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/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93
  94 /* NOTE:
  95  *      Implementing inode permission operations in /proc is almost
  96  *      certainly an error.  Permission checks need to happen during
  97  *      each system call not at open time.  The reason is that most of
  98  *      what we wish to check for permissions in /proc varies at runtime.
  99  *
 100  *      The classic example of a problem is opening file descriptors
 101  *      in /proc for a task before it execs a suid executable.
 102  */
 103
 104 struct pid_entry {
 105         char *name;
 106         int len;
 107         umode_t mode;
 108         const struct inode_operations *iop;
 109         const struct file_operations *fop;
 110         union proc_op op;
 111 };
 112
 113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 114         .name = (NAME),                                 \
 115         .len  = sizeof(NAME) - 1,                       \
 116         .mode = MODE,                                   \
 117         .iop  = IOP,                                    \
 118         .fop  = FOP,                                    \
 119         .op   = OP,                                     \
 120 }
 121
 122 #define DIR(NAME, MODE, iops, fops)     \
 123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 124 #define LNK(NAME, get_link)                                     \
 125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 126                 &proc_pid_link_inode_operations, NULL,          \
 127                 { .proc_get_link = get_link } )
 128 #define REG(NAME, MODE, fops)                           \
 129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 130 #define INF(NAME, MODE, read)                           \
 131         NOD(NAME, (S_IFREG|(MODE)),                     \
 132                 NULL, &proc_info_file_operations,       \
 133                 { .proc_read = read } )
 134 #define ONE(NAME, MODE, show)                           \
 135         NOD(NAME, (S_IFREG|(MODE)),                     \
 136                 NULL, &proc_single_file_operations,     \
 137                 { .proc_show = show } )
 138
 139 static int proc_fd_permission(struct inode *inode, int mask);
 140
 141 /*
 142  * Count the number of hardlinks for the pid_entry table, excluding the .
 143  * and .. links.
 144  */
 145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 146         unsigned int n)
 147 {
 148         unsigned int i;
 149         unsigned int count;
 150
 151         count = 0;
 152         for (i = 0; i < n; ++i) {
 153                 if (S_ISDIR(entries[i].mode))
 154                         ++count;
 155         }
 156
 157         return count;
 158 }
 159
 160 static int get_task_root(struct task_struct *task, struct path *root)
 161 {
 162         int result = -ENOENT;
 163
 164         task_lock(task);
 165         if (task->fs) {
 166                 get_fs_root(task->fs, root);
 167                 result = 0;
 168         }
 169         task_unlock(task);
 170         return result;
 171 }
 172
 173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 174 {
 175         struct task_struct *task = get_proc_task(dentry->d_inode);
 176         int result = -ENOENT;
 177
 178         if (task) {
 179                 task_lock(task);
 180                 if (task->fs) {
 181                         get_fs_pwd(task->fs, path);
 182                         result = 0;
 183                 }
 184                 task_unlock(task);
 185                 put_task_struct(task);
 186         }
 187         return result;
 188 }
 189
 190 static int proc_root_link(struct dentry *dentry, struct path *path)
 191 {
 192         struct task_struct *task = get_proc_task(dentry->d_inode);
 193         int result = -ENOENT;
 194
 195         if (task) {
 196                 result = get_task_root(task, path);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 203 {
 204         int res = 0;
 205         unsigned int len;
 206         struct mm_struct *mm = get_task_mm(task);
 207         if (!mm)
 208                 goto out;
 209         if (!mm->arg_end)
 210                 goto out_mm;    /* Shh! No looking before we're done */
 211
 212         len = mm->arg_end - mm->arg_start;
 213
 214         if (len > PAGE_SIZE)
 215                 len = PAGE_SIZE;
 216
 217         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 218
 219         // If the nul at the end of args has been overwritten, then
 220         // assume application is using setproctitle(3).
 221         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 222                 len = strnlen(buffer, res);
 223                 if (len < res) {
 224                     res = len;
 225                 } else {
 226                         len = mm->env_end - mm->env_start;
 227                         if (len > PAGE_SIZE - res)
 228                                 len = PAGE_SIZE - res;
 229                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 230                         res = strnlen(buffer, res);
 231                 }
 232         }
 233 out_mm:
 234         mmput(mm);
 235 out:
 236         return res;
 237 }
 238
 239 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 240 {
 241         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 242         int res = PTR_ERR(mm);
 243         if (mm && !IS_ERR(mm)) {
 244                 unsigned int nwords = 0;
 245                 do {
 246                         nwords += 2;
 247                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 248                 res = nwords * sizeof(mm->saved_auxv[0]);
 249                 if (res > PAGE_SIZE)
 250                         res = PAGE_SIZE;
 251                 memcpy(buffer, mm->saved_auxv, res);
 252                 mmput(mm);
 253         }
 254         return res;
 255 }
 256
 257
 258 #ifdef CONFIG_KALLSYMS
 259 /*
 260  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 261  * Returns the resolved symbol.  If that fails, simply return the address.
 262  */
 263 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 264 {
 265         unsigned long wchan;
 266         char symname[KSYM_NAME_LEN];
 267
 268         wchan = get_wchan(task);
 269
 270         if (lookup_symbol_name(wchan, symname) < 0)
 271                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 272                         return 0;
 273                 else
 274                         return sprintf(buffer, "%lu", wchan);
 275         else
 276                 return sprintf(buffer, "%s", symname);
 277 }
 278 #endif /* CONFIG_KALLSYMS */
 279
 280 static int lock_trace(struct task_struct *task)
 281 {
 282         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 283         if (err)
 284                 return err;
 285         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 286                 mutex_unlock(&task->signal->cred_guard_mutex);
 287                 return -EPERM;
 288         }
 289         return 0;
 290 }
 291
 292 static void unlock_trace(struct task_struct *task)
 293 {
 294         mutex_unlock(&task->signal->cred_guard_mutex);
 295 }
 296
 297 #ifdef CONFIG_STACKTRACE
 298
 299 #define MAX_STACK_TRACE_DEPTH   64
 300
 301 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 302                           struct pid *pid, struct task_struct *task)
 303 {
 304         struct stack_trace trace;
 305         unsigned long *entries;
 306         int err;
 307         int i;
 308
 309         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 310         if (!entries)
 311                 return -ENOMEM;
 312
 313         trace.nr_entries        = 0;
 314         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 315         trace.entries           = entries;
 316         trace.skip              = 0;
 317
 318         err = lock_trace(task);
 319         if (!err) {
 320                 save_stack_trace_tsk(task, &trace);
 321
 322                 for (i = 0; i < trace.nr_entries; i++) {
 323                         seq_printf(m, "[<%pK>] %pS\n",
 324                                    (void *)entries[i], (void *)entries[i]);
 325                 }
 326                 unlock_trace(task);
 327         }
 328         kfree(entries);
 329
 330         return err;
 331 }
 332 #endif
 333
 334 #ifdef CONFIG_SCHEDSTATS
 335 /*
 336  * Provides /proc/PID/schedstat
 337  */
 338 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 339 {
 340         return sprintf(buffer, "%llu %llu %lu\n",
 341                         (unsigned long long)task->se.sum_exec_runtime,
 342                         (unsigned long long)task->sched_info.run_delay,
 343                         task->sched_info.pcount);
 344 }
 345 #endif
 346
 347 #ifdef CONFIG_LATENCYTOP
 348 static int lstats_show_proc(struct seq_file *m, void *v)
 349 {
 350         int i;
 351         struct inode *inode = m->private;
 352         struct task_struct *task = get_proc_task(inode);
 353
 354         if (!task)
 355                 return -ESRCH;
 356         seq_puts(m, "Latency Top version : v0.1\n");
 357         for (i = 0; i < 32; i++) {
 358                 struct latency_record *lr = &task->latency_record[i];
 359                 if (lr->backtrace[0]) {
 360                         int q;
 361                         seq_printf(m, "%i %li %li",
 362                                    lr->count, lr->time, lr->max);
 363                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 364                                 unsigned long bt = lr->backtrace[q];
 365                                 if (!bt)
 366                                         break;
 367                                 if (bt == ULONG_MAX)
 368                                         break;
 369                                 seq_printf(m, " %ps", (void *)bt);
 370                         }
 371                         seq_putc(m, '\n');
 372                 }
 373
 374         }
 375         put_task_struct(task);
 376         return 0;
 377 }
 378
 379 static int lstats_open(struct inode *inode, struct file *file)
 380 {
 381         return single_open(file, lstats_show_proc, inode);
 382 }
 383
 384 static ssize_t lstats_write(struct file *file, const char __user *buf,
 385                             size_t count, loff_t *offs)
 386 {
 387         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 388
 389         if (!task)
 390                 return -ESRCH;
 391         clear_all_latency_tracing(task);
 392         put_task_struct(task);
 393
 394         return count;
 395 }
 396
 397 static const struct file_operations proc_lstats_operations = {
 398         .open           = lstats_open,
 399         .read           = seq_read,
 400         .write          = lstats_write,
 401         .llseek         = seq_lseek,
 402         .release        = single_release,
 403 };
 404
 405 #endif
 406
 407 static int proc_oom_score(struct task_struct *task, char *buffer)
 408 {
 409         unsigned long totalpages = totalram_pages + total_swap_pages;
 410         unsigned long points = 0;
 411
 412         read_lock(&tasklist_lock);
 413         if (pid_alive(task))
 414                 points = oom_badness(task, NULL, NULL, totalpages) *
 415                                                 1000 / totalpages;
 416         read_unlock(&tasklist_lock);
 417         return sprintf(buffer, "%lu\n", points);
 418 }
 419
 420 struct limit_names {
 421         char *name;
 422         char *unit;
 423 };
 424
 425 static const struct limit_names lnames[RLIM_NLIMITS] = {
 426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 428         [RLIMIT_DATA] = {"Max data size", "bytes"},
 429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 432         [RLIMIT_NPROC] = {"Max processes", "processes"},
 433         [RLIMIT_NOFILE] = {"Max open files", "files"},
 434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 435         [RLIMIT_AS] = {"Max address space", "bytes"},
 436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 439         [RLIMIT_NICE] = {"Max nice priority", NULL},
 440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 442 };
 443
 444 /* Display limits for a process */
 445 static int proc_pid_limits(struct task_struct *task, char *buffer)
 446 {
 447         unsigned int i;
 448         int count = 0;
 449         unsigned long flags;
 450         char *bufptr = buffer;
 451
 452         struct rlimit rlim[RLIM_NLIMITS];
 453
 454         if (!lock_task_sighand(task, &flags))
 455                 return 0;
 456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 457         unlock_task_sighand(task, &flags);
 458
 459         /*
 460          * print the file header
 461          */
 462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 463                         "Limit", "Soft Limit", "Hard Limit", "Units");
 464
 465         for (i = 0; i < RLIM_NLIMITS; i++) {
 466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 468                                          lnames[i].name, "unlimited");
 469                 else
 470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 471                                          lnames[i].name, rlim[i].rlim_cur);
 472
 473                 if (rlim[i].rlim_max == RLIM_INFINITY)
 474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 475                 else
 476                         count += sprintf(&bufptr[count], "%-20lu ",
 477                                          rlim[i].rlim_max);
 478
 479                 if (lnames[i].unit)
 480                         count += sprintf(&bufptr[count], "%-10s\n",
 481                                          lnames[i].unit);
 482                 else
 483                         count += sprintf(&bufptr[count], "\n");
 484         }
 485
 486         return count;
 487 }
 488
 489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 491 {
 492         long nr;
 493         unsigned long args[6], sp, pc;
 494         int res = lock_trace(task);
 495         if (res)
 496                 return res;
 497
 498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 499                 res = sprintf(buffer, "running\n");
 500         else if (nr < 0)
 501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 502         else
 503                 res = sprintf(buffer,
 504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 505                        nr,
 506                        args[0], args[1], args[2], args[3], args[4], args[5],
 507                        sp, pc);
 508         unlock_trace(task);
 509         return res;
 510 }
 511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 512
 513 /************************************************************************/
 514 /*                       Here the fs part begins                        */
 515 /************************************************************************/
 516
 517 /* permission checks */
 518 static int proc_fd_access_allowed(struct inode *inode)
 519 {
 520         struct task_struct *task;
 521         int allowed = 0;
 522         /* Allow access to a task's file descriptors if it is us or we
 523          * may use ptrace attach to the process and find out that
 524          * information.
 525          */
 526         task = get_proc_task(inode);
 527         if (task) {
 528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 529                 put_task_struct(task);
 530         }
 531         return allowed;
 532 }
 533
 534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 535 {
 536         int error;
 537         struct inode *inode = dentry->d_inode;
 538
 539         if (attr->ia_valid & ATTR_MODE)
 540                 return -EPERM;
 541
 542         error = inode_change_ok(inode, attr);
 543         if (error)
 544                 return error;
 545
 546         if ((attr->ia_valid & ATTR_SIZE) &&
 547             attr->ia_size != i_size_read(inode)) {
 548                 error = vmtruncate(inode, attr->ia_size);
 549                 if (error)
 550                         return error;
 551         }
 552
 553         setattr_copy(inode, attr);
 554         mark_inode_dirty(inode);
 555         return 0;
 556 }
 557
 558 /*
 559  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 560  * or euid/egid (for hide_pid_min=2)?
 561  */
 562 static bool has_pid_permissions(struct pid_namespace *pid,
 563                                  struct task_struct *task,
 564                                  int hide_pid_min)
 565 {
 566         if (pid->hide_pid < hide_pid_min)
 567                 return true;
 568         if (in_group_p(pid->pid_gid))
 569                 return true;
 570         return ptrace_may_access(task, PTRACE_MODE_READ);
 571 }
 572
 573
 574 static int proc_pid_permission(struct inode *inode, int mask)
 575 {
 576         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 577         struct task_struct *task;
 578         bool has_perms;
 579
 580         task = get_proc_task(inode);
 581         if (!task)
 582                 return -ESRCH;
 583         has_perms = has_pid_permissions(pid, task, 1);
 584         put_task_struct(task);
 585
 586         if (!has_perms) {
 587                 if (pid->hide_pid == 2) {
 588                         /*
 589                          * Let's make getdents(), stat(), and open()
 590                          * consistent with each other.  If a process
 591                          * may not stat() a file, it shouldn't be seen
 592                          * in procfs at all.
 593                          */
 594                         return -ENOENT;
 595                 }
 596
 597                 return -EPERM;
 598         }
 599         return generic_permission(inode, mask);
 600 }
 601
 602
 603
 604 static const struct inode_operations proc_def_inode_operations = {
 605         .setattr        = proc_setattr,
 606 };
 607
 608 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 609
 610 static ssize_t proc_info_read(struct file * file, char __user * buf,
 611                           size_t count, loff_t *ppos)
 612 {
 613         struct inode * inode = file->f_path.dentry->d_inode;
 614         unsigned long page;
 615         ssize_t length;
 616         struct task_struct *task = get_proc_task(inode);
 617
 618         length = -ESRCH;
 619         if (!task)
 620                 goto out_no_task;
 621
 622         if (count > PROC_BLOCK_SIZE)
 623                 count = PROC_BLOCK_SIZE;
 624
 625         length = -ENOMEM;
 626         if (!(page = __get_free_page(GFP_TEMPORARY)))
 627                 goto out;
 628
 629         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 630
 631         if (length >= 0)
 632                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 633         free_page(page);
 634 out:
 635         put_task_struct(task);
 636 out_no_task:
 637         return length;
 638 }
 639
 640 static const struct file_operations proc_info_file_operations = {
 641         .read           = proc_info_read,
 642         .llseek         = generic_file_llseek,
 643 };
 644
 645 static int proc_single_show(struct seq_file *m, void *v)
 646 {
 647         struct inode *inode = m->private;
 648         struct pid_namespace *ns;
 649         struct pid *pid;
 650         struct task_struct *task;
 651         int ret;
 652
 653         ns = inode->i_sb->s_fs_info;
 654         pid = proc_pid(inode);
 655         task = get_pid_task(pid, PIDTYPE_PID);
 656         if (!task)
 657                 return -ESRCH;
 658
 659         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 660
 661         put_task_struct(task);
 662         return ret;
 663 }
 664
 665 static int proc_single_open(struct inode *inode, struct file *filp)
 666 {
 667         return single_open(filp, proc_single_show, inode);
 668 }
 669
 670 static const struct file_operations proc_single_file_operations = {
 671         .open           = proc_single_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = single_release,
 675 };
 676
 677 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 678 {
 679         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 680         struct mm_struct *mm;
 681
 682         if (!task)
 683                 return -ESRCH;
 684
 685         mm = mm_access(task, mode);
 686         put_task_struct(task);
 687
 688         if (IS_ERR(mm))
 689                 return PTR_ERR(mm);
 690
 691         if (mm) {
 692                 /* ensure this mm_struct can't be freed */
 693                 atomic_inc(&mm->mm_count);
 694                 /* but do not pin its memory */
 695                 mmput(mm);
 696         }
 697
 698         /* OK to pass negative loff_t, we can catch out-of-range */
 699         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 700         file->private_data = mm;
 701
 702         return 0;
 703 }
 704
 705 static int mem_open(struct inode *inode, struct file *file)
 706 {
 707         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
 708 }
 709
 710 static ssize_t mem_rw(struct file *file, char __user *buf,
 711                         size_t count, loff_t *ppos, int write)
 712 {
 713         struct mm_struct *mm = file->private_data;
 714         unsigned long addr = *ppos;
 715         ssize_t copied;
 716         char *page;
 717
 718         if (!mm)
 719                 return 0;
 720
 721         page = (char *)__get_free_page(GFP_TEMPORARY);
 722         if (!page)
 723                 return -ENOMEM;
 724
 725         copied = 0;
 726         if (!atomic_inc_not_zero(&mm->mm_users))
 727                 goto free;
 728
 729         while (count > 0) {
 730                 int this_len = min_t(int, count, PAGE_SIZE);
 731
 732                 if (write && copy_from_user(page, buf, this_len)) {
 733                         copied = -EFAULT;
 734                         break;
 735                 }
 736
 737                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 738                 if (!this_len) {
 739                         if (!copied)
 740                                 copied = -EIO;
 741                         break;
 742                 }
 743
 744                 if (!write && copy_to_user(buf, page, this_len)) {
 745                         copied = -EFAULT;
 746                         break;
 747                 }
 748
 749                 buf += this_len;
 750                 addr += this_len;
 751                 copied += this_len;
 752                 count -= this_len;
 753         }
 754         *ppos = addr;
 755
 756         mmput(mm);
 757 free:
 758         free_page((unsigned long) page);
 759         return copied;
 760 }
 761
 762 static ssize_t mem_read(struct file *file, char __user *buf,
 763                         size_t count, loff_t *ppos)
 764 {
 765         return mem_rw(file, buf, count, ppos, 0);
 766 }
 767
 768 static ssize_t mem_write(struct file *file, const char __user *buf,
 769                          size_t count, loff_t *ppos)
 770 {
 771         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 772 }
 773
 774 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 775 {
 776         switch (orig) {
 777         case 0:
 778                 file->f_pos = offset;
 779                 break;
 780         case 1:
 781                 file->f_pos += offset;
 782                 break;
 783         default:
 784                 return -EINVAL;
 785         }
 786         force_successful_syscall_return();
 787         return file->f_pos;
 788 }
 789
 790 static int mem_release(struct inode *inode, struct file *file)
 791 {
 792         struct mm_struct *mm = file->private_data;
 793         if (mm)
 794                 mmdrop(mm);
 795         return 0;
 796 }
 797
 798 static const struct file_operations proc_mem_operations = {
 799         .llseek         = mem_lseek,
 800         .read           = mem_read,
 801         .write          = mem_write,
 802         .open           = mem_open,
 803         .release        = mem_release,
 804 };
 805
 806 static int environ_open(struct inode *inode, struct file *file)
 807 {
 808         return __mem_open(inode, file, PTRACE_MODE_READ);
 809 }
 810
 811 static ssize_t environ_read(struct file *file, char __user *buf,
 812                         size_t count, loff_t *ppos)
 813 {
 814         char *page;
 815         unsigned long src = *ppos;
 816         int ret = 0;
 817         struct mm_struct *mm = file->private_data;
 818
 819         if (!mm)
 820                 return 0;
 821
 822         page = (char *)__get_free_page(GFP_TEMPORARY);
 823         if (!page)
 824                 return -ENOMEM;
 825
 826         ret = 0;
 827         if (!atomic_inc_not_zero(&mm->mm_users))
 828                 goto free;
 829         while (count > 0) {
 830                 int this_len, retval, max_len;
 831
 832                 this_len = mm->env_end - (mm->env_start + src);
 833
 834                 if (this_len <= 0)
 835                         break;
 836
 837                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 838                 this_len = (this_len > max_len) ? max_len : this_len;
 839
 840                 retval = access_remote_vm(mm, (mm->env_start + src),
 841                         page, this_len, 0);
 842
 843                 if (retval <= 0) {
 844                         ret = retval;
 845                         break;
 846                 }
 847
 848                 if (copy_to_user(buf, page, retval)) {
 849                         ret = -EFAULT;
 850                         break;
 851                 }
 852
 853                 ret += retval;
 854                 src += retval;
 855                 buf += retval;
 856                 count -= retval;
 857         }
 858         *ppos = src;
 859         mmput(mm);
 860
 861 free:
 862         free_page((unsigned long) page);
 863         return ret;
 864 }
 865
 866 static const struct file_operations proc_environ_operations = {
 867         .open           = environ_open,
 868         .read           = environ_read,
 869         .llseek         = generic_file_llseek,
 870         .release        = mem_release,
 871 };
 872
 873 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 874                                 size_t count, loff_t *ppos)
 875 {
 876         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 877         char buffer[PROC_NUMBUF];
 878         size_t len;
 879         int oom_adjust = OOM_DISABLE;
 880         unsigned long flags;
 881
 882         if (!task)
 883                 return -ESRCH;
 884
 885         if (lock_task_sighand(task, &flags)) {
 886                 oom_adjust = task->signal->oom_adj;
 887                 unlock_task_sighand(task, &flags);
 888         }
 889
 890         put_task_struct(task);
 891
 892         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 893
 894         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 895 }
 896
 897 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 898                                 size_t count, loff_t *ppos)
 899 {
 900         struct task_struct *task;
 901         char buffer[PROC_NUMBUF];
 902         int oom_adjust;
 903         unsigned long flags;
 904         int err;
 905
 906         memset(buffer, 0, sizeof(buffer));
 907         if (count > sizeof(buffer) - 1)
 908                 count = sizeof(buffer) - 1;
 909         if (copy_from_user(buffer, buf, count)) {
 910                 err = -EFAULT;
 911                 goto out;
 912         }
 913
 914         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 915         if (err)
 916                 goto out;
 917         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 918              oom_adjust != OOM_DISABLE) {
 919                 err = -EINVAL;
 920                 goto out;
 921         }
 922
 923         task = get_proc_task(file->f_path.dentry->d_inode);
 924         if (!task) {
 925                 err = -ESRCH;
 926                 goto out;
 927         }
 928
 929         task_lock(task);
 930         if (!task->mm) {
 931                 err = -EINVAL;
 932                 goto err_task_lock;
 933         }
 934
 935         if (!lock_task_sighand(task, &flags)) {
 936                 err = -ESRCH;
 937                 goto err_task_lock;
 938         }
 939
 940         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 941                 err = -EACCES;
 942                 goto err_sighand;
 943         }
 944
 945         /*
 946          * Warn that /proc/pid/oom_adj is deprecated, see
 947          * Documentation/feature-removal-schedule.txt.
 948          */
 949         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 950                   current->comm, task_pid_nr(current), task_pid_nr(task),
 951                   task_pid_nr(task));
 952         task->signal->oom_adj = oom_adjust;
 953         /*
 954          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 955          * value is always attainable.
 956          */
 957         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 958                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 959         else
 960                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 961                                                                 -OOM_DISABLE;
 962         trace_oom_score_adj_update(task);
 963 err_sighand:
 964         unlock_task_sighand(task, &flags);
 965 err_task_lock:
 966         task_unlock(task);
 967         put_task_struct(task);
 968 out:
 969         return err < 0 ? err : count;
 970 }
 971
 972 static const struct file_operations proc_oom_adjust_operations = {
 973         .read           = oom_adjust_read,
 974         .write          = oom_adjust_write,
 975         .llseek         = generic_file_llseek,
 976 };
 977
 978 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 979                                         size_t count, loff_t *ppos)
 980 {
 981         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 982         char buffer[PROC_NUMBUF];
 983         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 984         unsigned long flags;
 985         size_t len;
 986
 987         if (!task)
 988                 return -ESRCH;
 989         if (lock_task_sighand(task, &flags)) {
 990                 oom_score_adj = task->signal->oom_score_adj;
 991                 unlock_task_sighand(task, &flags);
 992         }
 993         put_task_struct(task);
 994         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 995         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 996 }
 997
 998 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 999                                         size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task;
1002         char buffer[PROC_NUMBUF];
1003         unsigned long flags;
1004         int oom_score_adj;
1005         int err;
1006
1007         memset(buffer, 0, sizeof(buffer));
1008         if (count > sizeof(buffer) - 1)
1009                 count = sizeof(buffer) - 1;
1010         if (copy_from_user(buffer, buf, count)) {
1011                 err = -EFAULT;
1012                 goto out;
1013         }
1014
1015         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016         if (err)
1017                 goto out;
1018         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         task = get_proc_task(file->f_path.dentry->d_inode);
1025         if (!task) {
1026                 err = -ESRCH;
1027                 goto out;
1028         }
1029
1030         task_lock(task);
1031         if (!task->mm) {
1032                 err = -EINVAL;
1033                 goto err_task_lock;
1034         }
1035
1036         if (!lock_task_sighand(task, &flags)) {
1037                 err = -ESRCH;
1038                 goto err_task_lock;
1039         }
1040
1041         if (oom_score_adj < task->signal->oom_score_adj_min &&
1042                         !capable(CAP_SYS_RESOURCE)) {
1043                 err = -EACCES;
1044                 goto err_sighand;
1045         }
1046
1047         task->signal->oom_score_adj = oom_score_adj;
1048         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049                 task->signal->oom_score_adj_min = oom_score_adj;
1050         trace_oom_score_adj_update(task);
1051         /*
1052          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053          * always attainable.
1054          */
1055         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056                 task->signal->oom_adj = OOM_DISABLE;
1057         else
1058                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059                                                         OOM_SCORE_ADJ_MAX;
1060 err_sighand:
1061         unlock_task_sighand(task, &flags);
1062 err_task_lock:
1063         task_unlock(task);
1064         put_task_struct(task);
1065 out:
1066         return err < 0 ? err : count;
1067 }
1068
1069 static const struct file_operations proc_oom_score_adj_operations = {
1070         .read           = oom_score_adj_read,
1071         .write          = oom_score_adj_write,
1072         .llseek         = default_llseek,
1073 };
1074
1075 #ifdef CONFIG_AUDITSYSCALL
1076 #define TMPBUFLEN 21
1077 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078                                   size_t count, loff_t *ppos)
1079 {
1080         struct inode * inode = file->f_path.dentry->d_inode;
1081         struct task_struct *task = get_proc_task(inode);
1082         ssize_t length;
1083         char tmpbuf[TMPBUFLEN];
1084
1085         if (!task)
1086                 return -ESRCH;
1087         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088                                 audit_get_loginuid(task));
1089         put_task_struct(task);
1090         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091 }
1092
1093 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094                                    size_t count, loff_t *ppos)
1095 {
1096         struct inode * inode = file->f_path.dentry->d_inode;
1097         char *page, *tmp;
1098         ssize_t length;
1099         uid_t loginuid;
1100
1101         rcu_read_lock();
1102         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103                 rcu_read_unlock();
1104                 return -EPERM;
1105         }
1106         rcu_read_unlock();
1107
1108         if (count >= PAGE_SIZE)
1109                 count = PAGE_SIZE - 1;
1110
1111         if (*ppos != 0) {
1112                 /* No partial writes. */
1113                 return -EINVAL;
1114         }
1115         page = (char*)__get_free_page(GFP_TEMPORARY);
1116         if (!page)
1117                 return -ENOMEM;
1118         length = -EFAULT;
1119         if (copy_from_user(page, buf, count))
1120                 goto out_free_page;
1121
1122         page[count] = '\0';
1123         loginuid = simple_strtoul(page, &tmp, 10);
1124         if (tmp == page) {
1125                 length = -EINVAL;
1126                 goto out_free_page;
1127
1128         }
1129         length = audit_set_loginuid(loginuid);
1130         if (likely(length == 0))
1131                 length = count;
1132
1133 out_free_page:
1134         free_page((unsigned long) page);
1135         return length;
1136 }
1137
1138 static const struct file_operations proc_loginuid_operations = {
1139         .read           = proc_loginuid_read,
1140         .write          = proc_loginuid_write,
1141         .llseek         = generic_file_llseek,
1142 };
1143
1144 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145                                   size_t count, loff_t *ppos)
1146 {
1147         struct inode * inode = file->f_path.dentry->d_inode;
1148         struct task_struct *task = get_proc_task(inode);
1149         ssize_t length;
1150         char tmpbuf[TMPBUFLEN];
1151
1152         if (!task)
1153                 return -ESRCH;
1154         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155                                 audit_get_sessionid(task));
1156         put_task_struct(task);
1157         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158 }
1159
1160 static const struct file_operations proc_sessionid_operations = {
1161         .read           = proc_sessionid_read,
1162         .llseek         = generic_file_llseek,
1163 };
1164 #endif
1165
1166 #ifdef CONFIG_FAULT_INJECTION
1167 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168                                       size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171         char buffer[PROC_NUMBUF];
1172         size_t len;
1173         int make_it_fail;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         make_it_fail = task->make_it_fail;
1178         put_task_struct(task);
1179
1180         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183 }
1184
1185 static ssize_t proc_fault_inject_write(struct file * file,
1186                         const char __user * buf, size_t count, loff_t *ppos)
1187 {
1188         struct task_struct *task;
1189         char buffer[PROC_NUMBUF], *end;
1190         int make_it_fail;
1191
1192         if (!capable(CAP_SYS_RESOURCE))
1193                 return -EPERM;
1194         memset(buffer, 0, sizeof(buffer));
1195         if (count > sizeof(buffer) - 1)
1196                 count = sizeof(buffer) - 1;
1197         if (copy_from_user(buffer, buf, count))
1198                 return -EFAULT;
1199         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200         if (*end)
1201                 return -EINVAL;
1202         task = get_proc_task(file->f_dentry->d_inode);
1203         if (!task)
1204                 return -ESRCH;
1205         task->make_it_fail = make_it_fail;
1206         put_task_struct(task);
1207
1208         return count;
1209 }
1210
1211 static const struct file_operations proc_fault_inject_operations = {
1212         .read           = proc_fault_inject_read,
1213         .write          = proc_fault_inject_write,
1214         .llseek         = generic_file_llseek,
1215 };
1216 #endif
1217
1218
1219 #ifdef CONFIG_SCHED_DEBUG
1220 /*
1221  * Print out various scheduling related per-task fields:
1222  */
1223 static int sched_show(struct seq_file *m, void *v)
1224 {
1225         struct inode *inode = m->private;
1226         struct task_struct *p;
1227
1228         p = get_proc_task(inode);
1229         if (!p)
1230                 return -ESRCH;
1231         proc_sched_show_task(p, m);
1232
1233         put_task_struct(p);
1234
1235         return 0;
1236 }
1237
1238 static ssize_t
1239 sched_write(struct file *file, const char __user *buf,
1240             size_t count, loff_t *offset)
1241 {
1242         struct inode *inode = file->f_path.dentry->d_inode;
1243         struct task_struct *p;
1244
1245         p = get_proc_task(inode);
1246         if (!p)
1247                 return -ESRCH;
1248         proc_sched_set_task(p);
1249
1250         put_task_struct(p);
1251
1252         return count;
1253 }
1254
1255 static int sched_open(struct inode *inode, struct file *filp)
1256 {
1257         return single_open(filp, sched_show, inode);
1258 }
1259
1260 static const struct file_operations proc_pid_sched_operations = {
1261         .open           = sched_open,
1262         .read           = seq_read,
1263         .write          = sched_write,
1264         .llseek         = seq_lseek,
1265         .release        = single_release,
1266 };
1267
1268 #endif
1269
1270 #ifdef CONFIG_SCHED_AUTOGROUP
1271 /*
1272  * Print out autogroup related information:
1273  */
1274 static int sched_autogroup_show(struct seq_file *m, void *v)
1275 {
1276         struct inode *inode = m->private;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_autogroup_show_task(p, m);
1283
1284         put_task_struct(p);
1285
1286         return 0;
1287 }
1288
1289 static ssize_t
1290 sched_autogroup_write(struct file *file, const char __user *buf,
1291             size_t count, loff_t *offset)
1292 {
1293         struct inode *inode = file->f_path.dentry->d_inode;
1294         struct task_struct *p;
1295         char buffer[PROC_NUMBUF];
1296         int nice;
1297         int err;
1298
1299         memset(buffer, 0, sizeof(buffer));
1300         if (count > sizeof(buffer) - 1)
1301                 count = sizeof(buffer) - 1;
1302         if (copy_from_user(buffer, buf, count))
1303                 return -EFAULT;
1304
1305         err = kstrtoint(strstrip(buffer), 0, &nice);
1306         if (err < 0)
1307                 return err;
1308
1309         p = get_proc_task(inode);
1310         if (!p)
1311                 return -ESRCH;
1312
1313         err = proc_sched_autogroup_set_nice(p, nice);
1314         if (err)
1315                 count = err;
1316
1317         put_task_struct(p);
1318
1319         return count;
1320 }
1321
1322 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323 {
1324         int ret;
1325
1326         ret = single_open(filp, sched_autogroup_show, NULL);
1327         if (!ret) {
1328                 struct seq_file *m = filp->private_data;
1329
1330                 m->private = inode;
1331         }
1332         return ret;
1333 }
1334
1335 static const struct file_operations proc_pid_sched_autogroup_operations = {
1336         .open           = sched_autogroup_open,
1337         .read           = seq_read,
1338         .write          = sched_autogroup_write,
1339         .llseek         = seq_lseek,
1340         .release        = single_release,
1341 };
1342
1343 #endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345 static ssize_t comm_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[TASK_COMM_LEN];
1351
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361
1362         if (same_thread_group(current, p))
1363                 set_task_comm(p, buffer);
1364         else
1365                 count = -EINVAL;
1366
1367         put_task_struct(p);
1368
1369         return count;
1370 }
1371
1372 static int comm_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380
1381         task_lock(p);
1382         seq_printf(m, "%s\n", p->comm);
1383         task_unlock(p);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static int comm_open(struct inode *inode, struct file *filp)
1391 {
1392         return single_open(filp, comm_show, inode);
1393 }
1394
1395 static const struct file_operations proc_pid_set_comm_operations = {
1396         .open           = comm_open,
1397         .read           = seq_read,
1398         .write          = comm_write,
1399         .llseek         = seq_lseek,
1400         .release        = single_release,
1401 };
1402
1403 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404 {
1405         struct task_struct *task;
1406         struct mm_struct *mm;
1407         struct file *exe_file;
1408
1409         task = get_proc_task(dentry->d_inode);
1410         if (!task)
1411                 return -ENOENT;
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414         if (!mm)
1415                 return -ENOENT;
1416         exe_file = get_mm_exe_file(mm);
1417         mmput(mm);
1418         if (exe_file) {
1419                 *exe_path = exe_file->f_path;
1420                 path_get(&exe_file->f_path);
1421                 fput(exe_file);
1422                 return 0;
1423         } else
1424                 return -ENOENT;
1425 }
1426
1427 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428 {
1429         struct inode *inode = dentry->d_inode;
1430         struct path path;
1431         int error = -EACCES;
1432
1433         /* Are we allowed to snoop on the tasks file descriptors? */
1434         if (!proc_fd_access_allowed(inode))
1435                 goto out;
1436
1437         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1438         if (error)
1439                 goto out;
1440
1441         nd_jump_link(nd, &path);
1442         return NULL;
1443 out:
1444         return ERR_PTR(error);
1445 }
1446
1447 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1448 {
1449         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1450         char *pathname;
1451         int len;
1452
1453         if (!tmp)
1454                 return -ENOMEM;
1455
1456         pathname = d_path(path, tmp, PAGE_SIZE);
1457         len = PTR_ERR(pathname);
1458         if (IS_ERR(pathname))
1459                 goto out;
1460         len = tmp + PAGE_SIZE - 1 - pathname;
1461
1462         if (len > buflen)
1463                 len = buflen;
1464         if (copy_to_user(buffer, pathname, len))
1465                 len = -EFAULT;
1466  out:
1467         free_page((unsigned long)tmp);
1468         return len;
1469 }
1470
1471 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1472 {
1473         int error = -EACCES;
1474         struct inode *inode = dentry->d_inode;
1475         struct path path;
1476
1477         /* Are we allowed to snoop on the tasks file descriptors? */
1478         if (!proc_fd_access_allowed(inode))
1479                 goto out;
1480
1481         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1482         if (error)
1483                 goto out;
1484
1485         error = do_proc_readlink(&path, buffer, buflen);
1486         path_put(&path);
1487 out:
1488         return error;
1489 }
1490
1491 static const struct inode_operations proc_pid_link_inode_operations = {
1492         .readlink       = proc_pid_readlink,
1493         .follow_link    = proc_pid_follow_link,
1494         .setattr        = proc_setattr,
1495 };
1496
1497
1498 /* building an inode */
1499
1500 static int task_dumpable(struct task_struct *task)
1501 {
1502         int dumpable = 0;
1503         struct mm_struct *mm;
1504
1505         task_lock(task);
1506         mm = task->mm;
1507         if (mm)
1508                 dumpable = get_dumpable(mm);
1509         task_unlock(task);
1510         if(dumpable == 1)
1511                 return 1;
1512         return 0;
1513 }
1514
1515 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1516 {
1517         struct inode * inode;
1518         struct proc_inode *ei;
1519         const struct cred *cred;
1520
1521         /* We need a new inode */
1522
1523         inode = new_inode(sb);
1524         if (!inode)
1525                 goto out;
1526
1527         /* Common stuff */
1528         ei = PROC_I(inode);
1529         inode->i_ino = get_next_ino();
1530         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1531         inode->i_op = &proc_def_inode_operations;
1532
1533         /*
1534          * grab the reference to task.
1535          */
1536         ei->pid = get_task_pid(task, PIDTYPE_PID);
1537         if (!ei->pid)
1538                 goto out_unlock;
1539
1540         if (task_dumpable(task)) {
1541                 rcu_read_lock();
1542                 cred = __task_cred(task);
1543                 inode->i_uid = cred->euid;
1544                 inode->i_gid = cred->egid;
1545                 rcu_read_unlock();
1546         }
1547         security_task_to_inode(task, inode);
1548
1549 out:
1550         return inode;
1551
1552 out_unlock:
1553         iput(inode);
1554         return NULL;
1555 }
1556
1557 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1558 {
1559         struct inode *inode = dentry->d_inode;
1560         struct task_struct *task;
1561         const struct cred *cred;
1562         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1563
1564         generic_fillattr(inode, stat);
1565
1566         rcu_read_lock();
1567         stat->uid = GLOBAL_ROOT_UID;
1568         stat->gid = GLOBAL_ROOT_GID;
1569         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1570         if (task) {
1571                 if (!has_pid_permissions(pid, task, 2)) {
1572                         rcu_read_unlock();
1573                         /*
1574                          * This doesn't prevent learning whether PID exists,
1575                          * it only makes getattr() consistent with readdir().
1576                          */
1577                         return -ENOENT;
1578                 }
1579                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1580                     task_dumpable(task)) {
1581                         cred = __task_cred(task);
1582                         stat->uid = cred->euid;
1583                         stat->gid = cred->egid;
1584                 }
1585         }
1586         rcu_read_unlock();
1587         return 0;
1588 }
1589
1590 /* dentry stuff */
1591
1592 /*
1593  *      Exceptional case: normally we are not allowed to unhash a busy
1594  * directory. In this case, however, we can do it - no aliasing problems
1595  * due to the way we treat inodes.
1596  *
1597  * Rewrite the inode's ownerships here because the owning task may have
1598  * performed a setuid(), etc.
1599  *
1600  * Before the /proc/pid/status file was created the only way to read
1601  * the effective uid of a /process was to stat /proc/pid.  Reading
1602  * /proc/pid/status is slow enough that procps and other packages
1603  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1604  * made this apply to all per process world readable and executable
1605  * directories.
1606  */
1607 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1608 {
1609         struct inode *inode;
1610         struct task_struct *task;
1611         const struct cred *cred;
1612
1613         if (flags & LOOKUP_RCU)
1614                 return -ECHILD;
1615
1616         inode = dentry->d_inode;
1617         task = get_proc_task(inode);
1618
1619         if (task) {
1620                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1621                     task_dumpable(task)) {
1622                         rcu_read_lock();
1623                         cred = __task_cred(task);
1624                         inode->i_uid = cred->euid;
1625                         inode->i_gid = cred->egid;
1626                         rcu_read_unlock();
1627                 } else {
1628                         inode->i_uid = GLOBAL_ROOT_UID;
1629                         inode->i_gid = GLOBAL_ROOT_GID;
1630                 }
1631                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1632                 security_task_to_inode(task, inode);
1633                 put_task_struct(task);
1634                 return 1;
1635         }
1636         d_drop(dentry);
1637         return 0;
1638 }
1639
1640 static int pid_delete_dentry(const struct dentry * dentry)
1641 {
1642         /* Is the task we represent dead?
1643          * If so, then don't put the dentry on the lru list,
1644          * kill it immediately.
1645          */
1646         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1647 }
1648
1649 const struct dentry_operations pid_dentry_operations =
1650 {
1651         .d_revalidate   = pid_revalidate,
1652         .d_delete       = pid_delete_dentry,
1653 };
1654
1655 /* Lookups */
1656
1657 /*
1658  * Fill a directory entry.
1659  *
1660  * If possible create the dcache entry and derive our inode number and
1661  * file type from dcache entry.
1662  *
1663  * Since all of the proc inode numbers are dynamically generated, the inode
1664  * numbers do not exist until the inode is cache.  This means creating the
1665  * the dcache entry in readdir is necessary to keep the inode numbers
1666  * reported by readdir in sync with the inode numbers reported
1667  * by stat.
1668  */
1669 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1670         const char *name, int len,
1671         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1672 {
1673         struct dentry *child, *dir = filp->f_path.dentry;
1674         struct inode *inode;
1675         struct qstr qname;
1676         ino_t ino = 0;
1677         unsigned type = DT_UNKNOWN;
1678
1679         qname.name = name;
1680         qname.len  = len;
1681         qname.hash = full_name_hash(name, len);
1682
1683         child = d_lookup(dir, &qname);
1684         if (!child) {
1685                 struct dentry *new;
1686                 new = d_alloc(dir, &qname);
1687                 if (new) {
1688                         child = instantiate(dir->d_inode, new, task, ptr);
1689                         if (child)
1690                                 dput(new);
1691                         else
1692                                 child = new;
1693                 }
1694         }
1695         if (!child || IS_ERR(child) || !child->d_inode)
1696                 goto end_instantiate;
1697         inode = child->d_inode;
1698         if (inode) {
1699                 ino = inode->i_ino;
1700                 type = inode->i_mode >> 12;
1701         }
1702         dput(child);
1703 end_instantiate:
1704         if (!ino)
1705                 ino = find_inode_number(dir, &qname);
1706         if (!ino)
1707                 ino = 1;
1708         return filldir(dirent, name, len, filp->f_pos, ino, type);
1709 }
1710
1711 static unsigned name_to_int(struct dentry *dentry)
1712 {
1713         const char *name = dentry->d_name.name;
1714         int len = dentry->d_name.len;
1715         unsigned n = 0;
1716
1717         if (len > 1 && *name == '0')
1718                 goto out;
1719         while (len-- > 0) {
1720                 unsigned c = *name++ - '0';
1721                 if (c > 9)
1722                         goto out;
1723                 if (n >= (~0U-9)/10)
1724                         goto out;
1725                 n *= 10;
1726                 n += c;
1727         }
1728         return n;
1729 out:
1730         return ~0U;
1731 }
1732
1733 #define PROC_FDINFO_MAX 64
1734
1735 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1736 {
1737         struct task_struct *task = get_proc_task(inode);
1738         struct files_struct *files = NULL;
1739         struct file *file;
1740         int fd = proc_fd(inode);
1741
1742         if (task) {
1743                 files = get_files_struct(task);
1744                 put_task_struct(task);
1745         }
1746         if (files) {
1747                 /*
1748                  * We are not taking a ref to the file structure, so we must
1749                  * hold ->file_lock.
1750                  */
1751                 spin_lock(&files->file_lock);
1752                 file = fcheck_files(files, fd);
1753                 if (file) {
1754                         unsigned int f_flags;
1755                         struct fdtable *fdt;
1756
1757                         fdt = files_fdtable(files);
1758                         f_flags = file->f_flags & ~O_CLOEXEC;
1759                         if (close_on_exec(fd, fdt))
1760                                 f_flags |= O_CLOEXEC;
1761
1762                         if (path) {
1763                                 *path = file->f_path;
1764                                 path_get(&file->f_path);
1765                         }
1766                         if (info)
1767                                 snprintf(info, PROC_FDINFO_MAX,
1768                                          "pos:\t%lli\n"
1769                                          "flags:\t0%o\n",
1770                                          (long long) file->f_pos,
1771                                          f_flags);
1772                         spin_unlock(&files->file_lock);
1773                         put_files_struct(files);
1774                         return 0;
1775                 }
1776                 spin_unlock(&files->file_lock);
1777                 put_files_struct(files);
1778         }
1779         return -ENOENT;
1780 }
1781
1782 static int proc_fd_link(struct dentry *dentry, struct path *path)
1783 {
1784         return proc_fd_info(dentry->d_inode, path, NULL);
1785 }
1786
1787 static int tid_fd_revalidate(struct dentry *dentry, unsigned int flags)
1788 {
1789         struct inode *inode;
1790         struct task_struct *task;
1791         int fd;
1792         struct files_struct *files;
1793         const struct cred *cred;
1794
1795         if (flags & LOOKUP_RCU)
1796                 return -ECHILD;
1797
1798         inode = dentry->d_inode;
1799         task = get_proc_task(inode);
1800         fd = proc_fd(inode);
1801
1802         if (task) {
1803                 files = get_files_struct(task);
1804                 if (files) {
1805                         struct file *file;
1806                         rcu_read_lock();
1807                         file = fcheck_files(files, fd);
1808                         if (file) {
1809                                 unsigned f_mode = file->f_mode;
1810
1811                                 rcu_read_unlock();
1812                                 put_files_struct(files);
1813
1814                                 if (task_dumpable(task)) {
1815                                         rcu_read_lock();
1816                                         cred = __task_cred(task);
1817                                         inode->i_uid = cred->euid;
1818                                         inode->i_gid = cred->egid;
1819                                         rcu_read_unlock();
1820                                 } else {
1821                                         inode->i_uid = GLOBAL_ROOT_UID;
1822                                         inode->i_gid = GLOBAL_ROOT_GID;
1823                                 }
1824
1825                                 if (S_ISLNK(inode->i_mode)) {
1826                                         unsigned i_mode = S_IFLNK;
1827                                         if (f_mode & FMODE_READ)
1828                                                 i_mode |= S_IRUSR | S_IXUSR;
1829                                         if (f_mode & FMODE_WRITE)
1830                                                 i_mode |= S_IWUSR | S_IXUSR;
1831                                         inode->i_mode = i_mode;
1832                                 }
1833
1834                                 security_task_to_inode(task, inode);
1835                                 put_task_struct(task);
1836                                 return 1;
1837                         }
1838                         rcu_read_unlock();
1839                         put_files_struct(files);
1840                 }
1841                 put_task_struct(task);
1842         }
1843         d_drop(dentry);
1844         return 0;
1845 }
1846
1847 static const struct dentry_operations tid_fd_dentry_operations =
1848 {
1849         .d_revalidate   = tid_fd_revalidate,
1850         .d_delete       = pid_delete_dentry,
1851 };
1852
1853 static struct dentry *proc_fd_instantiate(struct inode *dir,
1854         struct dentry *dentry, struct task_struct *task, const void *ptr)
1855 {
1856         unsigned fd = (unsigned long)ptr;
1857         struct inode *inode;
1858         struct proc_inode *ei;
1859         struct dentry *error = ERR_PTR(-ENOENT);
1860
1861         inode = proc_pid_make_inode(dir->i_sb, task);
1862         if (!inode)
1863                 goto out;
1864         ei = PROC_I(inode);
1865         ei->fd = fd;
1866
1867         inode->i_mode = S_IFLNK;
1868         inode->i_op = &proc_pid_link_inode_operations;
1869         inode->i_size = 64;
1870         ei->op.proc_get_link = proc_fd_link;
1871         d_set_d_op(dentry, &tid_fd_dentry_operations);
1872         d_add(dentry, inode);
1873         /* Close the race of the process dying before we return the dentry */
1874         if (tid_fd_revalidate(dentry, 0))
1875                 error = NULL;
1876
1877  out:
1878         return error;
1879 }
1880
1881 static struct dentry *proc_lookupfd_common(struct inode *dir,
1882                                            struct dentry *dentry,
1883                                            instantiate_t instantiate)
1884 {
1885         struct task_struct *task = get_proc_task(dir);
1886         unsigned fd = name_to_int(dentry);
1887         struct dentry *result = ERR_PTR(-ENOENT);
1888
1889         if (!task)
1890                 goto out_no_task;
1891         if (fd == ~0U)
1892                 goto out;
1893
1894         result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1895 out:
1896         put_task_struct(task);
1897 out_no_task:
1898         return result;
1899 }
1900
1901 static int proc_readfd_common(struct file * filp, void * dirent,
1902                               filldir_t filldir, instantiate_t instantiate)
1903 {
1904         struct dentry *dentry = filp->f_path.dentry;
1905         struct inode *inode = dentry->d_inode;
1906         struct task_struct *p = get_proc_task(inode);
1907         unsigned int fd, ino;
1908         int retval;
1909         struct files_struct * files;
1910
1911         retval = -ENOENT;
1912         if (!p)
1913                 goto out_no_task;
1914         retval = 0;
1915
1916         fd = filp->f_pos;
1917         switch (fd) {
1918                 case 0:
1919                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1920                                 goto out;
1921                         filp->f_pos++;
1922                 case 1:
1923                         ino = parent_ino(dentry);
1924                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1925                                 goto out;
1926                         filp->f_pos++;
1927                 default:
1928                         files = get_files_struct(p);
1929                         if (!files)
1930                                 goto out;
1931                         rcu_read_lock();
1932                         for (fd = filp->f_pos-2;
1933                              fd < files_fdtable(files)->max_fds;
1934                              fd++, filp->f_pos++) {
1935                                 char name[PROC_NUMBUF];
1936                                 int len;
1937                                 int rv;
1938
1939                                 if (!fcheck_files(files, fd))
1940                                         continue;
1941                                 rcu_read_unlock();
1942
1943                                 len = snprintf(name, sizeof(name), "%d", fd);
1944                                 rv = proc_fill_cache(filp, dirent, filldir,
1945                                                      name, len, instantiate, p,
1946                                                      (void *)(unsigned long)fd);
1947                                 if (rv < 0)
1948                                         goto out_fd_loop;
1949                                 rcu_read_lock();
1950                         }
1951                         rcu_read_unlock();
1952 out_fd_loop:
1953                         put_files_struct(files);
1954         }
1955 out:
1956         put_task_struct(p);
1957 out_no_task:
1958         return retval;
1959 }
1960
1961 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1962                                     unsigned int flags)
1963 {
1964         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1965 }
1966
1967 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1968 {
1969         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1970 }
1971
1972 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1973                                       size_t len, loff_t *ppos)
1974 {
1975         char tmp[PROC_FDINFO_MAX];
1976         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1977         if (!err)
1978                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1979         return err;
1980 }
1981
1982 static const struct file_operations proc_fdinfo_file_operations = {
1983         .open           = nonseekable_open,
1984         .read           = proc_fdinfo_read,
1985         .llseek         = no_llseek,
1986 };
1987
1988 static const struct file_operations proc_fd_operations = {
1989         .read           = generic_read_dir,
1990         .readdir        = proc_readfd,
1991         .llseek         = default_llseek,
1992 };
1993
1994 #ifdef CONFIG_CHECKPOINT_RESTORE
1995
1996 /*
1997  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1998  * which represent vma start and end addresses.
1999  */
2000 static int dname_to_vma_addr(struct dentry *dentry,
2001                              unsigned long *start, unsigned long *end)
2002 {
2003         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2004                 return -EINVAL;
2005
2006         return 0;
2007 }
2008
2009 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2010 {
2011         unsigned long vm_start, vm_end;
2012         bool exact_vma_exists = false;
2013         struct mm_struct *mm = NULL;
2014         struct task_struct *task;
2015         const struct cred *cred;
2016         struct inode *inode;
2017         int status = 0;
2018
2019         if (flags & LOOKUP_RCU)
2020                 return -ECHILD;
2021
2022         if (!capable(CAP_SYS_ADMIN)) {
2023                 status = -EACCES;
2024                 goto out_notask;
2025         }
2026
2027         inode = dentry->d_inode;
2028         task = get_proc_task(inode);
2029         if (!task)
2030                 goto out_notask;
2031
2032         mm = mm_access(task, PTRACE_MODE_READ);
2033         if (IS_ERR_OR_NULL(mm))
2034                 goto out;
2035
2036         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2037                 down_read(&mm->mmap_sem);
2038                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2039                 up_read(&mm->mmap_sem);
2040         }
2041
2042         mmput(mm);
2043
2044         if (exact_vma_exists) {
2045                 if (task_dumpable(task)) {
2046                         rcu_read_lock();
2047                         cred = __task_cred(task);
2048                         inode->i_uid = cred->euid;
2049                         inode->i_gid = cred->egid;
2050                         rcu_read_unlock();
2051                 } else {
2052                         inode->i_uid = GLOBAL_ROOT_UID;
2053                         inode->i_gid = GLOBAL_ROOT_GID;
2054                 }
2055                 security_task_to_inode(task, inode);
2056                 status = 1;
2057         }
2058
2059 out:
2060         put_task_struct(task);
2061
2062 out_notask:
2063         if (status <= 0)
2064                 d_drop(dentry);
2065
2066         return status;
2067 }
2068
2069 static const struct dentry_operations tid_map_files_dentry_operations = {
2070         .d_revalidate   = map_files_d_revalidate,
2071         .d_delete       = pid_delete_dentry,
2072 };
2073
2074 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2075 {
2076         unsigned long vm_start, vm_end;
2077         struct vm_area_struct *vma;
2078         struct task_struct *task;
2079         struct mm_struct *mm;
2080         int rc;
2081
2082         rc = -ENOENT;
2083         task = get_proc_task(dentry->d_inode);
2084         if (!task)
2085                 goto out;
2086
2087         mm = get_task_mm(task);
2088         put_task_struct(task);
2089         if (!mm)
2090                 goto out;
2091
2092         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2093         if (rc)
2094                 goto out_mmput;
2095
2096         down_read(&mm->mmap_sem);
2097         vma = find_exact_vma(mm, vm_start, vm_end);
2098         if (vma && vma->vm_file) {
2099                 *path = vma->vm_file->f_path;
2100                 path_get(path);
2101                 rc = 0;
2102         }
2103         up_read(&mm->mmap_sem);
2104
2105 out_mmput:
2106         mmput(mm);
2107 out:
2108         return rc;
2109 }
2110
2111 struct map_files_info {
2112         struct file     *file;
2113         unsigned long   len;
2114         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2115 };
2116
2117 static struct dentry *
2118 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2119                            struct task_struct *task, const void *ptr)
2120 {
2121         const struct file *file = ptr;
2122         struct proc_inode *ei;
2123         struct inode *inode;
2124
2125         if (!file)
2126                 return ERR_PTR(-ENOENT);
2127
2128         inode = proc_pid_make_inode(dir->i_sb, task);
2129         if (!inode)
2130                 return ERR_PTR(-ENOENT);
2131
2132         ei = PROC_I(inode);
2133         ei->op.proc_get_link = proc_map_files_get_link;
2134
2135         inode->i_op = &proc_pid_link_inode_operations;
2136         inode->i_size = 64;
2137         inode->i_mode = S_IFLNK;
2138
2139         if (file->f_mode & FMODE_READ)
2140                 inode->i_mode |= S_IRUSR;
2141         if (file->f_mode & FMODE_WRITE)
2142                 inode->i_mode |= S_IWUSR;
2143
2144         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2145         d_add(dentry, inode);
2146
2147         return NULL;
2148 }
2149
2150 static struct dentry *proc_map_files_lookup(struct inode *dir,
2151                 struct dentry *dentry, unsigned int flags)
2152 {
2153         unsigned long vm_start, vm_end;
2154         struct vm_area_struct *vma;
2155         struct task_struct *task;
2156         struct dentry *result;
2157         struct mm_struct *mm;
2158
2159         result = ERR_PTR(-EACCES);
2160         if (!capable(CAP_SYS_ADMIN))
2161                 goto out;
2162
2163         result = ERR_PTR(-ENOENT);
2164         task = get_proc_task(dir);
2165         if (!task)
2166                 goto out;
2167
2168         result = ERR_PTR(-EACCES);
2169         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2170                 goto out_put_task;
2171
2172         result = ERR_PTR(-ENOENT);
2173         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2174                 goto out_put_task;
2175
2176         mm = get_task_mm(task);
2177         if (!mm)
2178                 goto out_put_task;
2179
2180         down_read(&mm->mmap_sem);
2181         vma = find_exact_vma(mm, vm_start, vm_end);
2182         if (!vma)
2183                 goto out_no_vma;
2184
2185         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2186
2187 out_no_vma:
2188         up_read(&mm->mmap_sem);
2189         mmput(mm);
2190 out_put_task:
2191         put_task_struct(task);
2192 out:
2193         return result;
2194 }
2195
2196 static const struct inode_operations proc_map_files_inode_operations = {
2197         .lookup         = proc_map_files_lookup,
2198         .permission     = proc_fd_permission,
2199         .setattr        = proc_setattr,
2200 };
2201
2202 static int
2203 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2204 {
2205         struct dentry *dentry = filp->f_path.dentry;
2206         struct inode *inode = dentry->d_inode;
2207         struct vm_area_struct *vma;
2208         struct task_struct *task;
2209         struct mm_struct *mm;
2210         ino_t ino;
2211         int ret;
2212
2213         ret = -EACCES;
2214         if (!capable(CAP_SYS_ADMIN))
2215                 goto out;
2216
2217         ret = -ENOENT;
2218         task = get_proc_task(inode);
2219         if (!task)
2220                 goto out;
2221
2222         ret = -EACCES;
2223         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2224                 goto out_put_task;
2225
2226         ret = 0;
2227         switch (filp->f_pos) {
2228         case 0:
2229                 ino = inode->i_ino;
2230                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2231                         goto out_put_task;
2232                 filp->f_pos++;
2233         case 1:
2234                 ino = parent_ino(dentry);
2235                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2236                         goto out_put_task;
2237                 filp->f_pos++;
2238         default:
2239         {
2240                 unsigned long nr_files, pos, i;
2241                 struct flex_array *fa = NULL;
2242                 struct map_files_info info;
2243                 struct map_files_info *p;
2244
2245                 mm = get_task_mm(task);
2246                 if (!mm)
2247                         goto out_put_task;
2248                 down_read(&mm->mmap_sem);
2249
2250                 nr_files = 0;
2251
2252                 /*
2253                  * We need two passes here:
2254                  *
2255                  *  1) Collect vmas of mapped files with mmap_sem taken
2256                  *  2) Release mmap_sem and instantiate entries
2257                  *
2258                  * otherwise we get lockdep complained, since filldir()
2259                  * routine might require mmap_sem taken in might_fault().
2260                  */
2261
2262                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2263                         if (vma->vm_file && ++pos > filp->f_pos)
2264                                 nr_files++;
2265                 }
2266
2267                 if (nr_files) {
2268                         fa = flex_array_alloc(sizeof(info), nr_files,
2269                                                 GFP_KERNEL);
2270                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2271                                                         GFP_KERNEL)) {
2272                                 ret = -ENOMEM;
2273                                 if (fa)
2274                                         flex_array_free(fa);
2275                                 up_read(&mm->mmap_sem);
2276                                 mmput(mm);
2277                                 goto out_put_task;
2278                         }
2279                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2280                                         vma = vma->vm_next) {
2281                                 if (!vma->vm_file)
2282                                         continue;
2283                                 if (++pos <= filp->f_pos)
2284                                         continue;
2285
2286                                 get_file(vma->vm_file);
2287                                 info.file = vma->vm_file;
2288                                 info.len = snprintf(info.name,
2289                                                 sizeof(info.name), "%lx-%lx",
2290                                                 vma->vm_start, vma->vm_end);
2291                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2292                                         BUG();
2293                         }
2294                 }
2295                 up_read(&mm->mmap_sem);
2296
2297                 for (i = 0; i < nr_files; i++) {
2298                         p = flex_array_get(fa, i);
2299                         ret = proc_fill_cache(filp, dirent, filldir,
2300                                               p->name, p->len,
2301                                               proc_map_files_instantiate,
2302                                               task, p->file);
2303                         if (ret)
2304                                 break;
2305                         filp->f_pos++;
2306                         fput(p->file);
2307                 }
2308                 for (; i < nr_files; i++) {
2309                         /*
2310                          * In case of error don't forget
2311                          * to put rest of file refs.
2312                          */
2313                         p = flex_array_get(fa, i);
2314                         fput(p->file);
2315                 }
2316                 if (fa)
2317                         flex_array_free(fa);
2318                 mmput(mm);
2319         }
2320         }
2321
2322 out_put_task:
2323         put_task_struct(task);
2324 out:
2325         return ret;
2326 }
2327
2328 static const struct file_operations proc_map_files_operations = {
2329         .read           = generic_read_dir,
2330         .readdir        = proc_map_files_readdir,
2331         .llseek         = default_llseek,
2332 };
2333
2334 #endif /* CONFIG_CHECKPOINT_RESTORE */
2335
2336 /*
2337  * /proc/pid/fd needs a special permission handler so that a process can still
2338  * access /proc/self/fd after it has executed a setuid().
2339  */
2340 static int proc_fd_permission(struct inode *inode, int mask)
2341 {
2342         int rv = generic_permission(inode, mask);
2343         if (rv == 0)
2344                 return 0;
2345         if (task_pid(current) == proc_pid(inode))
2346                 rv = 0;
2347         return rv;
2348 }
2349
2350 /*
2351  * proc directories can do almost nothing..
2352  */
2353 static const struct inode_operations proc_fd_inode_operations = {
2354         .lookup         = proc_lookupfd,
2355         .permission     = proc_fd_permission,
2356         .setattr        = proc_setattr,
2357 };
2358
2359 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2360         struct dentry *dentry, struct task_struct *task, const void *ptr)
2361 {
2362         unsigned fd = (unsigned long)ptr;
2363         struct inode *inode;
2364         struct proc_inode *ei;
2365         struct dentry *error = ERR_PTR(-ENOENT);
2366
2367         inode = proc_pid_make_inode(dir->i_sb, task);
2368         if (!inode)
2369                 goto out;
2370         ei = PROC_I(inode);
2371         ei->fd = fd;
2372         inode->i_mode = S_IFREG | S_IRUSR;
2373         inode->i_fop = &proc_fdinfo_file_operations;
2374         d_set_d_op(dentry, &tid_fd_dentry_operations);
2375         d_add(dentry, inode);
2376         /* Close the race of the process dying before we return the dentry */
2377         if (tid_fd_revalidate(dentry, 0))
2378                 error = NULL;
2379
2380  out:
2381         return error;
2382 }
2383
2384 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2385                                         struct dentry *dentry,
2386                                         unsigned int flags)
2387 {
2388         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2389 }
2390
2391 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2392 {
2393         return proc_readfd_common(filp, dirent, filldir,
2394                                   proc_fdinfo_instantiate);
2395 }
2396
2397 static const struct file_operations proc_fdinfo_operations = {
2398         .read           = generic_read_dir,
2399         .readdir        = proc_readfdinfo,
2400         .llseek         = default_llseek,
2401 };
2402
2403 /*
2404  * proc directories can do almost nothing..
2405  */
2406 static const struct inode_operations proc_fdinfo_inode_operations = {
2407         .lookup         = proc_lookupfdinfo,
2408         .setattr        = proc_setattr,
2409 };
2410
2411
2412 static struct dentry *proc_pident_instantiate(struct inode *dir,
2413         struct dentry *dentry, struct task_struct *task, const void *ptr)
2414 {
2415         const struct pid_entry *p = ptr;
2416         struct inode *inode;
2417         struct proc_inode *ei;
2418         struct dentry *error = ERR_PTR(-ENOENT);
2419
2420         inode = proc_pid_make_inode(dir->i_sb, task);
2421         if (!inode)
2422                 goto out;
2423
2424         ei = PROC_I(inode);
2425         inode->i_mode = p->mode;
2426         if (S_ISDIR(inode->i_mode))
2427                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2428         if (p->iop)
2429                 inode->i_op = p->iop;
2430         if (p->fop)
2431                 inode->i_fop = p->fop;
2432         ei->op = p->op;
2433         d_set_d_op(dentry, &pid_dentry_operations);
2434         d_add(dentry, inode);
2435         /* Close the race of the process dying before we return the dentry */
2436         if (pid_revalidate(dentry, 0))
2437                 error = NULL;
2438 out:
2439         return error;
2440 }
2441
2442 static struct dentry *proc_pident_lookup(struct inode *dir,
2443                                          struct dentry *dentry,
2444                                          const struct pid_entry *ents,
2445                                          unsigned int nents)
2446 {
2447         struct dentry *error;
2448         struct task_struct *task = get_proc_task(dir);
2449         const struct pid_entry *p, *last;
2450
2451         error = ERR_PTR(-ENOENT);
2452
2453         if (!task)
2454                 goto out_no_task;
2455
2456         /*
2457          * Yes, it does not scale. And it should not. Don't add
2458          * new entries into /proc/<tgid>/ without very good reasons.
2459          */
2460         last = &ents[nents - 1];
2461         for (p = ents; p <= last; p++) {
2462                 if (p->len != dentry->d_name.len)
2463                         continue;
2464                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2465                         break;
2466         }
2467         if (p > last)
2468                 goto out;
2469
2470         error = proc_pident_instantiate(dir, dentry, task, p);
2471 out:
2472         put_task_struct(task);
2473 out_no_task:
2474         return error;
2475 }
2476
2477 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2478         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2479 {
2480         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2481                                 proc_pident_instantiate, task, p);
2482 }
2483
2484 static int proc_pident_readdir(struct file *filp,
2485                 void *dirent, filldir_t filldir,
2486                 const struct pid_entry *ents, unsigned int nents)
2487 {
2488         int i;
2489         struct dentry *dentry = filp->f_path.dentry;
2490         struct inode *inode = dentry->d_inode;
2491         struct task_struct *task = get_proc_task(inode);
2492         const struct pid_entry *p, *last;
2493         ino_t ino;
2494         int ret;
2495
2496         ret = -ENOENT;
2497         if (!task)
2498                 goto out_no_task;
2499
2500         ret = 0;
2501         i = filp->f_pos;
2502         switch (i) {
2503         case 0:
2504                 ino = inode->i_ino;
2505                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2506                         goto out;
2507                 i++;
2508                 filp->f_pos++;
2509                 /* fall through */
2510         case 1:
2511                 ino = parent_ino(dentry);
2512                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2513                         goto out;
2514                 i++;
2515                 filp->f_pos++;
2516                 /* fall through */
2517         default:
2518                 i -= 2;
2519                 if (i >= nents) {
2520                         ret = 1;
2521                         goto out;
2522                 }
2523                 p = ents + i;
2524                 last = &ents[nents - 1];
2525                 while (p <= last) {
2526                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2527                                 goto out;
2528                         filp->f_pos++;
2529                         p++;
2530                 }
2531         }
2532
2533         ret = 1;
2534 out:
2535         put_task_struct(task);
2536 out_no_task:
2537         return ret;
2538 }
2539
2540 #ifdef CONFIG_SECURITY
2541 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2542                                   size_t count, loff_t *ppos)
2543 {
2544         struct inode * inode = file->f_path.dentry->d_inode;
2545         char *p = NULL;
2546         ssize_t length;
2547         struct task_struct *task = get_proc_task(inode);
2548
2549         if (!task)
2550                 return -ESRCH;
2551
2552         length = security_getprocattr(task,
2553                                       (char*)file->f_path.dentry->d_name.name,
2554                                       &p);
2555         put_task_struct(task);
2556         if (length > 0)
2557                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2558         kfree(p);
2559         return length;
2560 }
2561
2562 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2563                                    size_t count, loff_t *ppos)
2564 {
2565         struct inode * inode = file->f_path.dentry->d_inode;
2566         char *page;
2567         ssize_t length;
2568         struct task_struct *task = get_proc_task(inode);
2569
2570         length = -ESRCH;
2571         if (!task)
2572                 goto out_no_task;
2573         if (count > PAGE_SIZE)
2574                 count = PAGE_SIZE;
2575
2576         /* No partial writes. */
2577         length = -EINVAL;
2578         if (*ppos != 0)
2579                 goto out;
2580
2581         length = -ENOMEM;
2582         page = (char*)__get_free_page(GFP_TEMPORARY);
2583         if (!page)
2584                 goto out;
2585
2586         length = -EFAULT;
2587         if (copy_from_user(page, buf, count))
2588                 goto out_free;
2589
2590         /* Guard against adverse ptrace interaction */
2591         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2592         if (length < 0)
2593                 goto out_free;
2594
2595         length = security_setprocattr(task,
2596                                       (char*)file->f_path.dentry->d_name.name,
2597                                       (void*)page, count);
2598         mutex_unlock(&task->signal->cred_guard_mutex);
2599 out_free:
2600         free_page((unsigned long) page);
2601 out:
2602         put_task_struct(task);
2603 out_no_task:
2604         return length;
2605 }
2606
2607 static const struct file_operations proc_pid_attr_operations = {
2608         .read           = proc_pid_attr_read,
2609         .write          = proc_pid_attr_write,
2610         .llseek         = generic_file_llseek,
2611 };
2612
2613 static const struct pid_entry attr_dir_stuff[] = {
2614         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2616         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2617         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2618         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2619         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2620 };
2621
2622 static int proc_attr_dir_readdir(struct file * filp,
2623                              void * dirent, filldir_t filldir)
2624 {
2625         return proc_pident_readdir(filp,dirent,filldir,
2626                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2627 }
2628
2629 static const struct file_operations proc_attr_dir_operations = {
2630         .read           = generic_read_dir,
2631         .readdir        = proc_attr_dir_readdir,
2632         .llseek         = default_llseek,
2633 };
2634
2635 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2636                                 struct dentry *dentry, unsigned int flags)
2637 {
2638         return proc_pident_lookup(dir, dentry,
2639                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2640 }
2641
2642 static const struct inode_operations proc_attr_dir_inode_operations = {
2643         .lookup         = proc_attr_dir_lookup,
2644         .getattr        = pid_getattr,
2645         .setattr        = proc_setattr,
2646 };
2647
2648 #endif
2649
2650 #ifdef CONFIG_ELF_CORE
2651 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2652                                          size_t count, loff_t *ppos)
2653 {
2654         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2655         struct mm_struct *mm;
2656         char buffer[PROC_NUMBUF];
2657         size_t len;
2658         int ret;
2659
2660         if (!task)
2661                 return -ESRCH;
2662
2663         ret = 0;
2664         mm = get_task_mm(task);
2665         if (mm) {
2666                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2667                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2668                                 MMF_DUMP_FILTER_SHIFT));
2669                 mmput(mm);
2670                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2671         }
2672
2673         put_task_struct(task);
2674
2675         return ret;
2676 }
2677
2678 static ssize_t proc_coredump_filter_write(struct file *file,
2679                                           const char __user *buf,
2680                                           size_t count,
2681                                           loff_t *ppos)
2682 {
2683         struct task_struct *task;
2684         struct mm_struct *mm;
2685         char buffer[PROC_NUMBUF], *end;
2686         unsigned int val;
2687         int ret;
2688         int i;
2689         unsigned long mask;
2690
2691         ret = -EFAULT;
2692         memset(buffer, 0, sizeof(buffer));
2693         if (count > sizeof(buffer) - 1)
2694                 count = sizeof(buffer) - 1;
2695         if (copy_from_user(buffer, buf, count))
2696                 goto out_no_task;
2697
2698         ret = -EINVAL;
2699         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2700         if (*end == '\n')
2701                 end++;
2702         if (end - buffer == 0)
2703                 goto out_no_task;
2704
2705         ret = -ESRCH;
2706         task = get_proc_task(file->f_dentry->d_inode);
2707         if (!task)
2708                 goto out_no_task;
2709
2710         ret = end - buffer;
2711         mm = get_task_mm(task);
2712         if (!mm)
2713                 goto out_no_mm;
2714
2715         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2716                 if (val & mask)
2717                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2718                 else
2719                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2720         }
2721
2722         mmput(mm);
2723  out_no_mm:
2724         put_task_struct(task);
2725  out_no_task:
2726         return ret;
2727 }
2728
2729 static const struct file_operations proc_coredump_filter_operations = {
2730         .read           = proc_coredump_filter_read,
2731         .write          = proc_coredump_filter_write,
2732         .llseek         = generic_file_llseek,
2733 };
2734 #endif
2735
2736 /*
2737  * /proc/self:
2738  */
2739 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2740                               int buflen)
2741 {
2742         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2743         pid_t tgid = task_tgid_nr_ns(current, ns);
2744         char tmp[PROC_NUMBUF];
2745         if (!tgid)
2746                 return -ENOENT;
2747         sprintf(tmp, "%d", tgid);
2748         return vfs_readlink(dentry,buffer,buflen,tmp);
2749 }
2750
2751 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2752 {
2753         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2754         pid_t tgid = task_tgid_nr_ns(current, ns);
2755         char *name = ERR_PTR(-ENOENT);
2756         if (tgid) {
2757                 name = __getname();
2758                 if (!name)
2759                         name = ERR_PTR(-ENOMEM);
2760                 else
2761                         sprintf(name, "%d", tgid);
2762         }
2763         nd_set_link(nd, name);
2764         return NULL;
2765 }
2766
2767 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2768                                 void *cookie)
2769 {
2770         char *s = nd_get_link(nd);
2771         if (!IS_ERR(s))
2772                 __putname(s);
2773 }
2774
2775 static const struct inode_operations proc_self_inode_operations = {
2776         .readlink       = proc_self_readlink,
2777         .follow_link    = proc_self_follow_link,
2778         .put_link       = proc_self_put_link,
2779 };
2780
2781 /*
2782  * proc base
2783  *
2784  * These are the directory entries in the root directory of /proc
2785  * that properly belong to the /proc filesystem, as they describe
2786  * describe something that is process related.
2787  */
2788 static const struct pid_entry proc_base_stuff[] = {
2789         NOD("self", S_IFLNK|S_IRWXUGO,
2790                 &proc_self_inode_operations, NULL, {}),
2791 };
2792
2793 static struct dentry *proc_base_instantiate(struct inode *dir,
2794         struct dentry *dentry, struct task_struct *task, const void *ptr)
2795 {
2796         const struct pid_entry *p = ptr;
2797         struct inode *inode;
2798         struct proc_inode *ei;
2799         struct dentry *error;
2800
2801         /* Allocate the inode */
2802         error = ERR_PTR(-ENOMEM);
2803         inode = new_inode(dir->i_sb);
2804         if (!inode)
2805                 goto out;
2806
2807         /* Initialize the inode */
2808         ei = PROC_I(inode);
2809         inode->i_ino = get_next_ino();
2810         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2811
2812         /*
2813          * grab the reference to the task.
2814          */
2815         ei->pid = get_task_pid(task, PIDTYPE_PID);
2816         if (!ei->pid)
2817                 goto out_iput;
2818
2819         inode->i_mode = p->mode;
2820         if (S_ISDIR(inode->i_mode))
2821                 set_nlink(inode, 2);
2822         if (S_ISLNK(inode->i_mode))
2823                 inode->i_size = 64;
2824         if (p->iop)
2825                 inode->i_op = p->iop;
2826         if (p->fop)
2827                 inode->i_fop = p->fop;
2828         ei->op = p->op;
2829         d_add(dentry, inode);
2830         error = NULL;
2831 out:
2832         return error;
2833 out_iput:
2834         iput(inode);
2835         goto out;
2836 }
2837
2838 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2839 {
2840         struct dentry *error;
2841         struct task_struct *task = get_proc_task(dir);
2842         const struct pid_entry *p, *last;
2843
2844         error = ERR_PTR(-ENOENT);
2845
2846         if (!task)
2847                 goto out_no_task;
2848
2849         /* Lookup the directory entry */
2850         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2851         for (p = proc_base_stuff; p <= last; p++) {
2852                 if (p->len != dentry->d_name.len)
2853                         continue;
2854                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2855                         break;
2856         }
2857         if (p > last)
2858                 goto out;
2859
2860         error = proc_base_instantiate(dir, dentry, task, p);
2861
2862 out:
2863         put_task_struct(task);
2864 out_no_task:
2865         return error;
2866 }
2867
2868 static int proc_base_fill_cache(struct file *filp, void *dirent,
2869         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2870 {
2871         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2872                                 proc_base_instantiate, task, p);
2873 }
2874
2875 #ifdef CONFIG_TASK_IO_ACCOUNTING
2876 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2877 {
2878         struct task_io_accounting acct = task->ioac;
2879         unsigned long flags;
2880         int result;
2881
2882         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2883         if (result)
2884                 return result;
2885
2886         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2887                 result = -EACCES;
2888                 goto out_unlock;
2889         }
2890
2891         if (whole && lock_task_sighand(task, &flags)) {
2892                 struct task_struct *t = task;
2893
2894                 task_io_accounting_add(&acct, &task->signal->ioac);
2895                 while_each_thread(task, t)
2896                         task_io_accounting_add(&acct, &t->ioac);
2897
2898                 unlock_task_sighand(task, &flags);
2899         }
2900         result = sprintf(buffer,
2901                         "rchar: %llu\n"
2902                         "wchar: %llu\n"
2903                         "syscr: %llu\n"
2904                         "syscw: %llu\n"
2905                         "read_bytes: %llu\n"
2906                         "write_bytes: %llu\n"
2907                         "cancelled_write_bytes: %llu\n",
2908                         (unsigned long long)acct.rchar,
2909                         (unsigned long long)acct.wchar,
2910                         (unsigned long long)acct.syscr,
2911                         (unsigned long long)acct.syscw,
2912                         (unsigned long long)acct.read_bytes,
2913                         (unsigned long long)acct.write_bytes,
2914                         (unsigned long long)acct.cancelled_write_bytes);
2915 out_unlock:
2916         mutex_unlock(&task->signal->cred_guard_mutex);
2917         return result;
2918 }
2919
2920 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2921 {
2922         return do_io_accounting(task, buffer, 0);
2923 }
2924
2925 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2926 {
2927         return do_io_accounting(task, buffer, 1);
2928 }
2929 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2930
2931 #ifdef CONFIG_USER_NS
2932 static int proc_id_map_open(struct inode *inode, struct file *file,
2933         struct seq_operations *seq_ops)
2934 {
2935         struct user_namespace *ns = NULL;
2936         struct task_struct *task;
2937         struct seq_file *seq;
2938         int ret = -EINVAL;
2939
2940         task = get_proc_task(inode);
2941         if (task) {
2942                 rcu_read_lock();
2943                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2944                 rcu_read_unlock();
2945                 put_task_struct(task);
2946         }
2947         if (!ns)
2948                 goto err;
2949
2950         ret = seq_open(file, seq_ops);
2951         if (ret)
2952                 goto err_put_ns;
2953
2954         seq = file->private_data;
2955         seq->private = ns;
2956
2957         return 0;
2958 err_put_ns:
2959         put_user_ns(ns);
2960 err:
2961         return ret;
2962 }
2963
2964 static int proc_id_map_release(struct inode *inode, struct file *file)
2965 {
2966         struct seq_file *seq = file->private_data;
2967         struct user_namespace *ns = seq->private;
2968         put_user_ns(ns);
2969         return seq_release(inode, file);
2970 }
2971
2972 static int proc_uid_map_open(struct inode *inode, struct file *file)
2973 {
2974         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2975 }
2976
2977 static int proc_gid_map_open(struct inode *inode, struct file *file)
2978 {
2979         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2980 }
2981
2982 static const struct file_operations proc_uid_map_operations = {
2983         .open           = proc_uid_map_open,
2984         .write          = proc_uid_map_write,
2985         .read           = seq_read,
2986         .llseek         = seq_lseek,
2987         .release        = proc_id_map_release,
2988 };
2989
2990 static const struct file_operations proc_gid_map_operations = {
2991         .open           = proc_gid_map_open,
2992         .write          = proc_gid_map_write,
2993         .read           = seq_read,
2994         .llseek         = seq_lseek,
2995         .release        = proc_id_map_release,
2996 };
2997 #endif /* CONFIG_USER_NS */
2998
2999 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3000                                 struct pid *pid, struct task_struct *task)
3001 {
3002         int err = lock_trace(task);
3003         if (!err) {
3004                 seq_printf(m, "%08x\n", task->personality);
3005                 unlock_trace(task);
3006         }
3007         return err;
3008 }
3009
3010 /*
3011  * Thread groups
3012  */
3013 static const struct file_operations proc_task_operations;
3014 static const struct inode_operations proc_task_inode_operations;
3015
3016 static const struct pid_entry tgid_base_stuff[] = {
3017         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3018         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3019 #ifdef CONFIG_CHECKPOINT_RESTORE
3020         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3021 #endif
3022         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3023         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3024 #ifdef CONFIG_NET
3025         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3026 #endif
3027         REG("environ",    S_IRUSR, proc_environ_operations),
3028         INF("auxv",       S_IRUSR, proc_pid_auxv),
3029         ONE("status",     S_IRUGO, proc_pid_status),
3030         ONE("personality", S_IRUGO, proc_pid_personality),
3031         INF("limits",     S_IRUGO, proc_pid_limits),
3032 #ifdef CONFIG_SCHED_DEBUG
3033         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3034 #endif
3035 #ifdef CONFIG_SCHED_AUTOGROUP
3036         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3037 #endif
3038         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3039 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3040         INF("syscall",    S_IRUGO, proc_pid_syscall),
3041 #endif
3042         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3043         ONE("stat",       S_IRUGO, proc_tgid_stat),
3044         ONE("statm",      S_IRUGO, proc_pid_statm),
3045         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3046 #ifdef CONFIG_NUMA
3047         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3048 #endif
3049         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3050         LNK("cwd",        proc_cwd_link),
3051         LNK("root",       proc_root_link),
3052         LNK("exe",        proc_exe_link),
3053         REG("mounts",     S_IRUGO, proc_mounts_operations),
3054         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3055         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3056 #ifdef CONFIG_PROC_PAGE_MONITOR
3057         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3058         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3059         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3060 #endif
3061 #ifdef CONFIG_SECURITY
3062         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3063 #endif
3064 #ifdef CONFIG_KALLSYMS
3065         INF("wchan",      S_IRUGO, proc_pid_wchan),
3066 #endif
3067 #ifdef CONFIG_STACKTRACE
3068         ONE("stack",      S_IRUGO, proc_pid_stack),
3069 #endif
3070 #ifdef CONFIG_SCHEDSTATS
3071         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3072 #endif
3073 #ifdef CONFIG_LATENCYTOP
3074         REG("latency",  S_IRUGO, proc_lstats_operations),
3075 #endif
3076 #ifdef CONFIG_PROC_PID_CPUSET
3077         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3078 #endif
3079 #ifdef CONFIG_CGROUPS
3080         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3081 #endif
3082         INF("oom_score",  S_IRUGO, proc_oom_score),
3083         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3084         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3085 #ifdef CONFIG_AUDITSYSCALL
3086         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3087         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3088 #endif
3089 #ifdef CONFIG_FAULT_INJECTION
3090         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3091 #endif
3092 #ifdef CONFIG_ELF_CORE
3093         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3094 #endif
3095 #ifdef CONFIG_TASK_IO_ACCOUNTING
3096         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3097 #endif
3098 #ifdef CONFIG_HARDWALL
3099         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3100 #endif
3101 #ifdef CONFIG_USER_NS
3102         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3103         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3104 #endif
3105 };
3106
3107 static int proc_tgid_base_readdir(struct file * filp,
3108                              void * dirent, filldir_t filldir)
3109 {
3110         return proc_pident_readdir(filp,dirent,filldir,
3111                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3112 }
3113
3114 static const struct file_operations proc_tgid_base_operations = {
3115         .read           = generic_read_dir,
3116         .readdir        = proc_tgid_base_readdir,
3117         .llseek         = default_llseek,
3118 };
3119
3120 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3121 {
3122         return proc_pident_lookup(dir, dentry,
3123                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3124 }
3125
3126 static const struct inode_operations proc_tgid_base_inode_operations = {
3127         .lookup         = proc_tgid_base_lookup,
3128         .getattr        = pid_getattr,
3129         .setattr        = proc_setattr,
3130         .permission     = proc_pid_permission,
3131 };
3132
3133 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3134 {
3135         struct dentry *dentry, *leader, *dir;
3136         char buf[PROC_NUMBUF];
3137         struct qstr name;
3138
3139         name.name = buf;
3140         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3141         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3142         if (dentry) {
3143                 shrink_dcache_parent(dentry);
3144                 d_drop(dentry);
3145                 dput(dentry);
3146         }
3147
3148         name.name = buf;
3149         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3150         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3151         if (!leader)
3152                 goto out;
3153
3154         name.name = "task";
3155         name.len = strlen(name.name);
3156         dir = d_hash_and_lookup(leader, &name);
3157         if (!dir)
3158                 goto out_put_leader;
3159
3160         name.name = buf;
3161         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3162         dentry = d_hash_and_lookup(dir, &name);
3163         if (dentry) {
3164                 shrink_dcache_parent(dentry);
3165                 d_drop(dentry);
3166                 dput(dentry);
3167         }
3168
3169         dput(dir);
3170 out_put_leader:
3171         dput(leader);
3172 out:
3173         return;
3174 }
3175
3176 /**
3177  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3178  * @task: task that should be flushed.
3179  *
3180  * When flushing dentries from proc, one needs to flush them from global
3181  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3182  * in. This call is supposed to do all of this job.
3183  *
3184  * Looks in the dcache for
3185  * /proc/@pid
3186  * /proc/@tgid/task/@pid
3187  * if either directory is present flushes it and all of it'ts children
3188  * from the dcache.
3189  *
3190  * It is safe and reasonable to cache /proc entries for a task until
3191  * that task exits.  After that they just clog up the dcache with
3192  * useless entries, possibly causing useful dcache entries to be
3193  * flushed instead.  This routine is proved to flush those useless
3194  * dcache entries at process exit time.
3195  *
3196  * NOTE: This routine is just an optimization so it does not guarantee
3197  *       that no dcache entries will exist at process exit time it
3198  *       just makes it very unlikely that any will persist.
3199  */
3200
3201 void proc_flush_task(struct task_struct *task)
3202 {
3203         int i;
3204         struct pid *pid, *tgid;
3205         struct upid *upid;
3206
3207         pid = task_pid(task);
3208         tgid = task_tgid(task);
3209
3210         for (i = 0; i <= pid->level; i++) {
3211                 upid = &pid->numbers[i];
3212                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3213                                         tgid->numbers[i].nr);
3214         }
3215
3216         upid = &pid->numbers[pid->level];
3217         if (upid->nr == 1)
3218                 pid_ns_release_proc(upid->ns);
3219 }
3220
3221 static struct dentry *proc_pid_instantiate(struct inode *dir,
3222                                            struct dentry * dentry,
3223                                            struct task_struct *task, const void *ptr)
3224 {
3225         struct dentry *error = ERR_PTR(-ENOENT);
3226         struct inode *inode;
3227
3228         inode = proc_pid_make_inode(dir->i_sb, task);
3229         if (!inode)
3230                 goto out;
3231
3232         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3233         inode->i_op = &proc_tgid_base_inode_operations;
3234         inode->i_fop = &proc_tgid_base_operations;
3235         inode->i_flags|=S_IMMUTABLE;
3236
3237         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3238                                                   ARRAY_SIZE(tgid_base_stuff)));
3239
3240         d_set_d_op(dentry, &pid_dentry_operations);
3241
3242         d_add(dentry, inode);
3243         /* Close the race of the process dying before we return the dentry */
3244         if (pid_revalidate(dentry, 0))
3245                 error = NULL;
3246 out:
3247         return error;
3248 }
3249
3250 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3251 {
3252         struct dentry *result;
3253         struct task_struct *task;
3254         unsigned tgid;
3255         struct pid_namespace *ns;
3256
3257         result = proc_base_lookup(dir, dentry);
3258         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3259                 goto out;
3260
3261         tgid = name_to_int(dentry);
3262         if (tgid == ~0U)
3263                 goto out;
3264
3265         ns = dentry->d_sb->s_fs_info;
3266         rcu_read_lock();
3267         task = find_task_by_pid_ns(tgid, ns);
3268         if (task)
3269                 get_task_struct(task);
3270         rcu_read_unlock();
3271         if (!task)
3272                 goto out;
3273
3274         result = proc_pid_instantiate(dir, dentry, task, NULL);
3275         put_task_struct(task);
3276 out:
3277         return result;
3278 }
3279
3280 /*
3281  * Find the first task with tgid >= tgid
3282  *
3283  */
3284 struct tgid_iter {
3285         unsigned int tgid;
3286         struct task_struct *task;
3287 };
3288 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3289 {
3290         struct pid *pid;
3291
3292         if (iter.task)
3293                 put_task_struct(iter.task);
3294         rcu_read_lock();
3295 retry:
3296         iter.task = NULL;
3297         pid = find_ge_pid(iter.tgid, ns);
3298         if (pid) {
3299                 iter.tgid = pid_nr_ns(pid, ns);
3300                 iter.task = pid_task(pid, PIDTYPE_PID);
3301                 /* What we to know is if the pid we have find is the
3302                  * pid of a thread_group_leader.  Testing for task
3303                  * being a thread_group_leader is the obvious thing
3304                  * todo but there is a window when it fails, due to
3305                  * the pid transfer logic in de_thread.
3306                  *
3307                  * So we perform the straight forward test of seeing
3308                  * if the pid we have found is the pid of a thread
3309                  * group leader, and don't worry if the task we have
3310                  * found doesn't happen to be a thread group leader.
3311                  * As we don't care in the case of readdir.
3312                  */
3313                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3314                         iter.tgid += 1;
3315                         goto retry;
3316                 }
3317                 get_task_struct(iter.task);
3318         }
3319         rcu_read_unlock();
3320         return iter;
3321 }
3322
3323 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3324
3325 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3326         struct tgid_iter iter)
3327 {
3328         char name[PROC_NUMBUF];
3329         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3330         return proc_fill_cache(filp, dirent, filldir, name, len,
3331                                 proc_pid_instantiate, iter.task, NULL);
3332 }
3333
3334 static int fake_filldir(void *buf, const char *name, int namelen,
3335                         loff_t offset, u64 ino, unsigned d_type)
3336 {
3337         return 0;
3338 }
3339
3340 /* for the /proc/ directory itself, after non-process stuff has been done */
3341 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3342 {
3343         unsigned int nr;
3344         struct task_struct *reaper;
3345         struct tgid_iter iter;
3346         struct pid_namespace *ns;
3347         filldir_t __filldir;
3348
3349         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3350                 goto out_no_task;
3351         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3352
3353         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3354         if (!reaper)
3355                 goto out_no_task;
3356
3357         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3358                 const struct pid_entry *p = &proc_base_stuff[nr];
3359                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3360                         goto out;
3361         }
3362
3363         ns = filp->f_dentry->d_sb->s_fs_info;
3364         iter.task = NULL;
3365         iter.tgid = filp->f_pos - TGID_OFFSET;
3366         for (iter = next_tgid(ns, iter);
3367              iter.task;
3368              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3369                 if (has_pid_permissions(ns, iter.task, 2))
3370                         __filldir = filldir;
3371                 else
3372                         __filldir = fake_filldir;
3373
3374                 filp->f_pos = iter.tgid + TGID_OFFSET;
3375                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3376                         put_task_struct(iter.task);
3377                         goto out;
3378                 }
3379         }
3380         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3381 out:
3382         put_task_struct(reaper);
3383 out_no_task:
3384         return 0;
3385 }
3386
3387 /*
3388  * Tasks
3389  */
3390 static const struct pid_entry tid_base_stuff[] = {
3391         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3392         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3393         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3394         REG("environ",   S_IRUSR, proc_environ_operations),
3395         INF("auxv",      S_IRUSR, proc_pid_auxv),
3396         ONE("status",    S_IRUGO, proc_pid_status),
3397         ONE("personality", S_IRUGO, proc_pid_personality),
3398         INF("limits",    S_IRUGO, proc_pid_limits),
3399 #ifdef CONFIG_SCHED_DEBUG
3400         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3401 #endif
3402         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3403 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3404         INF("syscall",   S_IRUGO, proc_pid_syscall),
3405 #endif
3406         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3407         ONE("stat",      S_IRUGO, proc_tid_stat),
3408         ONE("statm",     S_IRUGO, proc_pid_statm),
3409         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3410 #ifdef CONFIG_CHECKPOINT_RESTORE
3411         REG("children",  S_IRUGO, proc_tid_children_operations),
3412 #endif
3413 #ifdef CONFIG_NUMA
3414         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3415 #endif
3416         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3417         LNK("cwd",       proc_cwd_link),
3418         LNK("root",      proc_root_link),
3419         LNK("exe",       proc_exe_link),
3420         REG("mounts",    S_IRUGO, proc_mounts_operations),
3421         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3422 #ifdef CONFIG_PROC_PAGE_MONITOR
3423         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3424         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3425         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3426 #endif
3427 #ifdef CONFIG_SECURITY
3428         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3429 #endif
3430 #ifdef CONFIG_KALLSYMS
3431         INF("wchan",     S_IRUGO, proc_pid_wchan),
3432 #endif
3433 #ifdef CONFIG_STACKTRACE
3434         ONE("stack",      S_IRUGO, proc_pid_stack),
3435 #endif
3436 #ifdef CONFIG_SCHEDSTATS
3437         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3438 #endif
3439 #ifdef CONFIG_LATENCYTOP
3440         REG("latency",  S_IRUGO, proc_lstats_operations),
3441 #endif
3442 #ifdef CONFIG_PROC_PID_CPUSET
3443         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3444 #endif
3445 #ifdef CONFIG_CGROUPS
3446         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3447 #endif
3448         INF("oom_score", S_IRUGO, proc_oom_score),
3449         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3450         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3451 #ifdef CONFIG_AUDITSYSCALL
3452         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3453         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3454 #endif
3455 #ifdef CONFIG_FAULT_INJECTION
3456         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3457 #endif
3458 #ifdef CONFIG_TASK_IO_ACCOUNTING
3459         INF("io",       S_IRUSR, proc_tid_io_accounting),
3460 #endif
3461 #ifdef CONFIG_HARDWALL
3462         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3463 #endif
3464 #ifdef CONFIG_USER_NS
3465         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3466         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3467 #endif
3468 };
3469
3470 static int proc_tid_base_readdir(struct file * filp,
3471                              void * dirent, filldir_t filldir)
3472 {
3473         return proc_pident_readdir(filp,dirent,filldir,
3474                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3475 }
3476
3477 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3478 {
3479         return proc_pident_lookup(dir, dentry,
3480                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3481 }
3482
3483 static const struct file_operations proc_tid_base_operations = {
3484         .read           = generic_read_dir,
3485         .readdir        = proc_tid_base_readdir,
3486         .llseek         = default_llseek,
3487 };
3488
3489 static const struct inode_operations proc_tid_base_inode_operations = {
3490         .lookup         = proc_tid_base_lookup,
3491         .getattr        = pid_getattr,
3492         .setattr        = proc_setattr,
3493 };
3494
3495 static struct dentry *proc_task_instantiate(struct inode *dir,
3496         struct dentry *dentry, struct task_struct *task, const void *ptr)
3497 {
3498         struct dentry *error = ERR_PTR(-ENOENT);
3499         struct inode *inode;
3500         inode = proc_pid_make_inode(dir->i_sb, task);
3501
3502         if (!inode)
3503                 goto out;
3504         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3505         inode->i_op = &proc_tid_base_inode_operations;
3506         inode->i_fop = &proc_tid_base_operations;
3507         inode->i_flags|=S_IMMUTABLE;
3508
3509         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3510                                                   ARRAY_SIZE(tid_base_stuff)));
3511
3512         d_set_d_op(dentry, &pid_dentry_operations);
3513
3514         d_add(dentry, inode);
3515         /* Close the race of the process dying before we return the dentry */
3516         if (pid_revalidate(dentry, 0))
3517                 error = NULL;
3518 out:
3519         return error;
3520 }
3521
3522 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3523 {
3524         struct dentry *result = ERR_PTR(-ENOENT);
3525         struct task_struct *task;
3526         struct task_struct *leader = get_proc_task(dir);
3527         unsigned tid;
3528         struct pid_namespace *ns;
3529
3530         if (!leader)
3531                 goto out_no_task;
3532
3533         tid = name_to_int(dentry);
3534         if (tid == ~0U)
3535                 goto out;
3536
3537         ns = dentry->d_sb->s_fs_info;
3538         rcu_read_lock();
3539         task = find_task_by_pid_ns(tid, ns);
3540         if (task)
3541                 get_task_struct(task);
3542         rcu_read_unlock();
3543         if (!task)
3544                 goto out;
3545         if (!same_thread_group(leader, task))
3546                 goto out_drop_task;
3547
3548         result = proc_task_instantiate(dir, dentry, task, NULL);
3549 out_drop_task:
3550         put_task_struct(task);
3551 out:
3552         put_task_struct(leader);
3553 out_no_task:
3554         return result;
3555 }
3556
3557 /*
3558  * Find the first tid of a thread group to return to user space.
3559  *
3560  * Usually this is just the thread group leader, but if the users
3561  * buffer was too small or there was a seek into the middle of the
3562  * directory we have more work todo.
3563  *
3564  * In the case of a short read we start with find_task_by_pid.
3565  *
3566  * In the case of a seek we start with the leader and walk nr
3567  * threads past it.
3568  */
3569 static struct task_struct *first_tid(struct task_struct *leader,
3570                 int tid, int nr, struct pid_namespace *ns)
3571 {
3572         struct task_struct *pos;
3573
3574         rcu_read_lock();
3575         /* Attempt to start with the pid of a thread */
3576         if (tid && (nr > 0)) {
3577                 pos = find_task_by_pid_ns(tid, ns);
3578                 if (pos && (pos->group_leader == leader))
3579                         goto found;
3580         }
3581
3582         /* If nr exceeds the number of threads there is nothing todo */
3583         pos = NULL;
3584         if (nr && nr >= get_nr_threads(leader))
3585                 goto out;
3586
3587         /* If we haven't found our starting place yet start
3588          * with the leader and walk nr threads forward.
3589          */
3590         for (pos = leader; nr > 0; --nr) {
3591                 pos = next_thread(pos);
3592                 if (pos == leader) {
3593                         pos = NULL;
3594                         goto out;
3595                 }
3596         }
3597 found:
3598         get_task_struct(pos);
3599 out:
3600         rcu_read_unlock();
3601         return pos;
3602 }
3603
3604 /*
3605  * Find the next thread in the thread list.
3606  * Return NULL if there is an error or no next thread.
3607  *
3608  * The reference to the input task_struct is released.
3609  */
3610 static struct task_struct *next_tid(struct task_struct *start)
3611 {
3612         struct task_struct *pos = NULL;
3613         rcu_read_lock();
3614         if (pid_alive(start)) {
3615                 pos = next_thread(start);
3616                 if (thread_group_leader(pos))
3617                         pos = NULL;
3618                 else
3619                         get_task_struct(pos);
3620         }
3621         rcu_read_unlock();
3622         put_task_struct(start);
3623         return pos;
3624 }
3625
3626 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3627         struct task_struct *task, int tid)
3628 {
3629         char name[PROC_NUMBUF];
3630         int len = snprintf(name, sizeof(name), "%d", tid);
3631         return proc_fill_cache(filp, dirent, filldir, name, len,
3632                                 proc_task_instantiate, task, NULL);
3633 }
3634
3635 /* for the /proc/TGID/task/ directories */
3636 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3637 {
3638         struct dentry *dentry = filp->f_path.dentry;
3639         struct inode *inode = dentry->d_inode;
3640         struct task_struct *leader = NULL;
3641         struct task_struct *task;
3642         int retval = -ENOENT;
3643         ino_t ino;
3644         int tid;
3645         struct pid_namespace *ns;
3646
3647         task = get_proc_task(inode);
3648         if (!task)
3649                 goto out_no_task;
3650         rcu_read_lock();
3651         if (pid_alive(task)) {
3652                 leader = task->group_leader;
3653                 get_task_struct(leader);
3654         }
3655         rcu_read_unlock();
3656         put_task_struct(task);
3657         if (!leader)
3658                 goto out_no_task;
3659         retval = 0;
3660
3661         switch ((unsigned long)filp->f_pos) {
3662         case 0:
3663                 ino = inode->i_ino;
3664                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3665                         goto out;
3666                 filp->f_pos++;
3667                 /* fall through */
3668         case 1:
3669                 ino = parent_ino(dentry);
3670                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3671                         goto out;
3672                 filp->f_pos++;
3673                 /* fall through */
3674         }
3675
3676         /* f_version caches the tgid value that the last readdir call couldn't
3677          * return. lseek aka telldir automagically resets f_version to 0.
3678          */
3679         ns = filp->f_dentry->d_sb->s_fs_info;
3680         tid = (int)filp->f_version;
3681         filp->f_version = 0;
3682         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3683              task;
3684              task = next_tid(task), filp->f_pos++) {
3685                 tid = task_pid_nr_ns(task, ns);
3686                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3687                         /* returning this tgid failed, save it as the first
3688                          * pid for the next readir call */
3689                         filp->f_version = (u64)tid;
3690                         put_task_struct(task);
3691                         break;
3692                 }
3693         }
3694 out:
3695         put_task_struct(leader);
3696 out_no_task:
3697         return retval;
3698 }
3699
3700 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3701 {
3702         struct inode *inode = dentry->d_inode;
3703         struct task_struct *p = get_proc_task(inode);
3704         generic_fillattr(inode, stat);
3705
3706         if (p) {
3707                 stat->nlink += get_nr_threads(p);
3708                 put_task_struct(p);
3709         }
3710
3711         return 0;
3712 }
3713
3714 static const struct inode_operations proc_task_inode_operations = {
3715         .lookup         = proc_task_lookup,
3716         .getattr        = proc_task_getattr,
3717         .setattr        = proc_setattr,
3718         .permission     = proc_pid_permission,
3719 };
3720
3721 static const struct file_operations proc_task_operations = {
3722         .read           = generic_read_dir,
3723         .readdir        = proc_task_readdir,
3724         .llseek         = default_llseek,
3725 };