fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/stacktrace.h>
  69 #include <linux/resource.h>
  70 #include <linux/module.h>
  71 #include <linux/mount.h>
  72 #include <linux/security.h>
  73 #include <linux/ptrace.h>
  74 #include <linux/tracehook.h>
  75 #include <linux/cgroup.h>
  76 #include <linux/cpuset.h>
  77 #include <linux/audit.h>
  78 #include <linux/poll.h>
  79 #include <linux/nsproxy.h>
  80 #include <linux/oom.h>
  81 #include <linux/elf.h>
  82 #include <linux/pid_namespace.h>
  83 #include <linux/fs_struct.h>
  84 #include "internal.h"
  85
  86 /* NOTE:
  87  *      Implementing inode permission operations in /proc is almost
  88  *      certainly an error.  Permission checks need to happen during
  89  *      each system call not at open time.  The reason is that most of
  90  *      what we wish to check for permissions in /proc varies at runtime.
  91  *
  92  *      The classic example of a problem is opening file descriptors
  93  *      in /proc for a task before it execs a suid executable.
  94  */
  95
  96 struct pid_entry {
  97         char *name;
  98         int len;
  99         mode_t mode;
 100         const struct inode_operations *iop;
 101         const struct file_operations *fop;
 102         union proc_op op;
 103 };
 104
 105 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 106         .name = (NAME),                                 \
 107         .len  = sizeof(NAME) - 1,                       \
 108         .mode = MODE,                                   \
 109         .iop  = IOP,                                    \
 110         .fop  = FOP,                                    \
 111         .op   = OP,                                     \
 112 }
 113
 114 #define DIR(NAME, MODE, iops, fops)     \
 115         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 116 #define LNK(NAME, get_link)                                     \
 117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 118                 &proc_pid_link_inode_operations, NULL,          \
 119                 { .proc_get_link = get_link } )
 120 #define REG(NAME, MODE, fops)                           \
 121         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 122 #define INF(NAME, MODE, read)                           \
 123         NOD(NAME, (S_IFREG|(MODE)),                     \
 124                 NULL, &proc_info_file_operations,       \
 125                 { .proc_read = read } )
 126 #define ONE(NAME, MODE, show)                           \
 127         NOD(NAME, (S_IFREG|(MODE)),                     \
 128                 NULL, &proc_single_file_operations,     \
 129                 { .proc_show = show } )
 130
 131 /*
 132  * Count the number of hardlinks for the pid_entry table, excluding the .
 133  * and .. links.
 134  */
 135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 136         unsigned int n)
 137 {
 138         unsigned int i;
 139         unsigned int count;
 140
 141         count = 0;
 142         for (i = 0; i < n; ++i) {
 143                 if (S_ISDIR(entries[i].mode))
 144                         ++count;
 145         }
 146
 147         return count;
 148 }
 149
 150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
 151 {
 152         struct fs_struct *fs;
 153         int result = -ENOENT;
 154
 155         task_lock(task);
 156         fs = task->fs;
 157         if (fs) {
 158                 read_lock(&fs->lock);
 159                 *path = root ? fs->root : fs->pwd;
 160                 path_get(path);
 161                 read_unlock(&fs->lock);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int get_nr_threads(struct task_struct *tsk)
 169 {
 170         unsigned long flags;
 171         int count = 0;
 172
 173         if (lock_task_sighand(tsk, &flags)) {
 174                 count = atomic_read(&tsk->signal->count);
 175                 unlock_task_sighand(tsk, &flags);
 176         }
 177         return count;
 178 }
 179
 180 static int proc_cwd_link(struct inode *inode, struct path *path)
 181 {
 182         struct task_struct *task = get_proc_task(inode);
 183         int result = -ENOENT;
 184
 185         if (task) {
 186                 result = get_fs_path(task, path, 0);
 187                 put_task_struct(task);
 188         }
 189         return result;
 190 }
 191
 192 static int proc_root_link(struct inode *inode, struct path *path)
 193 {
 194         struct task_struct *task = get_proc_task(inode);
 195         int result = -ENOENT;
 196
 197         if (task) {
 198                 result = get_fs_path(task, path, 1);
 199                 put_task_struct(task);
 200         }
 201         return result;
 202 }
 203
 204 /*
 205  * Return zero if current may access user memory in @task, -error if not.
 206  */
 207 static int check_mem_permission(struct task_struct *task)
 208 {
 209         /*
 210          * A task can always look at itself, in case it chooses
 211          * to use system calls instead of load instructions.
 212          */
 213         if (task == current)
 214                 return 0;
 215
 216         /*
 217          * If current is actively ptrace'ing, and would also be
 218          * permitted to freshly attach with ptrace now, permit it.
 219          */
 220         if (task_is_stopped_or_traced(task)) {
 221                 int match;
 222                 rcu_read_lock();
 223                 match = (tracehook_tracer_task(task) == current);
 224                 rcu_read_unlock();
 225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 226                         return 0;
 227         }
 228
 229         /*
 230          * Noone else is allowed.
 231          */
 232         return -EPERM;
 233 }
 234
 235 struct mm_struct *mm_for_maps(struct task_struct *task)
 236 {
 237         struct mm_struct *mm;
 238
 239         if (mutex_lock_killable(&task->cred_guard_mutex))
 240                 return NULL;
 241
 242         mm = get_task_mm(task);
 243         if (mm && mm != current->mm &&
 244                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 245                 mmput(mm);
 246                 mm = NULL;
 247         }
 248         mutex_unlock(&task->cred_guard_mutex);
 249
 250         return mm;
 251 }
 252
 253 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 254 {
 255         int res = 0;
 256         unsigned int len;
 257         struct mm_struct *mm = get_task_mm(task);
 258         if (!mm)
 259                 goto out;
 260         if (!mm->arg_end)
 261                 goto out_mm;    /* Shh! No looking before we're done */
 262
 263         len = mm->arg_end - mm->arg_start;
 264
 265         if (len > PAGE_SIZE)
 266                 len = PAGE_SIZE;
 267
 268         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 269
 270         // If the nul at the end of args has been overwritten, then
 271         // assume application is using setproctitle(3).
 272         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 273                 len = strnlen(buffer, res);
 274                 if (len < res) {
 275                     res = len;
 276                 } else {
 277                         len = mm->env_end - mm->env_start;
 278                         if (len > PAGE_SIZE - res)
 279                                 len = PAGE_SIZE - res;
 280                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 281                         res = strnlen(buffer, res);
 282                 }
 283         }
 284 out_mm:
 285         mmput(mm);
 286 out:
 287         return res;
 288 }
 289
 290 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 291 {
 292         int res = 0;
 293         struct mm_struct *mm = get_task_mm(task);
 294         if (mm) {
 295                 unsigned int nwords = 0;
 296                 do {
 297                         nwords += 2;
 298                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 299                 res = nwords * sizeof(mm->saved_auxv[0]);
 300                 if (res > PAGE_SIZE)
 301                         res = PAGE_SIZE;
 302                 memcpy(buffer, mm->saved_auxv, res);
 303                 mmput(mm);
 304         }
 305         return res;
 306 }
 307
 308
 309 #ifdef CONFIG_KALLSYMS
 310 /*
 311  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 312  * Returns the resolved symbol.  If that fails, simply return the address.
 313  */
 314 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 315 {
 316         unsigned long wchan;
 317         char symname[KSYM_NAME_LEN];
 318
 319         wchan = get_wchan(task);
 320
 321         if (lookup_symbol_name(wchan, symname) < 0)
 322                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 323                         return 0;
 324                 else
 325                         return sprintf(buffer, "%lu", wchan);
 326         else
 327                 return sprintf(buffer, "%s", symname);
 328 }
 329 #endif /* CONFIG_KALLSYMS */
 330
 331 #ifdef CONFIG_STACKTRACE
 332
 333 #define MAX_STACK_TRACE_DEPTH   64
 334
 335 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 336                           struct pid *pid, struct task_struct *task)
 337 {
 338         struct stack_trace trace;
 339         unsigned long *entries;
 340         int i;
 341
 342         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 343         if (!entries)
 344                 return -ENOMEM;
 345
 346         trace.nr_entries        = 0;
 347         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 348         trace.entries           = entries;
 349         trace.skip              = 0;
 350         save_stack_trace_tsk(task, &trace);
 351
 352         for (i = 0; i < trace.nr_entries; i++) {
 353                 seq_printf(m, "[<%p>] %pS\n",
 354                            (void *)entries[i], (void *)entries[i]);
 355         }
 356         kfree(entries);
 357
 358         return 0;
 359 }
 360 #endif
 361
 362 #ifdef CONFIG_SCHEDSTATS
 363 /*
 364  * Provides /proc/PID/schedstat
 365  */
 366 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 367 {
 368         return sprintf(buffer, "%llu %llu %lu\n",
 369                         (unsigned long long)task->se.sum_exec_runtime,
 370                         (unsigned long long)task->sched_info.run_delay,
 371                         task->sched_info.pcount);
 372 }
 373 #endif
 374
 375 #ifdef CONFIG_LATENCYTOP
 376 static int lstats_show_proc(struct seq_file *m, void *v)
 377 {
 378         int i;
 379         struct inode *inode = m->private;
 380         struct task_struct *task = get_proc_task(inode);
 381
 382         if (!task)
 383                 return -ESRCH;
 384         seq_puts(m, "Latency Top version : v0.1\n");
 385         for (i = 0; i < 32; i++) {
 386                 if (task->latency_record[i].backtrace[0]) {
 387                         int q;
 388                         seq_printf(m, "%i %li %li ",
 389                                 task->latency_record[i].count,
 390                                 task->latency_record[i].time,
 391                                 task->latency_record[i].max);
 392                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 393                                 char sym[KSYM_SYMBOL_LEN];
 394                                 char *c;
 395                                 if (!task->latency_record[i].backtrace[q])
 396                                         break;
 397                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 398                                         break;
 399                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 400                                 c = strchr(sym, '+');
 401                                 if (c)
 402                                         *c = 0;
 403                                 seq_printf(m, "%s ", sym);
 404                         }
 405                         seq_printf(m, "\n");
 406                 }
 407
 408         }
 409         put_task_struct(task);
 410         return 0;
 411 }
 412
 413 static int lstats_open(struct inode *inode, struct file *file)
 414 {
 415         return single_open(file, lstats_show_proc, inode);
 416 }
 417
 418 static ssize_t lstats_write(struct file *file, const char __user *buf,
 419                             size_t count, loff_t *offs)
 420 {
 421         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 422
 423         if (!task)
 424                 return -ESRCH;
 425         clear_all_latency_tracing(task);
 426         put_task_struct(task);
 427
 428         return count;
 429 }
 430
 431 static const struct file_operations proc_lstats_operations = {
 432         .open           = lstats_open,
 433         .read           = seq_read,
 434         .write          = lstats_write,
 435         .llseek         = seq_lseek,
 436         .release        = single_release,
 437 };
 438
 439 #endif
 440
 441 /* The badness from the OOM killer */
 442 unsigned long badness(struct task_struct *p, unsigned long uptime);
 443 static int proc_oom_score(struct task_struct *task, char *buffer)
 444 {
 445         unsigned long points;
 446         struct timespec uptime;
 447
 448         do_posix_clock_monotonic_gettime(&uptime);
 449         read_lock(&tasklist_lock);
 450         points = badness(task->group_leader, uptime.tv_sec);
 451         read_unlock(&tasklist_lock);
 452         return sprintf(buffer, "%lu\n", points);
 453 }
 454
 455 struct limit_names {
 456         char *name;
 457         char *unit;
 458 };
 459
 460 static const struct limit_names lnames[RLIM_NLIMITS] = {
 461         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 462         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 463         [RLIMIT_DATA] = {"Max data size", "bytes"},
 464         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 465         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 466         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 467         [RLIMIT_NPROC] = {"Max processes", "processes"},
 468         [RLIMIT_NOFILE] = {"Max open files", "files"},
 469         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 470         [RLIMIT_AS] = {"Max address space", "bytes"},
 471         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 472         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 473         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 474         [RLIMIT_NICE] = {"Max nice priority", NULL},
 475         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 476         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 477 };
 478
 479 /* Display limits for a process */
 480 static int proc_pid_limits(struct task_struct *task, char *buffer)
 481 {
 482         unsigned int i;
 483         int count = 0;
 484         unsigned long flags;
 485         char *bufptr = buffer;
 486
 487         struct rlimit rlim[RLIM_NLIMITS];
 488
 489         if (!lock_task_sighand(task, &flags))
 490                 return 0;
 491         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 492         unlock_task_sighand(task, &flags);
 493
 494         /*
 495          * print the file header
 496          */
 497         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 498                         "Limit", "Soft Limit", "Hard Limit", "Units");
 499
 500         for (i = 0; i < RLIM_NLIMITS; i++) {
 501                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 502                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 503                                          lnames[i].name, "unlimited");
 504                 else
 505                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 506                                          lnames[i].name, rlim[i].rlim_cur);
 507
 508                 if (rlim[i].rlim_max == RLIM_INFINITY)
 509                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 510                 else
 511                         count += sprintf(&bufptr[count], "%-20lu ",
 512                                          rlim[i].rlim_max);
 513
 514                 if (lnames[i].unit)
 515                         count += sprintf(&bufptr[count], "%-10s\n",
 516                                          lnames[i].unit);
 517                 else
 518                         count += sprintf(&bufptr[count], "\n");
 519         }
 520
 521         return count;
 522 }
 523
 524 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 525 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 526 {
 527         long nr;
 528         unsigned long args[6], sp, pc;
 529
 530         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 531                 return sprintf(buffer, "running\n");
 532
 533         if (nr < 0)
 534                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 535
 536         return sprintf(buffer,
 537                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 538                        nr,
 539                        args[0], args[1], args[2], args[3], args[4], args[5],
 540                        sp, pc);
 541 }
 542 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 543
 544 /************************************************************************/
 545 /*                       Here the fs part begins                        */
 546 /************************************************************************/
 547
 548 /* permission checks */
 549 static int proc_fd_access_allowed(struct inode *inode)
 550 {
 551         struct task_struct *task;
 552         int allowed = 0;
 553         /* Allow access to a task's file descriptors if it is us or we
 554          * may use ptrace attach to the process and find out that
 555          * information.
 556          */
 557         task = get_proc_task(inode);
 558         if (task) {
 559                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 560                 put_task_struct(task);
 561         }
 562         return allowed;
 563 }
 564
 565 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 566 {
 567         int error;
 568         struct inode *inode = dentry->d_inode;
 569
 570         if (attr->ia_valid & ATTR_MODE)
 571                 return -EPERM;
 572
 573         error = inode_change_ok(inode, attr);
 574         if (!error)
 575                 error = inode_setattr(inode, attr);
 576         return error;
 577 }
 578
 579 static const struct inode_operations proc_def_inode_operations = {
 580         .setattr        = proc_setattr,
 581 };
 582
 583 static int mounts_open_common(struct inode *inode, struct file *file,
 584                               const struct seq_operations *op)
 585 {
 586         struct task_struct *task = get_proc_task(inode);
 587         struct nsproxy *nsp;
 588         struct mnt_namespace *ns = NULL;
 589         struct path root;
 590         struct proc_mounts *p;
 591         int ret = -EINVAL;
 592
 593         if (task) {
 594                 rcu_read_lock();
 595                 nsp = task_nsproxy(task);
 596                 if (nsp) {
 597                         ns = nsp->mnt_ns;
 598                         if (ns)
 599                                 get_mnt_ns(ns);
 600                 }
 601                 rcu_read_unlock();
 602                 if (ns && get_fs_path(task, &root, 1) == 0)
 603                         ret = 0;
 604                 put_task_struct(task);
 605         }
 606
 607         if (!ns)
 608                 goto err;
 609         if (ret)
 610                 goto err_put_ns;
 611
 612         ret = -ENOMEM;
 613         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 614         if (!p)
 615                 goto err_put_path;
 616
 617         file->private_data = &p->m;
 618         ret = seq_open(file, op);
 619         if (ret)
 620                 goto err_free;
 621
 622         p->m.private = p;
 623         p->ns = ns;
 624         p->root = root;
 625         p->event = ns->event;
 626
 627         return 0;
 628
 629  err_free:
 630         kfree(p);
 631  err_put_path:
 632         path_put(&root);
 633  err_put_ns:
 634         put_mnt_ns(ns);
 635  err:
 636         return ret;
 637 }
 638
 639 static int mounts_release(struct inode *inode, struct file *file)
 640 {
 641         struct proc_mounts *p = file->private_data;
 642         path_put(&p->root);
 643         put_mnt_ns(p->ns);
 644         return seq_release(inode, file);
 645 }
 646
 647 static unsigned mounts_poll(struct file *file, poll_table *wait)
 648 {
 649         struct proc_mounts *p = file->private_data;
 650         unsigned res = POLLIN | POLLRDNORM;
 651
 652         poll_wait(file, &p->ns->poll, wait);
 653         if (mnt_had_events(p))
 654                 res |= POLLERR | POLLPRI;
 655
 656         return res;
 657 }
 658
 659 static int mounts_open(struct inode *inode, struct file *file)
 660 {
 661         return mounts_open_common(inode, file, &mounts_op);
 662 }
 663
 664 static const struct file_operations proc_mounts_operations = {
 665         .open           = mounts_open,
 666         .read           = seq_read,
 667         .llseek         = seq_lseek,
 668         .release        = mounts_release,
 669         .poll           = mounts_poll,
 670 };
 671
 672 static int mountinfo_open(struct inode *inode, struct file *file)
 673 {
 674         return mounts_open_common(inode, file, &mountinfo_op);
 675 }
 676
 677 static const struct file_operations proc_mountinfo_operations = {
 678         .open           = mountinfo_open,
 679         .read           = seq_read,
 680         .llseek         = seq_lseek,
 681         .release        = mounts_release,
 682         .poll           = mounts_poll,
 683 };
 684
 685 static int mountstats_open(struct inode *inode, struct file *file)
 686 {
 687         return mounts_open_common(inode, file, &mountstats_op);
 688 }
 689
 690 static const struct file_operations proc_mountstats_operations = {
 691         .open           = mountstats_open,
 692         .read           = seq_read,
 693         .llseek         = seq_lseek,
 694         .release        = mounts_release,
 695 };
 696
 697 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 698
 699 static ssize_t proc_info_read(struct file * file, char __user * buf,
 700                           size_t count, loff_t *ppos)
 701 {
 702         struct inode * inode = file->f_path.dentry->d_inode;
 703         unsigned long page;
 704         ssize_t length;
 705         struct task_struct *task = get_proc_task(inode);
 706
 707         length = -ESRCH;
 708         if (!task)
 709                 goto out_no_task;
 710
 711         if (count > PROC_BLOCK_SIZE)
 712                 count = PROC_BLOCK_SIZE;
 713
 714         length = -ENOMEM;
 715         if (!(page = __get_free_page(GFP_TEMPORARY)))
 716                 goto out;
 717
 718         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 719
 720         if (length >= 0)
 721                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 722         free_page(page);
 723 out:
 724         put_task_struct(task);
 725 out_no_task:
 726         return length;
 727 }
 728
 729 static const struct file_operations proc_info_file_operations = {
 730         .read           = proc_info_read,
 731 };
 732
 733 static int proc_single_show(struct seq_file *m, void *v)
 734 {
 735         struct inode *inode = m->private;
 736         struct pid_namespace *ns;
 737         struct pid *pid;
 738         struct task_struct *task;
 739         int ret;
 740
 741         ns = inode->i_sb->s_fs_info;
 742         pid = proc_pid(inode);
 743         task = get_pid_task(pid, PIDTYPE_PID);
 744         if (!task)
 745                 return -ESRCH;
 746
 747         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 748
 749         put_task_struct(task);
 750         return ret;
 751 }
 752
 753 static int proc_single_open(struct inode *inode, struct file *filp)
 754 {
 755         int ret;
 756         ret = single_open(filp, proc_single_show, NULL);
 757         if (!ret) {
 758                 struct seq_file *m = filp->private_data;
 759
 760                 m->private = inode;
 761         }
 762         return ret;
 763 }
 764
 765 static const struct file_operations proc_single_file_operations = {
 766         .open           = proc_single_open,
 767         .read           = seq_read,
 768         .llseek         = seq_lseek,
 769         .release        = single_release,
 770 };
 771
 772 static int mem_open(struct inode* inode, struct file* file)
 773 {
 774         file->private_data = (void*)((long)current->self_exec_id);
 775         return 0;
 776 }
 777
 778 static ssize_t mem_read(struct file * file, char __user * buf,
 779                         size_t count, loff_t *ppos)
 780 {
 781         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 782         char *page;
 783         unsigned long src = *ppos;
 784         int ret = -ESRCH;
 785         struct mm_struct *mm;
 786
 787         if (!task)
 788                 goto out_no_task;
 789
 790         if (check_mem_permission(task))
 791                 goto out;
 792
 793         ret = -ENOMEM;
 794         page = (char *)__get_free_page(GFP_TEMPORARY);
 795         if (!page)
 796                 goto out;
 797
 798         ret = 0;
 799
 800         mm = get_task_mm(task);
 801         if (!mm)
 802                 goto out_free;
 803
 804         ret = -EIO;
 805
 806         if (file->private_data != (void*)((long)current->self_exec_id))
 807                 goto out_put;
 808
 809         ret = 0;
 810
 811         while (count > 0) {
 812                 int this_len, retval;
 813
 814                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 815                 retval = access_process_vm(task, src, page, this_len, 0);
 816                 if (!retval || check_mem_permission(task)) {
 817                         if (!ret)
 818                                 ret = -EIO;
 819                         break;
 820                 }
 821
 822                 if (copy_to_user(buf, page, retval)) {
 823                         ret = -EFAULT;
 824                         break;
 825                 }
 826
 827                 ret += retval;
 828                 src += retval;
 829                 buf += retval;
 830                 count -= retval;
 831         }
 832         *ppos = src;
 833
 834 out_put:
 835         mmput(mm);
 836 out_free:
 837         free_page((unsigned long) page);
 838 out:
 839         put_task_struct(task);
 840 out_no_task:
 841         return ret;
 842 }
 843
 844 #define mem_write NULL
 845
 846 #ifndef mem_write
 847 /* This is a security hazard */
 848 static ssize_t mem_write(struct file * file, const char __user *buf,
 849                          size_t count, loff_t *ppos)
 850 {
 851         int copied;
 852         char *page;
 853         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 854         unsigned long dst = *ppos;
 855
 856         copied = -ESRCH;
 857         if (!task)
 858                 goto out_no_task;
 859
 860         if (check_mem_permission(task))
 861                 goto out;
 862
 863         copied = -ENOMEM;
 864         page = (char *)__get_free_page(GFP_TEMPORARY);
 865         if (!page)
 866                 goto out;
 867
 868         copied = 0;
 869         while (count > 0) {
 870                 int this_len, retval;
 871
 872                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 873                 if (copy_from_user(page, buf, this_len)) {
 874                         copied = -EFAULT;
 875                         break;
 876                 }
 877                 retval = access_process_vm(task, dst, page, this_len, 1);
 878                 if (!retval) {
 879                         if (!copied)
 880                                 copied = -EIO;
 881                         break;
 882                 }
 883                 copied += retval;
 884                 buf += retval;
 885                 dst += retval;
 886                 count -= retval;
 887         }
 888         *ppos = dst;
 889         free_page((unsigned long) page);
 890 out:
 891         put_task_struct(task);
 892 out_no_task:
 893         return copied;
 894 }
 895 #endif
 896
 897 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 898 {
 899         switch (orig) {
 900         case 0:
 901                 file->f_pos = offset;
 902                 break;
 903         case 1:
 904                 file->f_pos += offset;
 905                 break;
 906         default:
 907                 return -EINVAL;
 908         }
 909         force_successful_syscall_return();
 910         return file->f_pos;
 911 }
 912
 913 static const struct file_operations proc_mem_operations = {
 914         .llseek         = mem_lseek,
 915         .read           = mem_read,
 916         .write          = mem_write,
 917         .open           = mem_open,
 918 };
 919
 920 static ssize_t environ_read(struct file *file, char __user *buf,
 921                         size_t count, loff_t *ppos)
 922 {
 923         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 924         char *page;
 925         unsigned long src = *ppos;
 926         int ret = -ESRCH;
 927         struct mm_struct *mm;
 928
 929         if (!task)
 930                 goto out_no_task;
 931
 932         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 933                 goto out;
 934
 935         ret = -ENOMEM;
 936         page = (char *)__get_free_page(GFP_TEMPORARY);
 937         if (!page)
 938                 goto out;
 939
 940         ret = 0;
 941
 942         mm = get_task_mm(task);
 943         if (!mm)
 944                 goto out_free;
 945
 946         while (count > 0) {
 947                 int this_len, retval, max_len;
 948
 949                 this_len = mm->env_end - (mm->env_start + src);
 950
 951                 if (this_len <= 0)
 952                         break;
 953
 954                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 955                 this_len = (this_len > max_len) ? max_len : this_len;
 956
 957                 retval = access_process_vm(task, (mm->env_start + src),
 958                         page, this_len, 0);
 959
 960                 if (retval <= 0) {
 961                         ret = retval;
 962                         break;
 963                 }
 964
 965                 if (copy_to_user(buf, page, retval)) {
 966                         ret = -EFAULT;
 967                         break;
 968                 }
 969
 970                 ret += retval;
 971                 src += retval;
 972                 buf += retval;
 973                 count -= retval;
 974         }
 975         *ppos = src;
 976
 977         mmput(mm);
 978 out_free:
 979         free_page((unsigned long) page);
 980 out:
 981         put_task_struct(task);
 982 out_no_task:
 983         return ret;
 984 }
 985
 986 static const struct file_operations proc_environ_operations = {
 987         .read           = environ_read,
 988 };
 989
 990 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 991                                 size_t count, loff_t *ppos)
 992 {
 993         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 994         char buffer[PROC_NUMBUF];
 995         size_t len;
 996         int oom_adjust = OOM_DISABLE;
 997         unsigned long flags;
 998
 999         if (!task)
1000                 return -ESRCH;
1001
1002         if (lock_task_sighand(task, &flags)) {
1003                 oom_adjust = task->signal->oom_adj;
1004                 unlock_task_sighand(task, &flags);
1005         }
1006
1007         put_task_struct(task);
1008
1009         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1010
1011         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1012 }
1013
1014 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1015                                 size_t count, loff_t *ppos)
1016 {
1017         struct task_struct *task;
1018         char buffer[PROC_NUMBUF];
1019         long oom_adjust;
1020         unsigned long flags;
1021         int err;
1022
1023         memset(buffer, 0, sizeof(buffer));
1024         if (count > sizeof(buffer) - 1)
1025                 count = sizeof(buffer) - 1;
1026         if (copy_from_user(buffer, buf, count))
1027                 return -EFAULT;
1028
1029         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1030         if (err)
1031                 return -EINVAL;
1032         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1033              oom_adjust != OOM_DISABLE)
1034                 return -EINVAL;
1035
1036         task = get_proc_task(file->f_path.dentry->d_inode);
1037         if (!task)
1038                 return -ESRCH;
1039         if (!lock_task_sighand(task, &flags)) {
1040                 put_task_struct(task);
1041                 return -ESRCH;
1042         }
1043
1044         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1045                 unlock_task_sighand(task, &flags);
1046                 put_task_struct(task);
1047                 return -EACCES;
1048         }
1049
1050         task->signal->oom_adj = oom_adjust;
1051
1052         unlock_task_sighand(task, &flags);
1053         put_task_struct(task);
1054
1055         return count;
1056 }
1057
1058 static const struct file_operations proc_oom_adjust_operations = {
1059         .read           = oom_adjust_read,
1060         .write          = oom_adjust_write,
1061 };
1062
1063 #ifdef CONFIG_AUDITSYSCALL
1064 #define TMPBUFLEN 21
1065 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1066                                   size_t count, loff_t *ppos)
1067 {
1068         struct inode * inode = file->f_path.dentry->d_inode;
1069         struct task_struct *task = get_proc_task(inode);
1070         ssize_t length;
1071         char tmpbuf[TMPBUFLEN];
1072
1073         if (!task)
1074                 return -ESRCH;
1075         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1076                                 audit_get_loginuid(task));
1077         put_task_struct(task);
1078         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1079 }
1080
1081 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1082                                    size_t count, loff_t *ppos)
1083 {
1084         struct inode * inode = file->f_path.dentry->d_inode;
1085         char *page, *tmp;
1086         ssize_t length;
1087         uid_t loginuid;
1088
1089         if (!capable(CAP_AUDIT_CONTROL))
1090                 return -EPERM;
1091
1092         rcu_read_lock();
1093         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1094                 rcu_read_unlock();
1095                 return -EPERM;
1096         }
1097         rcu_read_unlock();
1098
1099         if (count >= PAGE_SIZE)
1100                 count = PAGE_SIZE - 1;
1101
1102         if (*ppos != 0) {
1103                 /* No partial writes. */
1104                 return -EINVAL;
1105         }
1106         page = (char*)__get_free_page(GFP_TEMPORARY);
1107         if (!page)
1108                 return -ENOMEM;
1109         length = -EFAULT;
1110         if (copy_from_user(page, buf, count))
1111                 goto out_free_page;
1112
1113         page[count] = '\0';
1114         loginuid = simple_strtoul(page, &tmp, 10);
1115         if (tmp == page) {
1116                 length = -EINVAL;
1117                 goto out_free_page;
1118
1119         }
1120         length = audit_set_loginuid(current, loginuid);
1121         if (likely(length == 0))
1122                 length = count;
1123
1124 out_free_page:
1125         free_page((unsigned long) page);
1126         return length;
1127 }
1128
1129 static const struct file_operations proc_loginuid_operations = {
1130         .read           = proc_loginuid_read,
1131         .write          = proc_loginuid_write,
1132 };
1133
1134 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1135                                   size_t count, loff_t *ppos)
1136 {
1137         struct inode * inode = file->f_path.dentry->d_inode;
1138         struct task_struct *task = get_proc_task(inode);
1139         ssize_t length;
1140         char tmpbuf[TMPBUFLEN];
1141
1142         if (!task)
1143                 return -ESRCH;
1144         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1145                                 audit_get_sessionid(task));
1146         put_task_struct(task);
1147         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1148 }
1149
1150 static const struct file_operations proc_sessionid_operations = {
1151         .read           = proc_sessionid_read,
1152 };
1153 #endif
1154
1155 #ifdef CONFIG_FAULT_INJECTION
1156 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1157                                       size_t count, loff_t *ppos)
1158 {
1159         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1160         char buffer[PROC_NUMBUF];
1161         size_t len;
1162         int make_it_fail;
1163
1164         if (!task)
1165                 return -ESRCH;
1166         make_it_fail = task->make_it_fail;
1167         put_task_struct(task);
1168
1169         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1170
1171         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1172 }
1173
1174 static ssize_t proc_fault_inject_write(struct file * file,
1175                         const char __user * buf, size_t count, loff_t *ppos)
1176 {
1177         struct task_struct *task;
1178         char buffer[PROC_NUMBUF], *end;
1179         int make_it_fail;
1180
1181         if (!capable(CAP_SYS_RESOURCE))
1182                 return -EPERM;
1183         memset(buffer, 0, sizeof(buffer));
1184         if (count > sizeof(buffer) - 1)
1185                 count = sizeof(buffer) - 1;
1186         if (copy_from_user(buffer, buf, count))
1187                 return -EFAULT;
1188         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1189         if (*end)
1190                 return -EINVAL;
1191         task = get_proc_task(file->f_dentry->d_inode);
1192         if (!task)
1193                 return -ESRCH;
1194         task->make_it_fail = make_it_fail;
1195         put_task_struct(task);
1196
1197         return count;
1198 }
1199
1200 static const struct file_operations proc_fault_inject_operations = {
1201         .read           = proc_fault_inject_read,
1202         .write          = proc_fault_inject_write,
1203 };
1204 #endif
1205
1206
1207 #ifdef CONFIG_SCHED_DEBUG
1208 /*
1209  * Print out various scheduling related per-task fields:
1210  */
1211 static int sched_show(struct seq_file *m, void *v)
1212 {
1213         struct inode *inode = m->private;
1214         struct task_struct *p;
1215
1216         p = get_proc_task(inode);
1217         if (!p)
1218                 return -ESRCH;
1219         proc_sched_show_task(p, m);
1220
1221         put_task_struct(p);
1222
1223         return 0;
1224 }
1225
1226 static ssize_t
1227 sched_write(struct file *file, const char __user *buf,
1228             size_t count, loff_t *offset)
1229 {
1230         struct inode *inode = file->f_path.dentry->d_inode;
1231         struct task_struct *p;
1232
1233         p = get_proc_task(inode);
1234         if (!p)
1235                 return -ESRCH;
1236         proc_sched_set_task(p);
1237
1238         put_task_struct(p);
1239
1240         return count;
1241 }
1242
1243 static int sched_open(struct inode *inode, struct file *filp)
1244 {
1245         int ret;
1246
1247         ret = single_open(filp, sched_show, NULL);
1248         if (!ret) {
1249                 struct seq_file *m = filp->private_data;
1250
1251                 m->private = inode;
1252         }
1253         return ret;
1254 }
1255
1256 static const struct file_operations proc_pid_sched_operations = {
1257         .open           = sched_open,
1258         .read           = seq_read,
1259         .write          = sched_write,
1260         .llseek         = seq_lseek,
1261         .release        = single_release,
1262 };
1263
1264 #endif
1265
1266 static ssize_t comm_write(struct file *file, const char __user *buf,
1267                                 size_t count, loff_t *offset)
1268 {
1269         struct inode *inode = file->f_path.dentry->d_inode;
1270         struct task_struct *p;
1271         char buffer[TASK_COMM_LEN];
1272
1273         memset(buffer, 0, sizeof(buffer));
1274         if (count > sizeof(buffer) - 1)
1275                 count = sizeof(buffer) - 1;
1276         if (copy_from_user(buffer, buf, count))
1277                 return -EFAULT;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282
1283         if (same_thread_group(current, p))
1284                 set_task_comm(p, buffer);
1285         else
1286                 count = -EINVAL;
1287
1288         put_task_struct(p);
1289
1290         return count;
1291 }
1292
1293 static int comm_show(struct seq_file *m, void *v)
1294 {
1295         struct inode *inode = m->private;
1296         struct task_struct *p;
1297
1298         p = get_proc_task(inode);
1299         if (!p)
1300                 return -ESRCH;
1301
1302         task_lock(p);
1303         seq_printf(m, "%s\n", p->comm);
1304         task_unlock(p);
1305
1306         put_task_struct(p);
1307
1308         return 0;
1309 }
1310
1311 static int comm_open(struct inode *inode, struct file *filp)
1312 {
1313         int ret;
1314
1315         ret = single_open(filp, comm_show, NULL);
1316         if (!ret) {
1317                 struct seq_file *m = filp->private_data;
1318
1319                 m->private = inode;
1320         }
1321         return ret;
1322 }
1323
1324 static const struct file_operations proc_pid_set_comm_operations = {
1325         .open           = comm_open,
1326         .read           = seq_read,
1327         .write          = comm_write,
1328         .llseek         = seq_lseek,
1329         .release        = single_release,
1330 };
1331
1332 /*
1333  * We added or removed a vma mapping the executable. The vmas are only mapped
1334  * during exec and are not mapped with the mmap system call.
1335  * Callers must hold down_write() on the mm's mmap_sem for these
1336  */
1337 void added_exe_file_vma(struct mm_struct *mm)
1338 {
1339         mm->num_exe_file_vmas++;
1340 }
1341
1342 void removed_exe_file_vma(struct mm_struct *mm)
1343 {
1344         mm->num_exe_file_vmas--;
1345         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1346                 fput(mm->exe_file);
1347                 mm->exe_file = NULL;
1348         }
1349
1350 }
1351
1352 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1353 {
1354         if (new_exe_file)
1355                 get_file(new_exe_file);
1356         if (mm->exe_file)
1357                 fput(mm->exe_file);
1358         mm->exe_file = new_exe_file;
1359         mm->num_exe_file_vmas = 0;
1360 }
1361
1362 struct file *get_mm_exe_file(struct mm_struct *mm)
1363 {
1364         struct file *exe_file;
1365
1366         /* We need mmap_sem to protect against races with removal of
1367          * VM_EXECUTABLE vmas */
1368         down_read(&mm->mmap_sem);
1369         exe_file = mm->exe_file;
1370         if (exe_file)
1371                 get_file(exe_file);
1372         up_read(&mm->mmap_sem);
1373         return exe_file;
1374 }
1375
1376 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1377 {
1378         /* It's safe to write the exe_file pointer without exe_file_lock because
1379          * this is called during fork when the task is not yet in /proc */
1380         newmm->exe_file = get_mm_exe_file(oldmm);
1381 }
1382
1383 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1384 {
1385         struct task_struct *task;
1386         struct mm_struct *mm;
1387         struct file *exe_file;
1388
1389         task = get_proc_task(inode);
1390         if (!task)
1391                 return -ENOENT;
1392         mm = get_task_mm(task);
1393         put_task_struct(task);
1394         if (!mm)
1395                 return -ENOENT;
1396         exe_file = get_mm_exe_file(mm);
1397         mmput(mm);
1398         if (exe_file) {
1399                 *exe_path = exe_file->f_path;
1400                 path_get(&exe_file->f_path);
1401                 fput(exe_file);
1402                 return 0;
1403         } else
1404                 return -ENOENT;
1405 }
1406
1407 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1408 {
1409         struct inode *inode = dentry->d_inode;
1410         int error = -EACCES;
1411
1412         /* We don't need a base pointer in the /proc filesystem */
1413         path_put(&nd->path);
1414
1415         /* Are we allowed to snoop on the tasks file descriptors? */
1416         if (!proc_fd_access_allowed(inode))
1417                 goto out;
1418
1419         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1420 out:
1421         return ERR_PTR(error);
1422 }
1423
1424 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1425 {
1426         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1427         char *pathname;
1428         int len;
1429
1430         if (!tmp)
1431                 return -ENOMEM;
1432
1433         pathname = d_path(path, tmp, PAGE_SIZE);
1434         len = PTR_ERR(pathname);
1435         if (IS_ERR(pathname))
1436                 goto out;
1437         len = tmp + PAGE_SIZE - 1 - pathname;
1438
1439         if (len > buflen)
1440                 len = buflen;
1441         if (copy_to_user(buffer, pathname, len))
1442                 len = -EFAULT;
1443  out:
1444         free_page((unsigned long)tmp);
1445         return len;
1446 }
1447
1448 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1449 {
1450         int error = -EACCES;
1451         struct inode *inode = dentry->d_inode;
1452         struct path path;
1453
1454         /* Are we allowed to snoop on the tasks file descriptors? */
1455         if (!proc_fd_access_allowed(inode))
1456                 goto out;
1457
1458         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1459         if (error)
1460                 goto out;
1461
1462         error = do_proc_readlink(&path, buffer, buflen);
1463         path_put(&path);
1464 out:
1465         return error;
1466 }
1467
1468 static const struct inode_operations proc_pid_link_inode_operations = {
1469         .readlink       = proc_pid_readlink,
1470         .follow_link    = proc_pid_follow_link,
1471         .setattr        = proc_setattr,
1472 };
1473
1474
1475 /* building an inode */
1476
1477 static int task_dumpable(struct task_struct *task)
1478 {
1479         int dumpable = 0;
1480         struct mm_struct *mm;
1481
1482         task_lock(task);
1483         mm = task->mm;
1484         if (mm)
1485                 dumpable = get_dumpable(mm);
1486         task_unlock(task);
1487         if(dumpable == 1)
1488                 return 1;
1489         return 0;
1490 }
1491
1492
1493 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1494 {
1495         struct inode * inode;
1496         struct proc_inode *ei;
1497         const struct cred *cred;
1498
1499         /* We need a new inode */
1500
1501         inode = new_inode(sb);
1502         if (!inode)
1503                 goto out;
1504
1505         /* Common stuff */
1506         ei = PROC_I(inode);
1507         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1508         inode->i_op = &proc_def_inode_operations;
1509
1510         /*
1511          * grab the reference to task.
1512          */
1513         ei->pid = get_task_pid(task, PIDTYPE_PID);
1514         if (!ei->pid)
1515                 goto out_unlock;
1516
1517         if (task_dumpable(task)) {
1518                 rcu_read_lock();
1519                 cred = __task_cred(task);
1520                 inode->i_uid = cred->euid;
1521                 inode->i_gid = cred->egid;
1522                 rcu_read_unlock();
1523         }
1524         security_task_to_inode(task, inode);
1525
1526 out:
1527         return inode;
1528
1529 out_unlock:
1530         iput(inode);
1531         return NULL;
1532 }
1533
1534 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1535 {
1536         struct inode *inode = dentry->d_inode;
1537         struct task_struct *task;
1538         const struct cred *cred;
1539
1540         generic_fillattr(inode, stat);
1541
1542         rcu_read_lock();
1543         stat->uid = 0;
1544         stat->gid = 0;
1545         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1546         if (task) {
1547                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1548                     task_dumpable(task)) {
1549                         cred = __task_cred(task);
1550                         stat->uid = cred->euid;
1551                         stat->gid = cred->egid;
1552                 }
1553         }
1554         rcu_read_unlock();
1555         return 0;
1556 }
1557
1558 /* dentry stuff */
1559
1560 /*
1561  *      Exceptional case: normally we are not allowed to unhash a busy
1562  * directory. In this case, however, we can do it - no aliasing problems
1563  * due to the way we treat inodes.
1564  *
1565  * Rewrite the inode's ownerships here because the owning task may have
1566  * performed a setuid(), etc.
1567  *
1568  * Before the /proc/pid/status file was created the only way to read
1569  * the effective uid of a /process was to stat /proc/pid.  Reading
1570  * /proc/pid/status is slow enough that procps and other packages
1571  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1572  * made this apply to all per process world readable and executable
1573  * directories.
1574  */
1575 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1576 {
1577         struct inode *inode = dentry->d_inode;
1578         struct task_struct *task = get_proc_task(inode);
1579         const struct cred *cred;
1580
1581         if (task) {
1582                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1583                     task_dumpable(task)) {
1584                         rcu_read_lock();
1585                         cred = __task_cred(task);
1586                         inode->i_uid = cred->euid;
1587                         inode->i_gid = cred->egid;
1588                         rcu_read_unlock();
1589                 } else {
1590                         inode->i_uid = 0;
1591                         inode->i_gid = 0;
1592                 }
1593                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1594                 security_task_to_inode(task, inode);
1595                 put_task_struct(task);
1596                 return 1;
1597         }
1598         d_drop(dentry);
1599         return 0;
1600 }
1601
1602 static int pid_delete_dentry(struct dentry * dentry)
1603 {
1604         /* Is the task we represent dead?
1605          * If so, then don't put the dentry on the lru list,
1606          * kill it immediately.
1607          */
1608         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1609 }
1610
1611 static const struct dentry_operations pid_dentry_operations =
1612 {
1613         .d_revalidate   = pid_revalidate,
1614         .d_delete       = pid_delete_dentry,
1615 };
1616
1617 /* Lookups */
1618
1619 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1620                                 struct task_struct *, const void *);
1621
1622 /*
1623  * Fill a directory entry.
1624  *
1625  * If possible create the dcache entry and derive our inode number and
1626  * file type from dcache entry.
1627  *
1628  * Since all of the proc inode numbers are dynamically generated, the inode
1629  * numbers do not exist until the inode is cache.  This means creating the
1630  * the dcache entry in readdir is necessary to keep the inode numbers
1631  * reported by readdir in sync with the inode numbers reported
1632  * by stat.
1633  */
1634 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1635         char *name, int len,
1636         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1637 {
1638         struct dentry *child, *dir = filp->f_path.dentry;
1639         struct inode *inode;
1640         struct qstr qname;
1641         ino_t ino = 0;
1642         unsigned type = DT_UNKNOWN;
1643
1644         qname.name = name;
1645         qname.len  = len;
1646         qname.hash = full_name_hash(name, len);
1647
1648         child = d_lookup(dir, &qname);
1649         if (!child) {
1650                 struct dentry *new;
1651                 new = d_alloc(dir, &qname);
1652                 if (new) {
1653                         child = instantiate(dir->d_inode, new, task, ptr);
1654                         if (child)
1655                                 dput(new);
1656                         else
1657                                 child = new;
1658                 }
1659         }
1660         if (!child || IS_ERR(child) || !child->d_inode)
1661                 goto end_instantiate;
1662         inode = child->d_inode;
1663         if (inode) {
1664                 ino = inode->i_ino;
1665                 type = inode->i_mode >> 12;
1666         }
1667         dput(child);
1668 end_instantiate:
1669         if (!ino)
1670                 ino = find_inode_number(dir, &qname);
1671         if (!ino)
1672                 ino = 1;
1673         return filldir(dirent, name, len, filp->f_pos, ino, type);
1674 }
1675
1676 static unsigned name_to_int(struct dentry *dentry)
1677 {
1678         const char *name = dentry->d_name.name;
1679         int len = dentry->d_name.len;
1680         unsigned n = 0;
1681
1682         if (len > 1 && *name == '0')
1683                 goto out;
1684         while (len-- > 0) {
1685                 unsigned c = *name++ - '0';
1686                 if (c > 9)
1687                         goto out;
1688                 if (n >= (~0U-9)/10)
1689                         goto out;
1690                 n *= 10;
1691                 n += c;
1692         }
1693         return n;
1694 out:
1695         return ~0U;
1696 }
1697
1698 #define PROC_FDINFO_MAX 64
1699
1700 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1701 {
1702         struct task_struct *task = get_proc_task(inode);
1703         struct files_struct *files = NULL;
1704         struct file *file;
1705         int fd = proc_fd(inode);
1706
1707         if (task) {
1708                 files = get_files_struct(task);
1709                 put_task_struct(task);
1710         }
1711         if (files) {
1712                 /*
1713                  * We are not taking a ref to the file structure, so we must
1714                  * hold ->file_lock.
1715                  */
1716                 spin_lock(&files->file_lock);
1717                 file = fcheck_files(files, fd);
1718                 if (file) {
1719                         if (path) {
1720                                 *path = file->f_path;
1721                                 path_get(&file->f_path);
1722                         }
1723                         if (info)
1724                                 snprintf(info, PROC_FDINFO_MAX,
1725                                          "pos:\t%lli\n"
1726                                          "flags:\t0%o\n",
1727                                          (long long) file->f_pos,
1728                                          file->f_flags);
1729                         spin_unlock(&files->file_lock);
1730                         put_files_struct(files);
1731                         return 0;
1732                 }
1733                 spin_unlock(&files->file_lock);
1734                 put_files_struct(files);
1735         }
1736         return -ENOENT;
1737 }
1738
1739 static int proc_fd_link(struct inode *inode, struct path *path)
1740 {
1741         return proc_fd_info(inode, path, NULL);
1742 }
1743
1744 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1745 {
1746         struct inode *inode = dentry->d_inode;
1747         struct task_struct *task = get_proc_task(inode);
1748         int fd = proc_fd(inode);
1749         struct files_struct *files;
1750         const struct cred *cred;
1751
1752         if (task) {
1753                 files = get_files_struct(task);
1754                 if (files) {
1755                         rcu_read_lock();
1756                         if (fcheck_files(files, fd)) {
1757                                 rcu_read_unlock();
1758                                 put_files_struct(files);
1759                                 if (task_dumpable(task)) {
1760                                         rcu_read_lock();
1761                                         cred = __task_cred(task);
1762                                         inode->i_uid = cred->euid;
1763                                         inode->i_gid = cred->egid;
1764                                         rcu_read_unlock();
1765                                 } else {
1766                                         inode->i_uid = 0;
1767                                         inode->i_gid = 0;
1768                                 }
1769                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1770                                 security_task_to_inode(task, inode);
1771                                 put_task_struct(task);
1772                                 return 1;
1773                         }
1774                         rcu_read_unlock();
1775                         put_files_struct(files);
1776                 }
1777                 put_task_struct(task);
1778         }
1779         d_drop(dentry);
1780         return 0;
1781 }
1782
1783 static const struct dentry_operations tid_fd_dentry_operations =
1784 {
1785         .d_revalidate   = tid_fd_revalidate,
1786         .d_delete       = pid_delete_dentry,
1787 };
1788
1789 static struct dentry *proc_fd_instantiate(struct inode *dir,
1790         struct dentry *dentry, struct task_struct *task, const void *ptr)
1791 {
1792         unsigned fd = *(const unsigned *)ptr;
1793         struct file *file;
1794         struct files_struct *files;
1795         struct inode *inode;
1796         struct proc_inode *ei;
1797         struct dentry *error = ERR_PTR(-ENOENT);
1798
1799         inode = proc_pid_make_inode(dir->i_sb, task);
1800         if (!inode)
1801                 goto out;
1802         ei = PROC_I(inode);
1803         ei->fd = fd;
1804         files = get_files_struct(task);
1805         if (!files)
1806                 goto out_iput;
1807         inode->i_mode = S_IFLNK;
1808
1809         /*
1810          * We are not taking a ref to the file structure, so we must
1811          * hold ->file_lock.
1812          */
1813         spin_lock(&files->file_lock);
1814         file = fcheck_files(files, fd);
1815         if (!file)
1816                 goto out_unlock;
1817         if (file->f_mode & FMODE_READ)
1818                 inode->i_mode |= S_IRUSR | S_IXUSR;
1819         if (file->f_mode & FMODE_WRITE)
1820                 inode->i_mode |= S_IWUSR | S_IXUSR;
1821         spin_unlock(&files->file_lock);
1822         put_files_struct(files);
1823
1824         inode->i_op = &proc_pid_link_inode_operations;
1825         inode->i_size = 64;
1826         ei->op.proc_get_link = proc_fd_link;
1827         dentry->d_op = &tid_fd_dentry_operations;
1828         d_add(dentry, inode);
1829         /* Close the race of the process dying before we return the dentry */
1830         if (tid_fd_revalidate(dentry, NULL))
1831                 error = NULL;
1832
1833  out:
1834         return error;
1835 out_unlock:
1836         spin_unlock(&files->file_lock);
1837         put_files_struct(files);
1838 out_iput:
1839         iput(inode);
1840         goto out;
1841 }
1842
1843 static struct dentry *proc_lookupfd_common(struct inode *dir,
1844                                            struct dentry *dentry,
1845                                            instantiate_t instantiate)
1846 {
1847         struct task_struct *task = get_proc_task(dir);
1848         unsigned fd = name_to_int(dentry);
1849         struct dentry *result = ERR_PTR(-ENOENT);
1850
1851         if (!task)
1852                 goto out_no_task;
1853         if (fd == ~0U)
1854                 goto out;
1855
1856         result = instantiate(dir, dentry, task, &fd);
1857 out:
1858         put_task_struct(task);
1859 out_no_task:
1860         return result;
1861 }
1862
1863 static int proc_readfd_common(struct file * filp, void * dirent,
1864                               filldir_t filldir, instantiate_t instantiate)
1865 {
1866         struct dentry *dentry = filp->f_path.dentry;
1867         struct inode *inode = dentry->d_inode;
1868         struct task_struct *p = get_proc_task(inode);
1869         unsigned int fd, ino;
1870         int retval;
1871         struct files_struct * files;
1872
1873         retval = -ENOENT;
1874         if (!p)
1875                 goto out_no_task;
1876         retval = 0;
1877
1878         fd = filp->f_pos;
1879         switch (fd) {
1880                 case 0:
1881                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1882                                 goto out;
1883                         filp->f_pos++;
1884                 case 1:
1885                         ino = parent_ino(dentry);
1886                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1887                                 goto out;
1888                         filp->f_pos++;
1889                 default:
1890                         files = get_files_struct(p);
1891                         if (!files)
1892                                 goto out;
1893                         rcu_read_lock();
1894                         for (fd = filp->f_pos-2;
1895                              fd < files_fdtable(files)->max_fds;
1896                              fd++, filp->f_pos++) {
1897                                 char name[PROC_NUMBUF];
1898                                 int len;
1899
1900                                 if (!fcheck_files(files, fd))
1901                                         continue;
1902                                 rcu_read_unlock();
1903
1904                                 len = snprintf(name, sizeof(name), "%d", fd);
1905                                 if (proc_fill_cache(filp, dirent, filldir,
1906                                                     name, len, instantiate,
1907                                                     p, &fd) < 0) {
1908                                         rcu_read_lock();
1909                                         break;
1910                                 }
1911                                 rcu_read_lock();
1912                         }
1913                         rcu_read_unlock();
1914                         put_files_struct(files);
1915         }
1916 out:
1917         put_task_struct(p);
1918 out_no_task:
1919         return retval;
1920 }
1921
1922 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1923                                     struct nameidata *nd)
1924 {
1925         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1926 }
1927
1928 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1929 {
1930         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1931 }
1932
1933 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1934                                       size_t len, loff_t *ppos)
1935 {
1936         char tmp[PROC_FDINFO_MAX];
1937         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1938         if (!err)
1939                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1940         return err;
1941 }
1942
1943 static const struct file_operations proc_fdinfo_file_operations = {
1944         .open           = nonseekable_open,
1945         .read           = proc_fdinfo_read,
1946 };
1947
1948 static const struct file_operations proc_fd_operations = {
1949         .read           = generic_read_dir,
1950         .readdir        = proc_readfd,
1951 };
1952
1953 /*
1954  * /proc/pid/fd needs a special permission handler so that a process can still
1955  * access /proc/self/fd after it has executed a setuid().
1956  */
1957 static int proc_fd_permission(struct inode *inode, int mask)
1958 {
1959         int rv;
1960
1961         rv = generic_permission(inode, mask, NULL);
1962         if (rv == 0)
1963                 return 0;
1964         if (task_pid(current) == proc_pid(inode))
1965                 rv = 0;
1966         return rv;
1967 }
1968
1969 /*
1970  * proc directories can do almost nothing..
1971  */
1972 static const struct inode_operations proc_fd_inode_operations = {
1973         .lookup         = proc_lookupfd,
1974         .permission     = proc_fd_permission,
1975         .setattr        = proc_setattr,
1976 };
1977
1978 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1979         struct dentry *dentry, struct task_struct *task, const void *ptr)
1980 {
1981         unsigned fd = *(unsigned *)ptr;
1982         struct inode *inode;
1983         struct proc_inode *ei;
1984         struct dentry *error = ERR_PTR(-ENOENT);
1985
1986         inode = proc_pid_make_inode(dir->i_sb, task);
1987         if (!inode)
1988                 goto out;
1989         ei = PROC_I(inode);
1990         ei->fd = fd;
1991         inode->i_mode = S_IFREG | S_IRUSR;
1992         inode->i_fop = &proc_fdinfo_file_operations;
1993         dentry->d_op = &tid_fd_dentry_operations;
1994         d_add(dentry, inode);
1995         /* Close the race of the process dying before we return the dentry */
1996         if (tid_fd_revalidate(dentry, NULL))
1997                 error = NULL;
1998
1999  out:
2000         return error;
2001 }
2002
2003 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2004                                         struct dentry *dentry,
2005                                         struct nameidata *nd)
2006 {
2007         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2008 }
2009
2010 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2011 {
2012         return proc_readfd_common(filp, dirent, filldir,
2013                                   proc_fdinfo_instantiate);
2014 }
2015
2016 static const struct file_operations proc_fdinfo_operations = {
2017         .read           = generic_read_dir,
2018         .readdir        = proc_readfdinfo,
2019 };
2020
2021 /*
2022  * proc directories can do almost nothing..
2023  */
2024 static const struct inode_operations proc_fdinfo_inode_operations = {
2025         .lookup         = proc_lookupfdinfo,
2026         .setattr        = proc_setattr,
2027 };
2028
2029
2030 static struct dentry *proc_pident_instantiate(struct inode *dir,
2031         struct dentry *dentry, struct task_struct *task, const void *ptr)
2032 {
2033         const struct pid_entry *p = ptr;
2034         struct inode *inode;
2035         struct proc_inode *ei;
2036         struct dentry *error = ERR_PTR(-ENOENT);
2037
2038         inode = proc_pid_make_inode(dir->i_sb, task);
2039         if (!inode)
2040                 goto out;
2041
2042         ei = PROC_I(inode);
2043         inode->i_mode = p->mode;
2044         if (S_ISDIR(inode->i_mode))
2045                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2046         if (p->iop)
2047                 inode->i_op = p->iop;
2048         if (p->fop)
2049                 inode->i_fop = p->fop;
2050         ei->op = p->op;
2051         dentry->d_op = &pid_dentry_operations;
2052         d_add(dentry, inode);
2053         /* Close the race of the process dying before we return the dentry */
2054         if (pid_revalidate(dentry, NULL))
2055                 error = NULL;
2056 out:
2057         return error;
2058 }
2059
2060 static struct dentry *proc_pident_lookup(struct inode *dir,
2061                                          struct dentry *dentry,
2062                                          const struct pid_entry *ents,
2063                                          unsigned int nents)
2064 {
2065         struct dentry *error;
2066         struct task_struct *task = get_proc_task(dir);
2067         const struct pid_entry *p, *last;
2068
2069         error = ERR_PTR(-ENOENT);
2070
2071         if (!task)
2072                 goto out_no_task;
2073
2074         /*
2075          * Yes, it does not scale. And it should not. Don't add
2076          * new entries into /proc/<tgid>/ without very good reasons.
2077          */
2078         last = &ents[nents - 1];
2079         for (p = ents; p <= last; p++) {
2080                 if (p->len != dentry->d_name.len)
2081                         continue;
2082                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2083                         break;
2084         }
2085         if (p > last)
2086                 goto out;
2087
2088         error = proc_pident_instantiate(dir, dentry, task, p);
2089 out:
2090         put_task_struct(task);
2091 out_no_task:
2092         return error;
2093 }
2094
2095 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2096         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2097 {
2098         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2099                                 proc_pident_instantiate, task, p);
2100 }
2101
2102 static int proc_pident_readdir(struct file *filp,
2103                 void *dirent, filldir_t filldir,
2104                 const struct pid_entry *ents, unsigned int nents)
2105 {
2106         int i;
2107         struct dentry *dentry = filp->f_path.dentry;
2108         struct inode *inode = dentry->d_inode;
2109         struct task_struct *task = get_proc_task(inode);
2110         const struct pid_entry *p, *last;
2111         ino_t ino;
2112         int ret;
2113
2114         ret = -ENOENT;
2115         if (!task)
2116                 goto out_no_task;
2117
2118         ret = 0;
2119         i = filp->f_pos;
2120         switch (i) {
2121         case 0:
2122                 ino = inode->i_ino;
2123                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2124                         goto out;
2125                 i++;
2126                 filp->f_pos++;
2127                 /* fall through */
2128         case 1:
2129                 ino = parent_ino(dentry);
2130                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2131                         goto out;
2132                 i++;
2133                 filp->f_pos++;
2134                 /* fall through */
2135         default:
2136                 i -= 2;
2137                 if (i >= nents) {
2138                         ret = 1;
2139                         goto out;
2140                 }
2141                 p = ents + i;
2142                 last = &ents[nents - 1];
2143                 while (p <= last) {
2144                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2145                                 goto out;
2146                         filp->f_pos++;
2147                         p++;
2148                 }
2149         }
2150
2151         ret = 1;
2152 out:
2153         put_task_struct(task);
2154 out_no_task:
2155         return ret;
2156 }
2157
2158 #ifdef CONFIG_SECURITY
2159 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2160                                   size_t count, loff_t *ppos)
2161 {
2162         struct inode * inode = file->f_path.dentry->d_inode;
2163         char *p = NULL;
2164         ssize_t length;
2165         struct task_struct *task = get_proc_task(inode);
2166
2167         if (!task)
2168                 return -ESRCH;
2169
2170         length = security_getprocattr(task,
2171                                       (char*)file->f_path.dentry->d_name.name,
2172                                       &p);
2173         put_task_struct(task);
2174         if (length > 0)
2175                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2176         kfree(p);
2177         return length;
2178 }
2179
2180 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2181                                    size_t count, loff_t *ppos)
2182 {
2183         struct inode * inode = file->f_path.dentry->d_inode;
2184         char *page;
2185         ssize_t length;
2186         struct task_struct *task = get_proc_task(inode);
2187
2188         length = -ESRCH;
2189         if (!task)
2190                 goto out_no_task;
2191         if (count > PAGE_SIZE)
2192                 count = PAGE_SIZE;
2193
2194         /* No partial writes. */
2195         length = -EINVAL;
2196         if (*ppos != 0)
2197                 goto out;
2198
2199         length = -ENOMEM;
2200         page = (char*)__get_free_page(GFP_TEMPORARY);
2201         if (!page)
2202                 goto out;
2203
2204         length = -EFAULT;
2205         if (copy_from_user(page, buf, count))
2206                 goto out_free;
2207
2208         /* Guard against adverse ptrace interaction */
2209         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2210         if (length < 0)
2211                 goto out_free;
2212
2213         length = security_setprocattr(task,
2214                                       (char*)file->f_path.dentry->d_name.name,
2215                                       (void*)page, count);
2216         mutex_unlock(&task->cred_guard_mutex);
2217 out_free:
2218         free_page((unsigned long) page);
2219 out:
2220         put_task_struct(task);
2221 out_no_task:
2222         return length;
2223 }
2224
2225 static const struct file_operations proc_pid_attr_operations = {
2226         .read           = proc_pid_attr_read,
2227         .write          = proc_pid_attr_write,
2228 };
2229
2230 static const struct pid_entry attr_dir_stuff[] = {
2231         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2232         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2233         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2234         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2235         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2236         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2237 };
2238
2239 static int proc_attr_dir_readdir(struct file * filp,
2240                              void * dirent, filldir_t filldir)
2241 {
2242         return proc_pident_readdir(filp,dirent,filldir,
2243                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2244 }
2245
2246 static const struct file_operations proc_attr_dir_operations = {
2247         .read           = generic_read_dir,
2248         .readdir        = proc_attr_dir_readdir,
2249 };
2250
2251 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2252                                 struct dentry *dentry, struct nameidata *nd)
2253 {
2254         return proc_pident_lookup(dir, dentry,
2255                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2256 }
2257
2258 static const struct inode_operations proc_attr_dir_inode_operations = {
2259         .lookup         = proc_attr_dir_lookup,
2260         .getattr        = pid_getattr,
2261         .setattr        = proc_setattr,
2262 };
2263
2264 #endif
2265
2266 #ifdef CONFIG_ELF_CORE
2267 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2268                                          size_t count, loff_t *ppos)
2269 {
2270         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2271         struct mm_struct *mm;
2272         char buffer[PROC_NUMBUF];
2273         size_t len;
2274         int ret;
2275
2276         if (!task)
2277                 return -ESRCH;
2278
2279         ret = 0;
2280         mm = get_task_mm(task);
2281         if (mm) {
2282                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2283                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2284                                 MMF_DUMP_FILTER_SHIFT));
2285                 mmput(mm);
2286                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2287         }
2288
2289         put_task_struct(task);
2290
2291         return ret;
2292 }
2293
2294 static ssize_t proc_coredump_filter_write(struct file *file,
2295                                           const char __user *buf,
2296                                           size_t count,
2297                                           loff_t *ppos)
2298 {
2299         struct task_struct *task;
2300         struct mm_struct *mm;
2301         char buffer[PROC_NUMBUF], *end;
2302         unsigned int val;
2303         int ret;
2304         int i;
2305         unsigned long mask;
2306
2307         ret = -EFAULT;
2308         memset(buffer, 0, sizeof(buffer));
2309         if (count > sizeof(buffer) - 1)
2310                 count = sizeof(buffer) - 1;
2311         if (copy_from_user(buffer, buf, count))
2312                 goto out_no_task;
2313
2314         ret = -EINVAL;
2315         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2316         if (*end == '\n')
2317                 end++;
2318         if (end - buffer == 0)
2319                 goto out_no_task;
2320
2321         ret = -ESRCH;
2322         task = get_proc_task(file->f_dentry->d_inode);
2323         if (!task)
2324                 goto out_no_task;
2325
2326         ret = end - buffer;
2327         mm = get_task_mm(task);
2328         if (!mm)
2329                 goto out_no_mm;
2330
2331         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2332                 if (val & mask)
2333                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2334                 else
2335                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2336         }
2337
2338         mmput(mm);
2339  out_no_mm:
2340         put_task_struct(task);
2341  out_no_task:
2342         return ret;
2343 }
2344
2345 static const struct file_operations proc_coredump_filter_operations = {
2346         .read           = proc_coredump_filter_read,
2347         .write          = proc_coredump_filter_write,
2348 };
2349 #endif
2350
2351 /*
2352  * /proc/self:
2353  */
2354 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2355                               int buflen)
2356 {
2357         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2358         pid_t tgid = task_tgid_nr_ns(current, ns);
2359         char tmp[PROC_NUMBUF];
2360         if (!tgid)
2361                 return -ENOENT;
2362         sprintf(tmp, "%d", tgid);
2363         return vfs_readlink(dentry,buffer,buflen,tmp);
2364 }
2365
2366 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2367 {
2368         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2369         pid_t tgid = task_tgid_nr_ns(current, ns);
2370         char *name = ERR_PTR(-ENOENT);
2371         if (tgid) {
2372                 name = __getname();
2373                 if (!name)
2374                         name = ERR_PTR(-ENOMEM);
2375                 else
2376                         sprintf(name, "%d", tgid);
2377         }
2378         nd_set_link(nd, name);
2379         return NULL;
2380 }
2381
2382 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2383                                 void *cookie)
2384 {
2385         char *s = nd_get_link(nd);
2386         if (!IS_ERR(s))
2387                 __putname(s);
2388 }
2389
2390 static const struct inode_operations proc_self_inode_operations = {
2391         .readlink       = proc_self_readlink,
2392         .follow_link    = proc_self_follow_link,
2393         .put_link       = proc_self_put_link,
2394 };
2395
2396 /*
2397  * proc base
2398  *
2399  * These are the directory entries in the root directory of /proc
2400  * that properly belong to the /proc filesystem, as they describe
2401  * describe something that is process related.
2402  */
2403 static const struct pid_entry proc_base_stuff[] = {
2404         NOD("self", S_IFLNK|S_IRWXUGO,
2405                 &proc_self_inode_operations, NULL, {}),
2406 };
2407
2408 /*
2409  *      Exceptional case: normally we are not allowed to unhash a busy
2410  * directory. In this case, however, we can do it - no aliasing problems
2411  * due to the way we treat inodes.
2412  */
2413 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2414 {
2415         struct inode *inode = dentry->d_inode;
2416         struct task_struct *task = get_proc_task(inode);
2417         if (task) {
2418                 put_task_struct(task);
2419                 return 1;
2420         }
2421         d_drop(dentry);
2422         return 0;
2423 }
2424
2425 static const struct dentry_operations proc_base_dentry_operations =
2426 {
2427         .d_revalidate   = proc_base_revalidate,
2428         .d_delete       = pid_delete_dentry,
2429 };
2430
2431 static struct dentry *proc_base_instantiate(struct inode *dir,
2432         struct dentry *dentry, struct task_struct *task, const void *ptr)
2433 {
2434         const struct pid_entry *p = ptr;
2435         struct inode *inode;
2436         struct proc_inode *ei;
2437         struct dentry *error = ERR_PTR(-EINVAL);
2438
2439         /* Allocate the inode */
2440         error = ERR_PTR(-ENOMEM);
2441         inode = new_inode(dir->i_sb);
2442         if (!inode)
2443                 goto out;
2444
2445         /* Initialize the inode */
2446         ei = PROC_I(inode);
2447         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2448
2449         /*
2450          * grab the reference to the task.
2451          */
2452         ei->pid = get_task_pid(task, PIDTYPE_PID);
2453         if (!ei->pid)
2454                 goto out_iput;
2455
2456         inode->i_mode = p->mode;
2457         if (S_ISDIR(inode->i_mode))
2458                 inode->i_nlink = 2;
2459         if (S_ISLNK(inode->i_mode))
2460                 inode->i_size = 64;
2461         if (p->iop)
2462                 inode->i_op = p->iop;
2463         if (p->fop)
2464                 inode->i_fop = p->fop;
2465         ei->op = p->op;
2466         dentry->d_op = &proc_base_dentry_operations;
2467         d_add(dentry, inode);
2468         error = NULL;
2469 out:
2470         return error;
2471 out_iput:
2472         iput(inode);
2473         goto out;
2474 }
2475
2476 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2477 {
2478         struct dentry *error;
2479         struct task_struct *task = get_proc_task(dir);
2480         const struct pid_entry *p, *last;
2481
2482         error = ERR_PTR(-ENOENT);
2483
2484         if (!task)
2485                 goto out_no_task;
2486
2487         /* Lookup the directory entry */
2488         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2489         for (p = proc_base_stuff; p <= last; p++) {
2490                 if (p->len != dentry->d_name.len)
2491                         continue;
2492                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2493                         break;
2494         }
2495         if (p > last)
2496                 goto out;
2497
2498         error = proc_base_instantiate(dir, dentry, task, p);
2499
2500 out:
2501         put_task_struct(task);
2502 out_no_task:
2503         return error;
2504 }
2505
2506 static int proc_base_fill_cache(struct file *filp, void *dirent,
2507         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2508 {
2509         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2510                                 proc_base_instantiate, task, p);
2511 }
2512
2513 #ifdef CONFIG_TASK_IO_ACCOUNTING
2514 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2515 {
2516         struct task_io_accounting acct = task->ioac;
2517         unsigned long flags;
2518
2519         if (whole && lock_task_sighand(task, &flags)) {
2520                 struct task_struct *t = task;
2521
2522                 task_io_accounting_add(&acct, &task->signal->ioac);
2523                 while_each_thread(task, t)
2524                         task_io_accounting_add(&acct, &t->ioac);
2525
2526                 unlock_task_sighand(task, &flags);
2527         }
2528         return sprintf(buffer,
2529                         "rchar: %llu\n"
2530                         "wchar: %llu\n"
2531                         "syscr: %llu\n"
2532                         "syscw: %llu\n"
2533                         "read_bytes: %llu\n"
2534                         "write_bytes: %llu\n"
2535                         "cancelled_write_bytes: %llu\n",
2536                         (unsigned long long)acct.rchar,
2537                         (unsigned long long)acct.wchar,
2538                         (unsigned long long)acct.syscr,
2539                         (unsigned long long)acct.syscw,
2540                         (unsigned long long)acct.read_bytes,
2541                         (unsigned long long)acct.write_bytes,
2542                         (unsigned long long)acct.cancelled_write_bytes);
2543 }
2544
2545 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2546 {
2547         return do_io_accounting(task, buffer, 0);
2548 }
2549
2550 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2551 {
2552         return do_io_accounting(task, buffer, 1);
2553 }
2554 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2555
2556 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2557                                 struct pid *pid, struct task_struct *task)
2558 {
2559         seq_printf(m, "%08x\n", task->personality);
2560         return 0;
2561 }
2562
2563 /*
2564  * Thread groups
2565  */
2566 static const struct file_operations proc_task_operations;
2567 static const struct inode_operations proc_task_inode_operations;
2568
2569 static const struct pid_entry tgid_base_stuff[] = {
2570         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2571         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2572         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2573 #ifdef CONFIG_NET
2574         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2575 #endif
2576         REG("environ",    S_IRUSR, proc_environ_operations),
2577         INF("auxv",       S_IRUSR, proc_pid_auxv),
2578         ONE("status",     S_IRUGO, proc_pid_status),
2579         ONE("personality", S_IRUSR, proc_pid_personality),
2580         INF("limits",     S_IRUSR, proc_pid_limits),
2581 #ifdef CONFIG_SCHED_DEBUG
2582         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2583 #endif
2584         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2585 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2586         INF("syscall",    S_IRUSR, proc_pid_syscall),
2587 #endif
2588         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2589         ONE("stat",       S_IRUGO, proc_tgid_stat),
2590         ONE("statm",      S_IRUGO, proc_pid_statm),
2591         REG("maps",       S_IRUGO, proc_maps_operations),
2592 #ifdef CONFIG_NUMA
2593         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2594 #endif
2595         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2596         LNK("cwd",        proc_cwd_link),
2597         LNK("root",       proc_root_link),
2598         LNK("exe",        proc_exe_link),
2599         REG("mounts",     S_IRUGO, proc_mounts_operations),
2600         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2601         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2602 #ifdef CONFIG_PROC_PAGE_MONITOR
2603         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2604         REG("smaps",      S_IRUGO, proc_smaps_operations),
2605         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2606 #endif
2607 #ifdef CONFIG_SECURITY
2608         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2609 #endif
2610 #ifdef CONFIG_KALLSYMS
2611         INF("wchan",      S_IRUGO, proc_pid_wchan),
2612 #endif
2613 #ifdef CONFIG_STACKTRACE
2614         ONE("stack",      S_IRUSR, proc_pid_stack),
2615 #endif
2616 #ifdef CONFIG_SCHEDSTATS
2617         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2618 #endif
2619 #ifdef CONFIG_LATENCYTOP
2620         REG("latency",  S_IRUGO, proc_lstats_operations),
2621 #endif
2622 #ifdef CONFIG_PROC_PID_CPUSET
2623         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2624 #endif
2625 #ifdef CONFIG_CGROUPS
2626         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2627 #endif
2628         INF("oom_score",  S_IRUGO, proc_oom_score),
2629         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2630 #ifdef CONFIG_AUDITSYSCALL
2631         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2632         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2633 #endif
2634 #ifdef CONFIG_FAULT_INJECTION
2635         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2636 #endif
2637 #ifdef CONFIG_ELF_CORE
2638         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2639 #endif
2640 #ifdef CONFIG_TASK_IO_ACCOUNTING
2641         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2642 #endif
2643 };
2644
2645 static int proc_tgid_base_readdir(struct file * filp,
2646                              void * dirent, filldir_t filldir)
2647 {
2648         return proc_pident_readdir(filp,dirent,filldir,
2649                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2650 }
2651
2652 static const struct file_operations proc_tgid_base_operations = {
2653         .read           = generic_read_dir,
2654         .readdir        = proc_tgid_base_readdir,
2655 };
2656
2657 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2658         return proc_pident_lookup(dir, dentry,
2659                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2660 }
2661
2662 static const struct inode_operations proc_tgid_base_inode_operations = {
2663         .lookup         = proc_tgid_base_lookup,
2664         .getattr        = pid_getattr,
2665         .setattr        = proc_setattr,
2666 };
2667
2668 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2669 {
2670         struct dentry *dentry, *leader, *dir;
2671         char buf[PROC_NUMBUF];
2672         struct qstr name;
2673
2674         name.name = buf;
2675         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2676         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2677         if (dentry) {
2678                 shrink_dcache_parent(dentry);
2679                 d_drop(dentry);
2680                 dput(dentry);
2681         }
2682
2683         name.name = buf;
2684         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2685         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2686         if (!leader)
2687                 goto out;
2688
2689         name.name = "task";
2690         name.len = strlen(name.name);
2691         dir = d_hash_and_lookup(leader, &name);
2692         if (!dir)
2693                 goto out_put_leader;
2694
2695         name.name = buf;
2696         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2697         dentry = d_hash_and_lookup(dir, &name);
2698         if (dentry) {
2699                 shrink_dcache_parent(dentry);
2700                 d_drop(dentry);
2701                 dput(dentry);
2702         }
2703
2704         dput(dir);
2705 out_put_leader:
2706         dput(leader);
2707 out:
2708         return;
2709 }
2710
2711 /**
2712  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2713  * @task: task that should be flushed.
2714  *
2715  * When flushing dentries from proc, one needs to flush them from global
2716  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2717  * in. This call is supposed to do all of this job.
2718  *
2719  * Looks in the dcache for
2720  * /proc/@pid
2721  * /proc/@tgid/task/@pid
2722  * if either directory is present flushes it and all of it'ts children
2723  * from the dcache.
2724  *
2725  * It is safe and reasonable to cache /proc entries for a task until
2726  * that task exits.  After that they just clog up the dcache with
2727  * useless entries, possibly causing useful dcache entries to be
2728  * flushed instead.  This routine is proved to flush those useless
2729  * dcache entries at process exit time.
2730  *
2731  * NOTE: This routine is just an optimization so it does not guarantee
2732  *       that no dcache entries will exist at process exit time it
2733  *       just makes it very unlikely that any will persist.
2734  */
2735
2736 void proc_flush_task(struct task_struct *task)
2737 {
2738         int i;
2739         struct pid *pid, *tgid;
2740         struct upid *upid;
2741
2742         pid = task_pid(task);
2743         tgid = task_tgid(task);
2744
2745         for (i = 0; i <= pid->level; i++) {
2746                 upid = &pid->numbers[i];
2747                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2748                                         tgid->numbers[i].nr);
2749         }
2750
2751         upid = &pid->numbers[pid->level];
2752         if (upid->nr == 1)
2753                 pid_ns_release_proc(upid->ns);
2754 }
2755
2756 static struct dentry *proc_pid_instantiate(struct inode *dir,
2757                                            struct dentry * dentry,
2758                                            struct task_struct *task, const void *ptr)
2759 {
2760         struct dentry *error = ERR_PTR(-ENOENT);
2761         struct inode *inode;
2762
2763         inode = proc_pid_make_inode(dir->i_sb, task);
2764         if (!inode)
2765                 goto out;
2766
2767         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2768         inode->i_op = &proc_tgid_base_inode_operations;
2769         inode->i_fop = &proc_tgid_base_operations;
2770         inode->i_flags|=S_IMMUTABLE;
2771
2772         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2773                 ARRAY_SIZE(tgid_base_stuff));
2774
2775         dentry->d_op = &pid_dentry_operations;
2776
2777         d_add(dentry, inode);
2778         /* Close the race of the process dying before we return the dentry */
2779         if (pid_revalidate(dentry, NULL))
2780                 error = NULL;
2781 out:
2782         return error;
2783 }
2784
2785 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2786 {
2787         struct dentry *result = ERR_PTR(-ENOENT);
2788         struct task_struct *task;
2789         unsigned tgid;
2790         struct pid_namespace *ns;
2791
2792         result = proc_base_lookup(dir, dentry);
2793         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2794                 goto out;
2795
2796         tgid = name_to_int(dentry);
2797         if (tgid == ~0U)
2798                 goto out;
2799
2800         ns = dentry->d_sb->s_fs_info;
2801         rcu_read_lock();
2802         task = find_task_by_pid_ns(tgid, ns);
2803         if (task)
2804                 get_task_struct(task);
2805         rcu_read_unlock();
2806         if (!task)
2807                 goto out;
2808
2809         result = proc_pid_instantiate(dir, dentry, task, NULL);
2810         put_task_struct(task);
2811 out:
2812         return result;
2813 }
2814
2815 /*
2816  * Find the first task with tgid >= tgid
2817  *
2818  */
2819 struct tgid_iter {
2820         unsigned int tgid;
2821         struct task_struct *task;
2822 };
2823 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2824 {
2825         struct pid *pid;
2826
2827         if (iter.task)
2828                 put_task_struct(iter.task);
2829         rcu_read_lock();
2830 retry:
2831         iter.task = NULL;
2832         pid = find_ge_pid(iter.tgid, ns);
2833         if (pid) {
2834                 iter.tgid = pid_nr_ns(pid, ns);
2835                 iter.task = pid_task(pid, PIDTYPE_PID);
2836                 /* What we to know is if the pid we have find is the
2837                  * pid of a thread_group_leader.  Testing for task
2838                  * being a thread_group_leader is the obvious thing
2839                  * todo but there is a window when it fails, due to
2840                  * the pid transfer logic in de_thread.
2841                  *
2842                  * So we perform the straight forward test of seeing
2843                  * if the pid we have found is the pid of a thread
2844                  * group leader, and don't worry if the task we have
2845                  * found doesn't happen to be a thread group leader.
2846                  * As we don't care in the case of readdir.
2847                  */
2848                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2849                         iter.tgid += 1;
2850                         goto retry;
2851                 }
2852                 get_task_struct(iter.task);
2853         }
2854         rcu_read_unlock();
2855         return iter;
2856 }
2857
2858 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2859
2860 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2861         struct tgid_iter iter)
2862 {
2863         char name[PROC_NUMBUF];
2864         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2865         return proc_fill_cache(filp, dirent, filldir, name, len,
2866                                 proc_pid_instantiate, iter.task, NULL);
2867 }
2868
2869 /* for the /proc/ directory itself, after non-process stuff has been done */
2870 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2871 {
2872         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2873         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2874         struct tgid_iter iter;
2875         struct pid_namespace *ns;
2876
2877         if (!reaper)
2878                 goto out_no_task;
2879
2880         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2881                 const struct pid_entry *p = &proc_base_stuff[nr];
2882                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2883                         goto out;
2884         }
2885
2886         ns = filp->f_dentry->d_sb->s_fs_info;
2887         iter.task = NULL;
2888         iter.tgid = filp->f_pos - TGID_OFFSET;
2889         for (iter = next_tgid(ns, iter);
2890              iter.task;
2891              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2892                 filp->f_pos = iter.tgid + TGID_OFFSET;
2893                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2894                         put_task_struct(iter.task);
2895                         goto out;
2896                 }
2897         }
2898         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2899 out:
2900         put_task_struct(reaper);
2901 out_no_task:
2902         return 0;
2903 }
2904
2905 /*
2906  * Tasks
2907  */
2908 static const struct pid_entry tid_base_stuff[] = {
2909         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2910         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2911         REG("environ",   S_IRUSR, proc_environ_operations),
2912         INF("auxv",      S_IRUSR, proc_pid_auxv),
2913         ONE("status",    S_IRUGO, proc_pid_status),
2914         ONE("personality", S_IRUSR, proc_pid_personality),
2915         INF("limits",    S_IRUSR, proc_pid_limits),
2916 #ifdef CONFIG_SCHED_DEBUG
2917         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2918 #endif
2919         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2920 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2921         INF("syscall",   S_IRUSR, proc_pid_syscall),
2922 #endif
2923         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2924         ONE("stat",      S_IRUGO, proc_tid_stat),
2925         ONE("statm",     S_IRUGO, proc_pid_statm),
2926         REG("maps",      S_IRUGO, proc_maps_operations),
2927 #ifdef CONFIG_NUMA
2928         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2929 #endif
2930         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2931         LNK("cwd",       proc_cwd_link),
2932         LNK("root",      proc_root_link),
2933         LNK("exe",       proc_exe_link),
2934         REG("mounts",    S_IRUGO, proc_mounts_operations),
2935         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2936 #ifdef CONFIG_PROC_PAGE_MONITOR
2937         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2938         REG("smaps",     S_IRUGO, proc_smaps_operations),
2939         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2940 #endif
2941 #ifdef CONFIG_SECURITY
2942         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2943 #endif
2944 #ifdef CONFIG_KALLSYMS
2945         INF("wchan",     S_IRUGO, proc_pid_wchan),
2946 #endif
2947 #ifdef CONFIG_STACKTRACE
2948         ONE("stack",      S_IRUSR, proc_pid_stack),
2949 #endif
2950 #ifdef CONFIG_SCHEDSTATS
2951         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2952 #endif
2953 #ifdef CONFIG_LATENCYTOP
2954         REG("latency",  S_IRUGO, proc_lstats_operations),
2955 #endif
2956 #ifdef CONFIG_PROC_PID_CPUSET
2957         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2958 #endif
2959 #ifdef CONFIG_CGROUPS
2960         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2961 #endif
2962         INF("oom_score", S_IRUGO, proc_oom_score),
2963         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2964 #ifdef CONFIG_AUDITSYSCALL
2965         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2966         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2967 #endif
2968 #ifdef CONFIG_FAULT_INJECTION
2969         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2970 #endif
2971 #ifdef CONFIG_TASK_IO_ACCOUNTING
2972         INF("io",       S_IRUGO, proc_tid_io_accounting),
2973 #endif
2974 };
2975
2976 static int proc_tid_base_readdir(struct file * filp,
2977                              void * dirent, filldir_t filldir)
2978 {
2979         return proc_pident_readdir(filp,dirent,filldir,
2980                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2981 }
2982
2983 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2984         return proc_pident_lookup(dir, dentry,
2985                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2986 }
2987
2988 static const struct file_operations proc_tid_base_operations = {
2989         .read           = generic_read_dir,
2990         .readdir        = proc_tid_base_readdir,
2991 };
2992
2993 static const struct inode_operations proc_tid_base_inode_operations = {
2994         .lookup         = proc_tid_base_lookup,
2995         .getattr        = pid_getattr,
2996         .setattr        = proc_setattr,
2997 };
2998
2999 static struct dentry *proc_task_instantiate(struct inode *dir,
3000         struct dentry *dentry, struct task_struct *task, const void *ptr)
3001 {
3002         struct dentry *error = ERR_PTR(-ENOENT);
3003         struct inode *inode;
3004         inode = proc_pid_make_inode(dir->i_sb, task);
3005
3006         if (!inode)
3007                 goto out;
3008         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3009         inode->i_op = &proc_tid_base_inode_operations;
3010         inode->i_fop = &proc_tid_base_operations;
3011         inode->i_flags|=S_IMMUTABLE;
3012
3013         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3014                 ARRAY_SIZE(tid_base_stuff));
3015
3016         dentry->d_op = &pid_dentry_operations;
3017
3018         d_add(dentry, inode);
3019         /* Close the race of the process dying before we return the dentry */
3020         if (pid_revalidate(dentry, NULL))
3021                 error = NULL;
3022 out:
3023         return error;
3024 }
3025
3026 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3027 {
3028         struct dentry *result = ERR_PTR(-ENOENT);
3029         struct task_struct *task;
3030         struct task_struct *leader = get_proc_task(dir);
3031         unsigned tid;
3032         struct pid_namespace *ns;
3033
3034         if (!leader)
3035                 goto out_no_task;
3036
3037         tid = name_to_int(dentry);
3038         if (tid == ~0U)
3039                 goto out;
3040
3041         ns = dentry->d_sb->s_fs_info;
3042         rcu_read_lock();
3043         task = find_task_by_pid_ns(tid, ns);
3044         if (task)
3045                 get_task_struct(task);
3046         rcu_read_unlock();
3047         if (!task)
3048                 goto out;
3049         if (!same_thread_group(leader, task))
3050                 goto out_drop_task;
3051
3052         result = proc_task_instantiate(dir, dentry, task, NULL);
3053 out_drop_task:
3054         put_task_struct(task);
3055 out:
3056         put_task_struct(leader);
3057 out_no_task:
3058         return result;
3059 }
3060
3061 /*
3062  * Find the first tid of a thread group to return to user space.
3063  *
3064  * Usually this is just the thread group leader, but if the users
3065  * buffer was too small or there was a seek into the middle of the
3066  * directory we have more work todo.
3067  *
3068  * In the case of a short read we start with find_task_by_pid.
3069  *
3070  * In the case of a seek we start with the leader and walk nr
3071  * threads past it.
3072  */
3073 static struct task_struct *first_tid(struct task_struct *leader,
3074                 int tid, int nr, struct pid_namespace *ns)
3075 {
3076         struct task_struct *pos;
3077
3078         rcu_read_lock();
3079         /* Attempt to start with the pid of a thread */
3080         if (tid && (nr > 0)) {
3081                 pos = find_task_by_pid_ns(tid, ns);
3082                 if (pos && (pos->group_leader == leader))
3083                         goto found;
3084         }
3085
3086         /* If nr exceeds the number of threads there is nothing todo */
3087         pos = NULL;
3088         if (nr && nr >= get_nr_threads(leader))
3089                 goto out;
3090
3091         /* If we haven't found our starting place yet start
3092          * with the leader and walk nr threads forward.
3093          */
3094         for (pos = leader; nr > 0; --nr) {
3095                 pos = next_thread(pos);
3096                 if (pos == leader) {
3097                         pos = NULL;
3098                         goto out;
3099                 }
3100         }
3101 found:
3102         get_task_struct(pos);
3103 out:
3104         rcu_read_unlock();
3105         return pos;
3106 }
3107
3108 /*
3109  * Find the next thread in the thread list.
3110  * Return NULL if there is an error or no next thread.
3111  *
3112  * The reference to the input task_struct is released.
3113  */
3114 static struct task_struct *next_tid(struct task_struct *start)
3115 {
3116         struct task_struct *pos = NULL;
3117         rcu_read_lock();
3118         if (pid_alive(start)) {
3119                 pos = next_thread(start);
3120                 if (thread_group_leader(pos))
3121                         pos = NULL;
3122                 else
3123                         get_task_struct(pos);
3124         }
3125         rcu_read_unlock();
3126         put_task_struct(start);
3127         return pos;
3128 }
3129
3130 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3131         struct task_struct *task, int tid)
3132 {
3133         char name[PROC_NUMBUF];
3134         int len = snprintf(name, sizeof(name), "%d", tid);
3135         return proc_fill_cache(filp, dirent, filldir, name, len,
3136                                 proc_task_instantiate, task, NULL);
3137 }
3138
3139 /* for the /proc/TGID/task/ directories */
3140 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3141 {
3142         struct dentry *dentry = filp->f_path.dentry;
3143         struct inode *inode = dentry->d_inode;
3144         struct task_struct *leader = NULL;
3145         struct task_struct *task;
3146         int retval = -ENOENT;
3147         ino_t ino;
3148         int tid;
3149         struct pid_namespace *ns;
3150
3151         task = get_proc_task(inode);
3152         if (!task)
3153                 goto out_no_task;
3154         rcu_read_lock();
3155         if (pid_alive(task)) {
3156                 leader = task->group_leader;
3157                 get_task_struct(leader);
3158         }
3159         rcu_read_unlock();
3160         put_task_struct(task);
3161         if (!leader)
3162                 goto out_no_task;
3163         retval = 0;
3164
3165         switch ((unsigned long)filp->f_pos) {
3166         case 0:
3167                 ino = inode->i_ino;
3168                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3169                         goto out;
3170                 filp->f_pos++;
3171                 /* fall through */
3172         case 1:
3173                 ino = parent_ino(dentry);
3174                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3175                         goto out;
3176                 filp->f_pos++;
3177                 /* fall through */
3178         }
3179
3180         /* f_version caches the tgid value that the last readdir call couldn't
3181          * return. lseek aka telldir automagically resets f_version to 0.
3182          */
3183         ns = filp->f_dentry->d_sb->s_fs_info;
3184         tid = (int)filp->f_version;
3185         filp->f_version = 0;
3186         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3187              task;
3188              task = next_tid(task), filp->f_pos++) {
3189                 tid = task_pid_nr_ns(task, ns);
3190                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3191                         /* returning this tgid failed, save it as the first
3192                          * pid for the next readir call */
3193                         filp->f_version = (u64)tid;
3194                         put_task_struct(task);
3195                         break;
3196                 }
3197         }
3198 out:
3199         put_task_struct(leader);
3200 out_no_task:
3201         return retval;
3202 }
3203
3204 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3205 {
3206         struct inode *inode = dentry->d_inode;
3207         struct task_struct *p = get_proc_task(inode);
3208         generic_fillattr(inode, stat);
3209
3210         if (p) {
3211                 stat->nlink += get_nr_threads(p);
3212                 put_task_struct(p);
3213         }
3214
3215         return 0;
3216 }
3217
3218 static const struct inode_operations proc_task_inode_operations = {
3219         .lookup         = proc_task_lookup,
3220         .getattr        = proc_task_getattr,
3221         .setattr        = proc_setattr,
3222 };
3223
3224 static const struct file_operations proc_task_operations = {
3225         .read           = generic_read_dir,
3226         .readdir        = proc_task_readdir,
3227 };