fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/init.h>
  57 #include <linux/capability.h>
  58 #include <linux/file.h>
  59 #include <linux/string.h>
  60 #include <linux/seq_file.h>
  61 #include <linux/namei.h>
  62 #include <linux/mnt_namespace.h>
  63 #include <linux/mm.h>
  64 #include <linux/rcupdate.h>
  65 #include <linux/kallsyms.h>
  66 #include <linux/resource.h>
  67 #include <linux/module.h>
  68 #include <linux/mount.h>
  69 #include <linux/security.h>
  70 #include <linux/ptrace.h>
  71 #include <linux/cgroup.h>
  72 #include <linux/cpuset.h>
  73 #include <linux/audit.h>
  74 #include <linux/poll.h>
  75 #include <linux/nsproxy.h>
  76 #include <linux/oom.h>
  77 #include <linux/elf.h>
  78 #include <linux/pid_namespace.h>
  79 #include "internal.h"
  80
  81 /* NOTE:
  82  *      Implementing inode permission operations in /proc is almost
  83  *      certainly an error.  Permission checks need to happen during
  84  *      each system call not at open time.  The reason is that most of
  85  *      what we wish to check for permissions in /proc varies at runtime.
  86  *
  87  *      The classic example of a problem is opening file descriptors
  88  *      in /proc for a task before it execs a suid executable.
  89  */
  90
  91 struct pid_entry {
  92         char *name;
  93         int len;
  94         mode_t mode;
  95         const struct inode_operations *iop;
  96         const struct file_operations *fop;
  97         union proc_op op;
  98 };
  99
 100 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 101         .name = (NAME),                                 \
 102         .len  = sizeof(NAME) - 1,                       \
 103         .mode = MODE,                                   \
 104         .iop  = IOP,                                    \
 105         .fop  = FOP,                                    \
 106         .op   = OP,                                     \
 107 }
 108
 109 #define DIR(NAME, MODE, OTYPE)                                                  \
 110         NOD(NAME, (S_IFDIR|(MODE)),                                             \
 111                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
 112                 {} )
 113 #define LNK(NAME, OTYPE)                                        \
 114         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 115                 &proc_pid_link_inode_operations, NULL,          \
 116                 { .proc_get_link = &proc_##OTYPE##_link } )
 117 #define REG(NAME, MODE, OTYPE)                          \
 118         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
 119                 &proc_##OTYPE##_operations, {})
 120 #define INF(NAME, MODE, OTYPE)                          \
 121         NOD(NAME, (S_IFREG|(MODE)),                     \
 122                 NULL, &proc_info_file_operations,       \
 123                 { .proc_read = &proc_##OTYPE } )
 124 #define ONE(NAME, MODE, OTYPE)                          \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_single_file_operations,     \
 127                 { .proc_show = &proc_##OTYPE } )
 128
 129 int maps_protect;
 130 EXPORT_SYMBOL(maps_protect);
 131
 132 static struct fs_struct *get_fs_struct(struct task_struct *task)
 133 {
 134         struct fs_struct *fs;
 135         task_lock(task);
 136         fs = task->fs;
 137         if(fs)
 138                 atomic_inc(&fs->count);
 139         task_unlock(task);
 140         return fs;
 141 }
 142
 143 static int get_nr_threads(struct task_struct *tsk)
 144 {
 145         /* Must be called with the rcu_read_lock held */
 146         unsigned long flags;
 147         int count = 0;
 148
 149         if (lock_task_sighand(tsk, &flags)) {
 150                 count = atomic_read(&tsk->signal->count);
 151                 unlock_task_sighand(tsk, &flags);
 152         }
 153         return count;
 154 }
 155
 156 static int proc_cwd_link(struct inode *inode, struct path *path)
 157 {
 158         struct task_struct *task = get_proc_task(inode);
 159         struct fs_struct *fs = NULL;
 160         int result = -ENOENT;
 161
 162         if (task) {
 163                 fs = get_fs_struct(task);
 164                 put_task_struct(task);
 165         }
 166         if (fs) {
 167                 read_lock(&fs->lock);
 168                 *path = fs->pwd;
 169                 path_get(&fs->pwd);
 170                 read_unlock(&fs->lock);
 171                 result = 0;
 172                 put_fs_struct(fs);
 173         }
 174         return result;
 175 }
 176
 177 static int proc_root_link(struct inode *inode, struct path *path)
 178 {
 179         struct task_struct *task = get_proc_task(inode);
 180         struct fs_struct *fs = NULL;
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 fs = get_fs_struct(task);
 185                 put_task_struct(task);
 186         }
 187         if (fs) {
 188                 read_lock(&fs->lock);
 189                 *path = fs->root;
 190                 path_get(&fs->root);
 191                 read_unlock(&fs->lock);
 192                 result = 0;
 193                 put_fs_struct(fs);
 194         }
 195         return result;
 196 }
 197
 198 #define MAY_PTRACE(task) \
 199         (task == current || \
 200         (task->parent == current && \
 201         (task->ptrace & PT_PTRACED) && \
 202          (task_is_stopped_or_traced(task)) && \
 203          security_ptrace(current,task) == 0))
 204
 205 struct mm_struct *mm_for_maps(struct task_struct *task)
 206 {
 207         struct mm_struct *mm = get_task_mm(task);
 208         if (!mm)
 209                 return NULL;
 210         down_read(&mm->mmap_sem);
 211         task_lock(task);
 212         if (task->mm != mm)
 213                 goto out;
 214         if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
 215                 goto out;
 216         task_unlock(task);
 217         return mm;
 218 out:
 219         task_unlock(task);
 220         up_read(&mm->mmap_sem);
 221         mmput(mm);
 222         return NULL;
 223 }
 224
 225 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 226 {
 227         int res = 0;
 228         unsigned int len;
 229         struct mm_struct *mm = get_task_mm(task);
 230         if (!mm)
 231                 goto out;
 232         if (!mm->arg_end)
 233                 goto out_mm;    /* Shh! No looking before we're done */
 234
 235         len = mm->arg_end - mm->arg_start;
 236
 237         if (len > PAGE_SIZE)
 238                 len = PAGE_SIZE;
 239
 240         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 241
 242         // If the nul at the end of args has been overwritten, then
 243         // assume application is using setproctitle(3).
 244         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 245                 len = strnlen(buffer, res);
 246                 if (len < res) {
 247                     res = len;
 248                 } else {
 249                         len = mm->env_end - mm->env_start;
 250                         if (len > PAGE_SIZE - res)
 251                                 len = PAGE_SIZE - res;
 252                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 253                         res = strnlen(buffer, res);
 254                 }
 255         }
 256 out_mm:
 257         mmput(mm);
 258 out:
 259         return res;
 260 }
 261
 262 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 263 {
 264         int res = 0;
 265         struct mm_struct *mm = get_task_mm(task);
 266         if (mm) {
 267                 unsigned int nwords = 0;
 268                 do
 269                         nwords += 2;
 270                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 271                 res = nwords * sizeof(mm->saved_auxv[0]);
 272                 if (res > PAGE_SIZE)
 273                         res = PAGE_SIZE;
 274                 memcpy(buffer, mm->saved_auxv, res);
 275                 mmput(mm);
 276         }
 277         return res;
 278 }
 279
 280
 281 #ifdef CONFIG_KALLSYMS
 282 /*
 283  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 284  * Returns the resolved symbol.  If that fails, simply return the address.
 285  */
 286 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 287 {
 288         unsigned long wchan;
 289         char symname[KSYM_NAME_LEN];
 290
 291         wchan = get_wchan(task);
 292
 293         if (lookup_symbol_name(wchan, symname) < 0)
 294                 return sprintf(buffer, "%lu", wchan);
 295         else
 296                 return sprintf(buffer, "%s", symname);
 297 }
 298 #endif /* CONFIG_KALLSYMS */
 299
 300 #ifdef CONFIG_SCHEDSTATS
 301 /*
 302  * Provides /proc/PID/schedstat
 303  */
 304 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 305 {
 306         return sprintf(buffer, "%llu %llu %lu\n",
 307                         task->sched_info.cpu_time,
 308                         task->sched_info.run_delay,
 309                         task->sched_info.pcount);
 310 }
 311 #endif
 312
 313 #ifdef CONFIG_LATENCYTOP
 314 static int lstats_show_proc(struct seq_file *m, void *v)
 315 {
 316         int i;
 317         struct inode *inode = m->private;
 318         struct task_struct *task = get_proc_task(inode);
 319
 320         if (!task)
 321                 return -ESRCH;
 322         seq_puts(m, "Latency Top version : v0.1\n");
 323         for (i = 0; i < 32; i++) {
 324                 if (task->latency_record[i].backtrace[0]) {
 325                         int q;
 326                         seq_printf(m, "%i %li %li ",
 327                                 task->latency_record[i].count,
 328                                 task->latency_record[i].time,
 329                                 task->latency_record[i].max);
 330                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 331                                 char sym[KSYM_NAME_LEN];
 332                                 char *c;
 333                                 if (!task->latency_record[i].backtrace[q])
 334                                         break;
 335                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 336                                         break;
 337                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 338                                 c = strchr(sym, '+');
 339                                 if (c)
 340                                         *c = 0;
 341                                 seq_printf(m, "%s ", sym);
 342                         }
 343                         seq_printf(m, "\n");
 344                 }
 345
 346         }
 347         put_task_struct(task);
 348         return 0;
 349 }
 350
 351 static int lstats_open(struct inode *inode, struct file *file)
 352 {
 353         return single_open(file, lstats_show_proc, inode);
 354 }
 355
 356 static ssize_t lstats_write(struct file *file, const char __user *buf,
 357                             size_t count, loff_t *offs)
 358 {
 359         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 360
 361         if (!task)
 362                 return -ESRCH;
 363         clear_all_latency_tracing(task);
 364         put_task_struct(task);
 365
 366         return count;
 367 }
 368
 369 static const struct file_operations proc_lstats_operations = {
 370         .open           = lstats_open,
 371         .read           = seq_read,
 372         .write          = lstats_write,
 373         .llseek         = seq_lseek,
 374         .release        = single_release,
 375 };
 376
 377 #endif
 378
 379 /* The badness from the OOM killer */
 380 unsigned long badness(struct task_struct *p, unsigned long uptime);
 381 static int proc_oom_score(struct task_struct *task, char *buffer)
 382 {
 383         unsigned long points;
 384         struct timespec uptime;
 385
 386         do_posix_clock_monotonic_gettime(&uptime);
 387         read_lock(&tasklist_lock);
 388         points = badness(task, uptime.tv_sec);
 389         read_unlock(&tasklist_lock);
 390         return sprintf(buffer, "%lu\n", points);
 391 }
 392
 393 struct limit_names {
 394         char *name;
 395         char *unit;
 396 };
 397
 398 static const struct limit_names lnames[RLIM_NLIMITS] = {
 399         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 400         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 401         [RLIMIT_DATA] = {"Max data size", "bytes"},
 402         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 403         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 404         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 405         [RLIMIT_NPROC] = {"Max processes", "processes"},
 406         [RLIMIT_NOFILE] = {"Max open files", "files"},
 407         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 408         [RLIMIT_AS] = {"Max address space", "bytes"},
 409         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 410         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 411         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 412         [RLIMIT_NICE] = {"Max nice priority", NULL},
 413         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 414         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 415 };
 416
 417 /* Display limits for a process */
 418 static int proc_pid_limits(struct task_struct *task, char *buffer)
 419 {
 420         unsigned int i;
 421         int count = 0;
 422         unsigned long flags;
 423         char *bufptr = buffer;
 424
 425         struct rlimit rlim[RLIM_NLIMITS];
 426
 427         rcu_read_lock();
 428         if (!lock_task_sighand(task,&flags)) {
 429                 rcu_read_unlock();
 430                 return 0;
 431         }
 432         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 433         unlock_task_sighand(task, &flags);
 434         rcu_read_unlock();
 435
 436         /*
 437          * print the file header
 438          */
 439         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 440                         "Limit", "Soft Limit", "Hard Limit", "Units");
 441
 442         for (i = 0; i < RLIM_NLIMITS; i++) {
 443                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 444                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 445                                          lnames[i].name, "unlimited");
 446                 else
 447                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 448                                          lnames[i].name, rlim[i].rlim_cur);
 449
 450                 if (rlim[i].rlim_max == RLIM_INFINITY)
 451                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 452                 else
 453                         count += sprintf(&bufptr[count], "%-20lu ",
 454                                          rlim[i].rlim_max);
 455
 456                 if (lnames[i].unit)
 457                         count += sprintf(&bufptr[count], "%-10s\n",
 458                                          lnames[i].unit);
 459                 else
 460                         count += sprintf(&bufptr[count], "\n");
 461         }
 462
 463         return count;
 464 }
 465
 466 /************************************************************************/
 467 /*                       Here the fs part begins                        */
 468 /************************************************************************/
 469
 470 /* permission checks */
 471 static int proc_fd_access_allowed(struct inode *inode)
 472 {
 473         struct task_struct *task;
 474         int allowed = 0;
 475         /* Allow access to a task's file descriptors if it is us or we
 476          * may use ptrace attach to the process and find out that
 477          * information.
 478          */
 479         task = get_proc_task(inode);
 480         if (task) {
 481                 allowed = ptrace_may_attach(task);
 482                 put_task_struct(task);
 483         }
 484         return allowed;
 485 }
 486
 487 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 488 {
 489         int error;
 490         struct inode *inode = dentry->d_inode;
 491
 492         if (attr->ia_valid & ATTR_MODE)
 493                 return -EPERM;
 494
 495         error = inode_change_ok(inode, attr);
 496         if (!error)
 497                 error = inode_setattr(inode, attr);
 498         return error;
 499 }
 500
 501 static const struct inode_operations proc_def_inode_operations = {
 502         .setattr        = proc_setattr,
 503 };
 504
 505 static int mounts_open_common(struct inode *inode, struct file *file,
 506                               const struct seq_operations *op)
 507 {
 508         struct task_struct *task = get_proc_task(inode);
 509         struct nsproxy *nsp;
 510         struct mnt_namespace *ns = NULL;
 511         struct fs_struct *fs = NULL;
 512         struct path root;
 513         struct proc_mounts *p;
 514         int ret = -EINVAL;
 515
 516         if (task) {
 517                 rcu_read_lock();
 518                 nsp = task_nsproxy(task);
 519                 if (nsp) {
 520                         ns = nsp->mnt_ns;
 521                         if (ns)
 522                                 get_mnt_ns(ns);
 523                 }
 524                 rcu_read_unlock();
 525                 if (ns)
 526                         fs = get_fs_struct(task);
 527                 put_task_struct(task);
 528         }
 529
 530         if (!ns)
 531                 goto err;
 532         if (!fs)
 533                 goto err_put_ns;
 534
 535         read_lock(&fs->lock);
 536         root = fs->root;
 537         path_get(&root);
 538         read_unlock(&fs->lock);
 539         put_fs_struct(fs);
 540
 541         ret = -ENOMEM;
 542         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 543         if (!p)
 544                 goto err_put_path;
 545
 546         file->private_data = &p->m;
 547         ret = seq_open(file, op);
 548         if (ret)
 549                 goto err_free;
 550
 551         p->m.private = p;
 552         p->ns = ns;
 553         p->root = root;
 554         p->event = ns->event;
 555
 556         return 0;
 557
 558  err_free:
 559         kfree(p);
 560  err_put_path:
 561         path_put(&root);
 562  err_put_ns:
 563         put_mnt_ns(ns);
 564  err:
 565         return ret;
 566 }
 567
 568 static int mounts_release(struct inode *inode, struct file *file)
 569 {
 570         struct proc_mounts *p = file->private_data;
 571         path_put(&p->root);
 572         put_mnt_ns(p->ns);
 573         return seq_release(inode, file);
 574 }
 575
 576 static unsigned mounts_poll(struct file *file, poll_table *wait)
 577 {
 578         struct proc_mounts *p = file->private_data;
 579         struct mnt_namespace *ns = p->ns;
 580         unsigned res = 0;
 581
 582         poll_wait(file, &ns->poll, wait);
 583
 584         spin_lock(&vfsmount_lock);
 585         if (p->event != ns->event) {
 586                 p->event = ns->event;
 587                 res = POLLERR;
 588         }
 589         spin_unlock(&vfsmount_lock);
 590
 591         return res;
 592 }
 593
 594 static int mounts_open(struct inode *inode, struct file *file)
 595 {
 596         return mounts_open_common(inode, file, &mounts_op);
 597 }
 598
 599 static const struct file_operations proc_mounts_operations = {
 600         .open           = mounts_open,
 601         .read           = seq_read,
 602         .llseek         = seq_lseek,
 603         .release        = mounts_release,
 604         .poll           = mounts_poll,
 605 };
 606
 607 static int mountstats_open(struct inode *inode, struct file *file)
 608 {
 609         return mounts_open_common(inode, file, &mountstats_op);
 610 }
 611
 612 static const struct file_operations proc_mountstats_operations = {
 613         .open           = mountstats_open,
 614         .read           = seq_read,
 615         .llseek         = seq_lseek,
 616         .release        = mounts_release,
 617 };
 618
 619 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 620
 621 static ssize_t proc_info_read(struct file * file, char __user * buf,
 622                           size_t count, loff_t *ppos)
 623 {
 624         struct inode * inode = file->f_path.dentry->d_inode;
 625         unsigned long page;
 626         ssize_t length;
 627         struct task_struct *task = get_proc_task(inode);
 628
 629         length = -ESRCH;
 630         if (!task)
 631                 goto out_no_task;
 632
 633         if (count > PROC_BLOCK_SIZE)
 634                 count = PROC_BLOCK_SIZE;
 635
 636         length = -ENOMEM;
 637         if (!(page = __get_free_page(GFP_TEMPORARY)))
 638                 goto out;
 639
 640         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 641
 642         if (length >= 0)
 643                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 644         free_page(page);
 645 out:
 646         put_task_struct(task);
 647 out_no_task:
 648         return length;
 649 }
 650
 651 static const struct file_operations proc_info_file_operations = {
 652         .read           = proc_info_read,
 653 };
 654
 655 static int proc_single_show(struct seq_file *m, void *v)
 656 {
 657         struct inode *inode = m->private;
 658         struct pid_namespace *ns;
 659         struct pid *pid;
 660         struct task_struct *task;
 661         int ret;
 662
 663         ns = inode->i_sb->s_fs_info;
 664         pid = proc_pid(inode);
 665         task = get_pid_task(pid, PIDTYPE_PID);
 666         if (!task)
 667                 return -ESRCH;
 668
 669         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 670
 671         put_task_struct(task);
 672         return ret;
 673 }
 674
 675 static int proc_single_open(struct inode *inode, struct file *filp)
 676 {
 677         int ret;
 678         ret = single_open(filp, proc_single_show, NULL);
 679         if (!ret) {
 680                 struct seq_file *m = filp->private_data;
 681
 682                 m->private = inode;
 683         }
 684         return ret;
 685 }
 686
 687 static const struct file_operations proc_single_file_operations = {
 688         .open           = proc_single_open,
 689         .read           = seq_read,
 690         .llseek         = seq_lseek,
 691         .release        = single_release,
 692 };
 693
 694 static int mem_open(struct inode* inode, struct file* file)
 695 {
 696         file->private_data = (void*)((long)current->self_exec_id);
 697         return 0;
 698 }
 699
 700 static ssize_t mem_read(struct file * file, char __user * buf,
 701                         size_t count, loff_t *ppos)
 702 {
 703         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 704         char *page;
 705         unsigned long src = *ppos;
 706         int ret = -ESRCH;
 707         struct mm_struct *mm;
 708
 709         if (!task)
 710                 goto out_no_task;
 711
 712         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 713                 goto out;
 714
 715         ret = -ENOMEM;
 716         page = (char *)__get_free_page(GFP_TEMPORARY);
 717         if (!page)
 718                 goto out;
 719
 720         ret = 0;
 721
 722         mm = get_task_mm(task);
 723         if (!mm)
 724                 goto out_free;
 725
 726         ret = -EIO;
 727
 728         if (file->private_data != (void*)((long)current->self_exec_id))
 729                 goto out_put;
 730
 731         ret = 0;
 732
 733         while (count > 0) {
 734                 int this_len, retval;
 735
 736                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 737                 retval = access_process_vm(task, src, page, this_len, 0);
 738                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 739                         if (!ret)
 740                                 ret = -EIO;
 741                         break;
 742                 }
 743
 744                 if (copy_to_user(buf, page, retval)) {
 745                         ret = -EFAULT;
 746                         break;
 747                 }
 748
 749                 ret += retval;
 750                 src += retval;
 751                 buf += retval;
 752                 count -= retval;
 753         }
 754         *ppos = src;
 755
 756 out_put:
 757         mmput(mm);
 758 out_free:
 759         free_page((unsigned long) page);
 760 out:
 761         put_task_struct(task);
 762 out_no_task:
 763         return ret;
 764 }
 765
 766 #define mem_write NULL
 767
 768 #ifndef mem_write
 769 /* This is a security hazard */
 770 static ssize_t mem_write(struct file * file, const char __user *buf,
 771                          size_t count, loff_t *ppos)
 772 {
 773         int copied;
 774         char *page;
 775         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 776         unsigned long dst = *ppos;
 777
 778         copied = -ESRCH;
 779         if (!task)
 780                 goto out_no_task;
 781
 782         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 783                 goto out;
 784
 785         copied = -ENOMEM;
 786         page = (char *)__get_free_page(GFP_TEMPORARY);
 787         if (!page)
 788                 goto out;
 789
 790         copied = 0;
 791         while (count > 0) {
 792                 int this_len, retval;
 793
 794                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 795                 if (copy_from_user(page, buf, this_len)) {
 796                         copied = -EFAULT;
 797                         break;
 798                 }
 799                 retval = access_process_vm(task, dst, page, this_len, 1);
 800                 if (!retval) {
 801                         if (!copied)
 802                                 copied = -EIO;
 803                         break;
 804                 }
 805                 copied += retval;
 806                 buf += retval;
 807                 dst += retval;
 808                 count -= retval;
 809         }
 810         *ppos = dst;
 811         free_page((unsigned long) page);
 812 out:
 813         put_task_struct(task);
 814 out_no_task:
 815         return copied;
 816 }
 817 #endif
 818
 819 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 820 {
 821         switch (orig) {
 822         case 0:
 823                 file->f_pos = offset;
 824                 break;
 825         case 1:
 826                 file->f_pos += offset;
 827                 break;
 828         default:
 829                 return -EINVAL;
 830         }
 831         force_successful_syscall_return();
 832         return file->f_pos;
 833 }
 834
 835 static const struct file_operations proc_mem_operations = {
 836         .llseek         = mem_lseek,
 837         .read           = mem_read,
 838         .write          = mem_write,
 839         .open           = mem_open,
 840 };
 841
 842 static ssize_t environ_read(struct file *file, char __user *buf,
 843                         size_t count, loff_t *ppos)
 844 {
 845         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 846         char *page;
 847         unsigned long src = *ppos;
 848         int ret = -ESRCH;
 849         struct mm_struct *mm;
 850
 851         if (!task)
 852                 goto out_no_task;
 853
 854         if (!ptrace_may_attach(task))
 855                 goto out;
 856
 857         ret = -ENOMEM;
 858         page = (char *)__get_free_page(GFP_TEMPORARY);
 859         if (!page)
 860                 goto out;
 861
 862         ret = 0;
 863
 864         mm = get_task_mm(task);
 865         if (!mm)
 866                 goto out_free;
 867
 868         while (count > 0) {
 869                 int this_len, retval, max_len;
 870
 871                 this_len = mm->env_end - (mm->env_start + src);
 872
 873                 if (this_len <= 0)
 874                         break;
 875
 876                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 877                 this_len = (this_len > max_len) ? max_len : this_len;
 878
 879                 retval = access_process_vm(task, (mm->env_start + src),
 880                         page, this_len, 0);
 881
 882                 if (retval <= 0) {
 883                         ret = retval;
 884                         break;
 885                 }
 886
 887                 if (copy_to_user(buf, page, retval)) {
 888                         ret = -EFAULT;
 889                         break;
 890                 }
 891
 892                 ret += retval;
 893                 src += retval;
 894                 buf += retval;
 895                 count -= retval;
 896         }
 897         *ppos = src;
 898
 899         mmput(mm);
 900 out_free:
 901         free_page((unsigned long) page);
 902 out:
 903         put_task_struct(task);
 904 out_no_task:
 905         return ret;
 906 }
 907
 908 static const struct file_operations proc_environ_operations = {
 909         .read           = environ_read,
 910 };
 911
 912 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 913                                 size_t count, loff_t *ppos)
 914 {
 915         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 916         char buffer[PROC_NUMBUF];
 917         size_t len;
 918         int oom_adjust;
 919
 920         if (!task)
 921                 return -ESRCH;
 922         oom_adjust = task->oomkilladj;
 923         put_task_struct(task);
 924
 925         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 926
 927         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 928 }
 929
 930 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 931                                 size_t count, loff_t *ppos)
 932 {
 933         struct task_struct *task;
 934         char buffer[PROC_NUMBUF], *end;
 935         int oom_adjust;
 936
 937         memset(buffer, 0, sizeof(buffer));
 938         if (count > sizeof(buffer) - 1)
 939                 count = sizeof(buffer) - 1;
 940         if (copy_from_user(buffer, buf, count))
 941                 return -EFAULT;
 942         oom_adjust = simple_strtol(buffer, &end, 0);
 943         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 944              oom_adjust != OOM_DISABLE)
 945                 return -EINVAL;
 946         if (*end == '\n')
 947                 end++;
 948         task = get_proc_task(file->f_path.dentry->d_inode);
 949         if (!task)
 950                 return -ESRCH;
 951         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
 952                 put_task_struct(task);
 953                 return -EACCES;
 954         }
 955         task->oomkilladj = oom_adjust;
 956         put_task_struct(task);
 957         if (end - buffer == 0)
 958                 return -EIO;
 959         return end - buffer;
 960 }
 961
 962 static const struct file_operations proc_oom_adjust_operations = {
 963         .read           = oom_adjust_read,
 964         .write          = oom_adjust_write,
 965 };
 966
 967 #ifdef CONFIG_AUDITSYSCALL
 968 #define TMPBUFLEN 21
 969 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 970                                   size_t count, loff_t *ppos)
 971 {
 972         struct inode * inode = file->f_path.dentry->d_inode;
 973         struct task_struct *task = get_proc_task(inode);
 974         ssize_t length;
 975         char tmpbuf[TMPBUFLEN];
 976
 977         if (!task)
 978                 return -ESRCH;
 979         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 980                                 audit_get_loginuid(task));
 981         put_task_struct(task);
 982         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 983 }
 984
 985 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 986                                    size_t count, loff_t *ppos)
 987 {
 988         struct inode * inode = file->f_path.dentry->d_inode;
 989         char *page, *tmp;
 990         ssize_t length;
 991         uid_t loginuid;
 992
 993         if (!capable(CAP_AUDIT_CONTROL))
 994                 return -EPERM;
 995
 996         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
 997                 return -EPERM;
 998
 999         if (count >= PAGE_SIZE)
1000                 count = PAGE_SIZE - 1;
1001
1002         if (*ppos != 0) {
1003                 /* No partial writes. */
1004                 return -EINVAL;
1005         }
1006         page = (char*)__get_free_page(GFP_TEMPORARY);
1007         if (!page)
1008                 return -ENOMEM;
1009         length = -EFAULT;
1010         if (copy_from_user(page, buf, count))
1011                 goto out_free_page;
1012
1013         page[count] = '\0';
1014         loginuid = simple_strtoul(page, &tmp, 10);
1015         if (tmp == page) {
1016                 length = -EINVAL;
1017                 goto out_free_page;
1018
1019         }
1020         length = audit_set_loginuid(current, loginuid);
1021         if (likely(length == 0))
1022                 length = count;
1023
1024 out_free_page:
1025         free_page((unsigned long) page);
1026         return length;
1027 }
1028
1029 static const struct file_operations proc_loginuid_operations = {
1030         .read           = proc_loginuid_read,
1031         .write          = proc_loginuid_write,
1032 };
1033
1034 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1035                                   size_t count, loff_t *ppos)
1036 {
1037         struct inode * inode = file->f_path.dentry->d_inode;
1038         struct task_struct *task = get_proc_task(inode);
1039         ssize_t length;
1040         char tmpbuf[TMPBUFLEN];
1041
1042         if (!task)
1043                 return -ESRCH;
1044         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1045                                 audit_get_sessionid(task));
1046         put_task_struct(task);
1047         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1048 }
1049
1050 static const struct file_operations proc_sessionid_operations = {
1051         .read           = proc_sessionid_read,
1052 };
1053 #endif
1054
1055 #ifdef CONFIG_FAULT_INJECTION
1056 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1057                                       size_t count, loff_t *ppos)
1058 {
1059         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1060         char buffer[PROC_NUMBUF];
1061         size_t len;
1062         int make_it_fail;
1063
1064         if (!task)
1065                 return -ESRCH;
1066         make_it_fail = task->make_it_fail;
1067         put_task_struct(task);
1068
1069         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1070
1071         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1072 }
1073
1074 static ssize_t proc_fault_inject_write(struct file * file,
1075                         const char __user * buf, size_t count, loff_t *ppos)
1076 {
1077         struct task_struct *task;
1078         char buffer[PROC_NUMBUF], *end;
1079         int make_it_fail;
1080
1081         if (!capable(CAP_SYS_RESOURCE))
1082                 return -EPERM;
1083         memset(buffer, 0, sizeof(buffer));
1084         if (count > sizeof(buffer) - 1)
1085                 count = sizeof(buffer) - 1;
1086         if (copy_from_user(buffer, buf, count))
1087                 return -EFAULT;
1088         make_it_fail = simple_strtol(buffer, &end, 0);
1089         if (*end == '\n')
1090                 end++;
1091         task = get_proc_task(file->f_dentry->d_inode);
1092         if (!task)
1093                 return -ESRCH;
1094         task->make_it_fail = make_it_fail;
1095         put_task_struct(task);
1096         if (end - buffer == 0)
1097                 return -EIO;
1098         return end - buffer;
1099 }
1100
1101 static const struct file_operations proc_fault_inject_operations = {
1102         .read           = proc_fault_inject_read,
1103         .write          = proc_fault_inject_write,
1104 };
1105 #endif
1106
1107
1108 #ifdef CONFIG_SCHED_DEBUG
1109 /*
1110  * Print out various scheduling related per-task fields:
1111  */
1112 static int sched_show(struct seq_file *m, void *v)
1113 {
1114         struct inode *inode = m->private;
1115         struct task_struct *p;
1116
1117         WARN_ON(!inode);
1118
1119         p = get_proc_task(inode);
1120         if (!p)
1121                 return -ESRCH;
1122         proc_sched_show_task(p, m);
1123
1124         put_task_struct(p);
1125
1126         return 0;
1127 }
1128
1129 static ssize_t
1130 sched_write(struct file *file, const char __user *buf,
1131             size_t count, loff_t *offset)
1132 {
1133         struct inode *inode = file->f_path.dentry->d_inode;
1134         struct task_struct *p;
1135
1136         WARN_ON(!inode);
1137
1138         p = get_proc_task(inode);
1139         if (!p)
1140                 return -ESRCH;
1141         proc_sched_set_task(p);
1142
1143         put_task_struct(p);
1144
1145         return count;
1146 }
1147
1148 static int sched_open(struct inode *inode, struct file *filp)
1149 {
1150         int ret;
1151
1152         ret = single_open(filp, sched_show, NULL);
1153         if (!ret) {
1154                 struct seq_file *m = filp->private_data;
1155
1156                 m->private = inode;
1157         }
1158         return ret;
1159 }
1160
1161 static const struct file_operations proc_pid_sched_operations = {
1162         .open           = sched_open,
1163         .read           = seq_read,
1164         .write          = sched_write,
1165         .llseek         = seq_lseek,
1166         .release        = single_release,
1167 };
1168
1169 #endif
1170
1171 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1172 {
1173         struct inode *inode = dentry->d_inode;
1174         int error = -EACCES;
1175
1176         /* We don't need a base pointer in the /proc filesystem */
1177         path_put(&nd->path);
1178
1179         /* Are we allowed to snoop on the tasks file descriptors? */
1180         if (!proc_fd_access_allowed(inode))
1181                 goto out;
1182
1183         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1184         nd->last_type = LAST_BIND;
1185 out:
1186         return ERR_PTR(error);
1187 }
1188
1189 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1190 {
1191         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1192         char *pathname;
1193         int len;
1194
1195         if (!tmp)
1196                 return -ENOMEM;
1197
1198         pathname = d_path(path, tmp, PAGE_SIZE);
1199         len = PTR_ERR(pathname);
1200         if (IS_ERR(pathname))
1201                 goto out;
1202         len = tmp + PAGE_SIZE - 1 - pathname;
1203
1204         if (len > buflen)
1205                 len = buflen;
1206         if (copy_to_user(buffer, pathname, len))
1207                 len = -EFAULT;
1208  out:
1209         free_page((unsigned long)tmp);
1210         return len;
1211 }
1212
1213 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1214 {
1215         int error = -EACCES;
1216         struct inode *inode = dentry->d_inode;
1217         struct path path;
1218
1219         /* Are we allowed to snoop on the tasks file descriptors? */
1220         if (!proc_fd_access_allowed(inode))
1221                 goto out;
1222
1223         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1224         if (error)
1225                 goto out;
1226
1227         error = do_proc_readlink(&path, buffer, buflen);
1228         path_put(&path);
1229 out:
1230         return error;
1231 }
1232
1233 static const struct inode_operations proc_pid_link_inode_operations = {
1234         .readlink       = proc_pid_readlink,
1235         .follow_link    = proc_pid_follow_link,
1236         .setattr        = proc_setattr,
1237 };
1238
1239
1240 /* building an inode */
1241
1242 static int task_dumpable(struct task_struct *task)
1243 {
1244         int dumpable = 0;
1245         struct mm_struct *mm;
1246
1247         task_lock(task);
1248         mm = task->mm;
1249         if (mm)
1250                 dumpable = get_dumpable(mm);
1251         task_unlock(task);
1252         if(dumpable == 1)
1253                 return 1;
1254         return 0;
1255 }
1256
1257
1258 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1259 {
1260         struct inode * inode;
1261         struct proc_inode *ei;
1262
1263         /* We need a new inode */
1264
1265         inode = new_inode(sb);
1266         if (!inode)
1267                 goto out;
1268
1269         /* Common stuff */
1270         ei = PROC_I(inode);
1271         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1272         inode->i_op = &proc_def_inode_operations;
1273
1274         /*
1275          * grab the reference to task.
1276          */
1277         ei->pid = get_task_pid(task, PIDTYPE_PID);
1278         if (!ei->pid)
1279                 goto out_unlock;
1280
1281         inode->i_uid = 0;
1282         inode->i_gid = 0;
1283         if (task_dumpable(task)) {
1284                 inode->i_uid = task->euid;
1285                 inode->i_gid = task->egid;
1286         }
1287         security_task_to_inode(task, inode);
1288
1289 out:
1290         return inode;
1291
1292 out_unlock:
1293         iput(inode);
1294         return NULL;
1295 }
1296
1297 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1298 {
1299         struct inode *inode = dentry->d_inode;
1300         struct task_struct *task;
1301         generic_fillattr(inode, stat);
1302
1303         rcu_read_lock();
1304         stat->uid = 0;
1305         stat->gid = 0;
1306         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1307         if (task) {
1308                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1309                     task_dumpable(task)) {
1310                         stat->uid = task->euid;
1311                         stat->gid = task->egid;
1312                 }
1313         }
1314         rcu_read_unlock();
1315         return 0;
1316 }
1317
1318 /* dentry stuff */
1319
1320 /*
1321  *      Exceptional case: normally we are not allowed to unhash a busy
1322  * directory. In this case, however, we can do it - no aliasing problems
1323  * due to the way we treat inodes.
1324  *
1325  * Rewrite the inode's ownerships here because the owning task may have
1326  * performed a setuid(), etc.
1327  *
1328  * Before the /proc/pid/status file was created the only way to read
1329  * the effective uid of a /process was to stat /proc/pid.  Reading
1330  * /proc/pid/status is slow enough that procps and other packages
1331  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1332  * made this apply to all per process world readable and executable
1333  * directories.
1334  */
1335 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1336 {
1337         struct inode *inode = dentry->d_inode;
1338         struct task_struct *task = get_proc_task(inode);
1339         if (task) {
1340                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1341                     task_dumpable(task)) {
1342                         inode->i_uid = task->euid;
1343                         inode->i_gid = task->egid;
1344                 } else {
1345                         inode->i_uid = 0;
1346                         inode->i_gid = 0;
1347                 }
1348                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1349                 security_task_to_inode(task, inode);
1350                 put_task_struct(task);
1351                 return 1;
1352         }
1353         d_drop(dentry);
1354         return 0;
1355 }
1356
1357 static int pid_delete_dentry(struct dentry * dentry)
1358 {
1359         /* Is the task we represent dead?
1360          * If so, then don't put the dentry on the lru list,
1361          * kill it immediately.
1362          */
1363         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1364 }
1365
1366 static struct dentry_operations pid_dentry_operations =
1367 {
1368         .d_revalidate   = pid_revalidate,
1369         .d_delete       = pid_delete_dentry,
1370 };
1371
1372 /* Lookups */
1373
1374 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1375                                 struct task_struct *, const void *);
1376
1377 /*
1378  * Fill a directory entry.
1379  *
1380  * If possible create the dcache entry and derive our inode number and
1381  * file type from dcache entry.
1382  *
1383  * Since all of the proc inode numbers are dynamically generated, the inode
1384  * numbers do not exist until the inode is cache.  This means creating the
1385  * the dcache entry in readdir is necessary to keep the inode numbers
1386  * reported by readdir in sync with the inode numbers reported
1387  * by stat.
1388  */
1389 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1390         char *name, int len,
1391         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1392 {
1393         struct dentry *child, *dir = filp->f_path.dentry;
1394         struct inode *inode;
1395         struct qstr qname;
1396         ino_t ino = 0;
1397         unsigned type = DT_UNKNOWN;
1398
1399         qname.name = name;
1400         qname.len  = len;
1401         qname.hash = full_name_hash(name, len);
1402
1403         child = d_lookup(dir, &qname);
1404         if (!child) {
1405                 struct dentry *new;
1406                 new = d_alloc(dir, &qname);
1407                 if (new) {
1408                         child = instantiate(dir->d_inode, new, task, ptr);
1409                         if (child)
1410                                 dput(new);
1411                         else
1412                                 child = new;
1413                 }
1414         }
1415         if (!child || IS_ERR(child) || !child->d_inode)
1416                 goto end_instantiate;
1417         inode = child->d_inode;
1418         if (inode) {
1419                 ino = inode->i_ino;
1420                 type = inode->i_mode >> 12;
1421         }
1422         dput(child);
1423 end_instantiate:
1424         if (!ino)
1425                 ino = find_inode_number(dir, &qname);
1426         if (!ino)
1427                 ino = 1;
1428         return filldir(dirent, name, len, filp->f_pos, ino, type);
1429 }
1430
1431 static unsigned name_to_int(struct dentry *dentry)
1432 {
1433         const char *name = dentry->d_name.name;
1434         int len = dentry->d_name.len;
1435         unsigned n = 0;
1436
1437         if (len > 1 && *name == '0')
1438                 goto out;
1439         while (len-- > 0) {
1440                 unsigned c = *name++ - '0';
1441                 if (c > 9)
1442                         goto out;
1443                 if (n >= (~0U-9)/10)
1444                         goto out;
1445                 n *= 10;
1446                 n += c;
1447         }
1448         return n;
1449 out:
1450         return ~0U;
1451 }
1452
1453 #define PROC_FDINFO_MAX 64
1454
1455 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1456 {
1457         struct task_struct *task = get_proc_task(inode);
1458         struct files_struct *files = NULL;
1459         struct file *file;
1460         int fd = proc_fd(inode);
1461
1462         if (task) {
1463                 files = get_files_struct(task);
1464                 put_task_struct(task);
1465         }
1466         if (files) {
1467                 /*
1468                  * We are not taking a ref to the file structure, so we must
1469                  * hold ->file_lock.
1470                  */
1471                 spin_lock(&files->file_lock);
1472                 file = fcheck_files(files, fd);
1473                 if (file) {
1474                         if (path) {
1475                                 *path = file->f_path;
1476                                 path_get(&file->f_path);
1477                         }
1478                         if (info)
1479                                 snprintf(info, PROC_FDINFO_MAX,
1480                                          "pos:\t%lli\n"
1481                                          "flags:\t0%o\n",
1482                                          (long long) file->f_pos,
1483                                          file->f_flags);
1484                         spin_unlock(&files->file_lock);
1485                         put_files_struct(files);
1486                         return 0;
1487                 }
1488                 spin_unlock(&files->file_lock);
1489                 put_files_struct(files);
1490         }
1491         return -ENOENT;
1492 }
1493
1494 static int proc_fd_link(struct inode *inode, struct path *path)
1495 {
1496         return proc_fd_info(inode, path, NULL);
1497 }
1498
1499 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1500 {
1501         struct inode *inode = dentry->d_inode;
1502         struct task_struct *task = get_proc_task(inode);
1503         int fd = proc_fd(inode);
1504         struct files_struct *files;
1505
1506         if (task) {
1507                 files = get_files_struct(task);
1508                 if (files) {
1509                         rcu_read_lock();
1510                         if (fcheck_files(files, fd)) {
1511                                 rcu_read_unlock();
1512                                 put_files_struct(files);
1513                                 if (task_dumpable(task)) {
1514                                         inode->i_uid = task->euid;
1515                                         inode->i_gid = task->egid;
1516                                 } else {
1517                                         inode->i_uid = 0;
1518                                         inode->i_gid = 0;
1519                                 }
1520                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1521                                 security_task_to_inode(task, inode);
1522                                 put_task_struct(task);
1523                                 return 1;
1524                         }
1525                         rcu_read_unlock();
1526                         put_files_struct(files);
1527                 }
1528                 put_task_struct(task);
1529         }
1530         d_drop(dentry);
1531         return 0;
1532 }
1533
1534 static struct dentry_operations tid_fd_dentry_operations =
1535 {
1536         .d_revalidate   = tid_fd_revalidate,
1537         .d_delete       = pid_delete_dentry,
1538 };
1539
1540 static struct dentry *proc_fd_instantiate(struct inode *dir,
1541         struct dentry *dentry, struct task_struct *task, const void *ptr)
1542 {
1543         unsigned fd = *(const unsigned *)ptr;
1544         struct file *file;
1545         struct files_struct *files;
1546         struct inode *inode;
1547         struct proc_inode *ei;
1548         struct dentry *error = ERR_PTR(-ENOENT);
1549
1550         inode = proc_pid_make_inode(dir->i_sb, task);
1551         if (!inode)
1552                 goto out;
1553         ei = PROC_I(inode);
1554         ei->fd = fd;
1555         files = get_files_struct(task);
1556         if (!files)
1557                 goto out_iput;
1558         inode->i_mode = S_IFLNK;
1559
1560         /*
1561          * We are not taking a ref to the file structure, so we must
1562          * hold ->file_lock.
1563          */
1564         spin_lock(&files->file_lock);
1565         file = fcheck_files(files, fd);
1566         if (!file)
1567                 goto out_unlock;
1568         if (file->f_mode & 1)
1569                 inode->i_mode |= S_IRUSR | S_IXUSR;
1570         if (file->f_mode & 2)
1571                 inode->i_mode |= S_IWUSR | S_IXUSR;
1572         spin_unlock(&files->file_lock);
1573         put_files_struct(files);
1574
1575         inode->i_op = &proc_pid_link_inode_operations;
1576         inode->i_size = 64;
1577         ei->op.proc_get_link = proc_fd_link;
1578         dentry->d_op = &tid_fd_dentry_operations;
1579         d_add(dentry, inode);
1580         /* Close the race of the process dying before we return the dentry */
1581         if (tid_fd_revalidate(dentry, NULL))
1582                 error = NULL;
1583
1584  out:
1585         return error;
1586 out_unlock:
1587         spin_unlock(&files->file_lock);
1588         put_files_struct(files);
1589 out_iput:
1590         iput(inode);
1591         goto out;
1592 }
1593
1594 static struct dentry *proc_lookupfd_common(struct inode *dir,
1595                                            struct dentry *dentry,
1596                                            instantiate_t instantiate)
1597 {
1598         struct task_struct *task = get_proc_task(dir);
1599         unsigned fd = name_to_int(dentry);
1600         struct dentry *result = ERR_PTR(-ENOENT);
1601
1602         if (!task)
1603                 goto out_no_task;
1604         if (fd == ~0U)
1605                 goto out;
1606
1607         result = instantiate(dir, dentry, task, &fd);
1608 out:
1609         put_task_struct(task);
1610 out_no_task:
1611         return result;
1612 }
1613
1614 static int proc_readfd_common(struct file * filp, void * dirent,
1615                               filldir_t filldir, instantiate_t instantiate)
1616 {
1617         struct dentry *dentry = filp->f_path.dentry;
1618         struct inode *inode = dentry->d_inode;
1619         struct task_struct *p = get_proc_task(inode);
1620         unsigned int fd, ino;
1621         int retval;
1622         struct files_struct * files;
1623
1624         retval = -ENOENT;
1625         if (!p)
1626                 goto out_no_task;
1627         retval = 0;
1628
1629         fd = filp->f_pos;
1630         switch (fd) {
1631                 case 0:
1632                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1633                                 goto out;
1634                         filp->f_pos++;
1635                 case 1:
1636                         ino = parent_ino(dentry);
1637                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1638                                 goto out;
1639                         filp->f_pos++;
1640                 default:
1641                         files = get_files_struct(p);
1642                         if (!files)
1643                                 goto out;
1644                         rcu_read_lock();
1645                         for (fd = filp->f_pos-2;
1646                              fd < files_fdtable(files)->max_fds;
1647                              fd++, filp->f_pos++) {
1648                                 char name[PROC_NUMBUF];
1649                                 int len;
1650
1651                                 if (!fcheck_files(files, fd))
1652                                         continue;
1653                                 rcu_read_unlock();
1654
1655                                 len = snprintf(name, sizeof(name), "%d", fd);
1656                                 if (proc_fill_cache(filp, dirent, filldir,
1657                                                     name, len, instantiate,
1658                                                     p, &fd) < 0) {
1659                                         rcu_read_lock();
1660                                         break;
1661                                 }
1662                                 rcu_read_lock();
1663                         }
1664                         rcu_read_unlock();
1665                         put_files_struct(files);
1666         }
1667 out:
1668         put_task_struct(p);
1669 out_no_task:
1670         return retval;
1671 }
1672
1673 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1674                                     struct nameidata *nd)
1675 {
1676         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1677 }
1678
1679 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1680 {
1681         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1682 }
1683
1684 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1685                                       size_t len, loff_t *ppos)
1686 {
1687         char tmp[PROC_FDINFO_MAX];
1688         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1689         if (!err)
1690                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1691         return err;
1692 }
1693
1694 static const struct file_operations proc_fdinfo_file_operations = {
1695         .open           = nonseekable_open,
1696         .read           = proc_fdinfo_read,
1697 };
1698
1699 static const struct file_operations proc_fd_operations = {
1700         .read           = generic_read_dir,
1701         .readdir        = proc_readfd,
1702 };
1703
1704 /*
1705  * /proc/pid/fd needs a special permission handler so that a process can still
1706  * access /proc/self/fd after it has executed a setuid().
1707  */
1708 static int proc_fd_permission(struct inode *inode, int mask,
1709                                 struct nameidata *nd)
1710 {
1711         int rv;
1712
1713         rv = generic_permission(inode, mask, NULL);
1714         if (rv == 0)
1715                 return 0;
1716         if (task_pid(current) == proc_pid(inode))
1717                 rv = 0;
1718         return rv;
1719 }
1720
1721 /*
1722  * proc directories can do almost nothing..
1723  */
1724 static const struct inode_operations proc_fd_inode_operations = {
1725         .lookup         = proc_lookupfd,
1726         .permission     = proc_fd_permission,
1727         .setattr        = proc_setattr,
1728 };
1729
1730 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1731         struct dentry *dentry, struct task_struct *task, const void *ptr)
1732 {
1733         unsigned fd = *(unsigned *)ptr;
1734         struct inode *inode;
1735         struct proc_inode *ei;
1736         struct dentry *error = ERR_PTR(-ENOENT);
1737
1738         inode = proc_pid_make_inode(dir->i_sb, task);
1739         if (!inode)
1740                 goto out;
1741         ei = PROC_I(inode);
1742         ei->fd = fd;
1743         inode->i_mode = S_IFREG | S_IRUSR;
1744         inode->i_fop = &proc_fdinfo_file_operations;
1745         dentry->d_op = &tid_fd_dentry_operations;
1746         d_add(dentry, inode);
1747         /* Close the race of the process dying before we return the dentry */
1748         if (tid_fd_revalidate(dentry, NULL))
1749                 error = NULL;
1750
1751  out:
1752         return error;
1753 }
1754
1755 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1756                                         struct dentry *dentry,
1757                                         struct nameidata *nd)
1758 {
1759         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1760 }
1761
1762 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1763 {
1764         return proc_readfd_common(filp, dirent, filldir,
1765                                   proc_fdinfo_instantiate);
1766 }
1767
1768 static const struct file_operations proc_fdinfo_operations = {
1769         .read           = generic_read_dir,
1770         .readdir        = proc_readfdinfo,
1771 };
1772
1773 /*
1774  * proc directories can do almost nothing..
1775  */
1776 static const struct inode_operations proc_fdinfo_inode_operations = {
1777         .lookup         = proc_lookupfdinfo,
1778         .setattr        = proc_setattr,
1779 };
1780
1781
1782 static struct dentry *proc_pident_instantiate(struct inode *dir,
1783         struct dentry *dentry, struct task_struct *task, const void *ptr)
1784 {
1785         const struct pid_entry *p = ptr;
1786         struct inode *inode;
1787         struct proc_inode *ei;
1788         struct dentry *error = ERR_PTR(-EINVAL);
1789
1790         inode = proc_pid_make_inode(dir->i_sb, task);
1791         if (!inode)
1792                 goto out;
1793
1794         ei = PROC_I(inode);
1795         inode->i_mode = p->mode;
1796         if (S_ISDIR(inode->i_mode))
1797                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1798         if (p->iop)
1799                 inode->i_op = p->iop;
1800         if (p->fop)
1801                 inode->i_fop = p->fop;
1802         ei->op = p->op;
1803         dentry->d_op = &pid_dentry_operations;
1804         d_add(dentry, inode);
1805         /* Close the race of the process dying before we return the dentry */
1806         if (pid_revalidate(dentry, NULL))
1807                 error = NULL;
1808 out:
1809         return error;
1810 }
1811
1812 static struct dentry *proc_pident_lookup(struct inode *dir,
1813                                          struct dentry *dentry,
1814                                          const struct pid_entry *ents,
1815                                          unsigned int nents)
1816 {
1817         struct inode *inode;
1818         struct dentry *error;
1819         struct task_struct *task = get_proc_task(dir);
1820         const struct pid_entry *p, *last;
1821
1822         error = ERR_PTR(-ENOENT);
1823         inode = NULL;
1824
1825         if (!task)
1826                 goto out_no_task;
1827
1828         /*
1829          * Yes, it does not scale. And it should not. Don't add
1830          * new entries into /proc/<tgid>/ without very good reasons.
1831          */
1832         last = &ents[nents - 1];
1833         for (p = ents; p <= last; p++) {
1834                 if (p->len != dentry->d_name.len)
1835                         continue;
1836                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1837                         break;
1838         }
1839         if (p > last)
1840                 goto out;
1841
1842         error = proc_pident_instantiate(dir, dentry, task, p);
1843 out:
1844         put_task_struct(task);
1845 out_no_task:
1846         return error;
1847 }
1848
1849 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1850         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1851 {
1852         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1853                                 proc_pident_instantiate, task, p);
1854 }
1855
1856 static int proc_pident_readdir(struct file *filp,
1857                 void *dirent, filldir_t filldir,
1858                 const struct pid_entry *ents, unsigned int nents)
1859 {
1860         int i;
1861         struct dentry *dentry = filp->f_path.dentry;
1862         struct inode *inode = dentry->d_inode;
1863         struct task_struct *task = get_proc_task(inode);
1864         const struct pid_entry *p, *last;
1865         ino_t ino;
1866         int ret;
1867
1868         ret = -ENOENT;
1869         if (!task)
1870                 goto out_no_task;
1871
1872         ret = 0;
1873         i = filp->f_pos;
1874         switch (i) {
1875         case 0:
1876                 ino = inode->i_ino;
1877                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1878                         goto out;
1879                 i++;
1880                 filp->f_pos++;
1881                 /* fall through */
1882         case 1:
1883                 ino = parent_ino(dentry);
1884                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1885                         goto out;
1886                 i++;
1887                 filp->f_pos++;
1888                 /* fall through */
1889         default:
1890                 i -= 2;
1891                 if (i >= nents) {
1892                         ret = 1;
1893                         goto out;
1894                 }
1895                 p = ents + i;
1896                 last = &ents[nents - 1];
1897                 while (p <= last) {
1898                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1899                                 goto out;
1900                         filp->f_pos++;
1901                         p++;
1902                 }
1903         }
1904
1905         ret = 1;
1906 out:
1907         put_task_struct(task);
1908 out_no_task:
1909         return ret;
1910 }
1911
1912 #ifdef CONFIG_SECURITY
1913 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1914                                   size_t count, loff_t *ppos)
1915 {
1916         struct inode * inode = file->f_path.dentry->d_inode;
1917         char *p = NULL;
1918         ssize_t length;
1919         struct task_struct *task = get_proc_task(inode);
1920
1921         if (!task)
1922                 return -ESRCH;
1923
1924         length = security_getprocattr(task,
1925                                       (char*)file->f_path.dentry->d_name.name,
1926                                       &p);
1927         put_task_struct(task);
1928         if (length > 0)
1929                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1930         kfree(p);
1931         return length;
1932 }
1933
1934 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1935                                    size_t count, loff_t *ppos)
1936 {
1937         struct inode * inode = file->f_path.dentry->d_inode;
1938         char *page;
1939         ssize_t length;
1940         struct task_struct *task = get_proc_task(inode);
1941
1942         length = -ESRCH;
1943         if (!task)
1944                 goto out_no_task;
1945         if (count > PAGE_SIZE)
1946                 count = PAGE_SIZE;
1947
1948         /* No partial writes. */
1949         length = -EINVAL;
1950         if (*ppos != 0)
1951                 goto out;
1952
1953         length = -ENOMEM;
1954         page = (char*)__get_free_page(GFP_TEMPORARY);
1955         if (!page)
1956                 goto out;
1957
1958         length = -EFAULT;
1959         if (copy_from_user(page, buf, count))
1960                 goto out_free;
1961
1962         length = security_setprocattr(task,
1963                                       (char*)file->f_path.dentry->d_name.name,
1964                                       (void*)page, count);
1965 out_free:
1966         free_page((unsigned long) page);
1967 out:
1968         put_task_struct(task);
1969 out_no_task:
1970         return length;
1971 }
1972
1973 static const struct file_operations proc_pid_attr_operations = {
1974         .read           = proc_pid_attr_read,
1975         .write          = proc_pid_attr_write,
1976 };
1977
1978 static const struct pid_entry attr_dir_stuff[] = {
1979         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1980         REG("prev",       S_IRUGO,         pid_attr),
1981         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1982         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1983         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1984         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1985 };
1986
1987 static int proc_attr_dir_readdir(struct file * filp,
1988                              void * dirent, filldir_t filldir)
1989 {
1990         return proc_pident_readdir(filp,dirent,filldir,
1991                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1992 }
1993
1994 static const struct file_operations proc_attr_dir_operations = {
1995         .read           = generic_read_dir,
1996         .readdir        = proc_attr_dir_readdir,
1997 };
1998
1999 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2000                                 struct dentry *dentry, struct nameidata *nd)
2001 {
2002         return proc_pident_lookup(dir, dentry,
2003                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2004 }
2005
2006 static const struct inode_operations proc_attr_dir_inode_operations = {
2007         .lookup         = proc_attr_dir_lookup,
2008         .getattr        = pid_getattr,
2009         .setattr        = proc_setattr,
2010 };
2011
2012 #endif
2013
2014 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2015 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2016                                          size_t count, loff_t *ppos)
2017 {
2018         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2019         struct mm_struct *mm;
2020         char buffer[PROC_NUMBUF];
2021         size_t len;
2022         int ret;
2023
2024         if (!task)
2025                 return -ESRCH;
2026
2027         ret = 0;
2028         mm = get_task_mm(task);
2029         if (mm) {
2030                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2031                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2032                                 MMF_DUMP_FILTER_SHIFT));
2033                 mmput(mm);
2034                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2035         }
2036
2037         put_task_struct(task);
2038
2039         return ret;
2040 }
2041
2042 static ssize_t proc_coredump_filter_write(struct file *file,
2043                                           const char __user *buf,
2044                                           size_t count,
2045                                           loff_t *ppos)
2046 {
2047         struct task_struct *task;
2048         struct mm_struct *mm;
2049         char buffer[PROC_NUMBUF], *end;
2050         unsigned int val;
2051         int ret;
2052         int i;
2053         unsigned long mask;
2054
2055         ret = -EFAULT;
2056         memset(buffer, 0, sizeof(buffer));
2057         if (count > sizeof(buffer) - 1)
2058                 count = sizeof(buffer) - 1;
2059         if (copy_from_user(buffer, buf, count))
2060                 goto out_no_task;
2061
2062         ret = -EINVAL;
2063         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2064         if (*end == '\n')
2065                 end++;
2066         if (end - buffer == 0)
2067                 goto out_no_task;
2068
2069         ret = -ESRCH;
2070         task = get_proc_task(file->f_dentry->d_inode);
2071         if (!task)
2072                 goto out_no_task;
2073
2074         ret = end - buffer;
2075         mm = get_task_mm(task);
2076         if (!mm)
2077                 goto out_no_mm;
2078
2079         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2080                 if (val & mask)
2081                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2082                 else
2083                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2084         }
2085
2086         mmput(mm);
2087  out_no_mm:
2088         put_task_struct(task);
2089  out_no_task:
2090         return ret;
2091 }
2092
2093 static const struct file_operations proc_coredump_filter_operations = {
2094         .read           = proc_coredump_filter_read,
2095         .write          = proc_coredump_filter_write,
2096 };
2097 #endif
2098
2099 /*
2100  * /proc/self:
2101  */
2102 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2103                               int buflen)
2104 {
2105         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2106         pid_t tgid = task_tgid_nr_ns(current, ns);
2107         char tmp[PROC_NUMBUF];
2108         if (!tgid)
2109                 return -ENOENT;
2110         sprintf(tmp, "%d", tgid);
2111         return vfs_readlink(dentry,buffer,buflen,tmp);
2112 }
2113
2114 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2115 {
2116         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2117         pid_t tgid = task_tgid_nr_ns(current, ns);
2118         char tmp[PROC_NUMBUF];
2119         if (!tgid)
2120                 return ERR_PTR(-ENOENT);
2121         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2122         return ERR_PTR(vfs_follow_link(nd,tmp));
2123 }
2124
2125 static const struct inode_operations proc_self_inode_operations = {
2126         .readlink       = proc_self_readlink,
2127         .follow_link    = proc_self_follow_link,
2128 };
2129
2130 /*
2131  * proc base
2132  *
2133  * These are the directory entries in the root directory of /proc
2134  * that properly belong to the /proc filesystem, as they describe
2135  * describe something that is process related.
2136  */
2137 static const struct pid_entry proc_base_stuff[] = {
2138         NOD("self", S_IFLNK|S_IRWXUGO,
2139                 &proc_self_inode_operations, NULL, {}),
2140 };
2141
2142 /*
2143  *      Exceptional case: normally we are not allowed to unhash a busy
2144  * directory. In this case, however, we can do it - no aliasing problems
2145  * due to the way we treat inodes.
2146  */
2147 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2148 {
2149         struct inode *inode = dentry->d_inode;
2150         struct task_struct *task = get_proc_task(inode);
2151         if (task) {
2152                 put_task_struct(task);
2153                 return 1;
2154         }
2155         d_drop(dentry);
2156         return 0;
2157 }
2158
2159 static struct dentry_operations proc_base_dentry_operations =
2160 {
2161         .d_revalidate   = proc_base_revalidate,
2162         .d_delete       = pid_delete_dentry,
2163 };
2164
2165 static struct dentry *proc_base_instantiate(struct inode *dir,
2166         struct dentry *dentry, struct task_struct *task, const void *ptr)
2167 {
2168         const struct pid_entry *p = ptr;
2169         struct inode *inode;
2170         struct proc_inode *ei;
2171         struct dentry *error = ERR_PTR(-EINVAL);
2172
2173         /* Allocate the inode */
2174         error = ERR_PTR(-ENOMEM);
2175         inode = new_inode(dir->i_sb);
2176         if (!inode)
2177                 goto out;
2178
2179         /* Initialize the inode */
2180         ei = PROC_I(inode);
2181         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2182
2183         /*
2184          * grab the reference to the task.
2185          */
2186         ei->pid = get_task_pid(task, PIDTYPE_PID);
2187         if (!ei->pid)
2188                 goto out_iput;
2189
2190         inode->i_uid = 0;
2191         inode->i_gid = 0;
2192         inode->i_mode = p->mode;
2193         if (S_ISDIR(inode->i_mode))
2194                 inode->i_nlink = 2;
2195         if (S_ISLNK(inode->i_mode))
2196                 inode->i_size = 64;
2197         if (p->iop)
2198                 inode->i_op = p->iop;
2199         if (p->fop)
2200                 inode->i_fop = p->fop;
2201         ei->op = p->op;
2202         dentry->d_op = &proc_base_dentry_operations;
2203         d_add(dentry, inode);
2204         error = NULL;
2205 out:
2206         return error;
2207 out_iput:
2208         iput(inode);
2209         goto out;
2210 }
2211
2212 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2213 {
2214         struct dentry *error;
2215         struct task_struct *task = get_proc_task(dir);
2216         const struct pid_entry *p, *last;
2217
2218         error = ERR_PTR(-ENOENT);
2219
2220         if (!task)
2221                 goto out_no_task;
2222
2223         /* Lookup the directory entry */
2224         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2225         for (p = proc_base_stuff; p <= last; p++) {
2226                 if (p->len != dentry->d_name.len)
2227                         continue;
2228                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2229                         break;
2230         }
2231         if (p > last)
2232                 goto out;
2233
2234         error = proc_base_instantiate(dir, dentry, task, p);
2235
2236 out:
2237         put_task_struct(task);
2238 out_no_task:
2239         return error;
2240 }
2241
2242 static int proc_base_fill_cache(struct file *filp, void *dirent,
2243         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2244 {
2245         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2246                                 proc_base_instantiate, task, p);
2247 }
2248
2249 #ifdef CONFIG_TASK_IO_ACCOUNTING
2250 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2251 {
2252         return sprintf(buffer,
2253 #ifdef CONFIG_TASK_XACCT
2254                         "rchar: %llu\n"
2255                         "wchar: %llu\n"
2256                         "syscr: %llu\n"
2257                         "syscw: %llu\n"
2258 #endif
2259                         "read_bytes: %llu\n"
2260                         "write_bytes: %llu\n"
2261                         "cancelled_write_bytes: %llu\n",
2262 #ifdef CONFIG_TASK_XACCT
2263                         (unsigned long long)task->rchar,
2264                         (unsigned long long)task->wchar,
2265                         (unsigned long long)task->syscr,
2266                         (unsigned long long)task->syscw,
2267 #endif
2268                         (unsigned long long)task->ioac.read_bytes,
2269                         (unsigned long long)task->ioac.write_bytes,
2270                         (unsigned long long)task->ioac.cancelled_write_bytes);
2271 }
2272 #endif
2273
2274 /*
2275  * Thread groups
2276  */
2277 static const struct file_operations proc_task_operations;
2278 static const struct inode_operations proc_task_inode_operations;
2279
2280 static const struct pid_entry tgid_base_stuff[] = {
2281         DIR("task",       S_IRUGO|S_IXUGO, task),
2282         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2283         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2284 #ifdef CONFIG_NET
2285         DIR("net",        S_IRUGO|S_IXUGO, net),
2286 #endif
2287         REG("environ",    S_IRUSR, environ),
2288         INF("auxv",       S_IRUSR, pid_auxv),
2289         ONE("status",     S_IRUGO, pid_status),
2290         INF("limits",     S_IRUSR, pid_limits),
2291 #ifdef CONFIG_SCHED_DEBUG
2292         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2293 #endif
2294         INF("cmdline",    S_IRUGO, pid_cmdline),
2295         ONE("stat",       S_IRUGO, tgid_stat),
2296         ONE("statm",      S_IRUGO, pid_statm),
2297         REG("maps",       S_IRUGO, maps),
2298 #ifdef CONFIG_NUMA
2299         REG("numa_maps",  S_IRUGO, numa_maps),
2300 #endif
2301         REG("mem",        S_IRUSR|S_IWUSR, mem),
2302         LNK("cwd",        cwd),
2303         LNK("root",       root),
2304         LNK("exe",        exe),
2305         REG("mounts",     S_IRUGO, mounts),
2306         REG("mountstats", S_IRUSR, mountstats),
2307 #ifdef CONFIG_PROC_PAGE_MONITOR
2308         REG("clear_refs", S_IWUSR, clear_refs),
2309         REG("smaps",      S_IRUGO, smaps),
2310         REG("pagemap",    S_IRUSR, pagemap),
2311 #endif
2312 #ifdef CONFIG_SECURITY
2313         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2314 #endif
2315 #ifdef CONFIG_KALLSYMS
2316         INF("wchan",      S_IRUGO, pid_wchan),
2317 #endif
2318 #ifdef CONFIG_SCHEDSTATS
2319         INF("schedstat",  S_IRUGO, pid_schedstat),
2320 #endif
2321 #ifdef CONFIG_LATENCYTOP
2322         REG("latency",  S_IRUGO, lstats),
2323 #endif
2324 #ifdef CONFIG_PROC_PID_CPUSET
2325         REG("cpuset",     S_IRUGO, cpuset),
2326 #endif
2327 #ifdef CONFIG_CGROUPS
2328         REG("cgroup",  S_IRUGO, cgroup),
2329 #endif
2330         INF("oom_score",  S_IRUGO, oom_score),
2331         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2332 #ifdef CONFIG_AUDITSYSCALL
2333         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2334         REG("sessionid",  S_IRUSR, sessionid),
2335 #endif
2336 #ifdef CONFIG_FAULT_INJECTION
2337         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2338 #endif
2339 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2340         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2341 #endif
2342 #ifdef CONFIG_TASK_IO_ACCOUNTING
2343         INF("io",       S_IRUGO, pid_io_accounting),
2344 #endif
2345 };
2346
2347 static int proc_tgid_base_readdir(struct file * filp,
2348                              void * dirent, filldir_t filldir)
2349 {
2350         return proc_pident_readdir(filp,dirent,filldir,
2351                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2352 }
2353
2354 static const struct file_operations proc_tgid_base_operations = {
2355         .read           = generic_read_dir,
2356         .readdir        = proc_tgid_base_readdir,
2357 };
2358
2359 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2360         return proc_pident_lookup(dir, dentry,
2361                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2362 }
2363
2364 static const struct inode_operations proc_tgid_base_inode_operations = {
2365         .lookup         = proc_tgid_base_lookup,
2366         .getattr        = pid_getattr,
2367         .setattr        = proc_setattr,
2368 };
2369
2370 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2371 {
2372         struct dentry *dentry, *leader, *dir;
2373         char buf[PROC_NUMBUF];
2374         struct qstr name;
2375
2376         name.name = buf;
2377         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2378         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2379         if (dentry) {
2380                 if (!(current->flags & PF_EXITING))
2381                         shrink_dcache_parent(dentry);
2382                 d_drop(dentry);
2383                 dput(dentry);
2384         }
2385
2386         if (tgid == 0)
2387                 goto out;
2388
2389         name.name = buf;
2390         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2391         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2392         if (!leader)
2393                 goto out;
2394
2395         name.name = "task";
2396         name.len = strlen(name.name);
2397         dir = d_hash_and_lookup(leader, &name);
2398         if (!dir)
2399                 goto out_put_leader;
2400
2401         name.name = buf;
2402         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2403         dentry = d_hash_and_lookup(dir, &name);
2404         if (dentry) {
2405                 shrink_dcache_parent(dentry);
2406                 d_drop(dentry);
2407                 dput(dentry);
2408         }
2409
2410         dput(dir);
2411 out_put_leader:
2412         dput(leader);
2413 out:
2414         return;
2415 }
2416
2417 /**
2418  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2419  * @task: task that should be flushed.
2420  *
2421  * When flushing dentries from proc, one needs to flush them from global
2422  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2423  * in. This call is supposed to do all of this job.
2424  *
2425  * Looks in the dcache for
2426  * /proc/@pid
2427  * /proc/@tgid/task/@pid
2428  * if either directory is present flushes it and all of it'ts children
2429  * from the dcache.
2430  *
2431  * It is safe and reasonable to cache /proc entries for a task until
2432  * that task exits.  After that they just clog up the dcache with
2433  * useless entries, possibly causing useful dcache entries to be
2434  * flushed instead.  This routine is proved to flush those useless
2435  * dcache entries at process exit time.
2436  *
2437  * NOTE: This routine is just an optimization so it does not guarantee
2438  *       that no dcache entries will exist at process exit time it
2439  *       just makes it very unlikely that any will persist.
2440  */
2441
2442 void proc_flush_task(struct task_struct *task)
2443 {
2444         int i;
2445         struct pid *pid, *tgid = NULL;
2446         struct upid *upid;
2447
2448         pid = task_pid(task);
2449         if (thread_group_leader(task))
2450                 tgid = task_tgid(task);
2451
2452         for (i = 0; i <= pid->level; i++) {
2453                 upid = &pid->numbers[i];
2454                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2455                         tgid ? tgid->numbers[i].nr : 0);
2456         }
2457
2458         upid = &pid->numbers[pid->level];
2459         if (upid->nr == 1)
2460                 pid_ns_release_proc(upid->ns);
2461 }
2462
2463 static struct dentry *proc_pid_instantiate(struct inode *dir,
2464                                            struct dentry * dentry,
2465                                            struct task_struct *task, const void *ptr)
2466 {
2467         struct dentry *error = ERR_PTR(-ENOENT);
2468         struct inode *inode;
2469
2470         inode = proc_pid_make_inode(dir->i_sb, task);
2471         if (!inode)
2472                 goto out;
2473
2474         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2475         inode->i_op = &proc_tgid_base_inode_operations;
2476         inode->i_fop = &proc_tgid_base_operations;
2477         inode->i_flags|=S_IMMUTABLE;
2478         inode->i_nlink = 5;
2479 #ifdef CONFIG_SECURITY
2480         inode->i_nlink += 1;
2481 #endif
2482
2483         dentry->d_op = &pid_dentry_operations;
2484
2485         d_add(dentry, inode);
2486         /* Close the race of the process dying before we return the dentry */
2487         if (pid_revalidate(dentry, NULL))
2488                 error = NULL;
2489 out:
2490         return error;
2491 }
2492
2493 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2494 {
2495         struct dentry *result = ERR_PTR(-ENOENT);
2496         struct task_struct *task;
2497         unsigned tgid;
2498         struct pid_namespace *ns;
2499
2500         result = proc_base_lookup(dir, dentry);
2501         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2502                 goto out;
2503
2504         tgid = name_to_int(dentry);
2505         if (tgid == ~0U)
2506                 goto out;
2507
2508         ns = dentry->d_sb->s_fs_info;
2509         rcu_read_lock();
2510         task = find_task_by_pid_ns(tgid, ns);
2511         if (task)
2512                 get_task_struct(task);
2513         rcu_read_unlock();
2514         if (!task)
2515                 goto out;
2516
2517         result = proc_pid_instantiate(dir, dentry, task, NULL);
2518         put_task_struct(task);
2519 out:
2520         return result;
2521 }
2522
2523 /*
2524  * Find the first task with tgid >= tgid
2525  *
2526  */
2527 struct tgid_iter {
2528         unsigned int tgid;
2529         struct task_struct *task;
2530 };
2531 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2532 {
2533         struct pid *pid;
2534
2535         if (iter.task)
2536                 put_task_struct(iter.task);
2537         rcu_read_lock();
2538 retry:
2539         iter.task = NULL;
2540         pid = find_ge_pid(iter.tgid, ns);
2541         if (pid) {
2542                 iter.tgid = pid_nr_ns(pid, ns);
2543                 iter.task = pid_task(pid, PIDTYPE_PID);
2544                 /* What we to know is if the pid we have find is the
2545                  * pid of a thread_group_leader.  Testing for task
2546                  * being a thread_group_leader is the obvious thing
2547                  * todo but there is a window when it fails, due to
2548                  * the pid transfer logic in de_thread.
2549                  *
2550                  * So we perform the straight forward test of seeing
2551                  * if the pid we have found is the pid of a thread
2552                  * group leader, and don't worry if the task we have
2553                  * found doesn't happen to be a thread group leader.
2554                  * As we don't care in the case of readdir.
2555                  */
2556                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2557                         iter.tgid += 1;
2558                         goto retry;
2559                 }
2560                 get_task_struct(iter.task);
2561         }
2562         rcu_read_unlock();
2563         return iter;
2564 }
2565
2566 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2567
2568 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2569         struct tgid_iter iter)
2570 {
2571         char name[PROC_NUMBUF];
2572         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2573         return proc_fill_cache(filp, dirent, filldir, name, len,
2574                                 proc_pid_instantiate, iter.task, NULL);
2575 }
2576
2577 /* for the /proc/ directory itself, after non-process stuff has been done */
2578 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2579 {
2580         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2581         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2582         struct tgid_iter iter;
2583         struct pid_namespace *ns;
2584
2585         if (!reaper)
2586                 goto out_no_task;
2587
2588         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2589                 const struct pid_entry *p = &proc_base_stuff[nr];
2590                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2591                         goto out;
2592         }
2593
2594         ns = filp->f_dentry->d_sb->s_fs_info;
2595         iter.task = NULL;
2596         iter.tgid = filp->f_pos - TGID_OFFSET;
2597         for (iter = next_tgid(ns, iter);
2598              iter.task;
2599              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2600                 filp->f_pos = iter.tgid + TGID_OFFSET;
2601                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2602                         put_task_struct(iter.task);
2603                         goto out;
2604                 }
2605         }
2606         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2607 out:
2608         put_task_struct(reaper);
2609 out_no_task:
2610         return 0;
2611 }
2612
2613 /*
2614  * Tasks
2615  */
2616 static const struct pid_entry tid_base_stuff[] = {
2617         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2618         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2619         REG("environ",   S_IRUSR, environ),
2620         INF("auxv",      S_IRUSR, pid_auxv),
2621         ONE("status",    S_IRUGO, pid_status),
2622         INF("limits",    S_IRUSR, pid_limits),
2623 #ifdef CONFIG_SCHED_DEBUG
2624         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2625 #endif
2626         INF("cmdline",   S_IRUGO, pid_cmdline),
2627         ONE("stat",      S_IRUGO, tid_stat),
2628         ONE("statm",     S_IRUGO, pid_statm),
2629         REG("maps",      S_IRUGO, maps),
2630 #ifdef CONFIG_NUMA
2631         REG("numa_maps", S_IRUGO, numa_maps),
2632 #endif
2633         REG("mem",       S_IRUSR|S_IWUSR, mem),
2634         LNK("cwd",       cwd),
2635         LNK("root",      root),
2636         LNK("exe",       exe),
2637         REG("mounts",    S_IRUGO, mounts),
2638 #ifdef CONFIG_PROC_PAGE_MONITOR
2639         REG("clear_refs", S_IWUSR, clear_refs),
2640         REG("smaps",     S_IRUGO, smaps),
2641         REG("pagemap",    S_IRUSR, pagemap),
2642 #endif
2643 #ifdef CONFIG_SECURITY
2644         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2645 #endif
2646 #ifdef CONFIG_KALLSYMS
2647         INF("wchan",     S_IRUGO, pid_wchan),
2648 #endif
2649 #ifdef CONFIG_SCHEDSTATS
2650         INF("schedstat", S_IRUGO, pid_schedstat),
2651 #endif
2652 #ifdef CONFIG_LATENCYTOP
2653         REG("latency",  S_IRUGO, lstats),
2654 #endif
2655 #ifdef CONFIG_PROC_PID_CPUSET
2656         REG("cpuset",    S_IRUGO, cpuset),
2657 #endif
2658 #ifdef CONFIG_CGROUPS
2659         REG("cgroup",  S_IRUGO, cgroup),
2660 #endif
2661         INF("oom_score", S_IRUGO, oom_score),
2662         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2663 #ifdef CONFIG_AUDITSYSCALL
2664         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2665         REG("sessionid",  S_IRUSR, sessionid),
2666 #endif
2667 #ifdef CONFIG_FAULT_INJECTION
2668         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2669 #endif
2670 };
2671
2672 static int proc_tid_base_readdir(struct file * filp,
2673                              void * dirent, filldir_t filldir)
2674 {
2675         return proc_pident_readdir(filp,dirent,filldir,
2676                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2677 }
2678
2679 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2680         return proc_pident_lookup(dir, dentry,
2681                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2682 }
2683
2684 static const struct file_operations proc_tid_base_operations = {
2685         .read           = generic_read_dir,
2686         .readdir        = proc_tid_base_readdir,
2687 };
2688
2689 static const struct inode_operations proc_tid_base_inode_operations = {
2690         .lookup         = proc_tid_base_lookup,
2691         .getattr        = pid_getattr,
2692         .setattr        = proc_setattr,
2693 };
2694
2695 static struct dentry *proc_task_instantiate(struct inode *dir,
2696         struct dentry *dentry, struct task_struct *task, const void *ptr)
2697 {
2698         struct dentry *error = ERR_PTR(-ENOENT);
2699         struct inode *inode;
2700         inode = proc_pid_make_inode(dir->i_sb, task);
2701
2702         if (!inode)
2703                 goto out;
2704         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2705         inode->i_op = &proc_tid_base_inode_operations;
2706         inode->i_fop = &proc_tid_base_operations;
2707         inode->i_flags|=S_IMMUTABLE;
2708         inode->i_nlink = 4;
2709 #ifdef CONFIG_SECURITY
2710         inode->i_nlink += 1;
2711 #endif
2712
2713         dentry->d_op = &pid_dentry_operations;
2714
2715         d_add(dentry, inode);
2716         /* Close the race of the process dying before we return the dentry */
2717         if (pid_revalidate(dentry, NULL))
2718                 error = NULL;
2719 out:
2720         return error;
2721 }
2722
2723 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2724 {
2725         struct dentry *result = ERR_PTR(-ENOENT);
2726         struct task_struct *task;
2727         struct task_struct *leader = get_proc_task(dir);
2728         unsigned tid;
2729         struct pid_namespace *ns;
2730
2731         if (!leader)
2732                 goto out_no_task;
2733
2734         tid = name_to_int(dentry);
2735         if (tid == ~0U)
2736                 goto out;
2737
2738         ns = dentry->d_sb->s_fs_info;
2739         rcu_read_lock();
2740         task = find_task_by_pid_ns(tid, ns);
2741         if (task)
2742                 get_task_struct(task);
2743         rcu_read_unlock();
2744         if (!task)
2745                 goto out;
2746         if (!same_thread_group(leader, task))
2747                 goto out_drop_task;
2748
2749         result = proc_task_instantiate(dir, dentry, task, NULL);
2750 out_drop_task:
2751         put_task_struct(task);
2752 out:
2753         put_task_struct(leader);
2754 out_no_task:
2755         return result;
2756 }
2757
2758 /*
2759  * Find the first tid of a thread group to return to user space.
2760  *
2761  * Usually this is just the thread group leader, but if the users
2762  * buffer was too small or there was a seek into the middle of the
2763  * directory we have more work todo.
2764  *
2765  * In the case of a short read we start with find_task_by_pid.
2766  *
2767  * In the case of a seek we start with the leader and walk nr
2768  * threads past it.
2769  */
2770 static struct task_struct *first_tid(struct task_struct *leader,
2771                 int tid, int nr, struct pid_namespace *ns)
2772 {
2773         struct task_struct *pos;
2774
2775         rcu_read_lock();
2776         /* Attempt to start with the pid of a thread */
2777         if (tid && (nr > 0)) {
2778                 pos = find_task_by_pid_ns(tid, ns);
2779                 if (pos && (pos->group_leader == leader))
2780                         goto found;
2781         }
2782
2783         /* If nr exceeds the number of threads there is nothing todo */
2784         pos = NULL;
2785         if (nr && nr >= get_nr_threads(leader))
2786                 goto out;
2787
2788         /* If we haven't found our starting place yet start
2789          * with the leader and walk nr threads forward.
2790          */
2791         for (pos = leader; nr > 0; --nr) {
2792                 pos = next_thread(pos);
2793                 if (pos == leader) {
2794                         pos = NULL;
2795                         goto out;
2796                 }
2797         }
2798 found:
2799         get_task_struct(pos);
2800 out:
2801         rcu_read_unlock();
2802         return pos;
2803 }
2804
2805 /*
2806  * Find the next thread in the thread list.
2807  * Return NULL if there is an error or no next thread.
2808  *
2809  * The reference to the input task_struct is released.
2810  */
2811 static struct task_struct *next_tid(struct task_struct *start)
2812 {
2813         struct task_struct *pos = NULL;
2814         rcu_read_lock();
2815         if (pid_alive(start)) {
2816                 pos = next_thread(start);
2817                 if (thread_group_leader(pos))
2818                         pos = NULL;
2819                 else
2820                         get_task_struct(pos);
2821         }
2822         rcu_read_unlock();
2823         put_task_struct(start);
2824         return pos;
2825 }
2826
2827 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2828         struct task_struct *task, int tid)
2829 {
2830         char name[PROC_NUMBUF];
2831         int len = snprintf(name, sizeof(name), "%d", tid);
2832         return proc_fill_cache(filp, dirent, filldir, name, len,
2833                                 proc_task_instantiate, task, NULL);
2834 }
2835
2836 /* for the /proc/TGID/task/ directories */
2837 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2838 {
2839         struct dentry *dentry = filp->f_path.dentry;
2840         struct inode *inode = dentry->d_inode;
2841         struct task_struct *leader = NULL;
2842         struct task_struct *task;
2843         int retval = -ENOENT;
2844         ino_t ino;
2845         int tid;
2846         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2847         struct pid_namespace *ns;
2848
2849         task = get_proc_task(inode);
2850         if (!task)
2851                 goto out_no_task;
2852         rcu_read_lock();
2853         if (pid_alive(task)) {
2854                 leader = task->group_leader;
2855                 get_task_struct(leader);
2856         }
2857         rcu_read_unlock();
2858         put_task_struct(task);
2859         if (!leader)
2860                 goto out_no_task;
2861         retval = 0;
2862
2863         switch (pos) {
2864         case 0:
2865                 ino = inode->i_ino;
2866                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2867                         goto out;
2868                 pos++;
2869                 /* fall through */
2870         case 1:
2871                 ino = parent_ino(dentry);
2872                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2873                         goto out;
2874                 pos++;
2875                 /* fall through */
2876         }
2877
2878         /* f_version caches the tgid value that the last readdir call couldn't
2879          * return. lseek aka telldir automagically resets f_version to 0.
2880          */
2881         ns = filp->f_dentry->d_sb->s_fs_info;
2882         tid = (int)filp->f_version;
2883         filp->f_version = 0;
2884         for (task = first_tid(leader, tid, pos - 2, ns);
2885              task;
2886              task = next_tid(task), pos++) {
2887                 tid = task_pid_nr_ns(task, ns);
2888                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2889                         /* returning this tgid failed, save it as the first
2890                          * pid for the next readir call */
2891                         filp->f_version = (u64)tid;
2892                         put_task_struct(task);
2893                         break;
2894                 }
2895         }
2896 out:
2897         filp->f_pos = pos;
2898         put_task_struct(leader);
2899 out_no_task:
2900         return retval;
2901 }
2902
2903 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2904 {
2905         struct inode *inode = dentry->d_inode;
2906         struct task_struct *p = get_proc_task(inode);
2907         generic_fillattr(inode, stat);
2908
2909         if (p) {
2910                 rcu_read_lock();
2911                 stat->nlink += get_nr_threads(p);
2912                 rcu_read_unlock();
2913                 put_task_struct(p);
2914         }
2915
2916         return 0;
2917 }
2918
2919 static const struct inode_operations proc_task_inode_operations = {
2920         .lookup         = proc_task_lookup,
2921         .getattr        = proc_task_getattr,
2922         .setattr        = proc_setattr,
2923 };
2924
2925 static const struct file_operations proc_task_operations = {
2926         .read           = generic_read_dir,
2927         .readdir        = proc_task_readdir,
2928 };