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         file->private_data = mm;
 699
 700         return 0;
 701 }
 702
 703 static int mem_open(struct inode *inode, struct file *file)
 704 {
 705         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 706
 707         /* OK to pass negative loff_t, we can catch out-of-range */
 708         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 709
 710         return ret;
 711 }
 712
 713 static ssize_t mem_rw(struct file *file, char __user *buf,
 714                         size_t count, loff_t *ppos, int write)
 715 {
 716         struct mm_struct *mm = file->private_data;
 717         unsigned long addr = *ppos;
 718         ssize_t copied;
 719         char *page;
 720
 721         if (!mm)
 722                 return 0;
 723
 724         page = (char *)__get_free_page(GFP_TEMPORARY);
 725         if (!page)
 726                 return -ENOMEM;
 727
 728         copied = 0;
 729         if (!atomic_inc_not_zero(&mm->mm_users))
 730                 goto free;
 731
 732         while (count > 0) {
 733                 int this_len = min_t(int, count, PAGE_SIZE);
 734
 735                 if (write && copy_from_user(page, buf, this_len)) {
 736                         copied = -EFAULT;
 737                         break;
 738                 }
 739
 740                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 741                 if (!this_len) {
 742                         if (!copied)
 743                                 copied = -EIO;
 744                         break;
 745                 }
 746
 747                 if (!write && copy_to_user(buf, page, this_len)) {
 748                         copied = -EFAULT;
 749                         break;
 750                 }
 751
 752                 buf += this_len;
 753                 addr += this_len;
 754                 copied += this_len;
 755                 count -= this_len;
 756         }
 757         *ppos = addr;
 758
 759         mmput(mm);
 760 free:
 761         free_page((unsigned long) page);
 762         return copied;
 763 }
 764
 765 static ssize_t mem_read(struct file *file, char __user *buf,
 766                         size_t count, loff_t *ppos)
 767 {
 768         return mem_rw(file, buf, count, ppos, 0);
 769 }
 770
 771 static ssize_t mem_write(struct file *file, const char __user *buf,
 772                          size_t count, loff_t *ppos)
 773 {
 774         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 775 }
 776
 777 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 778 {
 779         switch (orig) {
 780         case 0:
 781                 file->f_pos = offset;
 782                 break;
 783         case 1:
 784                 file->f_pos += offset;
 785                 break;
 786         default:
 787                 return -EINVAL;
 788         }
 789         force_successful_syscall_return();
 790         return file->f_pos;
 791 }
 792
 793 static int mem_release(struct inode *inode, struct file *file)
 794 {
 795         struct mm_struct *mm = file->private_data;
 796         if (mm)
 797                 mmdrop(mm);
 798         return 0;
 799 }
 800
 801 static const struct file_operations proc_mem_operations = {
 802         .llseek         = mem_lseek,
 803         .read           = mem_read,
 804         .write          = mem_write,
 805         .open           = mem_open,
 806         .release        = mem_release,
 807 };
 808
 809 static int environ_open(struct inode *inode, struct file *file)
 810 {
 811         return __mem_open(inode, file, PTRACE_MODE_READ);
 812 }
 813
 814 static ssize_t environ_read(struct file *file, char __user *buf,
 815                         size_t count, loff_t *ppos)
 816 {
 817         char *page;
 818         unsigned long src = *ppos;
 819         int ret = 0;
 820         struct mm_struct *mm = file->private_data;
 821
 822         if (!mm)
 823                 return 0;
 824
 825         page = (char *)__get_free_page(GFP_TEMPORARY);
 826         if (!page)
 827                 return -ENOMEM;
 828
 829         ret = 0;
 830         if (!atomic_inc_not_zero(&mm->mm_users))
 831                 goto free;
 832         while (count > 0) {
 833                 size_t this_len, max_len;
 834                 int retval;
 835
 836                 if (src >= (mm->env_end - mm->env_start))
 837                         break;
 838
 839                 this_len = mm->env_end - (mm->env_start + src);
 840
 841                 max_len = min_t(size_t, PAGE_SIZE, count);
 842                 this_len = min(max_len, this_len);
 843
 844                 retval = access_remote_vm(mm, (mm->env_start + src),
 845                         page, this_len, 0);
 846
 847                 if (retval <= 0) {
 848                         ret = retval;
 849                         break;
 850                 }
 851
 852                 if (copy_to_user(buf, page, retval)) {
 853                         ret = -EFAULT;
 854                         break;
 855                 }
 856
 857                 ret += retval;
 858                 src += retval;
 859                 buf += retval;
 860                 count -= retval;
 861         }
 862         *ppos = src;
 863         mmput(mm);
 864
 865 free:
 866         free_page((unsigned long) page);
 867         return ret;
 868 }
 869
 870 static const struct file_operations proc_environ_operations = {
 871         .open           = environ_open,
 872         .read           = environ_read,
 873         .llseek         = generic_file_llseek,
 874         .release        = mem_release,
 875 };
 876
 877 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 878                                 size_t count, loff_t *ppos)
 879 {
 880         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 881         char buffer[PROC_NUMBUF];
 882         size_t len;
 883         int oom_adjust = OOM_DISABLE;
 884         unsigned long flags;
 885
 886         if (!task)
 887                 return -ESRCH;
 888
 889         if (lock_task_sighand(task, &flags)) {
 890                 oom_adjust = task->signal->oom_adj;
 891                 unlock_task_sighand(task, &flags);
 892         }
 893
 894         put_task_struct(task);
 895
 896         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 897
 898         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 899 }
 900
 901 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 902                                 size_t count, loff_t *ppos)
 903 {
 904         struct task_struct *task;
 905         char buffer[PROC_NUMBUF];
 906         int oom_adjust;
 907         unsigned long flags;
 908         int err;
 909
 910         memset(buffer, 0, sizeof(buffer));
 911         if (count > sizeof(buffer) - 1)
 912                 count = sizeof(buffer) - 1;
 913         if (copy_from_user(buffer, buf, count)) {
 914                 err = -EFAULT;
 915                 goto out;
 916         }
 917
 918         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 919         if (err)
 920                 goto out;
 921         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 922              oom_adjust != OOM_DISABLE) {
 923                 err = -EINVAL;
 924                 goto out;
 925         }
 926
 927         task = get_proc_task(file->f_path.dentry->d_inode);
 928         if (!task) {
 929                 err = -ESRCH;
 930                 goto out;
 931         }
 932
 933         task_lock(task);
 934         if (!task->mm) {
 935                 err = -EINVAL;
 936                 goto err_task_lock;
 937         }
 938
 939         if (!lock_task_sighand(task, &flags)) {
 940                 err = -ESRCH;
 941                 goto err_task_lock;
 942         }
 943
 944         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 945                 err = -EACCES;
 946                 goto err_sighand;
 947         }
 948
 949         /*
 950          * Warn that /proc/pid/oom_adj is deprecated, see
 951          * Documentation/feature-removal-schedule.txt.
 952          */
 953         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 954                   current->comm, task_pid_nr(current), task_pid_nr(task),
 955                   task_pid_nr(task));
 956         task->signal->oom_adj = oom_adjust;
 957         /*
 958          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 959          * value is always attainable.
 960          */
 961         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 962                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 963         else
 964                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 965                                                                 -OOM_DISABLE;
 966         trace_oom_score_adj_update(task);
 967 err_sighand:
 968         unlock_task_sighand(task, &flags);
 969 err_task_lock:
 970         task_unlock(task);
 971         put_task_struct(task);
 972 out:
 973         return err < 0 ? err : count;
 974 }
 975
 976 static const struct file_operations proc_oom_adjust_operations = {
 977         .read           = oom_adjust_read,
 978         .write          = oom_adjust_write,
 979         .llseek         = generic_file_llseek,
 980 };
 981
 982 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 983                                         size_t count, loff_t *ppos)
 984 {
 985         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 986         char buffer[PROC_NUMBUF];
 987         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 988         unsigned long flags;
 989         size_t len;
 990
 991         if (!task)
 992                 return -ESRCH;
 993         if (lock_task_sighand(task, &flags)) {
 994                 oom_score_adj = task->signal->oom_score_adj;
 995                 unlock_task_sighand(task, &flags);
 996         }
 997         put_task_struct(task);
 998         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 999         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1000 }
1001
1002 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1003                                         size_t count, loff_t *ppos)
1004 {
1005         struct task_struct *task;
1006         char buffer[PROC_NUMBUF];
1007         unsigned long flags;
1008         int oom_score_adj;
1009         int err;
1010
1011         memset(buffer, 0, sizeof(buffer));
1012         if (count > sizeof(buffer) - 1)
1013                 count = sizeof(buffer) - 1;
1014         if (copy_from_user(buffer, buf, count)) {
1015                 err = -EFAULT;
1016                 goto out;
1017         }
1018
1019         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1020         if (err)
1021                 goto out;
1022         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1023                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1024                 err = -EINVAL;
1025                 goto out;
1026         }
1027
1028         task = get_proc_task(file->f_path.dentry->d_inode);
1029         if (!task) {
1030                 err = -ESRCH;
1031                 goto out;
1032         }
1033
1034         task_lock(task);
1035         if (!task->mm) {
1036                 err = -EINVAL;
1037                 goto err_task_lock;
1038         }
1039
1040         if (!lock_task_sighand(task, &flags)) {
1041                 err = -ESRCH;
1042                 goto err_task_lock;
1043         }
1044
1045         if (oom_score_adj < task->signal->oom_score_adj_min &&
1046                         !capable(CAP_SYS_RESOURCE)) {
1047                 err = -EACCES;
1048                 goto err_sighand;
1049         }
1050
1051         task->signal->oom_score_adj = oom_score_adj;
1052         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1053                 task->signal->oom_score_adj_min = oom_score_adj;
1054         trace_oom_score_adj_update(task);
1055         /*
1056          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1057          * always attainable.
1058          */
1059         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1060                 task->signal->oom_adj = OOM_DISABLE;
1061         else
1062                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1063                                                         OOM_SCORE_ADJ_MAX;
1064 err_sighand:
1065         unlock_task_sighand(task, &flags);
1066 err_task_lock:
1067         task_unlock(task);
1068         put_task_struct(task);
1069 out:
1070         return err < 0 ? err : count;
1071 }
1072
1073 static const struct file_operations proc_oom_score_adj_operations = {
1074         .read           = oom_score_adj_read,
1075         .write          = oom_score_adj_write,
1076         .llseek         = default_llseek,
1077 };
1078
1079 #ifdef CONFIG_AUDITSYSCALL
1080 #define TMPBUFLEN 21
1081 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1082                                   size_t count, loff_t *ppos)
1083 {
1084         struct inode * inode = file->f_path.dentry->d_inode;
1085         struct task_struct *task = get_proc_task(inode);
1086         ssize_t length;
1087         char tmpbuf[TMPBUFLEN];
1088
1089         if (!task)
1090                 return -ESRCH;
1091         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1092                            from_kuid(file->f_cred->user_ns,
1093                                      audit_get_loginuid(task)));
1094         put_task_struct(task);
1095         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1096 }
1097
1098 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1099                                    size_t count, loff_t *ppos)
1100 {
1101         struct inode * inode = file->f_path.dentry->d_inode;
1102         char *page, *tmp;
1103         ssize_t length;
1104         uid_t loginuid;
1105         kuid_t kloginuid;
1106
1107         rcu_read_lock();
1108         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1109                 rcu_read_unlock();
1110                 return -EPERM;
1111         }
1112         rcu_read_unlock();
1113
1114         if (count >= PAGE_SIZE)
1115                 count = PAGE_SIZE - 1;
1116
1117         if (*ppos != 0) {
1118                 /* No partial writes. */
1119                 return -EINVAL;
1120         }
1121         page = (char*)__get_free_page(GFP_TEMPORARY);
1122         if (!page)
1123                 return -ENOMEM;
1124         length = -EFAULT;
1125         if (copy_from_user(page, buf, count))
1126                 goto out_free_page;
1127
1128         page[count] = '\0';
1129         loginuid = simple_strtoul(page, &tmp, 10);
1130         if (tmp == page) {
1131                 length = -EINVAL;
1132                 goto out_free_page;
1133
1134         }
1135         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1136         if (!uid_valid(kloginuid)) {
1137                 length = -EINVAL;
1138                 goto out_free_page;
1139         }
1140
1141         length = audit_set_loginuid(kloginuid);
1142         if (likely(length == 0))
1143                 length = count;
1144
1145 out_free_page:
1146         free_page((unsigned long) page);
1147         return length;
1148 }
1149
1150 static const struct file_operations proc_loginuid_operations = {
1151         .read           = proc_loginuid_read,
1152         .write          = proc_loginuid_write,
1153         .llseek         = generic_file_llseek,
1154 };
1155
1156 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1157                                   size_t count, loff_t *ppos)
1158 {
1159         struct inode * inode = file->f_path.dentry->d_inode;
1160         struct task_struct *task = get_proc_task(inode);
1161         ssize_t length;
1162         char tmpbuf[TMPBUFLEN];
1163
1164         if (!task)
1165                 return -ESRCH;
1166         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1167                                 audit_get_sessionid(task));
1168         put_task_struct(task);
1169         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1170 }
1171
1172 static const struct file_operations proc_sessionid_operations = {
1173         .read           = proc_sessionid_read,
1174         .llseek         = generic_file_llseek,
1175 };
1176 #endif
1177
1178 #ifdef CONFIG_FAULT_INJECTION
1179 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1180                                       size_t count, loff_t *ppos)
1181 {
1182         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1183         char buffer[PROC_NUMBUF];
1184         size_t len;
1185         int make_it_fail;
1186
1187         if (!task)
1188                 return -ESRCH;
1189         make_it_fail = task->make_it_fail;
1190         put_task_struct(task);
1191
1192         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1193
1194         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1195 }
1196
1197 static ssize_t proc_fault_inject_write(struct file * file,
1198                         const char __user * buf, size_t count, loff_t *ppos)
1199 {
1200         struct task_struct *task;
1201         char buffer[PROC_NUMBUF], *end;
1202         int make_it_fail;
1203
1204         if (!capable(CAP_SYS_RESOURCE))
1205                 return -EPERM;
1206         memset(buffer, 0, sizeof(buffer));
1207         if (count > sizeof(buffer) - 1)
1208                 count = sizeof(buffer) - 1;
1209         if (copy_from_user(buffer, buf, count))
1210                 return -EFAULT;
1211         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1212         if (*end)
1213                 return -EINVAL;
1214         task = get_proc_task(file->f_dentry->d_inode);
1215         if (!task)
1216                 return -ESRCH;
1217         task->make_it_fail = make_it_fail;
1218         put_task_struct(task);
1219
1220         return count;
1221 }
1222
1223 static const struct file_operations proc_fault_inject_operations = {
1224         .read           = proc_fault_inject_read,
1225         .write          = proc_fault_inject_write,
1226         .llseek         = generic_file_llseek,
1227 };
1228 #endif
1229
1230
1231 #ifdef CONFIG_SCHED_DEBUG
1232 /*
1233  * Print out various scheduling related per-task fields:
1234  */
1235 static int sched_show(struct seq_file *m, void *v)
1236 {
1237         struct inode *inode = m->private;
1238         struct task_struct *p;
1239
1240         p = get_proc_task(inode);
1241         if (!p)
1242                 return -ESRCH;
1243         proc_sched_show_task(p, m);
1244
1245         put_task_struct(p);
1246
1247         return 0;
1248 }
1249
1250 static ssize_t
1251 sched_write(struct file *file, const char __user *buf,
1252             size_t count, loff_t *offset)
1253 {
1254         struct inode *inode = file->f_path.dentry->d_inode;
1255         struct task_struct *p;
1256
1257         p = get_proc_task(inode);
1258         if (!p)
1259                 return -ESRCH;
1260         proc_sched_set_task(p);
1261
1262         put_task_struct(p);
1263
1264         return count;
1265 }
1266
1267 static int sched_open(struct inode *inode, struct file *filp)
1268 {
1269         return single_open(filp, sched_show, inode);
1270 }
1271
1272 static const struct file_operations proc_pid_sched_operations = {
1273         .open           = sched_open,
1274         .read           = seq_read,
1275         .write          = sched_write,
1276         .llseek         = seq_lseek,
1277         .release        = single_release,
1278 };
1279
1280 #endif
1281
1282 #ifdef CONFIG_SCHED_AUTOGROUP
1283 /*
1284  * Print out autogroup related information:
1285  */
1286 static int sched_autogroup_show(struct seq_file *m, void *v)
1287 {
1288         struct inode *inode = m->private;
1289         struct task_struct *p;
1290
1291         p = get_proc_task(inode);
1292         if (!p)
1293                 return -ESRCH;
1294         proc_sched_autogroup_show_task(p, m);
1295
1296         put_task_struct(p);
1297
1298         return 0;
1299 }
1300
1301 static ssize_t
1302 sched_autogroup_write(struct file *file, const char __user *buf,
1303             size_t count, loff_t *offset)
1304 {
1305         struct inode *inode = file->f_path.dentry->d_inode;
1306         struct task_struct *p;
1307         char buffer[PROC_NUMBUF];
1308         int nice;
1309         int err;
1310
1311         memset(buffer, 0, sizeof(buffer));
1312         if (count > sizeof(buffer) - 1)
1313                 count = sizeof(buffer) - 1;
1314         if (copy_from_user(buffer, buf, count))
1315                 return -EFAULT;
1316
1317         err = kstrtoint(strstrip(buffer), 0, &nice);
1318         if (err < 0)
1319                 return err;
1320
1321         p = get_proc_task(inode);
1322         if (!p)
1323                 return -ESRCH;
1324
1325         err = proc_sched_autogroup_set_nice(p, nice);
1326         if (err)
1327                 count = err;
1328
1329         put_task_struct(p);
1330
1331         return count;
1332 }
1333
1334 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1335 {
1336         int ret;
1337
1338         ret = single_open(filp, sched_autogroup_show, NULL);
1339         if (!ret) {
1340                 struct seq_file *m = filp->private_data;
1341
1342                 m->private = inode;
1343         }
1344         return ret;
1345 }
1346
1347 static const struct file_operations proc_pid_sched_autogroup_operations = {
1348         .open           = sched_autogroup_open,
1349         .read           = seq_read,
1350         .write          = sched_autogroup_write,
1351         .llseek         = seq_lseek,
1352         .release        = single_release,
1353 };
1354
1355 #endif /* CONFIG_SCHED_AUTOGROUP */
1356
1357 static ssize_t comm_write(struct file *file, const char __user *buf,
1358                                 size_t count, loff_t *offset)
1359 {
1360         struct inode *inode = file->f_path.dentry->d_inode;
1361         struct task_struct *p;
1362         char buffer[TASK_COMM_LEN];
1363
1364         memset(buffer, 0, sizeof(buffer));
1365         if (count > sizeof(buffer) - 1)
1366                 count = sizeof(buffer) - 1;
1367         if (copy_from_user(buffer, buf, count))
1368                 return -EFAULT;
1369
1370         p = get_proc_task(inode);
1371         if (!p)
1372                 return -ESRCH;
1373
1374         if (same_thread_group(current, p))
1375                 set_task_comm(p, buffer);
1376         else
1377                 count = -EINVAL;
1378
1379         put_task_struct(p);
1380
1381         return count;
1382 }
1383
1384 static int comm_show(struct seq_file *m, void *v)
1385 {
1386         struct inode *inode = m->private;
1387         struct task_struct *p;
1388
1389         p = get_proc_task(inode);
1390         if (!p)
1391                 return -ESRCH;
1392
1393         task_lock(p);
1394         seq_printf(m, "%s\n", p->comm);
1395         task_unlock(p);
1396
1397         put_task_struct(p);
1398
1399         return 0;
1400 }
1401
1402 static int comm_open(struct inode *inode, struct file *filp)
1403 {
1404         return single_open(filp, comm_show, inode);
1405 }
1406
1407 static const struct file_operations proc_pid_set_comm_operations = {
1408         .open           = comm_open,
1409         .read           = seq_read,
1410         .write          = comm_write,
1411         .llseek         = seq_lseek,
1412         .release        = single_release,
1413 };
1414
1415 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1416 {
1417         struct task_struct *task;
1418         struct mm_struct *mm;
1419         struct file *exe_file;
1420
1421         task = get_proc_task(dentry->d_inode);
1422         if (!task)
1423                 return -ENOENT;
1424         mm = get_task_mm(task);
1425         put_task_struct(task);
1426         if (!mm)
1427                 return -ENOENT;
1428         exe_file = get_mm_exe_file(mm);
1429         mmput(mm);
1430         if (exe_file) {
1431                 *exe_path = exe_file->f_path;
1432                 path_get(&exe_file->f_path);
1433                 fput(exe_file);
1434                 return 0;
1435         } else
1436                 return -ENOENT;
1437 }
1438
1439 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1440 {
1441         struct inode *inode = dentry->d_inode;
1442         struct path path;
1443         int error = -EACCES;
1444
1445         /* Are we allowed to snoop on the tasks file descriptors? */
1446         if (!proc_fd_access_allowed(inode))
1447                 goto out;
1448
1449         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1450         if (error)
1451                 goto out;
1452
1453         nd_jump_link(nd, &path);
1454         return NULL;
1455 out:
1456         return ERR_PTR(error);
1457 }
1458
1459 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1460 {
1461         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1462         char *pathname;
1463         int len;
1464
1465         if (!tmp)
1466                 return -ENOMEM;
1467
1468         pathname = d_path(path, tmp, PAGE_SIZE);
1469         len = PTR_ERR(pathname);
1470         if (IS_ERR(pathname))
1471                 goto out;
1472         len = tmp + PAGE_SIZE - 1 - pathname;
1473
1474         if (len > buflen)
1475                 len = buflen;
1476         if (copy_to_user(buffer, pathname, len))
1477                 len = -EFAULT;
1478  out:
1479         free_page((unsigned long)tmp);
1480         return len;
1481 }
1482
1483 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1484 {
1485         int error = -EACCES;
1486         struct inode *inode = dentry->d_inode;
1487         struct path path;
1488
1489         /* Are we allowed to snoop on the tasks file descriptors? */
1490         if (!proc_fd_access_allowed(inode))
1491                 goto out;
1492
1493         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1494         if (error)
1495                 goto out;
1496
1497         error = do_proc_readlink(&path, buffer, buflen);
1498         path_put(&path);
1499 out:
1500         return error;
1501 }
1502
1503 static const struct inode_operations proc_pid_link_inode_operations = {
1504         .readlink       = proc_pid_readlink,
1505         .follow_link    = proc_pid_follow_link,
1506         .setattr        = proc_setattr,
1507 };
1508
1509
1510 /* building an inode */
1511
1512 static int task_dumpable(struct task_struct *task)
1513 {
1514         int dumpable = 0;
1515         struct mm_struct *mm;
1516
1517         task_lock(task);
1518         mm = task->mm;
1519         if (mm)
1520                 dumpable = get_dumpable(mm);
1521         task_unlock(task);
1522         if(dumpable == 1)
1523                 return 1;
1524         return 0;
1525 }
1526
1527 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1528 {
1529         struct inode * inode;
1530         struct proc_inode *ei;
1531         const struct cred *cred;
1532
1533         /* We need a new inode */
1534
1535         inode = new_inode(sb);
1536         if (!inode)
1537                 goto out;
1538
1539         /* Common stuff */
1540         ei = PROC_I(inode);
1541         inode->i_ino = get_next_ino();
1542         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1543         inode->i_op = &proc_def_inode_operations;
1544
1545         /*
1546          * grab the reference to task.
1547          */
1548         ei->pid = get_task_pid(task, PIDTYPE_PID);
1549         if (!ei->pid)
1550                 goto out_unlock;
1551
1552         if (task_dumpable(task)) {
1553                 rcu_read_lock();
1554                 cred = __task_cred(task);
1555                 inode->i_uid = cred->euid;
1556                 inode->i_gid = cred->egid;
1557                 rcu_read_unlock();
1558         }
1559         security_task_to_inode(task, inode);
1560
1561 out:
1562         return inode;
1563
1564 out_unlock:
1565         iput(inode);
1566         return NULL;
1567 }
1568
1569 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1570 {
1571         struct inode *inode = dentry->d_inode;
1572         struct task_struct *task;
1573         const struct cred *cred;
1574         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1575
1576         generic_fillattr(inode, stat);
1577
1578         rcu_read_lock();
1579         stat->uid = GLOBAL_ROOT_UID;
1580         stat->gid = GLOBAL_ROOT_GID;
1581         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1582         if (task) {
1583                 if (!has_pid_permissions(pid, task, 2)) {
1584                         rcu_read_unlock();
1585                         /*
1586                          * This doesn't prevent learning whether PID exists,
1587                          * it only makes getattr() consistent with readdir().
1588                          */
1589                         return -ENOENT;
1590                 }
1591                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1592                     task_dumpable(task)) {
1593                         cred = __task_cred(task);
1594                         stat->uid = cred->euid;
1595                         stat->gid = cred->egid;
1596                 }
1597         }
1598         rcu_read_unlock();
1599         return 0;
1600 }
1601
1602 /* dentry stuff */
1603
1604 /*
1605  *      Exceptional case: normally we are not allowed to unhash a busy
1606  * directory. In this case, however, we can do it - no aliasing problems
1607  * due to the way we treat inodes.
1608  *
1609  * Rewrite the inode's ownerships here because the owning task may have
1610  * performed a setuid(), etc.
1611  *
1612  * Before the /proc/pid/status file was created the only way to read
1613  * the effective uid of a /process was to stat /proc/pid.  Reading
1614  * /proc/pid/status is slow enough that procps and other packages
1615  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1616  * made this apply to all per process world readable and executable
1617  * directories.
1618  */
1619 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1620 {
1621         struct inode *inode;
1622         struct task_struct *task;
1623         const struct cred *cred;
1624
1625         if (flags & LOOKUP_RCU)
1626                 return -ECHILD;
1627
1628         inode = dentry->d_inode;
1629         task = get_proc_task(inode);
1630
1631         if (task) {
1632                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1633                     task_dumpable(task)) {
1634                         rcu_read_lock();
1635                         cred = __task_cred(task);
1636                         inode->i_uid = cred->euid;
1637                         inode->i_gid = cred->egid;
1638                         rcu_read_unlock();
1639                 } else {
1640                         inode->i_uid = GLOBAL_ROOT_UID;
1641                         inode->i_gid = GLOBAL_ROOT_GID;
1642                 }
1643                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1644                 security_task_to_inode(task, inode);
1645                 put_task_struct(task);
1646                 return 1;
1647         }
1648         d_drop(dentry);
1649         return 0;
1650 }
1651
1652 static int pid_delete_dentry(const struct dentry * dentry)
1653 {
1654         /* Is the task we represent dead?
1655          * If so, then don't put the dentry on the lru list,
1656          * kill it immediately.
1657          */
1658         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1659 }
1660
1661 const struct dentry_operations pid_dentry_operations =
1662 {
1663         .d_revalidate   = pid_revalidate,
1664         .d_delete       = pid_delete_dentry,
1665 };
1666
1667 /* Lookups */
1668
1669 /*
1670  * Fill a directory entry.
1671  *
1672  * If possible create the dcache entry and derive our inode number and
1673  * file type from dcache entry.
1674  *
1675  * Since all of the proc inode numbers are dynamically generated, the inode
1676  * numbers do not exist until the inode is cache.  This means creating the
1677  * the dcache entry in readdir is necessary to keep the inode numbers
1678  * reported by readdir in sync with the inode numbers reported
1679  * by stat.
1680  */
1681 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1682         const char *name, int len,
1683         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1684 {
1685         struct dentry *child, *dir = filp->f_path.dentry;
1686         struct inode *inode;
1687         struct qstr qname;
1688         ino_t ino = 0;
1689         unsigned type = DT_UNKNOWN;
1690
1691         qname.name = name;
1692         qname.len  = len;
1693         qname.hash = full_name_hash(name, len);
1694
1695         child = d_lookup(dir, &qname);
1696         if (!child) {
1697                 struct dentry *new;
1698                 new = d_alloc(dir, &qname);
1699                 if (new) {
1700                         child = instantiate(dir->d_inode, new, task, ptr);
1701                         if (child)
1702                                 dput(new);
1703                         else
1704                                 child = new;
1705                 }
1706         }
1707         if (!child || IS_ERR(child) || !child->d_inode)
1708                 goto end_instantiate;
1709         inode = child->d_inode;
1710         if (inode) {
1711                 ino = inode->i_ino;
1712                 type = inode->i_mode >> 12;
1713         }
1714         dput(child);
1715 end_instantiate:
1716         if (!ino)
1717                 ino = find_inode_number(dir, &qname);
1718         if (!ino)
1719                 ino = 1;
1720         return filldir(dirent, name, len, filp->f_pos, ino, type);
1721 }
1722
1723 static unsigned name_to_int(struct dentry *dentry)
1724 {
1725         const char *name = dentry->d_name.name;
1726         int len = dentry->d_name.len;
1727         unsigned n = 0;
1728
1729         if (len > 1 && *name == '0')
1730                 goto out;
1731         while (len-- > 0) {
1732                 unsigned c = *name++ - '0';
1733                 if (c > 9)
1734                         goto out;
1735                 if (n >= (~0U-9)/10)
1736                         goto out;
1737                 n *= 10;
1738                 n += c;
1739         }
1740         return n;
1741 out:
1742         return ~0U;
1743 }
1744
1745 #define PROC_FDINFO_MAX 64
1746
1747 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1748 {
1749         struct task_struct *task = get_proc_task(inode);
1750         struct files_struct *files = NULL;
1751         struct file *file;
1752         int fd = proc_fd(inode);
1753
1754         if (task) {
1755                 files = get_files_struct(task);
1756                 put_task_struct(task);
1757         }
1758         if (files) {
1759                 /*
1760                  * We are not taking a ref to the file structure, so we must
1761                  * hold ->file_lock.
1762                  */
1763                 spin_lock(&files->file_lock);
1764                 file = fcheck_files(files, fd);
1765                 if (file) {
1766                         unsigned int f_flags;
1767                         struct fdtable *fdt;
1768
1769                         fdt = files_fdtable(files);
1770                         f_flags = file->f_flags & ~O_CLOEXEC;
1771                         if (close_on_exec(fd, fdt))
1772                                 f_flags |= O_CLOEXEC;
1773
1774                         if (path) {
1775                                 *path = file->f_path;
1776                                 path_get(&file->f_path);
1777                         }
1778                         if (info)
1779                                 snprintf(info, PROC_FDINFO_MAX,
1780                                          "pos:\t%lli\n"
1781                                          "flags:\t0%o\n",
1782                                          (long long) file->f_pos,
1783                                          f_flags);
1784                         spin_unlock(&files->file_lock);
1785                         put_files_struct(files);
1786                         return 0;
1787                 }
1788                 spin_unlock(&files->file_lock);
1789                 put_files_struct(files);
1790         }
1791         return -ENOENT;
1792 }
1793
1794 static int proc_fd_link(struct dentry *dentry, struct path *path)
1795 {
1796         return proc_fd_info(dentry->d_inode, path, NULL);
1797 }
1798
1799 static int tid_fd_revalidate(struct dentry *dentry, unsigned int flags)
1800 {
1801         struct inode *inode;
1802         struct task_struct *task;
1803         int fd;
1804         struct files_struct *files;
1805         const struct cred *cred;
1806
1807         if (flags & LOOKUP_RCU)
1808                 return -ECHILD;
1809
1810         inode = dentry->d_inode;
1811         task = get_proc_task(inode);
1812         fd = proc_fd(inode);
1813
1814         if (task) {
1815                 files = get_files_struct(task);
1816                 if (files) {
1817                         struct file *file;
1818                         rcu_read_lock();
1819                         file = fcheck_files(files, fd);
1820                         if (file) {
1821                                 unsigned f_mode = file->f_mode;
1822
1823                                 rcu_read_unlock();
1824                                 put_files_struct(files);
1825
1826                                 if (task_dumpable(task)) {
1827                                         rcu_read_lock();
1828                                         cred = __task_cred(task);
1829                                         inode->i_uid = cred->euid;
1830                                         inode->i_gid = cred->egid;
1831                                         rcu_read_unlock();
1832                                 } else {
1833                                         inode->i_uid = GLOBAL_ROOT_UID;
1834                                         inode->i_gid = GLOBAL_ROOT_GID;
1835                                 }
1836
1837                                 if (S_ISLNK(inode->i_mode)) {
1838                                         unsigned i_mode = S_IFLNK;
1839                                         if (f_mode & FMODE_READ)
1840                                                 i_mode |= S_IRUSR | S_IXUSR;
1841                                         if (f_mode & FMODE_WRITE)
1842                                                 i_mode |= S_IWUSR | S_IXUSR;
1843                                         inode->i_mode = i_mode;
1844                                 }
1845
1846                                 security_task_to_inode(task, inode);
1847                                 put_task_struct(task);
1848                                 return 1;
1849                         }
1850                         rcu_read_unlock();
1851                         put_files_struct(files);
1852                 }
1853                 put_task_struct(task);
1854         }
1855         d_drop(dentry);
1856         return 0;
1857 }
1858
1859 static const struct dentry_operations tid_fd_dentry_operations =
1860 {
1861         .d_revalidate   = tid_fd_revalidate,
1862         .d_delete       = pid_delete_dentry,
1863 };
1864
1865 static struct dentry *proc_fd_instantiate(struct inode *dir,
1866         struct dentry *dentry, struct task_struct *task, const void *ptr)
1867 {
1868         unsigned fd = (unsigned long)ptr;
1869         struct inode *inode;
1870         struct proc_inode *ei;
1871         struct dentry *error = ERR_PTR(-ENOENT);
1872
1873         inode = proc_pid_make_inode(dir->i_sb, task);
1874         if (!inode)
1875                 goto out;
1876         ei = PROC_I(inode);
1877         ei->fd = fd;
1878
1879         inode->i_mode = S_IFLNK;
1880         inode->i_op = &proc_pid_link_inode_operations;
1881         inode->i_size = 64;
1882         ei->op.proc_get_link = proc_fd_link;
1883         d_set_d_op(dentry, &tid_fd_dentry_operations);
1884         d_add(dentry, inode);
1885         /* Close the race of the process dying before we return the dentry */
1886         if (tid_fd_revalidate(dentry, 0))
1887                 error = NULL;
1888
1889  out:
1890         return error;
1891 }
1892
1893 static struct dentry *proc_lookupfd_common(struct inode *dir,
1894                                            struct dentry *dentry,
1895                                            instantiate_t instantiate)
1896 {
1897         struct task_struct *task = get_proc_task(dir);
1898         unsigned fd = name_to_int(dentry);
1899         struct dentry *result = ERR_PTR(-ENOENT);
1900
1901         if (!task)
1902                 goto out_no_task;
1903         if (fd == ~0U)
1904                 goto out;
1905
1906         result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1907 out:
1908         put_task_struct(task);
1909 out_no_task:
1910         return result;
1911 }
1912
1913 static int proc_readfd_common(struct file * filp, void * dirent,
1914                               filldir_t filldir, instantiate_t instantiate)
1915 {
1916         struct dentry *dentry = filp->f_path.dentry;
1917         struct inode *inode = dentry->d_inode;
1918         struct task_struct *p = get_proc_task(inode);
1919         unsigned int fd, ino;
1920         int retval;
1921         struct files_struct * files;
1922
1923         retval = -ENOENT;
1924         if (!p)
1925                 goto out_no_task;
1926         retval = 0;
1927
1928         fd = filp->f_pos;
1929         switch (fd) {
1930                 case 0:
1931                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1932                                 goto out;
1933                         filp->f_pos++;
1934                 case 1:
1935                         ino = parent_ino(dentry);
1936                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1937                                 goto out;
1938                         filp->f_pos++;
1939                 default:
1940                         files = get_files_struct(p);
1941                         if (!files)
1942                                 goto out;
1943                         rcu_read_lock();
1944                         for (fd = filp->f_pos-2;
1945                              fd < files_fdtable(files)->max_fds;
1946                              fd++, filp->f_pos++) {
1947                                 char name[PROC_NUMBUF];
1948                                 int len;
1949                                 int rv;
1950
1951                                 if (!fcheck_files(files, fd))
1952                                         continue;
1953                                 rcu_read_unlock();
1954
1955                                 len = snprintf(name, sizeof(name), "%d", fd);
1956                                 rv = proc_fill_cache(filp, dirent, filldir,
1957                                                      name, len, instantiate, p,
1958                                                      (void *)(unsigned long)fd);
1959                                 if (rv < 0)
1960                                         goto out_fd_loop;
1961                                 rcu_read_lock();
1962                         }
1963                         rcu_read_unlock();
1964 out_fd_loop:
1965                         put_files_struct(files);
1966         }
1967 out:
1968         put_task_struct(p);
1969 out_no_task:
1970         return retval;
1971 }
1972
1973 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1974                                     unsigned int flags)
1975 {
1976         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1977 }
1978
1979 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1980 {
1981         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1982 }
1983
1984 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1985                                       size_t len, loff_t *ppos)
1986 {
1987         char tmp[PROC_FDINFO_MAX];
1988         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1989         if (!err)
1990                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1991         return err;
1992 }
1993
1994 static const struct file_operations proc_fdinfo_file_operations = {
1995         .open           = nonseekable_open,
1996         .read           = proc_fdinfo_read,
1997         .llseek         = no_llseek,
1998 };
1999
2000 static const struct file_operations proc_fd_operations = {
2001         .read           = generic_read_dir,
2002         .readdir        = proc_readfd,
2003         .llseek         = default_llseek,
2004 };
2005
2006 #ifdef CONFIG_CHECKPOINT_RESTORE
2007
2008 /*
2009  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2010  * which represent vma start and end addresses.
2011  */
2012 static int dname_to_vma_addr(struct dentry *dentry,
2013                              unsigned long *start, unsigned long *end)
2014 {
2015         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2016                 return -EINVAL;
2017
2018         return 0;
2019 }
2020
2021 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2022 {
2023         unsigned long vm_start, vm_end;
2024         bool exact_vma_exists = false;
2025         struct mm_struct *mm = NULL;
2026         struct task_struct *task;
2027         const struct cred *cred;
2028         struct inode *inode;
2029         int status = 0;
2030
2031         if (flags & LOOKUP_RCU)
2032                 return -ECHILD;
2033
2034         if (!capable(CAP_SYS_ADMIN)) {
2035                 status = -EACCES;
2036                 goto out_notask;
2037         }
2038
2039         inode = dentry->d_inode;
2040         task = get_proc_task(inode);
2041         if (!task)
2042                 goto out_notask;
2043
2044         mm = mm_access(task, PTRACE_MODE_READ);
2045         if (IS_ERR_OR_NULL(mm))
2046                 goto out;
2047
2048         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2049                 down_read(&mm->mmap_sem);
2050                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2051                 up_read(&mm->mmap_sem);
2052         }
2053
2054         mmput(mm);
2055
2056         if (exact_vma_exists) {
2057                 if (task_dumpable(task)) {
2058                         rcu_read_lock();
2059                         cred = __task_cred(task);
2060                         inode->i_uid = cred->euid;
2061                         inode->i_gid = cred->egid;
2062                         rcu_read_unlock();
2063                 } else {
2064                         inode->i_uid = GLOBAL_ROOT_UID;
2065                         inode->i_gid = GLOBAL_ROOT_GID;
2066                 }
2067                 security_task_to_inode(task, inode);
2068                 status = 1;
2069         }
2070
2071 out:
2072         put_task_struct(task);
2073
2074 out_notask:
2075         if (status <= 0)
2076                 d_drop(dentry);
2077
2078         return status;
2079 }
2080
2081 static const struct dentry_operations tid_map_files_dentry_operations = {
2082         .d_revalidate   = map_files_d_revalidate,
2083         .d_delete       = pid_delete_dentry,
2084 };
2085
2086 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2087 {
2088         unsigned long vm_start, vm_end;
2089         struct vm_area_struct *vma;
2090         struct task_struct *task;
2091         struct mm_struct *mm;
2092         int rc;
2093
2094         rc = -ENOENT;
2095         task = get_proc_task(dentry->d_inode);
2096         if (!task)
2097                 goto out;
2098
2099         mm = get_task_mm(task);
2100         put_task_struct(task);
2101         if (!mm)
2102                 goto out;
2103
2104         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2105         if (rc)
2106                 goto out_mmput;
2107
2108         down_read(&mm->mmap_sem);
2109         vma = find_exact_vma(mm, vm_start, vm_end);
2110         if (vma && vma->vm_file) {
2111                 *path = vma->vm_file->f_path;
2112                 path_get(path);
2113                 rc = 0;
2114         }
2115         up_read(&mm->mmap_sem);
2116
2117 out_mmput:
2118         mmput(mm);
2119 out:
2120         return rc;
2121 }
2122
2123 struct map_files_info {
2124         struct file     *file;
2125         unsigned long   len;
2126         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2127 };
2128
2129 static struct dentry *
2130 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2131                            struct task_struct *task, const void *ptr)
2132 {
2133         const struct file *file = ptr;
2134         struct proc_inode *ei;
2135         struct inode *inode;
2136
2137         if (!file)
2138                 return ERR_PTR(-ENOENT);
2139
2140         inode = proc_pid_make_inode(dir->i_sb, task);
2141         if (!inode)
2142                 return ERR_PTR(-ENOENT);
2143
2144         ei = PROC_I(inode);
2145         ei->op.proc_get_link = proc_map_files_get_link;
2146
2147         inode->i_op = &proc_pid_link_inode_operations;
2148         inode->i_size = 64;
2149         inode->i_mode = S_IFLNK;
2150
2151         if (file->f_mode & FMODE_READ)
2152                 inode->i_mode |= S_IRUSR;
2153         if (file->f_mode & FMODE_WRITE)
2154                 inode->i_mode |= S_IWUSR;
2155
2156         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2157         d_add(dentry, inode);
2158
2159         return NULL;
2160 }
2161
2162 static struct dentry *proc_map_files_lookup(struct inode *dir,
2163                 struct dentry *dentry, unsigned int flags)
2164 {
2165         unsigned long vm_start, vm_end;
2166         struct vm_area_struct *vma;
2167         struct task_struct *task;
2168         struct dentry *result;
2169         struct mm_struct *mm;
2170
2171         result = ERR_PTR(-EACCES);
2172         if (!capable(CAP_SYS_ADMIN))
2173                 goto out;
2174
2175         result = ERR_PTR(-ENOENT);
2176         task = get_proc_task(dir);
2177         if (!task)
2178                 goto out;
2179
2180         result = ERR_PTR(-EACCES);
2181         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2182                 goto out_put_task;
2183
2184         result = ERR_PTR(-ENOENT);
2185         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2186                 goto out_put_task;
2187
2188         mm = get_task_mm(task);
2189         if (!mm)
2190                 goto out_put_task;
2191
2192         down_read(&mm->mmap_sem);
2193         vma = find_exact_vma(mm, vm_start, vm_end);
2194         if (!vma)
2195                 goto out_no_vma;
2196
2197         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2198
2199 out_no_vma:
2200         up_read(&mm->mmap_sem);
2201         mmput(mm);
2202 out_put_task:
2203         put_task_struct(task);
2204 out:
2205         return result;
2206 }
2207
2208 static const struct inode_operations proc_map_files_inode_operations = {
2209         .lookup         = proc_map_files_lookup,
2210         .permission     = proc_fd_permission,
2211         .setattr        = proc_setattr,
2212 };
2213
2214 static int
2215 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2216 {
2217         struct dentry *dentry = filp->f_path.dentry;
2218         struct inode *inode = dentry->d_inode;
2219         struct vm_area_struct *vma;
2220         struct task_struct *task;
2221         struct mm_struct *mm;
2222         ino_t ino;
2223         int ret;
2224
2225         ret = -EACCES;
2226         if (!capable(CAP_SYS_ADMIN))
2227                 goto out;
2228
2229         ret = -ENOENT;
2230         task = get_proc_task(inode);
2231         if (!task)
2232                 goto out;
2233
2234         ret = -EACCES;
2235         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2236                 goto out_put_task;
2237
2238         ret = 0;
2239         switch (filp->f_pos) {
2240         case 0:
2241                 ino = inode->i_ino;
2242                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2243                         goto out_put_task;
2244                 filp->f_pos++;
2245         case 1:
2246                 ino = parent_ino(dentry);
2247                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2248                         goto out_put_task;
2249                 filp->f_pos++;
2250         default:
2251         {
2252                 unsigned long nr_files, pos, i;
2253                 struct flex_array *fa = NULL;
2254                 struct map_files_info info;
2255                 struct map_files_info *p;
2256
2257                 mm = get_task_mm(task);
2258                 if (!mm)
2259                         goto out_put_task;
2260                 down_read(&mm->mmap_sem);
2261
2262                 nr_files = 0;
2263
2264                 /*
2265                  * We need two passes here:
2266                  *
2267                  *  1) Collect vmas of mapped files with mmap_sem taken
2268                  *  2) Release mmap_sem and instantiate entries
2269                  *
2270                  * otherwise we get lockdep complained, since filldir()
2271                  * routine might require mmap_sem taken in might_fault().
2272                  */
2273
2274                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2275                         if (vma->vm_file && ++pos > filp->f_pos)
2276                                 nr_files++;
2277                 }
2278
2279                 if (nr_files) {
2280                         fa = flex_array_alloc(sizeof(info), nr_files,
2281                                                 GFP_KERNEL);
2282                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2283                                                         GFP_KERNEL)) {
2284                                 ret = -ENOMEM;
2285                                 if (fa)
2286                                         flex_array_free(fa);
2287                                 up_read(&mm->mmap_sem);
2288                                 mmput(mm);
2289                                 goto out_put_task;
2290                         }
2291                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2292                                         vma = vma->vm_next) {
2293                                 if (!vma->vm_file)
2294                                         continue;
2295                                 if (++pos <= filp->f_pos)
2296                                         continue;
2297
2298                                 get_file(vma->vm_file);
2299                                 info.file = vma->vm_file;
2300                                 info.len = snprintf(info.name,
2301                                                 sizeof(info.name), "%lx-%lx",
2302                                                 vma->vm_start, vma->vm_end);
2303                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2304                                         BUG();
2305                         }
2306                 }
2307                 up_read(&mm->mmap_sem);
2308
2309                 for (i = 0; i < nr_files; i++) {
2310                         p = flex_array_get(fa, i);
2311                         ret = proc_fill_cache(filp, dirent, filldir,
2312                                               p->name, p->len,
2313                                               proc_map_files_instantiate,
2314                                               task, p->file);
2315                         if (ret)
2316                                 break;
2317                         filp->f_pos++;
2318                         fput(p->file);
2319                 }
2320                 for (; i < nr_files; i++) {
2321                         /*
2322                          * In case of error don't forget
2323                          * to put rest of file refs.
2324                          */
2325                         p = flex_array_get(fa, i);
2326                         fput(p->file);
2327                 }
2328                 if (fa)
2329                         flex_array_free(fa);
2330                 mmput(mm);
2331         }
2332         }
2333
2334 out_put_task:
2335         put_task_struct(task);
2336 out:
2337         return ret;
2338 }
2339
2340 static const struct file_operations proc_map_files_operations = {
2341         .read           = generic_read_dir,
2342         .readdir        = proc_map_files_readdir,
2343         .llseek         = default_llseek,
2344 };
2345
2346 #endif /* CONFIG_CHECKPOINT_RESTORE */
2347
2348 /*
2349  * /proc/pid/fd needs a special permission handler so that a process can still
2350  * access /proc/self/fd after it has executed a setuid().
2351  */
2352 static int proc_fd_permission(struct inode *inode, int mask)
2353 {
2354         int rv = generic_permission(inode, mask);
2355         if (rv == 0)
2356                 return 0;
2357         if (task_pid(current) == proc_pid(inode))
2358                 rv = 0;
2359         return rv;
2360 }
2361
2362 /*
2363  * proc directories can do almost nothing..
2364  */
2365 static const struct inode_operations proc_fd_inode_operations = {
2366         .lookup         = proc_lookupfd,
2367         .permission     = proc_fd_permission,
2368         .setattr        = proc_setattr,
2369 };
2370
2371 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2372         struct dentry *dentry, struct task_struct *task, const void *ptr)
2373 {
2374         unsigned fd = (unsigned long)ptr;
2375         struct inode *inode;
2376         struct proc_inode *ei;
2377         struct dentry *error = ERR_PTR(-ENOENT);
2378
2379         inode = proc_pid_make_inode(dir->i_sb, task);
2380         if (!inode)
2381                 goto out;
2382         ei = PROC_I(inode);
2383         ei->fd = fd;
2384         inode->i_mode = S_IFREG | S_IRUSR;
2385         inode->i_fop = &proc_fdinfo_file_operations;
2386         d_set_d_op(dentry, &tid_fd_dentry_operations);
2387         d_add(dentry, inode);
2388         /* Close the race of the process dying before we return the dentry */
2389         if (tid_fd_revalidate(dentry, 0))
2390                 error = NULL;
2391
2392  out:
2393         return error;
2394 }
2395
2396 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2397                                         struct dentry *dentry,
2398                                         unsigned int flags)
2399 {
2400         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2401 }
2402
2403 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2404 {
2405         return proc_readfd_common(filp, dirent, filldir,
2406                                   proc_fdinfo_instantiate);
2407 }
2408
2409 static const struct file_operations proc_fdinfo_operations = {
2410         .read           = generic_read_dir,
2411         .readdir        = proc_readfdinfo,
2412         .llseek         = default_llseek,
2413 };
2414
2415 /*
2416  * proc directories can do almost nothing..
2417  */
2418 static const struct inode_operations proc_fdinfo_inode_operations = {
2419         .lookup         = proc_lookupfdinfo,
2420         .setattr        = proc_setattr,
2421 };
2422
2423
2424 static struct dentry *proc_pident_instantiate(struct inode *dir,
2425         struct dentry *dentry, struct task_struct *task, const void *ptr)
2426 {
2427         const struct pid_entry *p = ptr;
2428         struct inode *inode;
2429         struct proc_inode *ei;
2430         struct dentry *error = ERR_PTR(-ENOENT);
2431
2432         inode = proc_pid_make_inode(dir->i_sb, task);
2433         if (!inode)
2434                 goto out;
2435
2436         ei = PROC_I(inode);
2437         inode->i_mode = p->mode;
2438         if (S_ISDIR(inode->i_mode))
2439                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2440         if (p->iop)
2441                 inode->i_op = p->iop;
2442         if (p->fop)
2443                 inode->i_fop = p->fop;
2444         ei->op = p->op;
2445         d_set_d_op(dentry, &pid_dentry_operations);
2446         d_add(dentry, inode);
2447         /* Close the race of the process dying before we return the dentry */
2448         if (pid_revalidate(dentry, 0))
2449                 error = NULL;
2450 out:
2451         return error;
2452 }
2453
2454 static struct dentry *proc_pident_lookup(struct inode *dir,
2455                                          struct dentry *dentry,
2456                                          const struct pid_entry *ents,
2457                                          unsigned int nents)
2458 {
2459         struct dentry *error;
2460         struct task_struct *task = get_proc_task(dir);
2461         const struct pid_entry *p, *last;
2462
2463         error = ERR_PTR(-ENOENT);
2464
2465         if (!task)
2466                 goto out_no_task;
2467
2468         /*
2469          * Yes, it does not scale. And it should not. Don't add
2470          * new entries into /proc/<tgid>/ without very good reasons.
2471          */
2472         last = &ents[nents - 1];
2473         for (p = ents; p <= last; p++) {
2474                 if (p->len != dentry->d_name.len)
2475                         continue;
2476                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2477                         break;
2478         }
2479         if (p > last)
2480                 goto out;
2481
2482         error = proc_pident_instantiate(dir, dentry, task, p);
2483 out:
2484         put_task_struct(task);
2485 out_no_task:
2486         return error;
2487 }
2488
2489 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2490         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2491 {
2492         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2493                                 proc_pident_instantiate, task, p);
2494 }
2495
2496 static int proc_pident_readdir(struct file *filp,
2497                 void *dirent, filldir_t filldir,
2498                 const struct pid_entry *ents, unsigned int nents)
2499 {
2500         int i;
2501         struct dentry *dentry = filp->f_path.dentry;
2502         struct inode *inode = dentry->d_inode;
2503         struct task_struct *task = get_proc_task(inode);
2504         const struct pid_entry *p, *last;
2505         ino_t ino;
2506         int ret;
2507
2508         ret = -ENOENT;
2509         if (!task)
2510                 goto out_no_task;
2511
2512         ret = 0;
2513         i = filp->f_pos;
2514         switch (i) {
2515         case 0:
2516                 ino = inode->i_ino;
2517                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2518                         goto out;
2519                 i++;
2520                 filp->f_pos++;
2521                 /* fall through */
2522         case 1:
2523                 ino = parent_ino(dentry);
2524                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2525                         goto out;
2526                 i++;
2527                 filp->f_pos++;
2528                 /* fall through */
2529         default:
2530                 i -= 2;
2531                 if (i >= nents) {
2532                         ret = 1;
2533                         goto out;
2534                 }
2535                 p = ents + i;
2536                 last = &ents[nents - 1];
2537                 while (p <= last) {
2538                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2539                                 goto out;
2540                         filp->f_pos++;
2541                         p++;
2542                 }
2543         }
2544
2545         ret = 1;
2546 out:
2547         put_task_struct(task);
2548 out_no_task:
2549         return ret;
2550 }
2551
2552 #ifdef CONFIG_SECURITY
2553 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2554                                   size_t count, loff_t *ppos)
2555 {
2556         struct inode * inode = file->f_path.dentry->d_inode;
2557         char *p = NULL;
2558         ssize_t length;
2559         struct task_struct *task = get_proc_task(inode);
2560
2561         if (!task)
2562                 return -ESRCH;
2563
2564         length = security_getprocattr(task,
2565                                       (char*)file->f_path.dentry->d_name.name,
2566                                       &p);
2567         put_task_struct(task);
2568         if (length > 0)
2569                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2570         kfree(p);
2571         return length;
2572 }
2573
2574 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2575                                    size_t count, loff_t *ppos)
2576 {
2577         struct inode * inode = file->f_path.dentry->d_inode;
2578         char *page;
2579         ssize_t length;
2580         struct task_struct *task = get_proc_task(inode);
2581
2582         length = -ESRCH;
2583         if (!task)
2584                 goto out_no_task;
2585         if (count > PAGE_SIZE)
2586                 count = PAGE_SIZE;
2587
2588         /* No partial writes. */
2589         length = -EINVAL;
2590         if (*ppos != 0)
2591                 goto out;
2592
2593         length = -ENOMEM;
2594         page = (char*)__get_free_page(GFP_TEMPORARY);
2595         if (!page)
2596                 goto out;
2597
2598         length = -EFAULT;
2599         if (copy_from_user(page, buf, count))
2600                 goto out_free;
2601
2602         /* Guard against adverse ptrace interaction */
2603         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2604         if (length < 0)
2605                 goto out_free;
2606
2607         length = security_setprocattr(task,
2608                                       (char*)file->f_path.dentry->d_name.name,
2609                                       (void*)page, count);
2610         mutex_unlock(&task->signal->cred_guard_mutex);
2611 out_free:
2612         free_page((unsigned long) page);
2613 out:
2614         put_task_struct(task);
2615 out_no_task:
2616         return length;
2617 }
2618
2619 static const struct file_operations proc_pid_attr_operations = {
2620         .read           = proc_pid_attr_read,
2621         .write          = proc_pid_attr_write,
2622         .llseek         = generic_file_llseek,
2623 };
2624
2625 static const struct pid_entry attr_dir_stuff[] = {
2626         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2627         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2628         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2629         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2630         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2631         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2632 };
2633
2634 static int proc_attr_dir_readdir(struct file * filp,
2635                              void * dirent, filldir_t filldir)
2636 {
2637         return proc_pident_readdir(filp,dirent,filldir,
2638                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2639 }
2640
2641 static const struct file_operations proc_attr_dir_operations = {
2642         .read           = generic_read_dir,
2643         .readdir        = proc_attr_dir_readdir,
2644         .llseek         = default_llseek,
2645 };
2646
2647 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2648                                 struct dentry *dentry, unsigned int flags)
2649 {
2650         return proc_pident_lookup(dir, dentry,
2651                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2652 }
2653
2654 static const struct inode_operations proc_attr_dir_inode_operations = {
2655         .lookup         = proc_attr_dir_lookup,
2656         .getattr        = pid_getattr,
2657         .setattr        = proc_setattr,
2658 };
2659
2660 #endif
2661
2662 #ifdef CONFIG_ELF_CORE
2663 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2664                                          size_t count, loff_t *ppos)
2665 {
2666         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2667         struct mm_struct *mm;
2668         char buffer[PROC_NUMBUF];
2669         size_t len;
2670         int ret;
2671
2672         if (!task)
2673                 return -ESRCH;
2674
2675         ret = 0;
2676         mm = get_task_mm(task);
2677         if (mm) {
2678                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2679                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2680                                 MMF_DUMP_FILTER_SHIFT));
2681                 mmput(mm);
2682                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2683         }
2684
2685         put_task_struct(task);
2686
2687         return ret;
2688 }
2689
2690 static ssize_t proc_coredump_filter_write(struct file *file,
2691                                           const char __user *buf,
2692                                           size_t count,
2693                                           loff_t *ppos)
2694 {
2695         struct task_struct *task;
2696         struct mm_struct *mm;
2697         char buffer[PROC_NUMBUF], *end;
2698         unsigned int val;
2699         int ret;
2700         int i;
2701         unsigned long mask;
2702
2703         ret = -EFAULT;
2704         memset(buffer, 0, sizeof(buffer));
2705         if (count > sizeof(buffer) - 1)
2706                 count = sizeof(buffer) - 1;
2707         if (copy_from_user(buffer, buf, count))
2708                 goto out_no_task;
2709
2710         ret = -EINVAL;
2711         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2712         if (*end == '\n')
2713                 end++;
2714         if (end - buffer == 0)
2715                 goto out_no_task;
2716
2717         ret = -ESRCH;
2718         task = get_proc_task(file->f_dentry->d_inode);
2719         if (!task)
2720                 goto out_no_task;
2721
2722         ret = end - buffer;
2723         mm = get_task_mm(task);
2724         if (!mm)
2725                 goto out_no_mm;
2726
2727         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2728                 if (val & mask)
2729                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2730                 else
2731                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2732         }
2733
2734         mmput(mm);
2735  out_no_mm:
2736         put_task_struct(task);
2737  out_no_task:
2738         return ret;
2739 }
2740
2741 static const struct file_operations proc_coredump_filter_operations = {
2742         .read           = proc_coredump_filter_read,
2743         .write          = proc_coredump_filter_write,
2744         .llseek         = generic_file_llseek,
2745 };
2746 #endif
2747
2748 /*
2749  * /proc/self:
2750  */
2751 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2752                               int buflen)
2753 {
2754         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2755         pid_t tgid = task_tgid_nr_ns(current, ns);
2756         char tmp[PROC_NUMBUF];
2757         if (!tgid)
2758                 return -ENOENT;
2759         sprintf(tmp, "%d", tgid);
2760         return vfs_readlink(dentry,buffer,buflen,tmp);
2761 }
2762
2763 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2764 {
2765         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2766         pid_t tgid = task_tgid_nr_ns(current, ns);
2767         char *name = ERR_PTR(-ENOENT);
2768         if (tgid) {
2769                 name = __getname();
2770                 if (!name)
2771                         name = ERR_PTR(-ENOMEM);
2772                 else
2773                         sprintf(name, "%d", tgid);
2774         }
2775         nd_set_link(nd, name);
2776         return NULL;
2777 }
2778
2779 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2780                                 void *cookie)
2781 {
2782         char *s = nd_get_link(nd);
2783         if (!IS_ERR(s))
2784                 __putname(s);
2785 }
2786
2787 static const struct inode_operations proc_self_inode_operations = {
2788         .readlink       = proc_self_readlink,
2789         .follow_link    = proc_self_follow_link,
2790         .put_link       = proc_self_put_link,
2791 };
2792
2793 /*
2794  * proc base
2795  *
2796  * These are the directory entries in the root directory of /proc
2797  * that properly belong to the /proc filesystem, as they describe
2798  * describe something that is process related.
2799  */
2800 static const struct pid_entry proc_base_stuff[] = {
2801         NOD("self", S_IFLNK|S_IRWXUGO,
2802                 &proc_self_inode_operations, NULL, {}),
2803 };
2804
2805 static struct dentry *proc_base_instantiate(struct inode *dir,
2806         struct dentry *dentry, struct task_struct *task, const void *ptr)
2807 {
2808         const struct pid_entry *p = ptr;
2809         struct inode *inode;
2810         struct proc_inode *ei;
2811         struct dentry *error;
2812
2813         /* Allocate the inode */
2814         error = ERR_PTR(-ENOMEM);
2815         inode = new_inode(dir->i_sb);
2816         if (!inode)
2817                 goto out;
2818
2819         /* Initialize the inode */
2820         ei = PROC_I(inode);
2821         inode->i_ino = get_next_ino();
2822         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2823
2824         /*
2825          * grab the reference to the task.
2826          */
2827         ei->pid = get_task_pid(task, PIDTYPE_PID);
2828         if (!ei->pid)
2829                 goto out_iput;
2830
2831         inode->i_mode = p->mode;
2832         if (S_ISDIR(inode->i_mode))
2833                 set_nlink(inode, 2);
2834         if (S_ISLNK(inode->i_mode))
2835                 inode->i_size = 64;
2836         if (p->iop)
2837                 inode->i_op = p->iop;
2838         if (p->fop)
2839                 inode->i_fop = p->fop;
2840         ei->op = p->op;
2841         d_add(dentry, inode);
2842         error = NULL;
2843 out:
2844         return error;
2845 out_iput:
2846         iput(inode);
2847         goto out;
2848 }
2849
2850 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2851 {
2852         struct dentry *error;
2853         struct task_struct *task = get_proc_task(dir);
2854         const struct pid_entry *p, *last;
2855
2856         error = ERR_PTR(-ENOENT);
2857
2858         if (!task)
2859                 goto out_no_task;
2860
2861         /* Lookup the directory entry */
2862         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2863         for (p = proc_base_stuff; p <= last; p++) {
2864                 if (p->len != dentry->d_name.len)
2865                         continue;
2866                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2867                         break;
2868         }
2869         if (p > last)
2870                 goto out;
2871
2872         error = proc_base_instantiate(dir, dentry, task, p);
2873
2874 out:
2875         put_task_struct(task);
2876 out_no_task:
2877         return error;
2878 }
2879
2880 static int proc_base_fill_cache(struct file *filp, void *dirent,
2881         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2882 {
2883         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2884                                 proc_base_instantiate, task, p);
2885 }
2886
2887 #ifdef CONFIG_TASK_IO_ACCOUNTING
2888 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2889 {
2890         struct task_io_accounting acct = task->ioac;
2891         unsigned long flags;
2892         int result;
2893
2894         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2895         if (result)
2896                 return result;
2897
2898         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2899                 result = -EACCES;
2900                 goto out_unlock;
2901         }
2902
2903         if (whole && lock_task_sighand(task, &flags)) {
2904                 struct task_struct *t = task;
2905
2906                 task_io_accounting_add(&acct, &task->signal->ioac);
2907                 while_each_thread(task, t)
2908                         task_io_accounting_add(&acct, &t->ioac);
2909
2910                 unlock_task_sighand(task, &flags);
2911         }
2912         result = sprintf(buffer,
2913                         "rchar: %llu\n"
2914                         "wchar: %llu\n"
2915                         "syscr: %llu\n"
2916                         "syscw: %llu\n"
2917                         "read_bytes: %llu\n"
2918                         "write_bytes: %llu\n"
2919                         "cancelled_write_bytes: %llu\n",
2920                         (unsigned long long)acct.rchar,
2921                         (unsigned long long)acct.wchar,
2922                         (unsigned long long)acct.syscr,
2923                         (unsigned long long)acct.syscw,
2924                         (unsigned long long)acct.read_bytes,
2925                         (unsigned long long)acct.write_bytes,
2926                         (unsigned long long)acct.cancelled_write_bytes);
2927 out_unlock:
2928         mutex_unlock(&task->signal->cred_guard_mutex);
2929         return result;
2930 }
2931
2932 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2933 {
2934         return do_io_accounting(task, buffer, 0);
2935 }
2936
2937 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2938 {
2939         return do_io_accounting(task, buffer, 1);
2940 }
2941 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2942
2943 #ifdef CONFIG_USER_NS
2944 static int proc_id_map_open(struct inode *inode, struct file *file,
2945         struct seq_operations *seq_ops)
2946 {
2947         struct user_namespace *ns = NULL;
2948         struct task_struct *task;
2949         struct seq_file *seq;
2950         int ret = -EINVAL;
2951
2952         task = get_proc_task(inode);
2953         if (task) {
2954                 rcu_read_lock();
2955                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2956                 rcu_read_unlock();
2957                 put_task_struct(task);
2958         }
2959         if (!ns)
2960                 goto err;
2961
2962         ret = seq_open(file, seq_ops);
2963         if (ret)
2964                 goto err_put_ns;
2965
2966         seq = file->private_data;
2967         seq->private = ns;
2968
2969         return 0;
2970 err_put_ns:
2971         put_user_ns(ns);
2972 err:
2973         return ret;
2974 }
2975
2976 static int proc_id_map_release(struct inode *inode, struct file *file)
2977 {
2978         struct seq_file *seq = file->private_data;
2979         struct user_namespace *ns = seq->private;
2980         put_user_ns(ns);
2981         return seq_release(inode, file);
2982 }
2983
2984 static int proc_uid_map_open(struct inode *inode, struct file *file)
2985 {
2986         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2987 }
2988
2989 static int proc_gid_map_open(struct inode *inode, struct file *file)
2990 {
2991         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2992 }
2993
2994 static const struct file_operations proc_uid_map_operations = {
2995         .open           = proc_uid_map_open,
2996         .write          = proc_uid_map_write,
2997         .read           = seq_read,
2998         .llseek         = seq_lseek,
2999         .release        = proc_id_map_release,
3000 };
3001
3002 static const struct file_operations proc_gid_map_operations = {
3003         .open           = proc_gid_map_open,
3004         .write          = proc_gid_map_write,
3005         .read           = seq_read,
3006         .llseek         = seq_lseek,
3007         .release        = proc_id_map_release,
3008 };
3009 #endif /* CONFIG_USER_NS */
3010
3011 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3012                                 struct pid *pid, struct task_struct *task)
3013 {
3014         int err = lock_trace(task);
3015         if (!err) {
3016                 seq_printf(m, "%08x\n", task->personality);
3017                 unlock_trace(task);
3018         }
3019         return err;
3020 }
3021
3022 /*
3023  * Thread groups
3024  */
3025 static const struct file_operations proc_task_operations;
3026 static const struct inode_operations proc_task_inode_operations;
3027
3028 static const struct pid_entry tgid_base_stuff[] = {
3029         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3030         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3031 #ifdef CONFIG_CHECKPOINT_RESTORE
3032         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3033 #endif
3034         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3035         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3036 #ifdef CONFIG_NET
3037         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3038 #endif
3039         REG("environ",    S_IRUSR, proc_environ_operations),
3040         INF("auxv",       S_IRUSR, proc_pid_auxv),
3041         ONE("status",     S_IRUGO, proc_pid_status),
3042         ONE("personality", S_IRUGO, proc_pid_personality),
3043         INF("limits",     S_IRUGO, proc_pid_limits),
3044 #ifdef CONFIG_SCHED_DEBUG
3045         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3046 #endif
3047 #ifdef CONFIG_SCHED_AUTOGROUP
3048         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3049 #endif
3050         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3051 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3052         INF("syscall",    S_IRUGO, proc_pid_syscall),
3053 #endif
3054         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3055         ONE("stat",       S_IRUGO, proc_tgid_stat),
3056         ONE("statm",      S_IRUGO, proc_pid_statm),
3057         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3058 #ifdef CONFIG_NUMA
3059         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3060 #endif
3061         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3062         LNK("cwd",        proc_cwd_link),
3063         LNK("root",       proc_root_link),
3064         LNK("exe",        proc_exe_link),
3065         REG("mounts",     S_IRUGO, proc_mounts_operations),
3066         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3067         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3068 #ifdef CONFIG_PROC_PAGE_MONITOR
3069         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3070         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3071         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3072 #endif
3073 #ifdef CONFIG_SECURITY
3074         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3075 #endif
3076 #ifdef CONFIG_KALLSYMS
3077         INF("wchan",      S_IRUGO, proc_pid_wchan),
3078 #endif
3079 #ifdef CONFIG_STACKTRACE
3080         ONE("stack",      S_IRUGO, proc_pid_stack),
3081 #endif
3082 #ifdef CONFIG_SCHEDSTATS
3083         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3084 #endif
3085 #ifdef CONFIG_LATENCYTOP
3086         REG("latency",  S_IRUGO, proc_lstats_operations),
3087 #endif
3088 #ifdef CONFIG_PROC_PID_CPUSET
3089         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3090 #endif
3091 #ifdef CONFIG_CGROUPS
3092         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3093 #endif
3094         INF("oom_score",  S_IRUGO, proc_oom_score),
3095         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3096         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3097 #ifdef CONFIG_AUDITSYSCALL
3098         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3099         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3100 #endif
3101 #ifdef CONFIG_FAULT_INJECTION
3102         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3103 #endif
3104 #ifdef CONFIG_ELF_CORE
3105         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3106 #endif
3107 #ifdef CONFIG_TASK_IO_ACCOUNTING
3108         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3109 #endif
3110 #ifdef CONFIG_HARDWALL
3111         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3112 #endif
3113 #ifdef CONFIG_USER_NS
3114         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3115         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3116 #endif
3117 };
3118
3119 static int proc_tgid_base_readdir(struct file * filp,
3120                              void * dirent, filldir_t filldir)
3121 {
3122         return proc_pident_readdir(filp,dirent,filldir,
3123                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3124 }
3125
3126 static const struct file_operations proc_tgid_base_operations = {
3127         .read           = generic_read_dir,
3128         .readdir        = proc_tgid_base_readdir,
3129         .llseek         = default_llseek,
3130 };
3131
3132 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3133 {
3134         return proc_pident_lookup(dir, dentry,
3135                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3136 }
3137
3138 static const struct inode_operations proc_tgid_base_inode_operations = {
3139         .lookup         = proc_tgid_base_lookup,
3140         .getattr        = pid_getattr,
3141         .setattr        = proc_setattr,
3142         .permission     = proc_pid_permission,
3143 };
3144
3145 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3146 {
3147         struct dentry *dentry, *leader, *dir;
3148         char buf[PROC_NUMBUF];
3149         struct qstr name;
3150
3151         name.name = buf;
3152         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3153         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3154         if (dentry) {
3155                 shrink_dcache_parent(dentry);
3156                 d_drop(dentry);
3157                 dput(dentry);
3158         }
3159
3160         name.name = buf;
3161         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3162         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3163         if (!leader)
3164                 goto out;
3165
3166         name.name = "task";
3167         name.len = strlen(name.name);
3168         dir = d_hash_and_lookup(leader, &name);
3169         if (!dir)
3170                 goto out_put_leader;
3171
3172         name.name = buf;
3173         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3174         dentry = d_hash_and_lookup(dir, &name);
3175         if (dentry) {
3176                 shrink_dcache_parent(dentry);
3177                 d_drop(dentry);
3178                 dput(dentry);
3179         }
3180
3181         dput(dir);
3182 out_put_leader:
3183         dput(leader);
3184 out:
3185         return;
3186 }
3187
3188 /**
3189  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3190  * @task: task that should be flushed.
3191  *
3192  * When flushing dentries from proc, one needs to flush them from global
3193  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3194  * in. This call is supposed to do all of this job.
3195  *
3196  * Looks in the dcache for
3197  * /proc/@pid
3198  * /proc/@tgid/task/@pid
3199  * if either directory is present flushes it and all of it'ts children
3200  * from the dcache.
3201  *
3202  * It is safe and reasonable to cache /proc entries for a task until
3203  * that task exits.  After that they just clog up the dcache with
3204  * useless entries, possibly causing useful dcache entries to be
3205  * flushed instead.  This routine is proved to flush those useless
3206  * dcache entries at process exit time.
3207  *
3208  * NOTE: This routine is just an optimization so it does not guarantee
3209  *       that no dcache entries will exist at process exit time it
3210  *       just makes it very unlikely that any will persist.
3211  */
3212
3213 void proc_flush_task(struct task_struct *task)
3214 {
3215         int i;
3216         struct pid *pid, *tgid;
3217         struct upid *upid;
3218
3219         pid = task_pid(task);
3220         tgid = task_tgid(task);
3221
3222         for (i = 0; i <= pid->level; i++) {
3223                 upid = &pid->numbers[i];
3224                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3225                                         tgid->numbers[i].nr);
3226         }
3227
3228         upid = &pid->numbers[pid->level];
3229         if (upid->nr == 1)
3230                 pid_ns_release_proc(upid->ns);
3231 }
3232
3233 static struct dentry *proc_pid_instantiate(struct inode *dir,
3234                                            struct dentry * dentry,
3235                                            struct task_struct *task, const void *ptr)
3236 {
3237         struct dentry *error = ERR_PTR(-ENOENT);
3238         struct inode *inode;
3239
3240         inode = proc_pid_make_inode(dir->i_sb, task);
3241         if (!inode)
3242                 goto out;
3243
3244         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3245         inode->i_op = &proc_tgid_base_inode_operations;
3246         inode->i_fop = &proc_tgid_base_operations;
3247         inode->i_flags|=S_IMMUTABLE;
3248
3249         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3250                                                   ARRAY_SIZE(tgid_base_stuff)));
3251
3252         d_set_d_op(dentry, &pid_dentry_operations);
3253
3254         d_add(dentry, inode);
3255         /* Close the race of the process dying before we return the dentry */
3256         if (pid_revalidate(dentry, 0))
3257                 error = NULL;
3258 out:
3259         return error;
3260 }
3261
3262 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3263 {
3264         struct dentry *result;
3265         struct task_struct *task;
3266         unsigned tgid;
3267         struct pid_namespace *ns;
3268
3269         result = proc_base_lookup(dir, dentry);
3270         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3271                 goto out;
3272
3273         tgid = name_to_int(dentry);
3274         if (tgid == ~0U)
3275                 goto out;
3276
3277         ns = dentry->d_sb->s_fs_info;
3278         rcu_read_lock();
3279         task = find_task_by_pid_ns(tgid, ns);
3280         if (task)
3281                 get_task_struct(task);
3282         rcu_read_unlock();
3283         if (!task)
3284                 goto out;
3285
3286         result = proc_pid_instantiate(dir, dentry, task, NULL);
3287         put_task_struct(task);
3288 out:
3289         return result;
3290 }
3291
3292 /*
3293  * Find the first task with tgid >= tgid
3294  *
3295  */
3296 struct tgid_iter {
3297         unsigned int tgid;
3298         struct task_struct *task;
3299 };
3300 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3301 {
3302         struct pid *pid;
3303
3304         if (iter.task)
3305                 put_task_struct(iter.task);
3306         rcu_read_lock();
3307 retry:
3308         iter.task = NULL;
3309         pid = find_ge_pid(iter.tgid, ns);
3310         if (pid) {
3311                 iter.tgid = pid_nr_ns(pid, ns);
3312                 iter.task = pid_task(pid, PIDTYPE_PID);
3313                 /* What we to know is if the pid we have find is the
3314                  * pid of a thread_group_leader.  Testing for task
3315                  * being a thread_group_leader is the obvious thing
3316                  * todo but there is a window when it fails, due to
3317                  * the pid transfer logic in de_thread.
3318                  *
3319                  * So we perform the straight forward test of seeing
3320                  * if the pid we have found is the pid of a thread
3321                  * group leader, and don't worry if the task we have
3322                  * found doesn't happen to be a thread group leader.
3323                  * As we don't care in the case of readdir.
3324                  */
3325                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3326                         iter.tgid += 1;
3327                         goto retry;
3328                 }
3329                 get_task_struct(iter.task);
3330         }
3331         rcu_read_unlock();
3332         return iter;
3333 }
3334
3335 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3336
3337 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3338         struct tgid_iter iter)
3339 {
3340         char name[PROC_NUMBUF];
3341         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3342         return proc_fill_cache(filp, dirent, filldir, name, len,
3343                                 proc_pid_instantiate, iter.task, NULL);
3344 }
3345
3346 static int fake_filldir(void *buf, const char *name, int namelen,
3347                         loff_t offset, u64 ino, unsigned d_type)
3348 {
3349         return 0;
3350 }
3351
3352 /* for the /proc/ directory itself, after non-process stuff has been done */
3353 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3354 {
3355         unsigned int nr;
3356         struct task_struct *reaper;
3357         struct tgid_iter iter;
3358         struct pid_namespace *ns;
3359         filldir_t __filldir;
3360
3361         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3362                 goto out_no_task;
3363         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3364
3365         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3366         if (!reaper)
3367                 goto out_no_task;
3368
3369         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3370                 const struct pid_entry *p = &proc_base_stuff[nr];
3371                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3372                         goto out;
3373         }
3374
3375         ns = filp->f_dentry->d_sb->s_fs_info;
3376         iter.task = NULL;
3377         iter.tgid = filp->f_pos - TGID_OFFSET;
3378         for (iter = next_tgid(ns, iter);
3379              iter.task;
3380              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3381                 if (has_pid_permissions(ns, iter.task, 2))
3382                         __filldir = filldir;
3383                 else
3384                         __filldir = fake_filldir;
3385
3386                 filp->f_pos = iter.tgid + TGID_OFFSET;
3387                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3388                         put_task_struct(iter.task);
3389                         goto out;
3390                 }
3391         }
3392         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3393 out:
3394         put_task_struct(reaper);
3395 out_no_task:
3396         return 0;
3397 }
3398
3399 /*
3400  * Tasks
3401  */
3402 static const struct pid_entry tid_base_stuff[] = {
3403         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3404         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3405         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3406         REG("environ",   S_IRUSR, proc_environ_operations),
3407         INF("auxv",      S_IRUSR, proc_pid_auxv),
3408         ONE("status",    S_IRUGO, proc_pid_status),
3409         ONE("personality", S_IRUGO, proc_pid_personality),
3410         INF("limits",    S_IRUGO, proc_pid_limits),
3411 #ifdef CONFIG_SCHED_DEBUG
3412         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3413 #endif
3414         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3415 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3416         INF("syscall",   S_IRUGO, proc_pid_syscall),
3417 #endif
3418         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3419         ONE("stat",      S_IRUGO, proc_tid_stat),
3420         ONE("statm",     S_IRUGO, proc_pid_statm),
3421         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3422 #ifdef CONFIG_CHECKPOINT_RESTORE
3423         REG("children",  S_IRUGO, proc_tid_children_operations),
3424 #endif
3425 #ifdef CONFIG_NUMA
3426         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3427 #endif
3428         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3429         LNK("cwd",       proc_cwd_link),
3430         LNK("root",      proc_root_link),
3431         LNK("exe",       proc_exe_link),
3432         REG("mounts",    S_IRUGO, proc_mounts_operations),
3433         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3434 #ifdef CONFIG_PROC_PAGE_MONITOR
3435         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3436         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3437         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3438 #endif
3439 #ifdef CONFIG_SECURITY
3440         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3441 #endif
3442 #ifdef CONFIG_KALLSYMS
3443         INF("wchan",     S_IRUGO, proc_pid_wchan),
3444 #endif
3445 #ifdef CONFIG_STACKTRACE
3446         ONE("stack",      S_IRUGO, proc_pid_stack),
3447 #endif
3448 #ifdef CONFIG_SCHEDSTATS
3449         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3450 #endif
3451 #ifdef CONFIG_LATENCYTOP
3452         REG("latency",  S_IRUGO, proc_lstats_operations),
3453 #endif
3454 #ifdef CONFIG_PROC_PID_CPUSET
3455         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3456 #endif
3457 #ifdef CONFIG_CGROUPS
3458         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3459 #endif
3460         INF("oom_score", S_IRUGO, proc_oom_score),
3461         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3462         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3463 #ifdef CONFIG_AUDITSYSCALL
3464         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3465         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3466 #endif
3467 #ifdef CONFIG_FAULT_INJECTION
3468         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3469 #endif
3470 #ifdef CONFIG_TASK_IO_ACCOUNTING
3471         INF("io",       S_IRUSR, proc_tid_io_accounting),
3472 #endif
3473 #ifdef CONFIG_HARDWALL
3474         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3475 #endif
3476 #ifdef CONFIG_USER_NS
3477         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3478         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3479 #endif
3480 };
3481
3482 static int proc_tid_base_readdir(struct file * filp,
3483                              void * dirent, filldir_t filldir)
3484 {
3485         return proc_pident_readdir(filp,dirent,filldir,
3486                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3487 }
3488
3489 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3490 {
3491         return proc_pident_lookup(dir, dentry,
3492                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3493 }
3494
3495 static const struct file_operations proc_tid_base_operations = {
3496         .read           = generic_read_dir,
3497         .readdir        = proc_tid_base_readdir,
3498         .llseek         = default_llseek,
3499 };
3500
3501 static const struct inode_operations proc_tid_base_inode_operations = {
3502         .lookup         = proc_tid_base_lookup,
3503         .getattr        = pid_getattr,
3504         .setattr        = proc_setattr,
3505 };
3506
3507 static struct dentry *proc_task_instantiate(struct inode *dir,
3508         struct dentry *dentry, struct task_struct *task, const void *ptr)
3509 {
3510         struct dentry *error = ERR_PTR(-ENOENT);
3511         struct inode *inode;
3512         inode = proc_pid_make_inode(dir->i_sb, task);
3513
3514         if (!inode)
3515                 goto out;
3516         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3517         inode->i_op = &proc_tid_base_inode_operations;
3518         inode->i_fop = &proc_tid_base_operations;
3519         inode->i_flags|=S_IMMUTABLE;
3520
3521         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3522                                                   ARRAY_SIZE(tid_base_stuff)));
3523
3524         d_set_d_op(dentry, &pid_dentry_operations);
3525
3526         d_add(dentry, inode);
3527         /* Close the race of the process dying before we return the dentry */
3528         if (pid_revalidate(dentry, 0))
3529                 error = NULL;
3530 out:
3531         return error;
3532 }
3533
3534 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3535 {
3536         struct dentry *result = ERR_PTR(-ENOENT);
3537         struct task_struct *task;
3538         struct task_struct *leader = get_proc_task(dir);
3539         unsigned tid;
3540         struct pid_namespace *ns;
3541
3542         if (!leader)
3543                 goto out_no_task;
3544
3545         tid = name_to_int(dentry);
3546         if (tid == ~0U)
3547                 goto out;
3548
3549         ns = dentry->d_sb->s_fs_info;
3550         rcu_read_lock();
3551         task = find_task_by_pid_ns(tid, ns);
3552         if (task)
3553                 get_task_struct(task);
3554         rcu_read_unlock();
3555         if (!task)
3556                 goto out;
3557         if (!same_thread_group(leader, task))
3558                 goto out_drop_task;
3559
3560         result = proc_task_instantiate(dir, dentry, task, NULL);
3561 out_drop_task:
3562         put_task_struct(task);
3563 out:
3564         put_task_struct(leader);
3565 out_no_task:
3566         return result;
3567 }
3568
3569 /*
3570  * Find the first tid of a thread group to return to user space.
3571  *
3572  * Usually this is just the thread group leader, but if the users
3573  * buffer was too small or there was a seek into the middle of the
3574  * directory we have more work todo.
3575  *
3576  * In the case of a short read we start with find_task_by_pid.
3577  *
3578  * In the case of a seek we start with the leader and walk nr
3579  * threads past it.
3580  */
3581 static struct task_struct *first_tid(struct task_struct *leader,
3582                 int tid, int nr, struct pid_namespace *ns)
3583 {
3584         struct task_struct *pos;
3585
3586         rcu_read_lock();
3587         /* Attempt to start with the pid of a thread */
3588         if (tid && (nr > 0)) {
3589                 pos = find_task_by_pid_ns(tid, ns);
3590                 if (pos && (pos->group_leader == leader))
3591                         goto found;
3592         }
3593
3594         /* If nr exceeds the number of threads there is nothing todo */
3595         pos = NULL;
3596         if (nr && nr >= get_nr_threads(leader))
3597                 goto out;
3598
3599         /* If we haven't found our starting place yet start
3600          * with the leader and walk nr threads forward.
3601          */
3602         for (pos = leader; nr > 0; --nr) {
3603                 pos = next_thread(pos);
3604                 if (pos == leader) {
3605                         pos = NULL;
3606                         goto out;
3607                 }
3608         }
3609 found:
3610         get_task_struct(pos);
3611 out:
3612         rcu_read_unlock();
3613         return pos;
3614 }
3615
3616 /*
3617  * Find the next thread in the thread list.
3618  * Return NULL if there is an error or no next thread.
3619  *
3620  * The reference to the input task_struct is released.
3621  */
3622 static struct task_struct *next_tid(struct task_struct *start)
3623 {
3624         struct task_struct *pos = NULL;
3625         rcu_read_lock();
3626         if (pid_alive(start)) {
3627                 pos = next_thread(start);
3628                 if (thread_group_leader(pos))
3629                         pos = NULL;
3630                 else
3631                         get_task_struct(pos);
3632         }
3633         rcu_read_unlock();
3634         put_task_struct(start);
3635         return pos;
3636 }
3637
3638 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3639         struct task_struct *task, int tid)
3640 {
3641         char name[PROC_NUMBUF];
3642         int len = snprintf(name, sizeof(name), "%d", tid);
3643         return proc_fill_cache(filp, dirent, filldir, name, len,
3644                                 proc_task_instantiate, task, NULL);
3645 }
3646
3647 /* for the /proc/TGID/task/ directories */
3648 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3649 {
3650         struct dentry *dentry = filp->f_path.dentry;
3651         struct inode *inode = dentry->d_inode;
3652         struct task_struct *leader = NULL;
3653         struct task_struct *task;
3654         int retval = -ENOENT;
3655         ino_t ino;
3656         int tid;
3657         struct pid_namespace *ns;
3658
3659         task = get_proc_task(inode);
3660         if (!task)
3661                 goto out_no_task;
3662         rcu_read_lock();
3663         if (pid_alive(task)) {
3664                 leader = task->group_leader;
3665                 get_task_struct(leader);
3666         }
3667         rcu_read_unlock();
3668         put_task_struct(task);
3669         if (!leader)
3670                 goto out_no_task;
3671         retval = 0;
3672
3673         switch ((unsigned long)filp->f_pos) {
3674         case 0:
3675                 ino = inode->i_ino;
3676                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3677                         goto out;
3678                 filp->f_pos++;
3679                 /* fall through */
3680         case 1:
3681                 ino = parent_ino(dentry);
3682                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3683                         goto out;
3684                 filp->f_pos++;
3685                 /* fall through */
3686         }
3687
3688         /* f_version caches the tgid value that the last readdir call couldn't
3689          * return. lseek aka telldir automagically resets f_version to 0.
3690          */
3691         ns = filp->f_dentry->d_sb->s_fs_info;
3692         tid = (int)filp->f_version;
3693         filp->f_version = 0;
3694         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3695              task;
3696              task = next_tid(task), filp->f_pos++) {
3697                 tid = task_pid_nr_ns(task, ns);
3698                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3699                         /* returning this tgid failed, save it as the first
3700                          * pid for the next readir call */
3701                         filp->f_version = (u64)tid;
3702                         put_task_struct(task);
3703                         break;
3704                 }
3705         }
3706 out:
3707         put_task_struct(leader);
3708 out_no_task:
3709         return retval;
3710 }
3711
3712 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3713 {
3714         struct inode *inode = dentry->d_inode;
3715         struct task_struct *p = get_proc_task(inode);
3716         generic_fillattr(inode, stat);
3717
3718         if (p) {
3719                 stat->nlink += get_nr_threads(p);
3720                 put_task_struct(p);
3721         }
3722
3723         return 0;
3724 }
3725
3726 static const struct inode_operations proc_task_inode_operations = {
3727         .lookup         = proc_task_lookup,
3728         .getattr        = proc_task_getattr,
3729         .setattr        = proc_setattr,
3730         .permission     = proc_pid_permission,
3731 };
3732
3733 static const struct file_operations proc_task_operations = {
3734         .read           = generic_read_dir,
3735         .readdir        = proc_task_readdir,
3736         .llseek         = default_llseek,
3737 };