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/stacktrace.h>
  69 #include <linux/resource.h>
  70 #include <linux/module.h>
  71 #include <linux/mount.h>
  72 #include <linux/security.h>
  73 #include <linux/ptrace.h>
  74 #include <linux/tracehook.h>
  75 #include <linux/cgroup.h>
  76 #include <linux/cpuset.h>
  77 #include <linux/audit.h>
  78 #include <linux/poll.h>
  79 #include <linux/nsproxy.h>
  80 #include <linux/oom.h>
  81 #include <linux/elf.h>
  82 #include <linux/pid_namespace.h>
  83 #include <linux/fs_struct.h>
  84 #include "internal.h"
  85
  86 /* NOTE:
  87  *      Implementing inode permission operations in /proc is almost
  88  *      certainly an error.  Permission checks need to happen during
  89  *      each system call not at open time.  The reason is that most of
  90  *      what we wish to check for permissions in /proc varies at runtime.
  91  *
  92  *      The classic example of a problem is opening file descriptors
  93  *      in /proc for a task before it execs a suid executable.
  94  */
  95
  96 struct pid_entry {
  97         char *name;
  98         int len;
  99         mode_t mode;
 100         const struct inode_operations *iop;
 101         const struct file_operations *fop;
 102         union proc_op op;
 103 };
 104
 105 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 106         .name = (NAME),                                 \
 107         .len  = sizeof(NAME) - 1,                       \
 108         .mode = MODE,                                   \
 109         .iop  = IOP,                                    \
 110         .fop  = FOP,                                    \
 111         .op   = OP,                                     \
 112 }
 113
 114 #define DIR(NAME, MODE, iops, fops)     \
 115         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 116 #define LNK(NAME, get_link)                                     \
 117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 118                 &proc_pid_link_inode_operations, NULL,          \
 119                 { .proc_get_link = get_link } )
 120 #define REG(NAME, MODE, fops)                           \
 121         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 122 #define INF(NAME, MODE, read)                           \
 123         NOD(NAME, (S_IFREG|(MODE)),                     \
 124                 NULL, &proc_info_file_operations,       \
 125                 { .proc_read = read } )
 126 #define ONE(NAME, MODE, show)                           \
 127         NOD(NAME, (S_IFREG|(MODE)),                     \
 128                 NULL, &proc_single_file_operations,     \
 129                 { .proc_show = show } )
 130
 131 /*
 132  * Count the number of hardlinks for the pid_entry table, excluding the .
 133  * and .. links.
 134  */
 135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 136         unsigned int n)
 137 {
 138         unsigned int i;
 139         unsigned int count;
 140
 141         count = 0;
 142         for (i = 0; i < n; ++i) {
 143                 if (S_ISDIR(entries[i].mode))
 144                         ++count;
 145         }
 146
 147         return count;
 148 }
 149
 150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
 151 {
 152         struct fs_struct *fs;
 153         int result = -ENOENT;
 154
 155         task_lock(task);
 156         fs = task->fs;
 157         if (fs) {
 158                 read_lock(&fs->lock);
 159                 *path = root ? fs->root : fs->pwd;
 160                 path_get(path);
 161                 read_unlock(&fs->lock);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int get_nr_threads(struct task_struct *tsk)
 169 {
 170         unsigned long flags;
 171         int count = 0;
 172
 173         if (lock_task_sighand(tsk, &flags)) {
 174                 count = atomic_read(&tsk->signal->count);
 175                 unlock_task_sighand(tsk, &flags);
 176         }
 177         return count;
 178 }
 179
 180 static int proc_cwd_link(struct inode *inode, struct path *path)
 181 {
 182         struct task_struct *task = get_proc_task(inode);
 183         int result = -ENOENT;
 184
 185         if (task) {
 186                 result = get_fs_path(task, path, 0);
 187                 put_task_struct(task);
 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         int result = -ENOENT;
 196
 197         if (task) {
 198                 result = get_fs_path(task, path, 1);
 199                 put_task_struct(task);
 200         }
 201         return result;
 202 }
 203
 204 /*
 205  * Return zero if current may access user memory in @task, -error if not.
 206  */
 207 static int check_mem_permission(struct task_struct *task)
 208 {
 209         /*
 210          * A task can always look at itself, in case it chooses
 211          * to use system calls instead of load instructions.
 212          */
 213         if (task == current)
 214                 return 0;
 215
 216         /*
 217          * If current is actively ptrace'ing, and would also be
 218          * permitted to freshly attach with ptrace now, permit it.
 219          */
 220         if (task_is_stopped_or_traced(task)) {
 221                 int match;
 222                 rcu_read_lock();
 223                 match = (tracehook_tracer_task(task) == current);
 224                 rcu_read_unlock();
 225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 226                         return 0;
 227         }
 228
 229         /*
 230          * Noone else is allowed.
 231          */
 232         return -EPERM;
 233 }
 234
 235 struct mm_struct *mm_for_maps(struct task_struct *task)
 236 {
 237         struct mm_struct *mm = get_task_mm(task);
 238         if (!mm)
 239                 return NULL;
 240         if (mm != current->mm) {
 241                 /*
 242                  * task->mm can be changed before security check,
 243                  * in that case we must notice the change after.
 244                  */
 245                 if (!ptrace_may_access(task, PTRACE_MODE_READ) ||
 246                     mm != task->mm) {
 247                         mmput(mm);
 248                         return NULL;
 249                 }
 250         }
 251         down_read(&mm->mmap_sem);
 252         return mm;
 253 }
 254
 255 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 256 {
 257         int res = 0;
 258         unsigned int len;
 259         struct mm_struct *mm = get_task_mm(task);
 260         if (!mm)
 261                 goto out;
 262         if (!mm->arg_end)
 263                 goto out_mm;    /* Shh! No looking before we're done */
 264
 265         len = mm->arg_end - mm->arg_start;
 266
 267         if (len > PAGE_SIZE)
 268                 len = PAGE_SIZE;
 269
 270         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 271
 272         // If the nul at the end of args has been overwritten, then
 273         // assume application is using setproctitle(3).
 274         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 275                 len = strnlen(buffer, res);
 276                 if (len < res) {
 277                     res = len;
 278                 } else {
 279                         len = mm->env_end - mm->env_start;
 280                         if (len > PAGE_SIZE - res)
 281                                 len = PAGE_SIZE - res;
 282                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 283                         res = strnlen(buffer, res);
 284                 }
 285         }
 286 out_mm:
 287         mmput(mm);
 288 out:
 289         return res;
 290 }
 291
 292 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 293 {
 294         int res = 0;
 295         struct mm_struct *mm = get_task_mm(task);
 296         if (mm) {
 297                 unsigned int nwords = 0;
 298                 do {
 299                         nwords += 2;
 300                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 301                 res = nwords * sizeof(mm->saved_auxv[0]);
 302                 if (res > PAGE_SIZE)
 303                         res = PAGE_SIZE;
 304                 memcpy(buffer, mm->saved_auxv, res);
 305                 mmput(mm);
 306         }
 307         return res;
 308 }
 309
 310
 311 #ifdef CONFIG_KALLSYMS
 312 /*
 313  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 314  * Returns the resolved symbol.  If that fails, simply return the address.
 315  */
 316 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 317 {
 318         unsigned long wchan;
 319         char symname[KSYM_NAME_LEN];
 320
 321         wchan = get_wchan(task);
 322
 323         if (lookup_symbol_name(wchan, symname) < 0)
 324                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 325                         return 0;
 326                 else
 327                         return sprintf(buffer, "%lu", wchan);
 328         else
 329                 return sprintf(buffer, "%s", symname);
 330 }
 331 #endif /* CONFIG_KALLSYMS */
 332
 333 #ifdef CONFIG_STACKTRACE
 334
 335 #define MAX_STACK_TRACE_DEPTH   64
 336
 337 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 338                           struct pid *pid, struct task_struct *task)
 339 {
 340         struct stack_trace trace;
 341         unsigned long *entries;
 342         int i;
 343
 344         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 345         if (!entries)
 346                 return -ENOMEM;
 347
 348         trace.nr_entries        = 0;
 349         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 350         trace.entries           = entries;
 351         trace.skip              = 0;
 352         save_stack_trace_tsk(task, &trace);
 353
 354         for (i = 0; i < trace.nr_entries; i++) {
 355                 seq_printf(m, "[<%p>] %pS\n",
 356                            (void *)entries[i], (void *)entries[i]);
 357         }
 358         kfree(entries);
 359
 360         return 0;
 361 }
 362 #endif
 363
 364 #ifdef CONFIG_SCHEDSTATS
 365 /*
 366  * Provides /proc/PID/schedstat
 367  */
 368 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 369 {
 370         return sprintf(buffer, "%llu %llu %lu\n",
 371                         (unsigned long long)task->se.sum_exec_runtime,
 372                         (unsigned long long)task->sched_info.run_delay,
 373                         task->sched_info.pcount);
 374 }
 375 #endif
 376
 377 #ifdef CONFIG_LATENCYTOP
 378 static int lstats_show_proc(struct seq_file *m, void *v)
 379 {
 380         int i;
 381         struct inode *inode = m->private;
 382         struct task_struct *task = get_proc_task(inode);
 383
 384         if (!task)
 385                 return -ESRCH;
 386         seq_puts(m, "Latency Top version : v0.1\n");
 387         for (i = 0; i < 32; i++) {
 388                 if (task->latency_record[i].backtrace[0]) {
 389                         int q;
 390                         seq_printf(m, "%i %li %li ",
 391                                 task->latency_record[i].count,
 392                                 task->latency_record[i].time,
 393                                 task->latency_record[i].max);
 394                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 395                                 char sym[KSYM_SYMBOL_LEN];
 396                                 char *c;
 397                                 if (!task->latency_record[i].backtrace[q])
 398                                         break;
 399                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 400                                         break;
 401                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 402                                 c = strchr(sym, '+');
 403                                 if (c)
 404                                         *c = 0;
 405                                 seq_printf(m, "%s ", sym);
 406                         }
 407                         seq_printf(m, "\n");
 408                 }
 409
 410         }
 411         put_task_struct(task);
 412         return 0;
 413 }
 414
 415 static int lstats_open(struct inode *inode, struct file *file)
 416 {
 417         return single_open(file, lstats_show_proc, inode);
 418 }
 419
 420 static ssize_t lstats_write(struct file *file, const char __user *buf,
 421                             size_t count, loff_t *offs)
 422 {
 423         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 424
 425         if (!task)
 426                 return -ESRCH;
 427         clear_all_latency_tracing(task);
 428         put_task_struct(task);
 429
 430         return count;
 431 }
 432
 433 static const struct file_operations proc_lstats_operations = {
 434         .open           = lstats_open,
 435         .read           = seq_read,
 436         .write          = lstats_write,
 437         .llseek         = seq_lseek,
 438         .release        = single_release,
 439 };
 440
 441 #endif
 442
 443 /* The badness from the OOM killer */
 444 unsigned long badness(struct task_struct *p, unsigned long uptime);
 445 static int proc_oom_score(struct task_struct *task, char *buffer)
 446 {
 447         unsigned long points;
 448         struct timespec uptime;
 449
 450         do_posix_clock_monotonic_gettime(&uptime);
 451         read_lock(&tasklist_lock);
 452         points = badness(task, uptime.tv_sec);
 453         read_unlock(&tasklist_lock);
 454         return sprintf(buffer, "%lu\n", points);
 455 }
 456
 457 struct limit_names {
 458         char *name;
 459         char *unit;
 460 };
 461
 462 static const struct limit_names lnames[RLIM_NLIMITS] = {
 463         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 464         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 465         [RLIMIT_DATA] = {"Max data size", "bytes"},
 466         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 467         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 468         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 469         [RLIMIT_NPROC] = {"Max processes", "processes"},
 470         [RLIMIT_NOFILE] = {"Max open files", "files"},
 471         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 472         [RLIMIT_AS] = {"Max address space", "bytes"},
 473         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 474         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 475         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 476         [RLIMIT_NICE] = {"Max nice priority", NULL},
 477         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 478         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 479 };
 480
 481 /* Display limits for a process */
 482 static int proc_pid_limits(struct task_struct *task, char *buffer)
 483 {
 484         unsigned int i;
 485         int count = 0;
 486         unsigned long flags;
 487         char *bufptr = buffer;
 488
 489         struct rlimit rlim[RLIM_NLIMITS];
 490
 491         if (!lock_task_sighand(task, &flags))
 492                 return 0;
 493         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 494         unlock_task_sighand(task, &flags);
 495
 496         /*
 497          * print the file header
 498          */
 499         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 500                         "Limit", "Soft Limit", "Hard Limit", "Units");
 501
 502         for (i = 0; i < RLIM_NLIMITS; i++) {
 503                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 504                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 505                                          lnames[i].name, "unlimited");
 506                 else
 507                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 508                                          lnames[i].name, rlim[i].rlim_cur);
 509
 510                 if (rlim[i].rlim_max == RLIM_INFINITY)
 511                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 512                 else
 513                         count += sprintf(&bufptr[count], "%-20lu ",
 514                                          rlim[i].rlim_max);
 515
 516                 if (lnames[i].unit)
 517                         count += sprintf(&bufptr[count], "%-10s\n",
 518                                          lnames[i].unit);
 519                 else
 520                         count += sprintf(&bufptr[count], "\n");
 521         }
 522
 523         return count;
 524 }
 525
 526 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 527 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 528 {
 529         long nr;
 530         unsigned long args[6], sp, pc;
 531
 532         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 533                 return sprintf(buffer, "running\n");
 534
 535         if (nr < 0)
 536                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 537
 538         return sprintf(buffer,
 539                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 540                        nr,
 541                        args[0], args[1], args[2], args[3], args[4], args[5],
 542                        sp, pc);
 543 }
 544 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 545
 546 /************************************************************************/
 547 /*                       Here the fs part begins                        */
 548 /************************************************************************/
 549
 550 /* permission checks */
 551 static int proc_fd_access_allowed(struct inode *inode)
 552 {
 553         struct task_struct *task;
 554         int allowed = 0;
 555         /* Allow access to a task's file descriptors if it is us or we
 556          * may use ptrace attach to the process and find out that
 557          * information.
 558          */
 559         task = get_proc_task(inode);
 560         if (task) {
 561                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 562                 put_task_struct(task);
 563         }
 564         return allowed;
 565 }
 566
 567 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 568 {
 569         int error;
 570         struct inode *inode = dentry->d_inode;
 571
 572         if (attr->ia_valid & ATTR_MODE)
 573                 return -EPERM;
 574
 575         error = inode_change_ok(inode, attr);
 576         if (!error)
 577                 error = inode_setattr(inode, attr);
 578         return error;
 579 }
 580
 581 static const struct inode_operations proc_def_inode_operations = {
 582         .setattr        = proc_setattr,
 583 };
 584
 585 static int mounts_open_common(struct inode *inode, struct file *file,
 586                               const struct seq_operations *op)
 587 {
 588         struct task_struct *task = get_proc_task(inode);
 589         struct nsproxy *nsp;
 590         struct mnt_namespace *ns = NULL;
 591         struct path root;
 592         struct proc_mounts *p;
 593         int ret = -EINVAL;
 594
 595         if (task) {
 596                 rcu_read_lock();
 597                 nsp = task_nsproxy(task);
 598                 if (nsp) {
 599                         ns = nsp->mnt_ns;
 600                         if (ns)
 601                                 get_mnt_ns(ns);
 602                 }
 603                 rcu_read_unlock();
 604                 if (ns && get_fs_path(task, &root, 1) == 0)
 605                         ret = 0;
 606                 put_task_struct(task);
 607         }
 608
 609         if (!ns)
 610                 goto err;
 611         if (ret)
 612                 goto err_put_ns;
 613
 614         ret = -ENOMEM;
 615         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 616         if (!p)
 617                 goto err_put_path;
 618
 619         file->private_data = &p->m;
 620         ret = seq_open(file, op);
 621         if (ret)
 622                 goto err_free;
 623
 624         p->m.private = p;
 625         p->ns = ns;
 626         p->root = root;
 627         p->event = ns->event;
 628
 629         return 0;
 630
 631  err_free:
 632         kfree(p);
 633  err_put_path:
 634         path_put(&root);
 635  err_put_ns:
 636         put_mnt_ns(ns);
 637  err:
 638         return ret;
 639 }
 640
 641 static int mounts_release(struct inode *inode, struct file *file)
 642 {
 643         struct proc_mounts *p = file->private_data;
 644         path_put(&p->root);
 645         put_mnt_ns(p->ns);
 646         return seq_release(inode, file);
 647 }
 648
 649 static unsigned mounts_poll(struct file *file, poll_table *wait)
 650 {
 651         struct proc_mounts *p = file->private_data;
 652         struct mnt_namespace *ns = p->ns;
 653         unsigned res = POLLIN | POLLRDNORM;
 654
 655         poll_wait(file, &ns->poll, wait);
 656
 657         spin_lock(&vfsmount_lock);
 658         if (p->event != ns->event) {
 659                 p->event = ns->event;
 660                 res |= POLLERR | POLLPRI;
 661         }
 662         spin_unlock(&vfsmount_lock);
 663
 664         return res;
 665 }
 666
 667 static int mounts_open(struct inode *inode, struct file *file)
 668 {
 669         return mounts_open_common(inode, file, &mounts_op);
 670 }
 671
 672 static const struct file_operations proc_mounts_operations = {
 673         .open           = mounts_open,
 674         .read           = seq_read,
 675         .llseek         = seq_lseek,
 676         .release        = mounts_release,
 677         .poll           = mounts_poll,
 678 };
 679
 680 static int mountinfo_open(struct inode *inode, struct file *file)
 681 {
 682         return mounts_open_common(inode, file, &mountinfo_op);
 683 }
 684
 685 static const struct file_operations proc_mountinfo_operations = {
 686         .open           = mountinfo_open,
 687         .read           = seq_read,
 688         .llseek         = seq_lseek,
 689         .release        = mounts_release,
 690         .poll           = mounts_poll,
 691 };
 692
 693 static int mountstats_open(struct inode *inode, struct file *file)
 694 {
 695         return mounts_open_common(inode, file, &mountstats_op);
 696 }
 697
 698 static const struct file_operations proc_mountstats_operations = {
 699         .open           = mountstats_open,
 700         .read           = seq_read,
 701         .llseek         = seq_lseek,
 702         .release        = mounts_release,
 703 };
 704
 705 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 706
 707 static ssize_t proc_info_read(struct file * file, char __user * buf,
 708                           size_t count, loff_t *ppos)
 709 {
 710         struct inode * inode = file->f_path.dentry->d_inode;
 711         unsigned long page;
 712         ssize_t length;
 713         struct task_struct *task = get_proc_task(inode);
 714
 715         length = -ESRCH;
 716         if (!task)
 717                 goto out_no_task;
 718
 719         if (count > PROC_BLOCK_SIZE)
 720                 count = PROC_BLOCK_SIZE;
 721
 722         length = -ENOMEM;
 723         if (!(page = __get_free_page(GFP_TEMPORARY)))
 724                 goto out;
 725
 726         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 727
 728         if (length >= 0)
 729                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 730         free_page(page);
 731 out:
 732         put_task_struct(task);
 733 out_no_task:
 734         return length;
 735 }
 736
 737 static const struct file_operations proc_info_file_operations = {
 738         .read           = proc_info_read,
 739 };
 740
 741 static int proc_single_show(struct seq_file *m, void *v)
 742 {
 743         struct inode *inode = m->private;
 744         struct pid_namespace *ns;
 745         struct pid *pid;
 746         struct task_struct *task;
 747         int ret;
 748
 749         ns = inode->i_sb->s_fs_info;
 750         pid = proc_pid(inode);
 751         task = get_pid_task(pid, PIDTYPE_PID);
 752         if (!task)
 753                 return -ESRCH;
 754
 755         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 756
 757         put_task_struct(task);
 758         return ret;
 759 }
 760
 761 static int proc_single_open(struct inode *inode, struct file *filp)
 762 {
 763         int ret;
 764         ret = single_open(filp, proc_single_show, NULL);
 765         if (!ret) {
 766                 struct seq_file *m = filp->private_data;
 767
 768                 m->private = inode;
 769         }
 770         return ret;
 771 }
 772
 773 static const struct file_operations proc_single_file_operations = {
 774         .open           = proc_single_open,
 775         .read           = seq_read,
 776         .llseek         = seq_lseek,
 777         .release        = single_release,
 778 };
 779
 780 static int mem_open(struct inode* inode, struct file* file)
 781 {
 782         file->private_data = (void*)((long)current->self_exec_id);
 783         return 0;
 784 }
 785
 786 static ssize_t mem_read(struct file * file, char __user * buf,
 787                         size_t count, loff_t *ppos)
 788 {
 789         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 790         char *page;
 791         unsigned long src = *ppos;
 792         int ret = -ESRCH;
 793         struct mm_struct *mm;
 794
 795         if (!task)
 796                 goto out_no_task;
 797
 798         if (check_mem_permission(task))
 799                 goto out;
 800
 801         ret = -ENOMEM;
 802         page = (char *)__get_free_page(GFP_TEMPORARY);
 803         if (!page)
 804                 goto out;
 805
 806         ret = 0;
 807
 808         mm = get_task_mm(task);
 809         if (!mm)
 810                 goto out_free;
 811
 812         ret = -EIO;
 813
 814         if (file->private_data != (void*)((long)current->self_exec_id))
 815                 goto out_put;
 816
 817         ret = 0;
 818
 819         while (count > 0) {
 820                 int this_len, retval;
 821
 822                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 823                 retval = access_process_vm(task, src, page, this_len, 0);
 824                 if (!retval || check_mem_permission(task)) {
 825                         if (!ret)
 826                                 ret = -EIO;
 827                         break;
 828                 }
 829
 830                 if (copy_to_user(buf, page, retval)) {
 831                         ret = -EFAULT;
 832                         break;
 833                 }
 834
 835                 ret += retval;
 836                 src += retval;
 837                 buf += retval;
 838                 count -= retval;
 839         }
 840         *ppos = src;
 841
 842 out_put:
 843         mmput(mm);
 844 out_free:
 845         free_page((unsigned long) page);
 846 out:
 847         put_task_struct(task);
 848 out_no_task:
 849         return ret;
 850 }
 851
 852 #define mem_write NULL
 853
 854 #ifndef mem_write
 855 /* This is a security hazard */
 856 static ssize_t mem_write(struct file * file, const char __user *buf,
 857                          size_t count, loff_t *ppos)
 858 {
 859         int copied;
 860         char *page;
 861         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 862         unsigned long dst = *ppos;
 863
 864         copied = -ESRCH;
 865         if (!task)
 866                 goto out_no_task;
 867
 868         if (check_mem_permission(task))
 869                 goto out;
 870
 871         copied = -ENOMEM;
 872         page = (char *)__get_free_page(GFP_TEMPORARY);
 873         if (!page)
 874                 goto out;
 875
 876         copied = 0;
 877         while (count > 0) {
 878                 int this_len, retval;
 879
 880                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 881                 if (copy_from_user(page, buf, this_len)) {
 882                         copied = -EFAULT;
 883                         break;
 884                 }
 885                 retval = access_process_vm(task, dst, page, this_len, 1);
 886                 if (!retval) {
 887                         if (!copied)
 888                                 copied = -EIO;
 889                         break;
 890                 }
 891                 copied += retval;
 892                 buf += retval;
 893                 dst += retval;
 894                 count -= retval;
 895         }
 896         *ppos = dst;
 897         free_page((unsigned long) page);
 898 out:
 899         put_task_struct(task);
 900 out_no_task:
 901         return copied;
 902 }
 903 #endif
 904
 905 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 906 {
 907         switch (orig) {
 908         case 0:
 909                 file->f_pos = offset;
 910                 break;
 911         case 1:
 912                 file->f_pos += offset;
 913                 break;
 914         default:
 915                 return -EINVAL;
 916         }
 917         force_successful_syscall_return();
 918         return file->f_pos;
 919 }
 920
 921 static const struct file_operations proc_mem_operations = {
 922         .llseek         = mem_lseek,
 923         .read           = mem_read,
 924         .write          = mem_write,
 925         .open           = mem_open,
 926 };
 927
 928 static ssize_t environ_read(struct file *file, char __user *buf,
 929                         size_t count, loff_t *ppos)
 930 {
 931         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 932         char *page;
 933         unsigned long src = *ppos;
 934         int ret = -ESRCH;
 935         struct mm_struct *mm;
 936
 937         if (!task)
 938                 goto out_no_task;
 939
 940         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 941                 goto out;
 942
 943         ret = -ENOMEM;
 944         page = (char *)__get_free_page(GFP_TEMPORARY);
 945         if (!page)
 946                 goto out;
 947
 948         ret = 0;
 949
 950         mm = get_task_mm(task);
 951         if (!mm)
 952                 goto out_free;
 953
 954         while (count > 0) {
 955                 int this_len, retval, max_len;
 956
 957                 this_len = mm->env_end - (mm->env_start + src);
 958
 959                 if (this_len <= 0)
 960                         break;
 961
 962                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 963                 this_len = (this_len > max_len) ? max_len : this_len;
 964
 965                 retval = access_process_vm(task, (mm->env_start + src),
 966                         page, this_len, 0);
 967
 968                 if (retval <= 0) {
 969                         ret = retval;
 970                         break;
 971                 }
 972
 973                 if (copy_to_user(buf, page, retval)) {
 974                         ret = -EFAULT;
 975                         break;
 976                 }
 977
 978                 ret += retval;
 979                 src += retval;
 980                 buf += retval;
 981                 count -= retval;
 982         }
 983         *ppos = src;
 984
 985         mmput(mm);
 986 out_free:
 987         free_page((unsigned long) page);
 988 out:
 989         put_task_struct(task);
 990 out_no_task:
 991         return ret;
 992 }
 993
 994 static const struct file_operations proc_environ_operations = {
 995         .read           = environ_read,
 996 };
 997
 998 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 999                                 size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1002         char buffer[PROC_NUMBUF];
1003         size_t len;
1004         int oom_adjust;
1005
1006         if (!task)
1007                 return -ESRCH;
1008         task_lock(task);
1009         if (task->mm)
1010                 oom_adjust = task->mm->oom_adj;
1011         else
1012                 oom_adjust = OOM_DISABLE;
1013         task_unlock(task);
1014         put_task_struct(task);
1015
1016         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1017
1018         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1019 }
1020
1021 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1022                                 size_t count, loff_t *ppos)
1023 {
1024         struct task_struct *task;
1025         char buffer[PROC_NUMBUF], *end;
1026         int oom_adjust;
1027
1028         memset(buffer, 0, sizeof(buffer));
1029         if (count > sizeof(buffer) - 1)
1030                 count = sizeof(buffer) - 1;
1031         if (copy_from_user(buffer, buf, count))
1032                 return -EFAULT;
1033         oom_adjust = simple_strtol(buffer, &end, 0);
1034         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1035              oom_adjust != OOM_DISABLE)
1036                 return -EINVAL;
1037         if (*end == '\n')
1038                 end++;
1039         task = get_proc_task(file->f_path.dentry->d_inode);
1040         if (!task)
1041                 return -ESRCH;
1042         task_lock(task);
1043         if (!task->mm) {
1044                 task_unlock(task);
1045                 put_task_struct(task);
1046                 return -EINVAL;
1047         }
1048         if (oom_adjust < task->mm->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1049                 task_unlock(task);
1050                 put_task_struct(task);
1051                 return -EACCES;
1052         }
1053         task->mm->oom_adj = oom_adjust;
1054         task_unlock(task);
1055         put_task_struct(task);
1056         if (end - buffer == 0)
1057                 return -EIO;
1058         return end - buffer;
1059 }
1060
1061 static const struct file_operations proc_oom_adjust_operations = {
1062         .read           = oom_adjust_read,
1063         .write          = oom_adjust_write,
1064 };
1065
1066 #ifdef CONFIG_AUDITSYSCALL
1067 #define TMPBUFLEN 21
1068 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1069                                   size_t count, loff_t *ppos)
1070 {
1071         struct inode * inode = file->f_path.dentry->d_inode;
1072         struct task_struct *task = get_proc_task(inode);
1073         ssize_t length;
1074         char tmpbuf[TMPBUFLEN];
1075
1076         if (!task)
1077                 return -ESRCH;
1078         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1079                                 audit_get_loginuid(task));
1080         put_task_struct(task);
1081         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1082 }
1083
1084 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1085                                    size_t count, loff_t *ppos)
1086 {
1087         struct inode * inode = file->f_path.dentry->d_inode;
1088         char *page, *tmp;
1089         ssize_t length;
1090         uid_t loginuid;
1091
1092         if (!capable(CAP_AUDIT_CONTROL))
1093                 return -EPERM;
1094
1095         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1096                 return -EPERM;
1097
1098         if (count >= PAGE_SIZE)
1099                 count = PAGE_SIZE - 1;
1100
1101         if (*ppos != 0) {
1102                 /* No partial writes. */
1103                 return -EINVAL;
1104         }
1105         page = (char*)__get_free_page(GFP_TEMPORARY);
1106         if (!page)
1107                 return -ENOMEM;
1108         length = -EFAULT;
1109         if (copy_from_user(page, buf, count))
1110                 goto out_free_page;
1111
1112         page[count] = '\0';
1113         loginuid = simple_strtoul(page, &tmp, 10);
1114         if (tmp == page) {
1115                 length = -EINVAL;
1116                 goto out_free_page;
1117
1118         }
1119         length = audit_set_loginuid(current, loginuid);
1120         if (likely(length == 0))
1121                 length = count;
1122
1123 out_free_page:
1124         free_page((unsigned long) page);
1125         return length;
1126 }
1127
1128 static const struct file_operations proc_loginuid_operations = {
1129         .read           = proc_loginuid_read,
1130         .write          = proc_loginuid_write,
1131 };
1132
1133 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1134                                   size_t count, loff_t *ppos)
1135 {
1136         struct inode * inode = file->f_path.dentry->d_inode;
1137         struct task_struct *task = get_proc_task(inode);
1138         ssize_t length;
1139         char tmpbuf[TMPBUFLEN];
1140
1141         if (!task)
1142                 return -ESRCH;
1143         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1144                                 audit_get_sessionid(task));
1145         put_task_struct(task);
1146         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1147 }
1148
1149 static const struct file_operations proc_sessionid_operations = {
1150         .read           = proc_sessionid_read,
1151 };
1152 #endif
1153
1154 #ifdef CONFIG_FAULT_INJECTION
1155 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1156                                       size_t count, loff_t *ppos)
1157 {
1158         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1159         char buffer[PROC_NUMBUF];
1160         size_t len;
1161         int make_it_fail;
1162
1163         if (!task)
1164                 return -ESRCH;
1165         make_it_fail = task->make_it_fail;
1166         put_task_struct(task);
1167
1168         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1169
1170         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1171 }
1172
1173 static ssize_t proc_fault_inject_write(struct file * file,
1174                         const char __user * buf, size_t count, loff_t *ppos)
1175 {
1176         struct task_struct *task;
1177         char buffer[PROC_NUMBUF], *end;
1178         int make_it_fail;
1179
1180         if (!capable(CAP_SYS_RESOURCE))
1181                 return -EPERM;
1182         memset(buffer, 0, sizeof(buffer));
1183         if (count > sizeof(buffer) - 1)
1184                 count = sizeof(buffer) - 1;
1185         if (copy_from_user(buffer, buf, count))
1186                 return -EFAULT;
1187         make_it_fail = simple_strtol(buffer, &end, 0);
1188         if (*end == '\n')
1189                 end++;
1190         task = get_proc_task(file->f_dentry->d_inode);
1191         if (!task)
1192                 return -ESRCH;
1193         task->make_it_fail = make_it_fail;
1194         put_task_struct(task);
1195         if (end - buffer == 0)
1196                 return -EIO;
1197         return end - buffer;
1198 }
1199
1200 static const struct file_operations proc_fault_inject_operations = {
1201         .read           = proc_fault_inject_read,
1202         .write          = proc_fault_inject_write,
1203 };
1204 #endif
1205
1206
1207 #ifdef CONFIG_SCHED_DEBUG
1208 /*
1209  * Print out various scheduling related per-task fields:
1210  */
1211 static int sched_show(struct seq_file *m, void *v)
1212 {
1213         struct inode *inode = m->private;
1214         struct task_struct *p;
1215
1216         p = get_proc_task(inode);
1217         if (!p)
1218                 return -ESRCH;
1219         proc_sched_show_task(p, m);
1220
1221         put_task_struct(p);
1222
1223         return 0;
1224 }
1225
1226 static ssize_t
1227 sched_write(struct file *file, const char __user *buf,
1228             size_t count, loff_t *offset)
1229 {
1230         struct inode *inode = file->f_path.dentry->d_inode;
1231         struct task_struct *p;
1232
1233         p = get_proc_task(inode);
1234         if (!p)
1235                 return -ESRCH;
1236         proc_sched_set_task(p);
1237
1238         put_task_struct(p);
1239
1240         return count;
1241 }
1242
1243 static int sched_open(struct inode *inode, struct file *filp)
1244 {
1245         int ret;
1246
1247         ret = single_open(filp, sched_show, NULL);
1248         if (!ret) {
1249                 struct seq_file *m = filp->private_data;
1250
1251                 m->private = inode;
1252         }
1253         return ret;
1254 }
1255
1256 static const struct file_operations proc_pid_sched_operations = {
1257         .open           = sched_open,
1258         .read           = seq_read,
1259         .write          = sched_write,
1260         .llseek         = seq_lseek,
1261         .release        = single_release,
1262 };
1263
1264 #endif
1265
1266 /*
1267  * We added or removed a vma mapping the executable. The vmas are only mapped
1268  * during exec and are not mapped with the mmap system call.
1269  * Callers must hold down_write() on the mm's mmap_sem for these
1270  */
1271 void added_exe_file_vma(struct mm_struct *mm)
1272 {
1273         mm->num_exe_file_vmas++;
1274 }
1275
1276 void removed_exe_file_vma(struct mm_struct *mm)
1277 {
1278         mm->num_exe_file_vmas--;
1279         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1280                 fput(mm->exe_file);
1281                 mm->exe_file = NULL;
1282         }
1283
1284 }
1285
1286 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1287 {
1288         if (new_exe_file)
1289                 get_file(new_exe_file);
1290         if (mm->exe_file)
1291                 fput(mm->exe_file);
1292         mm->exe_file = new_exe_file;
1293         mm->num_exe_file_vmas = 0;
1294 }
1295
1296 struct file *get_mm_exe_file(struct mm_struct *mm)
1297 {
1298         struct file *exe_file;
1299
1300         /* We need mmap_sem to protect against races with removal of
1301          * VM_EXECUTABLE vmas */
1302         down_read(&mm->mmap_sem);
1303         exe_file = mm->exe_file;
1304         if (exe_file)
1305                 get_file(exe_file);
1306         up_read(&mm->mmap_sem);
1307         return exe_file;
1308 }
1309
1310 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1311 {
1312         /* It's safe to write the exe_file pointer without exe_file_lock because
1313          * this is called during fork when the task is not yet in /proc */
1314         newmm->exe_file = get_mm_exe_file(oldmm);
1315 }
1316
1317 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1318 {
1319         struct task_struct *task;
1320         struct mm_struct *mm;
1321         struct file *exe_file;
1322
1323         task = get_proc_task(inode);
1324         if (!task)
1325                 return -ENOENT;
1326         mm = get_task_mm(task);
1327         put_task_struct(task);
1328         if (!mm)
1329                 return -ENOENT;
1330         exe_file = get_mm_exe_file(mm);
1331         mmput(mm);
1332         if (exe_file) {
1333                 *exe_path = exe_file->f_path;
1334                 path_get(&exe_file->f_path);
1335                 fput(exe_file);
1336                 return 0;
1337         } else
1338                 return -ENOENT;
1339 }
1340
1341 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1342 {
1343         struct inode *inode = dentry->d_inode;
1344         int error = -EACCES;
1345
1346         /* We don't need a base pointer in the /proc filesystem */
1347         path_put(&nd->path);
1348
1349         /* Are we allowed to snoop on the tasks file descriptors? */
1350         if (!proc_fd_access_allowed(inode))
1351                 goto out;
1352
1353         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1354         nd->last_type = LAST_BIND;
1355 out:
1356         return ERR_PTR(error);
1357 }
1358
1359 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1360 {
1361         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1362         char *pathname;
1363         int len;
1364
1365         if (!tmp)
1366                 return -ENOMEM;
1367
1368         pathname = d_path(path, tmp, PAGE_SIZE);
1369         len = PTR_ERR(pathname);
1370         if (IS_ERR(pathname))
1371                 goto out;
1372         len = tmp + PAGE_SIZE - 1 - pathname;
1373
1374         if (len > buflen)
1375                 len = buflen;
1376         if (copy_to_user(buffer, pathname, len))
1377                 len = -EFAULT;
1378  out:
1379         free_page((unsigned long)tmp);
1380         return len;
1381 }
1382
1383 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1384 {
1385         int error = -EACCES;
1386         struct inode *inode = dentry->d_inode;
1387         struct path path;
1388
1389         /* Are we allowed to snoop on the tasks file descriptors? */
1390         if (!proc_fd_access_allowed(inode))
1391                 goto out;
1392
1393         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1394         if (error)
1395                 goto out;
1396
1397         error = do_proc_readlink(&path, buffer, buflen);
1398         path_put(&path);
1399 out:
1400         return error;
1401 }
1402
1403 static const struct inode_operations proc_pid_link_inode_operations = {
1404         .readlink       = proc_pid_readlink,
1405         .follow_link    = proc_pid_follow_link,
1406         .setattr        = proc_setattr,
1407 };
1408
1409
1410 /* building an inode */
1411
1412 static int task_dumpable(struct task_struct *task)
1413 {
1414         int dumpable = 0;
1415         struct mm_struct *mm;
1416
1417         task_lock(task);
1418         mm = task->mm;
1419         if (mm)
1420                 dumpable = get_dumpable(mm);
1421         task_unlock(task);
1422         if(dumpable == 1)
1423                 return 1;
1424         return 0;
1425 }
1426
1427
1428 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1429 {
1430         struct inode * inode;
1431         struct proc_inode *ei;
1432         const struct cred *cred;
1433
1434         /* We need a new inode */
1435
1436         inode = new_inode(sb);
1437         if (!inode)
1438                 goto out;
1439
1440         /* Common stuff */
1441         ei = PROC_I(inode);
1442         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1443         inode->i_op = &proc_def_inode_operations;
1444
1445         /*
1446          * grab the reference to task.
1447          */
1448         ei->pid = get_task_pid(task, PIDTYPE_PID);
1449         if (!ei->pid)
1450                 goto out_unlock;
1451
1452         if (task_dumpable(task)) {
1453                 rcu_read_lock();
1454                 cred = __task_cred(task);
1455                 inode->i_uid = cred->euid;
1456                 inode->i_gid = cred->egid;
1457                 rcu_read_unlock();
1458         }
1459         security_task_to_inode(task, inode);
1460
1461 out:
1462         return inode;
1463
1464 out_unlock:
1465         iput(inode);
1466         return NULL;
1467 }
1468
1469 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1470 {
1471         struct inode *inode = dentry->d_inode;
1472         struct task_struct *task;
1473         const struct cred *cred;
1474
1475         generic_fillattr(inode, stat);
1476
1477         rcu_read_lock();
1478         stat->uid = 0;
1479         stat->gid = 0;
1480         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1481         if (task) {
1482                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1483                     task_dumpable(task)) {
1484                         cred = __task_cred(task);
1485                         stat->uid = cred->euid;
1486                         stat->gid = cred->egid;
1487                 }
1488         }
1489         rcu_read_unlock();
1490         return 0;
1491 }
1492
1493 /* dentry stuff */
1494
1495 /*
1496  *      Exceptional case: normally we are not allowed to unhash a busy
1497  * directory. In this case, however, we can do it - no aliasing problems
1498  * due to the way we treat inodes.
1499  *
1500  * Rewrite the inode's ownerships here because the owning task may have
1501  * performed a setuid(), etc.
1502  *
1503  * Before the /proc/pid/status file was created the only way to read
1504  * the effective uid of a /process was to stat /proc/pid.  Reading
1505  * /proc/pid/status is slow enough that procps and other packages
1506  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1507  * made this apply to all per process world readable and executable
1508  * directories.
1509  */
1510 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1511 {
1512         struct inode *inode = dentry->d_inode;
1513         struct task_struct *task = get_proc_task(inode);
1514         const struct cred *cred;
1515
1516         if (task) {
1517                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1518                     task_dumpable(task)) {
1519                         rcu_read_lock();
1520                         cred = __task_cred(task);
1521                         inode->i_uid = cred->euid;
1522                         inode->i_gid = cred->egid;
1523                         rcu_read_unlock();
1524                 } else {
1525                         inode->i_uid = 0;
1526                         inode->i_gid = 0;
1527                 }
1528                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1529                 security_task_to_inode(task, inode);
1530                 put_task_struct(task);
1531                 return 1;
1532         }
1533         d_drop(dentry);
1534         return 0;
1535 }
1536
1537 static int pid_delete_dentry(struct dentry * dentry)
1538 {
1539         /* Is the task we represent dead?
1540          * If so, then don't put the dentry on the lru list,
1541          * kill it immediately.
1542          */
1543         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1544 }
1545
1546 static const struct dentry_operations pid_dentry_operations =
1547 {
1548         .d_revalidate   = pid_revalidate,
1549         .d_delete       = pid_delete_dentry,
1550 };
1551
1552 /* Lookups */
1553
1554 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1555                                 struct task_struct *, const void *);
1556
1557 /*
1558  * Fill a directory entry.
1559  *
1560  * If possible create the dcache entry and derive our inode number and
1561  * file type from dcache entry.
1562  *
1563  * Since all of the proc inode numbers are dynamically generated, the inode
1564  * numbers do not exist until the inode is cache.  This means creating the
1565  * the dcache entry in readdir is necessary to keep the inode numbers
1566  * reported by readdir in sync with the inode numbers reported
1567  * by stat.
1568  */
1569 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1570         char *name, int len,
1571         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1572 {
1573         struct dentry *child, *dir = filp->f_path.dentry;
1574         struct inode *inode;
1575         struct qstr qname;
1576         ino_t ino = 0;
1577         unsigned type = DT_UNKNOWN;
1578
1579         qname.name = name;
1580         qname.len  = len;
1581         qname.hash = full_name_hash(name, len);
1582
1583         child = d_lookup(dir, &qname);
1584         if (!child) {
1585                 struct dentry *new;
1586                 new = d_alloc(dir, &qname);
1587                 if (new) {
1588                         child = instantiate(dir->d_inode, new, task, ptr);
1589                         if (child)
1590                                 dput(new);
1591                         else
1592                                 child = new;
1593                 }
1594         }
1595         if (!child || IS_ERR(child) || !child->d_inode)
1596                 goto end_instantiate;
1597         inode = child->d_inode;
1598         if (inode) {
1599                 ino = inode->i_ino;
1600                 type = inode->i_mode >> 12;
1601         }
1602         dput(child);
1603 end_instantiate:
1604         if (!ino)
1605                 ino = find_inode_number(dir, &qname);
1606         if (!ino)
1607                 ino = 1;
1608         return filldir(dirent, name, len, filp->f_pos, ino, type);
1609 }
1610
1611 static unsigned name_to_int(struct dentry *dentry)
1612 {
1613         const char *name = dentry->d_name.name;
1614         int len = dentry->d_name.len;
1615         unsigned n = 0;
1616
1617         if (len > 1 && *name == '0')
1618                 goto out;
1619         while (len-- > 0) {
1620                 unsigned c = *name++ - '0';
1621                 if (c > 9)
1622                         goto out;
1623                 if (n >= (~0U-9)/10)
1624                         goto out;
1625                 n *= 10;
1626                 n += c;
1627         }
1628         return n;
1629 out:
1630         return ~0U;
1631 }
1632
1633 #define PROC_FDINFO_MAX 64
1634
1635 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1636 {
1637         struct task_struct *task = get_proc_task(inode);
1638         struct files_struct *files = NULL;
1639         struct file *file;
1640         int fd = proc_fd(inode);
1641
1642         if (task) {
1643                 files = get_files_struct(task);
1644                 put_task_struct(task);
1645         }
1646         if (files) {
1647                 /*
1648                  * We are not taking a ref to the file structure, so we must
1649                  * hold ->file_lock.
1650                  */
1651                 spin_lock(&files->file_lock);
1652                 file = fcheck_files(files, fd);
1653                 if (file) {
1654                         if (path) {
1655                                 *path = file->f_path;
1656                                 path_get(&file->f_path);
1657                         }
1658                         if (info)
1659                                 snprintf(info, PROC_FDINFO_MAX,
1660                                          "pos:\t%lli\n"
1661                                          "flags:\t0%o\n",
1662                                          (long long) file->f_pos,
1663                                          file->f_flags);
1664                         spin_unlock(&files->file_lock);
1665                         put_files_struct(files);
1666                         return 0;
1667                 }
1668                 spin_unlock(&files->file_lock);
1669                 put_files_struct(files);
1670         }
1671         return -ENOENT;
1672 }
1673
1674 static int proc_fd_link(struct inode *inode, struct path *path)
1675 {
1676         return proc_fd_info(inode, path, NULL);
1677 }
1678
1679 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1680 {
1681         struct inode *inode = dentry->d_inode;
1682         struct task_struct *task = get_proc_task(inode);
1683         int fd = proc_fd(inode);
1684         struct files_struct *files;
1685         const struct cred *cred;
1686
1687         if (task) {
1688                 files = get_files_struct(task);
1689                 if (files) {
1690                         rcu_read_lock();
1691                         if (fcheck_files(files, fd)) {
1692                                 rcu_read_unlock();
1693                                 put_files_struct(files);
1694                                 if (task_dumpable(task)) {
1695                                         rcu_read_lock();
1696                                         cred = __task_cred(task);
1697                                         inode->i_uid = cred->euid;
1698                                         inode->i_gid = cred->egid;
1699                                         rcu_read_unlock();
1700                                 } else {
1701                                         inode->i_uid = 0;
1702                                         inode->i_gid = 0;
1703                                 }
1704                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1705                                 security_task_to_inode(task, inode);
1706                                 put_task_struct(task);
1707                                 return 1;
1708                         }
1709                         rcu_read_unlock();
1710                         put_files_struct(files);
1711                 }
1712                 put_task_struct(task);
1713         }
1714         d_drop(dentry);
1715         return 0;
1716 }
1717
1718 static const struct dentry_operations tid_fd_dentry_operations =
1719 {
1720         .d_revalidate   = tid_fd_revalidate,
1721         .d_delete       = pid_delete_dentry,
1722 };
1723
1724 static struct dentry *proc_fd_instantiate(struct inode *dir,
1725         struct dentry *dentry, struct task_struct *task, const void *ptr)
1726 {
1727         unsigned fd = *(const unsigned *)ptr;
1728         struct file *file;
1729         struct files_struct *files;
1730         struct inode *inode;
1731         struct proc_inode *ei;
1732         struct dentry *error = ERR_PTR(-ENOENT);
1733
1734         inode = proc_pid_make_inode(dir->i_sb, task);
1735         if (!inode)
1736                 goto out;
1737         ei = PROC_I(inode);
1738         ei->fd = fd;
1739         files = get_files_struct(task);
1740         if (!files)
1741                 goto out_iput;
1742         inode->i_mode = S_IFLNK;
1743
1744         /*
1745          * We are not taking a ref to the file structure, so we must
1746          * hold ->file_lock.
1747          */
1748         spin_lock(&files->file_lock);
1749         file = fcheck_files(files, fd);
1750         if (!file)
1751                 goto out_unlock;
1752         if (file->f_mode & FMODE_READ)
1753                 inode->i_mode |= S_IRUSR | S_IXUSR;
1754         if (file->f_mode & FMODE_WRITE)
1755                 inode->i_mode |= S_IWUSR | S_IXUSR;
1756         spin_unlock(&files->file_lock);
1757         put_files_struct(files);
1758
1759         inode->i_op = &proc_pid_link_inode_operations;
1760         inode->i_size = 64;
1761         ei->op.proc_get_link = proc_fd_link;
1762         dentry->d_op = &tid_fd_dentry_operations;
1763         d_add(dentry, inode);
1764         /* Close the race of the process dying before we return the dentry */
1765         if (tid_fd_revalidate(dentry, NULL))
1766                 error = NULL;
1767
1768  out:
1769         return error;
1770 out_unlock:
1771         spin_unlock(&files->file_lock);
1772         put_files_struct(files);
1773 out_iput:
1774         iput(inode);
1775         goto out;
1776 }
1777
1778 static struct dentry *proc_lookupfd_common(struct inode *dir,
1779                                            struct dentry *dentry,
1780                                            instantiate_t instantiate)
1781 {
1782         struct task_struct *task = get_proc_task(dir);
1783         unsigned fd = name_to_int(dentry);
1784         struct dentry *result = ERR_PTR(-ENOENT);
1785
1786         if (!task)
1787                 goto out_no_task;
1788         if (fd == ~0U)
1789                 goto out;
1790
1791         result = instantiate(dir, dentry, task, &fd);
1792 out:
1793         put_task_struct(task);
1794 out_no_task:
1795         return result;
1796 }
1797
1798 static int proc_readfd_common(struct file * filp, void * dirent,
1799                               filldir_t filldir, instantiate_t instantiate)
1800 {
1801         struct dentry *dentry = filp->f_path.dentry;
1802         struct inode *inode = dentry->d_inode;
1803         struct task_struct *p = get_proc_task(inode);
1804         unsigned int fd, ino;
1805         int retval;
1806         struct files_struct * files;
1807
1808         retval = -ENOENT;
1809         if (!p)
1810                 goto out_no_task;
1811         retval = 0;
1812
1813         fd = filp->f_pos;
1814         switch (fd) {
1815                 case 0:
1816                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1817                                 goto out;
1818                         filp->f_pos++;
1819                 case 1:
1820                         ino = parent_ino(dentry);
1821                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1822                                 goto out;
1823                         filp->f_pos++;
1824                 default:
1825                         files = get_files_struct(p);
1826                         if (!files)
1827                                 goto out;
1828                         rcu_read_lock();
1829                         for (fd = filp->f_pos-2;
1830                              fd < files_fdtable(files)->max_fds;
1831                              fd++, filp->f_pos++) {
1832                                 char name[PROC_NUMBUF];
1833                                 int len;
1834
1835                                 if (!fcheck_files(files, fd))
1836                                         continue;
1837                                 rcu_read_unlock();
1838
1839                                 len = snprintf(name, sizeof(name), "%d", fd);
1840                                 if (proc_fill_cache(filp, dirent, filldir,
1841                                                     name, len, instantiate,
1842                                                     p, &fd) < 0) {
1843                                         rcu_read_lock();
1844                                         break;
1845                                 }
1846                                 rcu_read_lock();
1847                         }
1848                         rcu_read_unlock();
1849                         put_files_struct(files);
1850         }
1851 out:
1852         put_task_struct(p);
1853 out_no_task:
1854         return retval;
1855 }
1856
1857 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1858                                     struct nameidata *nd)
1859 {
1860         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1861 }
1862
1863 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1864 {
1865         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1866 }
1867
1868 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1869                                       size_t len, loff_t *ppos)
1870 {
1871         char tmp[PROC_FDINFO_MAX];
1872         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1873         if (!err)
1874                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1875         return err;
1876 }
1877
1878 static const struct file_operations proc_fdinfo_file_operations = {
1879         .open           = nonseekable_open,
1880         .read           = proc_fdinfo_read,
1881 };
1882
1883 static const struct file_operations proc_fd_operations = {
1884         .read           = generic_read_dir,
1885         .readdir        = proc_readfd,
1886 };
1887
1888 /*
1889  * /proc/pid/fd needs a special permission handler so that a process can still
1890  * access /proc/self/fd after it has executed a setuid().
1891  */
1892 static int proc_fd_permission(struct inode *inode, int mask)
1893 {
1894         int rv;
1895
1896         rv = generic_permission(inode, mask, NULL);
1897         if (rv == 0)
1898                 return 0;
1899         if (task_pid(current) == proc_pid(inode))
1900                 rv = 0;
1901         return rv;
1902 }
1903
1904 /*
1905  * proc directories can do almost nothing..
1906  */
1907 static const struct inode_operations proc_fd_inode_operations = {
1908         .lookup         = proc_lookupfd,
1909         .permission     = proc_fd_permission,
1910         .setattr        = proc_setattr,
1911 };
1912
1913 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1914         struct dentry *dentry, struct task_struct *task, const void *ptr)
1915 {
1916         unsigned fd = *(unsigned *)ptr;
1917         struct inode *inode;
1918         struct proc_inode *ei;
1919         struct dentry *error = ERR_PTR(-ENOENT);
1920
1921         inode = proc_pid_make_inode(dir->i_sb, task);
1922         if (!inode)
1923                 goto out;
1924         ei = PROC_I(inode);
1925         ei->fd = fd;
1926         inode->i_mode = S_IFREG | S_IRUSR;
1927         inode->i_fop = &proc_fdinfo_file_operations;
1928         dentry->d_op = &tid_fd_dentry_operations;
1929         d_add(dentry, inode);
1930         /* Close the race of the process dying before we return the dentry */
1931         if (tid_fd_revalidate(dentry, NULL))
1932                 error = NULL;
1933
1934  out:
1935         return error;
1936 }
1937
1938 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1939                                         struct dentry *dentry,
1940                                         struct nameidata *nd)
1941 {
1942         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1943 }
1944
1945 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1946 {
1947         return proc_readfd_common(filp, dirent, filldir,
1948                                   proc_fdinfo_instantiate);
1949 }
1950
1951 static const struct file_operations proc_fdinfo_operations = {
1952         .read           = generic_read_dir,
1953         .readdir        = proc_readfdinfo,
1954 };
1955
1956 /*
1957  * proc directories can do almost nothing..
1958  */
1959 static const struct inode_operations proc_fdinfo_inode_operations = {
1960         .lookup         = proc_lookupfdinfo,
1961         .setattr        = proc_setattr,
1962 };
1963
1964
1965 static struct dentry *proc_pident_instantiate(struct inode *dir,
1966         struct dentry *dentry, struct task_struct *task, const void *ptr)
1967 {
1968         const struct pid_entry *p = ptr;
1969         struct inode *inode;
1970         struct proc_inode *ei;
1971         struct dentry *error = ERR_PTR(-ENOENT);
1972
1973         inode = proc_pid_make_inode(dir->i_sb, task);
1974         if (!inode)
1975                 goto out;
1976
1977         ei = PROC_I(inode);
1978         inode->i_mode = p->mode;
1979         if (S_ISDIR(inode->i_mode))
1980                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1981         if (p->iop)
1982                 inode->i_op = p->iop;
1983         if (p->fop)
1984                 inode->i_fop = p->fop;
1985         ei->op = p->op;
1986         dentry->d_op = &pid_dentry_operations;
1987         d_add(dentry, inode);
1988         /* Close the race of the process dying before we return the dentry */
1989         if (pid_revalidate(dentry, NULL))
1990                 error = NULL;
1991 out:
1992         return error;
1993 }
1994
1995 static struct dentry *proc_pident_lookup(struct inode *dir,
1996                                          struct dentry *dentry,
1997                                          const struct pid_entry *ents,
1998                                          unsigned int nents)
1999 {
2000         struct dentry *error;
2001         struct task_struct *task = get_proc_task(dir);
2002         const struct pid_entry *p, *last;
2003
2004         error = ERR_PTR(-ENOENT);
2005
2006         if (!task)
2007                 goto out_no_task;
2008
2009         /*
2010          * Yes, it does not scale. And it should not. Don't add
2011          * new entries into /proc/<tgid>/ without very good reasons.
2012          */
2013         last = &ents[nents - 1];
2014         for (p = ents; p <= last; p++) {
2015                 if (p->len != dentry->d_name.len)
2016                         continue;
2017                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2018                         break;
2019         }
2020         if (p > last)
2021                 goto out;
2022
2023         error = proc_pident_instantiate(dir, dentry, task, p);
2024 out:
2025         put_task_struct(task);
2026 out_no_task:
2027         return error;
2028 }
2029
2030 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2031         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2032 {
2033         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2034                                 proc_pident_instantiate, task, p);
2035 }
2036
2037 static int proc_pident_readdir(struct file *filp,
2038                 void *dirent, filldir_t filldir,
2039                 const struct pid_entry *ents, unsigned int nents)
2040 {
2041         int i;
2042         struct dentry *dentry = filp->f_path.dentry;
2043         struct inode *inode = dentry->d_inode;
2044         struct task_struct *task = get_proc_task(inode);
2045         const struct pid_entry *p, *last;
2046         ino_t ino;
2047         int ret;
2048
2049         ret = -ENOENT;
2050         if (!task)
2051                 goto out_no_task;
2052
2053         ret = 0;
2054         i = filp->f_pos;
2055         switch (i) {
2056         case 0:
2057                 ino = inode->i_ino;
2058                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2059                         goto out;
2060                 i++;
2061                 filp->f_pos++;
2062                 /* fall through */
2063         case 1:
2064                 ino = parent_ino(dentry);
2065                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2066                         goto out;
2067                 i++;
2068                 filp->f_pos++;
2069                 /* fall through */
2070         default:
2071                 i -= 2;
2072                 if (i >= nents) {
2073                         ret = 1;
2074                         goto out;
2075                 }
2076                 p = ents + i;
2077                 last = &ents[nents - 1];
2078                 while (p <= last) {
2079                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2080                                 goto out;
2081                         filp->f_pos++;
2082                         p++;
2083                 }
2084         }
2085
2086         ret = 1;
2087 out:
2088         put_task_struct(task);
2089 out_no_task:
2090         return ret;
2091 }
2092
2093 #ifdef CONFIG_SECURITY
2094 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2095                                   size_t count, loff_t *ppos)
2096 {
2097         struct inode * inode = file->f_path.dentry->d_inode;
2098         char *p = NULL;
2099         ssize_t length;
2100         struct task_struct *task = get_proc_task(inode);
2101
2102         if (!task)
2103                 return -ESRCH;
2104
2105         length = security_getprocattr(task,
2106                                       (char*)file->f_path.dentry->d_name.name,
2107                                       &p);
2108         put_task_struct(task);
2109         if (length > 0)
2110                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2111         kfree(p);
2112         return length;
2113 }
2114
2115 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2116                                    size_t count, loff_t *ppos)
2117 {
2118         struct inode * inode = file->f_path.dentry->d_inode;
2119         char *page;
2120         ssize_t length;
2121         struct task_struct *task = get_proc_task(inode);
2122
2123         length = -ESRCH;
2124         if (!task)
2125                 goto out_no_task;
2126         if (count > PAGE_SIZE)
2127                 count = PAGE_SIZE;
2128
2129         /* No partial writes. */
2130         length = -EINVAL;
2131         if (*ppos != 0)
2132                 goto out;
2133
2134         length = -ENOMEM;
2135         page = (char*)__get_free_page(GFP_TEMPORARY);
2136         if (!page)
2137                 goto out;
2138
2139         length = -EFAULT;
2140         if (copy_from_user(page, buf, count))
2141                 goto out_free;
2142
2143         /* Guard against adverse ptrace interaction */
2144         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2145         if (length < 0)
2146                 goto out_free;
2147
2148         length = security_setprocattr(task,
2149                                       (char*)file->f_path.dentry->d_name.name,
2150                                       (void*)page, count);
2151         mutex_unlock(&task->cred_guard_mutex);
2152 out_free:
2153         free_page((unsigned long) page);
2154 out:
2155         put_task_struct(task);
2156 out_no_task:
2157         return length;
2158 }
2159
2160 static const struct file_operations proc_pid_attr_operations = {
2161         .read           = proc_pid_attr_read,
2162         .write          = proc_pid_attr_write,
2163 };
2164
2165 static const struct pid_entry attr_dir_stuff[] = {
2166         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2167         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2168         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2169         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2170         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2171         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2172 };
2173
2174 static int proc_attr_dir_readdir(struct file * filp,
2175                              void * dirent, filldir_t filldir)
2176 {
2177         return proc_pident_readdir(filp,dirent,filldir,
2178                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2179 }
2180
2181 static const struct file_operations proc_attr_dir_operations = {
2182         .read           = generic_read_dir,
2183         .readdir        = proc_attr_dir_readdir,
2184 };
2185
2186 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2187                                 struct dentry *dentry, struct nameidata *nd)
2188 {
2189         return proc_pident_lookup(dir, dentry,
2190                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2191 }
2192
2193 static const struct inode_operations proc_attr_dir_inode_operations = {
2194         .lookup         = proc_attr_dir_lookup,
2195         .getattr        = pid_getattr,
2196         .setattr        = proc_setattr,
2197 };
2198
2199 #endif
2200
2201 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2202 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2203                                          size_t count, loff_t *ppos)
2204 {
2205         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2206         struct mm_struct *mm;
2207         char buffer[PROC_NUMBUF];
2208         size_t len;
2209         int ret;
2210
2211         if (!task)
2212                 return -ESRCH;
2213
2214         ret = 0;
2215         mm = get_task_mm(task);
2216         if (mm) {
2217                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2218                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2219                                 MMF_DUMP_FILTER_SHIFT));
2220                 mmput(mm);
2221                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2222         }
2223
2224         put_task_struct(task);
2225
2226         return ret;
2227 }
2228
2229 static ssize_t proc_coredump_filter_write(struct file *file,
2230                                           const char __user *buf,
2231                                           size_t count,
2232                                           loff_t *ppos)
2233 {
2234         struct task_struct *task;
2235         struct mm_struct *mm;
2236         char buffer[PROC_NUMBUF], *end;
2237         unsigned int val;
2238         int ret;
2239         int i;
2240         unsigned long mask;
2241
2242         ret = -EFAULT;
2243         memset(buffer, 0, sizeof(buffer));
2244         if (count > sizeof(buffer) - 1)
2245                 count = sizeof(buffer) - 1;
2246         if (copy_from_user(buffer, buf, count))
2247                 goto out_no_task;
2248
2249         ret = -EINVAL;
2250         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2251         if (*end == '\n')
2252                 end++;
2253         if (end - buffer == 0)
2254                 goto out_no_task;
2255
2256         ret = -ESRCH;
2257         task = get_proc_task(file->f_dentry->d_inode);
2258         if (!task)
2259                 goto out_no_task;
2260
2261         ret = end - buffer;
2262         mm = get_task_mm(task);
2263         if (!mm)
2264                 goto out_no_mm;
2265
2266         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2267                 if (val & mask)
2268                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2269                 else
2270                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2271         }
2272
2273         mmput(mm);
2274  out_no_mm:
2275         put_task_struct(task);
2276  out_no_task:
2277         return ret;
2278 }
2279
2280 static const struct file_operations proc_coredump_filter_operations = {
2281         .read           = proc_coredump_filter_read,
2282         .write          = proc_coredump_filter_write,
2283 };
2284 #endif
2285
2286 /*
2287  * /proc/self:
2288  */
2289 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2290                               int buflen)
2291 {
2292         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2293         pid_t tgid = task_tgid_nr_ns(current, ns);
2294         char tmp[PROC_NUMBUF];
2295         if (!tgid)
2296                 return -ENOENT;
2297         sprintf(tmp, "%d", tgid);
2298         return vfs_readlink(dentry,buffer,buflen,tmp);
2299 }
2300
2301 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2302 {
2303         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2304         pid_t tgid = task_tgid_nr_ns(current, ns);
2305         char tmp[PROC_NUMBUF];
2306         if (!tgid)
2307                 return ERR_PTR(-ENOENT);
2308         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2309         return ERR_PTR(vfs_follow_link(nd,tmp));
2310 }
2311
2312 static const struct inode_operations proc_self_inode_operations = {
2313         .readlink       = proc_self_readlink,
2314         .follow_link    = proc_self_follow_link,
2315 };
2316
2317 /*
2318  * proc base
2319  *
2320  * These are the directory entries in the root directory of /proc
2321  * that properly belong to the /proc filesystem, as they describe
2322  * describe something that is process related.
2323  */
2324 static const struct pid_entry proc_base_stuff[] = {
2325         NOD("self", S_IFLNK|S_IRWXUGO,
2326                 &proc_self_inode_operations, NULL, {}),
2327 };
2328
2329 /*
2330  *      Exceptional case: normally we are not allowed to unhash a busy
2331  * directory. In this case, however, we can do it - no aliasing problems
2332  * due to the way we treat inodes.
2333  */
2334 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2335 {
2336         struct inode *inode = dentry->d_inode;
2337         struct task_struct *task = get_proc_task(inode);
2338         if (task) {
2339                 put_task_struct(task);
2340                 return 1;
2341         }
2342         d_drop(dentry);
2343         return 0;
2344 }
2345
2346 static const struct dentry_operations proc_base_dentry_operations =
2347 {
2348         .d_revalidate   = proc_base_revalidate,
2349         .d_delete       = pid_delete_dentry,
2350 };
2351
2352 static struct dentry *proc_base_instantiate(struct inode *dir,
2353         struct dentry *dentry, struct task_struct *task, const void *ptr)
2354 {
2355         const struct pid_entry *p = ptr;
2356         struct inode *inode;
2357         struct proc_inode *ei;
2358         struct dentry *error = ERR_PTR(-EINVAL);
2359
2360         /* Allocate the inode */
2361         error = ERR_PTR(-ENOMEM);
2362         inode = new_inode(dir->i_sb);
2363         if (!inode)
2364                 goto out;
2365
2366         /* Initialize the inode */
2367         ei = PROC_I(inode);
2368         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2369
2370         /*
2371          * grab the reference to the task.
2372          */
2373         ei->pid = get_task_pid(task, PIDTYPE_PID);
2374         if (!ei->pid)
2375                 goto out_iput;
2376
2377         inode->i_mode = p->mode;
2378         if (S_ISDIR(inode->i_mode))
2379                 inode->i_nlink = 2;
2380         if (S_ISLNK(inode->i_mode))
2381                 inode->i_size = 64;
2382         if (p->iop)
2383                 inode->i_op = p->iop;
2384         if (p->fop)
2385                 inode->i_fop = p->fop;
2386         ei->op = p->op;
2387         dentry->d_op = &proc_base_dentry_operations;
2388         d_add(dentry, inode);
2389         error = NULL;
2390 out:
2391         return error;
2392 out_iput:
2393         iput(inode);
2394         goto out;
2395 }
2396
2397 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2398 {
2399         struct dentry *error;
2400         struct task_struct *task = get_proc_task(dir);
2401         const struct pid_entry *p, *last;
2402
2403         error = ERR_PTR(-ENOENT);
2404
2405         if (!task)
2406                 goto out_no_task;
2407
2408         /* Lookup the directory entry */
2409         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2410         for (p = proc_base_stuff; p <= last; p++) {
2411                 if (p->len != dentry->d_name.len)
2412                         continue;
2413                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2414                         break;
2415         }
2416         if (p > last)
2417                 goto out;
2418
2419         error = proc_base_instantiate(dir, dentry, task, p);
2420
2421 out:
2422         put_task_struct(task);
2423 out_no_task:
2424         return error;
2425 }
2426
2427 static int proc_base_fill_cache(struct file *filp, void *dirent,
2428         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2429 {
2430         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2431                                 proc_base_instantiate, task, p);
2432 }
2433
2434 #ifdef CONFIG_TASK_IO_ACCOUNTING
2435 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2436 {
2437         struct task_io_accounting acct = task->ioac;
2438         unsigned long flags;
2439
2440         if (whole && lock_task_sighand(task, &flags)) {
2441                 struct task_struct *t = task;
2442
2443                 task_io_accounting_add(&acct, &task->signal->ioac);
2444                 while_each_thread(task, t)
2445                         task_io_accounting_add(&acct, &t->ioac);
2446
2447                 unlock_task_sighand(task, &flags);
2448         }
2449         return sprintf(buffer,
2450                         "rchar: %llu\n"
2451                         "wchar: %llu\n"
2452                         "syscr: %llu\n"
2453                         "syscw: %llu\n"
2454                         "read_bytes: %llu\n"
2455                         "write_bytes: %llu\n"
2456                         "cancelled_write_bytes: %llu\n",
2457                         (unsigned long long)acct.rchar,
2458                         (unsigned long long)acct.wchar,
2459                         (unsigned long long)acct.syscr,
2460                         (unsigned long long)acct.syscw,
2461                         (unsigned long long)acct.read_bytes,
2462                         (unsigned long long)acct.write_bytes,
2463                         (unsigned long long)acct.cancelled_write_bytes);
2464 }
2465
2466 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2467 {
2468         return do_io_accounting(task, buffer, 0);
2469 }
2470
2471 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2472 {
2473         return do_io_accounting(task, buffer, 1);
2474 }
2475 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2476
2477 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2478                                 struct pid *pid, struct task_struct *task)
2479 {
2480         seq_printf(m, "%08x\n", task->personality);
2481         return 0;
2482 }
2483
2484 /*
2485  * Thread groups
2486  */
2487 static const struct file_operations proc_task_operations;
2488 static const struct inode_operations proc_task_inode_operations;
2489
2490 static const struct pid_entry tgid_base_stuff[] = {
2491         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2492         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2493         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2494 #ifdef CONFIG_NET
2495         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2496 #endif
2497         REG("environ",    S_IRUSR, proc_environ_operations),
2498         INF("auxv",       S_IRUSR, proc_pid_auxv),
2499         ONE("status",     S_IRUGO, proc_pid_status),
2500         ONE("personality", S_IRUSR, proc_pid_personality),
2501         INF("limits",     S_IRUSR, proc_pid_limits),
2502 #ifdef CONFIG_SCHED_DEBUG
2503         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2504 #endif
2505 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2506         INF("syscall",    S_IRUSR, proc_pid_syscall),
2507 #endif
2508         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2509         ONE("stat",       S_IRUGO, proc_tgid_stat),
2510         ONE("statm",      S_IRUGO, proc_pid_statm),
2511         REG("maps",       S_IRUGO, proc_maps_operations),
2512 #ifdef CONFIG_NUMA
2513         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2514 #endif
2515         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2516         LNK("cwd",        proc_cwd_link),
2517         LNK("root",       proc_root_link),
2518         LNK("exe",        proc_exe_link),
2519         REG("mounts",     S_IRUGO, proc_mounts_operations),
2520         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2521         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2522 #ifdef CONFIG_PROC_PAGE_MONITOR
2523         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2524         REG("smaps",      S_IRUGO, proc_smaps_operations),
2525         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2526 #endif
2527 #ifdef CONFIG_SECURITY
2528         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2529 #endif
2530 #ifdef CONFIG_KALLSYMS
2531         INF("wchan",      S_IRUGO, proc_pid_wchan),
2532 #endif
2533 #ifdef CONFIG_STACKTRACE
2534         ONE("stack",      S_IRUSR, proc_pid_stack),
2535 #endif
2536 #ifdef CONFIG_SCHEDSTATS
2537         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2538 #endif
2539 #ifdef CONFIG_LATENCYTOP
2540         REG("latency",  S_IRUGO, proc_lstats_operations),
2541 #endif
2542 #ifdef CONFIG_PROC_PID_CPUSET
2543         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2544 #endif
2545 #ifdef CONFIG_CGROUPS
2546         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2547 #endif
2548         INF("oom_score",  S_IRUGO, proc_oom_score),
2549         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2550 #ifdef CONFIG_AUDITSYSCALL
2551         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2552         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2553 #endif
2554 #ifdef CONFIG_FAULT_INJECTION
2555         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2556 #endif
2557 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2558         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2559 #endif
2560 #ifdef CONFIG_TASK_IO_ACCOUNTING
2561         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2562 #endif
2563 };
2564
2565 static int proc_tgid_base_readdir(struct file * filp,
2566                              void * dirent, filldir_t filldir)
2567 {
2568         return proc_pident_readdir(filp,dirent,filldir,
2569                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2570 }
2571
2572 static const struct file_operations proc_tgid_base_operations = {
2573         .read           = generic_read_dir,
2574         .readdir        = proc_tgid_base_readdir,
2575 };
2576
2577 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2578         return proc_pident_lookup(dir, dentry,
2579                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2580 }
2581
2582 static const struct inode_operations proc_tgid_base_inode_operations = {
2583         .lookup         = proc_tgid_base_lookup,
2584         .getattr        = pid_getattr,
2585         .setattr        = proc_setattr,
2586 };
2587
2588 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2589 {
2590         struct dentry *dentry, *leader, *dir;
2591         char buf[PROC_NUMBUF];
2592         struct qstr name;
2593
2594         name.name = buf;
2595         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2596         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2597         if (dentry) {
2598                 if (!(current->flags & PF_EXITING))
2599                         shrink_dcache_parent(dentry);
2600                 d_drop(dentry);
2601                 dput(dentry);
2602         }
2603
2604         if (tgid == 0)
2605                 goto out;
2606
2607         name.name = buf;
2608         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2609         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2610         if (!leader)
2611                 goto out;
2612
2613         name.name = "task";
2614         name.len = strlen(name.name);
2615         dir = d_hash_and_lookup(leader, &name);
2616         if (!dir)
2617                 goto out_put_leader;
2618
2619         name.name = buf;
2620         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2621         dentry = d_hash_and_lookup(dir, &name);
2622         if (dentry) {
2623                 shrink_dcache_parent(dentry);
2624                 d_drop(dentry);
2625                 dput(dentry);
2626         }
2627
2628         dput(dir);
2629 out_put_leader:
2630         dput(leader);
2631 out:
2632         return;
2633 }
2634
2635 /**
2636  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2637  * @task: task that should be flushed.
2638  *
2639  * When flushing dentries from proc, one needs to flush them from global
2640  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2641  * in. This call is supposed to do all of this job.
2642  *
2643  * Looks in the dcache for
2644  * /proc/@pid
2645  * /proc/@tgid/task/@pid
2646  * if either directory is present flushes it and all of it'ts children
2647  * from the dcache.
2648  *
2649  * It is safe and reasonable to cache /proc entries for a task until
2650  * that task exits.  After that they just clog up the dcache with
2651  * useless entries, possibly causing useful dcache entries to be
2652  * flushed instead.  This routine is proved to flush those useless
2653  * dcache entries at process exit time.
2654  *
2655  * NOTE: This routine is just an optimization so it does not guarantee
2656  *       that no dcache entries will exist at process exit time it
2657  *       just makes it very unlikely that any will persist.
2658  */
2659
2660 void proc_flush_task(struct task_struct *task)
2661 {
2662         int i;
2663         struct pid *pid, *tgid = NULL;
2664         struct upid *upid;
2665
2666         pid = task_pid(task);
2667         if (thread_group_leader(task))
2668                 tgid = task_tgid(task);
2669
2670         for (i = 0; i <= pid->level; i++) {
2671                 upid = &pid->numbers[i];
2672                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2673                         tgid ? tgid->numbers[i].nr : 0);
2674         }
2675
2676         upid = &pid->numbers[pid->level];
2677         if (upid->nr == 1)
2678                 pid_ns_release_proc(upid->ns);
2679 }
2680
2681 static struct dentry *proc_pid_instantiate(struct inode *dir,
2682                                            struct dentry * dentry,
2683                                            struct task_struct *task, const void *ptr)
2684 {
2685         struct dentry *error = ERR_PTR(-ENOENT);
2686         struct inode *inode;
2687
2688         inode = proc_pid_make_inode(dir->i_sb, task);
2689         if (!inode)
2690                 goto out;
2691
2692         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2693         inode->i_op = &proc_tgid_base_inode_operations;
2694         inode->i_fop = &proc_tgid_base_operations;
2695         inode->i_flags|=S_IMMUTABLE;
2696
2697         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2698                 ARRAY_SIZE(tgid_base_stuff));
2699
2700         dentry->d_op = &pid_dentry_operations;
2701
2702         d_add(dentry, inode);
2703         /* Close the race of the process dying before we return the dentry */
2704         if (pid_revalidate(dentry, NULL))
2705                 error = NULL;
2706 out:
2707         return error;
2708 }
2709
2710 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2711 {
2712         struct dentry *result = ERR_PTR(-ENOENT);
2713         struct task_struct *task;
2714         unsigned tgid;
2715         struct pid_namespace *ns;
2716
2717         result = proc_base_lookup(dir, dentry);
2718         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2719                 goto out;
2720
2721         tgid = name_to_int(dentry);
2722         if (tgid == ~0U)
2723                 goto out;
2724
2725         ns = dentry->d_sb->s_fs_info;
2726         rcu_read_lock();
2727         task = find_task_by_pid_ns(tgid, ns);
2728         if (task)
2729                 get_task_struct(task);
2730         rcu_read_unlock();
2731         if (!task)
2732                 goto out;
2733
2734         result = proc_pid_instantiate(dir, dentry, task, NULL);
2735         put_task_struct(task);
2736 out:
2737         return result;
2738 }
2739
2740 /*
2741  * Find the first task with tgid >= tgid
2742  *
2743  */
2744 struct tgid_iter {
2745         unsigned int tgid;
2746         struct task_struct *task;
2747 };
2748 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2749 {
2750         struct pid *pid;
2751
2752         if (iter.task)
2753                 put_task_struct(iter.task);
2754         rcu_read_lock();
2755 retry:
2756         iter.task = NULL;
2757         pid = find_ge_pid(iter.tgid, ns);
2758         if (pid) {
2759                 iter.tgid = pid_nr_ns(pid, ns);
2760                 iter.task = pid_task(pid, PIDTYPE_PID);
2761                 /* What we to know is if the pid we have find is the
2762                  * pid of a thread_group_leader.  Testing for task
2763                  * being a thread_group_leader is the obvious thing
2764                  * todo but there is a window when it fails, due to
2765                  * the pid transfer logic in de_thread.
2766                  *
2767                  * So we perform the straight forward test of seeing
2768                  * if the pid we have found is the pid of a thread
2769                  * group leader, and don't worry if the task we have
2770                  * found doesn't happen to be a thread group leader.
2771                  * As we don't care in the case of readdir.
2772                  */
2773                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2774                         iter.tgid += 1;
2775                         goto retry;
2776                 }
2777                 get_task_struct(iter.task);
2778         }
2779         rcu_read_unlock();
2780         return iter;
2781 }
2782
2783 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2784
2785 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2786         struct tgid_iter iter)
2787 {
2788         char name[PROC_NUMBUF];
2789         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2790         return proc_fill_cache(filp, dirent, filldir, name, len,
2791                                 proc_pid_instantiate, iter.task, NULL);
2792 }
2793
2794 /* for the /proc/ directory itself, after non-process stuff has been done */
2795 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2796 {
2797         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2798         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2799         struct tgid_iter iter;
2800         struct pid_namespace *ns;
2801
2802         if (!reaper)
2803                 goto out_no_task;
2804
2805         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2806                 const struct pid_entry *p = &proc_base_stuff[nr];
2807                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2808                         goto out;
2809         }
2810
2811         ns = filp->f_dentry->d_sb->s_fs_info;
2812         iter.task = NULL;
2813         iter.tgid = filp->f_pos - TGID_OFFSET;
2814         for (iter = next_tgid(ns, iter);
2815              iter.task;
2816              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2817                 filp->f_pos = iter.tgid + TGID_OFFSET;
2818                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2819                         put_task_struct(iter.task);
2820                         goto out;
2821                 }
2822         }
2823         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2824 out:
2825         put_task_struct(reaper);
2826 out_no_task:
2827         return 0;
2828 }
2829
2830 /*
2831  * Tasks
2832  */
2833 static const struct pid_entry tid_base_stuff[] = {
2834         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2835         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2836         REG("environ",   S_IRUSR, proc_environ_operations),
2837         INF("auxv",      S_IRUSR, proc_pid_auxv),
2838         ONE("status",    S_IRUGO, proc_pid_status),
2839         ONE("personality", S_IRUSR, proc_pid_personality),
2840         INF("limits",    S_IRUSR, proc_pid_limits),
2841 #ifdef CONFIG_SCHED_DEBUG
2842         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2843 #endif
2844 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2845         INF("syscall",   S_IRUSR, proc_pid_syscall),
2846 #endif
2847         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2848         ONE("stat",      S_IRUGO, proc_tid_stat),
2849         ONE("statm",     S_IRUGO, proc_pid_statm),
2850         REG("maps",      S_IRUGO, proc_maps_operations),
2851 #ifdef CONFIG_NUMA
2852         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2853 #endif
2854         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2855         LNK("cwd",       proc_cwd_link),
2856         LNK("root",      proc_root_link),
2857         LNK("exe",       proc_exe_link),
2858         REG("mounts",    S_IRUGO, proc_mounts_operations),
2859         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2860 #ifdef CONFIG_PROC_PAGE_MONITOR
2861         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2862         REG("smaps",     S_IRUGO, proc_smaps_operations),
2863         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2864 #endif
2865 #ifdef CONFIG_SECURITY
2866         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2867 #endif
2868 #ifdef CONFIG_KALLSYMS
2869         INF("wchan",     S_IRUGO, proc_pid_wchan),
2870 #endif
2871 #ifdef CONFIG_STACKTRACE
2872         ONE("stack",      S_IRUSR, proc_pid_stack),
2873 #endif
2874 #ifdef CONFIG_SCHEDSTATS
2875         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2876 #endif
2877 #ifdef CONFIG_LATENCYTOP
2878         REG("latency",  S_IRUGO, proc_lstats_operations),
2879 #endif
2880 #ifdef CONFIG_PROC_PID_CPUSET
2881         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2882 #endif
2883 #ifdef CONFIG_CGROUPS
2884         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2885 #endif
2886         INF("oom_score", S_IRUGO, proc_oom_score),
2887         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2888 #ifdef CONFIG_AUDITSYSCALL
2889         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2890         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2891 #endif
2892 #ifdef CONFIG_FAULT_INJECTION
2893         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2894 #endif
2895 #ifdef CONFIG_TASK_IO_ACCOUNTING
2896         INF("io",       S_IRUGO, proc_tid_io_accounting),
2897 #endif
2898 };
2899
2900 static int proc_tid_base_readdir(struct file * filp,
2901                              void * dirent, filldir_t filldir)
2902 {
2903         return proc_pident_readdir(filp,dirent,filldir,
2904                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2905 }
2906
2907 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2908         return proc_pident_lookup(dir, dentry,
2909                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2910 }
2911
2912 static const struct file_operations proc_tid_base_operations = {
2913         .read           = generic_read_dir,
2914         .readdir        = proc_tid_base_readdir,
2915 };
2916
2917 static const struct inode_operations proc_tid_base_inode_operations = {
2918         .lookup         = proc_tid_base_lookup,
2919         .getattr        = pid_getattr,
2920         .setattr        = proc_setattr,
2921 };
2922
2923 static struct dentry *proc_task_instantiate(struct inode *dir,
2924         struct dentry *dentry, struct task_struct *task, const void *ptr)
2925 {
2926         struct dentry *error = ERR_PTR(-ENOENT);
2927         struct inode *inode;
2928         inode = proc_pid_make_inode(dir->i_sb, task);
2929
2930         if (!inode)
2931                 goto out;
2932         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2933         inode->i_op = &proc_tid_base_inode_operations;
2934         inode->i_fop = &proc_tid_base_operations;
2935         inode->i_flags|=S_IMMUTABLE;
2936
2937         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2938                 ARRAY_SIZE(tid_base_stuff));
2939
2940         dentry->d_op = &pid_dentry_operations;
2941
2942         d_add(dentry, inode);
2943         /* Close the race of the process dying before we return the dentry */
2944         if (pid_revalidate(dentry, NULL))
2945                 error = NULL;
2946 out:
2947         return error;
2948 }
2949
2950 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2951 {
2952         struct dentry *result = ERR_PTR(-ENOENT);
2953         struct task_struct *task;
2954         struct task_struct *leader = get_proc_task(dir);
2955         unsigned tid;
2956         struct pid_namespace *ns;
2957
2958         if (!leader)
2959                 goto out_no_task;
2960
2961         tid = name_to_int(dentry);
2962         if (tid == ~0U)
2963                 goto out;
2964
2965         ns = dentry->d_sb->s_fs_info;
2966         rcu_read_lock();
2967         task = find_task_by_pid_ns(tid, ns);
2968         if (task)
2969                 get_task_struct(task);
2970         rcu_read_unlock();
2971         if (!task)
2972                 goto out;
2973         if (!same_thread_group(leader, task))
2974                 goto out_drop_task;
2975
2976         result = proc_task_instantiate(dir, dentry, task, NULL);
2977 out_drop_task:
2978         put_task_struct(task);
2979 out:
2980         put_task_struct(leader);
2981 out_no_task:
2982         return result;
2983 }
2984
2985 /*
2986  * Find the first tid of a thread group to return to user space.
2987  *
2988  * Usually this is just the thread group leader, but if the users
2989  * buffer was too small or there was a seek into the middle of the
2990  * directory we have more work todo.
2991  *
2992  * In the case of a short read we start with find_task_by_pid.
2993  *
2994  * In the case of a seek we start with the leader and walk nr
2995  * threads past it.
2996  */
2997 static struct task_struct *first_tid(struct task_struct *leader,
2998                 int tid, int nr, struct pid_namespace *ns)
2999 {
3000         struct task_struct *pos;
3001
3002         rcu_read_lock();
3003         /* Attempt to start with the pid of a thread */
3004         if (tid && (nr > 0)) {
3005                 pos = find_task_by_pid_ns(tid, ns);
3006                 if (pos && (pos->group_leader == leader))
3007                         goto found;
3008         }
3009
3010         /* If nr exceeds the number of threads there is nothing todo */
3011         pos = NULL;
3012         if (nr && nr >= get_nr_threads(leader))
3013                 goto out;
3014
3015         /* If we haven't found our starting place yet start
3016          * with the leader and walk nr threads forward.
3017          */
3018         for (pos = leader; nr > 0; --nr) {
3019                 pos = next_thread(pos);
3020                 if (pos == leader) {
3021                         pos = NULL;
3022                         goto out;
3023                 }
3024         }
3025 found:
3026         get_task_struct(pos);
3027 out:
3028         rcu_read_unlock();
3029         return pos;
3030 }
3031
3032 /*
3033  * Find the next thread in the thread list.
3034  * Return NULL if there is an error or no next thread.
3035  *
3036  * The reference to the input task_struct is released.
3037  */
3038 static struct task_struct *next_tid(struct task_struct *start)
3039 {
3040         struct task_struct *pos = NULL;
3041         rcu_read_lock();
3042         if (pid_alive(start)) {
3043                 pos = next_thread(start);
3044                 if (thread_group_leader(pos))
3045                         pos = NULL;
3046                 else
3047                         get_task_struct(pos);
3048         }
3049         rcu_read_unlock();
3050         put_task_struct(start);
3051         return pos;
3052 }
3053
3054 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3055         struct task_struct *task, int tid)
3056 {
3057         char name[PROC_NUMBUF];
3058         int len = snprintf(name, sizeof(name), "%d", tid);
3059         return proc_fill_cache(filp, dirent, filldir, name, len,
3060                                 proc_task_instantiate, task, NULL);
3061 }
3062
3063 /* for the /proc/TGID/task/ directories */
3064 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3065 {
3066         struct dentry *dentry = filp->f_path.dentry;
3067         struct inode *inode = dentry->d_inode;
3068         struct task_struct *leader = NULL;
3069         struct task_struct *task;
3070         int retval = -ENOENT;
3071         ino_t ino;
3072         int tid;
3073         struct pid_namespace *ns;
3074
3075         task = get_proc_task(inode);
3076         if (!task)
3077                 goto out_no_task;
3078         rcu_read_lock();
3079         if (pid_alive(task)) {
3080                 leader = task->group_leader;
3081                 get_task_struct(leader);
3082         }
3083         rcu_read_unlock();
3084         put_task_struct(task);
3085         if (!leader)
3086                 goto out_no_task;
3087         retval = 0;
3088
3089         switch ((unsigned long)filp->f_pos) {
3090         case 0:
3091                 ino = inode->i_ino;
3092                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3093                         goto out;
3094                 filp->f_pos++;
3095                 /* fall through */
3096         case 1:
3097                 ino = parent_ino(dentry);
3098                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3099                         goto out;
3100                 filp->f_pos++;
3101                 /* fall through */
3102         }
3103
3104         /* f_version caches the tgid value that the last readdir call couldn't
3105          * return. lseek aka telldir automagically resets f_version to 0.
3106          */
3107         ns = filp->f_dentry->d_sb->s_fs_info;
3108         tid = (int)filp->f_version;
3109         filp->f_version = 0;
3110         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3111              task;
3112              task = next_tid(task), filp->f_pos++) {
3113                 tid = task_pid_nr_ns(task, ns);
3114                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3115                         /* returning this tgid failed, save it as the first
3116                          * pid for the next readir call */
3117                         filp->f_version = (u64)tid;
3118                         put_task_struct(task);
3119                         break;
3120                 }
3121         }
3122 out:
3123         put_task_struct(leader);
3124 out_no_task:
3125         return retval;
3126 }
3127
3128 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3129 {
3130         struct inode *inode = dentry->d_inode;
3131         struct task_struct *p = get_proc_task(inode);
3132         generic_fillattr(inode, stat);
3133
3134         if (p) {
3135                 stat->nlink += get_nr_threads(p);
3136                 put_task_struct(p);
3137         }
3138
3139         return 0;
3140 }
3141
3142 static const struct inode_operations proc_task_inode_operations = {
3143         .lookup         = proc_task_lookup,
3144         .getattr        = proc_task_getattr,
3145         .setattr        = proc_setattr,
3146 };
3147
3148 static const struct file_operations proc_task_operations = {
3149         .read           = generic_read_dir,
3150         .readdir        = proc_task_readdir,
3151 };