_BLOCK_SIZE)
 644                 count = PROC_BLOCK_SIZE;
 645
 646         length = -ENOMEM;
 647         if (!(page = __get_free_page(GFP_TEMPORARY)))
 648                 goto out;
 649
 650         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 651
 652         if (length >= 0)
 653                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 654         free_page(page);
 655 out:
 656         put_task_struct(task);
 657 out_no_task:
 658         return length;
 659 }
 660
 661 static const struct file_operations proc_info_file_operations = {
 662         .read           = proc_info_read,
 663 };
 664
 665 static int mem_open(struct inode* inode, struct file* file)
 666 {
 667         file->private_data = (void*)((long)current->self_exec_id);
 668         return 0;
 669 }
 670
 671 static ssize_t mem_read(struct file * file, char __user * buf,
 672                         size_t count, loff_t *ppos)
 673 {
 674         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 675         char *page;
 676         unsigned long src = *ppos;
 677         int ret = -ESRCH;
 678         struct mm_struct *mm;
 679
 680         if (!task)
 681                 goto out_no_task;
 682
 683         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 684                 goto out;
 685
 686         ret = -ENOMEM;
 687         page = (char *)__get_free_page(GFP_TEMPORARY);
 688         if (!page)
 689                 goto out;
 690
 691         ret = 0;
 692
 693         mm = get_task_mm(task);
 694         if (!mm)
 695                 goto out_free;
 696
 697         ret = -EIO;
 698
 699         if (file->private_data != (void*)((long)current->self_exec_id))
 700                 goto out_put;
 701
 702         ret = 0;
 703
 704         while (count > 0) {
 705                 int this_len, retval;
 706
 707                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 708                 retval = access_process_vm(task, src, page, this_len, 0);
 709                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 710                         if (!ret)
 711                                 ret = -EIO;
 712                         break;
 713                 }
 714
 715                 if (copy_to_user(buf, page, retval)) {
 716                         ret = -EFAULT;
 717                         break;
 718                 }
 719
 720                 ret += retval;
 721                 src += retval;
 722                 buf += retval;
 723                 count -= retval;
 724         }
 725         *ppos = src;
 726
 727 out_put:
 728         mmput(mm);
 729 out_free:
 730         free_page((unsigned long) page);
 731 out:
 732         put_task_struct(task);
 733 out_no_task:
 734         return ret;
 735 }
 736
 737 #define mem_write NULL
 738
 739 #ifndef mem_write
 740 /* This is a security hazard */
 741 static ssize_t mem_write(struct file * file, const char __user *buf,
 742                          size_t count, loff_t *ppos)
 743 {
 744         int copied;
 745         char *page;
 746         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 747         unsigned long dst = *ppos;
 748
 749         copied = -ESRCH;
 750         if (!task)
 751                 goto out_no_task;
 752
 753         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 754                 goto out;
 755
 756         copied = -ENOMEM;
 757         page = (char *)__get_free_page(GFP_TEMPORARY);
 758         if (!page)
 759                 goto out;
 760
 761         copied = 0;
 762         while (count > 0) {
 763                 int this_len, retval;
 764
 765                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 766                 if (copy_from_user(page, buf, this_len)) {
 767                         copied = -EFAULT;
 768                         break;
 769                 }
 770                 retval = access_process_vm(task, dst, page, this_len, 1);
 771                 if (!retval) {
 772                         if (!copied)
 773                                 copied = -EIO;
 774                         break;
 775                 }
 776                 copied += retval;
 777                 buf += retval;
 778                 dst += retval;
 779                 count -= retval;
 780         }
 781         *ppos = dst;
 782         free_page((unsigned long) page);
 783 out:
 784         put_task_struct(task);
 785 out_no_task:
 786         return copied;
 787 }
 788 #endif
 789
 790 static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
 791 {
 792         switch (orig) {
 793         case 0:
 794                 file->f_pos = offset;
 795                 break;
 796         case 1:
 797                 file->f_pos += offset;
 798                 break;
 799         default:
 800                 return -EINVAL;
 801         }
 802         force_successful_syscall_return();
 803         return file->f_pos;
 804 }
 805
 806 static const struct file_operations proc_mem_operations = {
 807         .llseek         = mem_lseek,
 808         .read           = mem_read,
 809         .write          = mem_write,
 810         .open           = mem_open,
 811 };
 812
 813 static ssize_t environ_read(struct file *file, char __user *buf,
 814                         size_t count, loff_t *ppos)
 815 {
 816         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 817         char *page;
 818         unsigned long src = *ppos;
 819         int ret = -ESRCH;
 820         struct mm_struct *mm;
 821
 822         if (!task)
 823                 goto out_no_task;
 824
 825         if (!ptrace_may_attach(task))
 826                 goto out;
 827
 828         ret = -ENOMEM;
 829         page = (char *)__get_free_page(GFP_TEMPORARY);
 830         if (!page)
 831                 goto out;
 832
 833         ret = 0;
 834
 835         mm = get_task_mm(task);
 836         if (!mm)
 837                 goto out_free;
 838
 839         while (count > 0) {
 840                 int this_len, retval, max_len;
 841
 842                 this_len = mm->env_end - (mm->env_start + src);
 843
 844                 if (this_len <= 0)
 845                         break;
 846
 847                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 848                 this_len = (this_len > max_len) ? max_len : this_len;
 849
 850                 retval = access_process_vm(task, (mm->env_start + src),
 851                         page, this_len, 0);
 852
 853                 if (retval <= 0) {
 854                         ret = retval;
 855                         break;
 856                 }
 857
 858                 if (copy_to_user(buf, page, retval)) {
 859                         ret = -EFAULT;
 860                         break;
 861                 }
 862
 863                 ret += retval;
 864                 src += retval;
 865                 buf += retval;
 866                 count -= retval;
 867         }
 868         *ppos = src;
 869
 870         mmput(mm);
 871 out_free:
 872         free_page((unsigned long) page);
 873 out:
 874         put_task_struct(task);
 875 out_no_task:
 876         return ret;
 877 }
 878
 879 static const struct file_operations proc_environ_operations = {
 880         .read           = environ_read,
 881 };
 882
 883 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 884                                 size_t count, loff_t *ppos)
 885 {
 886         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 887         char buffer[PROC_NUMBUF];
 888         size_t len;
 889         int oom_adjust;
 890
 891         if (!task)
 892                 return -ESRCH;
 893         oom_adjust = task->oomkilladj;
 894         put_task_struct(task);
 895
 896         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 897
 898         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 899 }
 900
 901 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 902                                 size_t count, loff_t *ppos)
 903 {
 904         struct task_struct *task;
 905         char buffer[PROC_NUMBUF], *end;
 906         int oom_adjust;
 907
 908         memset(buffer, 0, sizeof(buffer));
 909         if (count > sizeof(buffer) - 1)
 910                 count = sizeof(buffer) - 1;
 911         if (copy_from_user(buffer, buf, count))
 912                 return -EFAULT;
 913         oom_adjust = simple_strtol(buffer, &end, 0);
 914         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 915              oom_adjust != OOM_DISABLE)
 916                 return -EINVAL;
 917         if (*end == '\n')
 918                 end++;
 919         task = get_proc_task(file->f_path.dentry->d_inode);
 920         if (!task)
 921                 return -ESRCH;
 922         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
 923                 put_task_struct(task);
 924                 return -EACCES;
 925         }
 926         task->oomkilladj = oom_adjust;
 927         put_task_struct(task);
 928         if (end - buffer == 0)
 929                 return -EIO;
 930         return end - buffer;
 931 }
 932
 933 static const struct file_operations proc_oom_adjust_operations = {
 934         .read           = oom_adjust_read,
 935         .write          = oom_adjust_write,
 936 };
 937
 938 #ifdef CONFIG_MMU
 939 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 940                                 size_t count, loff_t *ppos)
 941 {
 942         struct task_struct *task;
 943         char buffer[PROC_NUMBUF], *end;
 944         struct mm_struct *mm;
 945
 946         memset(buffer, 0, sizeof(buffer));
 947         if (count > sizeof(buffer) - 1)
 948                 count = sizeof(buffer) - 1;
 949         if (copy_from_user(buffer, buf, count))
 950                 return -EFAULT;
 951         if (!simple_strtol(buffer, &end, 0))
 952                 return -EINVAL;
 953         if (*end == '\n')
 954                 end++;
 955         task = get_proc_task(file->f_path.dentry->d_inode);
 956         if (!task)
 957                 return -ESRCH;
 958         mm = get_task_mm(task);
 959         if (mm) {
 960                 clear_refs_smap(mm);
 961                 mmput(mm);
 962         }
 963         put_task_struct(task);
 964         if (end - buffer == 0)
 965                 return -EIO;
 966         return end - buffer;
 967 }
 968
 969 static struct file_operations proc_clear_refs_operations = {
 970         .write          = clear_refs_write,
 971 };
 972 #endif
 973
 974 #ifdef CONFIG_AUDITSYSCALL
 975 #define TMPBUFLEN 21
 976 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 977                                   size_t count, loff_t *ppos)
 978 {
 979         struct inode * inode = file->f_path.dentry->d_inode;
 980         struct task_struct *task = get_proc_task(inode);
 981         ssize_t length;
 982         char tmpbuf[TMPBUFLEN];
 983
 984         if (!task)
 985                 return -ESRCH;
 986         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 987                                 audit_get_loginuid(task->audit_context));
 988         put_task_struct(task);
 989         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 990 }
 991
 992 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 993                                    size_t count, loff_t *ppos)
 994 {
 995         struct inode * inode = file->f_path.dentry->d_inode;
 996         char *page, *tmp;
 997         ssize_t length;
 998         uid_t loginuid;
 999
1000         if (!capable(CAP_AUDIT_CONTROL))
1001                 return -EPERM;
1002
1003         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1004                 return -EPERM;
1005
1006         if (count >= PAGE_SIZE)
1007                 count = PAGE_SIZE - 1;
1008
1009         if (*ppos != 0) {
1010                 /* No partial writes. */
1011                 return -EINVAL;
1012         }
1013         page = (char*)__get_free_page(GFP_TEMPORARY);
1014         if (!page)
1015                 return -ENOMEM;
1016         length = -EFAULT;
1017         if (copy_from_user(page, buf, count))
1018                 goto out_free_page;
1019
1020         page[count] = '\0';
1021         loginuid = simple_strtoul(page, &tmp, 10);
1022         if (tmp == page) {
1023                 length = -EINVAL;
1024                 goto out_free_page;
1025
1026         }
1027         length = audit_set_loginuid(current, loginuid);
1028         if (likely(length == 0))
1029                 length = count;
1030
1031 out_free_page:
1032         free_page((unsigned long) page);
1033         return length;
1034 }
1035
1036 static const struct file_operations proc_loginuid_operations = {
1037         .read           = proc_loginuid_read,
1038         .write          = proc_loginuid_write,
1039 };
1040 #endif
1041
1042 #ifdef CONFIG_FAULT_INJECTION
1043 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1044                                       size_t count, loff_t *ppos)
1045 {
1046         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1047         char buffer[PROC_NUMBUF];
1048         size_t len;
1049         int make_it_fail;
1050
1051         if (!task)
1052                 return -ESRCH;
1053         make_it_fail = task->make_it_fail;
1054         put_task_struct(task);
1055
1056         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1057
1058         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1059 }
1060
1061 static ssize_t proc_fault_inject_write(struct file * file,
1062                         const char __user * buf, size_t count, loff_t *ppos)
1063 {
1064         struct task_struct *task;
1065         char buffer[PROC_NUMBUF], *end;
1066         int make_it_fail;
1067
1068         if (!capable(CAP_SYS_RESOURCE))
1069                 return -EPERM;
1070         memset(buffer, 0, sizeof(buffer));
1071         if (count > sizeof(buffer) - 1)
1072                 count = sizeof(buffer) - 1;
1073         if (copy_from_user(buffer, buf, count))
1074                 return -EFAULT;
1075         make_it_fail = simple_strtol(buffer, &end, 0);
1076         if (*end == '\n')
1077                 end++;
1078         task = get_proc_task(file->f_dentry->d_inode);
1079         if (!task)
1080                 return -ESRCH;
1081         task->make_it_fail = make_it_fail;
1082         put_task_struct(task);
1083         if (end - buffer == 0)
1084                 return -EIO;
1085         return end - buffer;
1086 }
1087
1088 static const struct file_operations proc_fault_inject_operations = {
1089         .read           = proc_fault_inject_read,
1090         .write          = proc_fault_inject_write,
1091 };
1092 #endif
1093
1094
1095 #ifdef CONFIG_SCHED_DEBUG
1096 /*
1097  * Print out various scheduling related per-task fields:
1098  */
1099 static int sched_show(struct seq_file *m, void *v)
1100 {
1101         struct inode *inode = m->private;
1102         struct task_struct *p;
1103
1104         WARN_ON(!inode);
1105
1106         p = get_proc_task(inode);
1107         if (!p)
1108                 return -ESRCH;
1109         proc_sched_show_task(p, m);
1110
1111         put_task_struct(p);
1112
1113         return 0;
1114 }
1115
1116 static ssize_t
1117 sched_write(struct file *file, const char __user *buf,
1118             size_t count, loff_t *offset)
1119 {
1120         struct inode *inode = file->f_path.dentry->d_inode;
1121         struct task_struct *p;
1122
1123         WARN_ON(!inode);
1124
1125         p = get_proc_task(inode);
1126         if (!p)
1127                 return -ESRCH;
1128         proc_sched_set_task(p);
1129
1130         put_task_struct(p);
1131
1132         return count;
1133 }
1134
1135 static int sched_open(struct inode *inode, struct file *filp)
1136 {
1137         int ret;
1138
1139         ret = single_open(filp, sched_show, NULL);
1140         if (!ret) {
1141                 struct seq_file *m = filp->private_data;
1142
1143                 m->private = inode;
1144         }
1145         return ret;
1146 }
1147
1148 static const struct file_operations proc_pid_sched_operations = {
1149         .open           = sched_open,
1150         .read           = seq_read,
1151         .write          = sched_write,
1152         .llseek         = seq_lseek,
1153         .release        = single_release,
1154 };
1155
1156 #endif
1157
1158 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1159 {
1160         struct inode *inode = dentry->d_inode;
1161         int error = -EACCES;
1162
1163         /* We don't need a base pointer in the /proc filesystem */
1164         path_release(nd);
1165
1166         /* Are we allowed to snoop on the tasks file descriptors? */
1167         if (!proc_fd_access_allowed(inode))
1168                 goto out;
1169
1170         error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
1171         nd->last_type = LAST_BIND;
1172 out:
1173         return ERR_PTR(error);
1174 }
1175
1176 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
1177                             char __user *buffer, int buflen)
1178 {
1179         struct inode * inode;
1180         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1181         char *path;
1182         int len;
1183
1184         if (!tmp)
1185                 return -ENOMEM;
1186
1187         inode = dentry->d_inode;
1188         path = d_path(dentry, mnt, tmp, PAGE_SIZE);
1189         len = PTR_ERR(path);
1190         if (IS_ERR(path))
1191                 goto out;
1192         len = tmp + PAGE_SIZE - 1 - path;
1193
1194         if (len > buflen)
1195                 len = buflen;
1196         if (copy_to_user(buffer, path, len))
1197                 len = -EFAULT;
1198  out:
1199         free_page((unsigned long)tmp);
1200         return len;
1201 }
1202
1203 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1204 {
1205         int error = -EACCES;
1206         struct inode *inode = dentry->d_inode;
1207         struct dentry *de;
1208         struct vfsmount *mnt = NULL;
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, &de, &mnt);
1215         if (error)
1216                 goto out;
1217
1218         error = do_proc_readlink(de, mnt, buffer, buflen);
1219         dput(de);
1220         mntput(mnt);
1221 out:
1222         return error;
1223 }
1224
1225 static const struct inode_operations proc_pid_link_inode_operations = {
1226         .readlink       = proc_pid_readlink,
1227         .follow_link    = proc_pid_follow_link,
1228         .setattr        = proc_setattr,
1229 };
1230
1231
1232 /* building an inode */
1233
1234 static int task_dumpable(struct task_struct *task)
1235 {
1236         int dumpable = 0;
1237         struct mm_struct *mm;
1238
1239         task_lock(task);
1240         mm = task->mm;
1241         if (mm)
1242                 dumpable = get_dumpable(mm);
1243         task_unlock(task);
1244         if(dumpable == 1)
1245                 return 1;
1246         return 0;
1247 }
1248
1249
1250 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1251 {
1252         struct inode * inode;
1253         struct proc_inode *ei;
1254
1255         /* We need a new inode */
1256
1257         inode = new_inode(sb);
1258         if (!inode)
1259                 goto out;
1260
1261         /* Common stuff */
1262         ei = PROC_I(inode);
1263         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1264         inode->i_op = &proc_def_inode_operations;
1265
1266         /*
1267          * grab the reference to task.
1268          */
1269         ei->pid = get_task_pid(task, PIDTYPE_PID);
1270         if (!ei->pid)
1271                 goto out_unlock;
1272
1273         inode->i_uid = 0;
1274         inode->i_gid = 0;
1275         if (task_dumpable(task)) {
1276                 inode->i_uid = task->euid;
1277                 inode->i_gid = task->egid;
1278         }
1279         security_task_to_inode(task, inode);
1280
1281 out:
1282         return inode;
1283
1284 out_unlock:
1285         iput(inode);
1286         return NULL;
1287 }
1288
1289 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1290 {
1291         struct inode *inode = dentry->d_inode;
1292         struct task_struct *task;
1293         generic_fillattr(inode, stat);
1294
1295         rcu_read_lock();
1296         stat->uid = 0;
1297         stat->gid = 0;
1298         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1299         if (task) {
1300                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1301                     task_dumpable(task)) {
1302                         stat->uid = task->euid;
1303                         stat->gid = task->egid;
1304                 }
1305         }
1306         rcu_read_unlock();
1307         return 0;
1308 }
1309
1310 /* dentry stuff */
1311
1312 /*
1313  *      Exceptional case: normally we are not allowed to unhash a busy
1314  * directory. In this case, however, we can do it - no aliasing problems
1315  * due to the way we treat inodes.
1316  *
1317  * Rewrite the inode's ownerships here because the owning task may have
1318  * performed a setuid(), etc.
1319  *
1320  * Before the /proc/pid/status file was created the only way to read
1321  * the effective uid of a /process was to stat /proc/pid.  Reading
1322  * /proc/pid/status is slow enough that procps and other packages
1323  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1324  * made this apply to all per process world readable and executable
1325  * directories.
1326  */
1327 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1328 {
1329         struct inode *inode = dentry->d_inode;
1330         struct task_struct *task = get_proc_task(inode);
1331         if (task) {
1332                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1333                     task_dumpable(task)) {
1334                         inode->i_uid = task->euid;
1335                         inode->i_gid = task->egid;
1336                 } else {
1337                         inode->i_uid = 0;
1338                         inode->i_gid = 0;
1339                 }
1340                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1341                 security_task_to_inode(task, inode);
1342                 put_task_struct(task);
1343                 return 1;
1344         }
1345         d_drop(dentry);
1346         return 0;
1347 }
1348
1349 static int pid_delete_dentry(struct dentry * dentry)
1350 {
1351         /* Is the task we represent dead?
1352          * If so, then don't put the dentry on the lru list,
1353          * kill it immediately.
1354          */
1355         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1356 }
1357
1358 static struct dentry_operations pid_dentry_operations =
1359 {
1360         .d_revalidate   = pid_revalidate,
1361         .d_delete       = pid_delete_dentry,
1362 };
1363
1364 /* Lookups */
1365
1366 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1367                                 struct task_struct *, const void *);
1368
1369 /*
1370  * Fill a directory entry.
1371  *
1372  * If possible create the dcache entry and derive our inode number and
1373  * file type from dcache entry.
1374  *
1375  * Since all of the proc inode numbers are dynamically generated, the inode
1376  * numbers do not exist until the inode is cache.  This means creating the
1377  * the dcache entry in readdir is necessary to keep the inode numbers
1378  * reported by readdir in sync with the inode numbers reported
1379  * by stat.
1380  */
1381 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1382         char *name, int len,
1383         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1384 {
1385         struct dentry *child, *dir = filp->f_path.dentry;
1386         struct inode *inode;
1387         struct qstr qname;
1388         ino_t ino = 0;
1389         unsigned type = DT_UNKNOWN;
1390
1391         qname.name = name;
1392         qname.len  = len;
1393         qname.hash = full_name_hash(name, len);
1394
1395         child = d_lookup(dir, &qname);
1396         if (!child) {
1397                 struct dentry *new;
1398                 new = d_alloc(dir, &qname);
1399                 if (new) {
1400                         child = instantiate(dir->d_inode, new, task, ptr);
1401                         if (child)
1402                                 dput(new);
1403                         else
1404                                 child = new;
1405                 }
1406         }
1407         if (!child || IS_ERR(child) || !child->d_inode)
1408                 goto end_instantiate;
1409         inode = child->d_inode;
1410         if (inode) {
1411                 ino = inode->i_ino;
1412                 type = inode->i_mode >> 12;
1413         }
1414         dput(child);
1415 end_instantiate:
1416         if (!ino)
1417                 ino = find_inode_number(dir, &qname);
1418         if (!ino)
1419                 ino = 1;
1420         return filldir(dirent, name, len, filp->f_pos, ino, type);
1421 }
1422
1423 static unsigned name_to_int(struct dentry *dentry)
1424 {
1425         const char *name = dentry->d_name.name;
1426         int len = dentry->d_name.len;
1427         unsigned n = 0;
1428
1429         if (len > 1 && *name == '0')
1430                 goto out;
1431         while (len-- > 0) {
1432                 unsigned c = *name++ - '0';
1433                 if (c > 9)
1434                         goto out;
1435                 if (n >= (~0U-9)/10)
1436                         goto out;
1437                 n *= 10;
1438                 n += c;
1439         }
1440         return n;
1441 out:
1442         return ~0U;
1443 }
1444
1445 #define PROC_FDINFO_MAX 64
1446
1447 static int proc_fd_info(struct inode *inode, struct dentry **dentry,
1448                         struct vfsmount **mnt, char *info)
1449 {
1450         struct task_struct *task = get_proc_task(inode);
1451         struct files_struct *files = NULL;
1452         struct file *file;
1453         int fd = proc_fd(inode);
1454
1455         if (task) {
1456                 files = get_files_struct(task);
1457                 put_task_struct(task);
1458         }
1459         if (files) {
1460                 /*
1461                  * We are not taking a ref to the file structure, so we must
1462                  * hold ->file_lock.
1463                  */
1464                 spin_lock(&files->file_lock);
1465                 file = fcheck_files(files, fd);
1466                 if (file) {
1467                         if (mnt)
1468                                 *mnt = mntget(file->f_path.mnt);
1469                         if (dentry)
1470                                 *dentry = dget(file->f_path.dentry);
1471                         if (info)
1472                                 snprintf(info, PROC_FDINFO_MAX,
1473                                          "pos:\t%lli\n"
1474                                          "flags:\t0%o\n",
1475                                          (long long) file->f_pos,
1476                                          file->f_flags);
1477                         spin_unlock(&files->file_lock);
1478                         put_files_struct(files);
1479                         return 0;
1480                 }
1481                 spin_unlock(&files->file_lock);
1482                 put_files_struct(files);
1483         }
1484         return -ENOENT;
1485 }
1486
1487 static int proc_fd_link(struct inode *inode, struct dentry **dentry,
1488                         struct vfsmount **mnt)
1489 {
1490         return proc_fd_info(inode, dentry, mnt, NULL);
1491 }
1492
1493 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1494 {
1495         struct inode *inode = dentry->d_inode;
1496         struct task_struct *task = get_proc_task(inode);
1497         int fd = proc_fd(inode);
1498         struct files_struct *files;
1499
1500         if (task) {
1501                 files = get_files_struct(task);
1502                 if (files) {
1503                         rcu_read_lock();
1504                         if (fcheck_files(files, fd)) {
1505                                 rcu_read_unlock();
1506                                 put_files_struct(files);
1507                                 if (task_dumpable(task)) {
1508                                         inode->i_uid = task->euid;
1509                                         inode->i_gid = task->egid;
1510                                 } else {
1511                                         inode->i_uid = 0;
1512                                         inode->i_gid = 0;
1513                                 }
1514                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1515                                 security_task_to_inode(task, inode);
1516                                 put_task_struct(task);
1517                                 return 1;
1518                         }
1519                         rcu_read_unlock();
1520                         put_files_struct(files);
1521                 }
1522                 put_task_struct(task);
1523         }
1524         d_drop(dentry);
1525         return 0;
1526 }
1527
1528 static struct dentry_operations tid_fd_dentry_operations =
1529 {
1530         .d_revalidate   = tid_fd_revalidate,
1531         .d_delete       = pid_delete_dentry,
1532 };
1533
1534 static struct dentry *proc_fd_instantiate(struct inode *dir,
1535         struct dentry *dentry, struct task_struct *task, const void *ptr)
1536 {
1537         unsigned fd = *(const unsigned *)ptr;
1538         struct file *file;
1539         struct files_struct *files;
1540         struct inode *inode;
1541         struct proc_inode *ei;
1542         struct dentry *error = ERR_PTR(-ENOENT);
1543
1544         inode = proc_pid_make_inode(dir->i_sb, task);
1545         if (!inode)
1546                 goto out;
1547         ei = PROC_I(inode);
1548         ei->fd = fd;
1549         files = get_files_struct(task);
1550         if (!files)
1551                 goto out_iput;
1552         inode->i_mode = S_IFLNK;
1553
1554         /*
1555          * We are not taking a ref to the file structure, so we must
1556          * hold ->file_lock.
1557          */
1558         spin_lock(&files->file_lock);
1559         file = fcheck_files(files, fd);
1560         if (!file)
1561                 goto out_unlock;
1562         if (file->f_mode & 1)
1563                 inode->i_mode |= S_IRUSR | S_IXUSR;
1564         if (file->f_mode & 2)
1565                 inode->i_mode |= S_IWUSR | S_IXUSR;
1566         spin_unlock(&files->file_lock);
1567         put_files_struct(files);
1568
1569         inode->i_op = &proc_pid_link_inode_operations;
1570         inode->i_size = 64;
1571         ei->op.proc_get_link = proc_fd_link;
1572         dentry->d_op = &tid_fd_dentry_operations;
1573         d_add(dentry, inode);
1574         /* Close the race of the process dying before we return the dentry */
1575         if (tid_fd_revalidate(dentry, NULL))
1576                 error = NULL;
1577
1578  out:
1579         return error;
1580 out_unlock:
1581         spin_unlock(&files->file_lock);
1582         put_files_struct(files);
1583 out_iput:
1584         iput(inode);
1585         goto out;
1586 }
1587
1588 static struct dentry *proc_lookupfd_common(struct inode *dir,
1589                                            struct dentry *dentry,
1590                                            instantiate_t instantiate)
1591 {
1592         struct task_struct *task = get_proc_task(dir);
1593         unsigned fd = name_to_int(dentry);
1594         struct dentry *result = ERR_PTR(-ENOENT);
1595
1596         if (!task)
1597                 goto out_no_task;
1598         if (fd == ~0U)
1599                 goto out;
1600
1601         result = instantiate(dir, dentry, task, &fd);
1602 out:
1603         put_task_struct(task);
1604 out_no_task:
1605         return result;
1606 }
1607
1608 static int proc_readfd_common(struct file * filp, void * dirent,
1609                               filldir_t filldir, instantiate_t instantiate)
1610 {
1611         struct dentry *dentry = filp->f_path.dentry;
1612         struct inode *inode = dentry->d_inode;
1613         struct task_struct *p = get_proc_task(inode);
1614         unsigned int fd, ino;
1615         int retval;
1616         struct files_struct * files;
1617         struct fdtable *fdt;
1618
1619         retval = -ENOENT;
1620         if (!p)
1621                 goto out_no_task;
1622         retval = 0;
1623
1624         fd = filp->f_pos;
1625         switch (fd) {
1626                 case 0:
1627                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1628                                 goto out;
1629                         filp->f_pos++;
1630                 case 1:
1631                         ino = parent_ino(dentry);
1632                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1633                                 goto out;
1634                         filp->f_pos++;
1635                 default:
1636                         files = get_files_struct(p);
1637                         if (!files)
1638                                 goto out;
1639                         rcu_read_lock();
1640                         fdt = files_fdtable(files);
1641                         for (fd = filp->f_pos-2;
1642                              fd < fdt->max_fds;
1643                              fd++, filp->f_pos++) {
1644                                 char name[PROC_NUMBUF];
1645                                 int len;
1646
1647                                 if (!fcheck_files(files, fd))
1648                                         continue;
1649                                 rcu_read_unlock();
1650
1651                                 len = snprintf(name, sizeof(name), "%d", fd);
1652                                 if (proc_fill_cache(filp, dirent, filldir,
1653                                                     name, len, instantiate,
1654                                                     p, &fd) < 0) {
1655                                         rcu_read_lock();
1656                                         break;
1657                                 }
1658                                 rcu_read_lock();
1659                         }
1660                         rcu_read_unlock();
1661                         put_files_struct(files);
1662         }
1663 out:
1664         put_task_struct(p);
1665 out_no_task:
1666         return retval;
1667 }
1668
1669 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1670                                     struct nameidata *nd)
1671 {
1672         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1673 }
1674
1675 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1676 {
1677         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1678 }
1679
1680 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1681                                       size_t len, loff_t *ppos)
1682 {
1683         char tmp[PROC_FDINFO_MAX];
1684         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp);
1685         if (!err)
1686                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1687         return err;
1688 }
1689
1690 static const struct file_operations proc_fdinfo_file_operations = {
1691         .open           = nonseekable_open,
1692         .read           = proc_fdinfo_read,
1693 };
1694
1695 static const struct file_operations proc_fd_operations = {
1696         .read           = generic_read_dir,
1697         .readdir        = proc_readfd,
1698 };
1699
1700 /*
1701  * /proc/pid/fd needs a special permission handler so that a process can still
1702  * access /proc/self/fd after it has executed a setuid().
1703  */
1704 static int proc_fd_permission(struct inode *inode, int mask,
1705                                 struct nameidata *nd)
1706 {
1707         int rv;
1708
1709         rv = generic_permission(inode, mask, NULL);
1710         if (rv == 0)
1711                 return 0;
1712         if (task_pid(current) == proc_pid(inode))
1713                 rv = 0;
1714         return rv;
1715 }
1716
1717 /*
1718  * proc directories can do almost nothing..
1719  */
1720 static const struct inode_operations proc_fd_inode_operations = {
1721         .lookup         = proc_lookupfd,
1722         .permission     = proc_fd_permission,
1723         .setattr        = proc_setattr,
1724 };
1725
1726 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1727         struct dentry *dentry, struct task_struct *task, const void *ptr)
1728 {
1729         unsigned fd = *(unsigned *)ptr;
1730         struct inode *inode;
1731         struct proc_inode *ei;
1732         struct dentry *error = ERR_PTR(-ENOENT);
1733
1734         inode = proc_pid_make_inode(dir->i_sb, task);
1735         if (!inode)
1736                 goto out;
1737         ei = PROC_I(inode);
1738         ei->fd = fd;
1739         inode->i_mode = S_IFREG | S_IRUSR;
1740         inode->i_fop = &proc_fdinfo_file_operations;
1741         dentry->d_op = &tid_fd_dentry_operations;
1742         d_add(dentry, inode);
1743         /* Close the race of the process dying before we return the dentry */
1744         if (tid_fd_revalidate(dentry, NULL))
1745                 error = NULL;
1746
1747  out:
1748         return error;
1749 }
1750
1751 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1752                                         struct dentry *dentry,
1753                                         struct nameidata *nd)
1754 {
1755         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1756 }
1757
1758 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1759 {
1760         return proc_readfd_common(filp, dirent, filldir,
1761                                   proc_fdinfo_instantiate);
1762 }
1763
1764 static const struct file_operations proc_fdinfo_operations = {
1765         .read           = generic_read_dir,
1766         .readdir        = proc_readfdinfo,
1767 };
1768
1769 /*
1770  * proc directories can do almost nothing..
1771  */
1772 static const struct inode_operations proc_fdinfo_inode_operations = {
1773         .lookup         = proc_lookupfdinfo,
1774         .setattr        = proc_setattr,
1775 };
1776
1777
1778 static struct dentry *proc_pident_instantiate(struct inode *dir,
1779         struct dentry *dentry, struct task_struct *task, const void *ptr)
1780 {
1781         const struct pid_entry *p = ptr;
1782         struct inode *inode;
1783         struct proc_inode *ei;
1784         struct dentry *error = ERR_PTR(-EINVAL);
1785
1786         inode = proc_pid_make_inode(dir->i_sb, task);
1787         if (!inode)
1788                 goto out;
1789
1790         ei = PROC_I(inode);
1791         inode->i_mode = p->mode;
1792         if (S_ISDIR(inode->i_mode))
1793                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1794         if (p->iop)
1795                 inode->i_op = p->iop;
1796         if (p->fop)
1797                 inode->i_fop = p->fop;
1798         ei->op = p->op;
1799         dentry->d_op = &pid_dentry_operations;
1800         d_add(dentry, inode);
1801         /* Close the race of the process dying before we return the dentry */
1802         if (pid_revalidate(dentry, NULL))
1803                 error = NULL;
1804 out:
1805         return error;
1806 }
1807
1808 static struct dentry *proc_pident_lookup(struct inode *dir,
1809                                          struct dentry *dentry,
1810                                          const struct pid_entry *ents,
1811                                          unsigned int nents)
1812 {
1813         struct inode *inode;
1814         struct dentry *error;
1815         struct task_struct *task = get_proc_task(dir);
1816         const struct pid_entry *p, *last;
1817
1818         error = ERR_PTR(-ENOENT);
1819         inode = NULL;
1820
1821         if (!task)
1822                 goto out_no_task;
1823
1824         /*
1825          * Yes, it does not scale. And it should not. Don't add
1826          * new entries into /proc/<tgid>/ without very good reasons.
1827          */
1828         last = &ents[nents - 1];
1829         for (p = ents; p <= last; p++) {
1830                 if (p->len != dentry->d_name.len)
1831                         continue;
1832                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1833                         break;
1834         }
1835         if (p > last)
1836                 goto out;
1837
1838         error = proc_pident_instantiate(dir, dentry, task, p);
1839 out:
1840         put_task_struct(task);
1841 out_no_task:
1842         return error;
1843 }
1844
1845 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1846         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1847 {
1848         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1849                                 proc_pident_instantiate, task, p);
1850 }
1851
1852 static int proc_pident_readdir(struct file *filp,
1853                 void *dirent, filldir_t filldir,
1854                 const struct pid_entry *ents, unsigned int nents)
1855 {
1856         int i;
1857         struct dentry *dentry = filp->f_path.dentry;
1858         struct inode *inode = dentry->d_inode;
1859         struct task_struct *task = get_proc_task(inode);
1860         const struct pid_entry *p, *last;
1861         ino_t ino;
1862         int ret;
1863
1864         ret = -ENOENT;
1865         if (!task)
1866                 goto out_no_task;
1867
1868         ret = 0;
1869         i = filp->f_pos;
1870         switch (i) {
1871         case 0:
1872                 ino = inode->i_ino;
1873                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1874                         goto out;
1875                 i++;
1876                 filp->f_pos++;
1877                 /* fall through */
1878         case 1:
1879                 ino = parent_ino(dentry);
1880                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1881                         goto out;
1882                 i++;
1883                 filp->f_pos++;
1884                 /* fall through */
1885         default:
1886                 i -= 2;
1887                 if (i >= nents) {
1888                         ret = 1;
1889                         goto out;
1890                 }
1891                 p = ents + i;
1892                 last = &ents[nents - 1];
1893                 while (p <= last) {
1894                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1895                                 goto out;
1896                         filp->f_pos++;
1897                         p++;
1898                 }
1899         }
1900
1901         ret = 1;
1902 out:
1903         put_task_struct(task);
1904 out_no_task:
1905         return ret;
1906 }
1907
1908 #ifdef CONFIG_SECURITY
1909 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1910                                   size_t count, loff_t *ppos)
1911 {
1912         struct inode * inode = file->f_path.dentry->d_inode;
1913         char *p = NULL;
1914         ssize_t length;
1915         struct task_struct *task = get_proc_task(inode);
1916
1917         if (!task)
1918                 return -ESRCH;
1919
1920         length = security_getprocattr(task,
1921                                       (char*)file->f_path.dentry->d_name.name,
1922                                       &p);
1923         put_task_struct(task);
1924         if (length > 0)
1925                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1926         kfree(p);
1927         return length;
1928 }
1929
1930 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1931                                    size_t count, loff_t *ppos)
1932 {
1933         struct inode * inode = file->f_path.dentry->d_inode;
1934         char *page;
1935         ssize_t length;
1936         struct task_struct *task = get_proc_task(inode);
1937
1938         length = -ESRCH;
1939         if (!task)
1940                 goto out_no_task;
1941         if (count > PAGE_SIZE)
1942                 count = PAGE_SIZE;
1943
1944         /* No partial writes. */
1945         length = -EINVAL;
1946         if (*ppos != 0)
1947                 goto out;
1948
1949         length = -ENOMEM;
1950         page = (char*)__get_free_page(GFP_TEMPORARY);
1951         if (!page)
1952                 goto out;
1953
1954         length = -EFAULT;
1955         if (copy_from_user(page, buf, count))
1956                 goto out_free;
1957
1958         length = security_setprocattr(task,
1959                                       (char*)file->f_path.dentry->d_name.name,
1960                                       (void*)page, count);
1961 out_free:
1962         free_page((unsigned long) page);
1963 out:
1964         put_task_struct(task);
1965 out_no_task:
1966         return length;
1967 }
1968
1969 static const struct file_operations proc_pid_attr_operations = {
1970         .read           = proc_pid_attr_read,
1971         .write          = proc_pid_attr_write,
1972 };
1973
1974 static const struct pid_entry attr_dir_stuff[] = {
1975         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1976         REG("prev",       S_IRUGO,         pid_attr),
1977         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1978         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1979         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1980         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1981 };
1982
1983 static int proc_attr_dir_readdir(struct file * filp,
1984                              void * dirent, filldir_t filldir)
1985 {
1986         return proc_pident_readdir(filp,dirent,filldir,
1987                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1988 }
1989
1990 static const struct file_operations proc_attr_dir_operations = {
1991         .read           = generic_read_dir,
1992         .readdir        = proc_attr_dir_readdir,
1993 };
1994
1995 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1996                                 struct dentry *dentry, struct nameidata *nd)
1997 {
1998         return proc_pident_lookup(dir, dentry,
1999                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2000 }
2001
2002 static const struct inode_operations proc_attr_dir_inode_operations = {
2003         .lookup         = proc_attr_dir_lookup,
2004         .getattr        = pid_getattr,
2005         .setattr        = proc_setattr,
2006 };
2007
2008 #endif
2009
2010 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2011 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2012                                          size_t count, loff_t *ppos)
2013 {
2014         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2015         struct mm_struct *mm;
2016         char buffer[PROC_NUMBUF];
2017         size_t len;
2018         int ret;
2019
2020         if (!task)
2021                 return -ESRCH;
2022
2023         ret = 0;
2024         mm = get_task_mm(task);
2025         if (mm) {
2026                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2027                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2028                                 MMF_DUMP_FILTER_SHIFT));
2029                 mmput(mm);
2030                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2031         }
2032
2033         put_task_struct(task);
2034
2035         return ret;
2036 }
2037
2038 static ssize_t proc_coredump_filter_write(struct file *file,
2039                                           const char __user *buf,
2040                                           size_t count,
2041                                           loff_t *ppos)
2042 {
2043         struct task_struct *task;
2044         struct mm_struct *mm;
2045         char buffer[PROC_NUMBUF], *end;
2046         unsigned int val;
2047         int ret;
2048         int i;
2049         unsigned long mask;
2050
2051         ret = -EFAULT;
2052         memset(buffer, 0, sizeof(buffer));
2053         if (count > sizeof(buffer) - 1)
2054                 count = sizeof(buffer) - 1;
2055         if (copy_from_user(buffer, buf, count))
2056                 goto out_no_task;
2057
2058         ret = -EINVAL;
2059         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2060         if (*end == '\n')
2061                 end++;
2062         if (end - buffer == 0)
2063                 goto out_no_task;
2064
2065         ret = -ESRCH;
2066         task = get_proc_task(file->f_dentry->d_inode);
2067         if (!task)
2068                 goto out_no_task;
2069
2070         ret = end - buffer;
2071         mm = get_task_mm(task);
2072         if (!mm)
2073                 goto out_no_mm;
2074
2075         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2076                 if (val & mask)
2077                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2078                 else
2079                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2080         }
2081
2082         mmput(mm);
2083  out_no_mm:
2084         put_task_struct(task);
2085  out_no_task:
2086         return ret;
2087 }
2088
2089 static const struct file_operations proc_coredump_filter_operations = {
2090         .read           = proc_coredump_filter_read,
2091         .write          = proc_coredump_filter_write,
2092 };
2093 #endif
2094
2095 /*
2096  * /proc/self:
2097  */
2098 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2099                               int buflen)
2100 {
2101         char tmp[PROC_NUMBUF];
2102         sprintf(tmp, "%d", task_tgid_vnr(current));
2103         return vfs_readlink(dentry,buffer,buflen,tmp);
2104 }
2105
2106 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2107 {
2108         char tmp[PROC_NUMBUF];
2109         sprintf(tmp, "%d", task_tgid_vnr(current));
2110         return ERR_PTR(vfs_follow_link(nd,tmp));
2111 }
2112
2113 static const struct inode_operations proc_self_inode_operations = {
2114         .readlink       = proc_self_readlink,
2115         .follow_link    = proc_self_follow_link,
2116 };
2117
2118 /*
2119  * proc base
2120  *
2121  * These are the directory entries in the root directory of /proc
2122  * that properly belong to the /proc filesystem, as they describe
2123  * describe something that is process related.
2124  */
2125 static const struct pid_entry proc_base_stuff[] = {
2126         NOD("self", S_IFLNK|S_IRWXUGO,
2127                 &proc_self_inode_operations, NULL, {}),
2128 };
2129
2130 /*
2131  *      Exceptional case: normally we are not allowed to unhash a busy
2132  * directory. In this case, however, we can do it - no aliasing problems
2133  * due to the way we treat inodes.
2134  */
2135 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2136 {
2137         struct inode *inode = dentry->d_inode;
2138         struct task_struct *task = get_proc_task(inode);
2139         if (task) {
2140                 put_task_struct(task);
2141                 return 1;
2142         }
2143         d_drop(dentry);
2144         return 0;
2145 }
2146
2147 static struct dentry_operations proc_base_dentry_operations =
2148 {
2149         .d_revalidate   = proc_base_revalidate,
2150         .d_delete       = pid_delete_dentry,
2151 };
2152
2153 static struct dentry *proc_base_instantiate(struct inode *dir,
2154         struct dentry *dentry, struct task_struct *task, const void *ptr)
2155 {
2156         const struct pid_entry *p = ptr;
2157         struct inode *inode;
2158         struct proc_inode *ei;
2159         struct dentry *error = ERR_PTR(-EINVAL);
2160
2161         /* Allocate the inode */
2162         error = ERR_PTR(-ENOMEM);
2163         inode = new_inode(dir->i_sb);
2164         if (!inode)
2165                 goto out;
2166
2167         /* Initialize the inode */
2168         ei = PROC_I(inode);
2169         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2170
2171         /*
2172          * grab the reference to the task.
2173          */
2174         ei->pid = get_task_pid(task, PIDTYPE_PID);
2175         if (!ei->pid)
2176                 goto out_iput;
2177
2178         inode->i_uid = 0;
2179         inode->i_gid = 0;
2180         inode->i_mode = p->mode;
2181         if (S_ISDIR(inode->i_mode))
2182                 inode->i_nlink = 2;
2183         if (S_ISLNK(inode->i_mode))
2184                 inode->i_size = 64;
2185         if (p->iop)
2186                 inode->i_op = p->iop;
2187         if (p->fop)
2188                 inode->i_fop = p->fop;
2189         ei->op = p->op;
2190         dentry->d_op = &proc_base_dentry_operations;
2191         d_add(dentry, inode);
2192         error = NULL;
2193 out:
2194         return error;
2195 out_iput:
2196         iput(inode);
2197         goto out;
2198 }
2199
2200 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2201 {
2202         struct dentry *error;
2203         struct task_struct *task = get_proc_task(dir);
2204         const struct pid_entry *p, *last;
2205
2206         error = ERR_PTR(-ENOENT);
2207
2208         if (!task)
2209                 goto out_no_task;
2210
2211         /* Lookup the directory entry */
2212         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2213         for (p = proc_base_stuff; p <= last; p++) {
2214                 if (p->len != dentry->d_name.len)
2215                         continue;
2216                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2217                         break;
2218         }
2219         if (p > last)
2220                 goto out;
2221
2222         error = proc_base_instantiate(dir, dentry, task, p);
2223
2224 out:
2225         put_task_struct(task);
2226 out_no_task:
2227         return error;
2228 }
2229
2230 static int proc_base_fill_cache(struct file *filp, void *dirent,
2231         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2232 {
2233         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2234                                 proc_base_instantiate, task, p);
2235 }
2236
2237 #ifdef CONFIG_TASK_IO_ACCOUNTING
2238 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2239 {
2240         return sprintf(buffer,
2241 #ifdef CONFIG_TASK_XACCT
2242                         "rchar: %llu\n"
2243                         "wchar: %llu\n"
2244                         "syscr: %llu\n"
2245                         "syscw: %llu\n"
2246 #endif
2247                         "read_bytes: %llu\n"
2248                         "write_bytes: %llu\n"
2249                         "cancelled_write_bytes: %llu\n",
2250 #ifdef CONFIG_TASK_XACCT
2251                         (unsigned long long)task->rchar,
2252                         (unsigned long long)task->wchar,
2253                         (unsigned long long)task->syscr,
2254                         (unsigned long long)task->syscw,
2255 #endif
2256                         (unsigned long long)task->ioac.read_bytes,
2257                         (unsigned long long)task->ioac.write_bytes,
2258                         (unsigned long long)task->ioac.cancelled_write_bytes);
2259 }
2260 #endif
2261
2262 /*
2263  * Thread groups
2264  */
2265 static const struct file_operations proc_task_operations;
2266 static const struct inode_operations proc_task_inode_operations;
2267
2268 static const struct pid_entry tgid_base_stuff[] = {
2269         DIR("task",       S_IRUGO|S_IXUGO, task),
2270         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2271         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2272         REG("environ",    S_IRUSR, environ),
2273         INF("auxv",       S_IRUSR, pid_auxv),
2274         INF("status",     S_IRUGO, pid_status),
2275         INF("limits",     S_IRUSR, pid_limits),
2276 #ifdef CONFIG_SCHED_DEBUG
2277         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2278 #endif
2279         INF("cmdline",    S_IRUGO, pid_cmdline),
2280         INF("stat",       S_IRUGO, tgid_stat),
2281         INF("statm",      S_IRUGO, pid_statm),
2282         REG("maps",       S_IRUGO, maps),
2283 #ifdef CONFIG_NUMA
2284         REG("numa_maps",  S_IRUGO, numa_maps),
2285 #endif
2286         REG("mem",        S_IRUSR|S_IWUSR, mem),
2287         LNK("cwd",        cwd),
2288         LNK("root",       root),
2289         LNK("exe",        exe),
2290         REG("mounts",     S_IRUGO, mounts),
2291         REG("mountstats", S_IRUSR, mountstats),
2292 #ifdef CONFIG_MMU
2293         REG("clear_refs", S_IWUSR, clear_refs),
2294         REG("smaps",      S_IRUGO, smaps),
2295 #endif
2296 #ifdef CONFIG_SECURITY
2297         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2298 #endif
2299 #ifdef CONFIG_KALLSYMS
2300         INF("wchan",      S_IRUGO, pid_wchan),
2301 #endif
2302 #ifdef CONFIG_SCHEDSTATS
2303         INF("schedstat",  S_IRUGO, pid_schedstat),
2304 #endif
2305 #ifdef CONFIG_LATENCYTOP
2306         REG("latency",  S_IRUGO, lstats),
2307 #endif
2308 #ifdef CONFIG_PROC_PID_CPUSET
2309         REG("cpuset",     S_IRUGO, cpuset),
2310 #endif
2311 #ifdef CONFIG_CGROUPS
2312         REG("cgroup",  S_IRUGO, cgroup),
2313 #endif
2314         INF("oom_score",  S_IRUGO, oom_score),
2315         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2316 #ifdef CONFIG_AUDITSYSCALL
2317         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2318 #endif
2319 #ifdef CONFIG_FAULT_INJECTION
2320         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2321 #endif
2322 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2323         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2324 #endif
2325 #ifdef CONFIG_TASK_IO_ACCOUNTING
2326         INF("io",       S_IRUGO, pid_io_accounting),
2327 #endif
2328 };
2329
2330 static int proc_tgid_base_readdir(struct file * filp,
2331                              void * dirent, filldir_t filldir)
2332 {
2333         return proc_pident_readdir(filp,dirent,filldir,
2334                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2335 }
2336
2337 static const struct file_operations proc_tgid_base_operations = {
2338         .read           = generic_read_dir,
2339         .readdir        = proc_tgid_base_readdir,
2340 };
2341
2342 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2343         return proc_pident_lookup(dir, dentry,
2344                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2345 }
2346
2347 static const struct inode_operations proc_tgid_base_inode_operations = {
2348         .lookup         = proc_tgid_base_lookup,
2349         .getattr        = pid_getattr,
2350         .setattr        = proc_setattr,
2351 };
2352
2353 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2354 {
2355         struct dentry *dentry, *leader, *dir;
2356         char buf[PROC_NUMBUF];
2357         struct qstr name;
2358
2359         name.name = buf;
2360         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2361         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2362         if (dentry) {
2363                 shrink_dcache_parent(dentry);
2364                 d_drop(dentry);
2365                 dput(dentry);
2366         }
2367
2368         if (tgid == 0)
2369                 goto out;
2370
2371         name.name = buf;
2372         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2373         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2374         if (!leader)
2375                 goto out;
2376
2377         name.name = "task";
2378         name.len = strlen(name.name);
2379         dir = d_hash_and_lookup(leader, &name);
2380         if (!dir)
2381                 goto out_put_leader;
2382
2383         name.name = buf;
2384         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2385         dentry = d_hash_and_lookup(dir, &name);
2386         if (dentry) {
2387                 shrink_dcache_parent(dentry);
2388                 d_drop(dentry);
2389                 dput(dentry);
2390         }
2391
2392         dput(dir);
2393 out_put_leader:
2394         dput(leader);
2395 out:
2396         return;
2397 }
2398
2399 /**
2400  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2401  * @task: task that should be flushed.
2402  *
2403  * When flushing dentries from proc, one needs to flush them from global
2404  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2405  * in. This call is supposed to do all of this job.
2406  *
2407  * Looks in the dcache for
2408  * /proc/@pid
2409  * /proc/@tgid/task/@pid
2410  * if either directory is present flushes it and all of it'ts children
2411  * from the dcache.
2412  *
2413  * It is safe and reasonable to cache /proc entries for a task until
2414  * that task exits.  After that they just clog up the dcache with
2415  * useless entries, possibly causing useful dcache entries to be
2416  * flushed instead.  This routine is proved to flush those useless
2417  * dcache entries at process exit time.
2418  *
2419  * NOTE: This routine is just an optimization so it does not guarantee
2420  *       that no dcache entries will exist at process exit time it
2421  *       just makes it very unlikely that any will persist.
2422  */
2423
2424 void proc_flush_task(struct task_struct *task)
2425 {
2426         int i;
2427         struct pid *pid, *tgid = NULL;
2428         struct upid *upid;
2429
2430         pid = task_pid(task);
2431         if (thread_group_leader(task))
2432                 tgid = task_tgid(task);
2433
2434         for (i = 0; i <= pid->level; i++) {
2435                 upid = &pid->numbers[i];
2436                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2437                         tgid ? tgid->numbers[i].nr : 0);
2438         }
2439
2440         upid = &pid->numbers[pid->level];
2441         if (upid->nr == 1)
2442                 pid_ns_release_proc(upid->ns);
2443 }
2444
2445 static struct dentry *proc_pid_instantiate(struct inode *dir,
2446                                            struct dentry * dentry,
2447                                            struct task_struct *task, const void *ptr)
2448 {
2449         struct dentry *error = ERR_PTR(-ENOENT);
2450         struct inode *inode;
2451
2452         inode = proc_pid_make_inode(dir->i_sb, task);
2453         if (!inode)
2454                 goto out;
2455
2456         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2457         inode->i_op = &proc_tgid_base_inode_operations;
2458         inode->i_fop = &proc_tgid_base_operations;
2459         inode->i_flags|=S_IMMUTABLE;
2460         inode->i_nlink = 5;
2461 #ifdef CONFIG_SECURITY
2462         inode->i_nlink += 1;
2463 #endif
2464
2465         dentry->d_op = &pid_dentry_operations;
2466
2467         d_add(dentry, inode);
2468         /* Close the race of the process dying before we return the dentry */
2469         if (pid_revalidate(dentry, NULL))
2470                 error = NULL;
2471 out:
2472         return error;
2473 }
2474
2475 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2476 {
2477         struct dentry *result = ERR_PTR(-ENOENT);
2478         struct task_struct *task;
2479         unsigned tgid;
2480         struct pid_namespace *ns;
2481
2482         result = proc_base_lookup(dir, dentry);
2483         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2484                 goto out;
2485
2486         tgid = name_to_int(dentry);
2487         if (tgid == ~0U)
2488                 goto out;
2489
2490         ns = dentry->d_sb->s_fs_info;
2491         rcu_read_lock();
2492         task = find_task_by_pid_ns(tgid, ns);
2493         if (task)
2494                 get_task_struct(task);
2495         rcu_read_unlock();
2496         if (!task)
2497                 goto out;
2498
2499         result = proc_pid_instantiate(dir, dentry, task, NULL);
2500         put_task_struct(task);
2501 out:
2502         return result;
2503 }
2504
2505 /*
2506  * Find the first task with tgid >= tgid
2507  *
2508  */
2509 struct tgid_iter {
2510         unsigned int tgid;
2511         struct task_struct *task;
2512 };
2513 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2514 {
2515         struct pid *pid;
2516
2517         if (iter.task)
2518                 put_task_struct(iter.task);
2519         rcu_read_lock();
2520 retry:
2521         iter.task = NULL;
2522         pid = find_ge_pid(iter.tgid, ns);
2523         if (pid) {
2524                 iter.tgid = pid_nr_ns(pid, ns);
2525                 iter.task = pid_task(pid, PIDTYPE_PID);
2526                 /* What we to know is if the pid we have find is the
2527                  * pid of a thread_group_leader.  Testing for task
2528                  * being a thread_group_leader is the obvious thing
2529                  * todo but there is a window when it fails, due to
2530                  * the pid transfer logic in de_thread.
2531                  *
2532                  * So we perform the straight forward test of seeing
2533                  * if the pid we have found is the pid of a thread
2534                  * group leader, and don't worry if the task we have
2535                  * found doesn't happen to be a thread group leader.
2536                  * As we don't care in the case of readdir.
2537                  */
2538                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2539                         iter.tgid += 1;
2540                         goto retry;
2541                 }
2542                 get_task_struct(iter.task);
2543         }
2544         rcu_read_unlock();
2545         return iter;
2546 }
2547
2548 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2549
2550 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2551         struct tgid_iter iter)
2552 {
2553         char name[PROC_NUMBUF];
2554         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2555         return proc_fill_cache(filp, dirent, filldir, name, len,
2556                                 proc_pid_instantiate, iter.task, NULL);
2557 }
2558
2559 /* for the /proc/ directory itself, after non-process stuff has been done */
2560 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2561 {
2562         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2563         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2564         struct tgid_iter iter;
2565         struct pid_namespace *ns;
2566
2567         if (!reaper)
2568                 goto out_no_task;
2569
2570         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2571                 const struct pid_entry *p = &proc_base_stuff[nr];
2572                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2573                         goto out;
2574         }
2575
2576         ns = filp->f_dentry->d_sb->s_fs_info;
2577         iter.task = NULL;
2578         iter.tgid = filp->f_pos - TGID_OFFSET;
2579         for (iter = next_tgid(ns, iter);
2580              iter.task;
2581              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2582                 filp->f_pos = iter.tgid + TGID_OFFSET;
2583                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2584                         put_task_struct(iter.task);
2585                         goto out;
2586                 }
2587         }
2588         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2589 out:
2590         put_task_struct(reaper);
2591 out_no_task:
2592         return 0;
2593 }
2594
2595 /*
2596  * Tasks
2597  */
2598 static const struct pid_entry tid_base_stuff[] = {
2599         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2600         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2601         REG("environ",   S_IRUSR, environ),
2602         INF("auxv",      S_IRUSR, pid_auxv),
2603         INF("status",    S_IRUGO, pid_status),
2604         INF("limits",    S_IRUSR, pid_limits),
2605 #ifdef CONFIG_SCHED_DEBUG
2606         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2607 #endif
2608         INF("cmdline",   S_IRUGO, pid_cmdline),
2609         INF("stat",      S_IRUGO, tid_stat),
2610         INF("statm",     S_IRUGO, pid_statm),
2611         REG("maps",      S_IRUGO, maps),
2612 #ifdef CONFIG_NUMA
2613         REG("numa_maps", S_IRUGO, numa_maps),
2614 #endif
2615         REG("mem",       S_IRUSR|S_IWUSR, mem),
2616         LNK("cwd",       cwd),
2617         LNK("root",      root),
2618         LNK("exe",       exe),
2619         REG("mounts",    S_IRUGO, mounts),
2620 #ifdef CONFIG_MMU
2621         REG("clear_refs", S_IWUSR, clear_refs),
2622         REG("smaps",     S_IRUGO, smaps),
2623 #endif
2624 #ifdef CONFIG_SECURITY
2625         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2626 #endif
2627 #ifdef CONFIG_KALLSYMS
2628         INF("wchan",     S_IRUGO, pid_wchan),
2629 #endif
2630 #ifdef CONFIG_SCHEDSTATS
2631         INF("schedstat", S_IRUGO, pid_schedstat),
2632 #endif
2633 #ifdef CONFIG_LATENCYTOP
2634         REG("latency",  S_IRUGO, lstats),
2635 #endif
2636 #ifdef CONFIG_PROC_PID_CPUSET
2637         REG("cpuset",    S_IRUGO, cpuset),
2638 #endif
2639 #ifdef CONFIG_CGROUPS
2640         REG("cgroup",  S_IRUGO, cgroup),
2641 #endif
2642         INF("oom_score", S_IRUGO, oom_score),
2643         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2644 #ifdef CONFIG_AUDITSYSCALL
2645         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2646 #endif
2647 #ifdef CONFIG_FAULT_INJECTION
2648         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2649 #endif
2650 };
2651
2652 static int proc_tid_base_readdir(struct file * filp,
2653                              void * dirent, filldir_t filldir)
2654 {
2655         return proc_pident_readdir(filp,dirent,filldir,
2656                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2657 }
2658
2659 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2660         return proc_pident_lookup(dir, dentry,
2661                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2662 }
2663
2664 static const struct file_operations proc_tid_base_operations = {
2665         .read           = generic_read_dir,
2666         .readdir        = proc_tid_base_readdir,
2667 };
2668
2669 static const struct inode_operations proc_tid_base_inode_operations = {
2670         .lookup         = proc_tid_base_lookup,
2671         .getattr        = pid_getattr,
2672         .setattr        = proc_setattr,
2673 };
2674
2675 static struct dentry *proc_task_instantiate(struct inode *dir,
2676         struct dentry *dentry, struct task_struct *task, const void *ptr)
2677 {
2678         struct dentry *error = ERR_PTR(-ENOENT);
2679         struct inode *inode;
2680         inode = proc_pid_make_inode(dir->i_sb, task);
2681
2682         if (!inode)
2683                 goto out;
2684         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2685         inode->i_op = &proc_tid_base_inode_operations;
2686         inode->i_fop = &proc_tid_base_operations;
2687         inode->i_flags|=S_IMMUTABLE;
2688         inode->i_nlink = 4;
2689 #ifdef CONFIG_SECURITY
2690         inode->i_nlink += 1;
2691 #endif
2692
2693         dentry->d_op = &pid_dentry_operations;
2694
2695         d_add(dentry, inode);
2696         /* Close the race of the process dying before we return the dentry */
2697         if (pid_revalidate(dentry, NULL))
2698                 error = NULL;
2699 out:
2700         return error;
2701 }
2702
2703 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2704 {
2705         struct dentry *result = ERR_PTR(-ENOENT);
2706         struct task_struct *task;
2707         struct task_struct *leader = get_proc_task(dir);
2708         unsigned tid;
2709         struct pid_namespace *ns;
2710
2711         if (!leader)
2712                 goto out_no_task;
2713
2714         tid = name_to_int(dentry);
2715         if (tid == ~0U)
2716                 goto out;
2717
2718         ns = dentry->d_sb->s_fs_info;
2719         rcu_read_lock();
2720         task = find_task_by_pid_ns(tid, ns);
2721         if (task)
2722                 get_task_struct(task);
2723         rcu_read_unlock();
2724         if (!task)
2725                 goto out;
2726         if (!same_thread_group(leader, task))
2727                 goto out_drop_task;
2728
2729         result = proc_task_instantiate(dir, dentry, task, NULL);
2730 out_drop_task:
2731         put_task_struct(task);
2732 out:
2733         put_task_struct(leader);
2734 out_no_task:
2735         return result;
2736 }
2737
2738 /*
2739  * Find the first tid of a thread group to return to user space.
2740  *
2741  * Usually this is just the thread group leader, but if the users
2742  * buffer was too small or there was a seek into the middle of the
2743  * directory we have more work todo.
2744  *
2745  * In the case of a short read we start with find_task_by_pid.
2746  *
2747  * In the case of a seek we start with the leader and walk nr
2748  * threads past it.
2749  */
2750 static struct task_struct *first_tid(struct task_struct *leader,
2751                 int tid, int nr, struct pid_namespace *ns)
2752 {
2753         struct task_struct *pos;
2754
2755         rcu_read_lock();
2756         /* Attempt to start with the pid of a thread */
2757         if (tid && (nr > 0)) {
2758                 pos = find_task_by_pid_ns(tid, ns);
2759                 if (pos && (pos->group_leader == leader))
2760                         goto found;
2761         }
2762
2763         /* If nr exceeds the number of threads there is nothing todo */
2764         pos = NULL;
2765         if (nr && nr >= get_nr_threads(leader))
2766                 goto out;
2767
2768         /* If we haven't found our starting place yet start
2769          * with the leader and walk nr threads forward.
2770          */
2771         for (pos = leader; nr > 0; --nr) {
2772                 pos = next_thread(pos);
2773                 if (pos == leader) {
2774                         pos = NULL;
2775                         goto out;
2776                 }
2777         }
2778 found:
2779         get_task_struct(pos);
2780 out:
2781         rcu_read_unlock();
2782         return pos;
2783 }
2784
2785 /*
2786  * Find the next thread in the thread list.
2787  * Return NULL if there is an error or no next thread.
2788  *
2789  * The reference to the input task_struct is released.
2790  */
2791 static struct task_struct *next_tid(struct task_struct *start)
2792 {
2793         struct task_struct *pos = NULL;
2794         rcu_read_lock();
2795         if (pid_alive(start)) {
2796                 pos = next_thread(start);
2797                 if (thread_group_leader(pos))
2798                         pos = NULL;
2799                 else
2800                         get_task_struct(pos);
2801         }
2802         rcu_read_unlock();
2803         put_task_struct(start);
2804         return pos;
2805 }
2806
2807 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2808         struct task_struct *task, int tid)
2809 {
2810         char name[PROC_NUMBUF];
2811         int len = snprintf(name, sizeof(name), "%d", tid);
2812         return proc_fill_cache(filp, dirent, filldir, name, len,
2813                                 proc_task_instantiate, task, NULL);
2814 }
2815
2816 /* for the /proc/TGID/task/ directories */
2817 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2818 {
2819         struct dentry *dentry = filp->f_path.dentry;
2820         struct inode *inode = dentry->d_inode;
2821         struct task_struct *leader = NULL;
2822         struct task_struct *task;
2823         int retval = -ENOENT;
2824         ino_t ino;
2825         int tid;
2826         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2827         struct pid_namespace *ns;
2828
2829         task = get_proc_task(inode);
2830         if (!task)
2831                 goto out_no_task;
2832         rcu_read_lock();
2833         if (pid_alive(task)) {
2834                 leader = task->group_leader;
2835                 get_task_struct(leader);
2836         }
2837         rcu_read_unlock();
2838         put_task_struct(task);
2839         if (!leader)
2840                 goto out_no_task;
2841         retval = 0;
2842
2843         switch (pos) {
2844         case 0:
2845                 ino = inode->i_ino;
2846                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2847                         goto out;
2848                 pos++;
2849                 /* fall through */
2850         case 1:
2851                 ino = parent_ino(dentry);
2852                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2853                         goto out;
2854                 pos++;
2855                 /* fall through */
2856         }
2857
2858         /* f_version caches the tgid value that the last readdir call couldn't
2859          * return. lseek aka telldir automagically resets f_version to 0.
2860          */
2861         ns = filp->f_dentry->d_sb->s_fs_info;
2862         tid = (int)filp->f_version;
2863         filp->f_version = 0;
2864         for (task = first_tid(leader, tid, pos - 2, ns);
2865              task;
2866              task = next_tid(task), pos++) {
2867                 tid = task_pid_nr_ns(task, ns);
2868                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2869                         /* returning this tgid failed, save it as the first
2870                          * pid for the next readir call */
2871                         filp->f_version = (u64)tid;
2872                         put_task_struct(task);
2873                         break;
2874                 }
2875         }
2876 out:
2877         filp->f_pos = pos;
2878         put_task_struct(leader);
2879 out_no_task:
2880         return retval;
2881 }
2882
2883 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2884 {
2885         struct inode *inode = dentry->d_inode;
2886         struct task_struct *p = get_proc_task(inode);
2887         generic_fillattr(inode, stat);
2888
2889         if (p) {
2890                 rcu_read_lock();
2891                 stat->nlink += get_nr_threads(p);
2892                 rcu_read_unlock();
2893                 put_task_struct(p);
2894         }
2895
2896         return 0;
2897 }
2898
2899 static const struct inode_operations proc_task_inode_operations = {
2900         .lookup         = proc_task_lookup,
2901         .getattr        = proc_task_getattr,
2902         .setattr        = proc_setattr,
2903 };
2904
2905 static const struct file_operations proc_task_operations = {
2906         .read           = generic_read_dir,
2907         .readdir        = proc_task_readdir,
2908 };