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