err.no Git - linux-2.6/blob - 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 mountinfo_open(struct inode *inode, struct file *file)
 608 {
 609         return mounts_open_common(inode, file, &mountinfo_op);
 610 }
 611
 612 static const struct file_operations proc_mountinfo_operations = {
 613         .open           = mountinfo_open,
 614         .read           = seq_read,
 615         .llseek         = seq_lseek,
 616         .release        = mounts_release,
 617         .poll           = mounts_poll,
 618 };
 619
 620 static int mountstats_open(struct inode *inode, struct file *file)
 621 {
 622         return mounts_open_common(inode, file, &mountstats_op);
 623 }
 624
 625 static const struct file_operations proc_mountstats_operations = {
 626         .open           = mountstats_open,
 627         .read           = seq_read,
 628         .llseek         = seq_lseek,
 629         .release        = mounts_release,
 630 };
 631
 632 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 633
 634 static ssize_t proc_info_read(struct file * file, char __user * buf,
 635                           size_t count, loff_t *ppos)
 636 {
 637         struct inode * inode = file->f_path.dentry->d_inode;
 638         unsigned long page;
 639         ssize_t length;
 640         struct task_struct *task = get_proc_task(inode);
 641
 642         length = -ESRCH;
 643         if (!task)
 644                 goto out_no_task;
 645
 646         if (count > PROC_BLOCK_SIZE)
 647                 count = PROC_BLOCK_SIZE;
 648
 649         length = -ENOMEM;
 650         if (!(page = __get_free_page(GFP_TEMPORARY)))
 651                 goto out;
 652
 653         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 654
 655         if (length >= 0)
 656                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 657         free_page(page);
 658 out:
 659         put_task_struct(task);
 660 out_no_task:
 661         return length;
 662 }
 663
 664 static const struct file_operations proc_info_file_operations = {
 665         .read           = proc_info_read,
 666 };
 667
 668 static int proc_single_show(struct seq_file *m, void *v)
 669 {
 670         struct inode *inode = m->private;
 671         struct pid_namespace *ns;
 672         struct pid *pid;
 673         struct task_struct *task;
 674         int ret;
 675
 676         ns = inode->i_sb->s_fs_info;
 677         pid = proc_pid(inode);
 678         task = get_pid_task(pid, PIDTYPE_PID);
 679         if (!task)
 680                 return -ESRCH;
 681
 682         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 683
 684         put_task_struct(task);
 685         return ret;
 686 }
 687
 688 static int proc_single_open(struct inode *inode, struct file *filp)
 689 {
 690         int ret;
 691         ret = single_open(filp, proc_single_show, NULL);
 692         if (!ret) {
 693                 struct seq_file *m = filp->private_data;
 694
 695                 m->private = inode;
 696         }
 697         return ret;
 698 }
 699
 700 static const struct file_operations proc_single_file_operations = {
 701         .open           = proc_single_open,
 702         .read           = seq_read,
 703         .llseek         = seq_lseek,
 704         .release        = single_release,
 705 };
 706
 707 static int mem_open(struct inode* inode, struct file* file)
 708 {
 709         file->private_data = (void*)((long)current->self_exec_id);
 710         return 0;
 711 }
 712
 713 static ssize_t mem_read(struct file * file, char __user * buf,
 714                         size_t count, loff_t *ppos)
 715 {
 716         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 717         char *page;
 718         unsigned long src = *ppos;
 719         int ret = -ESRCH;
 720         struct mm_struct *mm;
 721
 722         if (!task)
 723                 goto out_no_task;
 724
 725         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 726                 goto out;
 727
 728         ret = -ENOMEM;
 729         page = (char *)__get_free_page(GFP_TEMPORARY);
 730         if (!page)
 731                 goto out;
 732
 733         ret = 0;
 734
 735         mm = get_task_mm(task);
 736         if (!mm)
 737                 goto out_free;
 738
 739         ret = -EIO;
 740
 741         if (file->private_data != (void*)((long)current->self_exec_id))
 742                 goto out_put;
 743
 744         ret = 0;
 745
 746         while (count > 0) {
 747                 int this_len, retval;
 748
 749                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 750                 retval = access_process_vm(task, src, page, this_len, 0);
 751                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 752                         if (!ret)
 753                                 ret = -EIO;
 754                         break;
 755                 }
 756
 757                 if (copy_to_user(buf, page, retval)) {
 758                         ret = -EFAULT;
 759                         break;
 760                 }
 761
 762                 ret += retval;
 763                 src += retval;
 764                 buf += retval;
 765                 count -= retval;
 766         }
 767         *ppos = src;
 768
 769 out_put:
 770         mmput(mm);
 771 out_free:
 772         free_page((unsigned long) page);
 773 out:
 774         put_task_struct(task);
 775 out_no_task:
 776         return ret;
 777 }
 778
 779 #define mem_write NULL
 780
 781 #ifndef mem_write
 782 /* This is a security hazard */
 783 static ssize_t mem_write(struct file * file, const char __user *buf,
 784                          size_t count, loff_t *ppos)
 785 {
 786         int copied;
 787         char *page;
 788         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 789         unsigned long dst = *ppos;
 790
 791         copied = -ESRCH;
 792         if (!task)
 793                 goto out_no_task;
 794
 795         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 796                 goto out;
 797
 798         copied = -ENOMEM;
 799         page = (char *)__get_free_page(GFP_TEMPORARY);
 800         if (!page)
 801                 goto out;
 802
 803         copied = 0;
 804         while (count > 0) {
 805                 int this_len, retval;
 806
 807                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 808                 if (copy_from_user(page, buf, this_len)) {
 809                         copied = -EFAULT;
 810                         break;
 811                 }
 812                 retval = access_process_vm(task, dst, page, this_len, 1);
 813                 if (!retval) {
 814                         if (!copied)
 815                                 copied = -EIO;
 816                         break;
 817                 }
 818                 copied += retval;
 819                 buf += retval;
 820                 dst += retval;
 821                 count -= retval;
 822         }
 823         *ppos = dst;
 824         free_page((unsigned long) page);
 825 out:
 826         put_task_struct(task);
 827 out_no_task:
 828         return copied;
 829 }
 830 #endif
 831
 832 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 833 {
 834         switch (orig) {
 835         case 0:
 836                 file->f_pos = offset;
 837                 break;
 838         case 1:
 839                 file->f_pos += offset;
 840                 break;
 841         default:
 842                 return -EINVAL;
 843         }
 844         force_successful_syscall_return();
 845         return file->f_pos;
 846 }
 847
 848 static const struct file_operations proc_mem_operations = {
 849         .llseek         = mem_lseek,
 850         .read           = mem_read,
 851         .write          = mem_write,
 852         .open           = mem_open,
 853 };
 854
 855 static ssize_t environ_read(struct file *file, char __user *buf,
 856                         size_t count, loff_t *ppos)
 857 {
 858         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 859         char *page;
 860         unsigned long src = *ppos;
 861         int ret = -ESRCH;
 862         struct mm_struct *mm;
 863
 864         if (!task)
 865                 goto out_no_task;
 866
 867         if (!ptrace_may_attach(task))
 868                 goto out;
 869
 870         ret = -ENOMEM;
 871         page = (char *)__get_free_page(GFP_TEMPORARY);
 872         if (!page)
 873                 goto out;
 874
 875         ret = 0;
 876
 877         mm = get_task_mm(task);
 878         if (!mm)
 879                 goto out_free;
 880
 881         while (count > 0) {
 882                 int this_len, retval, max_len;
 883
 884                 this_len = mm->env_end - (mm->env_start + src);
 885
 886                 if (this_len <= 0)
 887                         break;
 888
 889                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 890                 this_len = (this_len > max_len) ? max_len : this_len;
 891
 892                 retval = access_process_vm(task, (mm->env_start + src),
 893                         page, this_len, 0);
 894
 895                 if (retval <= 0) {
 896                         ret = retval;
 897                         break;
 898                 }
 899
 900                 if (copy_to_user(buf, page, retval)) {
 901                         ret = -EFAULT;
 902                         break;
 903                 }
 904
 905                 ret += retval;
 906                 src += retval;
 907                 buf += retval;
 908                 count -= retval;
 909         }
 910         *ppos = src;
 911
 912         mmput(mm);
 913 out_free:
 914         free_page((unsigned long) page);
 915 out:
 916         put_task_struct(task);
 917 out_no_task:
 918         return ret;
 919 }
 920
 921 static const struct file_operations proc_environ_operations = {
 922         .read           = environ_read,
 923 };
 924
 925 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 926                                 size_t count, loff_t *ppos)
 927 {
 928         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 929         char buffer[PROC_NUMBUF];
 930         size_t len;
 931         int oom_adjust;
 932
 933         if (!task)
 934                 return -ESRCH;
 935         oom_adjust = task->oomkilladj;
 936         put_task_struct(task);
 937
 938         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 939
 940         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 941 }
 942
 943 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 944                                 size_t count, loff_t *ppos)
 945 {
 946         struct task_struct *task;
 947         char buffer[PROC_NUMBUF], *end;
 948         int oom_adjust;
 949
 950         memset(buffer, 0, sizeof(buffer));
 951         if (count > sizeof(buffer) - 1)
 952                 count = sizeof(buffer) - 1;
 953         if (copy_from_user(buffer, buf, count))
 954                 return -EFAULT;
 955         oom_adjust = simple_strtol(buffer, &end, 0);
 956         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 957              oom_adjust != OOM_DISABLE)
 958                 return -EINVAL;
 959         if (*end == '\n')
 960                 end++;
 961         task = get_proc_task(file->f_path.dentry->d_inode);
 962         if (!task)
 963                 return -ESRCH;
 964         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
 965                 put_task_struct(task);
 966                 return -EACCES;
 967         }
 968         task->oomkilladj = oom_adjust;
 969         put_task_struct(task);
 970         if (end - buffer == 0)
 971                 return -EIO;
 972         return end - buffer;
 973 }
 974
 975 static const struct file_operations proc_oom_adjust_operations = {
 976         .read           = oom_adjust_read,
 977         .write          = oom_adjust_write,
 978 };
 979
 980 #ifdef CONFIG_AUDITSYSCALL
 981 #define TMPBUFLEN 21
 982 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 983                                   size_t count, loff_t *ppos)
 984 {
 985         struct inode * inode = file->f_path.dentry->d_inode;
 986         struct task_struct *task = get_proc_task(inode);
 987         ssize_t length;
 988         char tmpbuf[TMPBUFLEN];
 989
 990         if (!task)
 991                 return -ESRCH;
 992         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 993                                 audit_get_loginuid(task));
 994         put_task_struct(task);
 995         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 996 }
 997
 998 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 999                                    size_t count, loff_t *ppos)
1000 {
1001         struct inode * inode = file->f_path.dentry->d_inode;
1002         char *page, *tmp;
1003         ssize_t length;
1004         uid_t loginuid;
1005
1006         if (!capable(CAP_AUDIT_CONTROL))
1007                 return -EPERM;
1008
1009         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1010                 return -EPERM;
1011
1012         if (count >= PAGE_SIZE)
1013                 count = PAGE_SIZE - 1;
1014
1015         if (*ppos != 0) {
1016                 /* No partial writes. */
1017                 return -EINVAL;
1018         }
1019         page = (char*)__get_free_page(GFP_TEMPORARY);
1020         if (!page)
1021                 return -ENOMEM;
1022         length = -EFAULT;
1023         if (copy_from_user(page, buf, count))
1024                 goto out_free_page;
1025
1026         page[count] = '\0';
1027         loginuid = simple_strtoul(page, &tmp, 10);
1028         if (tmp == page) {
1029                 length = -EINVAL;
1030                 goto out_free_page;
1031
1032         }
1033         length = audit_set_loginuid(current, loginuid);
1034         if (likely(length == 0))
1035                 length = count;
1036
1037 out_free_page:
1038         free_page((unsigned long) page);
1039         return length;
1040 }
1041
1042 static const struct file_operations proc_loginuid_operations = {
1043         .read           = proc_loginuid_read,
1044         .write          = proc_loginuid_write,
1045 };
1046
1047 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1048                                   size_t count, loff_t *ppos)
1049 {
1050         struct inode * inode = file->f_path.dentry->d_inode;
1051         struct task_struct *task = get_proc_task(inode);
1052         ssize_t length;
1053         char tmpbuf[TMPBUFLEN];
1054
1055         if (!task)
1056                 return -ESRCH;
1057         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1058                                 audit_get_sessionid(task));
1059         put_task_struct(task);
1060         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1061 }
1062
1063 static const struct file_operations proc_sessionid_operations = {
1064         .read           = proc_sessionid_read,
1065 };
1066 #endif
1067
1068 #ifdef CONFIG_FAULT_INJECTION
1069 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1070                                       size_t count, loff_t *ppos)
1071 {
1072         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1073         char buffer[PROC_NUMBUF];
1074         size_t len;
1075         int make_it_fail;
1076
1077         if (!task)
1078                 return -ESRCH;
1079         make_it_fail = task->make_it_fail;
1080         put_task_struct(task);
1081
1082         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1083
1084         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1085 }
1086
1087 static ssize_t proc_fault_inject_write(struct file * file,
1088                         const char __user * buf, size_t count, loff_t *ppos)
1089 {
1090         struct task_struct *task;
1091         char buffer[PROC_NUMBUF], *end;
1092         int make_it_fail;
1093
1094         if (!capable(CAP_SYS_RESOURCE))
1095                 return -EPERM;
1096         memset(buffer, 0, sizeof(buffer));
1097         if (count > sizeof(buffer) - 1)
1098                 count = sizeof(buffer) - 1;
1099         if (copy_from_user(buffer, buf, count))
1100                 return -EFAULT;
1101         make_it_fail = simple_strtol(buffer, &end, 0);
1102         if (*end == '\n')
1103                 end++;
1104         task = get_proc_task(file->f_dentry->d_inode);
1105         if (!task)
1106                 return -ESRCH;
1107         task->make_it_fail = make_it_fail;
1108         put_task_struct(task);
1109         if (end - buffer == 0)
1110                 return -EIO;
1111         return end - buffer;
1112 }
1113
1114 static const struct file_operations proc_fault_inject_operations = {
1115         .read           = proc_fault_inject_read,
1116         .write          = proc_fault_inject_write,
1117 };
1118 #endif
1119
1120
1121 #ifdef CONFIG_SCHED_DEBUG
1122 /*
1123  * Print out various scheduling related per-task fields:
1124  */
1125 static int sched_show(struct seq_file *m, void *v)
1126 {
1127         struct inode *inode = m->private;
1128         struct task_struct *p;
1129
1130         WARN_ON(!inode);
1131
1132         p = get_proc_task(inode);
1133         if (!p)
1134                 return -ESRCH;
1135         proc_sched_show_task(p, m);
1136
1137         put_task_struct(p);
1138
1139         return 0;
1140 }
1141
1142 static ssize_t
1143 sched_write(struct file *file, const char __user *buf,
1144             size_t count, loff_t *offset)
1145 {
1146         struct inode *inode = file->f_path.dentry->d_inode;
1147         struct task_struct *p;
1148
1149         WARN_ON(!inode);
1150
1151         p = get_proc_task(inode);
1152         if (!p)
1153                 return -ESRCH;
1154         proc_sched_set_task(p);
1155
1156         put_task_struct(p);
1157
1158         return count;
1159 }
1160
1161 static int sched_open(struct inode *inode, struct file *filp)
1162 {
1163         int ret;
1164
1165         ret = single_open(filp, sched_show, NULL);
1166         if (!ret) {
1167                 struct seq_file *m = filp->private_data;
1168
1169                 m->private = inode;
1170         }
1171         return ret;
1172 }
1173
1174 static const struct file_operations proc_pid_sched_operations = {
1175         .open           = sched_open,
1176         .read           = seq_read,
1177         .write          = sched_write,
1178         .llseek         = seq_lseek,
1179         .release        = single_release,
1180 };
1181
1182 #endif
1183
1184 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1185 {
1186         struct inode *inode = dentry->d_inode;
1187         int error = -EACCES;
1188
1189         /* We don't need a base pointer in the /proc filesystem */
1190         path_put(&nd->path);
1191
1192         /* Are we allowed to snoop on the tasks file descriptors? */
1193         if (!proc_fd_access_allowed(inode))
1194                 goto out;
1195
1196         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1197         nd->last_type = LAST_BIND;
1198 out:
1199         return ERR_PTR(error);
1200 }
1201
1202 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1203 {
1204         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1205         char *pathname;
1206         int len;
1207
1208         if (!tmp)
1209                 return -ENOMEM;
1210
1211         pathname = d_path(path, tmp, PAGE_SIZE);
1212         len = PTR_ERR(pathname);
1213         if (IS_ERR(pathname))
1214                 goto out;
1215         len = tmp + PAGE_SIZE - 1 - pathname;
1216
1217         if (len > buflen)
1218                 len = buflen;
1219         if (copy_to_user(buffer, pathname, len))
1220                 len = -EFAULT;
1221  out:
1222         free_page((unsigned long)tmp);
1223         return len;
1224 }
1225
1226 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1227 {
1228         int error = -EACCES;
1229         struct inode *inode = dentry->d_inode;
1230         struct path path;
1231
1232         /* Are we allowed to snoop on the tasks file descriptors? */
1233         if (!proc_fd_access_allowed(inode))
1234                 goto out;
1235
1236         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1237         if (error)
1238                 goto out;
1239
1240         error = do_proc_readlink(&path, buffer, buflen);
1241         path_put(&path);
1242 out:
1243         return error;
1244 }
1245
1246 static const struct inode_operations proc_pid_link_inode_operations = {
1247         .readlink       = proc_pid_readlink,
1248         .follow_link    = proc_pid_follow_link,
1249         .setattr        = proc_setattr,
1250 };
1251
1252
1253 /* building an inode */
1254
1255 static int task_dumpable(struct task_struct *task)
1256 {
1257         int dumpable = 0;
1258         struct mm_struct *mm;
1259
1260         task_lock(task);
1261         mm = task->mm;
1262         if (mm)
1263                 dumpable = get_dumpable(mm);
1264         task_unlock(task);
1265         if(dumpable == 1)
1266                 return 1;
1267         return 0;
1268 }
1269
1270
1271 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1272 {
1273         struct inode * inode;
1274         struct proc_inode *ei;
1275
1276         /* We need a new inode */
1277
1278         inode = new_inode(sb);
1279         if (!inode)
1280                 goto out;
1281
1282         /* Common stuff */
1283         ei = PROC_I(inode);
1284         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1285         inode->i_op = &proc_def_inode_operations;
1286
1287         /*
1288          * grab the reference to task.
1289          */
1290         ei->pid = get_task_pid(task, PIDTYPE_PID);
1291         if (!ei->pid)
1292                 goto out_unlock;
1293
1294         inode->i_uid = 0;
1295         inode->i_gid = 0;
1296         if (task_dumpable(task)) {
1297                 inode->i_uid = task->euid;
1298                 inode->i_gid = task->egid;
1299         }
1300         security_task_to_inode(task, inode);
1301
1302 out:
1303         return inode;
1304
1305 out_unlock:
1306         iput(inode);
1307         return NULL;
1308 }
1309
1310 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1311 {
1312         struct inode *inode = dentry->d_inode;
1313         struct task_struct *task;
1314         generic_fillattr(inode, stat);
1315
1316         rcu_read_lock();
1317         stat->uid = 0;
1318         stat->gid = 0;
1319         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1320         if (task) {
1321                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1322                     task_dumpable(task)) {
1323                         stat->uid = task->euid;
1324                         stat->gid = task->egid;
1325                 }
1326         }
1327         rcu_read_unlock();
1328         return 0;
1329 }
1330
1331 /* dentry stuff */
1332
1333 /*
1334  *      Exceptional case: normally we are not allowed to unhash a busy
1335  * directory. In this case, however, we can do it - no aliasing problems
1336  * due to the way we treat inodes.
1337  *
1338  * Rewrite the inode's ownerships here because the owning task may have
1339  * performed a setuid(), etc.
1340  *
1341  * Before the /proc/pid/status file was created the only way to read
1342  * the effective uid of a /process was to stat /proc/pid.  Reading
1343  * /proc/pid/status is slow enough that procps and other packages
1344  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1345  * made this apply to all per process world readable and executable
1346  * directories.
1347  */
1348 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1349 {
1350         struct inode *inode = dentry->d_inode;
1351         struct task_struct *task = get_proc_task(inode);
1352         if (task) {
1353                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1354                     task_dumpable(task)) {
1355                         inode->i_uid = task->euid;
1356                         inode->i_gid = task->egid;
1357                 } else {
1358                         inode->i_uid = 0;
1359                         inode->i_gid = 0;
1360                 }
1361                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1362                 security_task_to_inode(task, inode);
1363                 put_task_struct(task);
1364                 return 1;
1365         }
1366         d_drop(dentry);
1367         return 0;
1368 }
1369
1370 static int pid_delete_dentry(struct dentry * dentry)
1371 {
1372         /* Is the task we represent dead?
1373          * If so, then don't put the dentry on the lru list,
1374          * kill it immediately.
1375          */
1376         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1377 }
1378
1379 static struct dentry_operations pid_dentry_operations =
1380 {
1381         .d_revalidate   = pid_revalidate,
1382         .d_delete       = pid_delete_dentry,
1383 };
1384
1385 /* Lookups */
1386
1387 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1388                                 struct task_struct *, const void *);
1389
1390 /*
1391  * Fill a directory entry.
1392  *
1393  * If possible create the dcache entry and derive our inode number and
1394  * file type from dcache entry.
1395  *
1396  * Since all of the proc inode numbers are dynamically generated, the inode
1397  * numbers do not exist until the inode is cache.  This means creating the
1398  * the dcache entry in readdir is necessary to keep the inode numbers
1399  * reported by readdir in sync with the inode numbers reported
1400  * by stat.
1401  */
1402 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1403         char *name, int len,
1404         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1405 {
1406         struct dentry *child, *dir = filp->f_path.dentry;
1407         struct inode *inode;
1408         struct qstr qname;
1409         ino_t ino = 0;
1410         unsigned type = DT_UNKNOWN;
1411
1412         qname.name = name;
1413         qname.len  = len;
1414         qname.hash = full_name_hash(name, len);
1415
1416         child = d_lookup(dir, &qname);
1417         if (!child) {
1418                 struct dentry *new;
1419                 new = d_alloc(dir, &qname);
1420                 if (new) {
1421                         child = instantiate(dir->d_inode, new, task, ptr);
1422                         if (child)
1423                                 dput(new);
1424                         else
1425                                 child = new;
1426                 }
1427         }
1428         if (!child || IS_ERR(child) || !child->d_inode)
1429                 goto end_instantiate;
1430         inode = child->d_inode;
1431         if (inode) {
1432                 ino = inode->i_ino;
1433                 type = inode->i_mode >> 12;
1434         }
1435         dput(child);
1436 end_instantiate:
1437         if (!ino)
1438                 ino = find_inode_number(dir, &qname);
1439         if (!ino)
1440                 ino = 1;
1441         return filldir(dirent, name, len, filp->f_pos, ino, type);
1442 }
1443
1444 static unsigned name_to_int(struct dentry *dentry)
1445 {
1446         const char *name = dentry->d_name.name;
1447         int len = dentry->d_name.len;
1448         unsigned n = 0;
1449
1450         if (len > 1 && *name == '0')
1451                 goto out;
1452         while (len-- > 0) {
1453                 unsigned c = *name++ - '0';
1454                 if (c > 9)
1455                         goto out;
1456                 if (n >= (~0U-9)/10)
1457                         goto out;
1458                 n *= 10;
1459                 n += c;
1460         }
1461         return n;
1462 out:
1463         return ~0U;
1464 }
1465
1466 #define PROC_FDINFO_MAX 64
1467
1468 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1469 {
1470         struct task_struct *task = get_proc_task(inode);
1471         struct files_struct *files = NULL;
1472         struct file *file;
1473         int fd = proc_fd(inode);
1474
1475         if (task) {
1476                 files = get_files_struct(task);
1477                 put_task_struct(task);
1478         }
1479         if (files) {
1480                 /*
1481                  * We are not taking a ref to the file structure, so we must
1482                  * hold ->file_lock.
1483                  */
1484                 spin_lock(&files->file_lock);
1485                 file = fcheck_files(files, fd);
1486                 if (file) {
1487                         if (path) {
1488                                 *path = file->f_path;
1489                                 path_get(&file->f_path);
1490                         }
1491                         if (info)
1492                                 snprintf(info, PROC_FDINFO_MAX,
1493                                          "pos:\t%lli\n"
1494                                          "flags:\t0%o\n",
1495                                          (long long) file->f_pos,
1496                                          file->f_flags);
1497                         spin_unlock(&files->file_lock);
1498                         put_files_struct(files);
1499                         return 0;
1500                 }
1501                 spin_unlock(&files->file_lock);
1502                 put_files_struct(files);
1503         }
1504         return -ENOENT;
1505 }
1506
1507 static int proc_fd_link(struct inode *inode, struct path *path)
1508 {
1509         return proc_fd_info(inode, path, NULL);
1510 }
1511
1512 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1513 {
1514         struct inode *inode = dentry->d_inode;
1515         struct task_struct *task = get_proc_task(inode);
1516         int fd = proc_fd(inode);
1517         struct files_struct *files;
1518
1519         if (task) {
1520                 files = get_files_struct(task);
1521                 if (files) {
1522                         rcu_read_lock();
1523                         if (fcheck_files(files, fd)) {
1524                                 rcu_read_unlock();
1525                                 put_files_struct(files);
1526                                 if (task_dumpable(task)) {
1527                                         inode->i_uid = task->euid;
1528                                         inode->i_gid = task->egid;
1529                                 } else {
1530                                         inode->i_uid = 0;
1531                                         inode->i_gid = 0;
1532                                 }
1533                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1534                                 security_task_to_inode(task, inode);
1535                                 put_task_struct(task);
1536                                 return 1;
1537                         }
1538                         rcu_read_unlock();
1539                         put_files_struct(files);
1540                 }
1541                 put_task_struct(task);
1542         }
1543         d_drop(dentry);
1544         return 0;
1545 }
1546
1547 static struct dentry_operations tid_fd_dentry_operations =
1548 {
1549         .d_revalidate   = tid_fd_revalidate,
1550         .d_delete       = pid_delete_dentry,
1551 };
1552
1553 static struct dentry *proc_fd_instantiate(struct inode *dir,
1554         struct dentry *dentry, struct task_struct *task, const void *ptr)
1555 {
1556         unsigned fd = *(const unsigned *)ptr;
1557         struct file *file;
1558         struct files_struct *files;
1559         struct inode *inode;
1560         struct proc_inode *ei;
1561         struct dentry *error = ERR_PTR(-ENOENT);
1562
1563         inode = proc_pid_make_inode(dir->i_sb, task);
1564         if (!inode)
1565                 goto out;
1566         ei = PROC_I(inode);
1567         ei->fd = fd;
1568         files = get_files_struct(task);
1569         if (!files)
1570                 goto out_iput;
1571         inode->i_mode = S_IFLNK;
1572
1573         /*
1574          * We are not taking a ref to the file structure, so we must
1575          * hold ->file_lock.
1576          */
1577         spin_lock(&files->file_lock);
1578         file = fcheck_files(files, fd);
1579         if (!file)
1580                 goto out_unlock;
1581         if (file->f_mode & 1)
1582                 inode->i_mode |= S_IRUSR | S_IXUSR;
1583         if (file->f_mode & 2)
1584                 inode->i_mode |= S_IWUSR | S_IXUSR;
1585         spin_unlock(&files->file_lock);
1586         put_files_struct(files);
1587
1588         inode->i_op = &proc_pid_link_inode_operations;
1589         inode->i_size = 64;
1590         ei->op.proc_get_link = proc_fd_link;
1591         dentry->d_op = &tid_fd_dentry_operations;
1592         d_add(dentry, inode);
1593         /* Close the race of the process dying before we return the dentry */
1594         if (tid_fd_revalidate(dentry, NULL))
1595                 error = NULL;
1596
1597  out:
1598         return error;
1599 out_unlock:
1600         spin_unlock(&files->file_lock);
1601         put_files_struct(files);
1602 out_iput:
1603         iput(inode);
1604         goto out;
1605 }
1606
1607 static struct dentry *proc_lookupfd_common(struct inode *dir,
1608                                            struct dentry *dentry,
1609                                            instantiate_t instantiate)
1610 {
1611         struct task_struct *task = get_proc_task(dir);
1612         unsigned fd = name_to_int(dentry);
1613         struct dentry *result = ERR_PTR(-ENOENT);
1614
1615         if (!task)
1616                 goto out_no_task;
1617         if (fd == ~0U)
1618                 goto out;
1619
1620         result = instantiate(dir, dentry, task, &fd);
1621 out:
1622         put_task_struct(task);
1623 out_no_task:
1624         return result;
1625 }
1626
1627 static int proc_readfd_common(struct file * filp, void * dirent,
1628                               filldir_t filldir, instantiate_t instantiate)
1629 {
1630         struct dentry *dentry = filp->f_path.dentry;
1631         struct inode *inode = dentry->d_inode;
1632         struct task_struct *p = get_proc_task(inode);
1633         unsigned int fd, ino;
1634         int retval;
1635         struct files_struct * files;
1636
1637         retval = -ENOENT;
1638         if (!p)
1639                 goto out_no_task;
1640         retval = 0;
1641
1642         fd = filp->f_pos;
1643         switch (fd) {
1644                 case 0:
1645                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1646                                 goto out;
1647                         filp->f_pos++;
1648                 case 1:
1649                         ino = parent_ino(dentry);
1650                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1651                                 goto out;
1652                         filp->f_pos++;
1653                 default:
1654                         files = get_files_struct(p);
1655                         if (!files)
1656                                 goto out;
1657                         rcu_read_lock();
1658                         for (fd = filp->f_pos-2;
1659                              fd < files_fdtable(files)->max_fds;
1660                              fd++, filp->f_pos++) {
1661                                 char name[PROC_NUMBUF];
1662                                 int len;
1663
1664                                 if (!fcheck_files(files, fd))
1665                                         continue;
1666                                 rcu_read_unlock();
1667
1668                                 len = snprintf(name, sizeof(name), "%d", fd);
1669                                 if (proc_fill_cache(filp, dirent, filldir,
1670                                                     name, len, instantiate,
1671                                                     p, &fd) < 0) {
1672                                         rcu_read_lock();
1673                                         break;
1674                                 }
1675                                 rcu_read_lock();
1676                         }
1677                         rcu_read_unlock();
1678                         put_files_struct(files);
1679         }
1680 out:
1681         put_task_struct(p);
1682 out_no_task:
1683         return retval;
1684 }
1685
1686 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1687                                     struct nameidata *nd)
1688 {
1689         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1690 }
1691
1692 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1693 {
1694         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1695 }
1696
1697 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1698                                       size_t len, loff_t *ppos)
1699 {
1700         char tmp[PROC_FDINFO_MAX];
1701         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1702         if (!err)
1703                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1704         return err;
1705 }
1706
1707 static const struct file_operations proc_fdinfo_file_operations = {
1708         .open           = nonseekable_open,
1709         .read           = proc_fdinfo_read,
1710 };
1711
1712 static const struct file_operations proc_fd_operations = {
1713         .read           = generic_read_dir,
1714         .readdir        = proc_readfd,
1715 };
1716
1717 /*
1718  * /proc/pid/fd needs a special permission handler so that a process can still
1719  * access /proc/self/fd after it has executed a setuid().
1720  */
1721 static int proc_fd_permission(struct inode *inode, int mask,
1722                                 struct nameidata *nd)
1723 {
1724         int rv;
1725
1726         rv = generic_permission(inode, mask, NULL);
1727         if (rv == 0)
1728                 return 0;
1729         if (task_pid(current) == proc_pid(inode))
1730                 rv = 0;
1731         return rv;
1732 }
1733
1734 /*
1735  * proc directories can do almost nothing..
1736  */
1737 static const struct inode_operations proc_fd_inode_operations = {
1738         .lookup         = proc_lookupfd,
1739         .permission     = proc_fd_permission,
1740         .setattr        = proc_setattr,
1741 };
1742
1743 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1744         struct dentry *dentry, struct task_struct *task, const void *ptr)
1745 {
1746         unsigned fd = *(unsigned *)ptr;
1747         struct inode *inode;
1748         struct proc_inode *ei;
1749         struct dentry *error = ERR_PTR(-ENOENT);
1750
1751         inode = proc_pid_make_inode(dir->i_sb, task);
1752         if (!inode)
1753                 goto out;
1754         ei = PROC_I(inode);
1755         ei->fd = fd;
1756         inode->i_mode = S_IFREG | S_IRUSR;
1757         inode->i_fop = &proc_fdinfo_file_operations;
1758         dentry->d_op = &tid_fd_dentry_operations;
1759         d_add(dentry, inode);
1760         /* Close the race of the process dying before we return the dentry */
1761         if (tid_fd_revalidate(dentry, NULL))
1762                 error = NULL;
1763
1764  out:
1765         return error;
1766 }
1767
1768 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1769                                         struct dentry *dentry,
1770                                         struct nameidata *nd)
1771 {
1772         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1773 }
1774
1775 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1776 {
1777         return proc_readfd_common(filp, dirent, filldir,
1778                                   proc_fdinfo_instantiate);
1779 }
1780
1781 static const struct file_operations proc_fdinfo_operations = {
1782         .read           = generic_read_dir,
1783         .readdir        = proc_readfdinfo,
1784 };
1785
1786 /*
1787  * proc directories can do almost nothing..
1788  */
1789 static const struct inode_operations proc_fdinfo_inode_operations = {
1790         .lookup         = proc_lookupfdinfo,
1791         .setattr        = proc_setattr,
1792 };
1793
1794
1795 static struct dentry *proc_pident_instantiate(struct inode *dir,
1796         struct dentry *dentry, struct task_struct *task, const void *ptr)
1797 {
1798         const struct pid_entry *p = ptr;
1799         struct inode *inode;
1800         struct proc_inode *ei;
1801         struct dentry *error = ERR_PTR(-EINVAL);
1802
1803         inode = proc_pid_make_inode(dir->i_sb, task);
1804         if (!inode)
1805                 goto out;
1806
1807         ei = PROC_I(inode);
1808         inode->i_mode = p->mode;
1809         if (S_ISDIR(inode->i_mode))
1810                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1811         if (p->iop)
1812                 inode->i_op = p->iop;
1813         if (p->fop)
1814                 inode->i_fop = p->fop;
1815         ei->op = p->op;
1816         dentry->d_op = &pid_dentry_operations;
1817         d_add(dentry, inode);
1818         /* Close the race of the process dying before we return the dentry */
1819         if (pid_revalidate(dentry, NULL))
1820                 error = NULL;
1821 out:
1822         return error;
1823 }
1824
1825 static struct dentry *proc_pident_lookup(struct inode *dir,
1826                                          struct dentry *dentry,
1827                                          const struct pid_entry *ents,
1828                                          unsigned int nents)
1829 {
1830         struct inode *inode;
1831         struct dentry *error;
1832         struct task_struct *task = get_proc_task(dir);
1833         const struct pid_entry *p, *last;
1834
1835         error = ERR_PTR(-ENOENT);
1836         inode = NULL;
1837
1838         if (!task)
1839                 goto out_no_task;
1840
1841         /*
1842          * Yes, it does not scale. And it should not. Don't add
1843          * new entries into /proc/<tgid>/ without very good reasons.
1844          */
1845         last = &ents[nents - 1];
1846         for (p = ents; p <= last; p++) {
1847                 if (p->len != dentry->d_name.len)
1848                         continue;
1849                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1850                         break;
1851         }
1852         if (p > last)
1853                 goto out;
1854
1855         error = proc_pident_instantiate(dir, dentry, task, p);
1856 out:
1857         put_task_struct(task);
1858 out_no_task:
1859         return error;
1860 }
1861
1862 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1863         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1864 {
1865         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1866                                 proc_pident_instantiate, task, p);
1867 }
1868
1869 static int proc_pident_readdir(struct file *filp,
1870                 void *dirent, filldir_t filldir,
1871                 const struct pid_entry *ents, unsigned int nents)
1872 {
1873         int i;
1874         struct dentry *dentry = filp->f_path.dentry;
1875         struct inode *inode = dentry->d_inode;
1876         struct task_struct *task = get_proc_task(inode);
1877         const struct pid_entry *p, *last;
1878         ino_t ino;
1879         int ret;
1880
1881         ret = -ENOENT;
1882         if (!task)
1883                 goto out_no_task;
1884
1885         ret = 0;
1886         i = filp->f_pos;
1887         switch (i) {
1888         case 0:
1889                 ino = inode->i_ino;
1890                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1891                         goto out;
1892                 i++;
1893                 filp->f_pos++;
1894                 /* fall through */
1895         case 1:
1896                 ino = parent_ino(dentry);
1897                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1898                         goto out;
1899                 i++;
1900                 filp->f_pos++;
1901                 /* fall through */
1902         default:
1903                 i -= 2;
1904                 if (i >= nents) {
1905                         ret = 1;
1906                         goto out;
1907                 }
1908                 p = ents + i;
1909                 last = &ents[nents - 1];
1910                 while (p <= last) {
1911                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1912                                 goto out;
1913                         filp->f_pos++;
1914                         p++;
1915                 }
1916         }
1917
1918         ret = 1;
1919 out:
1920         put_task_struct(task);
1921 out_no_task:
1922         return ret;
1923 }
1924
1925 #ifdef CONFIG_SECURITY
1926 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1927                                   size_t count, loff_t *ppos)
1928 {
1929         struct inode * inode = file->f_path.dentry->d_inode;
1930         char *p = NULL;
1931         ssize_t length;
1932         struct task_struct *task = get_proc_task(inode);
1933
1934         if (!task)
1935                 return -ESRCH;
1936
1937         length = security_getprocattr(task,
1938                                       (char*)file->f_path.dentry->d_name.name,
1939                                       &p);
1940         put_task_struct(task);
1941         if (length > 0)
1942                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1943         kfree(p);
1944         return length;
1945 }
1946
1947 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1948                                    size_t count, loff_t *ppos)
1949 {
1950         struct inode * inode = file->f_path.dentry->d_inode;
1951         char *page;
1952         ssize_t length;
1953         struct task_struct *task = get_proc_task(inode);
1954
1955         length = -ESRCH;
1956         if (!task)
1957                 goto out_no_task;
1958         if (count > PAGE_SIZE)
1959                 count = PAGE_SIZE;
1960
1961         /* No partial writes. */
1962         length = -EINVAL;
1963         if (*ppos != 0)
1964                 goto out;
1965
1966         length = -ENOMEM;
1967         page = (char*)__get_free_page(GFP_TEMPORARY);
1968         if (!page)
1969                 goto out;
1970
1971         length = -EFAULT;
1972         if (copy_from_user(page, buf, count))
1973                 goto out_free;
1974
1975         length = security_setprocattr(task,
1976                                       (char*)file->f_path.dentry->d_name.name,
1977                                       (void*)page, count);
1978 out_free:
1979         free_page((unsigned long) page);
1980 out:
1981         put_task_struct(task);
1982 out_no_task:
1983         return length;
1984 }
1985
1986 static const struct file_operations proc_pid_attr_operations = {
1987         .read           = proc_pid_attr_read,
1988         .write          = proc_pid_attr_write,
1989 };
1990
1991 static const struct pid_entry attr_dir_stuff[] = {
1992         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1993         REG("prev",       S_IRUGO,         pid_attr),
1994         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1995         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1996         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1997         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1998 };
1999
2000 static int proc_attr_dir_readdir(struct file * filp,
2001                              void * dirent, filldir_t filldir)
2002 {
2003         return proc_pident_readdir(filp,dirent,filldir,
2004                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2005 }
2006
2007 static const struct file_operations proc_attr_dir_operations = {
2008         .read           = generic_read_dir,
2009         .readdir        = proc_attr_dir_readdir,
2010 };
2011
2012 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2013                                 struct dentry *dentry, struct nameidata *nd)
2014 {
2015         return proc_pident_lookup(dir, dentry,
2016                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2017 }
2018
2019 static const struct inode_operations proc_attr_dir_inode_operations = {
2020         .lookup         = proc_attr_dir_lookup,
2021         .getattr        = pid_getattr,
2022         .setattr        = proc_setattr,
2023 };
2024
2025 #endif
2026
2027 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2028 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2029                                          size_t count, loff_t *ppos)
2030 {
2031         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2032         struct mm_struct *mm;
2033         char buffer[PROC_NUMBUF];
2034         size_t len;
2035         int ret;
2036
2037         if (!task)
2038                 return -ESRCH;
2039
2040         ret = 0;
2041         mm = get_task_mm(task);
2042         if (mm) {
2043                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2044                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2045                                 MMF_DUMP_FILTER_SHIFT));
2046                 mmput(mm);
2047                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2048         }
2049
2050         put_task_struct(task);
2051
2052         return ret;
2053 }
2054
2055 static ssize_t proc_coredump_filter_write(struct file *file,
2056                                           const char __user *buf,
2057                                           size_t count,
2058                                           loff_t *ppos)
2059 {
2060         struct task_struct *task;
2061         struct mm_struct *mm;
2062         char buffer[PROC_NUMBUF], *end;
2063         unsigned int val;
2064         int ret;
2065         int i;
2066         unsigned long mask;
2067
2068         ret = -EFAULT;
2069         memset(buffer, 0, sizeof(buffer));
2070         if (count > sizeof(buffer) - 1)
2071                 count = sizeof(buffer) - 1;
2072         if (copy_from_user(buffer, buf, count))
2073                 goto out_no_task;
2074
2075         ret = -EINVAL;
2076         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2077         if (*end == '\n')
2078                 end++;
2079         if (end - buffer == 0)
2080                 goto out_no_task;
2081
2082         ret = -ESRCH;
2083         task = get_proc_task(file->f_dentry->d_inode);
2084         if (!task)
2085                 goto out_no_task;
2086
2087         ret = end - buffer;
2088         mm = get_task_mm(task);
2089         if (!mm)
2090                 goto out_no_mm;
2091
2092         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2093                 if (val & mask)
2094                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2095                 else
2096                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2097         }
2098
2099         mmput(mm);
2100  out_no_mm:
2101         put_task_struct(task);
2102  out_no_task:
2103         return ret;
2104 }
2105
2106 static const struct file_operations proc_coredump_filter_operations = {
2107         .read           = proc_coredump_filter_read,
2108         .write          = proc_coredump_filter_write,
2109 };
2110 #endif
2111
2112 /*
2113  * /proc/self:
2114  */
2115 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2116                               int buflen)
2117 {
2118         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2119         pid_t tgid = task_tgid_nr_ns(current, ns);
2120         char tmp[PROC_NUMBUF];
2121         if (!tgid)
2122                 return -ENOENT;
2123         sprintf(tmp, "%d", tgid);
2124         return vfs_readlink(dentry,buffer,buflen,tmp);
2125 }
2126
2127 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2128 {
2129         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2130         pid_t tgid = task_tgid_nr_ns(current, ns);
2131         char tmp[PROC_NUMBUF];
2132         if (!tgid)
2133                 return ERR_PTR(-ENOENT);
2134         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2135         return ERR_PTR(vfs_follow_link(nd,tmp));
2136 }
2137
2138 static const struct inode_operations proc_self_inode_operations = {
2139         .readlink       = proc_self_readlink,
2140         .follow_link    = proc_self_follow_link,
2141 };
2142
2143 /*
2144  * proc base
2145  *
2146  * These are the directory entries in the root directory of /proc
2147  * that properly belong to the /proc filesystem, as they describe
2148  * describe something that is process related.
2149  */
2150 static const struct pid_entry proc_base_stuff[] = {
2151         NOD("self", S_IFLNK|S_IRWXUGO,
2152                 &proc_self_inode_operations, NULL, {}),
2153 };
2154
2155 /*
2156  *      Exceptional case: normally we are not allowed to unhash a busy
2157  * directory. In this case, however, we can do it - no aliasing problems
2158  * due to the way we treat inodes.
2159  */
2160 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2161 {
2162         struct inode *inode = dentry->d_inode;
2163         struct task_struct *task = get_proc_task(inode);
2164         if (task) {
2165                 put_task_struct(task);
2166                 return 1;
2167         }
2168         d_drop(dentry);
2169         return 0;
2170 }
2171
2172 static struct dentry_operations proc_base_dentry_operations =
2173 {
2174         .d_revalidate   = proc_base_revalidate,
2175         .d_delete       = pid_delete_dentry,
2176 };
2177
2178 static struct dentry *proc_base_instantiate(struct inode *dir,
2179         struct dentry *dentry, struct task_struct *task, const void *ptr)
2180 {
2181         const struct pid_entry *p = ptr;
2182         struct inode *inode;
2183         struct proc_inode *ei;
2184         struct dentry *error = ERR_PTR(-EINVAL);
2185
2186         /* Allocate the inode */
2187         error = ERR_PTR(-ENOMEM);
2188         inode = new_inode(dir->i_sb);
2189         if (!inode)
2190                 goto out;
2191
2192         /* Initialize the inode */
2193         ei = PROC_I(inode);
2194         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2195
2196         /*
2197          * grab the reference to the task.
2198          */
2199         ei->pid = get_task_pid(task, PIDTYPE_PID);
2200         if (!ei->pid)
2201                 goto out_iput;
2202
2203         inode->i_uid = 0;
2204         inode->i_gid = 0;
2205         inode->i_mode = p->mode;
2206         if (S_ISDIR(inode->i_mode))
2207                 inode->i_nlink = 2;
2208         if (S_ISLNK(inode->i_mode))
2209                 inode->i_size = 64;
2210         if (p->iop)
2211                 inode->i_op = p->iop;
2212         if (p->fop)
2213                 inode->i_fop = p->fop;
2214         ei->op = p->op;
2215         dentry->d_op = &proc_base_dentry_operations;
2216         d_add(dentry, inode);
2217         error = NULL;
2218 out:
2219         return error;
2220 out_iput:
2221         iput(inode);
2222         goto out;
2223 }
2224
2225 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2226 {
2227         struct dentry *error;
2228         struct task_struct *task = get_proc_task(dir);
2229         const struct pid_entry *p, *last;
2230
2231         error = ERR_PTR(-ENOENT);
2232
2233         if (!task)
2234                 goto out_no_task;
2235
2236         /* Lookup the directory entry */
2237         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2238         for (p = proc_base_stuff; p <= last; p++) {
2239                 if (p->len != dentry->d_name.len)
2240                         continue;
2241                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2242                         break;
2243         }
2244         if (p > last)
2245                 goto out;
2246
2247         error = proc_base_instantiate(dir, dentry, task, p);
2248
2249 out:
2250         put_task_struct(task);
2251 out_no_task:
2252         return error;
2253 }
2254
2255 static int proc_base_fill_cache(struct file *filp, void *dirent,
2256         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2257 {
2258         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2259                                 proc_base_instantiate, task, p);
2260 }
2261
2262 #ifdef CONFIG_TASK_IO_ACCOUNTING
2263 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2264 {
2265         return sprintf(buffer,
2266 #ifdef CONFIG_TASK_XACCT
2267                         "rchar: %llu\n"
2268                         "wchar: %llu\n"
2269                         "syscr: %llu\n"
2270                         "syscw: %llu\n"
2271 #endif
2272                         "read_bytes: %llu\n"
2273                         "write_bytes: %llu\n"
2274                         "cancelled_write_bytes: %llu\n",
2275 #ifdef CONFIG_TASK_XACCT
2276                         (unsigned long long)task->rchar,
2277                         (unsigned long long)task->wchar,
2278                         (unsigned long long)task->syscr,
2279                         (unsigned long long)task->syscw,
2280 #endif
2281                         (unsigned long long)task->ioac.read_bytes,
2282                         (unsigned long long)task->ioac.write_bytes,
2283                         (unsigned long long)task->ioac.cancelled_write_bytes);
2284 }
2285 #endif
2286
2287 /*
2288  * Thread groups
2289  */
2290 static const struct file_operations proc_task_operations;
2291 static const struct inode_operations proc_task_inode_operations;
2292
2293 static const struct pid_entry tgid_base_stuff[] = {
2294         DIR("task",       S_IRUGO|S_IXUGO, task),
2295         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2296         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2297 #ifdef CONFIG_NET
2298         DIR("net",        S_IRUGO|S_IXUGO, net),
2299 #endif
2300         REG("environ",    S_IRUSR, environ),
2301         INF("auxv",       S_IRUSR, pid_auxv),
2302         ONE("status",     S_IRUGO, pid_status),
2303         INF("limits",     S_IRUSR, pid_limits),
2304 #ifdef CONFIG_SCHED_DEBUG
2305         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2306 #endif
2307         INF("cmdline",    S_IRUGO, pid_cmdline),
2308         ONE("stat",       S_IRUGO, tgid_stat),
2309         ONE("statm",      S_IRUGO, pid_statm),
2310         REG("maps",       S_IRUGO, maps),
2311 #ifdef CONFIG_NUMA
2312         REG("numa_maps",  S_IRUGO, numa_maps),
2313 #endif
2314         REG("mem",        S_IRUSR|S_IWUSR, mem),
2315         LNK("cwd",        cwd),
2316         LNK("root",       root),
2317         LNK("exe",        exe),
2318         REG("mounts",     S_IRUGO, mounts),
2319         REG("mountinfo",  S_IRUGO, mountinfo),
2320         REG("mountstats", S_IRUSR, mountstats),
2321 #ifdef CONFIG_PROC_PAGE_MONITOR
2322         REG("clear_refs", S_IWUSR, clear_refs),
2323         REG("smaps",      S_IRUGO, smaps),
2324         REG("pagemap",    S_IRUSR, pagemap),
2325 #endif
2326 #ifdef CONFIG_SECURITY
2327         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2328 #endif
2329 #ifdef CONFIG_KALLSYMS
2330         INF("wchan",      S_IRUGO, pid_wchan),
2331 #endif
2332 #ifdef CONFIG_SCHEDSTATS
2333         INF("schedstat",  S_IRUGO, pid_schedstat),
2334 #endif
2335 #ifdef CONFIG_LATENCYTOP
2336         REG("latency",  S_IRUGO, lstats),
2337 #endif
2338 #ifdef CONFIG_PROC_PID_CPUSET
2339         REG("cpuset",     S_IRUGO, cpuset),
2340 #endif
2341 #ifdef CONFIG_CGROUPS
2342         REG("cgroup",  S_IRUGO, cgroup),
2343 #endif
2344         INF("oom_score",  S_IRUGO, oom_score),
2345         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2346 #ifdef CONFIG_AUDITSYSCALL
2347         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2348         REG("sessionid",  S_IRUSR, sessionid),
2349 #endif
2350 #ifdef CONFIG_FAULT_INJECTION
2351         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2352 #endif
2353 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2354         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2355 #endif
2356 #ifdef CONFIG_TASK_IO_ACCOUNTING
2357         INF("io",       S_IRUGO, pid_io_accounting),
2358 #endif
2359 };
2360
2361 static int proc_tgid_base_readdir(struct file * filp,
2362                              void * dirent, filldir_t filldir)
2363 {
2364         return proc_pident_readdir(filp,dirent,filldir,
2365                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2366 }
2367
2368 static const struct file_operations proc_tgid_base_operations = {
2369         .read           = generic_read_dir,
2370         .readdir        = proc_tgid_base_readdir,
2371 };
2372
2373 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2374         return proc_pident_lookup(dir, dentry,
2375                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2376 }
2377
2378 static const struct inode_operations proc_tgid_base_inode_operations = {
2379         .lookup         = proc_tgid_base_lookup,
2380         .getattr        = pid_getattr,
2381         .setattr        = proc_setattr,
2382 };
2383
2384 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2385 {
2386         struct dentry *dentry, *leader, *dir;
2387         char buf[PROC_NUMBUF];
2388         struct qstr name;
2389
2390         name.name = buf;
2391         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2392         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2393         if (dentry) {
2394                 if (!(current->flags & PF_EXITING))
2395                         shrink_dcache_parent(dentry);
2396                 d_drop(dentry);
2397                 dput(dentry);
2398         }
2399
2400         if (tgid == 0)
2401                 goto out;
2402
2403         name.name = buf;
2404         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2405         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2406         if (!leader)
2407                 goto out;
2408
2409         name.name = "task";
2410         name.len = strlen(name.name);
2411         dir = d_hash_and_lookup(leader, &name);
2412         if (!dir)
2413                 goto out_put_leader;
2414
2415         name.name = buf;
2416         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2417         dentry = d_hash_and_lookup(dir, &name);
2418         if (dentry) {
2419                 shrink_dcache_parent(dentry);
2420                 d_drop(dentry);
2421                 dput(dentry);
2422         }
2423
2424         dput(dir);
2425 out_put_leader:
2426         dput(leader);
2427 out:
2428         return;
2429 }
2430
2431 /**
2432  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2433  * @task: task that should be flushed.
2434  *
2435  * When flushing dentries from proc, one needs to flush them from global
2436  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2437  * in. This call is supposed to do all of this job.
2438  *
2439  * Looks in the dcache for
2440  * /proc/@pid
2441  * /proc/@tgid/task/@pid
2442  * if either directory is present flushes it and all of it'ts children
2443  * from the dcache.
2444  *
2445  * It is safe and reasonable to cache /proc entries for a task until
2446  * that task exits.  After that they just clog up the dcache with
2447  * useless entries, possibly causing useful dcache entries to be
2448  * flushed instead.  This routine is proved to flush those useless
2449  * dcache entries at process exit time.
2450  *
2451  * NOTE: This routine is just an optimization so it does not guarantee
2452  *       that no dcache entries will exist at process exit time it
2453  *       just makes it very unlikely that any will persist.
2454  */
2455
2456 void proc_flush_task(struct task_struct *task)
2457 {
2458         int i;
2459         struct pid *pid, *tgid = NULL;
2460         struct upid *upid;
2461
2462         pid = task_pid(task);
2463         if (thread_group_leader(task))
2464                 tgid = task_tgid(task);
2465
2466         for (i = 0; i <= pid->level; i++) {
2467                 upid = &pid->numbers[i];
2468                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2469                         tgid ? tgid->numbers[i].nr : 0);
2470         }
2471
2472         upid = &pid->numbers[pid->level];
2473         if (upid->nr == 1)
2474                 pid_ns_release_proc(upid->ns);
2475 }
2476
2477 static struct dentry *proc_pid_instantiate(struct inode *dir,
2478                                            struct dentry * dentry,
2479                                            struct task_struct *task, const void *ptr)
2480 {
2481         struct dentry *error = ERR_PTR(-ENOENT);
2482         struct inode *inode;
2483
2484         inode = proc_pid_make_inode(dir->i_sb, task);
2485         if (!inode)
2486                 goto out;
2487
2488         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2489         inode->i_op = &proc_tgid_base_inode_operations;
2490         inode->i_fop = &proc_tgid_base_operations;
2491         inode->i_flags|=S_IMMUTABLE;
2492         inode->i_nlink = 5;
2493 #ifdef CONFIG_SECURITY
2494         inode->i_nlink += 1;
2495 #endif
2496
2497         dentry->d_op = &pid_dentry_operations;
2498
2499         d_add(dentry, inode);
2500         /* Close the race of the process dying before we return the dentry */
2501         if (pid_revalidate(dentry, NULL))
2502                 error = NULL;
2503 out:
2504         return error;
2505 }
2506
2507 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2508 {
2509         struct dentry *result = ERR_PTR(-ENOENT);
2510         struct task_struct *task;
2511         unsigned tgid;
2512         struct pid_namespace *ns;
2513
2514         result = proc_base_lookup(dir, dentry);
2515         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2516                 goto out;
2517
2518         tgid = name_to_int(dentry);
2519         if (tgid == ~0U)
2520                 goto out;
2521
2522         ns = dentry->d_sb->s_fs_info;
2523         rcu_read_lock();
2524         task = find_task_by_pid_ns(tgid, ns);
2525         if (task)
2526                 get_task_struct(task);
2527         rcu_read_unlock();
2528         if (!task)
2529                 goto out;
2530
2531         result = proc_pid_instantiate(dir, dentry, task, NULL);
2532         put_task_struct(task);
2533 out:
2534         return result;
2535 }
2536
2537 /*
2538  * Find the first task with tgid >= tgid
2539  *
2540  */
2541 struct tgid_iter {
2542         unsigned int tgid;
2543         struct task_struct *task;
2544 };
2545 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2546 {
2547         struct pid *pid;
2548
2549         if (iter.task)
2550                 put_task_struct(iter.task);
2551         rcu_read_lock();
2552 retry:
2553         iter.task = NULL;
2554         pid = find_ge_pid(iter.tgid, ns);
2555         if (pid) {
2556                 iter.tgid = pid_nr_ns(pid, ns);
2557                 iter.task = pid_task(pid, PIDTYPE_PID);
2558                 /* What we to know is if the pid we have find is the
2559                  * pid of a thread_group_leader.  Testing for task
2560                  * being a thread_group_leader is the obvious thing
2561                  * todo but there is a window when it fails, due to
2562                  * the pid transfer logic in de_thread.
2563                  *
2564                  * So we perform the straight forward test of seeing
2565                  * if the pid we have found is the pid of a thread
2566                  * group leader, and don't worry if the task we have
2567                  * found doesn't happen to be a thread group leader.
2568                  * As we don't care in the case of readdir.
2569                  */
2570                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2571                         iter.tgid += 1;
2572                         goto retry;
2573                 }
2574                 get_task_struct(iter.task);
2575         }
2576         rcu_read_unlock();
2577         return iter;
2578 }
2579
2580 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2581
2582 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2583         struct tgid_iter iter)
2584 {
2585         char name[PROC_NUMBUF];
2586         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2587         return proc_fill_cache(filp, dirent, filldir, name, len,
2588                                 proc_pid_instantiate, iter.task, NULL);
2589 }
2590
2591 /* for the /proc/ directory itself, after non-process stuff has been done */
2592 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2593 {
2594         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2595         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2596         struct tgid_iter iter;
2597         struct pid_namespace *ns;
2598
2599         if (!reaper)
2600                 goto out_no_task;
2601
2602         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2603                 const struct pid_entry *p = &proc_base_stuff[nr];
2604                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2605                         goto out;
2606         }
2607
2608         ns = filp->f_dentry->d_sb->s_fs_info;
2609         iter.task = NULL;
2610         iter.tgid = filp->f_pos - TGID_OFFSET;
2611         for (iter = next_tgid(ns, iter);
2612              iter.task;
2613              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2614                 filp->f_pos = iter.tgid + TGID_OFFSET;
2615                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2616                         put_task_struct(iter.task);
2617                         goto out;
2618                 }
2619         }
2620         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2621 out:
2622         put_task_struct(reaper);
2623 out_no_task:
2624         return 0;
2625 }
2626
2627 /*
2628  * Tasks
2629  */
2630 static const struct pid_entry tid_base_stuff[] = {
2631         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2632         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2633         REG("environ",   S_IRUSR, environ),
2634         INF("auxv",      S_IRUSR, pid_auxv),
2635         ONE("status",    S_IRUGO, pid_status),
2636         INF("limits",    S_IRUSR, pid_limits),
2637 #ifdef CONFIG_SCHED_DEBUG
2638         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2639 #endif
2640         INF("cmdline",   S_IRUGO, pid_cmdline),
2641         ONE("stat",      S_IRUGO, tid_stat),
2642         ONE("statm",     S_IRUGO, pid_statm),
2643         REG("maps",      S_IRUGO, maps),
2644 #ifdef CONFIG_NUMA
2645         REG("numa_maps", S_IRUGO, numa_maps),
2646 #endif
2647         REG("mem",       S_IRUSR|S_IWUSR, mem),
2648         LNK("cwd",       cwd),
2649         LNK("root",      root),
2650         LNK("exe",       exe),
2651         REG("mounts",    S_IRUGO, mounts),
2652         REG("mountinfo",  S_IRUGO, mountinfo),
2653 #ifdef CONFIG_PROC_PAGE_MONITOR
2654         REG("clear_refs", S_IWUSR, clear_refs),
2655         REG("smaps",     S_IRUGO, smaps),
2656         REG("pagemap",    S_IRUSR, pagemap),
2657 #endif
2658 #ifdef CONFIG_SECURITY
2659         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2660 #endif
2661 #ifdef CONFIG_KALLSYMS
2662         INF("wchan",     S_IRUGO, pid_wchan),
2663 #endif
2664 #ifdef CONFIG_SCHEDSTATS
2665         INF("schedstat", S_IRUGO, pid_schedstat),
2666 #endif
2667 #ifdef CONFIG_LATENCYTOP
2668         REG("latency",  S_IRUGO, lstats),
2669 #endif
2670 #ifdef CONFIG_PROC_PID_CPUSET
2671         REG("cpuset",    S_IRUGO, cpuset),
2672 #endif
2673 #ifdef CONFIG_CGROUPS
2674         REG("cgroup",  S_IRUGO, cgroup),
2675 #endif
2676         INF("oom_score", S_IRUGO, oom_score),
2677         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2678 #ifdef CONFIG_AUDITSYSCALL
2679         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2680         REG("sessionid",  S_IRUSR, sessionid),
2681 #endif
2682 #ifdef CONFIG_FAULT_INJECTION
2683         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2684 #endif
2685 };
2686
2687 static int proc_tid_base_readdir(struct file * filp,
2688                              void * dirent, filldir_t filldir)
2689 {
2690         return proc_pident_readdir(filp,dirent,filldir,
2691                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2692 }
2693
2694 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2695         return proc_pident_lookup(dir, dentry,
2696                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2697 }
2698
2699 static const struct file_operations proc_tid_base_operations = {
2700         .read           = generic_read_dir,
2701         .readdir        = proc_tid_base_readdir,
2702 };
2703
2704 static const struct inode_operations proc_tid_base_inode_operations = {
2705         .lookup         = proc_tid_base_lookup,
2706         .getattr        = pid_getattr,
2707         .setattr        = proc_setattr,
2708 };
2709
2710 static struct dentry *proc_task_instantiate(struct inode *dir,
2711         struct dentry *dentry, struct task_struct *task, const void *ptr)
2712 {
2713         struct dentry *error = ERR_PTR(-ENOENT);
2714         struct inode *inode;
2715         inode = proc_pid_make_inode(dir->i_sb, task);
2716
2717         if (!inode)
2718                 goto out;
2719         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2720         inode->i_op = &proc_tid_base_inode_operations;
2721         inode->i_fop = &proc_tid_base_operations;
2722         inode->i_flags|=S_IMMUTABLE;
2723         inode->i_nlink = 4;
2724 #ifdef CONFIG_SECURITY
2725         inode->i_nlink += 1;
2726 #endif
2727
2728         dentry->d_op = &pid_dentry_operations;
2729
2730         d_add(dentry, inode);
2731         /* Close the race of the process dying before we return the dentry */
2732         if (pid_revalidate(dentry, NULL))
2733                 error = NULL;
2734 out:
2735         return error;
2736 }
2737
2738 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2739 {
2740         struct dentry *result = ERR_PTR(-ENOENT);
2741         struct task_struct *task;
2742         struct task_struct *leader = get_proc_task(dir);
2743         unsigned tid;
2744         struct pid_namespace *ns;
2745
2746         if (!leader)
2747                 goto out_no_task;
2748
2749         tid = name_to_int(dentry);
2750         if (tid == ~0U)
2751                 goto out;
2752
2753         ns = dentry->d_sb->s_fs_info;
2754         rcu_read_lock();
2755         task = find_task_by_pid_ns(tid, ns);
2756         if (task)
2757                 get_task_struct(task);
2758         rcu_read_unlock();
2759         if (!task)
2760                 goto out;
2761         if (!same_thread_group(leader, task))
2762                 goto out_drop_task;
2763
2764         result = proc_task_instantiate(dir, dentry, task, NULL);
2765 out_drop_task:
2766         put_task_struct(task);
2767 out:
2768         put_task_struct(leader);
2769 out_no_task:
2770         return result;
2771 }
2772
2773 /*
2774  * Find the first tid of a thread group to return to user space.
2775  *
2776  * Usually this is just the thread group leader, but if the users
2777  * buffer was too small or there was a seek into the middle of the
2778  * directory we have more work todo.
2779  *
2780  * In the case of a short read we start with find_task_by_pid.
2781  *
2782  * In the case of a seek we start with the leader and walk nr
2783  * threads past it.
2784  */
2785 static struct task_struct *first_tid(struct task_struct *leader,
2786                 int tid, int nr, struct pid_namespace *ns)
2787 {
2788         struct task_struct *pos;
2789
2790         rcu_read_lock();
2791         /* Attempt to start with the pid of a thread */
2792         if (tid && (nr > 0)) {
2793                 pos = find_task_by_pid_ns(tid, ns);
2794                 if (pos && (pos->group_leader == leader))
2795                         goto found;
2796         }
2797
2798         /* If nr exceeds the number of threads there is nothing todo */
2799         pos = NULL;
2800         if (nr && nr >= get_nr_threads(leader))
2801                 goto out;
2802
2803         /* If we haven't found our starting place yet start
2804          * with the leader and walk nr threads forward.
2805          */
2806         for (pos = leader; nr > 0; --nr) {
2807                 pos = next_thread(pos);
2808                 if (pos == leader) {
2809                         pos = NULL;
2810                         goto out;
2811                 }
2812         }
2813 found:
2814         get_task_struct(pos);
2815 out:
2816         rcu_read_unlock();
2817         return pos;
2818 }
2819
2820 /*
2821  * Find the next thread in the thread list.
2822  * Return NULL if there is an error or no next thread.
2823  *
2824  * The reference to the input task_struct is released.
2825  */
2826 static struct task_struct *next_tid(struct task_struct *start)
2827 {
2828         struct task_struct *pos = NULL;
2829         rcu_read_lock();
2830         if (pid_alive(start)) {
2831                 pos = next_thread(start);
2832                 if (thread_group_leader(pos))
2833                         pos = NULL;
2834                 else
2835                         get_task_struct(pos);
2836         }
2837         rcu_read_unlock();
2838         put_task_struct(start);
2839         return pos;
2840 }
2841
2842 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2843         struct task_struct *task, int tid)
2844 {
2845         char name[PROC_NUMBUF];
2846         int len = snprintf(name, sizeof(name), "%d", tid);
2847         return proc_fill_cache(filp, dirent, filldir, name, len,
2848                                 proc_task_instantiate, task, NULL);
2849 }
2850
2851 /* for the /proc/TGID/task/ directories */
2852 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2853 {
2854         struct dentry *dentry = filp->f_path.dentry;
2855         struct inode *inode = dentry->d_inode;
2856         struct task_struct *leader = NULL;
2857         struct task_struct *task;
2858         int retval = -ENOENT;
2859         ino_t ino;
2860         int tid;
2861         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2862         struct pid_namespace *ns;
2863
2864         task = get_proc_task(inode);
2865         if (!task)
2866                 goto out_no_task;
2867         rcu_read_lock();
2868         if (pid_alive(task)) {
2869                 leader = task->group_leader;
2870                 get_task_struct(leader);
2871         }
2872         rcu_read_unlock();
2873         put_task_struct(task);
2874         if (!leader)
2875                 goto out_no_task;
2876         retval = 0;
2877
2878         switch (pos) {
2879         case 0:
2880                 ino = inode->i_ino;
2881                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2882                         goto out;
2883                 pos++;
2884                 /* fall through */
2885         case 1:
2886                 ino = parent_ino(dentry);
2887                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2888                         goto out;
2889                 pos++;
2890                 /* fall through */
2891         }
2892
2893         /* f_version caches the tgid value that the last readdir call couldn't
2894          * return. lseek aka telldir automagically resets f_version to 0.
2895          */
2896         ns = filp->f_dentry->d_sb->s_fs_info;
2897         tid = (int)filp->f_version;
2898         filp->f_version = 0;
2899         for (task = first_tid(leader, tid, pos - 2, ns);
2900              task;
2901              task = next_tid(task), pos++) {
2902                 tid = task_pid_nr_ns(task, ns);
2903                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2904                         /* returning this tgid failed, save it as the first
2905                          * pid for the next readir call */
2906                         filp->f_version = (u64)tid;
2907                         put_task_struct(task);
2908                         break;
2909                 }
2910         }
2911 out:
2912         filp->f_pos = pos;
2913         put_task_struct(leader);
2914 out_no_task:
2915         return retval;
2916 }
2917
2918 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2919 {
2920         struct inode *inode = dentry->d_inode;
2921         struct task_struct *p = get_proc_task(inode);
2922         generic_fillattr(inode, stat);
2923
2924         if (p) {
2925                 rcu_read_lock();
2926                 stat->nlink += get_nr_threads(p);
2927                 rcu_read_unlock();
2928                 put_task_struct(p);
2929         }
2930
2931         return 0;
2932 }
2933
2934 static const struct inode_operations proc_task_inode_operations = {
2935         .lookup         = proc_task_lookup,
2936         .getattr        = proc_task_getattr,
2937         .setattr        = proc_setattr,
2938 };
2939
2940 static const struct file_operations proc_task_operations = {
2941         .read           = generic_read_dir,
2942         .readdir        = proc_task_readdir,
2943 };