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 <linux/slab.h>
  85 #include "internal.h"
  86
  87 /* NOTE:
  88  *      Implementing inode permission operations in /proc is almost
  89  *      certainly an error.  Permission checks need to happen during
  90  *      each system call not at open time.  The reason is that most of
  91  *      what we wish to check for permissions in /proc varies at runtime.
  92  *
  93  *      The classic example of a problem is opening file descriptors
  94  *      in /proc for a task before it execs a suid executable.
  95  */
  96
  97 struct pid_entry {
  98         char *name;
  99         int len;
 100         mode_t mode;
 101         const struct inode_operations *iop;
 102         const struct file_operations *fop;
 103         union proc_op op;
 104 };
 105
 106 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 107         .name = (NAME),                                 \
 108         .len  = sizeof(NAME) - 1,                       \
 109         .mode = MODE,                                   \
 110         .iop  = IOP,                                    \
 111         .fop  = FOP,                                    \
 112         .op   = OP,                                     \
 113 }
 114
 115 #define DIR(NAME, MODE, iops, fops)     \
 116         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 117 #define LNK(NAME, get_link)                                     \
 118         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 119                 &proc_pid_link_inode_operations, NULL,          \
 120                 { .proc_get_link = get_link } )
 121 #define REG(NAME, MODE, fops)                           \
 122         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 123 #define INF(NAME, MODE, read)                           \
 124         NOD(NAME, (S_IFREG|(MODE)),                     \
 125                 NULL, &proc_info_file_operations,       \
 126                 { .proc_read = read } )
 127 #define ONE(NAME, MODE, show)                           \
 128         NOD(NAME, (S_IFREG|(MODE)),                     \
 129                 NULL, &proc_single_file_operations,     \
 130                 { .proc_show = show } )
 131
 132 /*
 133  * Count the number of hardlinks for the pid_entry table, excluding the .
 134  * and .. links.
 135  */
 136 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 137         unsigned int n)
 138 {
 139         unsigned int i;
 140         unsigned int count;
 141
 142         count = 0;
 143         for (i = 0; i < n; ++i) {
 144                 if (S_ISDIR(entries[i].mode))
 145                         ++count;
 146         }
 147
 148         return count;
 149 }
 150
 151 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
 152 {
 153         struct fs_struct *fs;
 154         int result = -ENOENT;
 155
 156         task_lock(task);
 157         fs = task->fs;
 158         if (fs) {
 159                 read_lock(&fs->lock);
 160                 *path = root ? fs->root : fs->pwd;
 161                 path_get(path);
 162                 read_unlock(&fs->lock);
 163                 result = 0;
 164         }
 165         task_unlock(task);
 166         return result;
 167 }
 168
 169 static int get_nr_threads(struct task_struct *tsk)
 170 {
 171         unsigned long flags;
 172         int count = 0;
 173
 174         if (lock_task_sighand(tsk, &flags)) {
 175                 count = atomic_read(&tsk->signal->count);
 176                 unlock_task_sighand(tsk, &flags);
 177         }
 178         return count;
 179 }
 180
 181 static int proc_cwd_link(struct inode *inode, struct path *path)
 182 {
 183         struct task_struct *task = get_proc_task(inode);
 184         int result = -ENOENT;
 185
 186         if (task) {
 187                 result = get_fs_path(task, path, 0);
 188                 put_task_struct(task);
 189         }
 190         return result;
 191 }
 192
 193 static int proc_root_link(struct inode *inode, struct path *path)
 194 {
 195         struct task_struct *task = get_proc_task(inode);
 196         int result = -ENOENT;
 197
 198         if (task) {
 199                 result = get_fs_path(task, path, 1);
 200                 put_task_struct(task);
 201         }
 202         return result;
 203 }
 204
 205 /*
 206  * Return zero if current may access user memory in @task, -error if not.
 207  */
 208 static int check_mem_permission(struct task_struct *task)
 209 {
 210         /*
 211          * A task can always look at itself, in case it chooses
 212          * to use system calls instead of load instructions.
 213          */
 214         if (task == current)
 215                 return 0;
 216
 217         /*
 218          * If current is actively ptrace'ing, and would also be
 219          * permitted to freshly attach with ptrace now, permit it.
 220          */
 221         if (task_is_stopped_or_traced(task)) {
 222                 int match;
 223                 rcu_read_lock();
 224                 match = (tracehook_tracer_task(task) == current);
 225                 rcu_read_unlock();
 226                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 227                         return 0;
 228         }
 229
 230         /*
 231          * Noone else is allowed.
 232          */
 233         return -EPERM;
 234 }
 235
 236 struct mm_struct *mm_for_maps(struct task_struct *task)
 237 {
 238         struct mm_struct *mm;
 239
 240         if (mutex_lock_killable(&task->cred_guard_mutex))
 241                 return NULL;
 242
 243         mm = get_task_mm(task);
 244         if (mm && mm != current->mm &&
 245                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 246                 mmput(mm);
 247                 mm = NULL;
 248         }
 249         mutex_unlock(&task->cred_guard_mutex);
 250
 251         return mm;
 252 }
 253
 254 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 255 {
 256         int res = 0;
 257         unsigned int len;
 258         struct mm_struct *mm = get_task_mm(task);
 259         if (!mm)
 260                 goto out;
 261         if (!mm->arg_end)
 262                 goto out_mm;    /* Shh! No looking before we're done */
 263
 264         len = mm->arg_end - mm->arg_start;
 265
 266         if (len > PAGE_SIZE)
 267                 len = PAGE_SIZE;
 268
 269         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 270
 271         // If the nul at the end of args has been overwritten, then
 272         // assume application is using setproctitle(3).
 273         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 274                 len = strnlen(buffer, res);
 275                 if (len < res) {
 276                     res = len;
 277                 } else {
 278                         len = mm->env_end - mm->env_start;
 279                         if (len > PAGE_SIZE - res)
 280                                 len = PAGE_SIZE - res;
 281                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 282                         res = strnlen(buffer, res);
 283                 }
 284         }
 285 out_mm:
 286         mmput(mm);
 287 out:
 288         return res;
 289 }
 290
 291 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 292 {
 293         int res = 0;
 294         struct mm_struct *mm = get_task_mm(task);
 295         if (mm) {
 296                 unsigned int nwords = 0;
 297                 do {
 298                         nwords += 2;
 299                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 300                 res = nwords * sizeof(mm->saved_auxv[0]);
 301                 if (res > PAGE_SIZE)
 302                         res = PAGE_SIZE;
 303                 memcpy(buffer, mm->saved_auxv, res);
 304                 mmput(mm);
 305         }
 306         return res;
 307 }
 308
 309
 310 #ifdef CONFIG_KALLSYMS
 311 /*
 312  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 313  * Returns the resolved symbol.  If that fails, simply return the address.
 314  */
 315 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 316 {
 317         unsigned long wchan;
 318         char symname[KSYM_NAME_LEN];
 319
 320         wchan = get_wchan(task);
 321
 322         if (lookup_symbol_name(wchan, symname) < 0)
 323                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 324                         return 0;
 325                 else
 326                         return sprintf(buffer, "%lu", wchan);
 327         else
 328                 return sprintf(buffer, "%s", symname);
 329 }
 330 #endif /* CONFIG_KALLSYMS */
 331
 332 #ifdef CONFIG_STACKTRACE
 333
 334 #define MAX_STACK_TRACE_DEPTH   64
 335
 336 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 337                           struct pid *pid, struct task_struct *task)
 338 {
 339         struct stack_trace trace;
 340         unsigned long *entries;
 341         int i;
 342
 343         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 344         if (!entries)
 345                 return -ENOMEM;
 346
 347         trace.nr_entries        = 0;
 348         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 349         trace.entries           = entries;
 350         trace.skip              = 0;
 351         save_stack_trace_tsk(task, &trace);
 352
 353         for (i = 0; i < trace.nr_entries; i++) {
 354                 seq_printf(m, "[<%p>] %pS\n",
 355                            (void *)entries[i], (void *)entries[i]);
 356         }
 357         kfree(entries);
 358
 359         return 0;
 360 }
 361 #endif
 362
 363 #ifdef CONFIG_SCHEDSTATS
 364 /*
 365  * Provides /proc/PID/schedstat
 366  */
 367 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 368 {
 369         return sprintf(buffer, "%llu %llu %lu\n",
 370                         (unsigned long long)task->se.sum_exec_runtime,
 371                         (unsigned long long)task->sched_info.run_delay,
 372                         task->sched_info.pcount);
 373 }
 374 #endif
 375
 376 #ifdef CONFIG_LATENCYTOP
 377 static int lstats_show_proc(struct seq_file *m, void *v)
 378 {
 379         int i;
 380         struct inode *inode = m->private;
 381         struct task_struct *task = get_proc_task(inode);
 382
 383         if (!task)
 384                 return -ESRCH;
 385         seq_puts(m, "Latency Top version : v0.1\n");
 386         for (i = 0; i < 32; i++) {
 387                 if (task->latency_record[i].backtrace[0]) {
 388                         int q;
 389                         seq_printf(m, "%i %li %li ",
 390                                 task->latency_record[i].count,
 391                                 task->latency_record[i].time,
 392                                 task->latency_record[i].max);
 393                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 394                                 char sym[KSYM_SYMBOL_LEN];
 395                                 char *c;
 396                                 if (!task->latency_record[i].backtrace[q])
 397                                         break;
 398                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 399                                         break;
 400                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 401                                 c = strchr(sym, '+');
 402                                 if (c)
 403                                         *c = 0;
 404                                 seq_printf(m, "%s ", sym);
 405                         }
 406                         seq_printf(m, "\n");
 407                 }
 408
 409         }
 410         put_task_struct(task);
 411         return 0;
 412 }
 413
 414 static int lstats_open(struct inode *inode, struct file *file)
 415 {
 416         return single_open(file, lstats_show_proc, inode);
 417 }
 418
 419 static ssize_t lstats_write(struct file *file, const char __user *buf,
 420                             size_t count, loff_t *offs)
 421 {
 422         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 423
 424         if (!task)
 425                 return -ESRCH;
 426         clear_all_latency_tracing(task);
 427         put_task_struct(task);
 428
 429         return count;
 430 }
 431
 432 static const struct file_operations proc_lstats_operations = {
 433         .open           = lstats_open,
 434         .read           = seq_read,
 435         .write          = lstats_write,
 436         .llseek         = seq_lseek,
 437         .release        = single_release,
 438 };
 439
 440 #endif
 441
 442 /* The badness from the OOM killer */
 443 unsigned long badness(struct task_struct *p, unsigned long uptime);
 444 static int proc_oom_score(struct task_struct *task, char *buffer)
 445 {
 446         unsigned long points = 0;
 447         struct timespec uptime;
 448
 449         do_posix_clock_monotonic_gettime(&uptime);
 450         read_lock(&tasklist_lock);
 451         if (pid_alive(task))
 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", "seconds"},
 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         unsigned res = POLLIN | POLLRDNORM;
 653
 654         poll_wait(file, &p->ns->poll, wait);
 655         if (mnt_had_events(p))
 656                 res |= POLLERR | POLLPRI;
 657
 658         return res;
 659 }
 660
 661 static int mounts_open(struct inode *inode, struct file *file)
 662 {
 663         return mounts_open_common(inode, file, &mounts_op);
 664 }
 665
 666 static const struct file_operations proc_mounts_operations = {
 667         .open           = mounts_open,
 668         .read           = seq_read,
 669         .llseek         = seq_lseek,
 670         .release        = mounts_release,
 671         .poll           = mounts_poll,
 672 };
 673
 674 static int mountinfo_open(struct inode *inode, struct file *file)
 675 {
 676         return mounts_open_common(inode, file, &mountinfo_op);
 677 }
 678
 679 static const struct file_operations proc_mountinfo_operations = {
 680         .open           = mountinfo_open,
 681         .read           = seq_read,
 682         .llseek         = seq_lseek,
 683         .release        = mounts_release,
 684         .poll           = mounts_poll,
 685 };
 686
 687 static int mountstats_open(struct inode *inode, struct file *file)
 688 {
 689         return mounts_open_common(inode, file, &mountstats_op);
 690 }
 691
 692 static const struct file_operations proc_mountstats_operations = {
 693         .open           = mountstats_open,
 694         .read           = seq_read,
 695         .llseek         = seq_lseek,
 696         .release        = mounts_release,
 697 };
 698
 699 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 700
 701 static ssize_t proc_info_read(struct file * file, char __user * buf,
 702                           size_t count, loff_t *ppos)
 703 {
 704         struct inode * inode = file->f_path.dentry->d_inode;
 705         unsigned long page;
 706         ssize_t length;
 707         struct task_struct *task = get_proc_task(inode);
 708
 709         length = -ESRCH;
 710         if (!task)
 711                 goto out_no_task;
 712
 713         if (count > PROC_BLOCK_SIZE)
 714                 count = PROC_BLOCK_SIZE;
 715
 716         length = -ENOMEM;
 717         if (!(page = __get_free_page(GFP_TEMPORARY)))
 718                 goto out;
 719
 720         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 721
 722         if (length >= 0)
 723                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 724         free_page(page);
 725 out:
 726         put_task_struct(task);
 727 out_no_task:
 728         return length;
 729 }
 730
 731 static const struct file_operations proc_info_file_operations = {
 732         .read           = proc_info_read,
 733 };
 734
 735 static int proc_single_show(struct seq_file *m, void *v)
 736 {
 737         struct inode *inode = m->private;
 738         struct pid_namespace *ns;
 739         struct pid *pid;
 740         struct task_struct *task;
 741         int ret;
 742
 743         ns = inode->i_sb->s_fs_info;
 744         pid = proc_pid(inode);
 745         task = get_pid_task(pid, PIDTYPE_PID);
 746         if (!task)
 747                 return -ESRCH;
 748
 749         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 750
 751         put_task_struct(task);
 752         return ret;
 753 }
 754
 755 static int proc_single_open(struct inode *inode, struct file *filp)
 756 {
 757         int ret;
 758         ret = single_open(filp, proc_single_show, NULL);
 759         if (!ret) {
 760                 struct seq_file *m = filp->private_data;
 761
 762                 m->private = inode;
 763         }
 764         return ret;
 765 }
 766
 767 static const struct file_operations proc_single_file_operations = {
 768         .open           = proc_single_open,
 769         .read           = seq_read,
 770         .llseek         = seq_lseek,
 771         .release        = single_release,
 772 };
 773
 774 static int mem_open(struct inode* inode, struct file* file)
 775 {
 776         file->private_data = (void*)((long)current->self_exec_id);
 777         return 0;
 778 }
 779
 780 static ssize_t mem_read(struct file * file, char __user * buf,
 781                         size_t count, loff_t *ppos)
 782 {
 783         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 784         char *page;
 785         unsigned long src = *ppos;
 786         int ret = -ESRCH;
 787         struct mm_struct *mm;
 788
 789         if (!task)
 790                 goto out_no_task;
 791
 792         if (check_mem_permission(task))
 793                 goto out;
 794
 795         ret = -ENOMEM;
 796         page = (char *)__get_free_page(GFP_TEMPORARY);
 797         if (!page)
 798                 goto out;
 799
 800         ret = 0;
 801
 802         mm = get_task_mm(task);
 803         if (!mm)
 804                 goto out_free;
 805
 806         ret = -EIO;
 807
 808         if (file->private_data != (void*)((long)current->self_exec_id))
 809                 goto out_put;
 810
 811         ret = 0;
 812
 813         while (count > 0) {
 814                 int this_len, retval;
 815
 816                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 817                 retval = access_process_vm(task, src, page, this_len, 0);
 818                 if (!retval || check_mem_permission(task)) {
 819                         if (!ret)
 820                                 ret = -EIO;
 821                         break;
 822                 }
 823
 824                 if (copy_to_user(buf, page, retval)) {
 825                         ret = -EFAULT;
 826                         break;
 827                 }
 828
 829                 ret += retval;
 830                 src += retval;
 831                 buf += retval;
 832                 count -= retval;
 833         }
 834         *ppos = src;
 835
 836 out_put:
 837         mmput(mm);
 838 out_free:
 839         free_page((unsigned long) page);
 840 out:
 841         put_task_struct(task);
 842 out_no_task:
 843         return ret;
 844 }
 845
 846 #define mem_write NULL
 847
 848 #ifndef mem_write
 849 /* This is a security hazard */
 850 static ssize_t mem_write(struct file * file, const char __user *buf,
 851                          size_t count, loff_t *ppos)
 852 {
 853         int copied;
 854         char *page;
 855         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 856         unsigned long dst = *ppos;
 857
 858         copied = -ESRCH;
 859         if (!task)
 860                 goto out_no_task;
 861
 862         if (check_mem_permission(task))
 863                 goto out;
 864
 865         copied = -ENOMEM;
 866         page = (char *)__get_free_page(GFP_TEMPORARY);
 867         if (!page)
 868                 goto out;
 869
 870         copied = 0;
 871         while (count > 0) {
 872                 int this_len, retval;
 873
 874                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 875                 if (copy_from_user(page, buf, this_len)) {
 876                         copied = -EFAULT;
 877                         break;
 878                 }
 879                 retval = access_process_vm(task, dst, page, this_len, 1);
 880                 if (!retval) {
 881                         if (!copied)
 882                                 copied = -EIO;
 883                         break;
 884                 }
 885                 copied += retval;
 886                 buf += retval;
 887                 dst += retval;
 888                 count -= retval;
 889         }
 890         *ppos = dst;
 891         free_page((unsigned long) page);
 892 out:
 893         put_task_struct(task);
 894 out_no_task:
 895         return copied;
 896 }
 897 #endif
 898
 899 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 900 {
 901         switch (orig) {
 902         case 0:
 903                 file->f_pos = offset;
 904                 break;
 905         case 1:
 906                 file->f_pos += offset;
 907                 break;
 908         default:
 909                 return -EINVAL;
 910         }
 911         force_successful_syscall_return();
 912         return file->f_pos;
 913 }
 914
 915 static const struct file_operations proc_mem_operations = {
 916         .llseek         = mem_lseek,
 917         .read           = mem_read,
 918         .write          = mem_write,
 919         .open           = mem_open,
 920 };
 921
 922 static ssize_t environ_read(struct file *file, char __user *buf,
 923                         size_t count, loff_t *ppos)
 924 {
 925         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 926         char *page;
 927         unsigned long src = *ppos;
 928         int ret = -ESRCH;
 929         struct mm_struct *mm;
 930
 931         if (!task)
 932                 goto out_no_task;
 933
 934         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 935                 goto out;
 936
 937         ret = -ENOMEM;
 938         page = (char *)__get_free_page(GFP_TEMPORARY);
 939         if (!page)
 940                 goto out;
 941
 942         ret = 0;
 943
 944         mm = get_task_mm(task);
 945         if (!mm)
 946                 goto out_free;
 947
 948         while (count > 0) {
 949                 int this_len, retval, max_len;
 950
 951                 this_len = mm->env_end - (mm->env_start + src);
 952
 953                 if (this_len <= 0)
 954                         break;
 955
 956                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 957                 this_len = (this_len > max_len) ? max_len : this_len;
 958
 959                 retval = access_process_vm(task, (mm->env_start + src),
 960                         page, this_len, 0);
 961
 962                 if (retval <= 0) {
 963                         ret = retval;
 964                         break;
 965                 }
 966
 967                 if (copy_to_user(buf, page, retval)) {
 968                         ret = -EFAULT;
 969                         break;
 970                 }
 971
 972                 ret += retval;
 973                 src += retval;
 974                 buf += retval;
 975                 count -= retval;
 976         }
 977         *ppos = src;
 978
 979         mmput(mm);
 980 out_free:
 981         free_page((unsigned long) page);
 982 out:
 983         put_task_struct(task);
 984 out_no_task:
 985         return ret;
 986 }
 987
 988 static const struct file_operations proc_environ_operations = {
 989         .read           = environ_read,
 990 };
 991
 992 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 993                                 size_t count, loff_t *ppos)
 994 {
 995         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 996         char buffer[PROC_NUMBUF];
 997         size_t len;
 998         int oom_adjust = OOM_DISABLE;
 999         unsigned long flags;
1000
1001         if (!task)
1002                 return -ESRCH;
1003
1004         if (lock_task_sighand(task, &flags)) {
1005                 oom_adjust = task->signal->oom_adj;
1006                 unlock_task_sighand(task, &flags);
1007         }
1008
1009         put_task_struct(task);
1010
1011         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1012
1013         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1014 }
1015
1016 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1017                                 size_t count, loff_t *ppos)
1018 {
1019         struct task_struct *task;
1020         char buffer[PROC_NUMBUF];
1021         long oom_adjust;
1022         unsigned long flags;
1023         int err;
1024
1025         memset(buffer, 0, sizeof(buffer));
1026         if (count > sizeof(buffer) - 1)
1027                 count = sizeof(buffer) - 1;
1028         if (copy_from_user(buffer, buf, count))
1029                 return -EFAULT;
1030
1031         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1032         if (err)
1033                 return -EINVAL;
1034         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1035              oom_adjust != OOM_DISABLE)
1036                 return -EINVAL;
1037
1038         task = get_proc_task(file->f_path.dentry->d_inode);
1039         if (!task)
1040                 return -ESRCH;
1041         if (!lock_task_sighand(task, &flags)) {
1042                 put_task_struct(task);
1043                 return -ESRCH;
1044         }
1045
1046         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1047                 unlock_task_sighand(task, &flags);
1048                 put_task_struct(task);
1049                 return -EACCES;
1050         }
1051
1052         task->signal->oom_adj = oom_adjust;
1053
1054         unlock_task_sighand(task, &flags);
1055         put_task_struct(task);
1056
1057         return count;
1058 }
1059
1060 static const struct file_operations proc_oom_adjust_operations = {
1061         .read           = oom_adjust_read,
1062         .write          = oom_adjust_write,
1063 };
1064
1065 #ifdef CONFIG_AUDITSYSCALL
1066 #define TMPBUFLEN 21
1067 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1068                                   size_t count, loff_t *ppos)
1069 {
1070         struct inode * inode = file->f_path.dentry->d_inode;
1071         struct task_struct *task = get_proc_task(inode);
1072         ssize_t length;
1073         char tmpbuf[TMPBUFLEN];
1074
1075         if (!task)
1076                 return -ESRCH;
1077         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1078                                 audit_get_loginuid(task));
1079         put_task_struct(task);
1080         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1081 }
1082
1083 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1084                                    size_t count, loff_t *ppos)
1085 {
1086         struct inode * inode = file->f_path.dentry->d_inode;
1087         char *page, *tmp;
1088         ssize_t length;
1089         uid_t loginuid;
1090
1091         if (!capable(CAP_AUDIT_CONTROL))
1092                 return -EPERM;
1093
1094         rcu_read_lock();
1095         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1096                 rcu_read_unlock();
1097                 return -EPERM;
1098         }
1099         rcu_read_unlock();
1100
1101         if (count >= PAGE_SIZE)
1102                 count = PAGE_SIZE - 1;
1103
1104         if (*ppos != 0) {
1105                 /* No partial writes. */
1106                 return -EINVAL;
1107         }
1108         page = (char*)__get_free_page(GFP_TEMPORARY);
1109         if (!page)
1110                 return -ENOMEM;
1111         length = -EFAULT;
1112         if (copy_from_user(page, buf, count))
1113                 goto out_free_page;
1114
1115         page[count] = '\0';
1116         loginuid = simple_strtoul(page, &tmp, 10);
1117         if (tmp == page) {
1118                 length = -EINVAL;
1119                 goto out_free_page;
1120
1121         }
1122         length = audit_set_loginuid(current, loginuid);
1123         if (likely(length == 0))
1124                 length = count;
1125
1126 out_free_page:
1127         free_page((unsigned long) page);
1128         return length;
1129 }
1130
1131 static const struct file_operations proc_loginuid_operations = {
1132         .read           = proc_loginuid_read,
1133         .write          = proc_loginuid_write,
1134 };
1135
1136 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1137                                   size_t count, loff_t *ppos)
1138 {
1139         struct inode * inode = file->f_path.dentry->d_inode;
1140         struct task_struct *task = get_proc_task(inode);
1141         ssize_t length;
1142         char tmpbuf[TMPBUFLEN];
1143
1144         if (!task)
1145                 return -ESRCH;
1146         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1147                                 audit_get_sessionid(task));
1148         put_task_struct(task);
1149         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1150 }
1151
1152 static const struct file_operations proc_sessionid_operations = {
1153         .read           = proc_sessionid_read,
1154 };
1155 #endif
1156
1157 #ifdef CONFIG_FAULT_INJECTION
1158 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1159                                       size_t count, loff_t *ppos)
1160 {
1161         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1162         char buffer[PROC_NUMBUF];
1163         size_t len;
1164         int make_it_fail;
1165
1166         if (!task)
1167                 return -ESRCH;
1168         make_it_fail = task->make_it_fail;
1169         put_task_struct(task);
1170
1171         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1172
1173         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1174 }
1175
1176 static ssize_t proc_fault_inject_write(struct file * file,
1177                         const char __user * buf, size_t count, loff_t *ppos)
1178 {
1179         struct task_struct *task;
1180         char buffer[PROC_NUMBUF], *end;
1181         int make_it_fail;
1182
1183         if (!capable(CAP_SYS_RESOURCE))
1184                 return -EPERM;
1185         memset(buffer, 0, sizeof(buffer));
1186         if (count > sizeof(buffer) - 1)
1187                 count = sizeof(buffer) - 1;
1188         if (copy_from_user(buffer, buf, count))
1189                 return -EFAULT;
1190         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1191         if (*end)
1192                 return -EINVAL;
1193         task = get_proc_task(file->f_dentry->d_inode);
1194         if (!task)
1195                 return -ESRCH;
1196         task->make_it_fail = make_it_fail;
1197         put_task_struct(task);
1198
1199         return count;
1200 }
1201
1202 static const struct file_operations proc_fault_inject_operations = {
1203         .read           = proc_fault_inject_read,
1204         .write          = proc_fault_inject_write,
1205 };
1206 #endif
1207
1208
1209 #ifdef CONFIG_SCHED_DEBUG
1210 /*
1211  * Print out various scheduling related per-task fields:
1212  */
1213 static int sched_show(struct seq_file *m, void *v)
1214 {
1215         struct inode *inode = m->private;
1216         struct task_struct *p;
1217
1218         p = get_proc_task(inode);
1219         if (!p)
1220                 return -ESRCH;
1221         proc_sched_show_task(p, m);
1222
1223         put_task_struct(p);
1224
1225         return 0;
1226 }
1227
1228 static ssize_t
1229 sched_write(struct file *file, const char __user *buf,
1230             size_t count, loff_t *offset)
1231 {
1232         struct inode *inode = file->f_path.dentry->d_inode;
1233         struct task_struct *p;
1234
1235         p = get_proc_task(inode);
1236         if (!p)
1237                 return -ESRCH;
1238         proc_sched_set_task(p);
1239
1240         put_task_struct(p);
1241
1242         return count;
1243 }
1244
1245 static int sched_open(struct inode *inode, struct file *filp)
1246 {
1247         int ret;
1248
1249         ret = single_open(filp, sched_show, NULL);
1250         if (!ret) {
1251                 struct seq_file *m = filp->private_data;
1252
1253                 m->private = inode;
1254         }
1255         return ret;
1256 }
1257
1258 static const struct file_operations proc_pid_sched_operations = {
1259         .open           = sched_open,
1260         .read           = seq_read,
1261         .write          = sched_write,
1262         .llseek         = seq_lseek,
1263         .release        = single_release,
1264 };
1265
1266 #endif
1267
1268 static ssize_t comm_write(struct file *file, const char __user *buf,
1269                                 size_t count, loff_t *offset)
1270 {
1271         struct inode *inode = file->f_path.dentry->d_inode;
1272         struct task_struct *p;
1273         char buffer[TASK_COMM_LEN];
1274
1275         memset(buffer, 0, sizeof(buffer));
1276         if (count > sizeof(buffer) - 1)
1277                 count = sizeof(buffer) - 1;
1278         if (copy_from_user(buffer, buf, count))
1279                 return -EFAULT;
1280
1281         p = get_proc_task(inode);
1282         if (!p)
1283                 return -ESRCH;
1284
1285         if (same_thread_group(current, p))
1286                 set_task_comm(p, buffer);
1287         else
1288                 count = -EINVAL;
1289
1290         put_task_struct(p);
1291
1292         return count;
1293 }
1294
1295 static int comm_show(struct seq_file *m, void *v)
1296 {
1297         struct inode *inode = m->private;
1298         struct task_struct *p;
1299
1300         p = get_proc_task(inode);
1301         if (!p)
1302                 return -ESRCH;
1303
1304         task_lock(p);
1305         seq_printf(m, "%s\n", p->comm);
1306         task_unlock(p);
1307
1308         put_task_struct(p);
1309
1310         return 0;
1311 }
1312
1313 static int comm_open(struct inode *inode, struct file *filp)
1314 {
1315         int ret;
1316
1317         ret = single_open(filp, comm_show, NULL);
1318         if (!ret) {
1319                 struct seq_file *m = filp->private_data;
1320
1321                 m->private = inode;
1322         }
1323         return ret;
1324 }
1325
1326 static const struct file_operations proc_pid_set_comm_operations = {
1327         .open           = comm_open,
1328         .read           = seq_read,
1329         .write          = comm_write,
1330         .llseek         = seq_lseek,
1331         .release        = single_release,
1332 };
1333
1334 /*
1335  * We added or removed a vma mapping the executable. The vmas are only mapped
1336  * during exec and are not mapped with the mmap system call.
1337  * Callers must hold down_write() on the mm's mmap_sem for these
1338  */
1339 void added_exe_file_vma(struct mm_struct *mm)
1340 {
1341         mm->num_exe_file_vmas++;
1342 }
1343
1344 void removed_exe_file_vma(struct mm_struct *mm)
1345 {
1346         mm->num_exe_file_vmas--;
1347         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1348                 fput(mm->exe_file);
1349                 mm->exe_file = NULL;
1350         }
1351
1352 }
1353
1354 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1355 {
1356         if (new_exe_file)
1357                 get_file(new_exe_file);
1358         if (mm->exe_file)
1359                 fput(mm->exe_file);
1360         mm->exe_file = new_exe_file;
1361         mm->num_exe_file_vmas = 0;
1362 }
1363
1364 struct file *get_mm_exe_file(struct mm_struct *mm)
1365 {
1366         struct file *exe_file;
1367
1368         /* We need mmap_sem to protect against races with removal of
1369          * VM_EXECUTABLE vmas */
1370         down_read(&mm->mmap_sem);
1371         exe_file = mm->exe_file;
1372         if (exe_file)
1373                 get_file(exe_file);
1374         up_read(&mm->mmap_sem);
1375         return exe_file;
1376 }
1377
1378 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1379 {
1380         /* It's safe to write the exe_file pointer without exe_file_lock because
1381          * this is called during fork when the task is not yet in /proc */
1382         newmm->exe_file = get_mm_exe_file(oldmm);
1383 }
1384
1385 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1386 {
1387         struct task_struct *task;
1388         struct mm_struct *mm;
1389         struct file *exe_file;
1390
1391         task = get_proc_task(inode);
1392         if (!task)
1393                 return -ENOENT;
1394         mm = get_task_mm(task);
1395         put_task_struct(task);
1396         if (!mm)
1397                 return -ENOENT;
1398         exe_file = get_mm_exe_file(mm);
1399         mmput(mm);
1400         if (exe_file) {
1401                 *exe_path = exe_file->f_path;
1402                 path_get(&exe_file->f_path);
1403                 fput(exe_file);
1404                 return 0;
1405         } else
1406                 return -ENOENT;
1407 }
1408
1409 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1410 {
1411         struct inode *inode = dentry->d_inode;
1412         int error = -EACCES;
1413
1414         /* We don't need a base pointer in the /proc filesystem */
1415         path_put(&nd->path);
1416
1417         /* Are we allowed to snoop on the tasks file descriptors? */
1418         if (!proc_fd_access_allowed(inode))
1419                 goto out;
1420
1421         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1422 out:
1423         return ERR_PTR(error);
1424 }
1425
1426 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1427 {
1428         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1429         char *pathname;
1430         int len;
1431
1432         if (!tmp)
1433                 return -ENOMEM;
1434
1435         pathname = d_path(path, tmp, PAGE_SIZE);
1436         len = PTR_ERR(pathname);
1437         if (IS_ERR(pathname))
1438                 goto out;
1439         len = tmp + PAGE_SIZE - 1 - pathname;
1440
1441         if (len > buflen)
1442                 len = buflen;
1443         if (copy_to_user(buffer, pathname, len))
1444                 len = -EFAULT;
1445  out:
1446         free_page((unsigned long)tmp);
1447         return len;
1448 }
1449
1450 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1451 {
1452         int error = -EACCES;
1453         struct inode *inode = dentry->d_inode;
1454         struct path path;
1455
1456         /* Are we allowed to snoop on the tasks file descriptors? */
1457         if (!proc_fd_access_allowed(inode))
1458                 goto out;
1459
1460         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1461         if (error)
1462                 goto out;
1463
1464         error = do_proc_readlink(&path, buffer, buflen);
1465         path_put(&path);
1466 out:
1467         return error;
1468 }
1469
1470 static const struct inode_operations proc_pid_link_inode_operations = {
1471         .readlink       = proc_pid_readlink,
1472         .follow_link    = proc_pid_follow_link,
1473         .setattr        = proc_setattr,
1474 };
1475
1476
1477 /* building an inode */
1478
1479 static int task_dumpable(struct task_struct *task)
1480 {
1481         int dumpable = 0;
1482         struct mm_struct *mm;
1483
1484         task_lock(task);
1485         mm = task->mm;
1486         if (mm)
1487                 dumpable = get_dumpable(mm);
1488         task_unlock(task);
1489         if(dumpable == 1)
1490                 return 1;
1491         return 0;
1492 }
1493
1494
1495 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1496 {
1497         struct inode * inode;
1498         struct proc_inode *ei;
1499         const struct cred *cred;
1500
1501         /* We need a new inode */
1502
1503         inode = new_inode(sb);
1504         if (!inode)
1505                 goto out;
1506
1507         /* Common stuff */
1508         ei = PROC_I(inode);
1509         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1510         inode->i_op = &proc_def_inode_operations;
1511
1512         /*
1513          * grab the reference to task.
1514          */
1515         ei->pid = get_task_pid(task, PIDTYPE_PID);
1516         if (!ei->pid)
1517                 goto out_unlock;
1518
1519         if (task_dumpable(task)) {
1520                 rcu_read_lock();
1521                 cred = __task_cred(task);
1522                 inode->i_uid = cred->euid;
1523                 inode->i_gid = cred->egid;
1524                 rcu_read_unlock();
1525         }
1526         security_task_to_inode(task, inode);
1527
1528 out:
1529         return inode;
1530
1531 out_unlock:
1532         iput(inode);
1533         return NULL;
1534 }
1535
1536 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1537 {
1538         struct inode *inode = dentry->d_inode;
1539         struct task_struct *task;
1540         const struct cred *cred;
1541
1542         generic_fillattr(inode, stat);
1543
1544         rcu_read_lock();
1545         stat->uid = 0;
1546         stat->gid = 0;
1547         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1548         if (task) {
1549                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1550                     task_dumpable(task)) {
1551                         cred = __task_cred(task);
1552                         stat->uid = cred->euid;
1553                         stat->gid = cred->egid;
1554                 }
1555         }
1556         rcu_read_unlock();
1557         return 0;
1558 }
1559
1560 /* dentry stuff */
1561
1562 /*
1563  *      Exceptional case: normally we are not allowed to unhash a busy
1564  * directory. In this case, however, we can do it - no aliasing problems
1565  * due to the way we treat inodes.
1566  *
1567  * Rewrite the inode's ownerships here because the owning task may have
1568  * performed a setuid(), etc.
1569  *
1570  * Before the /proc/pid/status file was created the only way to read
1571  * the effective uid of a /process was to stat /proc/pid.  Reading
1572  * /proc/pid/status is slow enough that procps and other packages
1573  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1574  * made this apply to all per process world readable and executable
1575  * directories.
1576  */
1577 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1578 {
1579         struct inode *inode = dentry->d_inode;
1580         struct task_struct *task = get_proc_task(inode);
1581         const struct cred *cred;
1582
1583         if (task) {
1584                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1585                     task_dumpable(task)) {
1586                         rcu_read_lock();
1587                         cred = __task_cred(task);
1588                         inode->i_uid = cred->euid;
1589                         inode->i_gid = cred->egid;
1590                         rcu_read_unlock();
1591                 } else {
1592                         inode->i_uid = 0;
1593                         inode->i_gid = 0;
1594                 }
1595                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1596                 security_task_to_inode(task, inode);
1597                 put_task_struct(task);
1598                 return 1;
1599         }
1600         d_drop(dentry);
1601         return 0;
1602 }
1603
1604 static int pid_delete_dentry(struct dentry * dentry)
1605 {
1606         /* Is the task we represent dead?
1607          * If so, then don't put the dentry on the lru list,
1608          * kill it immediately.
1609          */
1610         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1611 }
1612
1613 static const struct dentry_operations pid_dentry_operations =
1614 {
1615         .d_revalidate   = pid_revalidate,
1616         .d_delete       = pid_delete_dentry,
1617 };
1618
1619 /* Lookups */
1620
1621 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1622                                 struct task_struct *, const void *);
1623
1624 /*
1625  * Fill a directory entry.
1626  *
1627  * If possible create the dcache entry and derive our inode number and
1628  * file type from dcache entry.
1629  *
1630  * Since all of the proc inode numbers are dynamically generated, the inode
1631  * numbers do not exist until the inode is cache.  This means creating the
1632  * the dcache entry in readdir is necessary to keep the inode numbers
1633  * reported by readdir in sync with the inode numbers reported
1634  * by stat.
1635  */
1636 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1637         char *name, int len,
1638         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1639 {
1640         struct dentry *child, *dir = filp->f_path.dentry;
1641         struct inode *inode;
1642         struct qstr qname;
1643         ino_t ino = 0;
1644         unsigned type = DT_UNKNOWN;
1645
1646         qname.name = name;
1647         qname.len  = len;
1648         qname.hash = full_name_hash(name, len);
1649
1650         child = d_lookup(dir, &qname);
1651         if (!child) {
1652                 struct dentry *new;
1653                 new = d_alloc(dir, &qname);
1654                 if (new) {
1655                         child = instantiate(dir->d_inode, new, task, ptr);
1656                         if (child)
1657                                 dput(new);
1658                         else
1659                                 child = new;
1660                 }
1661         }
1662         if (!child || IS_ERR(child) || !child->d_inode)
1663                 goto end_instantiate;
1664         inode = child->d_inode;
1665         if (inode) {
1666                 ino = inode->i_ino;
1667                 type = inode->i_mode >> 12;
1668         }
1669         dput(child);
1670 end_instantiate:
1671         if (!ino)
1672                 ino = find_inode_number(dir, &qname);
1673         if (!ino)
1674                 ino = 1;
1675         return filldir(dirent, name, len, filp->f_pos, ino, type);
1676 }
1677
1678 static unsigned name_to_int(struct dentry *dentry)
1679 {
1680         const char *name = dentry->d_name.name;
1681         int len = dentry->d_name.len;
1682         unsigned n = 0;
1683
1684         if (len > 1 && *name == '0')
1685                 goto out;
1686         while (len-- > 0) {
1687                 unsigned c = *name++ - '0';
1688                 if (c > 9)
1689                         goto out;
1690                 if (n >= (~0U-9)/10)
1691                         goto out;
1692                 n *= 10;
1693                 n += c;
1694         }
1695         return n;
1696 out:
1697         return ~0U;
1698 }
1699
1700 #define PROC_FDINFO_MAX 64
1701
1702 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1703 {
1704         struct task_struct *task = get_proc_task(inode);
1705         struct files_struct *files = NULL;
1706         struct file *file;
1707         int fd = proc_fd(inode);
1708
1709         if (task) {
1710                 files = get_files_struct(task);
1711                 put_task_struct(task);
1712         }
1713         if (files) {
1714                 /*
1715                  * We are not taking a ref to the file structure, so we must
1716                  * hold ->file_lock.
1717                  */
1718                 spin_lock(&files->file_lock);
1719                 file = fcheck_files(files, fd);
1720                 if (file) {
1721                         if (path) {
1722                                 *path = file->f_path;
1723                                 path_get(&file->f_path);
1724                         }
1725                         if (info)
1726                                 snprintf(info, PROC_FDINFO_MAX,
1727                                          "pos:\t%lli\n"
1728                                          "flags:\t0%o\n",
1729                                          (long long) file->f_pos,
1730                                          file->f_flags);
1731                         spin_unlock(&files->file_lock);
1732                         put_files_struct(files);
1733                         return 0;
1734                 }
1735                 spin_unlock(&files->file_lock);
1736                 put_files_struct(files);
1737         }
1738         return -ENOENT;
1739 }
1740
1741 static int proc_fd_link(struct inode *inode, struct path *path)
1742 {
1743         return proc_fd_info(inode, path, NULL);
1744 }
1745
1746 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1747 {
1748         struct inode *inode = dentry->d_inode;
1749         struct task_struct *task = get_proc_task(inode);
1750         int fd = proc_fd(inode);
1751         struct files_struct *files;
1752         const struct cred *cred;
1753
1754         if (task) {
1755                 files = get_files_struct(task);
1756                 if (files) {
1757                         rcu_read_lock();
1758                         if (fcheck_files(files, fd)) {
1759                                 rcu_read_unlock();
1760                                 put_files_struct(files);
1761                                 if (task_dumpable(task)) {
1762                                         rcu_read_lock();
1763                                         cred = __task_cred(task);
1764                                         inode->i_uid = cred->euid;
1765                                         inode->i_gid = cred->egid;
1766                                         rcu_read_unlock();
1767                                 } else {
1768                                         inode->i_uid = 0;
1769                                         inode->i_gid = 0;
1770                                 }
1771                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1772                                 security_task_to_inode(task, inode);
1773                                 put_task_struct(task);
1774                                 return 1;
1775                         }
1776                         rcu_read_unlock();
1777                         put_files_struct(files);
1778                 }
1779                 put_task_struct(task);
1780         }
1781         d_drop(dentry);
1782         return 0;
1783 }
1784
1785 static const struct dentry_operations tid_fd_dentry_operations =
1786 {
1787         .d_revalidate   = tid_fd_revalidate,
1788         .d_delete       = pid_delete_dentry,
1789 };
1790
1791 static struct dentry *proc_fd_instantiate(struct inode *dir,
1792         struct dentry *dentry, struct task_struct *task, const void *ptr)
1793 {
1794         unsigned fd = *(const unsigned *)ptr;
1795         struct file *file;
1796         struct files_struct *files;
1797         struct inode *inode;
1798         struct proc_inode *ei;
1799         struct dentry *error = ERR_PTR(-ENOENT);
1800
1801         inode = proc_pid_make_inode(dir->i_sb, task);
1802         if (!inode)
1803                 goto out;
1804         ei = PROC_I(inode);
1805         ei->fd = fd;
1806         files = get_files_struct(task);
1807         if (!files)
1808                 goto out_iput;
1809         inode->i_mode = S_IFLNK;
1810
1811         /*
1812          * We are not taking a ref to the file structure, so we must
1813          * hold ->file_lock.
1814          */
1815         spin_lock(&files->file_lock);
1816         file = fcheck_files(files, fd);
1817         if (!file)
1818                 goto out_unlock;
1819         if (file->f_mode & FMODE_READ)
1820                 inode->i_mode |= S_IRUSR | S_IXUSR;
1821         if (file->f_mode & FMODE_WRITE)
1822                 inode->i_mode |= S_IWUSR | S_IXUSR;
1823         spin_unlock(&files->file_lock);
1824         put_files_struct(files);
1825
1826         inode->i_op = &proc_pid_link_inode_operations;
1827         inode->i_size = 64;
1828         ei->op.proc_get_link = proc_fd_link;
1829         dentry->d_op = &tid_fd_dentry_operations;
1830         d_add(dentry, inode);
1831         /* Close the race of the process dying before we return the dentry */
1832         if (tid_fd_revalidate(dentry, NULL))
1833                 error = NULL;
1834
1835  out:
1836         return error;
1837 out_unlock:
1838         spin_unlock(&files->file_lock);
1839         put_files_struct(files);
1840 out_iput:
1841         iput(inode);
1842         goto out;
1843 }
1844
1845 static struct dentry *proc_lookupfd_common(struct inode *dir,
1846                                            struct dentry *dentry,
1847                                            instantiate_t instantiate)
1848 {
1849         struct task_struct *task = get_proc_task(dir);
1850         unsigned fd = name_to_int(dentry);
1851         struct dentry *result = ERR_PTR(-ENOENT);
1852
1853         if (!task)
1854                 goto out_no_task;
1855         if (fd == ~0U)
1856                 goto out;
1857
1858         result = instantiate(dir, dentry, task, &fd);
1859 out:
1860         put_task_struct(task);
1861 out_no_task:
1862         return result;
1863 }
1864
1865 static int proc_readfd_common(struct file * filp, void * dirent,
1866                               filldir_t filldir, instantiate_t instantiate)
1867 {
1868         struct dentry *dentry = filp->f_path.dentry;
1869         struct inode *inode = dentry->d_inode;
1870         struct task_struct *p = get_proc_task(inode);
1871         unsigned int fd, ino;
1872         int retval;
1873         struct files_struct * files;
1874
1875         retval = -ENOENT;
1876         if (!p)
1877                 goto out_no_task;
1878         retval = 0;
1879
1880         fd = filp->f_pos;
1881         switch (fd) {
1882                 case 0:
1883                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1884                                 goto out;
1885                         filp->f_pos++;
1886                 case 1:
1887                         ino = parent_ino(dentry);
1888                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1889                                 goto out;
1890                         filp->f_pos++;
1891                 default:
1892                         files = get_files_struct(p);
1893                         if (!files)
1894                                 goto out;
1895                         rcu_read_lock();
1896                         for (fd = filp->f_pos-2;
1897                              fd < files_fdtable(files)->max_fds;
1898                              fd++, filp->f_pos++) {
1899                                 char name[PROC_NUMBUF];
1900                                 int len;
1901
1902                                 if (!fcheck_files(files, fd))
1903                                         continue;
1904                                 rcu_read_unlock();
1905
1906                                 len = snprintf(name, sizeof(name), "%d", fd);
1907                                 if (proc_fill_cache(filp, dirent, filldir,
1908                                                     name, len, instantiate,
1909                                                     p, &fd) < 0) {
1910                                         rcu_read_lock();
1911                                         break;
1912                                 }
1913                                 rcu_read_lock();
1914                         }
1915                         rcu_read_unlock();
1916                         put_files_struct(files);
1917         }
1918 out:
1919         put_task_struct(p);
1920 out_no_task:
1921         return retval;
1922 }
1923
1924 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1925                                     struct nameidata *nd)
1926 {
1927         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1928 }
1929
1930 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1931 {
1932         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1933 }
1934
1935 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1936                                       size_t len, loff_t *ppos)
1937 {
1938         char tmp[PROC_FDINFO_MAX];
1939         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1940         if (!err)
1941                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1942         return err;
1943 }
1944
1945 static const struct file_operations proc_fdinfo_file_operations = {
1946         .open           = nonseekable_open,
1947         .read           = proc_fdinfo_read,
1948 };
1949
1950 static const struct file_operations proc_fd_operations = {
1951         .read           = generic_read_dir,
1952         .readdir        = proc_readfd,
1953 };
1954
1955 /*
1956  * /proc/pid/fd needs a special permission handler so that a process can still
1957  * access /proc/self/fd after it has executed a setuid().
1958  */
1959 static int proc_fd_permission(struct inode *inode, int mask)
1960 {
1961         int rv;
1962
1963         rv = generic_permission(inode, mask, NULL);
1964         if (rv == 0)
1965                 return 0;
1966         if (task_pid(current) == proc_pid(inode))
1967                 rv = 0;
1968         return rv;
1969 }
1970
1971 /*
1972  * proc directories can do almost nothing..
1973  */
1974 static const struct inode_operations proc_fd_inode_operations = {
1975         .lookup         = proc_lookupfd,
1976         .permission     = proc_fd_permission,
1977         .setattr        = proc_setattr,
1978 };
1979
1980 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1981         struct dentry *dentry, struct task_struct *task, const void *ptr)
1982 {
1983         unsigned fd = *(unsigned *)ptr;
1984         struct inode *inode;
1985         struct proc_inode *ei;
1986         struct dentry *error = ERR_PTR(-ENOENT);
1987
1988         inode = proc_pid_make_inode(dir->i_sb, task);
1989         if (!inode)
1990                 goto out;
1991         ei = PROC_I(inode);
1992         ei->fd = fd;
1993         inode->i_mode = S_IFREG | S_IRUSR;
1994         inode->i_fop = &proc_fdinfo_file_operations;
1995         dentry->d_op = &tid_fd_dentry_operations;
1996         d_add(dentry, inode);
1997         /* Close the race of the process dying before we return the dentry */
1998         if (tid_fd_revalidate(dentry, NULL))
1999                 error = NULL;
2000
2001  out:
2002         return error;
2003 }
2004
2005 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2006                                         struct dentry *dentry,
2007                                         struct nameidata *nd)
2008 {
2009         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2010 }
2011
2012 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2013 {
2014         return proc_readfd_common(filp, dirent, filldir,
2015                                   proc_fdinfo_instantiate);
2016 }
2017
2018 static const struct file_operations proc_fdinfo_operations = {
2019         .read           = generic_read_dir,
2020         .readdir        = proc_readfdinfo,
2021 };
2022
2023 /*
2024  * proc directories can do almost nothing..
2025  */
2026 static const struct inode_operations proc_fdinfo_inode_operations = {
2027         .lookup         = proc_lookupfdinfo,
2028         .setattr        = proc_setattr,
2029 };
2030
2031
2032 static struct dentry *proc_pident_instantiate(struct inode *dir,
2033         struct dentry *dentry, struct task_struct *task, const void *ptr)
2034 {
2035         const struct pid_entry *p = ptr;
2036         struct inode *inode;
2037         struct proc_inode *ei;
2038         struct dentry *error = ERR_PTR(-ENOENT);
2039
2040         inode = proc_pid_make_inode(dir->i_sb, task);
2041         if (!inode)
2042                 goto out;
2043
2044         ei = PROC_I(inode);
2045         inode->i_mode = p->mode;
2046         if (S_ISDIR(inode->i_mode))
2047                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2048         if (p->iop)
2049                 inode->i_op = p->iop;
2050         if (p->fop)
2051                 inode->i_fop = p->fop;
2052         ei->op = p->op;
2053         dentry->d_op = &pid_dentry_operations;
2054         d_add(dentry, inode);
2055         /* Close the race of the process dying before we return the dentry */
2056         if (pid_revalidate(dentry, NULL))
2057                 error = NULL;
2058 out:
2059         return error;
2060 }
2061
2062 static struct dentry *proc_pident_lookup(struct inode *dir,
2063                                          struct dentry *dentry,
2064                                          const struct pid_entry *ents,
2065                                          unsigned int nents)
2066 {
2067         struct dentry *error;
2068         struct task_struct *task = get_proc_task(dir);
2069         const struct pid_entry *p, *last;
2070
2071         error = ERR_PTR(-ENOENT);
2072
2073         if (!task)
2074                 goto out_no_task;
2075
2076         /*
2077          * Yes, it does not scale. And it should not. Don't add
2078          * new entries into /proc/<tgid>/ without very good reasons.
2079          */
2080         last = &ents[nents - 1];
2081         for (p = ents; p <= last; p++) {
2082                 if (p->len != dentry->d_name.len)
2083                         continue;
2084                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2085                         break;
2086         }
2087         if (p > last)
2088                 goto out;
2089
2090         error = proc_pident_instantiate(dir, dentry, task, p);
2091 out:
2092         put_task_struct(task);
2093 out_no_task:
2094         return error;
2095 }
2096
2097 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2098         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2099 {
2100         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2101                                 proc_pident_instantiate, task, p);
2102 }
2103
2104 static int proc_pident_readdir(struct file *filp,
2105                 void *dirent, filldir_t filldir,
2106                 const struct pid_entry *ents, unsigned int nents)
2107 {
2108         int i;
2109         struct dentry *dentry = filp->f_path.dentry;
2110         struct inode *inode = dentry->d_inode;
2111         struct task_struct *task = get_proc_task(inode);
2112         const struct pid_entry *p, *last;
2113         ino_t ino;
2114         int ret;
2115
2116         ret = -ENOENT;
2117         if (!task)
2118                 goto out_no_task;
2119
2120         ret = 0;
2121         i = filp->f_pos;
2122         switch (i) {
2123         case 0:
2124                 ino = inode->i_ino;
2125                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2126                         goto out;
2127                 i++;
2128                 filp->f_pos++;
2129                 /* fall through */
2130         case 1:
2131                 ino = parent_ino(dentry);
2132                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2133                         goto out;
2134                 i++;
2135                 filp->f_pos++;
2136                 /* fall through */
2137         default:
2138                 i -= 2;
2139                 if (i >= nents) {
2140                         ret = 1;
2141                         goto out;
2142                 }
2143                 p = ents + i;
2144                 last = &ents[nents - 1];
2145                 while (p <= last) {
2146                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2147                                 goto out;
2148                         filp->f_pos++;
2149                         p++;
2150                 }
2151         }
2152
2153         ret = 1;
2154 out:
2155         put_task_struct(task);
2156 out_no_task:
2157         return ret;
2158 }
2159
2160 #ifdef CONFIG_SECURITY
2161 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2162                                   size_t count, loff_t *ppos)
2163 {
2164         struct inode * inode = file->f_path.dentry->d_inode;
2165         char *p = NULL;
2166         ssize_t length;
2167         struct task_struct *task = get_proc_task(inode);
2168
2169         if (!task)
2170                 return -ESRCH;
2171
2172         length = security_getprocattr(task,
2173                                       (char*)file->f_path.dentry->d_name.name,
2174                                       &p);
2175         put_task_struct(task);
2176         if (length > 0)
2177                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2178         kfree(p);
2179         return length;
2180 }
2181
2182 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2183                                    size_t count, loff_t *ppos)
2184 {
2185         struct inode * inode = file->f_path.dentry->d_inode;
2186         char *page;
2187         ssize_t length;
2188         struct task_struct *task = get_proc_task(inode);
2189
2190         length = -ESRCH;
2191         if (!task)
2192                 goto out_no_task;
2193         if (count > PAGE_SIZE)
2194                 count = PAGE_SIZE;
2195
2196         /* No partial writes. */
2197         length = -EINVAL;
2198         if (*ppos != 0)
2199                 goto out;
2200
2201         length = -ENOMEM;
2202         page = (char*)__get_free_page(GFP_TEMPORARY);
2203         if (!page)
2204                 goto out;
2205
2206         length = -EFAULT;
2207         if (copy_from_user(page, buf, count))
2208                 goto out_free;
2209
2210         /* Guard against adverse ptrace interaction */
2211         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2212         if (length < 0)
2213                 goto out_free;
2214
2215         length = security_setprocattr(task,
2216                                       (char*)file->f_path.dentry->d_name.name,
2217                                       (void*)page, count);
2218         mutex_unlock(&task->cred_guard_mutex);
2219 out_free:
2220         free_page((unsigned long) page);
2221 out:
2222         put_task_struct(task);
2223 out_no_task:
2224         return length;
2225 }
2226
2227 static const struct file_operations proc_pid_attr_operations = {
2228         .read           = proc_pid_attr_read,
2229         .write          = proc_pid_attr_write,
2230 };
2231
2232 static const struct pid_entry attr_dir_stuff[] = {
2233         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2234         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2235         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2236         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2237         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2238         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2239 };
2240
2241 static int proc_attr_dir_readdir(struct file * filp,
2242                              void * dirent, filldir_t filldir)
2243 {
2244         return proc_pident_readdir(filp,dirent,filldir,
2245                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2246 }
2247
2248 static const struct file_operations proc_attr_dir_operations = {
2249         .read           = generic_read_dir,
2250         .readdir        = proc_attr_dir_readdir,
2251 };
2252
2253 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2254                                 struct dentry *dentry, struct nameidata *nd)
2255 {
2256         return proc_pident_lookup(dir, dentry,
2257                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2258 }
2259
2260 static const struct inode_operations proc_attr_dir_inode_operations = {
2261         .lookup         = proc_attr_dir_lookup,
2262         .getattr        = pid_getattr,
2263         .setattr        = proc_setattr,
2264 };
2265
2266 #endif
2267
2268 #ifdef CONFIG_ELF_CORE
2269 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2270                                          size_t count, loff_t *ppos)
2271 {
2272         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2273         struct mm_struct *mm;
2274         char buffer[PROC_NUMBUF];
2275         size_t len;
2276         int ret;
2277
2278         if (!task)
2279                 return -ESRCH;
2280
2281         ret = 0;
2282         mm = get_task_mm(task);
2283         if (mm) {
2284                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2285                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2286                                 MMF_DUMP_FILTER_SHIFT));
2287                 mmput(mm);
2288                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2289         }
2290
2291         put_task_struct(task);
2292
2293         return ret;
2294 }
2295
2296 static ssize_t proc_coredump_filter_write(struct file *file,
2297                                           const char __user *buf,
2298                                           size_t count,
2299                                           loff_t *ppos)
2300 {
2301         struct task_struct *task;
2302         struct mm_struct *mm;
2303         char buffer[PROC_NUMBUF], *end;
2304         unsigned int val;
2305         int ret;
2306         int i;
2307         unsigned long mask;
2308
2309         ret = -EFAULT;
2310         memset(buffer, 0, sizeof(buffer));
2311         if (count > sizeof(buffer) - 1)
2312                 count = sizeof(buffer) - 1;
2313         if (copy_from_user(buffer, buf, count))
2314                 goto out_no_task;
2315
2316         ret = -EINVAL;
2317         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2318         if (*end == '\n')
2319                 end++;
2320         if (end - buffer == 0)
2321                 goto out_no_task;
2322
2323         ret = -ESRCH;
2324         task = get_proc_task(file->f_dentry->d_inode);
2325         if (!task)
2326                 goto out_no_task;
2327
2328         ret = end - buffer;
2329         mm = get_task_mm(task);
2330         if (!mm)
2331                 goto out_no_mm;
2332
2333         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2334                 if (val & mask)
2335                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2336                 else
2337                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2338         }
2339
2340         mmput(mm);
2341  out_no_mm:
2342         put_task_struct(task);
2343  out_no_task:
2344         return ret;
2345 }
2346
2347 static const struct file_operations proc_coredump_filter_operations = {
2348         .read           = proc_coredump_filter_read,
2349         .write          = proc_coredump_filter_write,
2350 };
2351 #endif
2352
2353 /*
2354  * /proc/self:
2355  */
2356 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2357                               int buflen)
2358 {
2359         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2360         pid_t tgid = task_tgid_nr_ns(current, ns);
2361         char tmp[PROC_NUMBUF];
2362         if (!tgid)
2363                 return -ENOENT;
2364         sprintf(tmp, "%d", tgid);
2365         return vfs_readlink(dentry,buffer,buflen,tmp);
2366 }
2367
2368 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2369 {
2370         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2371         pid_t tgid = task_tgid_nr_ns(current, ns);
2372         char *name = ERR_PTR(-ENOENT);
2373         if (tgid) {
2374                 name = __getname();
2375                 if (!name)
2376                         name = ERR_PTR(-ENOMEM);
2377                 else
2378                         sprintf(name, "%d", tgid);
2379         }
2380         nd_set_link(nd, name);
2381         return NULL;
2382 }
2383
2384 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2385                                 void *cookie)
2386 {
2387         char *s = nd_get_link(nd);
2388         if (!IS_ERR(s))
2389                 __putname(s);
2390 }
2391
2392 static const struct inode_operations proc_self_inode_operations = {
2393         .readlink       = proc_self_readlink,
2394         .follow_link    = proc_self_follow_link,
2395         .put_link       = proc_self_put_link,
2396 };
2397
2398 /*
2399  * proc base
2400  *
2401  * These are the directory entries in the root directory of /proc
2402  * that properly belong to the /proc filesystem, as they describe
2403  * describe something that is process related.
2404  */
2405 static const struct pid_entry proc_base_stuff[] = {
2406         NOD("self", S_IFLNK|S_IRWXUGO,
2407                 &proc_self_inode_operations, NULL, {}),
2408 };
2409
2410 /*
2411  *      Exceptional case: normally we are not allowed to unhash a busy
2412  * directory. In this case, however, we can do it - no aliasing problems
2413  * due to the way we treat inodes.
2414  */
2415 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2416 {
2417         struct inode *inode = dentry->d_inode;
2418         struct task_struct *task = get_proc_task(inode);
2419         if (task) {
2420                 put_task_struct(task);
2421                 return 1;
2422         }
2423         d_drop(dentry);
2424         return 0;
2425 }
2426
2427 static const struct dentry_operations proc_base_dentry_operations =
2428 {
2429         .d_revalidate   = proc_base_revalidate,
2430         .d_delete       = pid_delete_dentry,
2431 };
2432
2433 static struct dentry *proc_base_instantiate(struct inode *dir,
2434         struct dentry *dentry, struct task_struct *task, const void *ptr)
2435 {
2436         const struct pid_entry *p = ptr;
2437         struct inode *inode;
2438         struct proc_inode *ei;
2439         struct dentry *error = ERR_PTR(-EINVAL);
2440
2441         /* Allocate the inode */
2442         error = ERR_PTR(-ENOMEM);
2443         inode = new_inode(dir->i_sb);
2444         if (!inode)
2445                 goto out;
2446
2447         /* Initialize the inode */
2448         ei = PROC_I(inode);
2449         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2450
2451         /*
2452          * grab the reference to the task.
2453          */
2454         ei->pid = get_task_pid(task, PIDTYPE_PID);
2455         if (!ei->pid)
2456                 goto out_iput;
2457
2458         inode->i_mode = p->mode;
2459         if (S_ISDIR(inode->i_mode))
2460                 inode->i_nlink = 2;
2461         if (S_ISLNK(inode->i_mode))
2462                 inode->i_size = 64;
2463         if (p->iop)
2464                 inode->i_op = p->iop;
2465         if (p->fop)
2466                 inode->i_fop = p->fop;
2467         ei->op = p->op;
2468         dentry->d_op = &proc_base_dentry_operations;
2469         d_add(dentry, inode);
2470         error = NULL;
2471 out:
2472         return error;
2473 out_iput:
2474         iput(inode);
2475         goto out;
2476 }
2477
2478 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2479 {
2480         struct dentry *error;
2481         struct task_struct *task = get_proc_task(dir);
2482         const struct pid_entry *p, *last;
2483
2484         error = ERR_PTR(-ENOENT);
2485
2486         if (!task)
2487                 goto out_no_task;
2488
2489         /* Lookup the directory entry */
2490         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2491         for (p = proc_base_stuff; p <= last; p++) {
2492                 if (p->len != dentry->d_name.len)
2493                         continue;
2494                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2495                         break;
2496         }
2497         if (p > last)
2498                 goto out;
2499
2500         error = proc_base_instantiate(dir, dentry, task, p);
2501
2502 out:
2503         put_task_struct(task);
2504 out_no_task:
2505         return error;
2506 }
2507
2508 static int proc_base_fill_cache(struct file *filp, void *dirent,
2509         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2510 {
2511         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2512                                 proc_base_instantiate, task, p);
2513 }
2514
2515 #ifdef CONFIG_TASK_IO_ACCOUNTING
2516 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2517 {
2518         struct task_io_accounting acct = task->ioac;
2519         unsigned long flags;
2520
2521         if (whole && lock_task_sighand(task, &flags)) {
2522                 struct task_struct *t = task;
2523
2524                 task_io_accounting_add(&acct, &task->signal->ioac);
2525                 while_each_thread(task, t)
2526                         task_io_accounting_add(&acct, &t->ioac);
2527
2528                 unlock_task_sighand(task, &flags);
2529         }
2530         return sprintf(buffer,
2531                         "rchar: %llu\n"
2532                         "wchar: %llu\n"
2533                         "syscr: %llu\n"
2534                         "syscw: %llu\n"
2535                         "read_bytes: %llu\n"
2536                         "write_bytes: %llu\n"
2537                         "cancelled_write_bytes: %llu\n",
2538                         (unsigned long long)acct.rchar,
2539                         (unsigned long long)acct.wchar,
2540                         (unsigned long long)acct.syscr,
2541                         (unsigned long long)acct.syscw,
2542                         (unsigned long long)acct.read_bytes,
2543                         (unsigned long long)acct.write_bytes,
2544                         (unsigned long long)acct.cancelled_write_bytes);
2545 }
2546
2547 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2548 {
2549         return do_io_accounting(task, buffer, 0);
2550 }
2551
2552 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2553 {
2554         return do_io_accounting(task, buffer, 1);
2555 }
2556 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2557
2558 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2559                                 struct pid *pid, struct task_struct *task)
2560 {
2561         seq_printf(m, "%08x\n", task->personality);
2562         return 0;
2563 }
2564
2565 /*
2566  * Thread groups
2567  */
2568 static const struct file_operations proc_task_operations;
2569 static const struct inode_operations proc_task_inode_operations;
2570
2571 static const struct pid_entry tgid_base_stuff[] = {
2572         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2573         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2574         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2575 #ifdef CONFIG_NET
2576         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2577 #endif
2578         REG("environ",    S_IRUSR, proc_environ_operations),
2579         INF("auxv",       S_IRUSR, proc_pid_auxv),
2580         ONE("status",     S_IRUGO, proc_pid_status),
2581         ONE("personality", S_IRUSR, proc_pid_personality),
2582         INF("limits",     S_IRUSR, proc_pid_limits),
2583 #ifdef CONFIG_SCHED_DEBUG
2584         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2585 #endif
2586         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2587 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2588         INF("syscall",    S_IRUSR, proc_pid_syscall),
2589 #endif
2590         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2591         ONE("stat",       S_IRUGO, proc_tgid_stat),
2592         ONE("statm",      S_IRUGO, proc_pid_statm),
2593         REG("maps",       S_IRUGO, proc_maps_operations),
2594 #ifdef CONFIG_NUMA
2595         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2596 #endif
2597         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2598         LNK("cwd",        proc_cwd_link),
2599         LNK("root",       proc_root_link),
2600         LNK("exe",        proc_exe_link),
2601         REG("mounts",     S_IRUGO, proc_mounts_operations),
2602         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2603         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2604 #ifdef CONFIG_PROC_PAGE_MONITOR
2605         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2606         REG("smaps",      S_IRUGO, proc_smaps_operations),
2607         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2608 #endif
2609 #ifdef CONFIG_SECURITY
2610         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2611 #endif
2612 #ifdef CONFIG_KALLSYMS
2613         INF("wchan",      S_IRUGO, proc_pid_wchan),
2614 #endif
2615 #ifdef CONFIG_STACKTRACE
2616         ONE("stack",      S_IRUSR, proc_pid_stack),
2617 #endif
2618 #ifdef CONFIG_SCHEDSTATS
2619         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2620 #endif
2621 #ifdef CONFIG_LATENCYTOP
2622         REG("latency",  S_IRUGO, proc_lstats_operations),
2623 #endif
2624 #ifdef CONFIG_PROC_PID_CPUSET
2625         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2626 #endif
2627 #ifdef CONFIG_CGROUPS
2628         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2629 #endif
2630         INF("oom_score",  S_IRUGO, proc_oom_score),
2631         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2632 #ifdef CONFIG_AUDITSYSCALL
2633         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2634         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2635 #endif
2636 #ifdef CONFIG_FAULT_INJECTION
2637         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2638 #endif
2639 #ifdef CONFIG_ELF_CORE
2640         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2641 #endif
2642 #ifdef CONFIG_TASK_IO_ACCOUNTING
2643         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2644 #endif
2645 };
2646
2647 static int proc_tgid_base_readdir(struct file * filp,
2648                              void * dirent, filldir_t filldir)
2649 {
2650         return proc_pident_readdir(filp,dirent,filldir,
2651                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2652 }
2653
2654 static const struct file_operations proc_tgid_base_operations = {
2655         .read           = generic_read_dir,
2656         .readdir        = proc_tgid_base_readdir,
2657 };
2658
2659 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2660         return proc_pident_lookup(dir, dentry,
2661                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2662 }
2663
2664 static const struct inode_operations proc_tgid_base_inode_operations = {
2665         .lookup         = proc_tgid_base_lookup,
2666         .getattr        = pid_getattr,
2667         .setattr        = proc_setattr,
2668 };
2669
2670 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2671 {
2672         struct dentry *dentry, *leader, *dir;
2673         char buf[PROC_NUMBUF];
2674         struct qstr name;
2675
2676         name.name = buf;
2677         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2678         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2679         if (dentry) {
2680                 shrink_dcache_parent(dentry);
2681                 d_drop(dentry);
2682                 dput(dentry);
2683         }
2684
2685         name.name = buf;
2686         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2687         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2688         if (!leader)
2689                 goto out;
2690
2691         name.name = "task";
2692         name.len = strlen(name.name);
2693         dir = d_hash_and_lookup(leader, &name);
2694         if (!dir)
2695                 goto out_put_leader;
2696
2697         name.name = buf;
2698         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2699         dentry = d_hash_and_lookup(dir, &name);
2700         if (dentry) {
2701                 shrink_dcache_parent(dentry);
2702                 d_drop(dentry);
2703                 dput(dentry);
2704         }
2705
2706         dput(dir);
2707 out_put_leader:
2708         dput(leader);
2709 out:
2710         return;
2711 }
2712
2713 /**
2714  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2715  * @task: task that should be flushed.
2716  *
2717  * When flushing dentries from proc, one needs to flush them from global
2718  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2719  * in. This call is supposed to do all of this job.
2720  *
2721  * Looks in the dcache for
2722  * /proc/@pid
2723  * /proc/@tgid/task/@pid
2724  * if either directory is present flushes it and all of it'ts children
2725  * from the dcache.
2726  *
2727  * It is safe and reasonable to cache /proc entries for a task until
2728  * that task exits.  After that they just clog up the dcache with
2729  * useless entries, possibly causing useful dcache entries to be
2730  * flushed instead.  This routine is proved to flush those useless
2731  * dcache entries at process exit time.
2732  *
2733  * NOTE: This routine is just an optimization so it does not guarantee
2734  *       that no dcache entries will exist at process exit time it
2735  *       just makes it very unlikely that any will persist.
2736  */
2737
2738 void proc_flush_task(struct task_struct *task)
2739 {
2740         int i;
2741         struct pid *pid, *tgid;
2742         struct upid *upid;
2743
2744         pid = task_pid(task);
2745         tgid = task_tgid(task);
2746
2747         for (i = 0; i <= pid->level; i++) {
2748                 upid = &pid->numbers[i];
2749                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2750                                         tgid->numbers[i].nr);
2751         }
2752
2753         upid = &pid->numbers[pid->level];
2754         if (upid->nr == 1)
2755                 pid_ns_release_proc(upid->ns);
2756 }
2757
2758 static struct dentry *proc_pid_instantiate(struct inode *dir,
2759                                            struct dentry * dentry,
2760                                            struct task_struct *task, const void *ptr)
2761 {
2762         struct dentry *error = ERR_PTR(-ENOENT);
2763         struct inode *inode;
2764
2765         inode = proc_pid_make_inode(dir->i_sb, task);
2766         if (!inode)
2767                 goto out;
2768
2769         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2770         inode->i_op = &proc_tgid_base_inode_operations;
2771         inode->i_fop = &proc_tgid_base_operations;
2772         inode->i_flags|=S_IMMUTABLE;
2773
2774         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2775                 ARRAY_SIZE(tgid_base_stuff));
2776
2777         dentry->d_op = &pid_dentry_operations;
2778
2779         d_add(dentry, inode);
2780         /* Close the race of the process dying before we return the dentry */
2781         if (pid_revalidate(dentry, NULL))
2782                 error = NULL;
2783 out:
2784         return error;
2785 }
2786
2787 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2788 {
2789         struct dentry *result = ERR_PTR(-ENOENT);
2790         struct task_struct *task;
2791         unsigned tgid;
2792         struct pid_namespace *ns;
2793
2794         result = proc_base_lookup(dir, dentry);
2795         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2796                 goto out;
2797
2798         tgid = name_to_int(dentry);
2799         if (tgid == ~0U)
2800                 goto out;
2801
2802         ns = dentry->d_sb->s_fs_info;
2803         rcu_read_lock();
2804         task = find_task_by_pid_ns(tgid, ns);
2805         if (task)
2806                 get_task_struct(task);
2807         rcu_read_unlock();
2808         if (!task)
2809                 goto out;
2810
2811         result = proc_pid_instantiate(dir, dentry, task, NULL);
2812         put_task_struct(task);
2813 out:
2814         return result;
2815 }
2816
2817 /*
2818  * Find the first task with tgid >= tgid
2819  *
2820  */
2821 struct tgid_iter {
2822         unsigned int tgid;
2823         struct task_struct *task;
2824 };
2825 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2826 {
2827         struct pid *pid;
2828
2829         if (iter.task)
2830                 put_task_struct(iter.task);
2831         rcu_read_lock();
2832 retry:
2833         iter.task = NULL;
2834         pid = find_ge_pid(iter.tgid, ns);
2835         if (pid) {
2836                 iter.tgid = pid_nr_ns(pid, ns);
2837                 iter.task = pid_task(pid, PIDTYPE_PID);
2838                 /* What we to know is if the pid we have find is the
2839                  * pid of a thread_group_leader.  Testing for task
2840                  * being a thread_group_leader is the obvious thing
2841                  * todo but there is a window when it fails, due to
2842                  * the pid transfer logic in de_thread.
2843                  *
2844                  * So we perform the straight forward test of seeing
2845                  * if the pid we have found is the pid of a thread
2846                  * group leader, and don't worry if the task we have
2847                  * found doesn't happen to be a thread group leader.
2848                  * As we don't care in the case of readdir.
2849                  */
2850                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2851                         iter.tgid += 1;
2852                         goto retry;
2853                 }
2854                 get_task_struct(iter.task);
2855         }
2856         rcu_read_unlock();
2857         return iter;
2858 }
2859
2860 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2861
2862 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2863         struct tgid_iter iter)
2864 {
2865         char name[PROC_NUMBUF];
2866         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2867         return proc_fill_cache(filp, dirent, filldir, name, len,
2868                                 proc_pid_instantiate, iter.task, NULL);
2869 }
2870
2871 /* for the /proc/ directory itself, after non-process stuff has been done */
2872 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2873 {
2874         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2875         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2876         struct tgid_iter iter;
2877         struct pid_namespace *ns;
2878
2879         if (!reaper)
2880                 goto out_no_task;
2881
2882         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2883                 const struct pid_entry *p = &proc_base_stuff[nr];
2884                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2885                         goto out;
2886         }
2887
2888         ns = filp->f_dentry->d_sb->s_fs_info;
2889         iter.task = NULL;
2890         iter.tgid = filp->f_pos - TGID_OFFSET;
2891         for (iter = next_tgid(ns, iter);
2892              iter.task;
2893              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2894                 filp->f_pos = iter.tgid + TGID_OFFSET;
2895                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2896                         put_task_struct(iter.task);
2897                         goto out;
2898                 }
2899         }
2900         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2901 out:
2902         put_task_struct(reaper);
2903 out_no_task:
2904         return 0;
2905 }
2906
2907 /*
2908  * Tasks
2909  */
2910 static const struct pid_entry tid_base_stuff[] = {
2911         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2912         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2913         REG("environ",   S_IRUSR, proc_environ_operations),
2914         INF("auxv",      S_IRUSR, proc_pid_auxv),
2915         ONE("status",    S_IRUGO, proc_pid_status),
2916         ONE("personality", S_IRUSR, proc_pid_personality),
2917         INF("limits",    S_IRUSR, proc_pid_limits),
2918 #ifdef CONFIG_SCHED_DEBUG
2919         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2920 #endif
2921         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2922 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2923         INF("syscall",   S_IRUSR, proc_pid_syscall),
2924 #endif
2925         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2926         ONE("stat",      S_IRUGO, proc_tid_stat),
2927         ONE("statm",     S_IRUGO, proc_pid_statm),
2928         REG("maps",      S_IRUGO, proc_maps_operations),
2929 #ifdef CONFIG_NUMA
2930         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2931 #endif
2932         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2933         LNK("cwd",       proc_cwd_link),
2934         LNK("root",      proc_root_link),
2935         LNK("exe",       proc_exe_link),
2936         REG("mounts",    S_IRUGO, proc_mounts_operations),
2937         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2938 #ifdef CONFIG_PROC_PAGE_MONITOR
2939         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2940         REG("smaps",     S_IRUGO, proc_smaps_operations),
2941         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2942 #endif
2943 #ifdef CONFIG_SECURITY
2944         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2945 #endif
2946 #ifdef CONFIG_KALLSYMS
2947         INF("wchan",     S_IRUGO, proc_pid_wchan),
2948 #endif
2949 #ifdef CONFIG_STACKTRACE
2950         ONE("stack",      S_IRUSR, proc_pid_stack),
2951 #endif
2952 #ifdef CONFIG_SCHEDSTATS
2953         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2954 #endif
2955 #ifdef CONFIG_LATENCYTOP
2956         REG("latency",  S_IRUGO, proc_lstats_operations),
2957 #endif
2958 #ifdef CONFIG_PROC_PID_CPUSET
2959         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2960 #endif
2961 #ifdef CONFIG_CGROUPS
2962         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2963 #endif
2964         INF("oom_score", S_IRUGO, proc_oom_score),
2965         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2966 #ifdef CONFIG_AUDITSYSCALL
2967         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2968         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2969 #endif
2970 #ifdef CONFIG_FAULT_INJECTION
2971         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2972 #endif
2973 #ifdef CONFIG_TASK_IO_ACCOUNTING
2974         INF("io",       S_IRUGO, proc_tid_io_accounting),
2975 #endif
2976 };
2977
2978 static int proc_tid_base_readdir(struct file * filp,
2979                              void * dirent, filldir_t filldir)
2980 {
2981         return proc_pident_readdir(filp,dirent,filldir,
2982                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2983 }
2984
2985 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2986         return proc_pident_lookup(dir, dentry,
2987                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2988 }
2989
2990 static const struct file_operations proc_tid_base_operations = {
2991         .read           = generic_read_dir,
2992         .readdir        = proc_tid_base_readdir,
2993 };
2994
2995 static const struct inode_operations proc_tid_base_inode_operations = {
2996         .lookup         = proc_tid_base_lookup,
2997         .getattr        = pid_getattr,
2998         .setattr        = proc_setattr,
2999 };
3000
3001 static struct dentry *proc_task_instantiate(struct inode *dir,
3002         struct dentry *dentry, struct task_struct *task, const void *ptr)
3003 {
3004         struct dentry *error = ERR_PTR(-ENOENT);
3005         struct inode *inode;
3006         inode = proc_pid_make_inode(dir->i_sb, task);
3007
3008         if (!inode)
3009                 goto out;
3010         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3011         inode->i_op = &proc_tid_base_inode_operations;
3012         inode->i_fop = &proc_tid_base_operations;
3013         inode->i_flags|=S_IMMUTABLE;
3014
3015         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3016                 ARRAY_SIZE(tid_base_stuff));
3017
3018         dentry->d_op = &pid_dentry_operations;
3019
3020         d_add(dentry, inode);
3021         /* Close the race of the process dying before we return the dentry */
3022         if (pid_revalidate(dentry, NULL))
3023                 error = NULL;
3024 out:
3025         return error;
3026 }
3027
3028 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3029 {
3030         struct dentry *result = ERR_PTR(-ENOENT);
3031         struct task_struct *task;
3032         struct task_struct *leader = get_proc_task(dir);
3033         unsigned tid;
3034         struct pid_namespace *ns;
3035
3036         if (!leader)
3037                 goto out_no_task;
3038
3039         tid = name_to_int(dentry);
3040         if (tid == ~0U)
3041                 goto out;
3042
3043         ns = dentry->d_sb->s_fs_info;
3044         rcu_read_lock();
3045         task = find_task_by_pid_ns(tid, ns);
3046         if (task)
3047                 get_task_struct(task);
3048         rcu_read_unlock();
3049         if (!task)
3050                 goto out;
3051         if (!same_thread_group(leader, task))
3052                 goto out_drop_task;
3053
3054         result = proc_task_instantiate(dir, dentry, task, NULL);
3055 out_drop_task:
3056         put_task_struct(task);
3057 out:
3058         put_task_struct(leader);
3059 out_no_task:
3060         return result;
3061 }
3062
3063 /*
3064  * Find the first tid of a thread group to return to user space.
3065  *
3066  * Usually this is just the thread group leader, but if the users
3067  * buffer was too small or there was a seek into the middle of the
3068  * directory we have more work todo.
3069  *
3070  * In the case of a short read we start with find_task_by_pid.
3071  *
3072  * In the case of a seek we start with the leader and walk nr
3073  * threads past it.
3074  */
3075 static struct task_struct *first_tid(struct task_struct *leader,
3076                 int tid, int nr, struct pid_namespace *ns)
3077 {
3078         struct task_struct *pos;
3079
3080         rcu_read_lock();
3081         /* Attempt to start with the pid of a thread */
3082         if (tid && (nr > 0)) {
3083                 pos = find_task_by_pid_ns(tid, ns);
3084                 if (pos && (pos->group_leader == leader))
3085                         goto found;
3086         }
3087
3088         /* If nr exceeds the number of threads there is nothing todo */
3089         pos = NULL;
3090         if (nr && nr >= get_nr_threads(leader))
3091                 goto out;
3092
3093         /* If we haven't found our starting place yet start
3094          * with the leader and walk nr threads forward.
3095          */
3096         for (pos = leader; nr > 0; --nr) {
3097                 pos = next_thread(pos);
3098                 if (pos == leader) {
3099                         pos = NULL;
3100                         goto out;
3101                 }
3102         }
3103 found:
3104         get_task_struct(pos);
3105 out:
3106         rcu_read_unlock();
3107         return pos;
3108 }
3109
3110 /*
3111  * Find the next thread in the thread list.
3112  * Return NULL if there is an error or no next thread.
3113  *
3114  * The reference to the input task_struct is released.
3115  */
3116 static struct task_struct *next_tid(struct task_struct *start)
3117 {
3118         struct task_struct *pos = NULL;
3119         rcu_read_lock();
3120         if (pid_alive(start)) {
3121                 pos = next_thread(start);
3122                 if (thread_group_leader(pos))
3123                         pos = NULL;
3124                 else
3125                         get_task_struct(pos);
3126         }
3127         rcu_read_unlock();
3128         put_task_struct(start);
3129         return pos;
3130 }
3131
3132 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3133         struct task_struct *task, int tid)
3134 {
3135         char name[PROC_NUMBUF];
3136         int len = snprintf(name, sizeof(name), "%d", tid);
3137         return proc_fill_cache(filp, dirent, filldir, name, len,
3138                                 proc_task_instantiate, task, NULL);
3139 }
3140
3141 /* for the /proc/TGID/task/ directories */
3142 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3143 {
3144         struct dentry *dentry = filp->f_path.dentry;
3145         struct inode *inode = dentry->d_inode;
3146         struct task_struct *leader = NULL;
3147         struct task_struct *task;
3148         int retval = -ENOENT;
3149         ino_t ino;
3150         int tid;
3151         struct pid_namespace *ns;
3152
3153         task = get_proc_task(inode);
3154         if (!task)
3155                 goto out_no_task;
3156         rcu_read_lock();
3157         if (pid_alive(task)) {
3158                 leader = task->group_leader;
3159                 get_task_struct(leader);
3160         }
3161         rcu_read_unlock();
3162         put_task_struct(task);
3163         if (!leader)
3164                 goto out_no_task;
3165         retval = 0;
3166
3167         switch ((unsigned long)filp->f_pos) {
3168         case 0:
3169                 ino = inode->i_ino;
3170                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3171                         goto out;
3172                 filp->f_pos++;
3173                 /* fall through */
3174         case 1:
3175                 ino = parent_ino(dentry);
3176                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3177                         goto out;
3178                 filp->f_pos++;
3179                 /* fall through */
3180         }
3181
3182         /* f_version caches the tgid value that the last readdir call couldn't
3183          * return. lseek aka telldir automagically resets f_version to 0.
3184          */
3185         ns = filp->f_dentry->d_sb->s_fs_info;
3186         tid = (int)filp->f_version;
3187         filp->f_version = 0;
3188         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3189              task;
3190              task = next_tid(task), filp->f_pos++) {
3191                 tid = task_pid_nr_ns(task, ns);
3192                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3193                         /* returning this tgid failed, save it as the first
3194                          * pid for the next readir call */
3195                         filp->f_version = (u64)tid;
3196                         put_task_struct(task);
3197                         break;
3198                 }
3199         }
3200 out:
3201         put_task_struct(leader);
3202 out_no_task:
3203         return retval;
3204 }
3205
3206 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3207 {
3208         struct inode *inode = dentry->d_inode;
3209         struct task_struct *p = get_proc_task(inode);
3210         generic_fillattr(inode, stat);
3211
3212         if (p) {
3213                 stat->nlink += get_nr_threads(p);
3214                 put_task_struct(p);
3215         }
3216
3217         return 0;
3218 }
3219
3220 static const struct inode_operations proc_task_inode_operations = {
3221         .lookup         = proc_task_lookup,
3222         .getattr        = proc_task_getattr,
3223         .setattr        = proc_setattr,
3224 };
3225
3226 static const struct file_operations proc_task_operations = {
3227         .read           = generic_read_dir,
3228         .readdir        = proc_task_readdir,
3229 };