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/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #ifdef CONFIG_HARDWALL
  87 #include <asm/hardwall.h>
  88 #endif
  89 #include "internal.h"
  90
  91 /* NOTE:
  92  *      Implementing inode permission operations in /proc is almost
  93  *      certainly an error.  Permission checks need to happen during
  94  *      each system call not at open time.  The reason is that most of
  95  *      what we wish to check for permissions in /proc varies at runtime.
  96  *
  97  *      The classic example of a problem is opening file descriptors
  98  *      in /proc for a task before it execs a suid executable.
  99  */
 100
 101 struct pid_entry {
 102         char *name;
 103         int len;
 104         mode_t mode;
 105         const struct inode_operations *iop;
 106         const struct file_operations *fop;
 107         union proc_op op;
 108 };
 109
 110 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 111         .name = (NAME),                                 \
 112         .len  = sizeof(NAME) - 1,                       \
 113         .mode = MODE,                                   \
 114         .iop  = IOP,                                    \
 115         .fop  = FOP,                                    \
 116         .op   = OP,                                     \
 117 }
 118
 119 #define DIR(NAME, MODE, iops, fops)     \
 120         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 121 #define LNK(NAME, get_link)                                     \
 122         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 123                 &proc_pid_link_inode_operations, NULL,          \
 124                 { .proc_get_link = get_link } )
 125 #define REG(NAME, MODE, fops)                           \
 126         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 127 #define INF(NAME, MODE, read)                           \
 128         NOD(NAME, (S_IFREG|(MODE)),                     \
 129                 NULL, &proc_info_file_operations,       \
 130                 { .proc_read = read } )
 131 #define ONE(NAME, MODE, show)                           \
 132         NOD(NAME, (S_IFREG|(MODE)),                     \
 133                 NULL, &proc_single_file_operations,     \
 134                 { .proc_show = show } )
 135
 136 /*
 137  * Count the number of hardlinks for the pid_entry table, excluding the .
 138  * and .. links.
 139  */
 140 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 141         unsigned int n)
 142 {
 143         unsigned int i;
 144         unsigned int count;
 145
 146         count = 0;
 147         for (i = 0; i < n; ++i) {
 148                 if (S_ISDIR(entries[i].mode))
 149                         ++count;
 150         }
 151
 152         return count;
 153 }
 154
 155 static int get_task_root(struct task_struct *task, struct path *root)
 156 {
 157         int result = -ENOENT;
 158
 159         task_lock(task);
 160         if (task->fs) {
 161                 get_fs_root(task->fs, root);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int proc_cwd_link(struct inode *inode, struct path *path)
 169 {
 170         struct task_struct *task = get_proc_task(inode);
 171         int result = -ENOENT;
 172
 173         if (task) {
 174                 task_lock(task);
 175                 if (task->fs) {
 176                         get_fs_pwd(task->fs, path);
 177                         result = 0;
 178                 }
 179                 task_unlock(task);
 180                 put_task_struct(task);
 181         }
 182         return result;
 183 }
 184
 185 static int proc_root_link(struct inode *inode, struct path *path)
 186 {
 187         struct task_struct *task = get_proc_task(inode);
 188         int result = -ENOENT;
 189
 190         if (task) {
 191                 result = get_task_root(task, path);
 192                 put_task_struct(task);
 193         }
 194         return result;
 195 }
 196
 197 static struct mm_struct *__check_mem_permission(struct task_struct *task)
 198 {
 199         struct mm_struct *mm;
 200
 201         mm = get_task_mm(task);
 202         if (!mm)
 203                 return ERR_PTR(-EINVAL);
 204
 205         /*
 206          * A task can always look at itself, in case it chooses
 207          * to use system calls instead of load instructions.
 208          */
 209         if (task == current)
 210                 return mm;
 211
 212         /*
 213          * If current is actively ptrace'ing, and would also be
 214          * permitted to freshly attach with ptrace now, permit it.
 215          */
 216         if (task_is_stopped_or_traced(task)) {
 217                 int match;
 218                 rcu_read_lock();
 219                 match = (ptrace_parent(task) == current);
 220                 rcu_read_unlock();
 221                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 222                         return mm;
 223         }
 224
 225         /*
 226          * No one else is allowed.
 227          */
 228         mmput(mm);
 229         return ERR_PTR(-EPERM);
 230 }
 231
 232 /*
 233  * If current may access user memory in @task return a reference to the
 234  * corresponding mm, otherwise ERR_PTR.
 235  */
 236 static struct mm_struct *check_mem_permission(struct task_struct *task)
 237 {
 238         struct mm_struct *mm;
 239         int err;
 240
 241         /*
 242          * Avoid racing if task exec's as we might get a new mm but validate
 243          * against old credentials.
 244          */
 245         err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 246         if (err)
 247                 return ERR_PTR(err);
 248
 249         mm = __check_mem_permission(task);
 250         mutex_unlock(&task->signal->cred_guard_mutex);
 251
 252         return mm;
 253 }
 254
 255 struct mm_struct *mm_for_maps(struct task_struct *task)
 256 {
 257         struct mm_struct *mm;
 258         int err;
 259
 260         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 261         if (err)
 262                 return ERR_PTR(err);
 263
 264         mm = get_task_mm(task);
 265         if (mm && mm != current->mm &&
 266                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 267                 mmput(mm);
 268                 mm = ERR_PTR(-EACCES);
 269         }
 270         mutex_unlock(&task->signal->cred_guard_mutex);
 271
 272         return mm;
 273 }
 274
 275 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 276 {
 277         int res = 0;
 278         unsigned int len;
 279         struct mm_struct *mm = get_task_mm(task);
 280         if (!mm)
 281                 goto out;
 282         if (!mm->arg_end)
 283                 goto out_mm;    /* Shh! No looking before we're done */
 284
 285         len = mm->arg_end - mm->arg_start;
 286
 287         if (len > PAGE_SIZE)
 288                 len = PAGE_SIZE;
 289
 290         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 291
 292         // If the nul at the end of args has been overwritten, then
 293         // assume application is using setproctitle(3).
 294         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 295                 len = strnlen(buffer, res);
 296                 if (len < res) {
 297                     res = len;
 298                 } else {
 299                         len = mm->env_end - mm->env_start;
 300                         if (len > PAGE_SIZE - res)
 301                                 len = PAGE_SIZE - res;
 302                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 303                         res = strnlen(buffer, res);
 304                 }
 305         }
 306 out_mm:
 307         mmput(mm);
 308 out:
 309         return res;
 310 }
 311
 312 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 313 {
 314         struct mm_struct *mm = mm_for_maps(task);
 315         int res = PTR_ERR(mm);
 316         if (mm && !IS_ERR(mm)) {
 317                 unsigned int nwords = 0;
 318                 do {
 319                         nwords += 2;
 320                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 321                 res = nwords * sizeof(mm->saved_auxv[0]);
 322                 if (res > PAGE_SIZE)
 323                         res = PAGE_SIZE;
 324                 memcpy(buffer, mm->saved_auxv, res);
 325                 mmput(mm);
 326         }
 327         return res;
 328 }
 329
 330
 331 #ifdef CONFIG_KALLSYMS
 332 /*
 333  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 334  * Returns the resolved symbol.  If that fails, simply return the address.
 335  */
 336 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 337 {
 338         unsigned long wchan;
 339         char symname[KSYM_NAME_LEN];
 340
 341         wchan = get_wchan(task);
 342
 343         if (lookup_symbol_name(wchan, symname) < 0)
 344                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 345                         return 0;
 346                 else
 347                         return sprintf(buffer, "%lu", wchan);
 348         else
 349                 return sprintf(buffer, "%s", symname);
 350 }
 351 #endif /* CONFIG_KALLSYMS */
 352
 353 static int lock_trace(struct task_struct *task)
 354 {
 355         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 356         if (err)
 357                 return err;
 358         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 359                 mutex_unlock(&task->signal->cred_guard_mutex);
 360                 return -EPERM;
 361         }
 362         return 0;
 363 }
 364
 365 static void unlock_trace(struct task_struct *task)
 366 {
 367         mutex_unlock(&task->signal->cred_guard_mutex);
 368 }
 369
 370 #ifdef CONFIG_STACKTRACE
 371
 372 #define MAX_STACK_TRACE_DEPTH   64
 373
 374 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 375                           struct pid *pid, struct task_struct *task)
 376 {
 377         struct stack_trace trace;
 378         unsigned long *entries;
 379         int err;
 380         int i;
 381
 382         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 383         if (!entries)
 384                 return -ENOMEM;
 385
 386         trace.nr_entries        = 0;
 387         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 388         trace.entries           = entries;
 389         trace.skip              = 0;
 390
 391         err = lock_trace(task);
 392         if (!err) {
 393                 save_stack_trace_tsk(task, &trace);
 394
 395                 for (i = 0; i < trace.nr_entries; i++) {
 396                         seq_printf(m, "[<%pK>] %pS\n",
 397                                    (void *)entries[i], (void *)entries[i]);
 398                 }
 399                 unlock_trace(task);
 400         }
 401         kfree(entries);
 402
 403         return err;
 404 }
 405 #endif
 406
 407 #ifdef CONFIG_SCHEDSTATS
 408 /*
 409  * Provides /proc/PID/schedstat
 410  */
 411 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 412 {
 413         return sprintf(buffer, "%llu %llu %lu\n",
 414                         (unsigned long long)task->se.sum_exec_runtime,
 415                         (unsigned long long)task->sched_info.run_delay,
 416                         task->sched_info.pcount);
 417 }
 418 #endif
 419
 420 #ifdef CONFIG_LATENCYTOP
 421 static int lstats_show_proc(struct seq_file *m, void *v)
 422 {
 423         int i;
 424         struct inode *inode = m->private;
 425         struct task_struct *task = get_proc_task(inode);
 426
 427         if (!task)
 428                 return -ESRCH;
 429         seq_puts(m, "Latency Top version : v0.1\n");
 430         for (i = 0; i < 32; i++) {
 431                 struct latency_record *lr = &task->latency_record[i];
 432                 if (lr->backtrace[0]) {
 433                         int q;
 434                         seq_printf(m, "%i %li %li",
 435                                    lr->count, lr->time, lr->max);
 436                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 437                                 unsigned long bt = lr->backtrace[q];
 438                                 if (!bt)
 439                                         break;
 440                                 if (bt == ULONG_MAX)
 441                                         break;
 442                                 seq_printf(m, " %ps", (void *)bt);
 443                         }
 444                         seq_putc(m, '\n');
 445                 }
 446
 447         }
 448         put_task_struct(task);
 449         return 0;
 450 }
 451
 452 static int lstats_open(struct inode *inode, struct file *file)
 453 {
 454         return single_open(file, lstats_show_proc, inode);
 455 }
 456
 457 static ssize_t lstats_write(struct file *file, const char __user *buf,
 458                             size_t count, loff_t *offs)
 459 {
 460         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 461
 462         if (!task)
 463                 return -ESRCH;
 464         clear_all_latency_tracing(task);
 465         put_task_struct(task);
 466
 467         return count;
 468 }
 469
 470 static const struct file_operations proc_lstats_operations = {
 471         .open           = lstats_open,
 472         .read           = seq_read,
 473         .write          = lstats_write,
 474         .llseek         = seq_lseek,
 475         .release        = single_release,
 476 };
 477
 478 #endif
 479
 480 static int proc_oom_score(struct task_struct *task, char *buffer)
 481 {
 482         unsigned long points = 0;
 483
 484         read_lock(&tasklist_lock);
 485         if (pid_alive(task))
 486                 points = oom_badness(task, NULL, NULL,
 487                                         totalram_pages + total_swap_pages);
 488         read_unlock(&tasklist_lock);
 489         return sprintf(buffer, "%lu\n", points);
 490 }
 491
 492 struct limit_names {
 493         char *name;
 494         char *unit;
 495 };
 496
 497 static const struct limit_names lnames[RLIM_NLIMITS] = {
 498         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 499         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 500         [RLIMIT_DATA] = {"Max data size", "bytes"},
 501         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 502         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 503         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 504         [RLIMIT_NPROC] = {"Max processes", "processes"},
 505         [RLIMIT_NOFILE] = {"Max open files", "files"},
 506         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 507         [RLIMIT_AS] = {"Max address space", "bytes"},
 508         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 509         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 510         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 511         [RLIMIT_NICE] = {"Max nice priority", NULL},
 512         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 513         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 514 };
 515
 516 /* Display limits for a process */
 517 static int proc_pid_limits(struct task_struct *task, char *buffer)
 518 {
 519         unsigned int i;
 520         int count = 0;
 521         unsigned long flags;
 522         char *bufptr = buffer;
 523
 524         struct rlimit rlim[RLIM_NLIMITS];
 525
 526         if (!lock_task_sighand(task, &flags))
 527                 return 0;
 528         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 529         unlock_task_sighand(task, &flags);
 530
 531         /*
 532          * print the file header
 533          */
 534         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 535                         "Limit", "Soft Limit", "Hard Limit", "Units");
 536
 537         for (i = 0; i < RLIM_NLIMITS; i++) {
 538                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 539                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 540                                          lnames[i].name, "unlimited");
 541                 else
 542                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 543                                          lnames[i].name, rlim[i].rlim_cur);
 544
 545                 if (rlim[i].rlim_max == RLIM_INFINITY)
 546                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 547                 else
 548                         count += sprintf(&bufptr[count], "%-20lu ",
 549                                          rlim[i].rlim_max);
 550
 551                 if (lnames[i].unit)
 552                         count += sprintf(&bufptr[count], "%-10s\n",
 553                                          lnames[i].unit);
 554                 else
 555                         count += sprintf(&bufptr[count], "\n");
 556         }
 557
 558         return count;
 559 }
 560
 561 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 562 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 563 {
 564         long nr;
 565         unsigned long args[6], sp, pc;
 566         int res = lock_trace(task);
 567         if (res)
 568                 return res;
 569
 570         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 571                 res = sprintf(buffer, "running\n");
 572         else if (nr < 0)
 573                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 574         else
 575                 res = sprintf(buffer,
 576                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 577                        nr,
 578                        args[0], args[1], args[2], args[3], args[4], args[5],
 579                        sp, pc);
 580         unlock_trace(task);
 581         return res;
 582 }
 583 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 584
 585 /************************************************************************/
 586 /*                       Here the fs part begins                        */
 587 /************************************************************************/
 588
 589 /* permission checks */
 590 static int proc_fd_access_allowed(struct inode *inode)
 591 {
 592         struct task_struct *task;
 593         int allowed = 0;
 594         /* Allow access to a task's file descriptors if it is us or we
 595          * may use ptrace attach to the process and find out that
 596          * information.
 597          */
 598         task = get_proc_task(inode);
 599         if (task) {
 600                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 601                 put_task_struct(task);
 602         }
 603         return allowed;
 604 }
 605
 606 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 607 {
 608         int error;
 609         struct inode *inode = dentry->d_inode;
 610
 611         if (attr->ia_valid & ATTR_MODE)
 612                 return -EPERM;
 613
 614         error = inode_change_ok(inode, attr);
 615         if (error)
 616                 return error;
 617
 618         if ((attr->ia_valid & ATTR_SIZE) &&
 619             attr->ia_size != i_size_read(inode)) {
 620                 error = vmtruncate(inode, attr->ia_size);
 621                 if (error)
 622                         return error;
 623         }
 624
 625         setattr_copy(inode, attr);
 626         mark_inode_dirty(inode);
 627         return 0;
 628 }
 629
 630 static const struct inode_operations proc_def_inode_operations = {
 631         .setattr        = proc_setattr,
 632 };
 633
 634 static int mounts_open_common(struct inode *inode, struct file *file,
 635                               const struct seq_operations *op)
 636 {
 637         struct task_struct *task = get_proc_task(inode);
 638         struct nsproxy *nsp;
 639         struct mnt_namespace *ns = NULL;
 640         struct path root;
 641         struct proc_mounts *p;
 642         int ret = -EINVAL;
 643
 644         if (task) {
 645                 rcu_read_lock();
 646                 nsp = task_nsproxy(task);
 647                 if (nsp) {
 648                         ns = nsp->mnt_ns;
 649                         if (ns)
 650                                 get_mnt_ns(ns);
 651                 }
 652                 rcu_read_unlock();
 653                 if (ns && get_task_root(task, &root) == 0)
 654                         ret = 0;
 655                 put_task_struct(task);
 656         }
 657
 658         if (!ns)
 659                 goto err;
 660         if (ret)
 661                 goto err_put_ns;
 662
 663         ret = -ENOMEM;
 664         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 665         if (!p)
 666                 goto err_put_path;
 667
 668         file->private_data = &p->m;
 669         ret = seq_open(file, op);
 670         if (ret)
 671                 goto err_free;
 672
 673         p->m.private = p;
 674         p->ns = ns;
 675         p->root = root;
 676         p->m.poll_event = ns->event;
 677
 678         return 0;
 679
 680  err_free:
 681         kfree(p);
 682  err_put_path:
 683         path_put(&root);
 684  err_put_ns:
 685         put_mnt_ns(ns);
 686  err:
 687         return ret;
 688 }
 689
 690 static int mounts_release(struct inode *inode, struct file *file)
 691 {
 692         struct proc_mounts *p = file->private_data;
 693         path_put(&p->root);
 694         put_mnt_ns(p->ns);
 695         return seq_release(inode, file);
 696 }
 697
 698 static unsigned mounts_poll(struct file *file, poll_table *wait)
 699 {
 700         struct proc_mounts *p = file->private_data;
 701         unsigned res = POLLIN | POLLRDNORM;
 702
 703         poll_wait(file, &p->ns->poll, wait);
 704         if (mnt_had_events(p))
 705                 res |= POLLERR | POLLPRI;
 706
 707         return res;
 708 }
 709
 710 static int mounts_open(struct inode *inode, struct file *file)
 711 {
 712         return mounts_open_common(inode, file, &mounts_op);
 713 }
 714
 715 static const struct file_operations proc_mounts_operations = {
 716         .open           = mounts_open,
 717         .read           = seq_read,
 718         .llseek         = seq_lseek,
 719         .release        = mounts_release,
 720         .poll           = mounts_poll,
 721 };
 722
 723 static int mountinfo_open(struct inode *inode, struct file *file)
 724 {
 725         return mounts_open_common(inode, file, &mountinfo_op);
 726 }
 727
 728 static const struct file_operations proc_mountinfo_operations = {
 729         .open           = mountinfo_open,
 730         .read           = seq_read,
 731         .llseek         = seq_lseek,
 732         .release        = mounts_release,
 733         .poll           = mounts_poll,
 734 };
 735
 736 static int mountstats_open(struct inode *inode, struct file *file)
 737 {
 738         return mounts_open_common(inode, file, &mountstats_op);
 739 }
 740
 741 static const struct file_operations proc_mountstats_operations = {
 742         .open           = mountstats_open,
 743         .read           = seq_read,
 744         .llseek         = seq_lseek,
 745         .release        = mounts_release,
 746 };
 747
 748 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 749
 750 static ssize_t proc_info_read(struct file * file, char __user * buf,
 751                           size_t count, loff_t *ppos)
 752 {
 753         struct inode * inode = file->f_path.dentry->d_inode;
 754         unsigned long page;
 755         ssize_t length;
 756         struct task_struct *task = get_proc_task(inode);
 757
 758         length = -ESRCH;
 759         if (!task)
 760                 goto out_no_task;
 761
 762         if (count > PROC_BLOCK_SIZE)
 763                 count = PROC_BLOCK_SIZE;
 764
 765         length = -ENOMEM;
 766         if (!(page = __get_free_page(GFP_TEMPORARY)))
 767                 goto out;
 768
 769         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 770
 771         if (length >= 0)
 772                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 773         free_page(page);
 774 out:
 775         put_task_struct(task);
 776 out_no_task:
 777         return length;
 778 }
 779
 780 static const struct file_operations proc_info_file_operations = {
 781         .read           = proc_info_read,
 782         .llseek         = generic_file_llseek,
 783 };
 784
 785 static int proc_single_show(struct seq_file *m, void *v)
 786 {
 787         struct inode *inode = m->private;
 788         struct pid_namespace *ns;
 789         struct pid *pid;
 790         struct task_struct *task;
 791         int ret;
 792
 793         ns = inode->i_sb->s_fs_info;
 794         pid = proc_pid(inode);
 795         task = get_pid_task(pid, PIDTYPE_PID);
 796         if (!task)
 797                 return -ESRCH;
 798
 799         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 800
 801         put_task_struct(task);
 802         return ret;
 803 }
 804
 805 static int proc_single_open(struct inode *inode, struct file *filp)
 806 {
 807         return single_open(filp, proc_single_show, inode);
 808 }
 809
 810 static const struct file_operations proc_single_file_operations = {
 811         .open           = proc_single_open,
 812         .read           = seq_read,
 813         .llseek         = seq_lseek,
 814         .release        = single_release,
 815 };
 816
 817 static int mem_open(struct inode* inode, struct file* file)
 818 {
 819         file->private_data = (void*)((long)current->self_exec_id);
 820         /* OK to pass negative loff_t, we can catch out-of-range */
 821         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 822         return 0;
 823 }
 824
 825 static ssize_t mem_read(struct file * file, char __user * buf,
 826                         size_t count, loff_t *ppos)
 827 {
 828         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 829         char *page;
 830         unsigned long src = *ppos;
 831         int ret = -ESRCH;
 832         struct mm_struct *mm;
 833
 834         if (!task)
 835                 goto out_no_task;
 836
 837         ret = -ENOMEM;
 838         page = (char *)__get_free_page(GFP_TEMPORARY);
 839         if (!page)
 840                 goto out;
 841
 842         mm = check_mem_permission(task);
 843         ret = PTR_ERR(mm);
 844         if (IS_ERR(mm))
 845                 goto out_free;
 846
 847         ret = -EIO;
 848
 849         if (file->private_data != (void*)((long)current->self_exec_id))
 850                 goto out_put;
 851
 852         ret = 0;
 853
 854         while (count > 0) {
 855                 int this_len, retval;
 856
 857                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 858                 retval = access_remote_vm(mm, src, page, this_len, 0);
 859                 if (!retval) {
 860                         if (!ret)
 861                                 ret = -EIO;
 862                         break;
 863                 }
 864
 865                 if (copy_to_user(buf, page, retval)) {
 866                         ret = -EFAULT;
 867                         break;
 868                 }
 869
 870                 ret += retval;
 871                 src += retval;
 872                 buf += retval;
 873                 count -= retval;
 874         }
 875         *ppos = src;
 876
 877 out_put:
 878         mmput(mm);
 879 out_free:
 880         free_page((unsigned long) page);
 881 out:
 882         put_task_struct(task);
 883 out_no_task:
 884         return ret;
 885 }
 886
 887 static ssize_t mem_write(struct file * file, const char __user *buf,
 888                          size_t count, loff_t *ppos)
 889 {
 890         int copied;
 891         char *page;
 892         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 893         unsigned long dst = *ppos;
 894         struct mm_struct *mm;
 895
 896         copied = -ESRCH;
 897         if (!task)
 898                 goto out_no_task;
 899
 900         copied = -ENOMEM;
 901         page = (char *)__get_free_page(GFP_TEMPORARY);
 902         if (!page)
 903                 goto out_task;
 904
 905         mm = check_mem_permission(task);
 906         copied = PTR_ERR(mm);
 907         if (IS_ERR(mm))
 908                 goto out_free;
 909
 910         copied = -EIO;
 911         if (file->private_data != (void *)((long)current->self_exec_id))
 912                 goto out_mm;
 913
 914         copied = 0;
 915         while (count > 0) {
 916                 int this_len, retval;
 917
 918                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 919                 if (copy_from_user(page, buf, this_len)) {
 920                         copied = -EFAULT;
 921                         break;
 922                 }
 923                 retval = access_remote_vm(mm, dst, page, this_len, 1);
 924                 if (!retval) {
 925                         if (!copied)
 926                                 copied = -EIO;
 927                         break;
 928                 }
 929                 copied += retval;
 930                 buf += retval;
 931                 dst += retval;
 932                 count -= retval;
 933         }
 934         *ppos = dst;
 935
 936 out_mm:
 937         mmput(mm);
 938 out_free:
 939         free_page((unsigned long) page);
 940 out_task:
 941         put_task_struct(task);
 942 out_no_task:
 943         return copied;
 944 }
 945
 946 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 947 {
 948         switch (orig) {
 949         case 0:
 950                 file->f_pos = offset;
 951                 break;
 952         case 1:
 953                 file->f_pos += offset;
 954                 break;
 955         default:
 956                 return -EINVAL;
 957         }
 958         force_successful_syscall_return();
 959         return file->f_pos;
 960 }
 961
 962 static const struct file_operations proc_mem_operations = {
 963         .llseek         = mem_lseek,
 964         .read           = mem_read,
 965         .write          = mem_write,
 966         .open           = mem_open,
 967 };
 968
 969 static ssize_t environ_read(struct file *file, char __user *buf,
 970                         size_t count, loff_t *ppos)
 971 {
 972         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 973         char *page;
 974         unsigned long src = *ppos;
 975         int ret = -ESRCH;
 976         struct mm_struct *mm;
 977
 978         if (!task)
 979                 goto out_no_task;
 980
 981         ret = -ENOMEM;
 982         page = (char *)__get_free_page(GFP_TEMPORARY);
 983         if (!page)
 984                 goto out;
 985
 986
 987         mm = mm_for_maps(task);
 988         ret = PTR_ERR(mm);
 989         if (!mm || IS_ERR(mm))
 990                 goto out_free;
 991
 992         ret = 0;
 993         while (count > 0) {
 994                 int this_len, retval, max_len;
 995
 996                 this_len = mm->env_end - (mm->env_start + src);
 997
 998                 if (this_len <= 0)
 999                         break;
1000
1001                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
1002                 this_len = (this_len > max_len) ? max_len : this_len;
1003
1004                 retval = access_process_vm(task, (mm->env_start + src),
1005                         page, this_len, 0);
1006
1007                 if (retval <= 0) {
1008                         ret = retval;
1009                         break;
1010                 }
1011
1012                 if (copy_to_user(buf, page, retval)) {
1013                         ret = -EFAULT;
1014                         break;
1015                 }
1016
1017                 ret += retval;
1018                 src += retval;
1019                 buf += retval;
1020                 count -= retval;
1021         }
1022         *ppos = src;
1023
1024         mmput(mm);
1025 out_free:
1026         free_page((unsigned long) page);
1027 out:
1028         put_task_struct(task);
1029 out_no_task:
1030         return ret;
1031 }
1032
1033 static const struct file_operations proc_environ_operations = {
1034         .read           = environ_read,
1035         .llseek         = generic_file_llseek,
1036 };
1037
1038 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1039                                 size_t count, loff_t *ppos)
1040 {
1041         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1042         char buffer[PROC_NUMBUF];
1043         size_t len;
1044         int oom_adjust = OOM_DISABLE;
1045         unsigned long flags;
1046
1047         if (!task)
1048                 return -ESRCH;
1049
1050         if (lock_task_sighand(task, &flags)) {
1051                 oom_adjust = task->signal->oom_adj;
1052                 unlock_task_sighand(task, &flags);
1053         }
1054
1055         put_task_struct(task);
1056
1057         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1058
1059         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1060 }
1061
1062 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1063                                 size_t count, loff_t *ppos)
1064 {
1065         struct task_struct *task;
1066         char buffer[PROC_NUMBUF];
1067         int oom_adjust;
1068         unsigned long flags;
1069         int err;
1070
1071         memset(buffer, 0, sizeof(buffer));
1072         if (count > sizeof(buffer) - 1)
1073                 count = sizeof(buffer) - 1;
1074         if (copy_from_user(buffer, buf, count)) {
1075                 err = -EFAULT;
1076                 goto out;
1077         }
1078
1079         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
1080         if (err)
1081                 goto out;
1082         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1083              oom_adjust != OOM_DISABLE) {
1084                 err = -EINVAL;
1085                 goto out;
1086         }
1087
1088         task = get_proc_task(file->f_path.dentry->d_inode);
1089         if (!task) {
1090                 err = -ESRCH;
1091                 goto out;
1092         }
1093
1094         task_lock(task);
1095         if (!task->mm) {
1096                 err = -EINVAL;
1097                 goto err_task_lock;
1098         }
1099
1100         if (!lock_task_sighand(task, &flags)) {
1101                 err = -ESRCH;
1102                 goto err_task_lock;
1103         }
1104
1105         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1106                 err = -EACCES;
1107                 goto err_sighand;
1108         }
1109
1110         /*
1111          * Warn that /proc/pid/oom_adj is deprecated, see
1112          * Documentation/feature-removal-schedule.txt.
1113          */
1114         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1115                   current->comm, task_pid_nr(current), task_pid_nr(task),
1116                   task_pid_nr(task));
1117         task->signal->oom_adj = oom_adjust;
1118         /*
1119          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1120          * value is always attainable.
1121          */
1122         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1123                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1124         else
1125                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1126                                                                 -OOM_DISABLE;
1127 err_sighand:
1128         unlock_task_sighand(task, &flags);
1129 err_task_lock:
1130         task_unlock(task);
1131         put_task_struct(task);
1132 out:
1133         return err < 0 ? err : count;
1134 }
1135
1136 static const struct file_operations proc_oom_adjust_operations = {
1137         .read           = oom_adjust_read,
1138         .write          = oom_adjust_write,
1139         .llseek         = generic_file_llseek,
1140 };
1141
1142 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1143                                         size_t count, loff_t *ppos)
1144 {
1145         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1146         char buffer[PROC_NUMBUF];
1147         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1148         unsigned long flags;
1149         size_t len;
1150
1151         if (!task)
1152                 return -ESRCH;
1153         if (lock_task_sighand(task, &flags)) {
1154                 oom_score_adj = task->signal->oom_score_adj;
1155                 unlock_task_sighand(task, &flags);
1156         }
1157         put_task_struct(task);
1158         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1159         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1160 }
1161
1162 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1163                                         size_t count, loff_t *ppos)
1164 {
1165         struct task_struct *task;
1166         char buffer[PROC_NUMBUF];
1167         unsigned long flags;
1168         int oom_score_adj;
1169         int err;
1170
1171         memset(buffer, 0, sizeof(buffer));
1172         if (count > sizeof(buffer) - 1)
1173                 count = sizeof(buffer) - 1;
1174         if (copy_from_user(buffer, buf, count)) {
1175                 err = -EFAULT;
1176                 goto out;
1177         }
1178
1179         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1180         if (err)
1181                 goto out;
1182         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1183                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1184                 err = -EINVAL;
1185                 goto out;
1186         }
1187
1188         task = get_proc_task(file->f_path.dentry->d_inode);
1189         if (!task) {
1190                 err = -ESRCH;
1191                 goto out;
1192         }
1193
1194         task_lock(task);
1195         if (!task->mm) {
1196                 err = -EINVAL;
1197                 goto err_task_lock;
1198         }
1199
1200         if (!lock_task_sighand(task, &flags)) {
1201                 err = -ESRCH;
1202                 goto err_task_lock;
1203         }
1204
1205         if (oom_score_adj < task->signal->oom_score_adj_min &&
1206                         !capable(CAP_SYS_RESOURCE)) {
1207                 err = -EACCES;
1208                 goto err_sighand;
1209         }
1210
1211         task->signal->oom_score_adj = oom_score_adj;
1212         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1213                 task->signal->oom_score_adj_min = oom_score_adj;
1214         /*
1215          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1216          * always attainable.
1217          */
1218         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1219                 task->signal->oom_adj = OOM_DISABLE;
1220         else
1221                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1222                                                         OOM_SCORE_ADJ_MAX;
1223 err_sighand:
1224         unlock_task_sighand(task, &flags);
1225 err_task_lock:
1226         task_unlock(task);
1227         put_task_struct(task);
1228 out:
1229         return err < 0 ? err : count;
1230 }
1231
1232 static const struct file_operations proc_oom_score_adj_operations = {
1233         .read           = oom_score_adj_read,
1234         .write          = oom_score_adj_write,
1235         .llseek         = default_llseek,
1236 };
1237
1238 #ifdef CONFIG_AUDITSYSCALL
1239 #define TMPBUFLEN 21
1240 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1241                                   size_t count, loff_t *ppos)
1242 {
1243         struct inode * inode = file->f_path.dentry->d_inode;
1244         struct task_struct *task = get_proc_task(inode);
1245         ssize_t length;
1246         char tmpbuf[TMPBUFLEN];
1247
1248         if (!task)
1249                 return -ESRCH;
1250         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1251                                 audit_get_loginuid(task));
1252         put_task_struct(task);
1253         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1254 }
1255
1256 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1257                                    size_t count, loff_t *ppos)
1258 {
1259         struct inode * inode = file->f_path.dentry->d_inode;
1260         char *page, *tmp;
1261         ssize_t length;
1262         uid_t loginuid;
1263
1264         if (!capable(CAP_AUDIT_CONTROL))
1265                 return -EPERM;
1266
1267         rcu_read_lock();
1268         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1269                 rcu_read_unlock();
1270                 return -EPERM;
1271         }
1272         rcu_read_unlock();
1273
1274         if (count >= PAGE_SIZE)
1275                 count = PAGE_SIZE - 1;
1276
1277         if (*ppos != 0) {
1278                 /* No partial writes. */
1279                 return -EINVAL;
1280         }
1281         page = (char*)__get_free_page(GFP_TEMPORARY);
1282         if (!page)
1283                 return -ENOMEM;
1284         length = -EFAULT;
1285         if (copy_from_user(page, buf, count))
1286                 goto out_free_page;
1287
1288         page[count] = '\0';
1289         loginuid = simple_strtoul(page, &tmp, 10);
1290         if (tmp == page) {
1291                 length = -EINVAL;
1292                 goto out_free_page;
1293
1294         }
1295         length = audit_set_loginuid(current, loginuid);
1296         if (likely(length == 0))
1297                 length = count;
1298
1299 out_free_page:
1300         free_page((unsigned long) page);
1301         return length;
1302 }
1303
1304 static const struct file_operations proc_loginuid_operations = {
1305         .read           = proc_loginuid_read,
1306         .write          = proc_loginuid_write,
1307         .llseek         = generic_file_llseek,
1308 };
1309
1310 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1311                                   size_t count, loff_t *ppos)
1312 {
1313         struct inode * inode = file->f_path.dentry->d_inode;
1314         struct task_struct *task = get_proc_task(inode);
1315         ssize_t length;
1316         char tmpbuf[TMPBUFLEN];
1317
1318         if (!task)
1319                 return -ESRCH;
1320         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1321                                 audit_get_sessionid(task));
1322         put_task_struct(task);
1323         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1324 }
1325
1326 static const struct file_operations proc_sessionid_operations = {
1327         .read           = proc_sessionid_read,
1328         .llseek         = generic_file_llseek,
1329 };
1330 #endif
1331
1332 #ifdef CONFIG_FAULT_INJECTION
1333 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1334                                       size_t count, loff_t *ppos)
1335 {
1336         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1337         char buffer[PROC_NUMBUF];
1338         size_t len;
1339         int make_it_fail;
1340
1341         if (!task)
1342                 return -ESRCH;
1343         make_it_fail = task->make_it_fail;
1344         put_task_struct(task);
1345
1346         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1347
1348         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1349 }
1350
1351 static ssize_t proc_fault_inject_write(struct file * file,
1352                         const char __user * buf, size_t count, loff_t *ppos)
1353 {
1354         struct task_struct *task;
1355         char buffer[PROC_NUMBUF], *end;
1356         int make_it_fail;
1357
1358         if (!capable(CAP_SYS_RESOURCE))
1359                 return -EPERM;
1360         memset(buffer, 0, sizeof(buffer));
1361         if (count > sizeof(buffer) - 1)
1362                 count = sizeof(buffer) - 1;
1363         if (copy_from_user(buffer, buf, count))
1364                 return -EFAULT;
1365         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1366         if (*end)
1367                 return -EINVAL;
1368         task = get_proc_task(file->f_dentry->d_inode);
1369         if (!task)
1370                 return -ESRCH;
1371         task->make_it_fail = make_it_fail;
1372         put_task_struct(task);
1373
1374         return count;
1375 }
1376
1377 static const struct file_operations proc_fault_inject_operations = {
1378         .read           = proc_fault_inject_read,
1379         .write          = proc_fault_inject_write,
1380         .llseek         = generic_file_llseek,
1381 };
1382 #endif
1383
1384
1385 #ifdef CONFIG_SCHED_DEBUG
1386 /*
1387  * Print out various scheduling related per-task fields:
1388  */
1389 static int sched_show(struct seq_file *m, void *v)
1390 {
1391         struct inode *inode = m->private;
1392         struct task_struct *p;
1393
1394         p = get_proc_task(inode);
1395         if (!p)
1396                 return -ESRCH;
1397         proc_sched_show_task(p, m);
1398
1399         put_task_struct(p);
1400
1401         return 0;
1402 }
1403
1404 static ssize_t
1405 sched_write(struct file *file, const char __user *buf,
1406             size_t count, loff_t *offset)
1407 {
1408         struct inode *inode = file->f_path.dentry->d_inode;
1409         struct task_struct *p;
1410
1411         p = get_proc_task(inode);
1412         if (!p)
1413                 return -ESRCH;
1414         proc_sched_set_task(p);
1415
1416         put_task_struct(p);
1417
1418         return count;
1419 }
1420
1421 static int sched_open(struct inode *inode, struct file *filp)
1422 {
1423         return single_open(filp, sched_show, inode);
1424 }
1425
1426 static const struct file_operations proc_pid_sched_operations = {
1427         .open           = sched_open,
1428         .read           = seq_read,
1429         .write          = sched_write,
1430         .llseek         = seq_lseek,
1431         .release        = single_release,
1432 };
1433
1434 #endif
1435
1436 #ifdef CONFIG_SCHED_AUTOGROUP
1437 /*
1438  * Print out autogroup related information:
1439  */
1440 static int sched_autogroup_show(struct seq_file *m, void *v)
1441 {
1442         struct inode *inode = m->private;
1443         struct task_struct *p;
1444
1445         p = get_proc_task(inode);
1446         if (!p)
1447                 return -ESRCH;
1448         proc_sched_autogroup_show_task(p, m);
1449
1450         put_task_struct(p);
1451
1452         return 0;
1453 }
1454
1455 static ssize_t
1456 sched_autogroup_write(struct file *file, const char __user *buf,
1457             size_t count, loff_t *offset)
1458 {
1459         struct inode *inode = file->f_path.dentry->d_inode;
1460         struct task_struct *p;
1461         char buffer[PROC_NUMBUF];
1462         int nice;
1463         int err;
1464
1465         memset(buffer, 0, sizeof(buffer));
1466         if (count > sizeof(buffer) - 1)
1467                 count = sizeof(buffer) - 1;
1468         if (copy_from_user(buffer, buf, count))
1469                 return -EFAULT;
1470
1471         err = kstrtoint(strstrip(buffer), 0, &nice);
1472         if (err < 0)
1473                 return err;
1474
1475         p = get_proc_task(inode);
1476         if (!p)
1477                 return -ESRCH;
1478
1479         err = nice;
1480         err = proc_sched_autogroup_set_nice(p, &err);
1481         if (err)
1482                 count = err;
1483
1484         put_task_struct(p);
1485
1486         return count;
1487 }
1488
1489 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1490 {
1491         int ret;
1492
1493         ret = single_open(filp, sched_autogroup_show, NULL);
1494         if (!ret) {
1495                 struct seq_file *m = filp->private_data;
1496
1497                 m->private = inode;
1498         }
1499         return ret;
1500 }
1501
1502 static const struct file_operations proc_pid_sched_autogroup_operations = {
1503         .open           = sched_autogroup_open,
1504         .read           = seq_read,
1505         .write          = sched_autogroup_write,
1506         .llseek         = seq_lseek,
1507         .release        = single_release,
1508 };
1509
1510 #endif /* CONFIG_SCHED_AUTOGROUP */
1511
1512 static ssize_t comm_write(struct file *file, const char __user *buf,
1513                                 size_t count, loff_t *offset)
1514 {
1515         struct inode *inode = file->f_path.dentry->d_inode;
1516         struct task_struct *p;
1517         char buffer[TASK_COMM_LEN];
1518
1519         memset(buffer, 0, sizeof(buffer));
1520         if (count > sizeof(buffer) - 1)
1521                 count = sizeof(buffer) - 1;
1522         if (copy_from_user(buffer, buf, count))
1523                 return -EFAULT;
1524
1525         p = get_proc_task(inode);
1526         if (!p)
1527                 return -ESRCH;
1528
1529         if (same_thread_group(current, p))
1530                 set_task_comm(p, buffer);
1531         else
1532                 count = -EINVAL;
1533
1534         put_task_struct(p);
1535
1536         return count;
1537 }
1538
1539 static int comm_show(struct seq_file *m, void *v)
1540 {
1541         struct inode *inode = m->private;
1542         struct task_struct *p;
1543
1544         p = get_proc_task(inode);
1545         if (!p)
1546                 return -ESRCH;
1547
1548         task_lock(p);
1549         seq_printf(m, "%s\n", p->comm);
1550         task_unlock(p);
1551
1552         put_task_struct(p);
1553
1554         return 0;
1555 }
1556
1557 static int comm_open(struct inode *inode, struct file *filp)
1558 {
1559         return single_open(filp, comm_show, inode);
1560 }
1561
1562 static const struct file_operations proc_pid_set_comm_operations = {
1563         .open           = comm_open,
1564         .read           = seq_read,
1565         .write          = comm_write,
1566         .llseek         = seq_lseek,
1567         .release        = single_release,
1568 };
1569
1570 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1571 {
1572         struct task_struct *task;
1573         struct mm_struct *mm;
1574         struct file *exe_file;
1575
1576         task = get_proc_task(inode);
1577         if (!task)
1578                 return -ENOENT;
1579         mm = get_task_mm(task);
1580         put_task_struct(task);
1581         if (!mm)
1582                 return -ENOENT;
1583         exe_file = get_mm_exe_file(mm);
1584         mmput(mm);
1585         if (exe_file) {
1586                 *exe_path = exe_file->f_path;
1587                 path_get(&exe_file->f_path);
1588                 fput(exe_file);
1589                 return 0;
1590         } else
1591                 return -ENOENT;
1592 }
1593
1594 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1595 {
1596         struct inode *inode = dentry->d_inode;
1597         int error = -EACCES;
1598
1599         /* We don't need a base pointer in the /proc filesystem */
1600         path_put(&nd->path);
1601
1602         /* Are we allowed to snoop on the tasks file descriptors? */
1603         if (!proc_fd_access_allowed(inode))
1604                 goto out;
1605
1606         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1607 out:
1608         return ERR_PTR(error);
1609 }
1610
1611 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1612 {
1613         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1614         char *pathname;
1615         int len;
1616
1617         if (!tmp)
1618                 return -ENOMEM;
1619
1620         pathname = d_path(path, tmp, PAGE_SIZE);
1621         len = PTR_ERR(pathname);
1622         if (IS_ERR(pathname))
1623                 goto out;
1624         len = tmp + PAGE_SIZE - 1 - pathname;
1625
1626         if (len > buflen)
1627                 len = buflen;
1628         if (copy_to_user(buffer, pathname, len))
1629                 len = -EFAULT;
1630  out:
1631         free_page((unsigned long)tmp);
1632         return len;
1633 }
1634
1635 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1636 {
1637         int error = -EACCES;
1638         struct inode *inode = dentry->d_inode;
1639         struct path path;
1640
1641         /* Are we allowed to snoop on the tasks file descriptors? */
1642         if (!proc_fd_access_allowed(inode))
1643                 goto out;
1644
1645         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1646         if (error)
1647                 goto out;
1648
1649         error = do_proc_readlink(&path, buffer, buflen);
1650         path_put(&path);
1651 out:
1652         return error;
1653 }
1654
1655 static const struct inode_operations proc_pid_link_inode_operations = {
1656         .readlink       = proc_pid_readlink,
1657         .follow_link    = proc_pid_follow_link,
1658         .setattr        = proc_setattr,
1659 };
1660
1661
1662 /* building an inode */
1663
1664 static int task_dumpable(struct task_struct *task)
1665 {
1666         int dumpable = 0;
1667         struct mm_struct *mm;
1668
1669         task_lock(task);
1670         mm = task->mm;
1671         if (mm)
1672                 dumpable = get_dumpable(mm);
1673         task_unlock(task);
1674         if(dumpable == 1)
1675                 return 1;
1676         return 0;
1677 }
1678
1679 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1680 {
1681         struct inode * inode;
1682         struct proc_inode *ei;
1683         const struct cred *cred;
1684
1685         /* We need a new inode */
1686
1687         inode = new_inode(sb);
1688         if (!inode)
1689                 goto out;
1690
1691         /* Common stuff */
1692         ei = PROC_I(inode);
1693         inode->i_ino = get_next_ino();
1694         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1695         inode->i_op = &proc_def_inode_operations;
1696
1697         /*
1698          * grab the reference to task.
1699          */
1700         ei->pid = get_task_pid(task, PIDTYPE_PID);
1701         if (!ei->pid)
1702                 goto out_unlock;
1703
1704         if (task_dumpable(task)) {
1705                 rcu_read_lock();
1706                 cred = __task_cred(task);
1707                 inode->i_uid = cred->euid;
1708                 inode->i_gid = cred->egid;
1709                 rcu_read_unlock();
1710         }
1711         security_task_to_inode(task, inode);
1712
1713 out:
1714         return inode;
1715
1716 out_unlock:
1717         iput(inode);
1718         return NULL;
1719 }
1720
1721 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1722 {
1723         struct inode *inode = dentry->d_inode;
1724         struct task_struct *task;
1725         const struct cred *cred;
1726
1727         generic_fillattr(inode, stat);
1728
1729         rcu_read_lock();
1730         stat->uid = 0;
1731         stat->gid = 0;
1732         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1733         if (task) {
1734                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1735                     task_dumpable(task)) {
1736                         cred = __task_cred(task);
1737                         stat->uid = cred->euid;
1738                         stat->gid = cred->egid;
1739                 }
1740         }
1741         rcu_read_unlock();
1742         return 0;
1743 }
1744
1745 /* dentry stuff */
1746
1747 /*
1748  *      Exceptional case: normally we are not allowed to unhash a busy
1749  * directory. In this case, however, we can do it - no aliasing problems
1750  * due to the way we treat inodes.
1751  *
1752  * Rewrite the inode's ownerships here because the owning task may have
1753  * performed a setuid(), etc.
1754  *
1755  * Before the /proc/pid/status file was created the only way to read
1756  * the effective uid of a /process was to stat /proc/pid.  Reading
1757  * /proc/pid/status is slow enough that procps and other packages
1758  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1759  * made this apply to all per process world readable and executable
1760  * directories.
1761  */
1762 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1763 {
1764         struct inode *inode;
1765         struct task_struct *task;
1766         const struct cred *cred;
1767
1768         if (nd && nd->flags & LOOKUP_RCU)
1769                 return -ECHILD;
1770
1771         inode = dentry->d_inode;
1772         task = get_proc_task(inode);
1773
1774         if (task) {
1775                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1776                     task_dumpable(task)) {
1777                         rcu_read_lock();
1778                         cred = __task_cred(task);
1779                         inode->i_uid = cred->euid;
1780                         inode->i_gid = cred->egid;
1781                         rcu_read_unlock();
1782                 } else {
1783                         inode->i_uid = 0;
1784                         inode->i_gid = 0;
1785                 }
1786                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1787                 security_task_to_inode(task, inode);
1788                 put_task_struct(task);
1789                 return 1;
1790         }
1791         d_drop(dentry);
1792         return 0;
1793 }
1794
1795 static int pid_delete_dentry(const struct dentry * dentry)
1796 {
1797         /* Is the task we represent dead?
1798          * If so, then don't put the dentry on the lru list,
1799          * kill it immediately.
1800          */
1801         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1802 }
1803
1804 const struct dentry_operations pid_dentry_operations =
1805 {
1806         .d_revalidate   = pid_revalidate,
1807         .d_delete       = pid_delete_dentry,
1808 };
1809
1810 /* Lookups */
1811
1812 /*
1813  * Fill a directory entry.
1814  *
1815  * If possible create the dcache entry and derive our inode number and
1816  * file type from dcache entry.
1817  *
1818  * Since all of the proc inode numbers are dynamically generated, the inode
1819  * numbers do not exist until the inode is cache.  This means creating the
1820  * the dcache entry in readdir is necessary to keep the inode numbers
1821  * reported by readdir in sync with the inode numbers reported
1822  * by stat.
1823  */
1824 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1825         const char *name, int len,
1826         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1827 {
1828         struct dentry *child, *dir = filp->f_path.dentry;
1829         struct inode *inode;
1830         struct qstr qname;
1831         ino_t ino = 0;
1832         unsigned type = DT_UNKNOWN;
1833
1834         qname.name = name;
1835         qname.len  = len;
1836         qname.hash = full_name_hash(name, len);
1837
1838         child = d_lookup(dir, &qname);
1839         if (!child) {
1840                 struct dentry *new;
1841                 new = d_alloc(dir, &qname);
1842                 if (new) {
1843                         child = instantiate(dir->d_inode, new, task, ptr);
1844                         if (child)
1845                                 dput(new);
1846                         else
1847                                 child = new;
1848                 }
1849         }
1850         if (!child || IS_ERR(child) || !child->d_inode)
1851                 goto end_instantiate;
1852         inode = child->d_inode;
1853         if (inode) {
1854                 ino = inode->i_ino;
1855                 type = inode->i_mode >> 12;
1856         }
1857         dput(child);
1858 end_instantiate:
1859         if (!ino)
1860                 ino = find_inode_number(dir, &qname);
1861         if (!ino)
1862                 ino = 1;
1863         return filldir(dirent, name, len, filp->f_pos, ino, type);
1864 }
1865
1866 static unsigned name_to_int(struct dentry *dentry)
1867 {
1868         const char *name = dentry->d_name.name;
1869         int len = dentry->d_name.len;
1870         unsigned n = 0;
1871
1872         if (len > 1 && *name == '0')
1873                 goto out;
1874         while (len-- > 0) {
1875                 unsigned c = *name++ - '0';
1876                 if (c > 9)
1877                         goto out;
1878                 if (n >= (~0U-9)/10)
1879                         goto out;
1880                 n *= 10;
1881                 n += c;
1882         }
1883         return n;
1884 out:
1885         return ~0U;
1886 }
1887
1888 #define PROC_FDINFO_MAX 64
1889
1890 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1891 {
1892         struct task_struct *task = get_proc_task(inode);
1893         struct files_struct *files = NULL;
1894         struct file *file;
1895         int fd = proc_fd(inode);
1896
1897         if (task) {
1898                 files = get_files_struct(task);
1899                 put_task_struct(task);
1900         }
1901         if (files) {
1902                 /*
1903                  * We are not taking a ref to the file structure, so we must
1904                  * hold ->file_lock.
1905                  */
1906                 spin_lock(&files->file_lock);
1907                 file = fcheck_files(files, fd);
1908                 if (file) {
1909                         unsigned int f_flags;
1910                         struct fdtable *fdt;
1911
1912                         fdt = files_fdtable(files);
1913                         f_flags = file->f_flags & ~O_CLOEXEC;
1914                         if (FD_ISSET(fd, fdt->close_on_exec))
1915                                 f_flags |= O_CLOEXEC;
1916
1917                         if (path) {
1918                                 *path = file->f_path;
1919                                 path_get(&file->f_path);
1920                         }
1921                         if (info)
1922                                 snprintf(info, PROC_FDINFO_MAX,
1923                                          "pos:\t%lli\n"
1924                                          "flags:\t0%o\n",
1925                                          (long long) file->f_pos,
1926                                          f_flags);
1927                         spin_unlock(&files->file_lock);
1928                         put_files_struct(files);
1929                         return 0;
1930                 }
1931                 spin_unlock(&files->file_lock);
1932                 put_files_struct(files);
1933         }
1934         return -ENOENT;
1935 }
1936
1937 static int proc_fd_link(struct inode *inode, struct path *path)
1938 {
1939         return proc_fd_info(inode, path, NULL);
1940 }
1941
1942 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1943 {
1944         struct inode *inode;
1945         struct task_struct *task;
1946         int fd;
1947         struct files_struct *files;
1948         const struct cred *cred;
1949
1950         if (nd && nd->flags & LOOKUP_RCU)
1951                 return -ECHILD;
1952
1953         inode = dentry->d_inode;
1954         task = get_proc_task(inode);
1955         fd = proc_fd(inode);
1956
1957         if (task) {
1958                 files = get_files_struct(task);
1959                 if (files) {
1960                         rcu_read_lock();
1961                         if (fcheck_files(files, fd)) {
1962                                 rcu_read_unlock();
1963                                 put_files_struct(files);
1964                                 if (task_dumpable(task)) {
1965                                         rcu_read_lock();
1966                                         cred = __task_cred(task);
1967                                         inode->i_uid = cred->euid;
1968                                         inode->i_gid = cred->egid;
1969                                         rcu_read_unlock();
1970                                 } else {
1971                                         inode->i_uid = 0;
1972                                         inode->i_gid = 0;
1973                                 }
1974                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1975                                 security_task_to_inode(task, inode);
1976                                 put_task_struct(task);
1977                                 return 1;
1978                         }
1979                         rcu_read_unlock();
1980                         put_files_struct(files);
1981                 }
1982                 put_task_struct(task);
1983         }
1984         d_drop(dentry);
1985         return 0;
1986 }
1987
1988 static const struct dentry_operations tid_fd_dentry_operations =
1989 {
1990         .d_revalidate   = tid_fd_revalidate,
1991         .d_delete       = pid_delete_dentry,
1992 };
1993
1994 static struct dentry *proc_fd_instantiate(struct inode *dir,
1995         struct dentry *dentry, struct task_struct *task, const void *ptr)
1996 {
1997         unsigned fd = *(const unsigned *)ptr;
1998         struct file *file;
1999         struct files_struct *files;
2000         struct inode *inode;
2001         struct proc_inode *ei;
2002         struct dentry *error = ERR_PTR(-ENOENT);
2003
2004         inode = proc_pid_make_inode(dir->i_sb, task);
2005         if (!inode)
2006                 goto out;
2007         ei = PROC_I(inode);
2008         ei->fd = fd;
2009         files = get_files_struct(task);
2010         if (!files)
2011                 goto out_iput;
2012         inode->i_mode = S_IFLNK;
2013
2014         /*
2015          * We are not taking a ref to the file structure, so we must
2016          * hold ->file_lock.
2017          */
2018         spin_lock(&files->file_lock);
2019         file = fcheck_files(files, fd);
2020         if (!file)
2021                 goto out_unlock;
2022         if (file->f_mode & FMODE_READ)
2023                 inode->i_mode |= S_IRUSR | S_IXUSR;
2024         if (file->f_mode & FMODE_WRITE)
2025                 inode->i_mode |= S_IWUSR | S_IXUSR;
2026         spin_unlock(&files->file_lock);
2027         put_files_struct(files);
2028
2029         inode->i_op = &proc_pid_link_inode_operations;
2030         inode->i_size = 64;
2031         ei->op.proc_get_link = proc_fd_link;
2032         d_set_d_op(dentry, &tid_fd_dentry_operations);
2033         d_add(dentry, inode);
2034         /* Close the race of the process dying before we return the dentry */
2035         if (tid_fd_revalidate(dentry, NULL))
2036                 error = NULL;
2037
2038  out:
2039         return error;
2040 out_unlock:
2041         spin_unlock(&files->file_lock);
2042         put_files_struct(files);
2043 out_iput:
2044         iput(inode);
2045         goto out;
2046 }
2047
2048 static struct dentry *proc_lookupfd_common(struct inode *dir,
2049                                            struct dentry *dentry,
2050                                            instantiate_t instantiate)
2051 {
2052         struct task_struct *task = get_proc_task(dir);
2053         unsigned fd = name_to_int(dentry);
2054         struct dentry *result = ERR_PTR(-ENOENT);
2055
2056         if (!task)
2057                 goto out_no_task;
2058         if (fd == ~0U)
2059                 goto out;
2060
2061         result = instantiate(dir, dentry, task, &fd);
2062 out:
2063         put_task_struct(task);
2064 out_no_task:
2065         return result;
2066 }
2067
2068 static int proc_readfd_common(struct file * filp, void * dirent,
2069                               filldir_t filldir, instantiate_t instantiate)
2070 {
2071         struct dentry *dentry = filp->f_path.dentry;
2072         struct inode *inode = dentry->d_inode;
2073         struct task_struct *p = get_proc_task(inode);
2074         unsigned int fd, ino;
2075         int retval;
2076         struct files_struct * files;
2077
2078         retval = -ENOENT;
2079         if (!p)
2080                 goto out_no_task;
2081         retval = 0;
2082
2083         fd = filp->f_pos;
2084         switch (fd) {
2085                 case 0:
2086                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2087                                 goto out;
2088                         filp->f_pos++;
2089                 case 1:
2090                         ino = parent_ino(dentry);
2091                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2092                                 goto out;
2093                         filp->f_pos++;
2094                 default:
2095                         files = get_files_struct(p);
2096                         if (!files)
2097                                 goto out;
2098                         rcu_read_lock();
2099                         for (fd = filp->f_pos-2;
2100                              fd < files_fdtable(files)->max_fds;
2101                              fd++, filp->f_pos++) {
2102                                 char name[PROC_NUMBUF];
2103                                 int len;
2104
2105                                 if (!fcheck_files(files, fd))
2106                                         continue;
2107                                 rcu_read_unlock();
2108
2109                                 len = snprintf(name, sizeof(name), "%d", fd);
2110                                 if (proc_fill_cache(filp, dirent, filldir,
2111                                                     name, len, instantiate,
2112                                                     p, &fd) < 0) {
2113                                         rcu_read_lock();
2114                                         break;
2115                                 }
2116                                 rcu_read_lock();
2117                         }
2118                         rcu_read_unlock();
2119                         put_files_struct(files);
2120         }
2121 out:
2122         put_task_struct(p);
2123 out_no_task:
2124         return retval;
2125 }
2126
2127 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2128                                     struct nameidata *nd)
2129 {
2130         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2131 }
2132
2133 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2134 {
2135         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2136 }
2137
2138 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2139                                       size_t len, loff_t *ppos)
2140 {
2141         char tmp[PROC_FDINFO_MAX];
2142         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2143         if (!err)
2144                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2145         return err;
2146 }
2147
2148 static const struct file_operations proc_fdinfo_file_operations = {
2149         .open           = nonseekable_open,
2150         .read           = proc_fdinfo_read,
2151         .llseek         = no_llseek,
2152 };
2153
2154 static const struct file_operations proc_fd_operations = {
2155         .read           = generic_read_dir,
2156         .readdir        = proc_readfd,
2157         .llseek         = default_llseek,
2158 };
2159
2160 /*
2161  * /proc/pid/fd needs a special permission handler so that a process can still
2162  * access /proc/self/fd after it has executed a setuid().
2163  */
2164 static int proc_fd_permission(struct inode *inode, int mask)
2165 {
2166         int rv = generic_permission(inode, mask);
2167         if (rv == 0)
2168                 return 0;
2169         if (task_pid(current) == proc_pid(inode))
2170                 rv = 0;
2171         return rv;
2172 }
2173
2174 /*
2175  * proc directories can do almost nothing..
2176  */
2177 static const struct inode_operations proc_fd_inode_operations = {
2178         .lookup         = proc_lookupfd,
2179         .permission     = proc_fd_permission,
2180         .setattr        = proc_setattr,
2181 };
2182
2183 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2184         struct dentry *dentry, struct task_struct *task, const void *ptr)
2185 {
2186         unsigned fd = *(unsigned *)ptr;
2187         struct inode *inode;
2188         struct proc_inode *ei;
2189         struct dentry *error = ERR_PTR(-ENOENT);
2190
2191         inode = proc_pid_make_inode(dir->i_sb, task);
2192         if (!inode)
2193                 goto out;
2194         ei = PROC_I(inode);
2195         ei->fd = fd;
2196         inode->i_mode = S_IFREG | S_IRUSR;
2197         inode->i_fop = &proc_fdinfo_file_operations;
2198         d_set_d_op(dentry, &tid_fd_dentry_operations);
2199         d_add(dentry, inode);
2200         /* Close the race of the process dying before we return the dentry */
2201         if (tid_fd_revalidate(dentry, NULL))
2202                 error = NULL;
2203
2204  out:
2205         return error;
2206 }
2207
2208 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2209                                         struct dentry *dentry,
2210                                         struct nameidata *nd)
2211 {
2212         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2213 }
2214
2215 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2216 {
2217         return proc_readfd_common(filp, dirent, filldir,
2218                                   proc_fdinfo_instantiate);
2219 }
2220
2221 static const struct file_operations proc_fdinfo_operations = {
2222         .read           = generic_read_dir,
2223         .readdir        = proc_readfdinfo,
2224         .llseek         = default_llseek,
2225 };
2226
2227 /*
2228  * proc directories can do almost nothing..
2229  */
2230 static const struct inode_operations proc_fdinfo_inode_operations = {
2231         .lookup         = proc_lookupfdinfo,
2232         .setattr        = proc_setattr,
2233 };
2234
2235
2236 static struct dentry *proc_pident_instantiate(struct inode *dir,
2237         struct dentry *dentry, struct task_struct *task, const void *ptr)
2238 {
2239         const struct pid_entry *p = ptr;
2240         struct inode *inode;
2241         struct proc_inode *ei;
2242         struct dentry *error = ERR_PTR(-ENOENT);
2243
2244         inode = proc_pid_make_inode(dir->i_sb, task);
2245         if (!inode)
2246                 goto out;
2247
2248         ei = PROC_I(inode);
2249         inode->i_mode = p->mode;
2250         if (S_ISDIR(inode->i_mode))
2251                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2252         if (p->iop)
2253                 inode->i_op = p->iop;
2254         if (p->fop)
2255                 inode->i_fop = p->fop;
2256         ei->op = p->op;
2257         d_set_d_op(dentry, &pid_dentry_operations);
2258         d_add(dentry, inode);
2259         /* Close the race of the process dying before we return the dentry */
2260         if (pid_revalidate(dentry, NULL))
2261                 error = NULL;
2262 out:
2263         return error;
2264 }
2265
2266 static struct dentry *proc_pident_lookup(struct inode *dir,
2267                                          struct dentry *dentry,
2268                                          const struct pid_entry *ents,
2269                                          unsigned int nents)
2270 {
2271         struct dentry *error;
2272         struct task_struct *task = get_proc_task(dir);
2273         const struct pid_entry *p, *last;
2274
2275         error = ERR_PTR(-ENOENT);
2276
2277         if (!task)
2278                 goto out_no_task;
2279
2280         /*
2281          * Yes, it does not scale. And it should not. Don't add
2282          * new entries into /proc/<tgid>/ without very good reasons.
2283          */
2284         last = &ents[nents - 1];
2285         for (p = ents; p <= last; p++) {
2286                 if (p->len != dentry->d_name.len)
2287                         continue;
2288                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2289                         break;
2290         }
2291         if (p > last)
2292                 goto out;
2293
2294         error = proc_pident_instantiate(dir, dentry, task, p);
2295 out:
2296         put_task_struct(task);
2297 out_no_task:
2298         return error;
2299 }
2300
2301 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2302         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2303 {
2304         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2305                                 proc_pident_instantiate, task, p);
2306 }
2307
2308 static int proc_pident_readdir(struct file *filp,
2309                 void *dirent, filldir_t filldir,
2310                 const struct pid_entry *ents, unsigned int nents)
2311 {
2312         int i;
2313         struct dentry *dentry = filp->f_path.dentry;
2314         struct inode *inode = dentry->d_inode;
2315         struct task_struct *task = get_proc_task(inode);
2316         const struct pid_entry *p, *last;
2317         ino_t ino;
2318         int ret;
2319
2320         ret = -ENOENT;
2321         if (!task)
2322                 goto out_no_task;
2323
2324         ret = 0;
2325         i = filp->f_pos;
2326         switch (i) {
2327         case 0:
2328                 ino = inode->i_ino;
2329                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2330                         goto out;
2331                 i++;
2332                 filp->f_pos++;
2333                 /* fall through */
2334         case 1:
2335                 ino = parent_ino(dentry);
2336                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2337                         goto out;
2338                 i++;
2339                 filp->f_pos++;
2340                 /* fall through */
2341         default:
2342                 i -= 2;
2343                 if (i >= nents) {
2344                         ret = 1;
2345                         goto out;
2346                 }
2347                 p = ents + i;
2348                 last = &ents[nents - 1];
2349                 while (p <= last) {
2350                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2351                                 goto out;
2352                         filp->f_pos++;
2353                         p++;
2354                 }
2355         }
2356
2357         ret = 1;
2358 out:
2359         put_task_struct(task);
2360 out_no_task:
2361         return ret;
2362 }
2363
2364 #ifdef CONFIG_SECURITY
2365 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2366                                   size_t count, loff_t *ppos)
2367 {
2368         struct inode * inode = file->f_path.dentry->d_inode;
2369         char *p = NULL;
2370         ssize_t length;
2371         struct task_struct *task = get_proc_task(inode);
2372
2373         if (!task)
2374                 return -ESRCH;
2375
2376         length = security_getprocattr(task,
2377                                       (char*)file->f_path.dentry->d_name.name,
2378                                       &p);
2379         put_task_struct(task);
2380         if (length > 0)
2381                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2382         kfree(p);
2383         return length;
2384 }
2385
2386 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2387                                    size_t count, loff_t *ppos)
2388 {
2389         struct inode * inode = file->f_path.dentry->d_inode;
2390         char *page;
2391         ssize_t length;
2392         struct task_struct *task = get_proc_task(inode);
2393
2394         length = -ESRCH;
2395         if (!task)
2396                 goto out_no_task;
2397         if (count > PAGE_SIZE)
2398                 count = PAGE_SIZE;
2399
2400         /* No partial writes. */
2401         length = -EINVAL;
2402         if (*ppos != 0)
2403                 goto out;
2404
2405         length = -ENOMEM;
2406         page = (char*)__get_free_page(GFP_TEMPORARY);
2407         if (!page)
2408                 goto out;
2409
2410         length = -EFAULT;
2411         if (copy_from_user(page, buf, count))
2412                 goto out_free;
2413
2414         /* Guard against adverse ptrace interaction */
2415         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2416         if (length < 0)
2417                 goto out_free;
2418
2419         length = security_setprocattr(task,
2420                                       (char*)file->f_path.dentry->d_name.name,
2421                                       (void*)page, count);
2422         mutex_unlock(&task->signal->cred_guard_mutex);
2423 out_free:
2424         free_page((unsigned long) page);
2425 out:
2426         put_task_struct(task);
2427 out_no_task:
2428         return length;
2429 }
2430
2431 static const struct file_operations proc_pid_attr_operations = {
2432         .read           = proc_pid_attr_read,
2433         .write          = proc_pid_attr_write,
2434         .llseek         = generic_file_llseek,
2435 };
2436
2437 static const struct pid_entry attr_dir_stuff[] = {
2438         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2439         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2440         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2441         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2442         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2443         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2444 };
2445
2446 static int proc_attr_dir_readdir(struct file * filp,
2447                              void * dirent, filldir_t filldir)
2448 {
2449         return proc_pident_readdir(filp,dirent,filldir,
2450                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2451 }
2452
2453 static const struct file_operations proc_attr_dir_operations = {
2454         .read           = generic_read_dir,
2455         .readdir        = proc_attr_dir_readdir,
2456         .llseek         = default_llseek,
2457 };
2458
2459 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2460                                 struct dentry *dentry, struct nameidata *nd)
2461 {
2462         return proc_pident_lookup(dir, dentry,
2463                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2464 }
2465
2466 static const struct inode_operations proc_attr_dir_inode_operations = {
2467         .lookup         = proc_attr_dir_lookup,
2468         .getattr        = pid_getattr,
2469         .setattr        = proc_setattr,
2470 };
2471
2472 #endif
2473
2474 #ifdef CONFIG_ELF_CORE
2475 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2476                                          size_t count, loff_t *ppos)
2477 {
2478         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2479         struct mm_struct *mm;
2480         char buffer[PROC_NUMBUF];
2481         size_t len;
2482         int ret;
2483
2484         if (!task)
2485                 return -ESRCH;
2486
2487         ret = 0;
2488         mm = get_task_mm(task);
2489         if (mm) {
2490                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2491                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2492                                 MMF_DUMP_FILTER_SHIFT));
2493                 mmput(mm);
2494                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2495         }
2496
2497         put_task_struct(task);
2498
2499         return ret;
2500 }
2501
2502 static ssize_t proc_coredump_filter_write(struct file *file,
2503                                           const char __user *buf,
2504                                           size_t count,
2505                                           loff_t *ppos)
2506 {
2507         struct task_struct *task;
2508         struct mm_struct *mm;
2509         char buffer[PROC_NUMBUF], *end;
2510         unsigned int val;
2511         int ret;
2512         int i;
2513         unsigned long mask;
2514
2515         ret = -EFAULT;
2516         memset(buffer, 0, sizeof(buffer));
2517         if (count > sizeof(buffer) - 1)
2518                 count = sizeof(buffer) - 1;
2519         if (copy_from_user(buffer, buf, count))
2520                 goto out_no_task;
2521
2522         ret = -EINVAL;
2523         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2524         if (*end == '\n')
2525                 end++;
2526         if (end - buffer == 0)
2527                 goto out_no_task;
2528
2529         ret = -ESRCH;
2530         task = get_proc_task(file->f_dentry->d_inode);
2531         if (!task)
2532                 goto out_no_task;
2533
2534         ret = end - buffer;
2535         mm = get_task_mm(task);
2536         if (!mm)
2537                 goto out_no_mm;
2538
2539         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2540                 if (val & mask)
2541                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2542                 else
2543                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2544         }
2545
2546         mmput(mm);
2547  out_no_mm:
2548         put_task_struct(task);
2549  out_no_task:
2550         return ret;
2551 }
2552
2553 static const struct file_operations proc_coredump_filter_operations = {
2554         .read           = proc_coredump_filter_read,
2555         .write          = proc_coredump_filter_write,
2556         .llseek         = generic_file_llseek,
2557 };
2558 #endif
2559
2560 /*
2561  * /proc/self:
2562  */
2563 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2564                               int buflen)
2565 {
2566         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2567         pid_t tgid = task_tgid_nr_ns(current, ns);
2568         char tmp[PROC_NUMBUF];
2569         if (!tgid)
2570                 return -ENOENT;
2571         sprintf(tmp, "%d", tgid);
2572         return vfs_readlink(dentry,buffer,buflen,tmp);
2573 }
2574
2575 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2576 {
2577         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2578         pid_t tgid = task_tgid_nr_ns(current, ns);
2579         char *name = ERR_PTR(-ENOENT);
2580         if (tgid) {
2581                 name = __getname();
2582                 if (!name)
2583                         name = ERR_PTR(-ENOMEM);
2584                 else
2585                         sprintf(name, "%d", tgid);
2586         }
2587         nd_set_link(nd, name);
2588         return NULL;
2589 }
2590
2591 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2592                                 void *cookie)
2593 {
2594         char *s = nd_get_link(nd);
2595         if (!IS_ERR(s))
2596                 __putname(s);
2597 }
2598
2599 static const struct inode_operations proc_self_inode_operations = {
2600         .readlink       = proc_self_readlink,
2601         .follow_link    = proc_self_follow_link,
2602         .put_link       = proc_self_put_link,
2603 };
2604
2605 /*
2606  * proc base
2607  *
2608  * These are the directory entries in the root directory of /proc
2609  * that properly belong to the /proc filesystem, as they describe
2610  * describe something that is process related.
2611  */
2612 static const struct pid_entry proc_base_stuff[] = {
2613         NOD("self", S_IFLNK|S_IRWXUGO,
2614                 &proc_self_inode_operations, NULL, {}),
2615 };
2616
2617 static struct dentry *proc_base_instantiate(struct inode *dir,
2618         struct dentry *dentry, struct task_struct *task, const void *ptr)
2619 {
2620         const struct pid_entry *p = ptr;
2621         struct inode *inode;
2622         struct proc_inode *ei;
2623         struct dentry *error;
2624
2625         /* Allocate the inode */
2626         error = ERR_PTR(-ENOMEM);
2627         inode = new_inode(dir->i_sb);
2628         if (!inode)
2629                 goto out;
2630
2631         /* Initialize the inode */
2632         ei = PROC_I(inode);
2633         inode->i_ino = get_next_ino();
2634         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2635
2636         /*
2637          * grab the reference to the task.
2638          */
2639         ei->pid = get_task_pid(task, PIDTYPE_PID);
2640         if (!ei->pid)
2641                 goto out_iput;
2642
2643         inode->i_mode = p->mode;
2644         if (S_ISDIR(inode->i_mode))
2645                 set_nlink(inode, 2);
2646         if (S_ISLNK(inode->i_mode))
2647                 inode->i_size = 64;
2648         if (p->iop)
2649                 inode->i_op = p->iop;
2650         if (p->fop)
2651                 inode->i_fop = p->fop;
2652         ei->op = p->op;
2653         d_add(dentry, inode);
2654         error = NULL;
2655 out:
2656         return error;
2657 out_iput:
2658         iput(inode);
2659         goto out;
2660 }
2661
2662 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2663 {
2664         struct dentry *error;
2665         struct task_struct *task = get_proc_task(dir);
2666         const struct pid_entry *p, *last;
2667
2668         error = ERR_PTR(-ENOENT);
2669
2670         if (!task)
2671                 goto out_no_task;
2672
2673         /* Lookup the directory entry */
2674         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2675         for (p = proc_base_stuff; p <= last; p++) {
2676                 if (p->len != dentry->d_name.len)
2677                         continue;
2678                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2679                         break;
2680         }
2681         if (p > last)
2682                 goto out;
2683
2684         error = proc_base_instantiate(dir, dentry, task, p);
2685
2686 out:
2687         put_task_struct(task);
2688 out_no_task:
2689         return error;
2690 }
2691
2692 static int proc_base_fill_cache(struct file *filp, void *dirent,
2693         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2694 {
2695         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2696                                 proc_base_instantiate, task, p);
2697 }
2698
2699 #ifdef CONFIG_TASK_IO_ACCOUNTING
2700 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2701 {
2702         struct task_io_accounting acct = task->ioac;
2703         unsigned long flags;
2704         int result;
2705
2706         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2707         if (result)
2708                 return result;
2709
2710         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2711                 result = -EACCES;
2712                 goto out_unlock;
2713         }
2714
2715         if (whole && lock_task_sighand(task, &flags)) {
2716                 struct task_struct *t = task;
2717
2718                 task_io_accounting_add(&acct, &task->signal->ioac);
2719                 while_each_thread(task, t)
2720                         task_io_accounting_add(&acct, &t->ioac);
2721
2722                 unlock_task_sighand(task, &flags);
2723         }
2724         result = sprintf(buffer,
2725                         "rchar: %llu\n"
2726                         "wchar: %llu\n"
2727                         "syscr: %llu\n"
2728                         "syscw: %llu\n"
2729                         "read_bytes: %llu\n"
2730                         "write_bytes: %llu\n"
2731                         "cancelled_write_bytes: %llu\n",
2732                         (unsigned long long)acct.rchar,
2733                         (unsigned long long)acct.wchar,
2734                         (unsigned long long)acct.syscr,
2735                         (unsigned long long)acct.syscw,
2736                         (unsigned long long)acct.read_bytes,
2737                         (unsigned long long)acct.write_bytes,
2738                         (unsigned long long)acct.cancelled_write_bytes);
2739 out_unlock:
2740         mutex_unlock(&task->signal->cred_guard_mutex);
2741         return result;
2742 }
2743
2744 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2745 {
2746         return do_io_accounting(task, buffer, 0);
2747 }
2748
2749 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2750 {
2751         return do_io_accounting(task, buffer, 1);
2752 }
2753 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2754
2755 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2756                                 struct pid *pid, struct task_struct *task)
2757 {
2758         int err = lock_trace(task);
2759         if (!err) {
2760                 seq_printf(m, "%08x\n", task->personality);
2761                 unlock_trace(task);
2762         }
2763         return err;
2764 }
2765
2766 /*
2767  * Thread groups
2768  */
2769 static const struct file_operations proc_task_operations;
2770 static const struct inode_operations proc_task_inode_operations;
2771
2772 static const struct pid_entry tgid_base_stuff[] = {
2773         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2774         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2775         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2776         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2777 #ifdef CONFIG_NET
2778         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2779 #endif
2780         REG("environ",    S_IRUSR, proc_environ_operations),
2781         INF("auxv",       S_IRUSR, proc_pid_auxv),
2782         ONE("status",     S_IRUGO, proc_pid_status),
2783         ONE("personality", S_IRUGO, proc_pid_personality),
2784         INF("limits",     S_IRUGO, proc_pid_limits),
2785 #ifdef CONFIG_SCHED_DEBUG
2786         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2787 #endif
2788 #ifdef CONFIG_SCHED_AUTOGROUP
2789         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2790 #endif
2791         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2792 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2793         INF("syscall",    S_IRUGO, proc_pid_syscall),
2794 #endif
2795         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2796         ONE("stat",       S_IRUGO, proc_tgid_stat),
2797         ONE("statm",      S_IRUGO, proc_pid_statm),
2798         REG("maps",       S_IRUGO, proc_maps_operations),
2799 #ifdef CONFIG_NUMA
2800         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2801 #endif
2802         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2803         LNK("cwd",        proc_cwd_link),
2804         LNK("root",       proc_root_link),
2805         LNK("exe",        proc_exe_link),
2806         REG("mounts",     S_IRUGO, proc_mounts_operations),
2807         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2808         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2809 #ifdef CONFIG_PROC_PAGE_MONITOR
2810         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2811         REG("smaps",      S_IRUGO, proc_smaps_operations),
2812         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2813 #endif
2814 #ifdef CONFIG_SECURITY
2815         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2816 #endif
2817 #ifdef CONFIG_KALLSYMS
2818         INF("wchan",      S_IRUGO, proc_pid_wchan),
2819 #endif
2820 #ifdef CONFIG_STACKTRACE
2821         ONE("stack",      S_IRUGO, proc_pid_stack),
2822 #endif
2823 #ifdef CONFIG_SCHEDSTATS
2824         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2825 #endif
2826 #ifdef CONFIG_LATENCYTOP
2827         REG("latency",  S_IRUGO, proc_lstats_operations),
2828 #endif
2829 #ifdef CONFIG_PROC_PID_CPUSET
2830         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2831 #endif
2832 #ifdef CONFIG_CGROUPS
2833         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2834 #endif
2835         INF("oom_score",  S_IRUGO, proc_oom_score),
2836         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2837         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2838 #ifdef CONFIG_AUDITSYSCALL
2839         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2840         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2841 #endif
2842 #ifdef CONFIG_FAULT_INJECTION
2843         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2844 #endif
2845 #ifdef CONFIG_ELF_CORE
2846         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2847 #endif
2848 #ifdef CONFIG_TASK_IO_ACCOUNTING
2849         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2850 #endif
2851 #ifdef CONFIG_HARDWALL
2852         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2853 #endif
2854 };
2855
2856 static int proc_tgid_base_readdir(struct file * filp,
2857                              void * dirent, filldir_t filldir)
2858 {
2859         return proc_pident_readdir(filp,dirent,filldir,
2860                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2861 }
2862
2863 static const struct file_operations proc_tgid_base_operations = {
2864         .read           = generic_read_dir,
2865         .readdir        = proc_tgid_base_readdir,
2866         .llseek         = default_llseek,
2867 };
2868
2869 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2870         return proc_pident_lookup(dir, dentry,
2871                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2872 }
2873
2874 static const struct inode_operations proc_tgid_base_inode_operations = {
2875         .lookup         = proc_tgid_base_lookup,
2876         .getattr        = pid_getattr,
2877         .setattr        = proc_setattr,
2878 };
2879
2880 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2881 {
2882         struct dentry *dentry, *leader, *dir;
2883         char buf[PROC_NUMBUF];
2884         struct qstr name;
2885
2886         name.name = buf;
2887         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2888         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2889         if (dentry) {
2890                 shrink_dcache_parent(dentry);
2891                 d_drop(dentry);
2892                 dput(dentry);
2893         }
2894
2895         name.name = buf;
2896         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2897         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2898         if (!leader)
2899                 goto out;
2900
2901         name.name = "task";
2902         name.len = strlen(name.name);
2903         dir = d_hash_and_lookup(leader, &name);
2904         if (!dir)
2905                 goto out_put_leader;
2906
2907         name.name = buf;
2908         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2909         dentry = d_hash_and_lookup(dir, &name);
2910         if (dentry) {
2911                 shrink_dcache_parent(dentry);
2912                 d_drop(dentry);
2913                 dput(dentry);
2914         }
2915
2916         dput(dir);
2917 out_put_leader:
2918         dput(leader);
2919 out:
2920         return;
2921 }
2922
2923 /**
2924  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2925  * @task: task that should be flushed.
2926  *
2927  * When flushing dentries from proc, one needs to flush them from global
2928  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2929  * in. This call is supposed to do all of this job.
2930  *
2931  * Looks in the dcache for
2932  * /proc/@pid
2933  * /proc/@tgid/task/@pid
2934  * if either directory is present flushes it and all of it'ts children
2935  * from the dcache.
2936  *
2937  * It is safe and reasonable to cache /proc entries for a task until
2938  * that task exits.  After that they just clog up the dcache with
2939  * useless entries, possibly causing useful dcache entries to be
2940  * flushed instead.  This routine is proved to flush those useless
2941  * dcache entries at process exit time.
2942  *
2943  * NOTE: This routine is just an optimization so it does not guarantee
2944  *       that no dcache entries will exist at process exit time it
2945  *       just makes it very unlikely that any will persist.
2946  */
2947
2948 void proc_flush_task(struct task_struct *task)
2949 {
2950         int i;
2951         struct pid *pid, *tgid;
2952         struct upid *upid;
2953
2954         pid = task_pid(task);
2955         tgid = task_tgid(task);
2956
2957         for (i = 0; i <= pid->level; i++) {
2958                 upid = &pid->numbers[i];
2959                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2960                                         tgid->numbers[i].nr);
2961         }
2962
2963         upid = &pid->numbers[pid->level];
2964         if (upid->nr == 1)
2965                 pid_ns_release_proc(upid->ns);
2966 }
2967
2968 static struct dentry *proc_pid_instantiate(struct inode *dir,
2969                                            struct dentry * dentry,
2970                                            struct task_struct *task, const void *ptr)
2971 {
2972         struct dentry *error = ERR_PTR(-ENOENT);
2973         struct inode *inode;
2974
2975         inode = proc_pid_make_inode(dir->i_sb, task);
2976         if (!inode)
2977                 goto out;
2978
2979         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2980         inode->i_op = &proc_tgid_base_inode_operations;
2981         inode->i_fop = &proc_tgid_base_operations;
2982         inode->i_flags|=S_IMMUTABLE;
2983
2984         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2985                                                   ARRAY_SIZE(tgid_base_stuff)));
2986
2987         d_set_d_op(dentry, &pid_dentry_operations);
2988
2989         d_add(dentry, inode);
2990         /* Close the race of the process dying before we return the dentry */
2991         if (pid_revalidate(dentry, NULL))
2992                 error = NULL;
2993 out:
2994         return error;
2995 }
2996
2997 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2998 {
2999         struct dentry *result;
3000         struct task_struct *task;
3001         unsigned tgid;
3002         struct pid_namespace *ns;
3003
3004         result = proc_base_lookup(dir, dentry);
3005         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3006                 goto out;
3007
3008         tgid = name_to_int(dentry);
3009         if (tgid == ~0U)
3010                 goto out;
3011
3012         ns = dentry->d_sb->s_fs_info;
3013         rcu_read_lock();
3014         task = find_task_by_pid_ns(tgid, ns);
3015         if (task)
3016                 get_task_struct(task);
3017         rcu_read_unlock();
3018         if (!task)
3019                 goto out;
3020
3021         result = proc_pid_instantiate(dir, dentry, task, NULL);
3022         put_task_struct(task);
3023 out:
3024         return result;
3025 }
3026
3027 /*
3028  * Find the first task with tgid >= tgid
3029  *
3030  */
3031 struct tgid_iter {
3032         unsigned int tgid;
3033         struct task_struct *task;
3034 };
3035 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3036 {
3037         struct pid *pid;
3038
3039         if (iter.task)
3040                 put_task_struct(iter.task);
3041         rcu_read_lock();
3042 retry:
3043         iter.task = NULL;
3044         pid = find_ge_pid(iter.tgid, ns);
3045         if (pid) {
3046                 iter.tgid = pid_nr_ns(pid, ns);
3047                 iter.task = pid_task(pid, PIDTYPE_PID);
3048                 /* What we to know is if the pid we have find is the
3049                  * pid of a thread_group_leader.  Testing for task
3050                  * being a thread_group_leader is the obvious thing
3051                  * todo but there is a window when it fails, due to
3052                  * the pid transfer logic in de_thread.
3053                  *
3054                  * So we perform the straight forward test of seeing
3055                  * if the pid we have found is the pid of a thread
3056                  * group leader, and don't worry if the task we have
3057                  * found doesn't happen to be a thread group leader.
3058                  * As we don't care in the case of readdir.
3059                  */
3060                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3061                         iter.tgid += 1;
3062                         goto retry;
3063                 }
3064                 get_task_struct(iter.task);
3065         }
3066         rcu_read_unlock();
3067         return iter;
3068 }
3069
3070 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3071
3072 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3073         struct tgid_iter iter)
3074 {
3075         char name[PROC_NUMBUF];
3076         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3077         return proc_fill_cache(filp, dirent, filldir, name, len,
3078                                 proc_pid_instantiate, iter.task, NULL);
3079 }
3080
3081 /* for the /proc/ directory itself, after non-process stuff has been done */
3082 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3083 {
3084         unsigned int nr;
3085         struct task_struct *reaper;
3086         struct tgid_iter iter;
3087         struct pid_namespace *ns;
3088
3089         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3090                 goto out_no_task;
3091         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3092
3093         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3094         if (!reaper)
3095                 goto out_no_task;
3096
3097         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3098                 const struct pid_entry *p = &proc_base_stuff[nr];
3099                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3100                         goto out;
3101         }
3102
3103         ns = filp->f_dentry->d_sb->s_fs_info;
3104         iter.task = NULL;
3105         iter.tgid = filp->f_pos - TGID_OFFSET;
3106         for (iter = next_tgid(ns, iter);
3107              iter.task;
3108              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3109                 filp->f_pos = iter.tgid + TGID_OFFSET;
3110                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3111                         put_task_struct(iter.task);
3112                         goto out;
3113                 }
3114         }
3115         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3116 out:
3117         put_task_struct(reaper);
3118 out_no_task:
3119         return 0;
3120 }
3121
3122 /*
3123  * Tasks
3124  */
3125 static const struct pid_entry tid_base_stuff[] = {
3126         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3127         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3128         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3129         REG("environ",   S_IRUSR, proc_environ_operations),
3130         INF("auxv",      S_IRUSR, proc_pid_auxv),
3131         ONE("status",    S_IRUGO, proc_pid_status),
3132         ONE("personality", S_IRUGO, proc_pid_personality),
3133         INF("limits",    S_IRUGO, proc_pid_limits),
3134 #ifdef CONFIG_SCHED_DEBUG
3135         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3136 #endif
3137         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3138 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3139         INF("syscall",   S_IRUGO, proc_pid_syscall),
3140 #endif
3141         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3142         ONE("stat",      S_IRUGO, proc_tid_stat),
3143         ONE("statm",     S_IRUGO, proc_pid_statm),
3144         REG("maps",      S_IRUGO, proc_maps_operations),
3145 #ifdef CONFIG_NUMA
3146         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3147 #endif
3148         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3149         LNK("cwd",       proc_cwd_link),
3150         LNK("root",      proc_root_link),
3151         LNK("exe",       proc_exe_link),
3152         REG("mounts",    S_IRUGO, proc_mounts_operations),
3153         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3154 #ifdef CONFIG_PROC_PAGE_MONITOR
3155         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3156         REG("smaps",     S_IRUGO, proc_smaps_operations),
3157         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3158 #endif
3159 #ifdef CONFIG_SECURITY
3160         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3161 #endif
3162 #ifdef CONFIG_KALLSYMS
3163         INF("wchan",     S_IRUGO, proc_pid_wchan),
3164 #endif
3165 #ifdef CONFIG_STACKTRACE
3166         ONE("stack",      S_IRUGO, proc_pid_stack),
3167 #endif
3168 #ifdef CONFIG_SCHEDSTATS
3169         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3170 #endif
3171 #ifdef CONFIG_LATENCYTOP
3172         REG("latency",  S_IRUGO, proc_lstats_operations),
3173 #endif
3174 #ifdef CONFIG_PROC_PID_CPUSET
3175         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3176 #endif
3177 #ifdef CONFIG_CGROUPS
3178         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3179 #endif
3180         INF("oom_score", S_IRUGO, proc_oom_score),
3181         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3182         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3183 #ifdef CONFIG_AUDITSYSCALL
3184         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3185         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3186 #endif
3187 #ifdef CONFIG_FAULT_INJECTION
3188         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3189 #endif
3190 #ifdef CONFIG_TASK_IO_ACCOUNTING
3191         INF("io",       S_IRUSR, proc_tid_io_accounting),
3192 #endif
3193 #ifdef CONFIG_HARDWALL
3194         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3195 #endif
3196 };
3197
3198 static int proc_tid_base_readdir(struct file * filp,
3199                              void * dirent, filldir_t filldir)
3200 {
3201         return proc_pident_readdir(filp,dirent,filldir,
3202                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3203 }
3204
3205 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3206         return proc_pident_lookup(dir, dentry,
3207                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3208 }
3209
3210 static const struct file_operations proc_tid_base_operations = {
3211         .read           = generic_read_dir,
3212         .readdir        = proc_tid_base_readdir,
3213         .llseek         = default_llseek,
3214 };
3215
3216 static const struct inode_operations proc_tid_base_inode_operations = {
3217         .lookup         = proc_tid_base_lookup,
3218         .getattr        = pid_getattr,
3219         .setattr        = proc_setattr,
3220 };
3221
3222 static struct dentry *proc_task_instantiate(struct inode *dir,
3223         struct dentry *dentry, struct task_struct *task, const void *ptr)
3224 {
3225         struct dentry *error = ERR_PTR(-ENOENT);
3226         struct inode *inode;
3227         inode = proc_pid_make_inode(dir->i_sb, task);
3228
3229         if (!inode)
3230                 goto out;
3231         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3232         inode->i_op = &proc_tid_base_inode_operations;
3233         inode->i_fop = &proc_tid_base_operations;
3234         inode->i_flags|=S_IMMUTABLE;
3235
3236         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3237                                                   ARRAY_SIZE(tid_base_stuff)));
3238
3239         d_set_d_op(dentry, &pid_dentry_operations);
3240
3241         d_add(dentry, inode);
3242         /* Close the race of the process dying before we return the dentry */
3243         if (pid_revalidate(dentry, NULL))
3244                 error = NULL;
3245 out:
3246         return error;
3247 }
3248
3249 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3250 {
3251         struct dentry *result = ERR_PTR(-ENOENT);
3252         struct task_struct *task;
3253         struct task_struct *leader = get_proc_task(dir);
3254         unsigned tid;
3255         struct pid_namespace *ns;
3256
3257         if (!leader)
3258                 goto out_no_task;
3259
3260         tid = name_to_int(dentry);
3261         if (tid == ~0U)
3262                 goto out;
3263
3264         ns = dentry->d_sb->s_fs_info;
3265         rcu_read_lock();
3266         task = find_task_by_pid_ns(tid, ns);
3267         if (task)
3268                 get_task_struct(task);
3269         rcu_read_unlock();
3270         if (!task)
3271                 goto out;
3272         if (!same_thread_group(leader, task))
3273                 goto out_drop_task;
3274
3275         result = proc_task_instantiate(dir, dentry, task, NULL);
3276 out_drop_task:
3277         put_task_struct(task);
3278 out:
3279         put_task_struct(leader);
3280 out_no_task:
3281         return result;
3282 }
3283
3284 /*
3285  * Find the first tid of a thread group to return to user space.
3286  *
3287  * Usually this is just the thread group leader, but if the users
3288  * buffer was too small or there was a seek into the middle of the
3289  * directory we have more work todo.
3290  *
3291  * In the case of a short read we start with find_task_by_pid.
3292  *
3293  * In the case of a seek we start with the leader and walk nr
3294  * threads past it.
3295  */
3296 static struct task_struct *first_tid(struct task_struct *leader,
3297                 int tid, int nr, struct pid_namespace *ns)
3298 {
3299         struct task_struct *pos;
3300
3301         rcu_read_lock();
3302         /* Attempt to start with the pid of a thread */
3303         if (tid && (nr > 0)) {
3304                 pos = find_task_by_pid_ns(tid, ns);
3305                 if (pos && (pos->group_leader == leader))
3306                         goto found;
3307         }
3308
3309         /* If nr exceeds the number of threads there is nothing todo */
3310         pos = NULL;
3311         if (nr && nr >= get_nr_threads(leader))
3312                 goto out;
3313
3314         /* If we haven't found our starting place yet start
3315          * with the leader and walk nr threads forward.
3316          */
3317         for (pos = leader; nr > 0; --nr) {
3318                 pos = next_thread(pos);
3319                 if (pos == leader) {
3320                         pos = NULL;
3321                         goto out;
3322                 }
3323         }
3324 found:
3325         get_task_struct(pos);
3326 out:
3327         rcu_read_unlock();
3328         return pos;
3329 }
3330
3331 /*
3332  * Find the next thread in the thread list.
3333  * Return NULL if there is an error or no next thread.
3334  *
3335  * The reference to the input task_struct is released.
3336  */
3337 static struct task_struct *next_tid(struct task_struct *start)
3338 {
3339         struct task_struct *pos = NULL;
3340         rcu_read_lock();
3341         if (pid_alive(start)) {
3342                 pos = next_thread(start);
3343                 if (thread_group_leader(pos))
3344                         pos = NULL;
3345                 else
3346                         get_task_struct(pos);
3347         }
3348         rcu_read_unlock();
3349         put_task_struct(start);
3350         return pos;
3351 }
3352
3353 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3354         struct task_struct *task, int tid)
3355 {
3356         char name[PROC_NUMBUF];
3357         int len = snprintf(name, sizeof(name), "%d", tid);
3358         return proc_fill_cache(filp, dirent, filldir, name, len,
3359                                 proc_task_instantiate, task, NULL);
3360 }
3361
3362 /* for the /proc/TGID/task/ directories */
3363 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3364 {
3365         struct dentry *dentry = filp->f_path.dentry;
3366         struct inode *inode = dentry->d_inode;
3367         struct task_struct *leader = NULL;
3368         struct task_struct *task;
3369         int retval = -ENOENT;
3370         ino_t ino;
3371         int tid;
3372         struct pid_namespace *ns;
3373
3374         task = get_proc_task(inode);
3375         if (!task)
3376                 goto out_no_task;
3377         rcu_read_lock();
3378         if (pid_alive(task)) {
3379                 leader = task->group_leader;
3380                 get_task_struct(leader);
3381         }
3382         rcu_read_unlock();
3383         put_task_struct(task);
3384         if (!leader)
3385                 goto out_no_task;
3386         retval = 0;
3387
3388         switch ((unsigned long)filp->f_pos) {
3389         case 0:
3390                 ino = inode->i_ino;
3391                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3392                         goto out;
3393                 filp->f_pos++;
3394                 /* fall through */
3395         case 1:
3396                 ino = parent_ino(dentry);
3397                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3398                         goto out;
3399                 filp->f_pos++;
3400                 /* fall through */
3401         }
3402
3403         /* f_version caches the tgid value that the last readdir call couldn't
3404          * return. lseek aka telldir automagically resets f_version to 0.
3405          */
3406         ns = filp->f_dentry->d_sb->s_fs_info;
3407         tid = (int)filp->f_version;
3408         filp->f_version = 0;
3409         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3410              task;
3411              task = next_tid(task), filp->f_pos++) {
3412                 tid = task_pid_nr_ns(task, ns);
3413                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3414                         /* returning this tgid failed, save it as the first
3415                          * pid for the next readir call */
3416                         filp->f_version = (u64)tid;
3417                         put_task_struct(task);
3418                         break;
3419                 }
3420         }
3421 out:
3422         put_task_struct(leader);
3423 out_no_task:
3424         return retval;
3425 }
3426
3427 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3428 {
3429         struct inode *inode = dentry->d_inode;
3430         struct task_struct *p = get_proc_task(inode);
3431         generic_fillattr(inode, stat);
3432
3433         if (p) {
3434                 stat->nlink += get_nr_threads(p);
3435                 put_task_struct(p);
3436         }
3437
3438         return 0;
3439 }
3440
3441 static const struct inode_operations proc_task_inode_operations = {
3442         .lookup         = proc_task_lookup,
3443         .getattr        = proc_task_getattr,
3444         .setattr        = proc_setattr,
3445 };
3446
3447 static const struct file_operations proc_task_operations = {
3448         .read           = generic_read_dir,
3449         .readdir        = proc_task_readdir,
3450         .llseek         = default_llseek,
3451 };