SAFE public projects git trees. - safe/jmp/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/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/resource.h>
  69 #include <linux/module.h>
  70 #include <linux/mount.h>
  71 #include <linux/security.h>
  72 #include <linux/ptrace.h>
  73 #include <linux/tracehook.h>
  74 #include <linux/cgroup.h>
  75 #include <linux/cpuset.h>
  76 #include <linux/audit.h>
  77 #include <linux/poll.h>
  78 #include <linux/nsproxy.h>
  79 #include <linux/oom.h>
  80 #include <linux/elf.h>
  81 #include <linux/pid_namespace.h>
  82 #include "internal.h"
  83
  84 /* NOTE:
  85  *      Implementing inode permission operations in /proc is almost
  86  *      certainly an error.  Permission checks need to happen during
  87  *      each system call not at open time.  The reason is that most of
  88  *      what we wish to check for permissions in /proc varies at runtime.
  89  *
  90  *      The classic example of a problem is opening file descriptors
  91  *      in /proc for a task before it execs a suid executable.
  92  */
  93
  94 struct pid_entry {
  95         char *name;
  96         int len;
  97         mode_t mode;
  98         const struct inode_operations *iop;
  99         const struct file_operations *fop;
 100         union proc_op op;
 101 };
 102
 103 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 104         .name = (NAME),                                 \
 105         .len  = sizeof(NAME) - 1,                       \
 106         .mode = MODE,                                   \
 107         .iop  = IOP,                                    \
 108         .fop  = FOP,                                    \
 109         .op   = OP,                                     \
 110 }
 111
 112 #define DIR(NAME, MODE, iops, fops)     \
 113         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 114 #define LNK(NAME, get_link)                                     \
 115         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 116                 &proc_pid_link_inode_operations, NULL,          \
 117                 { .proc_get_link = get_link } )
 118 #define REG(NAME, MODE, fops)                           \
 119         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 120 #define INF(NAME, MODE, read)                           \
 121         NOD(NAME, (S_IFREG|(MODE)),                     \
 122                 NULL, &proc_info_file_operations,       \
 123                 { .proc_read = read } )
 124 #define ONE(NAME, MODE, show)                           \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_single_file_operations,     \
 127                 { .proc_show = show } )
 128
 129 /*
 130  * Count the number of hardlinks for the pid_entry table, excluding the .
 131  * and .. links.
 132  */
 133 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 134         unsigned int n)
 135 {
 136         unsigned int i;
 137         unsigned int count;
 138
 139         count = 0;
 140         for (i = 0; i < n; ++i) {
 141                 if (S_ISDIR(entries[i].mode))
 142                         ++count;
 143         }
 144
 145         return count;
 146 }
 147
 148 static struct fs_struct *get_fs_struct(struct task_struct *task)
 149 {
 150         struct fs_struct *fs;
 151         task_lock(task);
 152         fs = task->fs;
 153         if(fs)
 154                 atomic_inc(&fs->count);
 155         task_unlock(task);
 156         return fs;
 157 }
 158
 159 static int get_nr_threads(struct task_struct *tsk)
 160 {
 161         unsigned long flags;
 162         int count = 0;
 163
 164         if (lock_task_sighand(tsk, &flags)) {
 165                 count = atomic_read(&tsk->signal->count);
 166                 unlock_task_sighand(tsk, &flags);
 167         }
 168         return count;
 169 }
 170
 171 static int proc_cwd_link(struct inode *inode, struct path *path)
 172 {
 173         struct task_struct *task = get_proc_task(inode);
 174         struct fs_struct *fs = NULL;
 175         int result = -ENOENT;
 176
 177         if (task) {
 178                 fs = get_fs_struct(task);
 179                 put_task_struct(task);
 180         }
 181         if (fs) {
 182                 read_lock(&fs->lock);
 183                 *path = fs->pwd;
 184                 path_get(&fs->pwd);
 185                 read_unlock(&fs->lock);
 186                 result = 0;
 187                 put_fs_struct(fs);
 188         }
 189         return result;
 190 }
 191
 192 static int proc_root_link(struct inode *inode, struct path *path)
 193 {
 194         struct task_struct *task = get_proc_task(inode);
 195         struct fs_struct *fs = NULL;
 196         int result = -ENOENT;
 197
 198         if (task) {
 199                 fs = get_fs_struct(task);
 200                 put_task_struct(task);
 201         }
 202         if (fs) {
 203                 read_lock(&fs->lock);
 204                 *path = fs->root;
 205                 path_get(&fs->root);
 206                 read_unlock(&fs->lock);
 207                 result = 0;
 208                 put_fs_struct(fs);
 209         }
 210         return result;
 211 }
 212
 213 /*
 214  * Return zero if current may access user memory in @task, -error if not.
 215  */
 216 static int check_mem_permission(struct task_struct *task)
 217 {
 218         /*
 219          * A task can always look at itself, in case it chooses
 220          * to use system calls instead of load instructions.
 221          */
 222         if (task == current)
 223                 return 0;
 224
 225         /*
 226          * If current is actively ptrace'ing, and would also be
 227          * permitted to freshly attach with ptrace now, permit it.
 228          */
 229         if (task_is_stopped_or_traced(task)) {
 230                 int match;
 231                 rcu_read_lock();
 232                 match = (tracehook_tracer_task(task) == current);
 233                 rcu_read_unlock();
 234                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 235                         return 0;
 236         }
 237
 238         /*
 239          * Noone else is allowed.
 240          */
 241         return -EPERM;
 242 }
 243
 244 struct mm_struct *mm_for_maps(struct task_struct *task)
 245 {
 246         struct mm_struct *mm = get_task_mm(task);
 247         if (!mm)
 248                 return NULL;
 249         down_read(&mm->mmap_sem);
 250         task_lock(task);
 251         if (task->mm != mm)
 252                 goto out;
 253         if (task->mm != current->mm &&
 254             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
 255                 goto out;
 256         task_unlock(task);
 257         return mm;
 258 out:
 259         task_unlock(task);
 260         up_read(&mm->mmap_sem);
 261         mmput(mm);
 262         return NULL;
 263 }
 264
 265 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 266 {
 267         int res = 0;
 268         unsigned int len;
 269         struct mm_struct *mm = get_task_mm(task);
 270         if (!mm)
 271                 goto out;
 272         if (!mm->arg_end)
 273                 goto out_mm;    /* Shh! No looking before we're done */
 274
 275         len = mm->arg_end - mm->arg_start;
 276
 277         if (len > PAGE_SIZE)
 278                 len = PAGE_SIZE;
 279
 280         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 281
 282         // If the nul at the end of args has been overwritten, then
 283         // assume application is using setproctitle(3).
 284         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 285                 len = strnlen(buffer, res);
 286                 if (len < res) {
 287                     res = len;
 288                 } else {
 289                         len = mm->env_end - mm->env_start;
 290                         if (len > PAGE_SIZE - res)
 291                                 len = PAGE_SIZE - res;
 292                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 293                         res = strnlen(buffer, res);
 294                 }
 295         }
 296 out_mm:
 297         mmput(mm);
 298 out:
 299         return res;
 300 }
 301
 302 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 303 {
 304         int res = 0;
 305         struct mm_struct *mm = get_task_mm(task);
 306         if (mm) {
 307                 unsigned int nwords = 0;
 308                 do
 309                         nwords += 2;
 310                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 311                 res = nwords * sizeof(mm->saved_auxv[0]);
 312                 if (res > PAGE_SIZE)
 313                         res = PAGE_SIZE;
 314                 memcpy(buffer, mm->saved_auxv, res);
 315                 mmput(mm);
 316         }
 317         return res;
 318 }
 319
 320
 321 #ifdef CONFIG_KALLSYMS
 322 /*
 323  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 324  * Returns the resolved symbol.  If that fails, simply return the address.
 325  */
 326 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 327 {
 328         unsigned long wchan;
 329         char symname[KSYM_NAME_LEN];
 330
 331         wchan = get_wchan(task);
 332
 333         if (lookup_symbol_name(wchan, symname) < 0)
 334                 return sprintf(buffer, "%lu", wchan);
 335         else
 336                 return sprintf(buffer, "%s", symname);
 337 }
 338 #endif /* CONFIG_KALLSYMS */
 339
 340 #ifdef CONFIG_SCHEDSTATS
 341 /*
 342  * Provides /proc/PID/schedstat
 343  */
 344 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 345 {
 346         return sprintf(buffer, "%llu %llu %lu\n",
 347                         (unsigned long long)task->se.sum_exec_runtime,
 348                         (unsigned long long)task->sched_info.run_delay,
 349                         task->sched_info.pcount);
 350 }
 351 #endif
 352
 353 #ifdef CONFIG_LATENCYTOP
 354 static int lstats_show_proc(struct seq_file *m, void *v)
 355 {
 356         int i;
 357         struct inode *inode = m->private;
 358         struct task_struct *task = get_proc_task(inode);
 359
 360         if (!task)
 361                 return -ESRCH;
 362         seq_puts(m, "Latency Top version : v0.1\n");
 363         for (i = 0; i < 32; i++) {
 364                 if (task->latency_record[i].backtrace[0]) {
 365                         int q;
 366                         seq_printf(m, "%i %li %li ",
 367                                 task->latency_record[i].count,
 368                                 task->latency_record[i].time,
 369                                 task->latency_record[i].max);
 370                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 371                                 char sym[KSYM_SYMBOL_LEN];
 372                                 char *c;
 373                                 if (!task->latency_record[i].backtrace[q])
 374                                         break;
 375                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 376                                         break;
 377                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 378                                 c = strchr(sym, '+');
 379                                 if (c)
 380                                         *c = 0;
 381                                 seq_printf(m, "%s ", sym);
 382                         }
 383                         seq_printf(m, "\n");
 384                 }
 385
 386         }
 387         put_task_struct(task);
 388         return 0;
 389 }
 390
 391 static int lstats_open(struct inode *inode, struct file *file)
 392 {
 393         return single_open(file, lstats_show_proc, inode);
 394 }
 395
 396 static ssize_t lstats_write(struct file *file, const char __user *buf,
 397                             size_t count, loff_t *offs)
 398 {
 399         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 400
 401         if (!task)
 402                 return -ESRCH;
 403         clear_all_latency_tracing(task);
 404         put_task_struct(task);
 405
 406         return count;
 407 }
 408
 409 static const struct file_operations proc_lstats_operations = {
 410         .open           = lstats_open,
 411         .read           = seq_read,
 412         .write          = lstats_write,
 413         .llseek         = seq_lseek,
 414         .release        = single_release,
 415 };
 416
 417 #endif
 418
 419 /* The badness from the OOM killer */
 420 unsigned long badness(struct task_struct *p, unsigned long uptime);
 421 static int proc_oom_score(struct task_struct *task, char *buffer)
 422 {
 423         unsigned long points;
 424         struct timespec uptime;
 425
 426         do_posix_clock_monotonic_gettime(&uptime);
 427         read_lock(&tasklist_lock);
 428         points = badness(task, uptime.tv_sec);
 429         read_unlock(&tasklist_lock);
 430         return sprintf(buffer, "%lu\n", points);
 431 }
 432
 433 struct limit_names {
 434         char *name;
 435         char *unit;
 436 };
 437
 438 static const struct limit_names lnames[RLIM_NLIMITS] = {
 439         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 440         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 441         [RLIMIT_DATA] = {"Max data size", "bytes"},
 442         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 443         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 444         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 445         [RLIMIT_NPROC] = {"Max processes", "processes"},
 446         [RLIMIT_NOFILE] = {"Max open files", "files"},
 447         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 448         [RLIMIT_AS] = {"Max address space", "bytes"},
 449         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 450         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 451         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 452         [RLIMIT_NICE] = {"Max nice priority", NULL},
 453         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 454         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 455 };
 456
 457 /* Display limits for a process */
 458 static int proc_pid_limits(struct task_struct *task, char *buffer)
 459 {
 460         unsigned int i;
 461         int count = 0;
 462         unsigned long flags;
 463         char *bufptr = buffer;
 464
 465         struct rlimit rlim[RLIM_NLIMITS];
 466
 467         if (!lock_task_sighand(task, &flags))
 468                 return 0;
 469         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 470         unlock_task_sighand(task, &flags);
 471
 472         /*
 473          * print the file header
 474          */
 475         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 476                         "Limit", "Soft Limit", "Hard Limit", "Units");
 477
 478         for (i = 0; i < RLIM_NLIMITS; i++) {
 479                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 480                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 481                                          lnames[i].name, "unlimited");
 482                 else
 483                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 484                                          lnames[i].name, rlim[i].rlim_cur);
 485
 486                 if (rlim[i].rlim_max == RLIM_INFINITY)
 487                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 488                 else
 489                         count += sprintf(&bufptr[count], "%-20lu ",
 490                                          rlim[i].rlim_max);
 491
 492                 if (lnames[i].unit)
 493                         count += sprintf(&bufptr[count], "%-10s\n",
 494                                          lnames[i].unit);
 495                 else
 496                         count += sprintf(&bufptr[count], "\n");
 497         }
 498
 499         return count;
 500 }
 501
 502 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 503 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 504 {
 505         long nr;
 506         unsigned long args[6], sp, pc;
 507
 508         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 509                 return sprintf(buffer, "running\n");
 510
 511         if (nr < 0)
 512                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 513
 514         return sprintf(buffer,
 515                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 516                        nr,
 517                        args[0], args[1], args[2], args[3], args[4], args[5],
 518                        sp, pc);
 519 }
 520 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 521
 522 /************************************************************************/
 523 /*                       Here the fs part begins                        */
 524 /************************************************************************/
 525
 526 /* permission checks */
 527 static int proc_fd_access_allowed(struct inode *inode)
 528 {
 529         struct task_struct *task;
 530         int allowed = 0;
 531         /* Allow access to a task's file descriptors if it is us or we
 532          * may use ptrace attach to the process and find out that
 533          * information.
 534          */
 535         task = get_proc_task(inode);
 536         if (task) {
 537                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 538                 put_task_struct(task);
 539         }
 540         return allowed;
 541 }
 542
 543 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 544 {
 545         int error;
 546         struct inode *inode = dentry->d_inode;
 547
 548         if (attr->ia_valid & ATTR_MODE)
 549                 return -EPERM;
 550
 551         error = inode_change_ok(inode, attr);
 552         if (!error)
 553                 error = inode_setattr(inode, attr);
 554         return error;
 555 }
 556
 557 static const struct inode_operations proc_def_inode_operations = {
 558         .setattr        = proc_setattr,
 559 };
 560
 561 static int mounts_open_common(struct inode *inode, struct file *file,
 562                               const struct seq_operations *op)
 563 {
 564         struct task_struct *task = get_proc_task(inode);
 565         struct nsproxy *nsp;
 566         struct mnt_namespace *ns = NULL;
 567         struct fs_struct *fs = NULL;
 568         struct path root;
 569         struct proc_mounts *p;
 570         int ret = -EINVAL;
 571
 572         if (task) {
 573                 rcu_read_lock();
 574                 nsp = task_nsproxy(task);
 575                 if (nsp) {
 576                         ns = nsp->mnt_ns;
 577                         if (ns)
 578                                 get_mnt_ns(ns);
 579                 }
 580                 rcu_read_unlock();
 581                 if (ns)
 582                         fs = get_fs_struct(task);
 583                 put_task_struct(task);
 584         }
 585
 586         if (!ns)
 587                 goto err;
 588         if (!fs)
 589                 goto err_put_ns;
 590
 591         read_lock(&fs->lock);
 592         root = fs->root;
 593         path_get(&root);
 594         read_unlock(&fs->lock);
 595         put_fs_struct(fs);
 596
 597         ret = -ENOMEM;
 598         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 599         if (!p)
 600                 goto err_put_path;
 601
 602         file->private_data = &p->m;
 603         ret = seq_open(file, op);
 604         if (ret)
 605                 goto err_free;
 606
 607         p->m.private = p;
 608         p->ns = ns;
 609         p->root = root;
 610         p->event = ns->event;
 611
 612         return 0;
 613
 614  err_free:
 615         kfree(p);
 616  err_put_path:
 617         path_put(&root);
 618  err_put_ns:
 619         put_mnt_ns(ns);
 620  err:
 621         return ret;
 622 }
 623
 624 static int mounts_release(struct inode *inode, struct file *file)
 625 {
 626         struct proc_mounts *p = file->private_data;
 627         path_put(&p->root);
 628         put_mnt_ns(p->ns);
 629         return seq_release(inode, file);
 630 }
 631
 632 static unsigned mounts_poll(struct file *file, poll_table *wait)
 633 {
 634         struct proc_mounts *p = file->private_data;
 635         struct mnt_namespace *ns = p->ns;
 636         unsigned res = 0;
 637
 638         poll_wait(file, &ns->poll, wait);
 639
 640         spin_lock(&vfsmount_lock);
 641         if (p->event != ns->event) {
 642                 p->event = ns->event;
 643                 res = POLLERR;
 644         }
 645         spin_unlock(&vfsmount_lock);
 646
 647         return res;
 648 }
 649
 650 static int mounts_open(struct inode *inode, struct file *file)
 651 {
 652         return mounts_open_common(inode, file, &mounts_op);
 653 }
 654
 655 static const struct file_operations proc_mounts_operations = {
 656         .open           = mounts_open,
 657         .read           = seq_read,
 658         .llseek         = seq_lseek,
 659         .release        = mounts_release,
 660         .poll           = mounts_poll,
 661 };
 662
 663 static int mountinfo_open(struct inode *inode, struct file *file)
 664 {
 665         return mounts_open_common(inode, file, &mountinfo_op);
 666 }
 667
 668 static const struct file_operations proc_mountinfo_operations = {
 669         .open           = mountinfo_open,
 670         .read           = seq_read,
 671         .llseek         = seq_lseek,
 672         .release        = mounts_release,
 673         .poll           = mounts_poll,
 674 };
 675
 676 static int mountstats_open(struct inode *inode, struct file *file)
 677 {
 678         return mounts_open_common(inode, file, &mountstats_op);
 679 }
 680
 681 static const struct file_operations proc_mountstats_operations = {
 682         .open           = mountstats_open,
 683         .read           = seq_read,
 684         .llseek         = seq_lseek,
 685         .release        = mounts_release,
 686 };
 687
 688 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 689
 690 static ssize_t proc_info_read(struct file * file, char __user * buf,
 691                           size_t count, loff_t *ppos)
 692 {
 693         struct inode * inode = file->f_path.dentry->d_inode;
 694         unsigned long page;
 695         ssize_t length;
 696         struct task_struct *task = get_proc_task(inode);
 697
 698         length = -ESRCH;
 699         if (!task)
 700                 goto out_no_task;
 701
 702         if (count > PROC_BLOCK_SIZE)
 703                 count = PROC_BLOCK_SIZE;
 704
 705         length = -ENOMEM;
 706         if (!(page = __get_free_page(GFP_TEMPORARY)))
 707                 goto out;
 708
 709         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 710
 711         if (length >= 0)
 712                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 713         free_page(page);
 714 out:
 715         put_task_struct(task);
 716 out_no_task:
 717         return length;
 718 }
 719
 720 static const struct file_operations proc_info_file_operations = {
 721         .read           = proc_info_read,
 722 };
 723
 724 static int proc_single_show(struct seq_file *m, void *v)
 725 {
 726         struct inode *inode = m->private;
 727         struct pid_namespace *ns;
 728         struct pid *pid;
 729         struct task_struct *task;
 730         int ret;
 731
 732         ns = inode->i_sb->s_fs_info;
 733         pid = proc_pid(inode);
 734         task = get_pid_task(pid, PIDTYPE_PID);
 735         if (!task)
 736                 return -ESRCH;
 737
 738         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 739
 740         put_task_struct(task);
 741         return ret;
 742 }
 743
 744 static int proc_single_open(struct inode *inode, struct file *filp)
 745 {
 746         int ret;
 747         ret = single_open(filp, proc_single_show, NULL);
 748         if (!ret) {
 749                 struct seq_file *m = filp->private_data;
 750
 751                 m->private = inode;
 752         }
 753         return ret;
 754 }
 755
 756 static const struct file_operations proc_single_file_operations = {
 757         .open           = proc_single_open,
 758         .read           = seq_read,
 759         .llseek         = seq_lseek,
 760         .release        = single_release,
 761 };
 762
 763 static int mem_open(struct inode* inode, struct file* file)
 764 {
 765         file->private_data = (void*)((long)current->self_exec_id);
 766         return 0;
 767 }
 768
 769 static ssize_t mem_read(struct file * file, char __user * buf,
 770                         size_t count, loff_t *ppos)
 771 {
 772         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 773         char *page;
 774         unsigned long src = *ppos;
 775         int ret = -ESRCH;
 776         struct mm_struct *mm;
 777
 778         if (!task)
 779                 goto out_no_task;
 780
 781         if (check_mem_permission(task))
 782                 goto out;
 783
 784         ret = -ENOMEM;
 785         page = (char *)__get_free_page(GFP_TEMPORARY);
 786         if (!page)
 787                 goto out;
 788
 789         ret = 0;
 790
 791         mm = get_task_mm(task);
 792         if (!mm)
 793                 goto out_free;
 794
 795         ret = -EIO;
 796
 797         if (file->private_data != (void*)((long)current->self_exec_id))
 798                 goto out_put;
 799
 800         ret = 0;
 801
 802         while (count > 0) {
 803                 int this_len, retval;
 804
 805                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 806                 retval = access_process_vm(task, src, page, this_len, 0);
 807                 if (!retval || check_mem_permission(task)) {
 808                         if (!ret)
 809                                 ret = -EIO;
 810                         break;
 811                 }
 812
 813                 if (copy_to_user(buf, page, retval)) {
 814                         ret = -EFAULT;
 815                         break;
 816                 }
 817
 818                 ret += retval;
 819                 src += retval;
 820                 buf += retval;
 821                 count -= retval;
 822         }
 823         *ppos = src;
 824
 825 out_put:
 826         mmput(mm);
 827 out_free:
 828         free_page((unsigned long) page);
 829 out:
 830         put_task_struct(task);
 831 out_no_task:
 832         return ret;
 833 }
 834
 835 #define mem_write NULL
 836
 837 #ifndef mem_write
 838 /* This is a security hazard */
 839 static ssize_t mem_write(struct file * file, const char __user *buf,
 840                          size_t count, loff_t *ppos)
 841 {
 842         int copied;
 843         char *page;
 844         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 845         unsigned long dst = *ppos;
 846
 847         copied = -ESRCH;
 848         if (!task)
 849                 goto out_no_task;
 850
 851         if (check_mem_permission(task))
 852                 goto out;
 853
 854         copied = -ENOMEM;
 855         page = (char *)__get_free_page(GFP_TEMPORARY);
 856         if (!page)
 857                 goto out;
 858
 859         copied = 0;
 860         while (count > 0) {
 861                 int this_len, retval;
 862
 863                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 864                 if (copy_from_user(page, buf, this_len)) {
 865                         copied = -EFAULT;
 866                         break;
 867                 }
 868                 retval = access_process_vm(task, dst, page, this_len, 1);
 869                 if (!retval) {
 870                         if (!copied)
 871                                 copied = -EIO;
 872                         break;
 873                 }
 874                 copied += retval;
 875                 buf += retval;
 876                 dst += retval;
 877                 count -= retval;
 878         }
 879         *ppos = dst;
 880         free_page((unsigned long) page);
 881 out:
 882         put_task_struct(task);
 883 out_no_task:
 884         return copied;
 885 }
 886 #endif
 887
 888 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 889 {
 890         switch (orig) {
 891         case 0:
 892                 file->f_pos = offset;
 893                 break;
 894         case 1:
 895                 file->f_pos += offset;
 896                 break;
 897         default:
 898                 return -EINVAL;
 899         }
 900         force_successful_syscall_return();
 901         return file->f_pos;
 902 }
 903
 904 static const struct file_operations proc_mem_operations = {
 905         .llseek         = mem_lseek,
 906         .read           = mem_read,
 907         .write          = mem_write,
 908         .open           = mem_open,
 909 };
 910
 911 static ssize_t environ_read(struct file *file, char __user *buf,
 912                         size_t count, loff_t *ppos)
 913 {
 914         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 915         char *page;
 916         unsigned long src = *ppos;
 917         int ret = -ESRCH;
 918         struct mm_struct *mm;
 919
 920         if (!task)
 921                 goto out_no_task;
 922
 923         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 924                 goto out;
 925
 926         ret = -ENOMEM;
 927         page = (char *)__get_free_page(GFP_TEMPORARY);
 928         if (!page)
 929                 goto out;
 930
 931         ret = 0;
 932
 933         mm = get_task_mm(task);
 934         if (!mm)
 935                 goto out_free;
 936
 937         while (count > 0) {
 938                 int this_len, retval, max_len;
 939
 940                 this_len = mm->env_end - (mm->env_start + src);
 941
 942                 if (this_len <= 0)
 943                         break;
 944
 945                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 946                 this_len = (this_len > max_len) ? max_len : this_len;
 947
 948                 retval = access_process_vm(task, (mm->env_start + src),
 949                         page, this_len, 0);
 950
 951                 if (retval <= 0) {
 952                         ret = retval;
 953                         break;
 954                 }
 955
 956                 if (copy_to_user(buf, page, retval)) {
 957                         ret = -EFAULT;
 958                         break;
 959                 }
 960
 961                 ret += retval;
 962                 src += retval;
 963                 buf += retval;
 964                 count -= retval;
 965         }
 966         *ppos = src;
 967
 968         mmput(mm);
 969 out_free:
 970         free_page((unsigned long) page);
 971 out:
 972         put_task_struct(task);
 973 out_no_task:
 974         return ret;
 975 }
 976
 977 static const struct file_operations proc_environ_operations = {
 978         .read           = environ_read,
 979 };
 980
 981 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 982                                 size_t count, loff_t *ppos)
 983 {
 984         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 985         char buffer[PROC_NUMBUF];
 986         size_t len;
 987         int oom_adjust;
 988
 989         if (!task)
 990                 return -ESRCH;
 991         oom_adjust = task->oomkilladj;
 992         put_task_struct(task);
 993
 994         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 995
 996         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 997 }
 998
 999 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1000                                 size_t count, loff_t *ppos)
1001 {
1002         struct task_struct *task;
1003         char buffer[PROC_NUMBUF], *end;
1004         int oom_adjust;
1005
1006         memset(buffer, 0, sizeof(buffer));
1007         if (count > sizeof(buffer) - 1)
1008                 count = sizeof(buffer) - 1;
1009         if (copy_from_user(buffer, buf, count))
1010                 return -EFAULT;
1011         oom_adjust = simple_strtol(buffer, &end, 0);
1012         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1013              oom_adjust != OOM_DISABLE)
1014                 return -EINVAL;
1015         if (*end == '\n')
1016                 end++;
1017         task = get_proc_task(file->f_path.dentry->d_inode);
1018         if (!task)
1019                 return -ESRCH;
1020         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1021                 put_task_struct(task);
1022                 return -EACCES;
1023         }
1024         task->oomkilladj = oom_adjust;
1025         put_task_struct(task);
1026         if (end - buffer == 0)
1027                 return -EIO;
1028         return end - buffer;
1029 }
1030
1031 static const struct file_operations proc_oom_adjust_operations = {
1032         .read           = oom_adjust_read,
1033         .write          = oom_adjust_write,
1034 };
1035
1036 #ifdef CONFIG_AUDITSYSCALL
1037 #define TMPBUFLEN 21
1038 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1039                                   size_t count, loff_t *ppos)
1040 {
1041         struct inode * inode = file->f_path.dentry->d_inode;
1042         struct task_struct *task = get_proc_task(inode);
1043         ssize_t length;
1044         char tmpbuf[TMPBUFLEN];
1045
1046         if (!task)
1047                 return -ESRCH;
1048         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1049                                 audit_get_loginuid(task));
1050         put_task_struct(task);
1051         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1052 }
1053
1054 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1055                                    size_t count, loff_t *ppos)
1056 {
1057         struct inode * inode = file->f_path.dentry->d_inode;
1058         char *page, *tmp;
1059         ssize_t length;
1060         uid_t loginuid;
1061
1062         if (!capable(CAP_AUDIT_CONTROL))
1063                 return -EPERM;
1064
1065         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1066                 return -EPERM;
1067
1068         if (count >= PAGE_SIZE)
1069                 count = PAGE_SIZE - 1;
1070
1071         if (*ppos != 0) {
1072                 /* No partial writes. */
1073                 return -EINVAL;
1074         }
1075         page = (char*)__get_free_page(GFP_TEMPORARY);
1076         if (!page)
1077                 return -ENOMEM;
1078         length = -EFAULT;
1079         if (copy_from_user(page, buf, count))
1080                 goto out_free_page;
1081
1082         page[count] = '\0';
1083         loginuid = simple_strtoul(page, &tmp, 10);
1084         if (tmp == page) {
1085                 length = -EINVAL;
1086                 goto out_free_page;
1087
1088         }
1089         length = audit_set_loginuid(current, loginuid);
1090         if (likely(length == 0))
1091                 length = count;
1092
1093 out_free_page:
1094         free_page((unsigned long) page);
1095         return length;
1096 }
1097
1098 static const struct file_operations proc_loginuid_operations = {
1099         .read           = proc_loginuid_read,
1100         .write          = proc_loginuid_write,
1101 };
1102
1103 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1104                                   size_t count, loff_t *ppos)
1105 {
1106         struct inode * inode = file->f_path.dentry->d_inode;
1107         struct task_struct *task = get_proc_task(inode);
1108         ssize_t length;
1109         char tmpbuf[TMPBUFLEN];
1110
1111         if (!task)
1112                 return -ESRCH;
1113         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1114                                 audit_get_sessionid(task));
1115         put_task_struct(task);
1116         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1117 }
1118
1119 static const struct file_operations proc_sessionid_operations = {
1120         .read           = proc_sessionid_read,
1121 };
1122 #endif
1123
1124 #ifdef CONFIG_FAULT_INJECTION
1125 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1126                                       size_t count, loff_t *ppos)
1127 {
1128         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1129         char buffer[PROC_NUMBUF];
1130         size_t len;
1131         int make_it_fail;
1132
1133         if (!task)
1134                 return -ESRCH;
1135         make_it_fail = task->make_it_fail;
1136         put_task_struct(task);
1137
1138         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1139
1140         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1141 }
1142
1143 static ssize_t proc_fault_inject_write(struct file * file,
1144                         const char __user * buf, size_t count, loff_t *ppos)
1145 {
1146         struct task_struct *task;
1147         char buffer[PROC_NUMBUF], *end;
1148         int make_it_fail;
1149
1150         if (!capable(CAP_SYS_RESOURCE))
1151                 return -EPERM;
1152         memset(buffer, 0, sizeof(buffer));
1153         if (count > sizeof(buffer) - 1)
1154                 count = sizeof(buffer) - 1;
1155         if (copy_from_user(buffer, buf, count))
1156                 return -EFAULT;
1157         make_it_fail = simple_strtol(buffer, &end, 0);
1158         if (*end == '\n')
1159                 end++;
1160         task = get_proc_task(file->f_dentry->d_inode);
1161         if (!task)
1162                 return -ESRCH;
1163         task->make_it_fail = make_it_fail;
1164         put_task_struct(task);
1165         if (end - buffer == 0)
1166                 return -EIO;
1167         return end - buffer;
1168 }
1169
1170 static const struct file_operations proc_fault_inject_operations = {
1171         .read           = proc_fault_inject_read,
1172         .write          = proc_fault_inject_write,
1173 };
1174 #endif
1175
1176
1177 #ifdef CONFIG_SCHED_DEBUG
1178 /*
1179  * Print out various scheduling related per-task fields:
1180  */
1181 static int sched_show(struct seq_file *m, void *v)
1182 {
1183         struct inode *inode = m->private;
1184         struct task_struct *p;
1185
1186         p = get_proc_task(inode);
1187         if (!p)
1188                 return -ESRCH;
1189         proc_sched_show_task(p, m);
1190
1191         put_task_struct(p);
1192
1193         return 0;
1194 }
1195
1196 static ssize_t
1197 sched_write(struct file *file, const char __user *buf,
1198             size_t count, loff_t *offset)
1199 {
1200         struct inode *inode = file->f_path.dentry->d_inode;
1201         struct task_struct *p;
1202
1203         p = get_proc_task(inode);
1204         if (!p)
1205                 return -ESRCH;
1206         proc_sched_set_task(p);
1207
1208         put_task_struct(p);
1209
1210         return count;
1211 }
1212
1213 static int sched_open(struct inode *inode, struct file *filp)
1214 {
1215         int ret;
1216
1217         ret = single_open(filp, sched_show, NULL);
1218         if (!ret) {
1219                 struct seq_file *m = filp->private_data;
1220
1221                 m->private = inode;
1222         }
1223         return ret;
1224 }
1225
1226 static const struct file_operations proc_pid_sched_operations = {
1227         .open           = sched_open,
1228         .read           = seq_read,
1229         .write          = sched_write,
1230         .llseek         = seq_lseek,
1231         .release        = single_release,
1232 };
1233
1234 #endif
1235
1236 /*
1237  * We added or removed a vma mapping the executable. The vmas are only mapped
1238  * during exec and are not mapped with the mmap system call.
1239  * Callers must hold down_write() on the mm's mmap_sem for these
1240  */
1241 void added_exe_file_vma(struct mm_struct *mm)
1242 {
1243         mm->num_exe_file_vmas++;
1244 }
1245
1246 void removed_exe_file_vma(struct mm_struct *mm)
1247 {
1248         mm->num_exe_file_vmas--;
1249         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1250                 fput(mm->exe_file);
1251                 mm->exe_file = NULL;
1252         }
1253
1254 }
1255
1256 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1257 {
1258         if (new_exe_file)
1259                 get_file(new_exe_file);
1260         if (mm->exe_file)
1261                 fput(mm->exe_file);
1262         mm->exe_file = new_exe_file;
1263         mm->num_exe_file_vmas = 0;
1264 }
1265
1266 struct file *get_mm_exe_file(struct mm_struct *mm)
1267 {
1268         struct file *exe_file;
1269
1270         /* We need mmap_sem to protect against races with removal of
1271          * VM_EXECUTABLE vmas */
1272         down_read(&mm->mmap_sem);
1273         exe_file = mm->exe_file;
1274         if (exe_file)
1275                 get_file(exe_file);
1276         up_read(&mm->mmap_sem);
1277         return exe_file;
1278 }
1279
1280 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1281 {
1282         /* It's safe to write the exe_file pointer without exe_file_lock because
1283          * this is called during fork when the task is not yet in /proc */
1284         newmm->exe_file = get_mm_exe_file(oldmm);
1285 }
1286
1287 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1288 {
1289         struct task_struct *task;
1290         struct mm_struct *mm;
1291         struct file *exe_file;
1292
1293         task = get_proc_task(inode);
1294         if (!task)
1295                 return -ENOENT;
1296         mm = get_task_mm(task);
1297         put_task_struct(task);
1298         if (!mm)
1299                 return -ENOENT;
1300         exe_file = get_mm_exe_file(mm);
1301         mmput(mm);
1302         if (exe_file) {
1303                 *exe_path = exe_file->f_path;
1304                 path_get(&exe_file->f_path);
1305                 fput(exe_file);
1306                 return 0;
1307         } else
1308                 return -ENOENT;
1309 }
1310
1311 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1312 {
1313         struct inode *inode = dentry->d_inode;
1314         int error = -EACCES;
1315
1316         /* We don't need a base pointer in the /proc filesystem */
1317         path_put(&nd->path);
1318
1319         /* Are we allowed to snoop on the tasks file descriptors? */
1320         if (!proc_fd_access_allowed(inode))
1321                 goto out;
1322
1323         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1324         nd->last_type = LAST_BIND;
1325 out:
1326         return ERR_PTR(error);
1327 }
1328
1329 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1330 {
1331         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1332         char *pathname;
1333         int len;
1334
1335         if (!tmp)
1336                 return -ENOMEM;
1337
1338         pathname = d_path(path, tmp, PAGE_SIZE);
1339         len = PTR_ERR(pathname);
1340         if (IS_ERR(pathname))
1341                 goto out;
1342         len = tmp + PAGE_SIZE - 1 - pathname;
1343
1344         if (len > buflen)
1345                 len = buflen;
1346         if (copy_to_user(buffer, pathname, len))
1347                 len = -EFAULT;
1348  out:
1349         free_page((unsigned long)tmp);
1350         return len;
1351 }
1352
1353 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1354 {
1355         int error = -EACCES;
1356         struct inode *inode = dentry->d_inode;
1357         struct path path;
1358
1359         /* Are we allowed to snoop on the tasks file descriptors? */
1360         if (!proc_fd_access_allowed(inode))
1361                 goto out;
1362
1363         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1364         if (error)
1365                 goto out;
1366
1367         error = do_proc_readlink(&path, buffer, buflen);
1368         path_put(&path);
1369 out:
1370         return error;
1371 }
1372
1373 static const struct inode_operations proc_pid_link_inode_operations = {
1374         .readlink       = proc_pid_readlink,
1375         .follow_link    = proc_pid_follow_link,
1376         .setattr        = proc_setattr,
1377 };
1378
1379
1380 /* building an inode */
1381
1382 static int task_dumpable(struct task_struct *task)
1383 {
1384         int dumpable = 0;
1385         struct mm_struct *mm;
1386
1387         task_lock(task);
1388         mm = task->mm;
1389         if (mm)
1390                 dumpable = get_dumpable(mm);
1391         task_unlock(task);
1392         if(dumpable == 1)
1393                 return 1;
1394         return 0;
1395 }
1396
1397
1398 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1399 {
1400         struct inode * inode;
1401         struct proc_inode *ei;
1402         const struct cred *cred;
1403
1404         /* We need a new inode */
1405
1406         inode = new_inode(sb);
1407         if (!inode)
1408                 goto out;
1409
1410         /* Common stuff */
1411         ei = PROC_I(inode);
1412         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1413         inode->i_op = &proc_def_inode_operations;
1414
1415         /*
1416          * grab the reference to task.
1417          */
1418         ei->pid = get_task_pid(task, PIDTYPE_PID);
1419         if (!ei->pid)
1420                 goto out_unlock;
1421
1422         inode->i_uid = 0;
1423         inode->i_gid = 0;
1424         if (task_dumpable(task)) {
1425                 rcu_read_lock();
1426                 cred = __task_cred(task);
1427                 inode->i_uid = cred->euid;
1428                 inode->i_gid = cred->egid;
1429                 rcu_read_unlock();
1430         }
1431         security_task_to_inode(task, inode);
1432
1433 out:
1434         return inode;
1435
1436 out_unlock:
1437         iput(inode);
1438         return NULL;
1439 }
1440
1441 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1442 {
1443         struct inode *inode = dentry->d_inode;
1444         struct task_struct *task;
1445         const struct cred *cred;
1446
1447         generic_fillattr(inode, stat);
1448
1449         rcu_read_lock();
1450         stat->uid = 0;
1451         stat->gid = 0;
1452         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1453         if (task) {
1454                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1455                     task_dumpable(task)) {
1456                         cred = __task_cred(task);
1457                         stat->uid = cred->euid;
1458                         stat->gid = cred->egid;
1459                 }
1460         }
1461         rcu_read_unlock();
1462         return 0;
1463 }
1464
1465 /* dentry stuff */
1466
1467 /*
1468  *      Exceptional case: normally we are not allowed to unhash a busy
1469  * directory. In this case, however, we can do it - no aliasing problems
1470  * due to the way we treat inodes.
1471  *
1472  * Rewrite the inode's ownerships here because the owning task may have
1473  * performed a setuid(), etc.
1474  *
1475  * Before the /proc/pid/status file was created the only way to read
1476  * the effective uid of a /process was to stat /proc/pid.  Reading
1477  * /proc/pid/status is slow enough that procps and other packages
1478  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1479  * made this apply to all per process world readable and executable
1480  * directories.
1481  */
1482 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1483 {
1484         struct inode *inode = dentry->d_inode;
1485         struct task_struct *task = get_proc_task(inode);
1486         const struct cred *cred;
1487
1488         if (task) {
1489                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1490                     task_dumpable(task)) {
1491                         rcu_read_lock();
1492                         cred = __task_cred(task);
1493                         inode->i_uid = cred->euid;
1494                         inode->i_gid = cred->egid;
1495                         rcu_read_unlock();
1496                 } else {
1497                         inode->i_uid = 0;
1498                         inode->i_gid = 0;
1499                 }
1500                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1501                 security_task_to_inode(task, inode);
1502                 put_task_struct(task);
1503                 return 1;
1504         }
1505         d_drop(dentry);
1506         return 0;
1507 }
1508
1509 static int pid_delete_dentry(struct dentry * dentry)
1510 {
1511         /* Is the task we represent dead?
1512          * If so, then don't put the dentry on the lru list,
1513          * kill it immediately.
1514          */
1515         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1516 }
1517
1518 static struct dentry_operations pid_dentry_operations =
1519 {
1520         .d_revalidate   = pid_revalidate,
1521         .d_delete       = pid_delete_dentry,
1522 };
1523
1524 /* Lookups */
1525
1526 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1527                                 struct task_struct *, const void *);
1528
1529 /*
1530  * Fill a directory entry.
1531  *
1532  * If possible create the dcache entry and derive our inode number and
1533  * file type from dcache entry.
1534  *
1535  * Since all of the proc inode numbers are dynamically generated, the inode
1536  * numbers do not exist until the inode is cache.  This means creating the
1537  * the dcache entry in readdir is necessary to keep the inode numbers
1538  * reported by readdir in sync with the inode numbers reported
1539  * by stat.
1540  */
1541 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1542         char *name, int len,
1543         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1544 {
1545         struct dentry *child, *dir = filp->f_path.dentry;
1546         struct inode *inode;
1547         struct qstr qname;
1548         ino_t ino = 0;
1549         unsigned type = DT_UNKNOWN;
1550
1551         qname.name = name;
1552         qname.len  = len;
1553         qname.hash = full_name_hash(name, len);
1554
1555         child = d_lookup(dir, &qname);
1556         if (!child) {
1557                 struct dentry *new;
1558                 new = d_alloc(dir, &qname);
1559                 if (new) {
1560                         child = instantiate(dir->d_inode, new, task, ptr);
1561                         if (child)
1562                                 dput(new);
1563                         else
1564                                 child = new;
1565                 }
1566         }
1567         if (!child || IS_ERR(child) || !child->d_inode)
1568                 goto end_instantiate;
1569         inode = child->d_inode;
1570         if (inode) {
1571                 ino = inode->i_ino;
1572                 type = inode->i_mode >> 12;
1573         }
1574         dput(child);
1575 end_instantiate:
1576         if (!ino)
1577                 ino = find_inode_number(dir, &qname);
1578         if (!ino)
1579                 ino = 1;
1580         return filldir(dirent, name, len, filp->f_pos, ino, type);
1581 }
1582
1583 static unsigned name_to_int(struct dentry *dentry)
1584 {
1585         const char *name = dentry->d_name.name;
1586         int len = dentry->d_name.len;
1587         unsigned n = 0;
1588
1589         if (len > 1 && *name == '0')
1590                 goto out;
1591         while (len-- > 0) {
1592                 unsigned c = *name++ - '0';
1593                 if (c > 9)
1594                         goto out;
1595                 if (n >= (~0U-9)/10)
1596                         goto out;
1597                 n *= 10;
1598                 n += c;
1599         }
1600         return n;
1601 out:
1602         return ~0U;
1603 }
1604
1605 #define PROC_FDINFO_MAX 64
1606
1607 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1608 {
1609         struct task_struct *task = get_proc_task(inode);
1610         struct files_struct *files = NULL;
1611         struct file *file;
1612         int fd = proc_fd(inode);
1613
1614         if (task) {
1615                 files = get_files_struct(task);
1616                 put_task_struct(task);
1617         }
1618         if (files) {
1619                 /*
1620                  * We are not taking a ref to the file structure, so we must
1621                  * hold ->file_lock.
1622                  */
1623                 spin_lock(&files->file_lock);
1624                 file = fcheck_files(files, fd);
1625                 if (file) {
1626                         if (path) {
1627                                 *path = file->f_path;
1628                                 path_get(&file->f_path);
1629                         }
1630                         if (info)
1631                                 snprintf(info, PROC_FDINFO_MAX,
1632                                          "pos:\t%lli\n"
1633                                          "flags:\t0%o\n",
1634                                          (long long) file->f_pos,
1635                                          file->f_flags);
1636                         spin_unlock(&files->file_lock);
1637                         put_files_struct(files);
1638                         return 0;
1639                 }
1640                 spin_unlock(&files->file_lock);
1641                 put_files_struct(files);
1642         }
1643         return -ENOENT;
1644 }
1645
1646 static int proc_fd_link(struct inode *inode, struct path *path)
1647 {
1648         return proc_fd_info(inode, path, NULL);
1649 }
1650
1651 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1652 {
1653         struct inode *inode = dentry->d_inode;
1654         struct task_struct *task = get_proc_task(inode);
1655         int fd = proc_fd(inode);
1656         struct files_struct *files;
1657         const struct cred *cred;
1658
1659         if (task) {
1660                 files = get_files_struct(task);
1661                 if (files) {
1662                         rcu_read_lock();
1663                         if (fcheck_files(files, fd)) {
1664                                 rcu_read_unlock();
1665                                 put_files_struct(files);
1666                                 if (task_dumpable(task)) {
1667                                         rcu_read_lock();
1668                                         cred = __task_cred(task);
1669                                         inode->i_uid = cred->euid;
1670                                         inode->i_gid = cred->egid;
1671                                         rcu_read_unlock();
1672                                 } else {
1673                                         inode->i_uid = 0;
1674                                         inode->i_gid = 0;
1675                                 }
1676                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1677                                 security_task_to_inode(task, inode);
1678                                 put_task_struct(task);
1679                                 return 1;
1680                         }
1681                         rcu_read_unlock();
1682                         put_files_struct(files);
1683                 }
1684                 put_task_struct(task);
1685         }
1686         d_drop(dentry);
1687         return 0;
1688 }
1689
1690 static struct dentry_operations tid_fd_dentry_operations =
1691 {
1692         .d_revalidate   = tid_fd_revalidate,
1693         .d_delete       = pid_delete_dentry,
1694 };
1695
1696 static struct dentry *proc_fd_instantiate(struct inode *dir,
1697         struct dentry *dentry, struct task_struct *task, const void *ptr)
1698 {
1699         unsigned fd = *(const unsigned *)ptr;
1700         struct file *file;
1701         struct files_struct *files;
1702         struct inode *inode;
1703         struct proc_inode *ei;
1704         struct dentry *error = ERR_PTR(-ENOENT);
1705
1706         inode = proc_pid_make_inode(dir->i_sb, task);
1707         if (!inode)
1708                 goto out;
1709         ei = PROC_I(inode);
1710         ei->fd = fd;
1711         files = get_files_struct(task);
1712         if (!files)
1713                 goto out_iput;
1714         inode->i_mode = S_IFLNK;
1715
1716         /*
1717          * We are not taking a ref to the file structure, so we must
1718          * hold ->file_lock.
1719          */
1720         spin_lock(&files->file_lock);
1721         file = fcheck_files(files, fd);
1722         if (!file)
1723                 goto out_unlock;
1724         if (file->f_mode & FMODE_READ)
1725                 inode->i_mode |= S_IRUSR | S_IXUSR;
1726         if (file->f_mode & FMODE_WRITE)
1727                 inode->i_mode |= S_IWUSR | S_IXUSR;
1728         spin_unlock(&files->file_lock);
1729         put_files_struct(files);
1730
1731         inode->i_op = &proc_pid_link_inode_operations;
1732         inode->i_size = 64;
1733         ei->op.proc_get_link = proc_fd_link;
1734         dentry->d_op = &tid_fd_dentry_operations;
1735         d_add(dentry, inode);
1736         /* Close the race of the process dying before we return the dentry */
1737         if (tid_fd_revalidate(dentry, NULL))
1738                 error = NULL;
1739
1740  out:
1741         return error;
1742 out_unlock:
1743         spin_unlock(&files->file_lock);
1744         put_files_struct(files);
1745 out_iput:
1746         iput(inode);
1747         goto out;
1748 }
1749
1750 static struct dentry *proc_lookupfd_common(struct inode *dir,
1751                                            struct dentry *dentry,
1752                                            instantiate_t instantiate)
1753 {
1754         struct task_struct *task = get_proc_task(dir);
1755         unsigned fd = name_to_int(dentry);
1756         struct dentry *result = ERR_PTR(-ENOENT);
1757
1758         if (!task)
1759                 goto out_no_task;
1760         if (fd == ~0U)
1761                 goto out;
1762
1763         result = instantiate(dir, dentry, task, &fd);
1764 out:
1765         put_task_struct(task);
1766 out_no_task:
1767         return result;
1768 }
1769
1770 static int proc_readfd_common(struct file * filp, void * dirent,
1771                               filldir_t filldir, instantiate_t instantiate)
1772 {
1773         struct dentry *dentry = filp->f_path.dentry;
1774         struct inode *inode = dentry->d_inode;
1775         struct task_struct *p = get_proc_task(inode);
1776         unsigned int fd, ino;
1777         int retval;
1778         struct files_struct * files;
1779
1780         retval = -ENOENT;
1781         if (!p)
1782                 goto out_no_task;
1783         retval = 0;
1784
1785         fd = filp->f_pos;
1786         switch (fd) {
1787                 case 0:
1788                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1789                                 goto out;
1790                         filp->f_pos++;
1791                 case 1:
1792                         ino = parent_ino(dentry);
1793                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1794                                 goto out;
1795                         filp->f_pos++;
1796                 default:
1797                         files = get_files_struct(p);
1798                         if (!files)
1799                                 goto out;
1800                         rcu_read_lock();
1801                         for (fd = filp->f_pos-2;
1802                              fd < files_fdtable(files)->max_fds;
1803                              fd++, filp->f_pos++) {
1804                                 char name[PROC_NUMBUF];
1805                                 int len;
1806
1807                                 if (!fcheck_files(files, fd))
1808                                         continue;
1809                                 rcu_read_unlock();
1810
1811                                 len = snprintf(name, sizeof(name), "%d", fd);
1812                                 if (proc_fill_cache(filp, dirent, filldir,
1813                                                     name, len, instantiate,
1814                                                     p, &fd) < 0) {
1815                                         rcu_read_lock();
1816                                         break;
1817                                 }
1818                                 rcu_read_lock();
1819                         }
1820                         rcu_read_unlock();
1821                         put_files_struct(files);
1822         }
1823 out:
1824         put_task_struct(p);
1825 out_no_task:
1826         return retval;
1827 }
1828
1829 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1830                                     struct nameidata *nd)
1831 {
1832         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1833 }
1834
1835 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1836 {
1837         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1838 }
1839
1840 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1841                                       size_t len, loff_t *ppos)
1842 {
1843         char tmp[PROC_FDINFO_MAX];
1844         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1845         if (!err)
1846                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1847         return err;
1848 }
1849
1850 static const struct file_operations proc_fdinfo_file_operations = {
1851         .open           = nonseekable_open,
1852         .read           = proc_fdinfo_read,
1853 };
1854
1855 static const struct file_operations proc_fd_operations = {
1856         .read           = generic_read_dir,
1857         .readdir        = proc_readfd,
1858 };
1859
1860 /*
1861  * /proc/pid/fd needs a special permission handler so that a process can still
1862  * access /proc/self/fd after it has executed a setuid().
1863  */
1864 static int proc_fd_permission(struct inode *inode, int mask)
1865 {
1866         int rv;
1867
1868         rv = generic_permission(inode, mask, NULL);
1869         if (rv == 0)
1870                 return 0;
1871         if (task_pid(current) == proc_pid(inode))
1872                 rv = 0;
1873         return rv;
1874 }
1875
1876 /*
1877  * proc directories can do almost nothing..
1878  */
1879 static const struct inode_operations proc_fd_inode_operations = {
1880         .lookup         = proc_lookupfd,
1881         .permission     = proc_fd_permission,
1882         .setattr        = proc_setattr,
1883 };
1884
1885 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1886         struct dentry *dentry, struct task_struct *task, const void *ptr)
1887 {
1888         unsigned fd = *(unsigned *)ptr;
1889         struct inode *inode;
1890         struct proc_inode *ei;
1891         struct dentry *error = ERR_PTR(-ENOENT);
1892
1893         inode = proc_pid_make_inode(dir->i_sb, task);
1894         if (!inode)
1895                 goto out;
1896         ei = PROC_I(inode);
1897         ei->fd = fd;
1898         inode->i_mode = S_IFREG | S_IRUSR;
1899         inode->i_fop = &proc_fdinfo_file_operations;
1900         dentry->d_op = &tid_fd_dentry_operations;
1901         d_add(dentry, inode);
1902         /* Close the race of the process dying before we return the dentry */
1903         if (tid_fd_revalidate(dentry, NULL))
1904                 error = NULL;
1905
1906  out:
1907         return error;
1908 }
1909
1910 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1911                                         struct dentry *dentry,
1912                                         struct nameidata *nd)
1913 {
1914         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1915 }
1916
1917 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1918 {
1919         return proc_readfd_common(filp, dirent, filldir,
1920                                   proc_fdinfo_instantiate);
1921 }
1922
1923 static const struct file_operations proc_fdinfo_operations = {
1924         .read           = generic_read_dir,
1925         .readdir        = proc_readfdinfo,
1926 };
1927
1928 /*
1929  * proc directories can do almost nothing..
1930  */
1931 static const struct inode_operations proc_fdinfo_inode_operations = {
1932         .lookup         = proc_lookupfdinfo,
1933         .setattr        = proc_setattr,
1934 };
1935
1936
1937 static struct dentry *proc_pident_instantiate(struct inode *dir,
1938         struct dentry *dentry, struct task_struct *task, const void *ptr)
1939 {
1940         const struct pid_entry *p = ptr;
1941         struct inode *inode;
1942         struct proc_inode *ei;
1943         struct dentry *error = ERR_PTR(-EINVAL);
1944
1945         inode = proc_pid_make_inode(dir->i_sb, task);
1946         if (!inode)
1947                 goto out;
1948
1949         ei = PROC_I(inode);
1950         inode->i_mode = p->mode;
1951         if (S_ISDIR(inode->i_mode))
1952                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1953         if (p->iop)
1954                 inode->i_op = p->iop;
1955         if (p->fop)
1956                 inode->i_fop = p->fop;
1957         ei->op = p->op;
1958         dentry->d_op = &pid_dentry_operations;
1959         d_add(dentry, inode);
1960         /* Close the race of the process dying before we return the dentry */
1961         if (pid_revalidate(dentry, NULL))
1962                 error = NULL;
1963 out:
1964         return error;
1965 }
1966
1967 static struct dentry *proc_pident_lookup(struct inode *dir,
1968                                          struct dentry *dentry,
1969                                          const struct pid_entry *ents,
1970                                          unsigned int nents)
1971 {
1972         struct inode *inode;
1973         struct dentry *error;
1974         struct task_struct *task = get_proc_task(dir);
1975         const struct pid_entry *p, *last;
1976
1977         error = ERR_PTR(-ENOENT);
1978         inode = NULL;
1979
1980         if (!task)
1981                 goto out_no_task;
1982
1983         /*
1984          * Yes, it does not scale. And it should not. Don't add
1985          * new entries into /proc/<tgid>/ without very good reasons.
1986          */
1987         last = &ents[nents - 1];
1988         for (p = ents; p <= last; p++) {
1989                 if (p->len != dentry->d_name.len)
1990                         continue;
1991                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1992                         break;
1993         }
1994         if (p > last)
1995                 goto out;
1996
1997         error = proc_pident_instantiate(dir, dentry, task, p);
1998 out:
1999         put_task_struct(task);
2000 out_no_task:
2001         return error;
2002 }
2003
2004 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2005         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2006 {
2007         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2008                                 proc_pident_instantiate, task, p);
2009 }
2010
2011 static int proc_pident_readdir(struct file *filp,
2012                 void *dirent, filldir_t filldir,
2013                 const struct pid_entry *ents, unsigned int nents)
2014 {
2015         int i;
2016         struct dentry *dentry = filp->f_path.dentry;
2017         struct inode *inode = dentry->d_inode;
2018         struct task_struct *task = get_proc_task(inode);
2019         const struct pid_entry *p, *last;
2020         ino_t ino;
2021         int ret;
2022
2023         ret = -ENOENT;
2024         if (!task)
2025                 goto out_no_task;
2026
2027         ret = 0;
2028         i = filp->f_pos;
2029         switch (i) {
2030         case 0:
2031                 ino = inode->i_ino;
2032                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2033                         goto out;
2034                 i++;
2035                 filp->f_pos++;
2036                 /* fall through */
2037         case 1:
2038                 ino = parent_ino(dentry);
2039                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2040                         goto out;
2041                 i++;
2042                 filp->f_pos++;
2043                 /* fall through */
2044         default:
2045                 i -= 2;
2046                 if (i >= nents) {
2047                         ret = 1;
2048                         goto out;
2049                 }
2050                 p = ents + i;
2051                 last = &ents[nents - 1];
2052                 while (p <= last) {
2053                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2054                                 goto out;
2055                         filp->f_pos++;
2056                         p++;
2057                 }
2058         }
2059
2060         ret = 1;
2061 out:
2062         put_task_struct(task);
2063 out_no_task:
2064         return ret;
2065 }
2066
2067 #ifdef CONFIG_SECURITY
2068 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2069                                   size_t count, loff_t *ppos)
2070 {
2071         struct inode * inode = file->f_path.dentry->d_inode;
2072         char *p = NULL;
2073         ssize_t length;
2074         struct task_struct *task = get_proc_task(inode);
2075
2076         if (!task)
2077                 return -ESRCH;
2078
2079         length = security_getprocattr(task,
2080                                       (char*)file->f_path.dentry->d_name.name,
2081                                       &p);
2082         put_task_struct(task);
2083         if (length > 0)
2084                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2085         kfree(p);
2086         return length;
2087 }
2088
2089 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2090                                    size_t count, loff_t *ppos)
2091 {
2092         struct inode * inode = file->f_path.dentry->d_inode;
2093         char *page;
2094         ssize_t length;
2095         struct task_struct *task = get_proc_task(inode);
2096
2097         length = -ESRCH;
2098         if (!task)
2099                 goto out_no_task;
2100         if (count > PAGE_SIZE)
2101                 count = PAGE_SIZE;
2102
2103         /* No partial writes. */
2104         length = -EINVAL;
2105         if (*ppos != 0)
2106                 goto out;
2107
2108         length = -ENOMEM;
2109         page = (char*)__get_free_page(GFP_TEMPORARY);
2110         if (!page)
2111                 goto out;
2112
2113         length = -EFAULT;
2114         if (copy_from_user(page, buf, count))
2115                 goto out_free;
2116
2117         length = security_setprocattr(task,
2118                                       (char*)file->f_path.dentry->d_name.name,
2119                                       (void*)page, count);
2120 out_free:
2121         free_page((unsigned long) page);
2122 out:
2123         put_task_struct(task);
2124 out_no_task:
2125         return length;
2126 }
2127
2128 static const struct file_operations proc_pid_attr_operations = {
2129         .read           = proc_pid_attr_read,
2130         .write          = proc_pid_attr_write,
2131 };
2132
2133 static const struct pid_entry attr_dir_stuff[] = {
2134         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2135         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2136         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2137         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2138         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2139         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2140 };
2141
2142 static int proc_attr_dir_readdir(struct file * filp,
2143                              void * dirent, filldir_t filldir)
2144 {
2145         return proc_pident_readdir(filp,dirent,filldir,
2146                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2147 }
2148
2149 static const struct file_operations proc_attr_dir_operations = {
2150         .read           = generic_read_dir,
2151         .readdir        = proc_attr_dir_readdir,
2152 };
2153
2154 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2155                                 struct dentry *dentry, struct nameidata *nd)
2156 {
2157         return proc_pident_lookup(dir, dentry,
2158                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2159 }
2160
2161 static const struct inode_operations proc_attr_dir_inode_operations = {
2162         .lookup         = proc_attr_dir_lookup,
2163         .getattr        = pid_getattr,
2164         .setattr        = proc_setattr,
2165 };
2166
2167 #endif
2168
2169 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2170 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2171                                          size_t count, loff_t *ppos)
2172 {
2173         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2174         struct mm_struct *mm;
2175         char buffer[PROC_NUMBUF];
2176         size_t len;
2177         int ret;
2178
2179         if (!task)
2180                 return -ESRCH;
2181
2182         ret = 0;
2183         mm = get_task_mm(task);
2184         if (mm) {
2185                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2186                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2187                                 MMF_DUMP_FILTER_SHIFT));
2188                 mmput(mm);
2189                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2190         }
2191
2192         put_task_struct(task);
2193
2194         return ret;
2195 }
2196
2197 static ssize_t proc_coredump_filter_write(struct file *file,
2198                                           const char __user *buf,
2199                                           size_t count,
2200                                           loff_t *ppos)
2201 {
2202         struct task_struct *task;
2203         struct mm_struct *mm;
2204         char buffer[PROC_NUMBUF], *end;
2205         unsigned int val;
2206         int ret;
2207         int i;
2208         unsigned long mask;
2209
2210         ret = -EFAULT;
2211         memset(buffer, 0, sizeof(buffer));
2212         if (count > sizeof(buffer) - 1)
2213                 count = sizeof(buffer) - 1;
2214         if (copy_from_user(buffer, buf, count))
2215                 goto out_no_task;
2216
2217         ret = -EINVAL;
2218         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2219         if (*end == '\n')
2220                 end++;
2221         if (end - buffer == 0)
2222                 goto out_no_task;
2223
2224         ret = -ESRCH;
2225         task = get_proc_task(file->f_dentry->d_inode);
2226         if (!task)
2227                 goto out_no_task;
2228
2229         ret = end - buffer;
2230         mm = get_task_mm(task);
2231         if (!mm)
2232                 goto out_no_mm;
2233
2234         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2235                 if (val & mask)
2236                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2237                 else
2238                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2239         }
2240
2241         mmput(mm);
2242  out_no_mm:
2243         put_task_struct(task);
2244  out_no_task:
2245         return ret;
2246 }
2247
2248 static const struct file_operations proc_coredump_filter_operations = {
2249         .read           = proc_coredump_filter_read,
2250         .write          = proc_coredump_filter_write,
2251 };
2252 #endif
2253
2254 /*
2255  * /proc/self:
2256  */
2257 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2258                               int buflen)
2259 {
2260         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2261         pid_t tgid = task_tgid_nr_ns(current, ns);
2262         char tmp[PROC_NUMBUF];
2263         if (!tgid)
2264                 return -ENOENT;
2265         sprintf(tmp, "%d", tgid);
2266         return vfs_readlink(dentry,buffer,buflen,tmp);
2267 }
2268
2269 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2270 {
2271         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2272         pid_t tgid = task_tgid_nr_ns(current, ns);
2273         char tmp[PROC_NUMBUF];
2274         if (!tgid)
2275                 return ERR_PTR(-ENOENT);
2276         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2277         return ERR_PTR(vfs_follow_link(nd,tmp));
2278 }
2279
2280 static const struct inode_operations proc_self_inode_operations = {
2281         .readlink       = proc_self_readlink,
2282         .follow_link    = proc_self_follow_link,
2283 };
2284
2285 /*
2286  * proc base
2287  *
2288  * These are the directory entries in the root directory of /proc
2289  * that properly belong to the /proc filesystem, as they describe
2290  * describe something that is process related.
2291  */
2292 static const struct pid_entry proc_base_stuff[] = {
2293         NOD("self", S_IFLNK|S_IRWXUGO,
2294                 &proc_self_inode_operations, NULL, {}),
2295 };
2296
2297 /*
2298  *      Exceptional case: normally we are not allowed to unhash a busy
2299  * directory. In this case, however, we can do it - no aliasing problems
2300  * due to the way we treat inodes.
2301  */
2302 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2303 {
2304         struct inode *inode = dentry->d_inode;
2305         struct task_struct *task = get_proc_task(inode);
2306         if (task) {
2307                 put_task_struct(task);
2308                 return 1;
2309         }
2310         d_drop(dentry);
2311         return 0;
2312 }
2313
2314 static struct dentry_operations proc_base_dentry_operations =
2315 {
2316         .d_revalidate   = proc_base_revalidate,
2317         .d_delete       = pid_delete_dentry,
2318 };
2319
2320 static struct dentry *proc_base_instantiate(struct inode *dir,
2321         struct dentry *dentry, struct task_struct *task, const void *ptr)
2322 {
2323         const struct pid_entry *p = ptr;
2324         struct inode *inode;
2325         struct proc_inode *ei;
2326         struct dentry *error = ERR_PTR(-EINVAL);
2327
2328         /* Allocate the inode */
2329         error = ERR_PTR(-ENOMEM);
2330         inode = new_inode(dir->i_sb);
2331         if (!inode)
2332                 goto out;
2333
2334         /* Initialize the inode */
2335         ei = PROC_I(inode);
2336         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2337
2338         /*
2339          * grab the reference to the task.
2340          */
2341         ei->pid = get_task_pid(task, PIDTYPE_PID);
2342         if (!ei->pid)
2343                 goto out_iput;
2344
2345         inode->i_uid = 0;
2346         inode->i_gid = 0;
2347         inode->i_mode = p->mode;
2348         if (S_ISDIR(inode->i_mode))
2349                 inode->i_nlink = 2;
2350         if (S_ISLNK(inode->i_mode))
2351                 inode->i_size = 64;
2352         if (p->iop)
2353                 inode->i_op = p->iop;
2354         if (p->fop)
2355                 inode->i_fop = p->fop;
2356         ei->op = p->op;
2357         dentry->d_op = &proc_base_dentry_operations;
2358         d_add(dentry, inode);
2359         error = NULL;
2360 out:
2361         return error;
2362 out_iput:
2363         iput(inode);
2364         goto out;
2365 }
2366
2367 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2368 {
2369         struct dentry *error;
2370         struct task_struct *task = get_proc_task(dir);
2371         const struct pid_entry *p, *last;
2372
2373         error = ERR_PTR(-ENOENT);
2374
2375         if (!task)
2376                 goto out_no_task;
2377
2378         /* Lookup the directory entry */
2379         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2380         for (p = proc_base_stuff; p <= last; p++) {
2381                 if (p->len != dentry->d_name.len)
2382                         continue;
2383                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2384                         break;
2385         }
2386         if (p > last)
2387                 goto out;
2388
2389         error = proc_base_instantiate(dir, dentry, task, p);
2390
2391 out:
2392         put_task_struct(task);
2393 out_no_task:
2394         return error;
2395 }
2396
2397 static int proc_base_fill_cache(struct file *filp, void *dirent,
2398         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2399 {
2400         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2401                                 proc_base_instantiate, task, p);
2402 }
2403
2404 #ifdef CONFIG_TASK_IO_ACCOUNTING
2405 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2406 {
2407         struct task_io_accounting acct = task->ioac;
2408         unsigned long flags;
2409
2410         if (whole && lock_task_sighand(task, &flags)) {
2411                 struct task_struct *t = task;
2412
2413                 task_io_accounting_add(&acct, &task->signal->ioac);
2414                 while_each_thread(task, t)
2415                         task_io_accounting_add(&acct, &t->ioac);
2416
2417                 unlock_task_sighand(task, &flags);
2418         }
2419         return sprintf(buffer,
2420                         "rchar: %llu\n"
2421                         "wchar: %llu\n"
2422                         "syscr: %llu\n"
2423                         "syscw: %llu\n"
2424                         "read_bytes: %llu\n"
2425                         "write_bytes: %llu\n"
2426                         "cancelled_write_bytes: %llu\n",
2427                         (unsigned long long)acct.rchar,
2428                         (unsigned long long)acct.wchar,
2429                         (unsigned long long)acct.syscr,
2430                         (unsigned long long)acct.syscw,
2431                         (unsigned long long)acct.read_bytes,
2432                         (unsigned long long)acct.write_bytes,
2433                         (unsigned long long)acct.cancelled_write_bytes);
2434 }
2435
2436 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2437 {
2438         return do_io_accounting(task, buffer, 0);
2439 }
2440
2441 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2442 {
2443         return do_io_accounting(task, buffer, 1);
2444 }
2445 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2446
2447 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2448                                 struct pid *pid, struct task_struct *task)
2449 {
2450         seq_printf(m, "%08x\n", task->personality);
2451         return 0;
2452 }
2453
2454 /*
2455  * Thread groups
2456  */
2457 static const struct file_operations proc_task_operations;
2458 static const struct inode_operations proc_task_inode_operations;
2459
2460 static const struct pid_entry tgid_base_stuff[] = {
2461         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2462         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2463         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2464 #ifdef CONFIG_NET
2465         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2466 #endif
2467         REG("environ",    S_IRUSR, proc_environ_operations),
2468         INF("auxv",       S_IRUSR, proc_pid_auxv),
2469         ONE("status",     S_IRUGO, proc_pid_status),
2470         ONE("personality", S_IRUSR, proc_pid_personality),
2471         INF("limits",     S_IRUSR, proc_pid_limits),
2472 #ifdef CONFIG_SCHED_DEBUG
2473         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2474 #endif
2475 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2476         INF("syscall",    S_IRUSR, proc_pid_syscall),
2477 #endif
2478         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2479         ONE("stat",       S_IRUGO, proc_tgid_stat),
2480         ONE("statm",      S_IRUGO, proc_pid_statm),
2481         REG("maps",       S_IRUGO, proc_maps_operations),
2482 #ifdef CONFIG_NUMA
2483         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2484 #endif
2485         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2486         LNK("cwd",        proc_cwd_link),
2487         LNK("root",       proc_root_link),
2488         LNK("exe",        proc_exe_link),
2489         REG("mounts",     S_IRUGO, proc_mounts_operations),
2490         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2491         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2492 #ifdef CONFIG_PROC_PAGE_MONITOR
2493         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2494         REG("smaps",      S_IRUGO, proc_smaps_operations),
2495         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2496 #endif
2497 #ifdef CONFIG_SECURITY
2498         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2499 #endif
2500 #ifdef CONFIG_KALLSYMS
2501         INF("wchan",      S_IRUGO, proc_pid_wchan),
2502 #endif
2503 #ifdef CONFIG_SCHEDSTATS
2504         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2505 #endif
2506 #ifdef CONFIG_LATENCYTOP
2507         REG("latency",  S_IRUGO, proc_lstats_operations),
2508 #endif
2509 #ifdef CONFIG_PROC_PID_CPUSET
2510         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2511 #endif
2512 #ifdef CONFIG_CGROUPS
2513         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2514 #endif
2515         INF("oom_score",  S_IRUGO, proc_oom_score),
2516         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2517 #ifdef CONFIG_AUDITSYSCALL
2518         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2519         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2520 #endif
2521 #ifdef CONFIG_FAULT_INJECTION
2522         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2523 #endif
2524 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2525         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2526 #endif
2527 #ifdef CONFIG_TASK_IO_ACCOUNTING
2528         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2529 #endif
2530 };
2531
2532 static int proc_tgid_base_readdir(struct file * filp,
2533                              void * dirent, filldir_t filldir)
2534 {
2535         return proc_pident_readdir(filp,dirent,filldir,
2536                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2537 }
2538
2539 static const struct file_operations proc_tgid_base_operations = {
2540         .read           = generic_read_dir,
2541         .readdir        = proc_tgid_base_readdir,
2542 };
2543
2544 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2545         return proc_pident_lookup(dir, dentry,
2546                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2547 }
2548
2549 static const struct inode_operations proc_tgid_base_inode_operations = {
2550         .lookup         = proc_tgid_base_lookup,
2551         .getattr        = pid_getattr,
2552         .setattr        = proc_setattr,
2553 };
2554
2555 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2556 {
2557         struct dentry *dentry, *leader, *dir;
2558         char buf[PROC_NUMBUF];
2559         struct qstr name;
2560
2561         name.name = buf;
2562         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2563         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2564         if (dentry) {
2565                 if (!(current->flags & PF_EXITING))
2566                         shrink_dcache_parent(dentry);
2567                 d_drop(dentry);
2568                 dput(dentry);
2569         }
2570
2571         if (tgid == 0)
2572                 goto out;
2573
2574         name.name = buf;
2575         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2576         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2577         if (!leader)
2578                 goto out;
2579
2580         name.name = "task";
2581         name.len = strlen(name.name);
2582         dir = d_hash_and_lookup(leader, &name);
2583         if (!dir)
2584                 goto out_put_leader;
2585
2586         name.name = buf;
2587         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2588         dentry = d_hash_and_lookup(dir, &name);
2589         if (dentry) {
2590                 shrink_dcache_parent(dentry);
2591                 d_drop(dentry);
2592                 dput(dentry);
2593         }
2594
2595         dput(dir);
2596 out_put_leader:
2597         dput(leader);
2598 out:
2599         return;
2600 }
2601
2602 /**
2603  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2604  * @task: task that should be flushed.
2605  *
2606  * When flushing dentries from proc, one needs to flush them from global
2607  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2608  * in. This call is supposed to do all of this job.
2609  *
2610  * Looks in the dcache for
2611  * /proc/@pid
2612  * /proc/@tgid/task/@pid
2613  * if either directory is present flushes it and all of it'ts children
2614  * from the dcache.
2615  *
2616  * It is safe and reasonable to cache /proc entries for a task until
2617  * that task exits.  After that they just clog up the dcache with
2618  * useless entries, possibly causing useful dcache entries to be
2619  * flushed instead.  This routine is proved to flush those useless
2620  * dcache entries at process exit time.
2621  *
2622  * NOTE: This routine is just an optimization so it does not guarantee
2623  *       that no dcache entries will exist at process exit time it
2624  *       just makes it very unlikely that any will persist.
2625  */
2626
2627 void proc_flush_task(struct task_struct *task)
2628 {
2629         int i;
2630         struct pid *pid, *tgid = NULL;
2631         struct upid *upid;
2632
2633         pid = task_pid(task);
2634         if (thread_group_leader(task))
2635                 tgid = task_tgid(task);
2636
2637         for (i = 0; i <= pid->level; i++) {
2638                 upid = &pid->numbers[i];
2639                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2640                         tgid ? tgid->numbers[i].nr : 0);
2641         }
2642
2643         upid = &pid->numbers[pid->level];
2644         if (upid->nr == 1)
2645                 pid_ns_release_proc(upid->ns);
2646 }
2647
2648 static struct dentry *proc_pid_instantiate(struct inode *dir,
2649                                            struct dentry * dentry,
2650                                            struct task_struct *task, const void *ptr)
2651 {
2652         struct dentry *error = ERR_PTR(-ENOENT);
2653         struct inode *inode;
2654
2655         inode = proc_pid_make_inode(dir->i_sb, task);
2656         if (!inode)
2657                 goto out;
2658
2659         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2660         inode->i_op = &proc_tgid_base_inode_operations;
2661         inode->i_fop = &proc_tgid_base_operations;
2662         inode->i_flags|=S_IMMUTABLE;
2663
2664         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2665                 ARRAY_SIZE(tgid_base_stuff));
2666
2667         dentry->d_op = &pid_dentry_operations;
2668
2669         d_add(dentry, inode);
2670         /* Close the race of the process dying before we return the dentry */
2671         if (pid_revalidate(dentry, NULL))
2672                 error = NULL;
2673 out:
2674         return error;
2675 }
2676
2677 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2678 {
2679         struct dentry *result = ERR_PTR(-ENOENT);
2680         struct task_struct *task;
2681         unsigned tgid;
2682         struct pid_namespace *ns;
2683
2684         result = proc_base_lookup(dir, dentry);
2685         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2686                 goto out;
2687
2688         tgid = name_to_int(dentry);
2689         if (tgid == ~0U)
2690                 goto out;
2691
2692         ns = dentry->d_sb->s_fs_info;
2693         rcu_read_lock();
2694         task = find_task_by_pid_ns(tgid, ns);
2695         if (task)
2696                 get_task_struct(task);
2697         rcu_read_unlock();
2698         if (!task)
2699                 goto out;
2700
2701         result = proc_pid_instantiate(dir, dentry, task, NULL);
2702         put_task_struct(task);
2703 out:
2704         return result;
2705 }
2706
2707 /*
2708  * Find the first task with tgid >= tgid
2709  *
2710  */
2711 struct tgid_iter {
2712         unsigned int tgid;
2713         struct task_struct *task;
2714 };
2715 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2716 {
2717         struct pid *pid;
2718
2719         if (iter.task)
2720                 put_task_struct(iter.task);
2721         rcu_read_lock();
2722 retry:
2723         iter.task = NULL;
2724         pid = find_ge_pid(iter.tgid, ns);
2725         if (pid) {
2726                 iter.tgid = pid_nr_ns(pid, ns);
2727                 iter.task = pid_task(pid, PIDTYPE_PID);
2728                 /* What we to know is if the pid we have find is the
2729                  * pid of a thread_group_leader.  Testing for task
2730                  * being a thread_group_leader is the obvious thing
2731                  * todo but there is a window when it fails, due to
2732                  * the pid transfer logic in de_thread.
2733                  *
2734                  * So we perform the straight forward test of seeing
2735                  * if the pid we have found is the pid of a thread
2736                  * group leader, and don't worry if the task we have
2737                  * found doesn't happen to be a thread group leader.
2738                  * As we don't care in the case of readdir.
2739                  */
2740                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2741                         iter.tgid += 1;
2742                         goto retry;
2743                 }
2744                 get_task_struct(iter.task);
2745         }
2746         rcu_read_unlock();
2747         return iter;
2748 }
2749
2750 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2751
2752 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2753         struct tgid_iter iter)
2754 {
2755         char name[PROC_NUMBUF];
2756         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2757         return proc_fill_cache(filp, dirent, filldir, name, len,
2758                                 proc_pid_instantiate, iter.task, NULL);
2759 }
2760
2761 /* for the /proc/ directory itself, after non-process stuff has been done */
2762 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2763 {
2764         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2765         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2766         struct tgid_iter iter;
2767         struct pid_namespace *ns;
2768
2769         if (!reaper)
2770                 goto out_no_task;
2771
2772         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2773                 const struct pid_entry *p = &proc_base_stuff[nr];
2774                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2775                         goto out;
2776         }
2777
2778         ns = filp->f_dentry->d_sb->s_fs_info;
2779         iter.task = NULL;
2780         iter.tgid = filp->f_pos - TGID_OFFSET;
2781         for (iter = next_tgid(ns, iter);
2782              iter.task;
2783              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2784                 filp->f_pos = iter.tgid + TGID_OFFSET;
2785                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2786                         put_task_struct(iter.task);
2787                         goto out;
2788                 }
2789         }
2790         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2791 out:
2792         put_task_struct(reaper);
2793 out_no_task:
2794         return 0;
2795 }
2796
2797 /*
2798  * Tasks
2799  */
2800 static const struct pid_entry tid_base_stuff[] = {
2801         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2802         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2803         REG("environ",   S_IRUSR, proc_environ_operations),
2804         INF("auxv",      S_IRUSR, proc_pid_auxv),
2805         ONE("status",    S_IRUGO, proc_pid_status),
2806         ONE("personality", S_IRUSR, proc_pid_personality),
2807         INF("limits",    S_IRUSR, proc_pid_limits),
2808 #ifdef CONFIG_SCHED_DEBUG
2809         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2810 #endif
2811 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2812         INF("syscall",   S_IRUSR, proc_pid_syscall),
2813 #endif
2814         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2815         ONE("stat",      S_IRUGO, proc_tid_stat),
2816         ONE("statm",     S_IRUGO, proc_pid_statm),
2817         REG("maps",      S_IRUGO, proc_maps_operations),
2818 #ifdef CONFIG_NUMA
2819         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2820 #endif
2821         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2822         LNK("cwd",       proc_cwd_link),
2823         LNK("root",      proc_root_link),
2824         LNK("exe",       proc_exe_link),
2825         REG("mounts",    S_IRUGO, proc_mounts_operations),
2826         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2827 #ifdef CONFIG_PROC_PAGE_MONITOR
2828         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2829         REG("smaps",     S_IRUGO, proc_smaps_operations),
2830         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2831 #endif
2832 #ifdef CONFIG_SECURITY
2833         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2834 #endif
2835 #ifdef CONFIG_KALLSYMS
2836         INF("wchan",     S_IRUGO, proc_pid_wchan),
2837 #endif
2838 #ifdef CONFIG_SCHEDSTATS
2839         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2840 #endif
2841 #ifdef CONFIG_LATENCYTOP
2842         REG("latency",  S_IRUGO, proc_lstats_operations),
2843 #endif
2844 #ifdef CONFIG_PROC_PID_CPUSET
2845         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2846 #endif
2847 #ifdef CONFIG_CGROUPS
2848         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2849 #endif
2850         INF("oom_score", S_IRUGO, proc_oom_score),
2851         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2852 #ifdef CONFIG_AUDITSYSCALL
2853         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2854         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2855 #endif
2856 #ifdef CONFIG_FAULT_INJECTION
2857         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2858 #endif
2859 #ifdef CONFIG_TASK_IO_ACCOUNTING
2860         INF("io",       S_IRUGO, proc_tid_io_accounting),
2861 #endif
2862 };
2863
2864 static int proc_tid_base_readdir(struct file * filp,
2865                              void * dirent, filldir_t filldir)
2866 {
2867         return proc_pident_readdir(filp,dirent,filldir,
2868                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2869 }
2870
2871 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2872         return proc_pident_lookup(dir, dentry,
2873                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2874 }
2875
2876 static const struct file_operations proc_tid_base_operations = {
2877         .read           = generic_read_dir,
2878         .readdir        = proc_tid_base_readdir,
2879 };
2880
2881 static const struct inode_operations proc_tid_base_inode_operations = {
2882         .lookup         = proc_tid_base_lookup,
2883         .getattr        = pid_getattr,
2884         .setattr        = proc_setattr,
2885 };
2886
2887 static struct dentry *proc_task_instantiate(struct inode *dir,
2888         struct dentry *dentry, struct task_struct *task, const void *ptr)
2889 {
2890         struct dentry *error = ERR_PTR(-ENOENT);
2891         struct inode *inode;
2892         inode = proc_pid_make_inode(dir->i_sb, task);
2893
2894         if (!inode)
2895                 goto out;
2896         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2897         inode->i_op = &proc_tid_base_inode_operations;
2898         inode->i_fop = &proc_tid_base_operations;
2899         inode->i_flags|=S_IMMUTABLE;
2900
2901         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2902                 ARRAY_SIZE(tid_base_stuff));
2903
2904         dentry->d_op = &pid_dentry_operations;
2905
2906         d_add(dentry, inode);
2907         /* Close the race of the process dying before we return the dentry */
2908         if (pid_revalidate(dentry, NULL))
2909                 error = NULL;
2910 out:
2911         return error;
2912 }
2913
2914 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2915 {
2916         struct dentry *result = ERR_PTR(-ENOENT);
2917         struct task_struct *task;
2918         struct task_struct *leader = get_proc_task(dir);
2919         unsigned tid;
2920         struct pid_namespace *ns;
2921
2922         if (!leader)
2923                 goto out_no_task;
2924
2925         tid = name_to_int(dentry);
2926         if (tid == ~0U)
2927                 goto out;
2928
2929         ns = dentry->d_sb->s_fs_info;
2930         rcu_read_lock();
2931         task = find_task_by_pid_ns(tid, ns);
2932         if (task)
2933                 get_task_struct(task);
2934         rcu_read_unlock();
2935         if (!task)
2936                 goto out;
2937         if (!same_thread_group(leader, task))
2938                 goto out_drop_task;
2939
2940         result = proc_task_instantiate(dir, dentry, task, NULL);
2941 out_drop_task:
2942         put_task_struct(task);
2943 out:
2944         put_task_struct(leader);
2945 out_no_task:
2946         return result;
2947 }
2948
2949 /*
2950  * Find the first tid of a thread group to return to user space.
2951  *
2952  * Usually this is just the thread group leader, but if the users
2953  * buffer was too small or there was a seek into the middle of the
2954  * directory we have more work todo.
2955  *
2956  * In the case of a short read we start with find_task_by_pid.
2957  *
2958  * In the case of a seek we start with the leader and walk nr
2959  * threads past it.
2960  */
2961 static struct task_struct *first_tid(struct task_struct *leader,
2962                 int tid, int nr, struct pid_namespace *ns)
2963 {
2964         struct task_struct *pos;
2965
2966         rcu_read_lock();
2967         /* Attempt to start with the pid of a thread */
2968         if (tid && (nr > 0)) {
2969                 pos = find_task_by_pid_ns(tid, ns);
2970                 if (pos && (pos->group_leader == leader))
2971                         goto found;
2972         }
2973
2974         /* If nr exceeds the number of threads there is nothing todo */
2975         pos = NULL;
2976         if (nr && nr >= get_nr_threads(leader))
2977                 goto out;
2978
2979         /* If we haven't found our starting place yet start
2980          * with the leader and walk nr threads forward.
2981          */
2982         for (pos = leader; nr > 0; --nr) {
2983                 pos = next_thread(pos);
2984                 if (pos == leader) {
2985                         pos = NULL;
2986                         goto out;
2987                 }
2988         }
2989 found:
2990         get_task_struct(pos);
2991 out:
2992         rcu_read_unlock();
2993         return pos;
2994 }
2995
2996 /*
2997  * Find the next thread in the thread list.
2998  * Return NULL if there is an error or no next thread.
2999  *
3000  * The reference to the input task_struct is released.
3001  */
3002 static struct task_struct *next_tid(struct task_struct *start)
3003 {
3004         struct task_struct *pos = NULL;
3005         rcu_read_lock();
3006         if (pid_alive(start)) {
3007                 pos = next_thread(start);
3008                 if (thread_group_leader(pos))
3009                         pos = NULL;
3010                 else
3011                         get_task_struct(pos);
3012         }
3013         rcu_read_unlock();
3014         put_task_struct(start);
3015         return pos;
3016 }
3017
3018 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3019         struct task_struct *task, int tid)
3020 {
3021         char name[PROC_NUMBUF];
3022         int len = snprintf(name, sizeof(name), "%d", tid);
3023         return proc_fill_cache(filp, dirent, filldir, name, len,
3024                                 proc_task_instantiate, task, NULL);
3025 }
3026
3027 /* for the /proc/TGID/task/ directories */
3028 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3029 {
3030         struct dentry *dentry = filp->f_path.dentry;
3031         struct inode *inode = dentry->d_inode;
3032         struct task_struct *leader = NULL;
3033         struct task_struct *task;
3034         int retval = -ENOENT;
3035         ino_t ino;
3036         int tid;
3037         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
3038         struct pid_namespace *ns;
3039
3040         task = get_proc_task(inode);
3041         if (!task)
3042                 goto out_no_task;
3043         rcu_read_lock();
3044         if (pid_alive(task)) {
3045                 leader = task->group_leader;
3046                 get_task_struct(leader);
3047         }
3048         rcu_read_unlock();
3049         put_task_struct(task);
3050         if (!leader)
3051                 goto out_no_task;
3052         retval = 0;
3053
3054         switch (pos) {
3055         case 0:
3056                 ino = inode->i_ino;
3057                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
3058                         goto out;
3059                 pos++;
3060                 /* fall through */
3061         case 1:
3062                 ino = parent_ino(dentry);
3063                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
3064                         goto out;
3065                 pos++;
3066                 /* fall through */
3067         }
3068
3069         /* f_version caches the tgid value that the last readdir call couldn't
3070          * return. lseek aka telldir automagically resets f_version to 0.
3071          */
3072         ns = filp->f_dentry->d_sb->s_fs_info;
3073         tid = (int)filp->f_version;
3074         filp->f_version = 0;
3075         for (task = first_tid(leader, tid, pos - 2, ns);
3076              task;
3077              task = next_tid(task), pos++) {
3078                 tid = task_pid_nr_ns(task, ns);
3079                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3080                         /* returning this tgid failed, save it as the first
3081                          * pid for the next readir call */
3082                         filp->f_version = (u64)tid;
3083                         put_task_struct(task);
3084                         break;
3085                 }
3086         }
3087 out:
3088         filp->f_pos = pos;
3089         put_task_struct(leader);
3090 out_no_task:
3091         return retval;
3092 }
3093
3094 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3095 {
3096         struct inode *inode = dentry->d_inode;
3097         struct task_struct *p = get_proc_task(inode);
3098         generic_fillattr(inode, stat);
3099
3100         if (p) {
3101                 stat->nlink += get_nr_threads(p);
3102                 put_task_struct(p);
3103         }
3104
3105         return 0;
3106 }
3107
3108 static const struct inode_operations proc_task_inode_operations = {
3109         .lookup         = proc_task_lookup,
3110         .getattr        = proc_task_getattr,
3111         .setattr        = proc_setattr,
3112 };
3113
3114 static const struct file_operations proc_task_operations = {
3115         .read           = generic_read_dir,
3116         .readdir        = proc_task_readdir,
3117 };